linux/net/tls/tls_sw.c

2585 lines
65 KiB
C
Raw Normal View History

/*
* Copyright (c) 2016-2017, Mellanox Technologies. All rights reserved.
* Copyright (c) 2016-2017, Dave Watson <davejwatson@fb.com>. All rights reserved.
* Copyright (c) 2016-2017, Lance Chao <lancerchao@fb.com>. All rights reserved.
* Copyright (c) 2016, Fridolin Pokorny <fridolin.pokorny@gmail.com>. All rights reserved.
* Copyright (c) 2016, Nikos Mavrogiannopoulos <nmav@gnutls.org>. All rights reserved.
* Copyright (c) 2018, Covalent IO, Inc. http://covalent.io
*
* This software is available to you under a choice of one of two
* licenses. You may choose to be licensed under the terms of the GNU
* General Public License (GPL) Version 2, available from the file
* COPYING in the main directory of this source tree, or the
* OpenIB.org BSD license below:
*
* Redistribution and use in source and binary forms, with or
* without modification, are permitted provided that the following
* conditions are met:
*
* - Redistributions of source code must retain the above
* copyright notice, this list of conditions and the following
* disclaimer.
*
* - Redistributions in binary form must reproduce the above
* copyright notice, this list of conditions and the following
* disclaimer in the documentation and/or other materials
* provided with the distribution.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#include <linux/bug.h>
#include <linux/sched/signal.h>
#include <linux/module.h>
#include <linux/splice.h>
#include <crypto/aead.h>
#include <net/strparser.h>
#include <net/tls.h>
#include "tls.h"
struct tls_decrypt_arg {
bool zc;
bool async;
u8 tail;
};
struct tls_decrypt_ctx {
u8 iv[MAX_IV_SIZE];
u8 aad[TLS_MAX_AAD_SIZE];
u8 tail;
struct scatterlist sg[];
};
noinline void tls_err_abort(struct sock *sk, int err)
{
WARN_ON_ONCE(err >= 0);
/* sk->sk_err should contain a positive error code. */
sk->sk_err = -err;
sk_error_report(sk);
}
static int __skb_nsg(struct sk_buff *skb, int offset, int len,
unsigned int recursion_level)
{
int start = skb_headlen(skb);
int i, chunk = start - offset;
struct sk_buff *frag_iter;
int elt = 0;
if (unlikely(recursion_level >= 24))
return -EMSGSIZE;
if (chunk > 0) {
if (chunk > len)
chunk = len;
elt++;
len -= chunk;
if (len == 0)
return elt;
offset += chunk;
}
for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
int end;
WARN_ON(start > offset + len);
end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
chunk = end - offset;
if (chunk > 0) {
if (chunk > len)
chunk = len;
elt++;
len -= chunk;
if (len == 0)
return elt;
offset += chunk;
}
start = end;
}
if (unlikely(skb_has_frag_list(skb))) {
skb_walk_frags(skb, frag_iter) {
int end, ret;
WARN_ON(start > offset + len);
end = start + frag_iter->len;
chunk = end - offset;
if (chunk > 0) {
if (chunk > len)
chunk = len;
ret = __skb_nsg(frag_iter, offset - start, chunk,
recursion_level + 1);
if (unlikely(ret < 0))
return ret;
elt += ret;
len -= chunk;
if (len == 0)
return elt;
offset += chunk;
}
start = end;
}
}
BUG_ON(len);
return elt;
}
/* Return the number of scatterlist elements required to completely map the
* skb, or -EMSGSIZE if the recursion depth is exceeded.
*/
static int skb_nsg(struct sk_buff *skb, int offset, int len)
{
return __skb_nsg(skb, offset, len, 0);
}
static int tls_padding_length(struct tls_prot_info *prot, struct sk_buff *skb,
struct tls_decrypt_arg *darg)
{
struct strp_msg *rxm = strp_msg(skb);
struct tls_msg *tlm = tls_msg(skb);
int sub = 0;
/* Determine zero-padding length */
if (prot->version == TLS_1_3_VERSION) {
int offset = rxm->full_len - TLS_TAG_SIZE - 1;
char content_type = darg->zc ? darg->tail : 0;
int err;
while (content_type == 0) {
if (offset < prot->prepend_size)
return -EBADMSG;
err = skb_copy_bits(skb, rxm->offset + offset,
&content_type, 1);
if (err)
return err;
if (content_type)
break;
sub++;
offset--;
}
tlm->control = content_type;
}
return sub;
}
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
static void tls_decrypt_done(struct crypto_async_request *req, int err)
{
struct aead_request *aead_req = (struct aead_request *)req;
struct scatterlist *sgout = aead_req->dst;
struct scatterlist *sgin = aead_req->src;
tls: async support causes out-of-bounds access in crypto APIs When async support was added it needed to access the sk from the async callback to report errors up the stack. The patch tried to use space after the aead request struct by directly setting the reqsize field in aead_request. This is an internal field that should not be used outside the crypto APIs. It is used by the crypto code to define extra space for private structures used in the crypto context. Users of the API then use crypto_aead_reqsize() and add the returned amount of bytes to the end of the request memory allocation before posting the request to encrypt/decrypt APIs. So this breaks (with general protection fault and KASAN error, if enabled) because the request sent to decrypt is shorter than required causing the crypto API out-of-bounds errors. Also it seems unlikely the sk is even valid by the time it gets to the callback because of memset in crypto layer. Anyways, fix this by holding the sk in the skb->sk field when the callback is set up and because the skb is already passed through to the callback handler via void* we can access it in the handler. Then in the handler we need to be careful to NULL the pointer again before kfree_skb. I added comments on both the setup (in tls_do_decryption) and when we clear it from the crypto callback handler tls_decrypt_done(). After this selftests pass again and fixes KASAN errors/warnings. Fixes: 94524d8fc965 ("net/tls: Add support for async decryption of tls records") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Reviewed-by: Vakul Garg <Vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-14 20:01:46 +00:00
struct tls_sw_context_rx *ctx;
struct tls_context *tls_ctx;
struct tls_prot_info *prot;
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
struct scatterlist *sg;
tls: async support causes out-of-bounds access in crypto APIs When async support was added it needed to access the sk from the async callback to report errors up the stack. The patch tried to use space after the aead request struct by directly setting the reqsize field in aead_request. This is an internal field that should not be used outside the crypto APIs. It is used by the crypto code to define extra space for private structures used in the crypto context. Users of the API then use crypto_aead_reqsize() and add the returned amount of bytes to the end of the request memory allocation before posting the request to encrypt/decrypt APIs. So this breaks (with general protection fault and KASAN error, if enabled) because the request sent to decrypt is shorter than required causing the crypto API out-of-bounds errors. Also it seems unlikely the sk is even valid by the time it gets to the callback because of memset in crypto layer. Anyways, fix this by holding the sk in the skb->sk field when the callback is set up and because the skb is already passed through to the callback handler via void* we can access it in the handler. Then in the handler we need to be careful to NULL the pointer again before kfree_skb. I added comments on both the setup (in tls_do_decryption) and when we clear it from the crypto callback handler tls_decrypt_done(). After this selftests pass again and fixes KASAN errors/warnings. Fixes: 94524d8fc965 ("net/tls: Add support for async decryption of tls records") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Reviewed-by: Vakul Garg <Vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-14 20:01:46 +00:00
struct sk_buff *skb;
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
unsigned int pages;
tls: async support causes out-of-bounds access in crypto APIs When async support was added it needed to access the sk from the async callback to report errors up the stack. The patch tried to use space after the aead request struct by directly setting the reqsize field in aead_request. This is an internal field that should not be used outside the crypto APIs. It is used by the crypto code to define extra space for private structures used in the crypto context. Users of the API then use crypto_aead_reqsize() and add the returned amount of bytes to the end of the request memory allocation before posting the request to encrypt/decrypt APIs. So this breaks (with general protection fault and KASAN error, if enabled) because the request sent to decrypt is shorter than required causing the crypto API out-of-bounds errors. Also it seems unlikely the sk is even valid by the time it gets to the callback because of memset in crypto layer. Anyways, fix this by holding the sk in the skb->sk field when the callback is set up and because the skb is already passed through to the callback handler via void* we can access it in the handler. Then in the handler we need to be careful to NULL the pointer again before kfree_skb. I added comments on both the setup (in tls_do_decryption) and when we clear it from the crypto callback handler tls_decrypt_done(). After this selftests pass again and fixes KASAN errors/warnings. Fixes: 94524d8fc965 ("net/tls: Add support for async decryption of tls records") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Reviewed-by: Vakul Garg <Vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-14 20:01:46 +00:00
skb = (struct sk_buff *)req->data;
tls_ctx = tls_get_ctx(skb->sk);
ctx = tls_sw_ctx_rx(tls_ctx);
prot = &tls_ctx->prot_info;
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
/* Propagate if there was an err */
if (err) {
if (err == -EBADMSG)
TLS_INC_STATS(sock_net(skb->sk),
LINUX_MIB_TLSDECRYPTERROR);
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
ctx->async_wait.err = err;
tls: async support causes out-of-bounds access in crypto APIs When async support was added it needed to access the sk from the async callback to report errors up the stack. The patch tried to use space after the aead request struct by directly setting the reqsize field in aead_request. This is an internal field that should not be used outside the crypto APIs. It is used by the crypto code to define extra space for private structures used in the crypto context. Users of the API then use crypto_aead_reqsize() and add the returned amount of bytes to the end of the request memory allocation before posting the request to encrypt/decrypt APIs. So this breaks (with general protection fault and KASAN error, if enabled) because the request sent to decrypt is shorter than required causing the crypto API out-of-bounds errors. Also it seems unlikely the sk is even valid by the time it gets to the callback because of memset in crypto layer. Anyways, fix this by holding the sk in the skb->sk field when the callback is set up and because the skb is already passed through to the callback handler via void* we can access it in the handler. Then in the handler we need to be careful to NULL the pointer again before kfree_skb. I added comments on both the setup (in tls_do_decryption) and when we clear it from the crypto callback handler tls_decrypt_done(). After this selftests pass again and fixes KASAN errors/warnings. Fixes: 94524d8fc965 ("net/tls: Add support for async decryption of tls records") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Reviewed-by: Vakul Garg <Vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-14 20:01:46 +00:00
tls_err_abort(skb->sk, err);
} else {
struct strp_msg *rxm = strp_msg(skb);
/* No TLS 1.3 support with async crypto */
WARN_ON(prot->tail_size);
rxm->offset += prot->prepend_size;
rxm->full_len -= prot->overhead_size;
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
}
tls: async support causes out-of-bounds access in crypto APIs When async support was added it needed to access the sk from the async callback to report errors up the stack. The patch tried to use space after the aead request struct by directly setting the reqsize field in aead_request. This is an internal field that should not be used outside the crypto APIs. It is used by the crypto code to define extra space for private structures used in the crypto context. Users of the API then use crypto_aead_reqsize() and add the returned amount of bytes to the end of the request memory allocation before posting the request to encrypt/decrypt APIs. So this breaks (with general protection fault and KASAN error, if enabled) because the request sent to decrypt is shorter than required causing the crypto API out-of-bounds errors. Also it seems unlikely the sk is even valid by the time it gets to the callback because of memset in crypto layer. Anyways, fix this by holding the sk in the skb->sk field when the callback is set up and because the skb is already passed through to the callback handler via void* we can access it in the handler. Then in the handler we need to be careful to NULL the pointer again before kfree_skb. I added comments on both the setup (in tls_do_decryption) and when we clear it from the crypto callback handler tls_decrypt_done(). After this selftests pass again and fixes KASAN errors/warnings. Fixes: 94524d8fc965 ("net/tls: Add support for async decryption of tls records") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Reviewed-by: Vakul Garg <Vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-14 20:01:46 +00:00
/* After using skb->sk to propagate sk through crypto async callback
* we need to NULL it again.
*/
skb->sk = NULL;
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
/* Free the destination pages if skb was not decrypted inplace */
if (sgout != sgin) {
/* Skip the first S/G entry as it points to AAD */
for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) {
if (!sg)
break;
put_page(sg_page(sg));
}
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
}
kfree(aead_req);
2020-05-22 20:10:31 +00:00
spin_lock_bh(&ctx->decrypt_compl_lock);
if (!atomic_dec_return(&ctx->decrypt_pending))
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
complete(&ctx->async_wait.completion);
2020-05-22 20:10:31 +00:00
spin_unlock_bh(&ctx->decrypt_compl_lock);
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
}
static int tls_do_decryption(struct sock *sk,
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
struct sk_buff *skb,
struct scatterlist *sgin,
struct scatterlist *sgout,
char *iv_recv,
size_t data_len,
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
struct aead_request *aead_req,
struct tls_decrypt_arg *darg)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_prot_info *prot = &tls_ctx->prot_info;
struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
int ret;
aead_request_set_tfm(aead_req, ctx->aead_recv);
aead_request_set_ad(aead_req, prot->aad_size);
aead_request_set_crypt(aead_req, sgin, sgout,
data_len + prot->tag_size,
(u8 *)iv_recv);
if (darg->async) {
tls: async support causes out-of-bounds access in crypto APIs When async support was added it needed to access the sk from the async callback to report errors up the stack. The patch tried to use space after the aead request struct by directly setting the reqsize field in aead_request. This is an internal field that should not be used outside the crypto APIs. It is used by the crypto code to define extra space for private structures used in the crypto context. Users of the API then use crypto_aead_reqsize() and add the returned amount of bytes to the end of the request memory allocation before posting the request to encrypt/decrypt APIs. So this breaks (with general protection fault and KASAN error, if enabled) because the request sent to decrypt is shorter than required causing the crypto API out-of-bounds errors. Also it seems unlikely the sk is even valid by the time it gets to the callback because of memset in crypto layer. Anyways, fix this by holding the sk in the skb->sk field when the callback is set up and because the skb is already passed through to the callback handler via void* we can access it in the handler. Then in the handler we need to be careful to NULL the pointer again before kfree_skb. I added comments on both the setup (in tls_do_decryption) and when we clear it from the crypto callback handler tls_decrypt_done(). After this selftests pass again and fixes KASAN errors/warnings. Fixes: 94524d8fc965 ("net/tls: Add support for async decryption of tls records") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Reviewed-by: Vakul Garg <Vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-14 20:01:46 +00:00
/* Using skb->sk to push sk through to crypto async callback
* handler. This allows propagating errors up to the socket
* if needed. It _must_ be cleared in the async handler
* before consume_skb is called. We _know_ skb->sk is NULL
tls: async support causes out-of-bounds access in crypto APIs When async support was added it needed to access the sk from the async callback to report errors up the stack. The patch tried to use space after the aead request struct by directly setting the reqsize field in aead_request. This is an internal field that should not be used outside the crypto APIs. It is used by the crypto code to define extra space for private structures used in the crypto context. Users of the API then use crypto_aead_reqsize() and add the returned amount of bytes to the end of the request memory allocation before posting the request to encrypt/decrypt APIs. So this breaks (with general protection fault and KASAN error, if enabled) because the request sent to decrypt is shorter than required causing the crypto API out-of-bounds errors. Also it seems unlikely the sk is even valid by the time it gets to the callback because of memset in crypto layer. Anyways, fix this by holding the sk in the skb->sk field when the callback is set up and because the skb is already passed through to the callback handler via void* we can access it in the handler. Then in the handler we need to be careful to NULL the pointer again before kfree_skb. I added comments on both the setup (in tls_do_decryption) and when we clear it from the crypto callback handler tls_decrypt_done(). After this selftests pass again and fixes KASAN errors/warnings. Fixes: 94524d8fc965 ("net/tls: Add support for async decryption of tls records") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Reviewed-by: Vakul Garg <Vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-14 20:01:46 +00:00
* because it is a clone from strparser.
*/
skb->sk = sk;
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
aead_request_set_callback(aead_req,
CRYPTO_TFM_REQ_MAY_BACKLOG,
tls_decrypt_done, skb);
atomic_inc(&ctx->decrypt_pending);
} else {
aead_request_set_callback(aead_req,
CRYPTO_TFM_REQ_MAY_BACKLOG,
crypto_req_done, &ctx->async_wait);
}
ret = crypto_aead_decrypt(aead_req);
if (ret == -EINPROGRESS) {
if (darg->async)
return 0;
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
ret = crypto_wait_req(ret, &ctx->async_wait);
}
darg->async = false;
return ret;
}
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
static void tls_trim_both_msgs(struct sock *sk, int target_size)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_prot_info *prot = &tls_ctx->prot_info;
struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
struct tls_rec *rec = ctx->open_rec;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
sk_msg_trim(sk, &rec->msg_plaintext, target_size);
if (target_size > 0)
target_size += prot->overhead_size;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
sk_msg_trim(sk, &rec->msg_encrypted, target_size);
}
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
static int tls_alloc_encrypted_msg(struct sock *sk, int len)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
struct tls_rec *rec = ctx->open_rec;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
struct sk_msg *msg_en = &rec->msg_encrypted;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
return sk_msg_alloc(sk, msg_en, len, 0);
}
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
static int tls_clone_plaintext_msg(struct sock *sk, int required)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_prot_info *prot = &tls_ctx->prot_info;
struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
struct tls_rec *rec = ctx->open_rec;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
struct sk_msg *msg_pl = &rec->msg_plaintext;
struct sk_msg *msg_en = &rec->msg_encrypted;
int skip, len;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
/* We add page references worth len bytes from encrypted sg
* at the end of plaintext sg. It is guaranteed that msg_en
* has enough required room (ensured by caller).
*/
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
len = required - msg_pl->sg.size;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
/* Skip initial bytes in msg_en's data to be able to use
* same offset of both plain and encrypted data.
*/
skip = prot->prepend_size + msg_pl->sg.size;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
}
static struct tls_rec *tls_get_rec(struct sock *sk)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_prot_info *prot = &tls_ctx->prot_info;
struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
struct sk_msg *msg_pl, *msg_en;
struct tls_rec *rec;
int mem_size;
mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
rec = kzalloc(mem_size, sk->sk_allocation);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (!rec)
return NULL;
msg_pl = &rec->msg_plaintext;
msg_en = &rec->msg_encrypted;
sk_msg_init(msg_pl);
sk_msg_init(msg_en);
sg_init_table(rec->sg_aead_in, 2);
sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size);
sg_unmark_end(&rec->sg_aead_in[1]);
sg_init_table(rec->sg_aead_out, 2);
sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size);
sg_unmark_end(&rec->sg_aead_out[1]);
return rec;
}
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
{
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
sk_msg_free(sk, &rec->msg_encrypted);
sk_msg_free(sk, &rec->msg_plaintext);
kfree(rec);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
}
static void tls_free_open_rec(struct sock *sk)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
struct tls_rec *rec = ctx->open_rec;
if (rec) {
tls_free_rec(sk, rec);
ctx->open_rec = NULL;
}
}
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
int tls_tx_records(struct sock *sk, int flags)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
struct tls_rec *rec, *tmp;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
struct sk_msg *msg_en;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
int tx_flags, rc = 0;
if (tls_is_partially_sent_record(tls_ctx)) {
net/tls: Fixed race condition in async encryption On processors with multi-engine crypto accelerators, it is possible that multiple records get encrypted in parallel and their encryption completion is notified to different cpus in multicore processor. This leads to the situation where tls_encrypt_done() starts executing in parallel on different cores. In current implementation, encrypted records are queued to tx_ready_list in tls_encrypt_done(). This requires addition to linked list 'tx_ready_list' to be protected. As tls_decrypt_done() could be executing in irq content, it is not possible to protect linked list addition operation using a lock. To fix the problem, we remove linked list addition operation from the irq context. We do tx_ready_list addition/removal operation from application context only and get rid of possible multiple access to the linked list. Before starting encryption on the record, we add it to the tail of tx_ready_list. To prevent tls_tx_records() from transmitting it, we mark the record with a new flag 'tx_ready' in 'struct tls_rec'. When record encryption gets completed, tls_encrypt_done() has to only update the 'tx_ready' flag to true & linked list add operation is not required. The changed logic brings some other side benefits. Since the records are always submitted in tls sequence number order for encryption, the tx_ready_list always remains sorted and addition of new records to it does not have to traverse the linked list. Lastly, we renamed tx_ready_list in 'struct tls_sw_context_tx' to 'tx_list'. This is because now, the some of the records at the tail are not ready to transmit. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption") Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-24 10:05:56 +00:00
rec = list_first_entry(&ctx->tx_list,
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
struct tls_rec, list);
if (flags == -1)
tx_flags = rec->tx_flags;
else
tx_flags = flags;
rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
if (rc)
goto tx_err;
/* Full record has been transmitted.
net/tls: Fixed race condition in async encryption On processors with multi-engine crypto accelerators, it is possible that multiple records get encrypted in parallel and their encryption completion is notified to different cpus in multicore processor. This leads to the situation where tls_encrypt_done() starts executing in parallel on different cores. In current implementation, encrypted records are queued to tx_ready_list in tls_encrypt_done(). This requires addition to linked list 'tx_ready_list' to be protected. As tls_decrypt_done() could be executing in irq content, it is not possible to protect linked list addition operation using a lock. To fix the problem, we remove linked list addition operation from the irq context. We do tx_ready_list addition/removal operation from application context only and get rid of possible multiple access to the linked list. Before starting encryption on the record, we add it to the tail of tx_ready_list. To prevent tls_tx_records() from transmitting it, we mark the record with a new flag 'tx_ready' in 'struct tls_rec'. When record encryption gets completed, tls_encrypt_done() has to only update the 'tx_ready' flag to true & linked list add operation is not required. The changed logic brings some other side benefits. Since the records are always submitted in tls sequence number order for encryption, the tx_ready_list always remains sorted and addition of new records to it does not have to traverse the linked list. Lastly, we renamed tx_ready_list in 'struct tls_sw_context_tx' to 'tx_list'. This is because now, the some of the records at the tail are not ready to transmit. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption") Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-24 10:05:56 +00:00
* Remove the head of tx_list
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
*/
list_del(&rec->list);
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
sk_msg_free(sk, &rec->msg_plaintext);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
kfree(rec);
}
net/tls: Fixed race condition in async encryption On processors with multi-engine crypto accelerators, it is possible that multiple records get encrypted in parallel and their encryption completion is notified to different cpus in multicore processor. This leads to the situation where tls_encrypt_done() starts executing in parallel on different cores. In current implementation, encrypted records are queued to tx_ready_list in tls_encrypt_done(). This requires addition to linked list 'tx_ready_list' to be protected. As tls_decrypt_done() could be executing in irq content, it is not possible to protect linked list addition operation using a lock. To fix the problem, we remove linked list addition operation from the irq context. We do tx_ready_list addition/removal operation from application context only and get rid of possible multiple access to the linked list. Before starting encryption on the record, we add it to the tail of tx_ready_list. To prevent tls_tx_records() from transmitting it, we mark the record with a new flag 'tx_ready' in 'struct tls_rec'. When record encryption gets completed, tls_encrypt_done() has to only update the 'tx_ready' flag to true & linked list add operation is not required. The changed logic brings some other side benefits. Since the records are always submitted in tls sequence number order for encryption, the tx_ready_list always remains sorted and addition of new records to it does not have to traverse the linked list. Lastly, we renamed tx_ready_list in 'struct tls_sw_context_tx' to 'tx_list'. This is because now, the some of the records at the tail are not ready to transmit. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption") Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-24 10:05:56 +00:00
/* Tx all ready records */
list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
if (READ_ONCE(rec->tx_ready)) {
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (flags == -1)
tx_flags = rec->tx_flags;
else
tx_flags = flags;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
msg_en = &rec->msg_encrypted;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
rc = tls_push_sg(sk, tls_ctx,
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
&msg_en->sg.data[msg_en->sg.curr],
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
0, tx_flags);
if (rc)
goto tx_err;
list_del(&rec->list);
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
sk_msg_free(sk, &rec->msg_plaintext);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
kfree(rec);
} else {
break;
}
}
tx_err:
if (rc < 0 && rc != -EAGAIN)
tls_err_abort(sk, -EBADMSG);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
return rc;
}
static void tls_encrypt_done(struct crypto_async_request *req, int err)
{
struct aead_request *aead_req = (struct aead_request *)req;
struct sock *sk = req->data;
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_prot_info *prot = &tls_ctx->prot_info;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
struct scatterlist *sge;
struct sk_msg *msg_en;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
struct tls_rec *rec;
bool ready = false;
int pending;
rec = container_of(aead_req, struct tls_rec, aead_req);
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
msg_en = &rec->msg_encrypted;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
sge = sk_msg_elem(msg_en, msg_en->sg.curr);
sge->offset -= prot->prepend_size;
sge->length += prot->prepend_size;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
/* Check if error is previously set on socket */
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (err || sk->sk_err) {
rec = NULL;
/* If err is already set on socket, return the same code */
if (sk->sk_err) {
ctx->async_wait.err = -sk->sk_err;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
} else {
ctx->async_wait.err = err;
tls_err_abort(sk, err);
}
}
net/tls: Fixed race condition in async encryption On processors with multi-engine crypto accelerators, it is possible that multiple records get encrypted in parallel and their encryption completion is notified to different cpus in multicore processor. This leads to the situation where tls_encrypt_done() starts executing in parallel on different cores. In current implementation, encrypted records are queued to tx_ready_list in tls_encrypt_done(). This requires addition to linked list 'tx_ready_list' to be protected. As tls_decrypt_done() could be executing in irq content, it is not possible to protect linked list addition operation using a lock. To fix the problem, we remove linked list addition operation from the irq context. We do tx_ready_list addition/removal operation from application context only and get rid of possible multiple access to the linked list. Before starting encryption on the record, we add it to the tail of tx_ready_list. To prevent tls_tx_records() from transmitting it, we mark the record with a new flag 'tx_ready' in 'struct tls_rec'. When record encryption gets completed, tls_encrypt_done() has to only update the 'tx_ready' flag to true & linked list add operation is not required. The changed logic brings some other side benefits. Since the records are always submitted in tls sequence number order for encryption, the tx_ready_list always remains sorted and addition of new records to it does not have to traverse the linked list. Lastly, we renamed tx_ready_list in 'struct tls_sw_context_tx' to 'tx_list'. This is because now, the some of the records at the tail are not ready to transmit. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption") Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-24 10:05:56 +00:00
if (rec) {
struct tls_rec *first_rec;
/* Mark the record as ready for transmission */
smp_store_mb(rec->tx_ready, true);
/* If received record is at head of tx_list, schedule tx */
first_rec = list_first_entry(&ctx->tx_list,
struct tls_rec, list);
if (rec == first_rec)
ready = true;
}
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
2020-05-22 20:10:31 +00:00
spin_lock_bh(&ctx->encrypt_compl_lock);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
pending = atomic_dec_return(&ctx->encrypt_pending);
2020-05-22 20:10:31 +00:00
if (!pending && ctx->async_notify)
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
complete(&ctx->async_wait.completion);
2020-05-22 20:10:31 +00:00
spin_unlock_bh(&ctx->encrypt_compl_lock);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (!ready)
return;
/* Schedule the transmission */
if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
schedule_delayed_work(&ctx->tx_work.work, 1);
}
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
static int tls_do_encryption(struct sock *sk,
struct tls_context *tls_ctx,
tls: fix use-after-free in tls_push_record syzkaller managed to trigger a use-after-free in tls like the following: BUG: KASAN: use-after-free in tls_push_record.constprop.15+0x6a2/0x810 [tls] Write of size 1 at addr ffff88037aa08000 by task a.out/2317 CPU: 3 PID: 2317 Comm: a.out Not tainted 4.17.0+ #144 Hardware name: LENOVO 20FBCTO1WW/20FBCTO1WW, BIOS N1FET47W (1.21 ) 11/28/2016 Call Trace: dump_stack+0x71/0xab print_address_description+0x6a/0x280 kasan_report+0x258/0x380 ? tls_push_record.constprop.15+0x6a2/0x810 [tls] tls_push_record.constprop.15+0x6a2/0x810 [tls] tls_sw_push_pending_record+0x2e/0x40 [tls] tls_sk_proto_close+0x3fe/0x710 [tls] ? tcp_check_oom+0x4c0/0x4c0 ? tls_write_space+0x260/0x260 [tls] ? kmem_cache_free+0x88/0x1f0 inet_release+0xd6/0x1b0 __sock_release+0xc0/0x240 sock_close+0x11/0x20 __fput+0x22d/0x660 task_work_run+0x114/0x1a0 do_exit+0x71a/0x2780 ? mm_update_next_owner+0x650/0x650 ? handle_mm_fault+0x2f5/0x5f0 ? __do_page_fault+0x44f/0xa50 ? mm_fault_error+0x2d0/0x2d0 do_group_exit+0xde/0x300 __x64_sys_exit_group+0x3a/0x50 do_syscall_64+0x9a/0x300 ? page_fault+0x8/0x30 entry_SYSCALL_64_after_hwframe+0x44/0xa9 This happened through fault injection where aead_req allocation in tls_do_encryption() eventually failed and we returned -ENOMEM from the function. Turns out that the use-after-free is triggered from tls_sw_sendmsg() in the second tls_push_record(). The error then triggers a jump to waiting for memory in sk_stream_wait_memory() resp. returning immediately in case of MSG_DONTWAIT. What follows is the trim_both_sgl(sk, orig_size), which drops elements from the sg list added via tls_sw_sendmsg(). Now the use-after-free gets triggered when the socket is being closed, where tls_sk_proto_close() callback is invoked. The tls_complete_pending_work() will figure that there's a pending closed tls record to be flushed and thus calls into the tls_push_pending_closed_record() from there. ctx->push_pending_record() is called from the latter, which is the tls_sw_push_pending_record() from sw path. This again calls into tls_push_record(). And here the tls_fill_prepend() will panic since the buffer address has been freed earlier via trim_both_sgl(). One way to fix it is to move the aead request allocation out of tls_do_encryption() early into tls_push_record(). This means we don't prep the tls header and advance state to the TLS_PENDING_CLOSED_RECORD before allocation which could potentially fail happened. That fixes the issue on my side. Fixes: 3c4d7559159b ("tls: kernel TLS support") Reported-by: syzbot+5c74af81c547738e1684@syzkaller.appspotmail.com Reported-by: syzbot+709f2810a6a05f11d4d3@syzkaller.appspotmail.com Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-06-15 01:07:45 +00:00
struct tls_sw_context_tx *ctx,
struct aead_request *aead_req,
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
size_t data_len, u32 start)
{
struct tls_prot_info *prot = &tls_ctx->prot_info;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
struct tls_rec *rec = ctx->open_rec;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
struct sk_msg *msg_en = &rec->msg_encrypted;
struct scatterlist *sge = sk_msg_elem(msg_en, start);
int rc, iv_offset = 0;
/* For CCM based ciphers, first byte of IV is a constant */
switch (prot->cipher_type) {
case TLS_CIPHER_AES_CCM_128:
rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
iv_offset = 1;
break;
case TLS_CIPHER_SM4_CCM:
rec->iv_data[0] = TLS_SM4_CCM_IV_B0_BYTE;
iv_offset = 1;
break;
}
memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
prot->iv_size + prot->salt_size);
tls_xor_iv_with_seq(prot, rec->iv_data + iv_offset,
tls_ctx->tx.rec_seq);
sge->offset += prot->prepend_size;
sge->length -= prot->prepend_size;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
msg_en->sg.curr = start;
aead_request_set_tfm(aead_req, ctx->aead_send);
aead_request_set_ad(aead_req, prot->aad_size);
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
aead_request_set_crypt(aead_req, rec->sg_aead_in,
rec->sg_aead_out,
data_len, rec->iv_data);
aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
tls_encrypt_done, sk);
net/tls: Fixed race condition in async encryption On processors with multi-engine crypto accelerators, it is possible that multiple records get encrypted in parallel and their encryption completion is notified to different cpus in multicore processor. This leads to the situation where tls_encrypt_done() starts executing in parallel on different cores. In current implementation, encrypted records are queued to tx_ready_list in tls_encrypt_done(). This requires addition to linked list 'tx_ready_list' to be protected. As tls_decrypt_done() could be executing in irq content, it is not possible to protect linked list addition operation using a lock. To fix the problem, we remove linked list addition operation from the irq context. We do tx_ready_list addition/removal operation from application context only and get rid of possible multiple access to the linked list. Before starting encryption on the record, we add it to the tail of tx_ready_list. To prevent tls_tx_records() from transmitting it, we mark the record with a new flag 'tx_ready' in 'struct tls_rec'. When record encryption gets completed, tls_encrypt_done() has to only update the 'tx_ready' flag to true & linked list add operation is not required. The changed logic brings some other side benefits. Since the records are always submitted in tls sequence number order for encryption, the tx_ready_list always remains sorted and addition of new records to it does not have to traverse the linked list. Lastly, we renamed tx_ready_list in 'struct tls_sw_context_tx' to 'tx_list'. This is because now, the some of the records at the tail are not ready to transmit. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption") Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-24 10:05:56 +00:00
/* Add the record in tx_list */
list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
atomic_inc(&ctx->encrypt_pending);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
rc = crypto_aead_encrypt(aead_req);
if (!rc || rc != -EINPROGRESS) {
atomic_dec(&ctx->encrypt_pending);
sge->offset -= prot->prepend_size;
sge->length += prot->prepend_size;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
}
net/tls: Fixed race condition in async encryption On processors with multi-engine crypto accelerators, it is possible that multiple records get encrypted in parallel and their encryption completion is notified to different cpus in multicore processor. This leads to the situation where tls_encrypt_done() starts executing in parallel on different cores. In current implementation, encrypted records are queued to tx_ready_list in tls_encrypt_done(). This requires addition to linked list 'tx_ready_list' to be protected. As tls_decrypt_done() could be executing in irq content, it is not possible to protect linked list addition operation using a lock. To fix the problem, we remove linked list addition operation from the irq context. We do tx_ready_list addition/removal operation from application context only and get rid of possible multiple access to the linked list. Before starting encryption on the record, we add it to the tail of tx_ready_list. To prevent tls_tx_records() from transmitting it, we mark the record with a new flag 'tx_ready' in 'struct tls_rec'. When record encryption gets completed, tls_encrypt_done() has to only update the 'tx_ready' flag to true & linked list add operation is not required. The changed logic brings some other side benefits. Since the records are always submitted in tls sequence number order for encryption, the tx_ready_list always remains sorted and addition of new records to it does not have to traverse the linked list. Lastly, we renamed tx_ready_list in 'struct tls_sw_context_tx' to 'tx_list'. This is because now, the some of the records at the tail are not ready to transmit. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption") Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-24 10:05:56 +00:00
if (!rc) {
WRITE_ONCE(rec->tx_ready, true);
} else if (rc != -EINPROGRESS) {
list_del(&rec->list);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
return rc;
net/tls: Fixed race condition in async encryption On processors with multi-engine crypto accelerators, it is possible that multiple records get encrypted in parallel and their encryption completion is notified to different cpus in multicore processor. This leads to the situation where tls_encrypt_done() starts executing in parallel on different cores. In current implementation, encrypted records are queued to tx_ready_list in tls_encrypt_done(). This requires addition to linked list 'tx_ready_list' to be protected. As tls_decrypt_done() could be executing in irq content, it is not possible to protect linked list addition operation using a lock. To fix the problem, we remove linked list addition operation from the irq context. We do tx_ready_list addition/removal operation from application context only and get rid of possible multiple access to the linked list. Before starting encryption on the record, we add it to the tail of tx_ready_list. To prevent tls_tx_records() from transmitting it, we mark the record with a new flag 'tx_ready' in 'struct tls_rec'. When record encryption gets completed, tls_encrypt_done() has to only update the 'tx_ready' flag to true & linked list add operation is not required. The changed logic brings some other side benefits. Since the records are always submitted in tls sequence number order for encryption, the tx_ready_list always remains sorted and addition of new records to it does not have to traverse the linked list. Lastly, we renamed tx_ready_list in 'struct tls_sw_context_tx' to 'tx_list'. This is because now, the some of the records at the tail are not ready to transmit. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption") Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-24 10:05:56 +00:00
}
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
/* Unhook the record from context if encryption is not failure */
ctx->open_rec = NULL;
tls_advance_record_sn(sk, prot, &tls_ctx->tx);
return rc;
}
static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
struct tls_rec **to, struct sk_msg *msg_opl,
struct sk_msg *msg_oen, u32 split_point,
u32 tx_overhead_size, u32 *orig_end)
{
u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
struct scatterlist *sge, *osge, *nsge;
u32 orig_size = msg_opl->sg.size;
struct scatterlist tmp = { };
struct sk_msg *msg_npl;
struct tls_rec *new;
int ret;
new = tls_get_rec(sk);
if (!new)
return -ENOMEM;
ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size +
tx_overhead_size, 0);
if (ret < 0) {
tls_free_rec(sk, new);
return ret;
}
*orig_end = msg_opl->sg.end;
i = msg_opl->sg.start;
sge = sk_msg_elem(msg_opl, i);
while (apply && sge->length) {
if (sge->length > apply) {
u32 len = sge->length - apply;
get_page(sg_page(sge));
sg_set_page(&tmp, sg_page(sge), len,
sge->offset + apply);
sge->length = apply;
bytes += apply;
apply = 0;
} else {
apply -= sge->length;
bytes += sge->length;
}
sk_msg_iter_var_next(i);
if (i == msg_opl->sg.end)
break;
sge = sk_msg_elem(msg_opl, i);
}
msg_opl->sg.end = i;
msg_opl->sg.curr = i;
msg_opl->sg.copybreak = 0;
msg_opl->apply_bytes = 0;
msg_opl->sg.size = bytes;
msg_npl = &new->msg_plaintext;
msg_npl->apply_bytes = apply;
msg_npl->sg.size = orig_size - bytes;
j = msg_npl->sg.start;
nsge = sk_msg_elem(msg_npl, j);
if (tmp.length) {
memcpy(nsge, &tmp, sizeof(*nsge));
sk_msg_iter_var_next(j);
nsge = sk_msg_elem(msg_npl, j);
}
osge = sk_msg_elem(msg_opl, i);
while (osge->length) {
memcpy(nsge, osge, sizeof(*nsge));
sg_unmark_end(nsge);
sk_msg_iter_var_next(i);
sk_msg_iter_var_next(j);
if (i == *orig_end)
break;
osge = sk_msg_elem(msg_opl, i);
nsge = sk_msg_elem(msg_npl, j);
}
msg_npl->sg.end = j;
msg_npl->sg.curr = j;
msg_npl->sg.copybreak = 0;
*to = new;
return 0;
}
static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
struct tls_rec *from, u32 orig_end)
{
struct sk_msg *msg_npl = &from->msg_plaintext;
struct sk_msg *msg_opl = &to->msg_plaintext;
struct scatterlist *osge, *nsge;
u32 i, j;
i = msg_opl->sg.end;
sk_msg_iter_var_prev(i);
j = msg_npl->sg.start;
osge = sk_msg_elem(msg_opl, i);
nsge = sk_msg_elem(msg_npl, j);
if (sg_page(osge) == sg_page(nsge) &&
osge->offset + osge->length == nsge->offset) {
osge->length += nsge->length;
put_page(sg_page(nsge));
}
msg_opl->sg.end = orig_end;
msg_opl->sg.curr = orig_end;
msg_opl->sg.copybreak = 0;
msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
msg_opl->sg.size += msg_npl->sg.size;
sk_msg_free(sk, &to->msg_encrypted);
sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted);
kfree(from);
}
static int tls_push_record(struct sock *sk, int flags,
unsigned char record_type)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_prot_info *prot = &tls_ctx->prot_info;
struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
treewide: Remove uninitialized_var() usage Using uninitialized_var() is dangerous as it papers over real bugs[1] (or can in the future), and suppresses unrelated compiler warnings (e.g. "unused variable"). If the compiler thinks it is uninitialized, either simply initialize the variable or make compiler changes. In preparation for removing[2] the[3] macro[4], remove all remaining needless uses with the following script: git grep '\buninitialized_var\b' | cut -d: -f1 | sort -u | \ xargs perl -pi -e \ 's/\buninitialized_var\(([^\)]+)\)/\1/g; s:\s*/\* (GCC be quiet|to make compiler happy) \*/$::g;' drivers/video/fbdev/riva/riva_hw.c was manually tweaked to avoid pathological white-space. No outstanding warnings were found building allmodconfig with GCC 9.3.0 for x86_64, i386, arm64, arm, powerpc, powerpc64le, s390x, mips, sparc64, alpha, and m68k. [1] https://lore.kernel.org/lkml/20200603174714.192027-1-glider@google.com/ [2] https://lore.kernel.org/lkml/CA+55aFw+Vbj0i=1TGqCR5vQkCzWJ0QxK6CernOU6eedsudAixw@mail.gmail.com/ [3] https://lore.kernel.org/lkml/CA+55aFwgbgqhbp1fkxvRKEpzyR5J8n1vKT1VZdz9knmPuXhOeg@mail.gmail.com/ [4] https://lore.kernel.org/lkml/CA+55aFz2500WfbKXAx8s67wrm9=yVJu65TpLgN_ybYNv0VEOKA@mail.gmail.com/ Reviewed-by: Leon Romanovsky <leonro@mellanox.com> # drivers/infiniband and mlx4/mlx5 Acked-by: Jason Gunthorpe <jgg@mellanox.com> # IB Acked-by: Kalle Valo <kvalo@codeaurora.org> # wireless drivers Reviewed-by: Chao Yu <yuchao0@huawei.com> # erofs Signed-off-by: Kees Cook <keescook@chromium.org>
2020-06-03 20:09:38 +00:00
u32 i, split_point, orig_end;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
struct sk_msg *msg_pl, *msg_en;
tls: fix use-after-free in tls_push_record syzkaller managed to trigger a use-after-free in tls like the following: BUG: KASAN: use-after-free in tls_push_record.constprop.15+0x6a2/0x810 [tls] Write of size 1 at addr ffff88037aa08000 by task a.out/2317 CPU: 3 PID: 2317 Comm: a.out Not tainted 4.17.0+ #144 Hardware name: LENOVO 20FBCTO1WW/20FBCTO1WW, BIOS N1FET47W (1.21 ) 11/28/2016 Call Trace: dump_stack+0x71/0xab print_address_description+0x6a/0x280 kasan_report+0x258/0x380 ? tls_push_record.constprop.15+0x6a2/0x810 [tls] tls_push_record.constprop.15+0x6a2/0x810 [tls] tls_sw_push_pending_record+0x2e/0x40 [tls] tls_sk_proto_close+0x3fe/0x710 [tls] ? tcp_check_oom+0x4c0/0x4c0 ? tls_write_space+0x260/0x260 [tls] ? kmem_cache_free+0x88/0x1f0 inet_release+0xd6/0x1b0 __sock_release+0xc0/0x240 sock_close+0x11/0x20 __fput+0x22d/0x660 task_work_run+0x114/0x1a0 do_exit+0x71a/0x2780 ? mm_update_next_owner+0x650/0x650 ? handle_mm_fault+0x2f5/0x5f0 ? __do_page_fault+0x44f/0xa50 ? mm_fault_error+0x2d0/0x2d0 do_group_exit+0xde/0x300 __x64_sys_exit_group+0x3a/0x50 do_syscall_64+0x9a/0x300 ? page_fault+0x8/0x30 entry_SYSCALL_64_after_hwframe+0x44/0xa9 This happened through fault injection where aead_req allocation in tls_do_encryption() eventually failed and we returned -ENOMEM from the function. Turns out that the use-after-free is triggered from tls_sw_sendmsg() in the second tls_push_record(). The error then triggers a jump to waiting for memory in sk_stream_wait_memory() resp. returning immediately in case of MSG_DONTWAIT. What follows is the trim_both_sgl(sk, orig_size), which drops elements from the sg list added via tls_sw_sendmsg(). Now the use-after-free gets triggered when the socket is being closed, where tls_sk_proto_close() callback is invoked. The tls_complete_pending_work() will figure that there's a pending closed tls record to be flushed and thus calls into the tls_push_pending_closed_record() from there. ctx->push_pending_record() is called from the latter, which is the tls_sw_push_pending_record() from sw path. This again calls into tls_push_record(). And here the tls_fill_prepend() will panic since the buffer address has been freed earlier via trim_both_sgl(). One way to fix it is to move the aead request allocation out of tls_do_encryption() early into tls_push_record(). This means we don't prep the tls header and advance state to the TLS_PENDING_CLOSED_RECORD before allocation which could potentially fail happened. That fixes the issue on my side. Fixes: 3c4d7559159b ("tls: kernel TLS support") Reported-by: syzbot+5c74af81c547738e1684@syzkaller.appspotmail.com Reported-by: syzbot+709f2810a6a05f11d4d3@syzkaller.appspotmail.com Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-06-15 01:07:45 +00:00
struct aead_request *req;
bool split;
int rc;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (!rec)
return 0;
tls: fix use-after-free in tls_push_record syzkaller managed to trigger a use-after-free in tls like the following: BUG: KASAN: use-after-free in tls_push_record.constprop.15+0x6a2/0x810 [tls] Write of size 1 at addr ffff88037aa08000 by task a.out/2317 CPU: 3 PID: 2317 Comm: a.out Not tainted 4.17.0+ #144 Hardware name: LENOVO 20FBCTO1WW/20FBCTO1WW, BIOS N1FET47W (1.21 ) 11/28/2016 Call Trace: dump_stack+0x71/0xab print_address_description+0x6a/0x280 kasan_report+0x258/0x380 ? tls_push_record.constprop.15+0x6a2/0x810 [tls] tls_push_record.constprop.15+0x6a2/0x810 [tls] tls_sw_push_pending_record+0x2e/0x40 [tls] tls_sk_proto_close+0x3fe/0x710 [tls] ? tcp_check_oom+0x4c0/0x4c0 ? tls_write_space+0x260/0x260 [tls] ? kmem_cache_free+0x88/0x1f0 inet_release+0xd6/0x1b0 __sock_release+0xc0/0x240 sock_close+0x11/0x20 __fput+0x22d/0x660 task_work_run+0x114/0x1a0 do_exit+0x71a/0x2780 ? mm_update_next_owner+0x650/0x650 ? handle_mm_fault+0x2f5/0x5f0 ? __do_page_fault+0x44f/0xa50 ? mm_fault_error+0x2d0/0x2d0 do_group_exit+0xde/0x300 __x64_sys_exit_group+0x3a/0x50 do_syscall_64+0x9a/0x300 ? page_fault+0x8/0x30 entry_SYSCALL_64_after_hwframe+0x44/0xa9 This happened through fault injection where aead_req allocation in tls_do_encryption() eventually failed and we returned -ENOMEM from the function. Turns out that the use-after-free is triggered from tls_sw_sendmsg() in the second tls_push_record(). The error then triggers a jump to waiting for memory in sk_stream_wait_memory() resp. returning immediately in case of MSG_DONTWAIT. What follows is the trim_both_sgl(sk, orig_size), which drops elements from the sg list added via tls_sw_sendmsg(). Now the use-after-free gets triggered when the socket is being closed, where tls_sk_proto_close() callback is invoked. The tls_complete_pending_work() will figure that there's a pending closed tls record to be flushed and thus calls into the tls_push_pending_closed_record() from there. ctx->push_pending_record() is called from the latter, which is the tls_sw_push_pending_record() from sw path. This again calls into tls_push_record(). And here the tls_fill_prepend() will panic since the buffer address has been freed earlier via trim_both_sgl(). One way to fix it is to move the aead request allocation out of tls_do_encryption() early into tls_push_record(). This means we don't prep the tls header and advance state to the TLS_PENDING_CLOSED_RECORD before allocation which could potentially fail happened. That fixes the issue on my side. Fixes: 3c4d7559159b ("tls: kernel TLS support") Reported-by: syzbot+5c74af81c547738e1684@syzkaller.appspotmail.com Reported-by: syzbot+709f2810a6a05f11d4d3@syzkaller.appspotmail.com Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-06-15 01:07:45 +00:00
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
msg_pl = &rec->msg_plaintext;
msg_en = &rec->msg_encrypted;
split_point = msg_pl->apply_bytes;
split = split_point && split_point < msg_pl->sg.size;
bpf: Sockmap/tls, tls_sw can create a plaintext buf > encrypt buf It is possible to build a plaintext buffer using push helper that is larger than the allocated encrypt buffer. When this record is pushed to crypto layers this can result in a NULL pointer dereference because the crypto API expects the encrypt buffer is large enough to fit the plaintext buffer. Kernel splat below. To resolve catch the cases this can happen and split the buffer into two records to send individually. Unfortunately, there is still one case to handle where the split creates a zero sized buffer. In this case we merge the buffers and unmark the split. This happens when apply is zero and user pushed data beyond encrypt buffer. This fixes the original case as well because the split allocated an encrypt buffer larger than the plaintext buffer and the merge simply moves the pointers around so we now have a reference to the new (larger) encrypt buffer. Perhaps its not ideal but it seems the best solution for a fixes branch and avoids handling these two cases, (a) apply that needs split and (b) non apply case. The are edge cases anyways so optimizing them seems not necessary unless someone wants later in next branches. [ 306.719107] BUG: kernel NULL pointer dereference, address: 0000000000000008 [...] [ 306.747260] RIP: 0010:scatterwalk_copychunks+0x12f/0x1b0 [...] [ 306.770350] Call Trace: [ 306.770956] scatterwalk_map_and_copy+0x6c/0x80 [ 306.772026] gcm_enc_copy_hash+0x4b/0x50 [ 306.772925] gcm_hash_crypt_remain_continue+0xef/0x110 [ 306.774138] gcm_hash_crypt_continue+0xa1/0xb0 [ 306.775103] ? gcm_hash_crypt_continue+0xa1/0xb0 [ 306.776103] gcm_hash_assoc_remain_continue+0x94/0xa0 [ 306.777170] gcm_hash_assoc_continue+0x9d/0xb0 [ 306.778239] gcm_hash_init_continue+0x8f/0xa0 [ 306.779121] gcm_hash+0x73/0x80 [ 306.779762] gcm_encrypt_continue+0x6d/0x80 [ 306.780582] crypto_gcm_encrypt+0xcb/0xe0 [ 306.781474] crypto_aead_encrypt+0x1f/0x30 [ 306.782353] tls_push_record+0x3b9/0xb20 [tls] [ 306.783314] ? sk_psock_msg_verdict+0x199/0x300 [ 306.784287] bpf_exec_tx_verdict+0x3f2/0x680 [tls] [ 306.785357] tls_sw_sendmsg+0x4a3/0x6a0 [tls] test_sockmap test signature to trigger bug, [TEST]: (1, 1, 1, sendmsg, pass,redir,start 1,end 2,pop (1,2),ktls,): Fixes: d3b18ad31f93d ("tls: add bpf support to sk_msg handling") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com> Cc: stable@vger.kernel.org Link: https://lore.kernel.org/bpf/20200111061206.8028-7-john.fastabend@gmail.com
2020-01-11 06:12:04 +00:00
if (unlikely((!split &&
msg_pl->sg.size +
prot->overhead_size > msg_en->sg.size) ||
(split &&
split_point +
prot->overhead_size > msg_en->sg.size))) {
split = true;
split_point = msg_en->sg.size;
}
if (split) {
rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en,
split_point, prot->overhead_size,
&orig_end);
if (rc < 0)
return rc;
bpf: Sockmap/tls, tls_sw can create a plaintext buf > encrypt buf It is possible to build a plaintext buffer using push helper that is larger than the allocated encrypt buffer. When this record is pushed to crypto layers this can result in a NULL pointer dereference because the crypto API expects the encrypt buffer is large enough to fit the plaintext buffer. Kernel splat below. To resolve catch the cases this can happen and split the buffer into two records to send individually. Unfortunately, there is still one case to handle where the split creates a zero sized buffer. In this case we merge the buffers and unmark the split. This happens when apply is zero and user pushed data beyond encrypt buffer. This fixes the original case as well because the split allocated an encrypt buffer larger than the plaintext buffer and the merge simply moves the pointers around so we now have a reference to the new (larger) encrypt buffer. Perhaps its not ideal but it seems the best solution for a fixes branch and avoids handling these two cases, (a) apply that needs split and (b) non apply case. The are edge cases anyways so optimizing them seems not necessary unless someone wants later in next branches. [ 306.719107] BUG: kernel NULL pointer dereference, address: 0000000000000008 [...] [ 306.747260] RIP: 0010:scatterwalk_copychunks+0x12f/0x1b0 [...] [ 306.770350] Call Trace: [ 306.770956] scatterwalk_map_and_copy+0x6c/0x80 [ 306.772026] gcm_enc_copy_hash+0x4b/0x50 [ 306.772925] gcm_hash_crypt_remain_continue+0xef/0x110 [ 306.774138] gcm_hash_crypt_continue+0xa1/0xb0 [ 306.775103] ? gcm_hash_crypt_continue+0xa1/0xb0 [ 306.776103] gcm_hash_assoc_remain_continue+0x94/0xa0 [ 306.777170] gcm_hash_assoc_continue+0x9d/0xb0 [ 306.778239] gcm_hash_init_continue+0x8f/0xa0 [ 306.779121] gcm_hash+0x73/0x80 [ 306.779762] gcm_encrypt_continue+0x6d/0x80 [ 306.780582] crypto_gcm_encrypt+0xcb/0xe0 [ 306.781474] crypto_aead_encrypt+0x1f/0x30 [ 306.782353] tls_push_record+0x3b9/0xb20 [tls] [ 306.783314] ? sk_psock_msg_verdict+0x199/0x300 [ 306.784287] bpf_exec_tx_verdict+0x3f2/0x680 [tls] [ 306.785357] tls_sw_sendmsg+0x4a3/0x6a0 [tls] test_sockmap test signature to trigger bug, [TEST]: (1, 1, 1, sendmsg, pass,redir,start 1,end 2,pop (1,2),ktls,): Fixes: d3b18ad31f93d ("tls: add bpf support to sk_msg handling") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com> Cc: stable@vger.kernel.org Link: https://lore.kernel.org/bpf/20200111061206.8028-7-john.fastabend@gmail.com
2020-01-11 06:12:04 +00:00
/* This can happen if above tls_split_open_record allocates
* a single large encryption buffer instead of two smaller
* ones. In this case adjust pointers and continue without
* split.
*/
if (!msg_pl->sg.size) {
tls_merge_open_record(sk, rec, tmp, orig_end);
msg_pl = &rec->msg_plaintext;
msg_en = &rec->msg_encrypted;
split = false;
}
sk_msg_trim(sk, msg_en, msg_pl->sg.size +
prot->overhead_size);
}
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
rec->tx_flags = flags;
req = &rec->aead_req;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
i = msg_pl->sg.end;
sk_msg_iter_var_prev(i);
rec->content_type = record_type;
if (prot->version == TLS_1_3_VERSION) {
/* Add content type to end of message. No padding added */
sg_set_buf(&rec->sg_content_type, &rec->content_type, 1);
sg_mark_end(&rec->sg_content_type);
sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1,
&rec->sg_content_type);
} else {
sg_mark_end(sk_msg_elem(msg_pl, i));
}
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (msg_pl->sg.end < msg_pl->sg.start) {
sg_chain(&msg_pl->sg.data[msg_pl->sg.start],
MAX_SKB_FRAGS - msg_pl->sg.start + 1,
msg_pl->sg.data);
}
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
i = msg_pl->sg.start;
sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]);
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
i = msg_en->sg.end;
sk_msg_iter_var_prev(i);
sg_mark_end(sk_msg_elem(msg_en, i));
i = msg_en->sg.start;
sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size,
tls_ctx->tx.rec_seq, record_type, prot);
tls_fill_prepend(tls_ctx,
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
page_address(sg_page(&msg_en->sg.data[i])) +
msg_en->sg.data[i].offset,
msg_pl->sg.size + prot->tail_size,
record_type);
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
tls_ctx->pending_open_record_frags = false;
rc = tls_do_encryption(sk, tls_ctx, ctx, req,
msg_pl->sg.size + prot->tail_size, i);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (rc < 0) {
if (rc != -EINPROGRESS) {
tls_err_abort(sk, -EBADMSG);
if (split) {
tls_ctx->pending_open_record_frags = true;
tls_merge_open_record(sk, rec, tmp, orig_end);
}
}
ctx->async_capable = 1;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
return rc;
} else if (split) {
msg_pl = &tmp->msg_plaintext;
msg_en = &tmp->msg_encrypted;
sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size);
tls_ctx->pending_open_record_frags = true;
ctx->open_rec = tmp;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
}
net/tls: Fixed race condition in async encryption On processors with multi-engine crypto accelerators, it is possible that multiple records get encrypted in parallel and their encryption completion is notified to different cpus in multicore processor. This leads to the situation where tls_encrypt_done() starts executing in parallel on different cores. In current implementation, encrypted records are queued to tx_ready_list in tls_encrypt_done(). This requires addition to linked list 'tx_ready_list' to be protected. As tls_decrypt_done() could be executing in irq content, it is not possible to protect linked list addition operation using a lock. To fix the problem, we remove linked list addition operation from the irq context. We do tx_ready_list addition/removal operation from application context only and get rid of possible multiple access to the linked list. Before starting encryption on the record, we add it to the tail of tx_ready_list. To prevent tls_tx_records() from transmitting it, we mark the record with a new flag 'tx_ready' in 'struct tls_rec'. When record encryption gets completed, tls_encrypt_done() has to only update the 'tx_ready' flag to true & linked list add operation is not required. The changed logic brings some other side benefits. Since the records are always submitted in tls sequence number order for encryption, the tx_ready_list always remains sorted and addition of new records to it does not have to traverse the linked list. Lastly, we renamed tx_ready_list in 'struct tls_sw_context_tx' to 'tx_list'. This is because now, the some of the records at the tail are not ready to transmit. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption") Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-24 10:05:56 +00:00
return tls_tx_records(sk, flags);
}
static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk,
bool full_record, u8 record_type,
ssize_t *copied, int flags)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
struct sk_msg msg_redir = { };
struct sk_psock *psock;
struct sock *sk_redir;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
struct tls_rec *rec;
bpf: sk_msg, sock{map|hash} redirect through ULP A sockmap program that redirects through a kTLS ULP enabled socket will not work correctly because the ULP layer is skipped. This fixes the behavior to call through the ULP layer on redirect to ensure any operations required on the data stream at the ULP layer continue to be applied. To do this we add an internal flag MSG_SENDPAGE_NOPOLICY to avoid calling the BPF layer on a redirected message. This is required to avoid calling the BPF layer multiple times (possibly recursively) which is not the current/expected behavior without ULPs. In the future we may add a redirect flag if users _do_ want the policy applied again but this would need to work for both ULP and non-ULP sockets and be opt-in to avoid breaking existing programs. Also to avoid polluting the flag space with an internal flag we reuse the flag space overlapping MSG_SENDPAGE_NOPOLICY with MSG_WAITFORONE. Here WAITFORONE is specific to recv path and SENDPAGE_NOPOLICY is only used for sendpage hooks. The last thing to verify is user space API is masked correctly to ensure the flag can not be set by user. (Note this needs to be true regardless because we have internal flags already in-use that user space should not be able to set). But for completeness we have two UAPI paths into sendpage, sendfile and splice. In the sendfile case the function do_sendfile() zero's flags, ./fs/read_write.c: static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos, size_t count, loff_t max) { ... fl = 0; #if 0 /* * We need to debate whether we can enable this or not. The * man page documents EAGAIN return for the output at least, * and the application is arguably buggy if it doesn't expect * EAGAIN on a non-blocking file descriptor. */ if (in.file->f_flags & O_NONBLOCK) fl = SPLICE_F_NONBLOCK; #endif file_start_write(out.file); retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl); } In the splice case the pipe_to_sendpage "actor" is used which masks flags with SPLICE_F_MORE. ./fs/splice.c: static int pipe_to_sendpage(struct pipe_inode_info *pipe, struct pipe_buffer *buf, struct splice_desc *sd) { ... more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0; ... } Confirming what we expect that internal flags are in fact internal to socket side. Fixes: d3b18ad31f93 ("tls: add bpf support to sk_msg handling") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-12-20 19:35:35 +00:00
bool enospc, policy;
int err = 0, send;
u32 delta = 0;
bpf: sk_msg, sock{map|hash} redirect through ULP A sockmap program that redirects through a kTLS ULP enabled socket will not work correctly because the ULP layer is skipped. This fixes the behavior to call through the ULP layer on redirect to ensure any operations required on the data stream at the ULP layer continue to be applied. To do this we add an internal flag MSG_SENDPAGE_NOPOLICY to avoid calling the BPF layer on a redirected message. This is required to avoid calling the BPF layer multiple times (possibly recursively) which is not the current/expected behavior without ULPs. In the future we may add a redirect flag if users _do_ want the policy applied again but this would need to work for both ULP and non-ULP sockets and be opt-in to avoid breaking existing programs. Also to avoid polluting the flag space with an internal flag we reuse the flag space overlapping MSG_SENDPAGE_NOPOLICY with MSG_WAITFORONE. Here WAITFORONE is specific to recv path and SENDPAGE_NOPOLICY is only used for sendpage hooks. The last thing to verify is user space API is masked correctly to ensure the flag can not be set by user. (Note this needs to be true regardless because we have internal flags already in-use that user space should not be able to set). But for completeness we have two UAPI paths into sendpage, sendfile and splice. In the sendfile case the function do_sendfile() zero's flags, ./fs/read_write.c: static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos, size_t count, loff_t max) { ... fl = 0; #if 0 /* * We need to debate whether we can enable this or not. The * man page documents EAGAIN return for the output at least, * and the application is arguably buggy if it doesn't expect * EAGAIN on a non-blocking file descriptor. */ if (in.file->f_flags & O_NONBLOCK) fl = SPLICE_F_NONBLOCK; #endif file_start_write(out.file); retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl); } In the splice case the pipe_to_sendpage "actor" is used which masks flags with SPLICE_F_MORE. ./fs/splice.c: static int pipe_to_sendpage(struct pipe_inode_info *pipe, struct pipe_buffer *buf, struct splice_desc *sd) { ... more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0; ... } Confirming what we expect that internal flags are in fact internal to socket side. Fixes: d3b18ad31f93 ("tls: add bpf support to sk_msg handling") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-12-20 19:35:35 +00:00
policy = !(flags & MSG_SENDPAGE_NOPOLICY);
psock = sk_psock_get(sk);
if (!psock || !policy) {
err = tls_push_record(sk, flags, record_type);
if (err && sk->sk_err == EBADMSG) {
*copied -= sk_msg_free(sk, msg);
tls_free_open_rec(sk);
err = -sk->sk_err;
}
if (psock)
sk_psock_put(sk, psock);
return err;
}
more_data:
enospc = sk_msg_full(msg);
if (psock->eval == __SK_NONE) {
delta = msg->sg.size;
psock->eval = sk_psock_msg_verdict(sk, psock, msg);
delta -= msg->sg.size;
}
if (msg->cork_bytes && msg->cork_bytes > msg->sg.size &&
!enospc && !full_record) {
err = -ENOSPC;
goto out_err;
}
msg->cork_bytes = 0;
send = msg->sg.size;
if (msg->apply_bytes && msg->apply_bytes < send)
send = msg->apply_bytes;
switch (psock->eval) {
case __SK_PASS:
err = tls_push_record(sk, flags, record_type);
if (err && sk->sk_err == EBADMSG) {
*copied -= sk_msg_free(sk, msg);
tls_free_open_rec(sk);
err = -sk->sk_err;
goto out_err;
}
break;
case __SK_REDIRECT:
sk_redir = psock->sk_redir;
memcpy(&msg_redir, msg, sizeof(*msg));
if (msg->apply_bytes < send)
msg->apply_bytes = 0;
else
msg->apply_bytes -= send;
sk_msg_return_zero(sk, msg, send);
msg->sg.size -= send;
release_sock(sk);
err = tcp_bpf_sendmsg_redir(sk_redir, &msg_redir, send, flags);
lock_sock(sk);
if (err < 0) {
*copied -= sk_msg_free_nocharge(sk, &msg_redir);
msg->sg.size = 0;
}
if (msg->sg.size == 0)
tls_free_open_rec(sk);
break;
case __SK_DROP:
default:
sk_msg_free_partial(sk, msg, send);
if (msg->apply_bytes < send)
msg->apply_bytes = 0;
else
msg->apply_bytes -= send;
if (msg->sg.size == 0)
tls_free_open_rec(sk);
*copied -= (send + delta);
err = -EACCES;
}
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (likely(!err)) {
bool reset_eval = !ctx->open_rec;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
rec = ctx->open_rec;
if (rec) {
msg = &rec->msg_plaintext;
if (!msg->apply_bytes)
reset_eval = true;
}
if (reset_eval) {
psock->eval = __SK_NONE;
if (psock->sk_redir) {
sock_put(psock->sk_redir);
psock->sk_redir = NULL;
}
}
if (rec)
goto more_data;
}
out_err:
sk_psock_put(sk, psock);
return err;
}
static int tls_sw_push_pending_record(struct sock *sk, int flags)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
struct tls_rec *rec = ctx->open_rec;
struct sk_msg *msg_pl;
size_t copied;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (!rec)
return 0;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
msg_pl = &rec->msg_plaintext;
copied = msg_pl->sg.size;
if (!copied)
return 0;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA,
&copied, flags);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
}
int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
{
long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_prot_info *prot = &tls_ctx->prot_info;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
bool async_capable = ctx->async_capable;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
unsigned char record_type = TLS_RECORD_TYPE_DATA;
bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
bool eor = !(msg->msg_flags & MSG_MORE);
size_t try_to_copy;
ssize_t copied = 0;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
struct sk_msg *msg_pl, *msg_en;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
struct tls_rec *rec;
int required_size;
int num_async = 0;
bool full_record;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
int record_room;
int num_zc = 0;
int orig_size;
int ret = 0;
2020-05-22 20:10:31 +00:00
int pending;
if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
MSG_CMSG_COMPAT))
return -EOPNOTSUPP;
net/tls: add a TX lock TLS TX needs to release and re-acquire the socket lock if send buffer fills up. TLS SW TX path currently depends on only allowing one thread to enter the function by the abuse of sk_write_pending. If another writer is already waiting for memory no new ones are allowed in. This has two problems: - writers don't wake other threads up when they leave the kernel; meaning that this scheme works for single extra thread (second application thread or delayed work) because memory becoming available will send a wake up request, but as Mallesham and Pooja report with larger number of threads it leads to threads being put to sleep indefinitely; - the delayed work does not get _scheduled_ but it may _run_ when other writers are present leading to crashes as writers don't expect state to change under their feet (same records get pushed and freed multiple times); it's hard to reliably bail from the work, however, because the mere presence of a writer does not guarantee that the writer will push pending records before exiting. Ensuring wakeups always happen will make the code basically open code a mutex. Just use a mutex. The TLS HW TX path does not have any locking (not even the sk_write_pending hack), yet it uses a per-socket sg_tx_data array to push records. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption of records for performance") Reported-by: Mallesham Jatharakonda <mallesh537@gmail.com> Reported-by: Pooja Trivedi <poojatrivedi@gmail.com> Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com> Reviewed-by: Simon Horman <simon.horman@netronome.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-11-05 22:24:35 +00:00
mutex_lock(&tls_ctx->tx_lock);
lock_sock(sk);
if (unlikely(msg->msg_controllen)) {
ret = tls_process_cmsg(sk, msg, &record_type);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (ret) {
if (ret == -EINPROGRESS)
num_async++;
else if (ret != -EAGAIN)
goto send_end;
}
}
while (msg_data_left(msg)) {
if (sk->sk_err) {
ret = -sk->sk_err;
goto send_end;
}
if (ctx->open_rec)
rec = ctx->open_rec;
else
rec = ctx->open_rec = tls_get_rec(sk);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (!rec) {
ret = -ENOMEM;
goto send_end;
}
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
msg_pl = &rec->msg_plaintext;
msg_en = &rec->msg_encrypted;
orig_size = msg_pl->sg.size;
full_record = false;
try_to_copy = msg_data_left(msg);
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
if (try_to_copy >= record_room) {
try_to_copy = record_room;
full_record = true;
}
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
required_size = msg_pl->sg.size + try_to_copy +
prot->overhead_size;
if (!sk_stream_memory_free(sk))
goto wait_for_sndbuf;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
alloc_encrypted:
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
ret = tls_alloc_encrypted_msg(sk, required_size);
if (ret) {
if (ret != -ENOSPC)
goto wait_for_memory;
/* Adjust try_to_copy according to the amount that was
* actually allocated. The difference is due
* to max sg elements limit
*/
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
try_to_copy -= required_size - msg_en->sg.size;
full_record = true;
}
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (!is_kvec && (full_record || eor) && !async_capable) {
u32 first = msg_pl->sg.end;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter,
msg_pl, try_to_copy);
if (ret)
goto fallback_to_reg_send;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
num_zc++;
copied += try_to_copy;
sk_msg_sg_copy_set(msg_pl, first);
ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
record_type, &copied,
msg->msg_flags);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (ret) {
if (ret == -EINPROGRESS)
num_async++;
else if (ret == -ENOMEM)
goto wait_for_memory;
else if (ctx->open_rec && ret == -ENOSPC)
goto rollback_iter;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
else if (ret != -EAGAIN)
goto send_end;
}
continue;
rollback_iter:
copied -= try_to_copy;
sk_msg_sg_copy_clear(msg_pl, first);
iov_iter_revert(&msg->msg_iter,
msg_pl->sg.size - orig_size);
fallback_to_reg_send:
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
sk_msg_trim(sk, msg_pl, orig_size);
}
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
required_size = msg_pl->sg.size + try_to_copy;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
ret = tls_clone_plaintext_msg(sk, required_size);
if (ret) {
if (ret != -ENOSPC)
goto send_end;
/* Adjust try_to_copy according to the amount that was
* actually allocated. The difference is due
* to max sg elements limit
*/
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
try_to_copy -= required_size - msg_pl->sg.size;
full_record = true;
sk_msg_trim(sk, msg_en,
msg_pl->sg.size + prot->overhead_size);
}
if (try_to_copy) {
ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter,
msg_pl, try_to_copy);
if (ret < 0)
goto trim_sgl;
}
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
/* Open records defined only if successfully copied, otherwise
* we would trim the sg but not reset the open record frags.
*/
tls_ctx->pending_open_record_frags = true;
copied += try_to_copy;
if (full_record || eor) {
ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
record_type, &copied,
msg->msg_flags);
if (ret) {
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (ret == -EINPROGRESS)
num_async++;
else if (ret == -ENOMEM)
goto wait_for_memory;
else if (ret != -EAGAIN) {
if (ret == -ENOSPC)
ret = 0;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
goto send_end;
}
}
}
continue;
wait_for_sndbuf:
set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
wait_for_memory:
ret = sk_stream_wait_memory(sk, &timeo);
if (ret) {
trim_sgl:
if (ctx->open_rec)
tls_trim_both_msgs(sk, orig_size);
goto send_end;
}
if (ctx->open_rec && msg_en->sg.size < required_size)
goto alloc_encrypted;
}
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (!num_async) {
goto send_end;
} else if (num_zc) {
/* Wait for pending encryptions to get completed */
2020-05-22 20:10:31 +00:00
spin_lock_bh(&ctx->encrypt_compl_lock);
ctx->async_notify = true;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
2020-05-22 20:10:31 +00:00
pending = atomic_read(&ctx->encrypt_pending);
spin_unlock_bh(&ctx->encrypt_compl_lock);
if (pending)
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
else
reinit_completion(&ctx->async_wait.completion);
2020-05-22 20:10:31 +00:00
/* There can be no concurrent accesses, since we have no
* pending encrypt operations
*/
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
WRITE_ONCE(ctx->async_notify, false);
if (ctx->async_wait.err) {
ret = ctx->async_wait.err;
copied = 0;
}
}
/* Transmit if any encryptions have completed */
if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
cancel_delayed_work(&ctx->tx_work.work);
tls_tx_records(sk, msg->msg_flags);
}
send_end:
ret = sk_stream_error(sk, msg->msg_flags, ret);
release_sock(sk);
net/tls: add a TX lock TLS TX needs to release and re-acquire the socket lock if send buffer fills up. TLS SW TX path currently depends on only allowing one thread to enter the function by the abuse of sk_write_pending. If another writer is already waiting for memory no new ones are allowed in. This has two problems: - writers don't wake other threads up when they leave the kernel; meaning that this scheme works for single extra thread (second application thread or delayed work) because memory becoming available will send a wake up request, but as Mallesham and Pooja report with larger number of threads it leads to threads being put to sleep indefinitely; - the delayed work does not get _scheduled_ but it may _run_ when other writers are present leading to crashes as writers don't expect state to change under their feet (same records get pushed and freed multiple times); it's hard to reliably bail from the work, however, because the mere presence of a writer does not guarantee that the writer will push pending records before exiting. Ensuring wakeups always happen will make the code basically open code a mutex. Just use a mutex. The TLS HW TX path does not have any locking (not even the sk_write_pending hack), yet it uses a per-socket sg_tx_data array to push records. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption of records for performance") Reported-by: Mallesham Jatharakonda <mallesh537@gmail.com> Reported-by: Pooja Trivedi <poojatrivedi@gmail.com> Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com> Reviewed-by: Simon Horman <simon.horman@netronome.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-11-05 22:24:35 +00:00
mutex_unlock(&tls_ctx->tx_lock);
return copied > 0 ? copied : ret;
}
static int tls_sw_do_sendpage(struct sock *sk, struct page *page,
int offset, size_t size, int flags)
{
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
struct tls_prot_info *prot = &tls_ctx->prot_info;
unsigned char record_type = TLS_RECORD_TYPE_DATA;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
struct sk_msg *msg_pl;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
struct tls_rec *rec;
int num_async = 0;
ssize_t copied = 0;
bool full_record;
int record_room;
int ret = 0;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
bool eor;
tls: prevent oversized sendfile() hangs by ignoring MSG_MORE We got multiple reports that multi_chunk_sendfile test case from tls selftest fails. This was sort of expected, as the original fix was never applied (see it in the first Link:). The test in question uses sendfile() with count larger than the size of the underlying file. This will make splice set MSG_MORE on all sendpage calls, meaning TLS will never close and flush the last partial record. Eric seem to have addressed a similar problem in commit 35f9c09fe9c7 ("tcp: tcp_sendpages() should call tcp_push() once") by introducing MSG_SENDPAGE_NOTLAST. Unlike MSG_MORE MSG_SENDPAGE_NOTLAST is not set on the last call of a "pipefull" of data (PIPE_DEF_BUFFERS == 16, so every 16 pages or whenever we run out of data). Having a break every 16 pages should be fine, TLS can pack exactly 4 pages into a record, so for aligned reads there should be no difference, unaligned may see one extra record per sendpage(). Sticking to TCP semantics seems preferable to modifying splice, but we can revisit it if real life scenarios show a regression. Reported-by: Vadim Fedorenko <vfedorenko@novek.ru> Reported-by: Seth Forshee <seth.forshee@canonical.com> Link: https://lore.kernel.org/netdev/1591392508-14592-1-git-send-email-pooja.trivedi@stackpath.com/ Fixes: 3c4d7559159b ("tls: kernel TLS support") Signed-off-by: Jakub Kicinski <kuba@kernel.org> Tested-by: Seth Forshee <seth.forshee@canonical.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-06-18 20:34:06 +00:00
eor = !(flags & MSG_SENDPAGE_NOTLAST);
sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
/* Call the sk_stream functions to manage the sndbuf mem. */
while (size > 0) {
size_t copy, required_size;
if (sk->sk_err) {
ret = -sk->sk_err;
goto sendpage_end;
}
if (ctx->open_rec)
rec = ctx->open_rec;
else
rec = ctx->open_rec = tls_get_rec(sk);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (!rec) {
ret = -ENOMEM;
goto sendpage_end;
}
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
msg_pl = &rec->msg_plaintext;
full_record = false;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
copy = size;
if (copy >= record_room) {
copy = record_room;
full_record = true;
}
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
required_size = msg_pl->sg.size + copy + prot->overhead_size;
if (!sk_stream_memory_free(sk))
goto wait_for_sndbuf;
alloc_payload:
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
ret = tls_alloc_encrypted_msg(sk, required_size);
if (ret) {
if (ret != -ENOSPC)
goto wait_for_memory;
/* Adjust copy according to the amount that was
* actually allocated. The difference is due
* to max sg elements limit
*/
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
copy -= required_size - msg_pl->sg.size;
full_record = true;
}
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
sk_msg_page_add(msg_pl, page, copy, offset);
sk_mem_charge(sk, copy);
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
offset += copy;
size -= copy;
copied += copy;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
tls_ctx->pending_open_record_frags = true;
if (full_record || eor || sk_msg_full(msg_pl)) {
ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
record_type, &copied, flags);
if (ret) {
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (ret == -EINPROGRESS)
num_async++;
else if (ret == -ENOMEM)
goto wait_for_memory;
else if (ret != -EAGAIN) {
if (ret == -ENOSPC)
ret = 0;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
goto sendpage_end;
}
}
}
continue;
wait_for_sndbuf:
set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
wait_for_memory:
ret = sk_stream_wait_memory(sk, &timeo);
if (ret) {
if (ctx->open_rec)
tls_trim_both_msgs(sk, msg_pl->sg.size);
goto sendpage_end;
}
if (ctx->open_rec)
goto alloc_payload;
}
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (num_async) {
/* Transmit if any encryptions have completed */
if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
cancel_delayed_work(&ctx->tx_work.work);
tls_tx_records(sk, flags);
}
}
sendpage_end:
ret = sk_stream_error(sk, flags, ret);
return copied > 0 ? copied : ret;
}
int tls_sw_sendpage_locked(struct sock *sk, struct page *page,
int offset, size_t size, int flags)
{
if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY |
MSG_NO_SHARED_FRAGS))
return -EOPNOTSUPP;
return tls_sw_do_sendpage(sk, page, offset, size, flags);
}
bpf: sk_msg, sock{map|hash} redirect through ULP A sockmap program that redirects through a kTLS ULP enabled socket will not work correctly because the ULP layer is skipped. This fixes the behavior to call through the ULP layer on redirect to ensure any operations required on the data stream at the ULP layer continue to be applied. To do this we add an internal flag MSG_SENDPAGE_NOPOLICY to avoid calling the BPF layer on a redirected message. This is required to avoid calling the BPF layer multiple times (possibly recursively) which is not the current/expected behavior without ULPs. In the future we may add a redirect flag if users _do_ want the policy applied again but this would need to work for both ULP and non-ULP sockets and be opt-in to avoid breaking existing programs. Also to avoid polluting the flag space with an internal flag we reuse the flag space overlapping MSG_SENDPAGE_NOPOLICY with MSG_WAITFORONE. Here WAITFORONE is specific to recv path and SENDPAGE_NOPOLICY is only used for sendpage hooks. The last thing to verify is user space API is masked correctly to ensure the flag can not be set by user. (Note this needs to be true regardless because we have internal flags already in-use that user space should not be able to set). But for completeness we have two UAPI paths into sendpage, sendfile and splice. In the sendfile case the function do_sendfile() zero's flags, ./fs/read_write.c: static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos, size_t count, loff_t max) { ... fl = 0; #if 0 /* * We need to debate whether we can enable this or not. The * man page documents EAGAIN return for the output at least, * and the application is arguably buggy if it doesn't expect * EAGAIN on a non-blocking file descriptor. */ if (in.file->f_flags & O_NONBLOCK) fl = SPLICE_F_NONBLOCK; #endif file_start_write(out.file); retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl); } In the splice case the pipe_to_sendpage "actor" is used which masks flags with SPLICE_F_MORE. ./fs/splice.c: static int pipe_to_sendpage(struct pipe_inode_info *pipe, struct pipe_buffer *buf, struct splice_desc *sd) { ... more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0; ... } Confirming what we expect that internal flags are in fact internal to socket side. Fixes: d3b18ad31f93 ("tls: add bpf support to sk_msg handling") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-12-20 19:35:35 +00:00
int tls_sw_sendpage(struct sock *sk, struct page *page,
int offset, size_t size, int flags)
{
net/tls: add a TX lock TLS TX needs to release and re-acquire the socket lock if send buffer fills up. TLS SW TX path currently depends on only allowing one thread to enter the function by the abuse of sk_write_pending. If another writer is already waiting for memory no new ones are allowed in. This has two problems: - writers don't wake other threads up when they leave the kernel; meaning that this scheme works for single extra thread (second application thread or delayed work) because memory becoming available will send a wake up request, but as Mallesham and Pooja report with larger number of threads it leads to threads being put to sleep indefinitely; - the delayed work does not get _scheduled_ but it may _run_ when other writers are present leading to crashes as writers don't expect state to change under their feet (same records get pushed and freed multiple times); it's hard to reliably bail from the work, however, because the mere presence of a writer does not guarantee that the writer will push pending records before exiting. Ensuring wakeups always happen will make the code basically open code a mutex. Just use a mutex. The TLS HW TX path does not have any locking (not even the sk_write_pending hack), yet it uses a per-socket sg_tx_data array to push records. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption of records for performance") Reported-by: Mallesham Jatharakonda <mallesh537@gmail.com> Reported-by: Pooja Trivedi <poojatrivedi@gmail.com> Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com> Reviewed-by: Simon Horman <simon.horman@netronome.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-11-05 22:24:35 +00:00
struct tls_context *tls_ctx = tls_get_ctx(sk);
bpf: sk_msg, sock{map|hash} redirect through ULP A sockmap program that redirects through a kTLS ULP enabled socket will not work correctly because the ULP layer is skipped. This fixes the behavior to call through the ULP layer on redirect to ensure any operations required on the data stream at the ULP layer continue to be applied. To do this we add an internal flag MSG_SENDPAGE_NOPOLICY to avoid calling the BPF layer on a redirected message. This is required to avoid calling the BPF layer multiple times (possibly recursively) which is not the current/expected behavior without ULPs. In the future we may add a redirect flag if users _do_ want the policy applied again but this would need to work for both ULP and non-ULP sockets and be opt-in to avoid breaking existing programs. Also to avoid polluting the flag space with an internal flag we reuse the flag space overlapping MSG_SENDPAGE_NOPOLICY with MSG_WAITFORONE. Here WAITFORONE is specific to recv path and SENDPAGE_NOPOLICY is only used for sendpage hooks. The last thing to verify is user space API is masked correctly to ensure the flag can not be set by user. (Note this needs to be true regardless because we have internal flags already in-use that user space should not be able to set). But for completeness we have two UAPI paths into sendpage, sendfile and splice. In the sendfile case the function do_sendfile() zero's flags, ./fs/read_write.c: static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos, size_t count, loff_t max) { ... fl = 0; #if 0 /* * We need to debate whether we can enable this or not. The * man page documents EAGAIN return for the output at least, * and the application is arguably buggy if it doesn't expect * EAGAIN on a non-blocking file descriptor. */ if (in.file->f_flags & O_NONBLOCK) fl = SPLICE_F_NONBLOCK; #endif file_start_write(out.file); retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl); } In the splice case the pipe_to_sendpage "actor" is used which masks flags with SPLICE_F_MORE. ./fs/splice.c: static int pipe_to_sendpage(struct pipe_inode_info *pipe, struct pipe_buffer *buf, struct splice_desc *sd) { ... more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0; ... } Confirming what we expect that internal flags are in fact internal to socket side. Fixes: d3b18ad31f93 ("tls: add bpf support to sk_msg handling") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-12-20 19:35:35 +00:00
int ret;
if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY))
return -EOPNOTSUPP;
bpf: sk_msg, sock{map|hash} redirect through ULP A sockmap program that redirects through a kTLS ULP enabled socket will not work correctly because the ULP layer is skipped. This fixes the behavior to call through the ULP layer on redirect to ensure any operations required on the data stream at the ULP layer continue to be applied. To do this we add an internal flag MSG_SENDPAGE_NOPOLICY to avoid calling the BPF layer on a redirected message. This is required to avoid calling the BPF layer multiple times (possibly recursively) which is not the current/expected behavior without ULPs. In the future we may add a redirect flag if users _do_ want the policy applied again but this would need to work for both ULP and non-ULP sockets and be opt-in to avoid breaking existing programs. Also to avoid polluting the flag space with an internal flag we reuse the flag space overlapping MSG_SENDPAGE_NOPOLICY with MSG_WAITFORONE. Here WAITFORONE is specific to recv path and SENDPAGE_NOPOLICY is only used for sendpage hooks. The last thing to verify is user space API is masked correctly to ensure the flag can not be set by user. (Note this needs to be true regardless because we have internal flags already in-use that user space should not be able to set). But for completeness we have two UAPI paths into sendpage, sendfile and splice. In the sendfile case the function do_sendfile() zero's flags, ./fs/read_write.c: static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos, size_t count, loff_t max) { ... fl = 0; #if 0 /* * We need to debate whether we can enable this or not. The * man page documents EAGAIN return for the output at least, * and the application is arguably buggy if it doesn't expect * EAGAIN on a non-blocking file descriptor. */ if (in.file->f_flags & O_NONBLOCK) fl = SPLICE_F_NONBLOCK; #endif file_start_write(out.file); retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl); } In the splice case the pipe_to_sendpage "actor" is used which masks flags with SPLICE_F_MORE. ./fs/splice.c: static int pipe_to_sendpage(struct pipe_inode_info *pipe, struct pipe_buffer *buf, struct splice_desc *sd) { ... more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0; ... } Confirming what we expect that internal flags are in fact internal to socket side. Fixes: d3b18ad31f93 ("tls: add bpf support to sk_msg handling") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-12-20 19:35:35 +00:00
net/tls: add a TX lock TLS TX needs to release and re-acquire the socket lock if send buffer fills up. TLS SW TX path currently depends on only allowing one thread to enter the function by the abuse of sk_write_pending. If another writer is already waiting for memory no new ones are allowed in. This has two problems: - writers don't wake other threads up when they leave the kernel; meaning that this scheme works for single extra thread (second application thread or delayed work) because memory becoming available will send a wake up request, but as Mallesham and Pooja report with larger number of threads it leads to threads being put to sleep indefinitely; - the delayed work does not get _scheduled_ but it may _run_ when other writers are present leading to crashes as writers don't expect state to change under their feet (same records get pushed and freed multiple times); it's hard to reliably bail from the work, however, because the mere presence of a writer does not guarantee that the writer will push pending records before exiting. Ensuring wakeups always happen will make the code basically open code a mutex. Just use a mutex. The TLS HW TX path does not have any locking (not even the sk_write_pending hack), yet it uses a per-socket sg_tx_data array to push records. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption of records for performance") Reported-by: Mallesham Jatharakonda <mallesh537@gmail.com> Reported-by: Pooja Trivedi <poojatrivedi@gmail.com> Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com> Reviewed-by: Simon Horman <simon.horman@netronome.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-11-05 22:24:35 +00:00
mutex_lock(&tls_ctx->tx_lock);
bpf: sk_msg, sock{map|hash} redirect through ULP A sockmap program that redirects through a kTLS ULP enabled socket will not work correctly because the ULP layer is skipped. This fixes the behavior to call through the ULP layer on redirect to ensure any operations required on the data stream at the ULP layer continue to be applied. To do this we add an internal flag MSG_SENDPAGE_NOPOLICY to avoid calling the BPF layer on a redirected message. This is required to avoid calling the BPF layer multiple times (possibly recursively) which is not the current/expected behavior without ULPs. In the future we may add a redirect flag if users _do_ want the policy applied again but this would need to work for both ULP and non-ULP sockets and be opt-in to avoid breaking existing programs. Also to avoid polluting the flag space with an internal flag we reuse the flag space overlapping MSG_SENDPAGE_NOPOLICY with MSG_WAITFORONE. Here WAITFORONE is specific to recv path and SENDPAGE_NOPOLICY is only used for sendpage hooks. The last thing to verify is user space API is masked correctly to ensure the flag can not be set by user. (Note this needs to be true regardless because we have internal flags already in-use that user space should not be able to set). But for completeness we have two UAPI paths into sendpage, sendfile and splice. In the sendfile case the function do_sendfile() zero's flags, ./fs/read_write.c: static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos, size_t count, loff_t max) { ... fl = 0; #if 0 /* * We need to debate whether we can enable this or not. The * man page documents EAGAIN return for the output at least, * and the application is arguably buggy if it doesn't expect * EAGAIN on a non-blocking file descriptor. */ if (in.file->f_flags & O_NONBLOCK) fl = SPLICE_F_NONBLOCK; #endif file_start_write(out.file); retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl); } In the splice case the pipe_to_sendpage "actor" is used which masks flags with SPLICE_F_MORE. ./fs/splice.c: static int pipe_to_sendpage(struct pipe_inode_info *pipe, struct pipe_buffer *buf, struct splice_desc *sd) { ... more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0; ... } Confirming what we expect that internal flags are in fact internal to socket side. Fixes: d3b18ad31f93 ("tls: add bpf support to sk_msg handling") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-12-20 19:35:35 +00:00
lock_sock(sk);
ret = tls_sw_do_sendpage(sk, page, offset, size, flags);
release_sock(sk);
net/tls: add a TX lock TLS TX needs to release and re-acquire the socket lock if send buffer fills up. TLS SW TX path currently depends on only allowing one thread to enter the function by the abuse of sk_write_pending. If another writer is already waiting for memory no new ones are allowed in. This has two problems: - writers don't wake other threads up when they leave the kernel; meaning that this scheme works for single extra thread (second application thread or delayed work) because memory becoming available will send a wake up request, but as Mallesham and Pooja report with larger number of threads it leads to threads being put to sleep indefinitely; - the delayed work does not get _scheduled_ but it may _run_ when other writers are present leading to crashes as writers don't expect state to change under their feet (same records get pushed and freed multiple times); it's hard to reliably bail from the work, however, because the mere presence of a writer does not guarantee that the writer will push pending records before exiting. Ensuring wakeups always happen will make the code basically open code a mutex. Just use a mutex. The TLS HW TX path does not have any locking (not even the sk_write_pending hack), yet it uses a per-socket sg_tx_data array to push records. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption of records for performance") Reported-by: Mallesham Jatharakonda <mallesh537@gmail.com> Reported-by: Pooja Trivedi <poojatrivedi@gmail.com> Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com> Reviewed-by: Simon Horman <simon.horman@netronome.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-11-05 22:24:35 +00:00
mutex_unlock(&tls_ctx->tx_lock);
bpf: sk_msg, sock{map|hash} redirect through ULP A sockmap program that redirects through a kTLS ULP enabled socket will not work correctly because the ULP layer is skipped. This fixes the behavior to call through the ULP layer on redirect to ensure any operations required on the data stream at the ULP layer continue to be applied. To do this we add an internal flag MSG_SENDPAGE_NOPOLICY to avoid calling the BPF layer on a redirected message. This is required to avoid calling the BPF layer multiple times (possibly recursively) which is not the current/expected behavior without ULPs. In the future we may add a redirect flag if users _do_ want the policy applied again but this would need to work for both ULP and non-ULP sockets and be opt-in to avoid breaking existing programs. Also to avoid polluting the flag space with an internal flag we reuse the flag space overlapping MSG_SENDPAGE_NOPOLICY with MSG_WAITFORONE. Here WAITFORONE is specific to recv path and SENDPAGE_NOPOLICY is only used for sendpage hooks. The last thing to verify is user space API is masked correctly to ensure the flag can not be set by user. (Note this needs to be true regardless because we have internal flags already in-use that user space should not be able to set). But for completeness we have two UAPI paths into sendpage, sendfile and splice. In the sendfile case the function do_sendfile() zero's flags, ./fs/read_write.c: static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos, size_t count, loff_t max) { ... fl = 0; #if 0 /* * We need to debate whether we can enable this or not. The * man page documents EAGAIN return for the output at least, * and the application is arguably buggy if it doesn't expect * EAGAIN on a non-blocking file descriptor. */ if (in.file->f_flags & O_NONBLOCK) fl = SPLICE_F_NONBLOCK; #endif file_start_write(out.file); retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl); } In the splice case the pipe_to_sendpage "actor" is used which masks flags with SPLICE_F_MORE. ./fs/splice.c: static int pipe_to_sendpage(struct pipe_inode_info *pipe, struct pipe_buffer *buf, struct splice_desc *sd) { ... more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0; ... } Confirming what we expect that internal flags are in fact internal to socket side. Fixes: d3b18ad31f93 ("tls: add bpf support to sk_msg handling") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-12-20 19:35:35 +00:00
return ret;
}
static struct sk_buff *tls_wait_data(struct sock *sk, struct sk_psock *psock,
bool nonblock, long timeo, int *err)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
struct sk_buff *skb;
DEFINE_WAIT_FUNC(wait, woken_wake_function);
while (!(skb = ctx->recv_pkt) && sk_psock_queue_empty(psock)) {
if (sk->sk_err) {
*err = sock_error(sk);
return NULL;
}
if (!skb_queue_empty(&sk->sk_receive_queue)) {
__strp_unpause(&ctx->strp);
if (ctx->recv_pkt)
return ctx->recv_pkt;
}
if (sk->sk_shutdown & RCV_SHUTDOWN)
return NULL;
if (sock_flag(sk, SOCK_DONE))
return NULL;
if (nonblock || !timeo) {
*err = -EAGAIN;
return NULL;
}
add_wait_queue(sk_sleep(sk), &wait);
sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
sk_wait_event(sk, &timeo,
ctx->recv_pkt != skb ||
!sk_psock_queue_empty(psock),
&wait);
sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
remove_wait_queue(sk_sleep(sk), &wait);
/* Handle signals */
if (signal_pending(current)) {
*err = sock_intr_errno(timeo);
return NULL;
}
}
return skb;
}
static int tls_setup_from_iter(struct iov_iter *from,
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
int length, int *pages_used,
struct scatterlist *to,
int to_max_pages)
{
int rc = 0, i = 0, num_elem = *pages_used, maxpages;
struct page *pages[MAX_SKB_FRAGS];
unsigned int size = 0;
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
ssize_t copied, use;
size_t offset;
while (length > 0) {
i = 0;
maxpages = to_max_pages - num_elem;
if (maxpages == 0) {
rc = -EFAULT;
goto out;
}
copied = iov_iter_get_pages(from, pages,
length,
maxpages, &offset);
if (copied <= 0) {
rc = -EFAULT;
goto out;
}
iov_iter_advance(from, copied);
length -= copied;
size += copied;
while (copied) {
use = min_t(int, copied, PAGE_SIZE - offset);
sg_set_page(&to[num_elem],
pages[i], use, offset);
sg_unmark_end(&to[num_elem]);
/* We do not uncharge memory from this API */
offset = 0;
copied -= use;
i++;
num_elem++;
}
}
/* Mark the end in the last sg entry if newly added */
if (num_elem > *pages_used)
sg_mark_end(&to[num_elem - 1]);
out:
if (rc)
iov_iter_revert(from, size);
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
*pages_used = num_elem;
return rc;
}
/* This function decrypts the input skb into either out_iov or in out_sg
* or in skb buffers itself. The input parameter 'zc' indicates if
* zero-copy mode needs to be tried or not. With zero-copy mode, either
* out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
* NULL, then the decryption happens inside skb buffers itself, i.e.
* zero-copy gets disabled and 'zc' is updated.
*/
static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
struct iov_iter *out_iov,
struct scatterlist *out_sg,
struct tls_decrypt_arg *darg)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
struct tls_prot_info *prot = &tls_ctx->prot_info;
int n_sgin, n_sgout, aead_size, err, pages = 0;
struct strp_msg *rxm = strp_msg(skb);
struct tls_msg *tlm = tls_msg(skb);
struct aead_request *aead_req;
struct sk_buff *unused;
struct scatterlist *sgin = NULL;
struct scatterlist *sgout = NULL;
const int data_len = rxm->full_len - prot->overhead_size;
int tail_pages = !!prot->tail_size;
struct tls_decrypt_ctx *dctx;
int iv_offset = 0;
u8 *mem;
if (darg->zc && (out_iov || out_sg)) {
if (out_iov)
n_sgout = 1 + tail_pages +
iov_iter_npages_cap(out_iov, INT_MAX, data_len);
else
n_sgout = sg_nents(out_sg);
n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size,
rxm->full_len - prot->prepend_size);
} else {
n_sgout = 0;
darg->zc = false;
n_sgin = skb_cow_data(skb, 0, &unused);
}
if (n_sgin < 1)
return -EBADMSG;
/* Increment to accommodate AAD */
n_sgin = n_sgin + 1;
/* Allocate a single block of memory which contains
* aead_req || tls_decrypt_ctx.
* Both structs are variable length.
*/
aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
mem = kmalloc(aead_size + struct_size(dctx, sg, n_sgin + n_sgout),
sk->sk_allocation);
if (!mem)
return -ENOMEM;
/* Segment the allocated memory */
aead_req = (struct aead_request *)mem;
dctx = (struct tls_decrypt_ctx *)(mem + aead_size);
sgin = &dctx->sg[0];
sgout = &dctx->sg[n_sgin];
/* For CCM based ciphers, first byte of nonce+iv is a constant */
switch (prot->cipher_type) {
case TLS_CIPHER_AES_CCM_128:
dctx->iv[0] = TLS_AES_CCM_IV_B0_BYTE;
iv_offset = 1;
break;
case TLS_CIPHER_SM4_CCM:
dctx->iv[0] = TLS_SM4_CCM_IV_B0_BYTE;
iv_offset = 1;
break;
}
/* Prepare IV */
if (prot->version == TLS_1_3_VERSION ||
prot->cipher_type == TLS_CIPHER_CHACHA20_POLY1305) {
memcpy(&dctx->iv[iv_offset], tls_ctx->rx.iv,
2022-03-31 07:04:28 +00:00
prot->iv_size + prot->salt_size);
} else {
err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
&dctx->iv[iv_offset] + prot->salt_size,
prot->iv_size);
if (err < 0)
goto exit_free;
memcpy(&dctx->iv[iv_offset], tls_ctx->rx.iv, prot->salt_size);
}
tls_xor_iv_with_seq(prot, &dctx->iv[iv_offset], tls_ctx->rx.rec_seq);
/* Prepare AAD */
tls_make_aad(dctx->aad, rxm->full_len - prot->overhead_size +
prot->tail_size,
tls_ctx->rx.rec_seq, tlm->control, prot);
/* Prepare sgin */
sg_init_table(sgin, n_sgin);
sg_set_buf(&sgin[0], dctx->aad, prot->aad_size);
err = skb_to_sgvec(skb, &sgin[1],
rxm->offset + prot->prepend_size,
rxm->full_len - prot->prepend_size);
if (err < 0)
goto exit_free;
if (n_sgout) {
if (out_iov) {
sg_init_table(sgout, n_sgout);
sg_set_buf(&sgout[0], dctx->aad, prot->aad_size);
err = tls_setup_from_iter(out_iov, data_len,
&pages, &sgout[1],
(n_sgout - 1 - tail_pages));
if (err < 0)
goto fallback_to_reg_recv;
if (prot->tail_size) {
sg_unmark_end(&sgout[pages]);
sg_set_buf(&sgout[pages + 1], &dctx->tail,
prot->tail_size);
sg_mark_end(&sgout[pages + 1]);
}
} else if (out_sg) {
memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
} else {
goto fallback_to_reg_recv;
}
} else {
fallback_to_reg_recv:
sgout = sgin;
pages = 0;
darg->zc = false;
}
/* Prepare and submit AEAD request */
err = tls_do_decryption(sk, skb, sgin, sgout, dctx->iv,
data_len + prot->tail_size, aead_req, darg);
if (darg->async)
return 0;
if (prot->tail_size)
darg->tail = dctx->tail;
/* Release the pages in case iov was mapped to pages */
for (; pages > 0; pages--)
put_page(sg_page(&sgout[pages]));
exit_free:
kfree(mem);
return err;
}
static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
struct iov_iter *dest,
struct tls_decrypt_arg *darg)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_prot_info *prot = &tls_ctx->prot_info;
struct strp_msg *rxm = strp_msg(skb);
struct tls_msg *tlm = tls_msg(skb);
int pad, err;
if (tlm->decrypted) {
darg->zc = false;
darg->async = false;
return 0;
}
if (tls_ctx->rx_conf == TLS_HW) {
err = tls_device_decrypted(sk, tls_ctx, skb, rxm);
if (err < 0)
return err;
if (err > 0) {
tlm->decrypted = 1;
darg->zc = false;
darg->async = false;
goto decrypt_done;
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
}
}
err = decrypt_internal(sk, skb, dest, NULL, darg);
if (err < 0) {
if (err == -EBADMSG)
TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTERROR);
return err;
}
if (darg->async)
goto decrypt_next;
/* If opportunistic TLS 1.3 ZC failed retry without ZC */
if (unlikely(darg->zc && prot->version == TLS_1_3_VERSION &&
darg->tail != TLS_RECORD_TYPE_DATA)) {
darg->zc = false;
TLS_INC_STATS(sock_net(sk), LINUX_MIN_TLSDECRYPTRETRY);
return decrypt_skb_update(sk, skb, dest, darg);
}
decrypt_done:
pad = tls_padding_length(prot, skb, darg);
if (pad < 0)
return pad;
rxm->full_len -= pad;
rxm->offset += prot->prepend_size;
rxm->full_len -= prot->overhead_size;
tlm->decrypted = 1;
decrypt_next:
tls_advance_record_sn(sk, prot, &tls_ctx->rx);
return 0;
}
int decrypt_skb(struct sock *sk, struct sk_buff *skb,
struct scatterlist *sgout)
{
struct tls_decrypt_arg darg = { .zc = true, };
return decrypt_internal(sk, skb, NULL, sgout, &darg);
}
static int tls_record_content_type(struct msghdr *msg, struct tls_msg *tlm,
u8 *control)
{
int err;
if (!*control) {
*control = tlm->control;
if (!*control)
return -EBADMSG;
err = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
sizeof(*control), control);
if (*control != TLS_RECORD_TYPE_DATA) {
if (err || msg->msg_flags & MSG_CTRUNC)
return -EIO;
}
} else if (*control != tlm->control) {
return 0;
}
return 1;
}
/* This function traverses the rx_list in tls receive context to copies the
* decrypted records into the buffer provided by caller zero copy is not
* true. Further, the records are removed from the rx_list if it is not a peek
* case and the record has been consumed completely.
*/
static int process_rx_list(struct tls_sw_context_rx *ctx,
struct msghdr *msg,
u8 *control,
size_t skip,
size_t len,
bool zc,
bool is_peek)
{
struct sk_buff *skb = skb_peek(&ctx->rx_list);
struct tls_msg *tlm;
ssize_t copied = 0;
int err;
while (skip && skb) {
struct strp_msg *rxm = strp_msg(skb);
tlm = tls_msg(skb);
err = tls_record_content_type(msg, tlm, control);
if (err <= 0)
goto out;
if (skip < rxm->full_len)
break;
skip = skip - rxm->full_len;
skb = skb_peek_next(skb, &ctx->rx_list);
}
while (len && skb) {
struct sk_buff *next_skb;
struct strp_msg *rxm = strp_msg(skb);
int chunk = min_t(unsigned int, rxm->full_len - skip, len);
tlm = tls_msg(skb);
err = tls_record_content_type(msg, tlm, control);
if (err <= 0)
goto out;
if (!zc || (rxm->full_len - skip) > len) {
err = skb_copy_datagram_msg(skb, rxm->offset + skip,
msg, chunk);
if (err < 0)
goto out;
}
len = len - chunk;
copied = copied + chunk;
/* Consume the data from record if it is non-peek case*/
if (!is_peek) {
rxm->offset = rxm->offset + chunk;
rxm->full_len = rxm->full_len - chunk;
/* Return if there is unconsumed data in the record */
if (rxm->full_len - skip)
break;
}
/* The remaining skip-bytes must lie in 1st record in rx_list.
* So from the 2nd record, 'skip' should be 0.
*/
skip = 0;
if (msg)
msg->msg_flags |= MSG_EOR;
next_skb = skb_peek_next(skb, &ctx->rx_list);
if (!is_peek) {
__skb_unlink(skb, &ctx->rx_list);
consume_skb(skb);
}
skb = next_skb;
}
err = 0;
out:
return copied ? : err;
}
static void
tls_read_flush_backlog(struct sock *sk, struct tls_prot_info *prot,
size_t len_left, size_t decrypted, ssize_t done,
size_t *flushed_at)
{
size_t max_rec;
if (len_left <= decrypted)
return;
max_rec = prot->overhead_size - prot->tail_size + TLS_MAX_PAYLOAD_SIZE;
if (done - *flushed_at < SZ_128K && tcp_inq(sk) > max_rec)
return;
*flushed_at = done;
sk_flush_backlog(sk);
}
int tls_sw_recvmsg(struct sock *sk,
struct msghdr *msg,
size_t len,
int flags,
int *addr_len)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
struct tls_prot_info *prot = &tls_ctx->prot_info;
struct sk_psock *psock;
unsigned char control = 0;
ssize_t decrypted = 0;
size_t flushed_at = 0;
struct strp_msg *rxm;
struct tls_msg *tlm;
struct sk_buff *skb;
ssize_t copied = 0;
bool async = false;
int target, err = 0;
long timeo;
bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
bool is_peek = flags & MSG_PEEK;
bpf: Fix running sk_skb program types with ktls KTLS uses a stream parser to collect TLS messages and send them to the upper layer tls receive handler. This ensures the tls receiver has a full TLS header to parse when it is run. However, when a socket has BPF_SK_SKB_STREAM_VERDICT program attached before KTLS is enabled we end up with two stream parsers running on the same socket. The result is both try to run on the same socket. First the KTLS stream parser runs and calls read_sock() which will tcp_read_sock which in turn calls tcp_rcv_skb(). This dequeues the skb from the sk_receive_queue. When this is done KTLS code then data_ready() callback which because we stacked KTLS on top of the bpf stream verdict program has been replaced with sk_psock_start_strp(). This will in turn kick the stream parser again and eventually do the same thing KTLS did above calling into tcp_rcv_skb() and dequeuing a skb from the sk_receive_queue. At this point the data stream is broke. Part of the stream was handled by the KTLS side some other bytes may have been handled by the BPF side. Generally this results in either missing data or more likely a "Bad Message" complaint from the kTLS receive handler as the BPF program steals some bytes meant to be in a TLS header and/or the TLS header length is no longer correct. We've already broke the idealized model where we can stack ULPs in any order with generic callbacks on the TX side to handle this. So in this patch we do the same thing but for RX side. We add a sk_psock_strp_enabled() helper so TLS can learn a BPF verdict program is running and add a tls_sw_has_ctx_rx() helper so BPF side can learn there is a TLS ULP on the socket. Then on BPF side we omit calling our stream parser to avoid breaking the data stream for the KTLS receiver. Then on the KTLS side we call BPF_SK_SKB_STREAM_VERDICT once the KTLS receiver is done with the packet but before it posts the msg to userspace. This gives us symmetry between the TX and RX halfs and IMO makes it usable again. On the TX side we process packets in this order BPF -> TLS -> TCP and on the receive side in the reverse order TCP -> TLS -> BPF. Discovered while testing OpenSSL 3.0 Alpha2.0 release. Fixes: d829e9c4112b5 ("tls: convert to generic sk_msg interface") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Link: https://lore.kernel.org/bpf/159079361946.5745.605854335665044485.stgit@john-Precision-5820-Tower Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2020-05-29 23:06:59 +00:00
bool bpf_strp_enabled;
bool zc_capable;
if (unlikely(flags & MSG_ERRQUEUE))
return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
psock = sk_psock_get(sk);
lock_sock(sk);
bpf: Fix running sk_skb program types with ktls KTLS uses a stream parser to collect TLS messages and send them to the upper layer tls receive handler. This ensures the tls receiver has a full TLS header to parse when it is run. However, when a socket has BPF_SK_SKB_STREAM_VERDICT program attached before KTLS is enabled we end up with two stream parsers running on the same socket. The result is both try to run on the same socket. First the KTLS stream parser runs and calls read_sock() which will tcp_read_sock which in turn calls tcp_rcv_skb(). This dequeues the skb from the sk_receive_queue. When this is done KTLS code then data_ready() callback which because we stacked KTLS on top of the bpf stream verdict program has been replaced with sk_psock_start_strp(). This will in turn kick the stream parser again and eventually do the same thing KTLS did above calling into tcp_rcv_skb() and dequeuing a skb from the sk_receive_queue. At this point the data stream is broke. Part of the stream was handled by the KTLS side some other bytes may have been handled by the BPF side. Generally this results in either missing data or more likely a "Bad Message" complaint from the kTLS receive handler as the BPF program steals some bytes meant to be in a TLS header and/or the TLS header length is no longer correct. We've already broke the idealized model where we can stack ULPs in any order with generic callbacks on the TX side to handle this. So in this patch we do the same thing but for RX side. We add a sk_psock_strp_enabled() helper so TLS can learn a BPF verdict program is running and add a tls_sw_has_ctx_rx() helper so BPF side can learn there is a TLS ULP on the socket. Then on BPF side we omit calling our stream parser to avoid breaking the data stream for the KTLS receiver. Then on the KTLS side we call BPF_SK_SKB_STREAM_VERDICT once the KTLS receiver is done with the packet but before it posts the msg to userspace. This gives us symmetry between the TX and RX halfs and IMO makes it usable again. On the TX side we process packets in this order BPF -> TLS -> TCP and on the receive side in the reverse order TCP -> TLS -> BPF. Discovered while testing OpenSSL 3.0 Alpha2.0 release. Fixes: d829e9c4112b5 ("tls: convert to generic sk_msg interface") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Link: https://lore.kernel.org/bpf/159079361946.5745.605854335665044485.stgit@john-Precision-5820-Tower Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2020-05-29 23:06:59 +00:00
bpf_strp_enabled = sk_psock_strp_enabled(psock);
/* If crypto failed the connection is broken */
err = ctx->async_wait.err;
if (err)
goto end;
/* Process pending decrypted records. It must be non-zero-copy */
err = process_rx_list(ctx, msg, &control, 0, len, false, is_peek);
if (err < 0)
goto end;
copied = err;
if (len <= copied)
goto end;
target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
len = len - copied;
timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
zc_capable = !bpf_strp_enabled && !is_kvec && !is_peek &&
ctx->zc_capable;
decrypted = 0;
while (len && (decrypted + copied < target || ctx->recv_pkt)) {
struct tls_decrypt_arg darg = {};
2022-04-08 18:31:25 +00:00
int to_decrypt, chunk;
skb = tls_wait_data(sk, psock, flags & MSG_DONTWAIT, timeo, &err);
if (!skb) {
if (psock) {
chunk = sk_msg_recvmsg(sk, psock, msg, len,
flags);
if (chunk > 0)
goto leave_on_list;
}
goto recv_end;
}
rxm = strp_msg(skb);
tlm = tls_msg(skb);
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
to_decrypt = rxm->full_len - prot->overhead_size;
if (zc_capable && to_decrypt <= len &&
tlm->control == TLS_RECORD_TYPE_DATA)
darg.zc = true;
/* Do not use async mode if record is non-data */
if (tlm->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled)
darg.async = ctx->async_capable;
else
darg.async = false;
err = decrypt_skb_update(sk, skb, &msg->msg_iter, &darg);
if (err < 0) {
tls_err_abort(sk, -EBADMSG);
goto recv_end;
}
async |= darg.async;
/* If the type of records being processed is not known yet,
* set it to record type just dequeued. If it is already known,
* but does not match the record type just dequeued, go to end.
* We always get record type here since for tls1.2, record type
* is known just after record is dequeued from stream parser.
* For tls1.3, we disable async.
*/
err = tls_record_content_type(msg, tlm, &control);
if (err <= 0)
goto recv_end;
/* periodically flush backlog, and feed strparser */
tls_read_flush_backlog(sk, prot, len, to_decrypt,
decrypted + copied, &flushed_at);
ctx->recv_pkt = NULL;
__strp_unpause(&ctx->strp);
__skb_queue_tail(&ctx->rx_list, skb);
2022-04-08 18:31:25 +00:00
if (async) {
/* TLS 1.2-only, to_decrypt must be text length */
chunk = min_t(int, to_decrypt, len);
leave_on_list:
decrypted += chunk;
len -= chunk;
continue;
2022-04-08 18:31:25 +00:00
}
/* TLS 1.3 may have updated the length by more than overhead */
chunk = rxm->full_len;
if (!darg.zc) {
bool partially_consumed = chunk > len;
bpf: Fix running sk_skb program types with ktls KTLS uses a stream parser to collect TLS messages and send them to the upper layer tls receive handler. This ensures the tls receiver has a full TLS header to parse when it is run. However, when a socket has BPF_SK_SKB_STREAM_VERDICT program attached before KTLS is enabled we end up with two stream parsers running on the same socket. The result is both try to run on the same socket. First the KTLS stream parser runs and calls read_sock() which will tcp_read_sock which in turn calls tcp_rcv_skb(). This dequeues the skb from the sk_receive_queue. When this is done KTLS code then data_ready() callback which because we stacked KTLS on top of the bpf stream verdict program has been replaced with sk_psock_start_strp(). This will in turn kick the stream parser again and eventually do the same thing KTLS did above calling into tcp_rcv_skb() and dequeuing a skb from the sk_receive_queue. At this point the data stream is broke. Part of the stream was handled by the KTLS side some other bytes may have been handled by the BPF side. Generally this results in either missing data or more likely a "Bad Message" complaint from the kTLS receive handler as the BPF program steals some bytes meant to be in a TLS header and/or the TLS header length is no longer correct. We've already broke the idealized model where we can stack ULPs in any order with generic callbacks on the TX side to handle this. So in this patch we do the same thing but for RX side. We add a sk_psock_strp_enabled() helper so TLS can learn a BPF verdict program is running and add a tls_sw_has_ctx_rx() helper so BPF side can learn there is a TLS ULP on the socket. Then on BPF side we omit calling our stream parser to avoid breaking the data stream for the KTLS receiver. Then on the KTLS side we call BPF_SK_SKB_STREAM_VERDICT once the KTLS receiver is done with the packet but before it posts the msg to userspace. This gives us symmetry between the TX and RX halfs and IMO makes it usable again. On the TX side we process packets in this order BPF -> TLS -> TCP and on the receive side in the reverse order TCP -> TLS -> BPF. Discovered while testing OpenSSL 3.0 Alpha2.0 release. Fixes: d829e9c4112b5 ("tls: convert to generic sk_msg interface") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Link: https://lore.kernel.org/bpf/159079361946.5745.605854335665044485.stgit@john-Precision-5820-Tower Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2020-05-29 23:06:59 +00:00
if (bpf_strp_enabled) {
/* BPF may try to queue the skb */
__skb_unlink(skb, &ctx->rx_list);
bpf: Fix running sk_skb program types with ktls KTLS uses a stream parser to collect TLS messages and send them to the upper layer tls receive handler. This ensures the tls receiver has a full TLS header to parse when it is run. However, when a socket has BPF_SK_SKB_STREAM_VERDICT program attached before KTLS is enabled we end up with two stream parsers running on the same socket. The result is both try to run on the same socket. First the KTLS stream parser runs and calls read_sock() which will tcp_read_sock which in turn calls tcp_rcv_skb(). This dequeues the skb from the sk_receive_queue. When this is done KTLS code then data_ready() callback which because we stacked KTLS on top of the bpf stream verdict program has been replaced with sk_psock_start_strp(). This will in turn kick the stream parser again and eventually do the same thing KTLS did above calling into tcp_rcv_skb() and dequeuing a skb from the sk_receive_queue. At this point the data stream is broke. Part of the stream was handled by the KTLS side some other bytes may have been handled by the BPF side. Generally this results in either missing data or more likely a "Bad Message" complaint from the kTLS receive handler as the BPF program steals some bytes meant to be in a TLS header and/or the TLS header length is no longer correct. We've already broke the idealized model where we can stack ULPs in any order with generic callbacks on the TX side to handle this. So in this patch we do the same thing but for RX side. We add a sk_psock_strp_enabled() helper so TLS can learn a BPF verdict program is running and add a tls_sw_has_ctx_rx() helper so BPF side can learn there is a TLS ULP on the socket. Then on BPF side we omit calling our stream parser to avoid breaking the data stream for the KTLS receiver. Then on the KTLS side we call BPF_SK_SKB_STREAM_VERDICT once the KTLS receiver is done with the packet but before it posts the msg to userspace. This gives us symmetry between the TX and RX halfs and IMO makes it usable again. On the TX side we process packets in this order BPF -> TLS -> TCP and on the receive side in the reverse order TCP -> TLS -> BPF. Discovered while testing OpenSSL 3.0 Alpha2.0 release. Fixes: d829e9c4112b5 ("tls: convert to generic sk_msg interface") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Link: https://lore.kernel.org/bpf/159079361946.5745.605854335665044485.stgit@john-Precision-5820-Tower Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2020-05-29 23:06:59 +00:00
err = sk_psock_tls_strp_read(psock, skb);
if (err != __SK_PASS) {
rxm->offset = rxm->offset + rxm->full_len;
rxm->full_len = 0;
if (err == __SK_DROP)
consume_skb(skb);
continue;
}
__skb_queue_tail(&ctx->rx_list, skb);
bpf: Fix running sk_skb program types with ktls KTLS uses a stream parser to collect TLS messages and send them to the upper layer tls receive handler. This ensures the tls receiver has a full TLS header to parse when it is run. However, when a socket has BPF_SK_SKB_STREAM_VERDICT program attached before KTLS is enabled we end up with two stream parsers running on the same socket. The result is both try to run on the same socket. First the KTLS stream parser runs and calls read_sock() which will tcp_read_sock which in turn calls tcp_rcv_skb(). This dequeues the skb from the sk_receive_queue. When this is done KTLS code then data_ready() callback which because we stacked KTLS on top of the bpf stream verdict program has been replaced with sk_psock_start_strp(). This will in turn kick the stream parser again and eventually do the same thing KTLS did above calling into tcp_rcv_skb() and dequeuing a skb from the sk_receive_queue. At this point the data stream is broke. Part of the stream was handled by the KTLS side some other bytes may have been handled by the BPF side. Generally this results in either missing data or more likely a "Bad Message" complaint from the kTLS receive handler as the BPF program steals some bytes meant to be in a TLS header and/or the TLS header length is no longer correct. We've already broke the idealized model where we can stack ULPs in any order with generic callbacks on the TX side to handle this. So in this patch we do the same thing but for RX side. We add a sk_psock_strp_enabled() helper so TLS can learn a BPF verdict program is running and add a tls_sw_has_ctx_rx() helper so BPF side can learn there is a TLS ULP on the socket. Then on BPF side we omit calling our stream parser to avoid breaking the data stream for the KTLS receiver. Then on the KTLS side we call BPF_SK_SKB_STREAM_VERDICT once the KTLS receiver is done with the packet but before it posts the msg to userspace. This gives us symmetry between the TX and RX halfs and IMO makes it usable again. On the TX side we process packets in this order BPF -> TLS -> TCP and on the receive side in the reverse order TCP -> TLS -> BPF. Discovered while testing OpenSSL 3.0 Alpha2.0 release. Fixes: d829e9c4112b5 ("tls: convert to generic sk_msg interface") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Link: https://lore.kernel.org/bpf/159079361946.5745.605854335665044485.stgit@john-Precision-5820-Tower Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2020-05-29 23:06:59 +00:00
}
if (partially_consumed)
chunk = len;
err = skb_copy_datagram_msg(skb, rxm->offset,
msg, chunk);
if (err < 0)
goto recv_end;
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
if (is_peek)
goto leave_on_list;
if (partially_consumed) {
rxm->offset += chunk;
rxm->full_len -= chunk;
goto leave_on_list;
}
}
decrypted += chunk;
len -= chunk;
__skb_unlink(skb, &ctx->rx_list);
consume_skb(skb);
/* Return full control message to userspace before trying
* to parse another message type
*/
msg->msg_flags |= MSG_EOR;
if (control != TLS_RECORD_TYPE_DATA)
break;
}
recv_end:
if (async) {
int ret, pending;
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
/* Wait for all previously submitted records to be decrypted */
2020-05-22 20:10:31 +00:00
spin_lock_bh(&ctx->decrypt_compl_lock);
reinit_completion(&ctx->async_wait.completion);
2020-05-22 20:10:31 +00:00
pending = atomic_read(&ctx->decrypt_pending);
spin_unlock_bh(&ctx->decrypt_compl_lock);
if (pending) {
ret = crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
if (ret) {
if (err >= 0 || err == -EINPROGRESS)
err = ret;
decrypted = 0;
goto end;
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
}
}
2020-05-22 20:10:31 +00:00
/* Drain records from the rx_list & copy if required */
if (is_peek || is_kvec)
err = process_rx_list(ctx, msg, &control, copied,
decrypted, false, is_peek);
else
err = process_rx_list(ctx, msg, &control, 0,
decrypted, true, is_peek);
decrypted = max(err, 0);
net/tls: Add support for async decryption of tls records When tls records are decrypted using asynchronous acclerators such as NXP CAAM engine, the crypto apis return -EINPROGRESS. Presently, on getting -EINPROGRESS, the tls record processing stops till the time the crypto accelerator finishes off and returns the result. This incurs a context switch and is not an efficient way of accessing the crypto accelerators. Crypto accelerators work efficient when they are queued with multiple crypto jobs without having to wait for the previous ones to complete. The patch submits multiple crypto requests without having to wait for for previous ones to complete. This has been implemented for records which are decrypted in zero-copy mode. At the end of recvmsg(), we wait for all the asynchronous decryption requests to complete. The references to records which have been sent for async decryption are dropped. For cases where record decryption is not possible in zero-copy mode, asynchronous decryption is not used and we wait for decryption crypto api to complete. For crypto requests executing in async fashion, the memory for aead_request, sglists and skb etc is freed from the decryption completion handler. The decryption completion handler wakesup the sleeping user context when recvmsg() flags that it has done sending all the decryption requests and there are no more decryption requests pending to be completed. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Reviewed-by: Dave Watson <davejwatson@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-08-29 09:56:55 +00:00
}
copied += decrypted;
end:
release_sock(sk);
if (psock)
sk_psock_put(sk, psock);
return copied ? : err;
}
ssize_t tls_sw_splice_read(struct socket *sock, loff_t *ppos,
struct pipe_inode_info *pipe,
size_t len, unsigned int flags)
{
struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
struct strp_msg *rxm = NULL;
struct sock *sk = sock->sk;
struct tls_msg *tlm;
struct sk_buff *skb;
ssize_t copied = 0;
bool from_queue;
int err = 0;
long timeo;
int chunk;
lock_sock(sk);
timeo = sock_rcvtimeo(sk, flags & SPLICE_F_NONBLOCK);
from_queue = !skb_queue_empty(&ctx->rx_list);
if (from_queue) {
skb = __skb_dequeue(&ctx->rx_list);
} else {
struct tls_decrypt_arg darg = {};
skb = tls_wait_data(sk, NULL, flags & SPLICE_F_NONBLOCK, timeo,
&err);
if (!skb)
goto splice_read_end;
err = decrypt_skb_update(sk, skb, NULL, &darg);
if (err < 0) {
tls_err_abort(sk, -EBADMSG);
goto splice_read_end;
}
}
rxm = strp_msg(skb);
tlm = tls_msg(skb);
/* splice does not support reading control messages */
if (tlm->control != TLS_RECORD_TYPE_DATA) {
err = -EINVAL;
goto splice_read_end;
}
chunk = min_t(unsigned int, rxm->full_len, len);
copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
if (copied < 0)
goto splice_read_end;
if (!from_queue) {
ctx->recv_pkt = NULL;
__strp_unpause(&ctx->strp);
}
if (chunk < rxm->full_len) {
__skb_queue_head(&ctx->rx_list, skb);
rxm->offset += len;
rxm->full_len -= len;
} else {
consume_skb(skb);
}
splice_read_end:
release_sock(sk);
return copied ? : err;
}
bool tls_sw_sock_is_readable(struct sock *sk)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
bool ingress_empty = true;
struct sk_psock *psock;
rcu_read_lock();
psock = sk_psock(sk);
if (psock)
ingress_empty = list_empty(&psock->ingress_msg);
rcu_read_unlock();
return !ingress_empty || ctx->recv_pkt ||
!skb_queue_empty(&ctx->rx_list);
}
static int tls_read_size(struct strparser *strp, struct sk_buff *skb)
{
struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
struct tls_prot_info *prot = &tls_ctx->prot_info;
char header[TLS_HEADER_SIZE + MAX_IV_SIZE];
struct strp_msg *rxm = strp_msg(skb);
struct tls_msg *tlm = tls_msg(skb);
size_t cipher_overhead;
size_t data_len = 0;
int ret;
/* Verify that we have a full TLS header, or wait for more data */
if (rxm->offset + prot->prepend_size > skb->len)
return 0;
/* Sanity-check size of on-stack buffer. */
if (WARN_ON(prot->prepend_size > sizeof(header))) {
ret = -EINVAL;
goto read_failure;
}
/* Linearize header to local buffer */
ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size);
if (ret < 0)
goto read_failure;
tlm->decrypted = 0;
tlm->control = header[0];
data_len = ((header[4] & 0xFF) | (header[3] << 8));
cipher_overhead = prot->tag_size;
if (prot->version != TLS_1_3_VERSION &&
prot->cipher_type != TLS_CIPHER_CHACHA20_POLY1305)
cipher_overhead += prot->iv_size;
if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
prot->tail_size) {
ret = -EMSGSIZE;
goto read_failure;
}
if (data_len < cipher_overhead) {
ret = -EBADMSG;
goto read_failure;
}
/* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
if (header[1] != TLS_1_2_VERSION_MINOR ||
header[2] != TLS_1_2_VERSION_MAJOR) {
ret = -EINVAL;
goto read_failure;
}
net/tls: add kernel-driven TLS RX resync TLS offload device may lose sync with the TCP stream if packets arrive out of order. Drivers can currently request a resync at a specific TCP sequence number. When a record is found starting at that sequence number kernel will inform the device of the corresponding record number. This requires the device to constantly scan the stream for a known pattern (constant bytes of the header) after sync is lost. This patch adds an alternative approach which is entirely under the control of the kernel. Kernel tracks records it had to fully decrypt, even though TLS socket is in TLS_HW mode. If multiple records did not have any decrypted parts - it's a pretty strong indication that the device is out of sync. We choose the min number of fully encrypted records to be 2, which should hopefully be more than will get retransmitted at a time. After kernel decides the device is out of sync it schedules a resync request. If the TCP socket is empty the resync gets performed immediately. If socket is not empty we leave the record parser to resync when next record comes. Before resync in message parser we peek at the TCP socket and don't attempt the sync if the socket already has some of the next record queued. On resync failure (encrypted data continues to flow in) we retry with exponential backoff, up to once every 128 records (with a 16k record thats at most once every 2M of data). Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com> Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-06-11 04:40:02 +00:00
tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE,
TCP_SKB_CB(skb)->seq + rxm->offset);
return data_len + TLS_HEADER_SIZE;
read_failure:
tls_err_abort(strp->sk, ret);
return ret;
}
static void tls_queue(struct strparser *strp, struct sk_buff *skb)
{
struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
ctx->recv_pkt = skb;
strp_pause(strp);
ctx->saved_data_ready(strp->sk);
}
static void tls_data_ready(struct sock *sk)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
struct sk_psock *psock;
strp_data_ready(&ctx->strp);
psock = sk_psock_get(sk);
if (psock) {
if (!list_empty(&psock->ingress_msg))
ctx->saved_data_ready(sk);
sk_psock_put(sk, psock);
}
}
void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
{
struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask);
set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask);
cancel_delayed_work_sync(&ctx->tx_work.work);
}
void tls_sw_release_resources_tx(struct sock *sk)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
struct tls_rec *rec, *tmp;
net/tls: race causes kernel panic BUG: kernel NULL pointer dereference, address: 00000000000000b8 #PF: supervisor read access in kernel mode #PF: error_code(0x0000) - not-present page PGD 80000008b6fef067 P4D 80000008b6fef067 PUD 8b6fe6067 PMD 0 Oops: 0000 [#1] SMP PTI CPU: 12 PID: 23871 Comm: kworker/12:80 Kdump: loaded Tainted: G S 5.9.0-rc3+ #1 Hardware name: Supermicro X10SRA-F/X10SRA-F, BIOS 2.1 03/29/2018 Workqueue: events tx_work_handler [tls] RIP: 0010:tx_work_handler+0x1b/0x70 [tls] Code: dc fe ff ff e8 16 d4 a3 f6 66 0f 1f 44 00 00 0f 1f 44 00 00 55 53 48 8b 6f 58 48 8b bd a0 04 00 00 48 85 ff 74 1c 48 8b 47 28 <48> 8b 90 b8 00 00 00 83 e2 02 75 0c f0 48 0f ba b0 b8 00 00 00 00 RSP: 0018:ffffa44ace61fe88 EFLAGS: 00010286 RAX: 0000000000000000 RBX: ffff91da9e45cc30 RCX: dead000000000122 RDX: 0000000000000001 RSI: ffff91da9e45cc38 RDI: ffff91d95efac200 RBP: ffff91da133fd780 R08: 0000000000000000 R09: 000073746e657665 R10: 8080808080808080 R11: 0000000000000000 R12: ffff91dad7d30700 R13: ffff91dab6561080 R14: 0ffff91dad7d3070 R15: ffff91da9e45cc38 FS: 0000000000000000(0000) GS:ffff91dad7d00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00000000000000b8 CR3: 0000000906478003 CR4: 00000000003706e0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Call Trace: process_one_work+0x1a7/0x370 worker_thread+0x30/0x370 ? process_one_work+0x370/0x370 kthread+0x114/0x130 ? kthread_park+0x80/0x80 ret_from_fork+0x22/0x30 tls_sw_release_resources_tx() waits for encrypt_pending, which can have race, so we need similar changes as in commit 0cada33241d9de205522e3858b18e506ca5cce2c here as well. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption of records for performance") Signed-off-by: Rohit Maheshwari <rohitm@chelsio.com> Acked-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2020-09-24 06:58:45 +00:00
int pending;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
/* Wait for any pending async encryptions to complete */
net/tls: race causes kernel panic BUG: kernel NULL pointer dereference, address: 00000000000000b8 #PF: supervisor read access in kernel mode #PF: error_code(0x0000) - not-present page PGD 80000008b6fef067 P4D 80000008b6fef067 PUD 8b6fe6067 PMD 0 Oops: 0000 [#1] SMP PTI CPU: 12 PID: 23871 Comm: kworker/12:80 Kdump: loaded Tainted: G S 5.9.0-rc3+ #1 Hardware name: Supermicro X10SRA-F/X10SRA-F, BIOS 2.1 03/29/2018 Workqueue: events tx_work_handler [tls] RIP: 0010:tx_work_handler+0x1b/0x70 [tls] Code: dc fe ff ff e8 16 d4 a3 f6 66 0f 1f 44 00 00 0f 1f 44 00 00 55 53 48 8b 6f 58 48 8b bd a0 04 00 00 48 85 ff 74 1c 48 8b 47 28 <48> 8b 90 b8 00 00 00 83 e2 02 75 0c f0 48 0f ba b0 b8 00 00 00 00 RSP: 0018:ffffa44ace61fe88 EFLAGS: 00010286 RAX: 0000000000000000 RBX: ffff91da9e45cc30 RCX: dead000000000122 RDX: 0000000000000001 RSI: ffff91da9e45cc38 RDI: ffff91d95efac200 RBP: ffff91da133fd780 R08: 0000000000000000 R09: 000073746e657665 R10: 8080808080808080 R11: 0000000000000000 R12: ffff91dad7d30700 R13: ffff91dab6561080 R14: 0ffff91dad7d3070 R15: ffff91da9e45cc38 FS: 0000000000000000(0000) GS:ffff91dad7d00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00000000000000b8 CR3: 0000000906478003 CR4: 00000000003706e0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Call Trace: process_one_work+0x1a7/0x370 worker_thread+0x30/0x370 ? process_one_work+0x370/0x370 kthread+0x114/0x130 ? kthread_park+0x80/0x80 ret_from_fork+0x22/0x30 tls_sw_release_resources_tx() waits for encrypt_pending, which can have race, so we need similar changes as in commit 0cada33241d9de205522e3858b18e506ca5cce2c here as well. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption of records for performance") Signed-off-by: Rohit Maheshwari <rohitm@chelsio.com> Acked-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2020-09-24 06:58:45 +00:00
spin_lock_bh(&ctx->encrypt_compl_lock);
ctx->async_notify = true;
pending = atomic_read(&ctx->encrypt_pending);
spin_unlock_bh(&ctx->encrypt_compl_lock);
if (pending)
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
tls_tx_records(sk, -1);
net/tls: Fixed race condition in async encryption On processors with multi-engine crypto accelerators, it is possible that multiple records get encrypted in parallel and their encryption completion is notified to different cpus in multicore processor. This leads to the situation where tls_encrypt_done() starts executing in parallel on different cores. In current implementation, encrypted records are queued to tx_ready_list in tls_encrypt_done(). This requires addition to linked list 'tx_ready_list' to be protected. As tls_decrypt_done() could be executing in irq content, it is not possible to protect linked list addition operation using a lock. To fix the problem, we remove linked list addition operation from the irq context. We do tx_ready_list addition/removal operation from application context only and get rid of possible multiple access to the linked list. Before starting encryption on the record, we add it to the tail of tx_ready_list. To prevent tls_tx_records() from transmitting it, we mark the record with a new flag 'tx_ready' in 'struct tls_rec'. When record encryption gets completed, tls_encrypt_done() has to only update the 'tx_ready' flag to true & linked list add operation is not required. The changed logic brings some other side benefits. Since the records are always submitted in tls sequence number order for encryption, the tx_ready_list always remains sorted and addition of new records to it does not have to traverse the linked list. Lastly, we renamed tx_ready_list in 'struct tls_sw_context_tx' to 'tx_list'. This is because now, the some of the records at the tail are not ready to transmit. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption") Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-24 10:05:56 +00:00
/* Free up un-sent records in tx_list. First, free
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
* the partially sent record if any at head of tx_list.
*/
if (tls_ctx->partially_sent_record) {
tls_free_partial_record(sk, tls_ctx);
net/tls: Fixed race condition in async encryption On processors with multi-engine crypto accelerators, it is possible that multiple records get encrypted in parallel and their encryption completion is notified to different cpus in multicore processor. This leads to the situation where tls_encrypt_done() starts executing in parallel on different cores. In current implementation, encrypted records are queued to tx_ready_list in tls_encrypt_done(). This requires addition to linked list 'tx_ready_list' to be protected. As tls_decrypt_done() could be executing in irq content, it is not possible to protect linked list addition operation using a lock. To fix the problem, we remove linked list addition operation from the irq context. We do tx_ready_list addition/removal operation from application context only and get rid of possible multiple access to the linked list. Before starting encryption on the record, we add it to the tail of tx_ready_list. To prevent tls_tx_records() from transmitting it, we mark the record with a new flag 'tx_ready' in 'struct tls_rec'. When record encryption gets completed, tls_encrypt_done() has to only update the 'tx_ready' flag to true & linked list add operation is not required. The changed logic brings some other side benefits. Since the records are always submitted in tls sequence number order for encryption, the tx_ready_list always remains sorted and addition of new records to it does not have to traverse the linked list. Lastly, we renamed tx_ready_list in 'struct tls_sw_context_tx' to 'tx_list'. This is because now, the some of the records at the tail are not ready to transmit. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption") Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-24 10:05:56 +00:00
rec = list_first_entry(&ctx->tx_list,
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
struct tls_rec, list);
list_del(&rec->list);
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
sk_msg_free(sk, &rec->msg_plaintext);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
kfree(rec);
}
net/tls: Fixed race condition in async encryption On processors with multi-engine crypto accelerators, it is possible that multiple records get encrypted in parallel and their encryption completion is notified to different cpus in multicore processor. This leads to the situation where tls_encrypt_done() starts executing in parallel on different cores. In current implementation, encrypted records are queued to tx_ready_list in tls_encrypt_done(). This requires addition to linked list 'tx_ready_list' to be protected. As tls_decrypt_done() could be executing in irq content, it is not possible to protect linked list addition operation using a lock. To fix the problem, we remove linked list addition operation from the irq context. We do tx_ready_list addition/removal operation from application context only and get rid of possible multiple access to the linked list. Before starting encryption on the record, we add it to the tail of tx_ready_list. To prevent tls_tx_records() from transmitting it, we mark the record with a new flag 'tx_ready' in 'struct tls_rec'. When record encryption gets completed, tls_encrypt_done() has to only update the 'tx_ready' flag to true & linked list add operation is not required. The changed logic brings some other side benefits. Since the records are always submitted in tls sequence number order for encryption, the tx_ready_list always remains sorted and addition of new records to it does not have to traverse the linked list. Lastly, we renamed tx_ready_list in 'struct tls_sw_context_tx' to 'tx_list'. This is because now, the some of the records at the tail are not ready to transmit. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption") Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-24 10:05:56 +00:00
list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
list_del(&rec->list);
tls: convert to generic sk_msg interface Convert kTLS over to make use of sk_msg interface for plaintext and encrypted scattergather data, so it reuses all the sk_msg helpers and data structure which later on in a second step enables to glue this to BPF. This also allows to remove quite a bit of open coded helpers which are covered by the sk_msg API. Recent changes in kTLs 80ece6a03aaf ("tls: Remove redundant vars from tls record structure") and 4e6d47206c32 ("tls: Add support for inplace records encryption") changed the data path handling a bit; while we've kept the latter optimization intact, we had to undo the former change to better fit the sk_msg model, hence the sg_aead_in and sg_aead_out have been brought back and are linked into the sk_msg sgs. Now the kTLS record contains a msg_plaintext and msg_encrypted sk_msg each. In the original code, the zerocopy_from_iter() has been used out of TX but also RX path. For the strparser skb-based RX path, we've left the zerocopy_from_iter() in decrypt_internal() mostly untouched, meaning it has been moved into tls_setup_from_iter() with charging logic removed (as not used from RX). Given RX path is not based on sk_msg objects, we haven't pursued setting up a dummy sk_msg to call into sk_msg_zerocopy_from_iter(), but it could be an option to prusue in a later step. Joint work with John. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-10-13 00:45:59 +00:00
sk_msg_free(sk, &rec->msg_encrypted);
sk_msg_free(sk, &rec->msg_plaintext);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
kfree(rec);
}
crypto_free_aead(ctx->aead_send);
tls_free_open_rec(sk);
}
void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
{
struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
kfree(ctx);
}
void tls_sw_release_resources_rx(struct sock *sk)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
kfree(tls_ctx->rx.rec_seq);
kfree(tls_ctx->rx.iv);
if (ctx->aead_recv) {
kfree_skb(ctx->recv_pkt);
ctx->recv_pkt = NULL;
__skb_queue_purge(&ctx->rx_list);
crypto_free_aead(ctx->aead_recv);
strp_stop(&ctx->strp);
/* If tls_sw_strparser_arm() was not called (cleanup paths)
* we still want to strp_stop(), but sk->sk_data_ready was
* never swapped.
*/
if (ctx->saved_data_ready) {
write_lock_bh(&sk->sk_callback_lock);
sk->sk_data_ready = ctx->saved_data_ready;
write_unlock_bh(&sk->sk_callback_lock);
}
}
}
void tls_sw_strparser_done(struct tls_context *tls_ctx)
{
struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
strp_done(&ctx->strp);
}
void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
{
struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
kfree(ctx);
}
void tls_sw_free_resources_rx(struct sock *sk)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
tls_sw_release_resources_rx(sk);
tls_sw_free_ctx_rx(tls_ctx);
}
net/tls: Fixed race condition in async encryption On processors with multi-engine crypto accelerators, it is possible that multiple records get encrypted in parallel and their encryption completion is notified to different cpus in multicore processor. This leads to the situation where tls_encrypt_done() starts executing in parallel on different cores. In current implementation, encrypted records are queued to tx_ready_list in tls_encrypt_done(). This requires addition to linked list 'tx_ready_list' to be protected. As tls_decrypt_done() could be executing in irq content, it is not possible to protect linked list addition operation using a lock. To fix the problem, we remove linked list addition operation from the irq context. We do tx_ready_list addition/removal operation from application context only and get rid of possible multiple access to the linked list. Before starting encryption on the record, we add it to the tail of tx_ready_list. To prevent tls_tx_records() from transmitting it, we mark the record with a new flag 'tx_ready' in 'struct tls_rec'. When record encryption gets completed, tls_encrypt_done() has to only update the 'tx_ready' flag to true & linked list add operation is not required. The changed logic brings some other side benefits. Since the records are always submitted in tls sequence number order for encryption, the tx_ready_list always remains sorted and addition of new records to it does not have to traverse the linked list. Lastly, we renamed tx_ready_list in 'struct tls_sw_context_tx' to 'tx_list'. This is because now, the some of the records at the tail are not ready to transmit. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption") Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-24 10:05:56 +00:00
/* The work handler to transmitt the encrypted records in tx_list */
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
static void tx_work_handler(struct work_struct *work)
{
struct delayed_work *delayed_work = to_delayed_work(work);
struct tx_work *tx_work = container_of(delayed_work,
struct tx_work, work);
struct sock *sk = tx_work->sk;
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_sw_context_tx *ctx;
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
if (unlikely(!tls_ctx))
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
return;
ctx = tls_sw_ctx_tx(tls_ctx);
if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
return;
if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
return;
net/tls: add a TX lock TLS TX needs to release and re-acquire the socket lock if send buffer fills up. TLS SW TX path currently depends on only allowing one thread to enter the function by the abuse of sk_write_pending. If another writer is already waiting for memory no new ones are allowed in. This has two problems: - writers don't wake other threads up when they leave the kernel; meaning that this scheme works for single extra thread (second application thread or delayed work) because memory becoming available will send a wake up request, but as Mallesham and Pooja report with larger number of threads it leads to threads being put to sleep indefinitely; - the delayed work does not get _scheduled_ but it may _run_ when other writers are present leading to crashes as writers don't expect state to change under their feet (same records get pushed and freed multiple times); it's hard to reliably bail from the work, however, because the mere presence of a writer does not guarantee that the writer will push pending records before exiting. Ensuring wakeups always happen will make the code basically open code a mutex. Just use a mutex. The TLS HW TX path does not have any locking (not even the sk_write_pending hack), yet it uses a per-socket sg_tx_data array to push records. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption of records for performance") Reported-by: Mallesham Jatharakonda <mallesh537@gmail.com> Reported-by: Pooja Trivedi <poojatrivedi@gmail.com> Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com> Reviewed-by: Simon Horman <simon.horman@netronome.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-11-05 22:24:35 +00:00
mutex_lock(&tls_ctx->tx_lock);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
lock_sock(sk);
tls_tx_records(sk, -1);
release_sock(sk);
net/tls: add a TX lock TLS TX needs to release and re-acquire the socket lock if send buffer fills up. TLS SW TX path currently depends on only allowing one thread to enter the function by the abuse of sk_write_pending. If another writer is already waiting for memory no new ones are allowed in. This has two problems: - writers don't wake other threads up when they leave the kernel; meaning that this scheme works for single extra thread (second application thread or delayed work) because memory becoming available will send a wake up request, but as Mallesham and Pooja report with larger number of threads it leads to threads being put to sleep indefinitely; - the delayed work does not get _scheduled_ but it may _run_ when other writers are present leading to crashes as writers don't expect state to change under their feet (same records get pushed and freed multiple times); it's hard to reliably bail from the work, however, because the mere presence of a writer does not guarantee that the writer will push pending records before exiting. Ensuring wakeups always happen will make the code basically open code a mutex. Just use a mutex. The TLS HW TX path does not have any locking (not even the sk_write_pending hack), yet it uses a per-socket sg_tx_data array to push records. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption of records for performance") Reported-by: Mallesham Jatharakonda <mallesh537@gmail.com> Reported-by: Pooja Trivedi <poojatrivedi@gmail.com> Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com> Reviewed-by: Simon Horman <simon.horman@netronome.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-11-05 22:24:35 +00:00
mutex_unlock(&tls_ctx->tx_lock);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
}
static bool tls_is_tx_ready(struct tls_sw_context_tx *ctx)
{
struct tls_rec *rec;
rec = list_first_entry(&ctx->tx_list, struct tls_rec, list);
if (!rec)
return false;
return READ_ONCE(rec->tx_ready);
}
void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
{
struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
/* Schedule the transmission if tx list is ready */
if (tls_is_tx_ready(tx_ctx) &&
!test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
schedule_delayed_work(&tx_ctx->tx_work.work, 0);
}
void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
{
struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
write_lock_bh(&sk->sk_callback_lock);
rx_ctx->saved_data_ready = sk->sk_data_ready;
sk->sk_data_ready = tls_data_ready;
write_unlock_bh(&sk->sk_callback_lock);
strp_check_rcv(&rx_ctx->strp);
}
void tls_update_rx_zc_capable(struct tls_context *tls_ctx)
{
struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
rx_ctx->zc_capable = tls_ctx->rx_no_pad ||
tls_ctx->prot_info.version != TLS_1_3_VERSION;
}
int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
{
struct tls_context *tls_ctx = tls_get_ctx(sk);
struct tls_prot_info *prot = &tls_ctx->prot_info;
struct tls_crypto_info *crypto_info;
struct tls_sw_context_tx *sw_ctx_tx = NULL;
struct tls_sw_context_rx *sw_ctx_rx = NULL;
struct cipher_context *cctx;
struct crypto_aead **aead;
struct strp_callbacks cb;
u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
struct crypto_tfm *tfm;
char *iv, *rec_seq, *key, *salt, *cipher_name;
size_t keysize;
int rc = 0;
if (!ctx) {
rc = -EINVAL;
goto out;
}
if (tx) {
if (!ctx->priv_ctx_tx) {
sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
if (!sw_ctx_tx) {
rc = -ENOMEM;
goto out;
}
ctx->priv_ctx_tx = sw_ctx_tx;
} else {
sw_ctx_tx =
(struct tls_sw_context_tx *)ctx->priv_ctx_tx;
}
} else {
if (!ctx->priv_ctx_rx) {
sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
if (!sw_ctx_rx) {
rc = -ENOMEM;
goto out;
}
ctx->priv_ctx_rx = sw_ctx_rx;
} else {
sw_ctx_rx =
(struct tls_sw_context_rx *)ctx->priv_ctx_rx;
}
}
if (tx) {
crypto_init_wait(&sw_ctx_tx->async_wait);
2020-05-22 20:10:31 +00:00
spin_lock_init(&sw_ctx_tx->encrypt_compl_lock);
crypto_info = &ctx->crypto_send.info;
cctx = &ctx->tx;
aead = &sw_ctx_tx->aead_send;
net/tls: Fixed race condition in async encryption On processors with multi-engine crypto accelerators, it is possible that multiple records get encrypted in parallel and their encryption completion is notified to different cpus in multicore processor. This leads to the situation where tls_encrypt_done() starts executing in parallel on different cores. In current implementation, encrypted records are queued to tx_ready_list in tls_encrypt_done(). This requires addition to linked list 'tx_ready_list' to be protected. As tls_decrypt_done() could be executing in irq content, it is not possible to protect linked list addition operation using a lock. To fix the problem, we remove linked list addition operation from the irq context. We do tx_ready_list addition/removal operation from application context only and get rid of possible multiple access to the linked list. Before starting encryption on the record, we add it to the tail of tx_ready_list. To prevent tls_tx_records() from transmitting it, we mark the record with a new flag 'tx_ready' in 'struct tls_rec'. When record encryption gets completed, tls_encrypt_done() has to only update the 'tx_ready' flag to true & linked list add operation is not required. The changed logic brings some other side benefits. Since the records are always submitted in tls sequence number order for encryption, the tx_ready_list always remains sorted and addition of new records to it does not have to traverse the linked list. Lastly, we renamed tx_ready_list in 'struct tls_sw_context_tx' to 'tx_list'. This is because now, the some of the records at the tail are not ready to transmit. Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption") Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-24 10:05:56 +00:00
INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
net/tls: Add support for async encryption of records for performance In current implementation, tls records are encrypted & transmitted serially. Till the time the previously submitted user data is encrypted, the implementation waits and on finish starts transmitting the record. This approach of encrypt-one record at a time is inefficient when asynchronous crypto accelerators are used. For each record, there are overheads of interrupts, driver softIRQ scheduling etc. Also the crypto accelerator sits idle most of time while an encrypted record's pages are handed over to tcp stack for transmission. This patch enables encryption of multiple records in parallel when an async capable crypto accelerator is present in system. This is achieved by allowing the user space application to send more data using sendmsg() even while previously issued data is being processed by crypto accelerator. This requires returning the control back to user space application after submitting encryption request to accelerator. This also means that zero-copy mode of encryption cannot be used with async accelerator as we must be done with user space application buffer before returning from sendmsg(). There can be multiple records in flight to/from the accelerator. Each of the record is represented by 'struct tls_rec'. This is used to store the memory pages for the record. After the records are encrypted, they are added in a linked list called tx_ready_list which contains encrypted tls records sorted as per tls sequence number. The records from tx_ready_list are transmitted using a newly introduced function called tls_tx_records(). The tx_ready_list is polled for any record ready to be transmitted in sendmsg(), sendpage() after initiating encryption of new tls records. This achieves parallel encryption and transmission of records when async accelerator is present. There could be situation when crypto accelerator completes encryption later than polling of tx_ready_list by sendmsg()/sendpage(). Therefore we need a deferred work context to be able to transmit records from tx_ready_list. The deferred work context gets scheduled if applications are not sending much data through the socket. If the applications issue sendmsg()/sendpage() in quick succession, then the scheduling of tx_work_handler gets cancelled as the tx_ready_list would be polled from application's context itself. This saves scheduling overhead of deferred work. The patch also brings some side benefit. We are able to get rid of the concept of CLOSED record. This is because the records once closed are either encrypted and then placed into tx_ready_list or if encryption fails, the socket error is set. This simplifies the kernel tls sendpath. However since tls_device.c is still using macros, accessory functions for CLOSED records have been retained. Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-21 04:16:13 +00:00
INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
sw_ctx_tx->tx_work.sk = sk;
} else {
crypto_init_wait(&sw_ctx_rx->async_wait);
2020-05-22 20:10:31 +00:00
spin_lock_init(&sw_ctx_rx->decrypt_compl_lock);
crypto_info = &ctx->crypto_recv.info;
cctx = &ctx->rx;
skb_queue_head_init(&sw_ctx_rx->rx_list);
aead = &sw_ctx_rx->aead_recv;
}
switch (crypto_info->cipher_type) {
case TLS_CIPHER_AES_GCM_128: {
struct tls12_crypto_info_aes_gcm_128 *gcm_128_info;
gcm_128_info = (void *)crypto_info;
nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
iv = gcm_128_info->iv;
rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
rec_seq = gcm_128_info->rec_seq;
keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE;
key = gcm_128_info->key;
salt = gcm_128_info->salt;
salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
cipher_name = "gcm(aes)";
break;
}
case TLS_CIPHER_AES_GCM_256: {
struct tls12_crypto_info_aes_gcm_256 *gcm_256_info;
gcm_256_info = (void *)crypto_info;
nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE;
iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
iv = gcm_256_info->iv;
rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE;
rec_seq = gcm_256_info->rec_seq;
keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE;
key = gcm_256_info->key;
salt = gcm_256_info->salt;
salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE;
cipher_name = "gcm(aes)";
break;
}
case TLS_CIPHER_AES_CCM_128: {
struct tls12_crypto_info_aes_ccm_128 *ccm_128_info;
ccm_128_info = (void *)crypto_info;
nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE;
iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
iv = ccm_128_info->iv;
rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE;
rec_seq = ccm_128_info->rec_seq;
keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE;
key = ccm_128_info->key;
salt = ccm_128_info->salt;
salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE;
cipher_name = "ccm(aes)";
break;
}
case TLS_CIPHER_CHACHA20_POLY1305: {
struct tls12_crypto_info_chacha20_poly1305 *chacha20_poly1305_info;
chacha20_poly1305_info = (void *)crypto_info;
nonce_size = 0;
tag_size = TLS_CIPHER_CHACHA20_POLY1305_TAG_SIZE;
iv_size = TLS_CIPHER_CHACHA20_POLY1305_IV_SIZE;
iv = chacha20_poly1305_info->iv;
rec_seq_size = TLS_CIPHER_CHACHA20_POLY1305_REC_SEQ_SIZE;
rec_seq = chacha20_poly1305_info->rec_seq;
keysize = TLS_CIPHER_CHACHA20_POLY1305_KEY_SIZE;
key = chacha20_poly1305_info->key;
salt = chacha20_poly1305_info->salt;
salt_size = TLS_CIPHER_CHACHA20_POLY1305_SALT_SIZE;
cipher_name = "rfc7539(chacha20,poly1305)";
break;
}
case TLS_CIPHER_SM4_GCM: {
struct tls12_crypto_info_sm4_gcm *sm4_gcm_info;
sm4_gcm_info = (void *)crypto_info;
nonce_size = TLS_CIPHER_SM4_GCM_IV_SIZE;
tag_size = TLS_CIPHER_SM4_GCM_TAG_SIZE;
iv_size = TLS_CIPHER_SM4_GCM_IV_SIZE;
iv = sm4_gcm_info->iv;
rec_seq_size = TLS_CIPHER_SM4_GCM_REC_SEQ_SIZE;
rec_seq = sm4_gcm_info->rec_seq;
keysize = TLS_CIPHER_SM4_GCM_KEY_SIZE;
key = sm4_gcm_info->key;
salt = sm4_gcm_info->salt;
salt_size = TLS_CIPHER_SM4_GCM_SALT_SIZE;
cipher_name = "gcm(sm4)";
break;
}
case TLS_CIPHER_SM4_CCM: {
struct tls12_crypto_info_sm4_ccm *sm4_ccm_info;
sm4_ccm_info = (void *)crypto_info;
nonce_size = TLS_CIPHER_SM4_CCM_IV_SIZE;
tag_size = TLS_CIPHER_SM4_CCM_TAG_SIZE;
iv_size = TLS_CIPHER_SM4_CCM_IV_SIZE;
iv = sm4_ccm_info->iv;
rec_seq_size = TLS_CIPHER_SM4_CCM_REC_SEQ_SIZE;
rec_seq = sm4_ccm_info->rec_seq;
keysize = TLS_CIPHER_SM4_CCM_KEY_SIZE;
key = sm4_ccm_info->key;
salt = sm4_ccm_info->salt;
salt_size = TLS_CIPHER_SM4_CCM_SALT_SIZE;
cipher_name = "ccm(sm4)";
break;
}
default:
rc = -EINVAL;
goto free_priv;
}
if (crypto_info->version == TLS_1_3_VERSION) {
nonce_size = 0;
prot->aad_size = TLS_HEADER_SIZE;
prot->tail_size = 1;
} else {
prot->aad_size = TLS_AAD_SPACE_SIZE;
prot->tail_size = 0;
}
/* Sanity-check the sizes for stack allocations. */
if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE ||
rec_seq_size > TLS_MAX_REC_SEQ_SIZE || tag_size != TLS_TAG_SIZE ||
prot->aad_size > TLS_MAX_AAD_SIZE) {
rc = -EINVAL;
goto free_priv;
}
prot->version = crypto_info->version;
prot->cipher_type = crypto_info->cipher_type;
prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
prot->tag_size = tag_size;
prot->overhead_size = prot->prepend_size +
prot->tag_size + prot->tail_size;
prot->iv_size = iv_size;
prot->salt_size = salt_size;
cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL);
if (!cctx->iv) {
rc = -ENOMEM;
goto free_priv;
}
/* Note: 128 & 256 bit salt are the same size */
prot->rec_seq_size = rec_seq_size;
memcpy(cctx->iv, salt, salt_size);
memcpy(cctx->iv + salt_size, iv, iv_size);
cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
if (!cctx->rec_seq) {
rc = -ENOMEM;
goto free_iv;
}
if (!*aead) {
*aead = crypto_alloc_aead(cipher_name, 0, 0);
if (IS_ERR(*aead)) {
rc = PTR_ERR(*aead);
*aead = NULL;
goto free_rec_seq;
}
}
ctx->push_pending_record = tls_sw_push_pending_record;
rc = crypto_aead_setkey(*aead, key, keysize);
if (rc)
goto free_aead;
rc = crypto_aead_setauthsize(*aead, prot->tag_size);
if (rc)
goto free_aead;
if (sw_ctx_rx) {
tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
tls_update_rx_zc_capable(ctx);
sw_ctx_rx->async_capable =
crypto_info->version != TLS_1_3_VERSION &&
!!(tfm->__crt_alg->cra_flags & CRYPTO_ALG_ASYNC);
/* Set up strparser */
memset(&cb, 0, sizeof(cb));
cb.rcv_msg = tls_queue;
cb.parse_msg = tls_read_size;
strp_init(&sw_ctx_rx->strp, sk, &cb);
}
goto out;
free_aead:
crypto_free_aead(*aead);
*aead = NULL;
free_rec_seq:
kfree(cctx->rec_seq);
cctx->rec_seq = NULL;
free_iv:
kfree(cctx->iv);
cctx->iv = NULL;
free_priv:
if (tx) {
kfree(ctx->priv_ctx_tx);
ctx->priv_ctx_tx = NULL;
} else {
kfree(ctx->priv_ctx_rx);
ctx->priv_ctx_rx = NULL;
}
out:
return rc;
}