Pull random number generator updates from Jason Donenfeld:
"These updates continue to refine the work began in 5.17 and 5.18 of
modernizing the RNG's crypto and streamlining and documenting its
code.
New for 5.19, the updates aim to improve entropy collection methods
and make some initial decisions regarding the "premature next" problem
and our threat model. The cloc utility now reports that random.c is
931 lines of code and 466 lines of comments, not that basic metrics
like that mean all that much, but at the very least it tells you that
this is very much a manageable driver now.
Here's a summary of the various updates:
- The random_get_entropy() function now always returns something at
least minimally useful. This is the primary entropy source in most
collectors, which in the best case expands to something like RDTSC,
but prior to this change, in the worst case it would just return 0,
contributing nothing. For 5.19, additional architectures are wired
up, and architectures that are entirely missing a cycle counter now
have a generic fallback path, which uses the highest resolution
clock available from the timekeeping subsystem.
Some of those clocks can actually be quite good, despite the CPU
not having a cycle counter of its own, and going off-core for a
stamp is generally thought to increase jitter, something positive
from the perspective of entropy gathering. Done very early on in
the development cycle, this has been sitting in next getting some
testing for a while now and has relevant acks from the archs, so it
should be pretty well tested and fine, but is nonetheless the thing
I'll be keeping my eye on most closely.
- Of particular note with the random_get_entropy() improvements is
MIPS, which, on CPUs that lack the c0 count register, will now
combine the high-speed but short-cycle c0 random register with the
lower-speed but long-cycle generic fallback path.
- With random_get_entropy() now always returning something useful,
the interrupt handler now collects entropy in a consistent
construction.
- Rather than comparing two samples of random_get_entropy() for the
jitter dance, the algorithm now tests many samples, and uses the
amount of differing ones to determine whether or not jitter entropy
is usable and how laborious it must be. The problem with comparing
only two samples was that if the cycle counter was extremely slow,
but just so happened to be on the cusp of a change, the slowness
wouldn't be detected. Taking many samples fixes that to some
degree.
This, combined with the other improvements to random_get_entropy(),
should make future unification of /dev/random and /dev/urandom
maybe more possible. At the very least, were we to attempt it again
today (we're not), it wouldn't break any of Guenter's test rigs
that broke when we tried it with 5.18. So, not today, but perhaps
down the road, that's something we can revisit.
- We attempt to reseed the RNG immediately upon waking up from system
suspend or hibernation, making use of the various timestamps about
suspend time and such available, as well as the usual inputs such
as RDRAND when available.
- Batched randomness now falls back to ordinary randomness before the
RNG is initialized. This provides more consistent guarantees to the
types of random numbers being returned by the various accessors.
- The "pre-init injection" code is now gone for good. I suspect you
in particular will be happy to read that, as I recall you
expressing your distaste for it a few months ago. Instead, to avoid
a "premature first" issue, while still allowing for maximal amount
of entropy availability during system boot, the first 128 bits of
estimated entropy are used immediately as it arrives, with the next
128 bits being buffered. And, as before, after the RNG has been
fully initialized, it winds up reseeding anyway a few seconds later
in most cases. This resulted in a pretty big simplification of the
initialization code and let us remove various ad-hoc mechanisms
like the ugly crng_pre_init_inject().
- The RNG no longer pretends to handle the "premature next" security
model, something that various academics and other RNG designs have
tried to care about in the past. After an interesting mailing list
thread, these issues are thought to be a) mainly academic and not
practical at all, and b) actively harming the real security of the
RNG by delaying new entropy additions after a potential compromise,
making a potentially bad situation even worse. As well, in the
first place, our RNG never even properly handled the premature next
issue, so removing an incomplete solution to a fake problem was
particularly nice.
This allowed for numerous other simplifications in the code, which
is a lot cleaner as a consequence. If you didn't see it before,
https://lore.kernel.org/lkml/YmlMGx6+uigkGiZ0@zx2c4.com/ may be a
thread worth skimming through.
- While the interrupt handler received a separate code path years ago
that avoids locks by using per-cpu data structures and a faster
mixing algorithm, in order to reduce interrupt latency, input and
disk events that are triggered in hardirq handlers were still
hitting locks and more expensive algorithms. Those are now
redirected to use the faster per-cpu data structures.
- Rather than having the fake-crypto almost-siphash-based random32
implementation be used right and left, and in many places where
cryptographically secure randomness is desirable, the batched
entropy code is now fast enough to replace that.
- As usual, numerous code quality and documentation cleanups. For
example, the initialization state machine now uses enum symbolic
constants instead of just hard coding numbers everywhere.
- Since the RNG initializes once, and then is always initialized
thereafter, a pretty heavy amount of code used during that
initialization is never used again. It is now completely cordoned
off using static branches and it winds up in the .text.unlikely
section so that it doesn't reduce cache compactness after the RNG
is ready.
- A variety of functions meant for waiting on the RNG to be
initialized were only used by vsprintf, and in not a particularly
optimal way. Replacing that usage with a more ordinary setup made
it possible to remove those functions.
- A cleanup of how we warn userspace about the use of uninitialized
/dev/urandom and uninitialized get_random_bytes() usage.
Interestingly, with the change you merged for 5.18 that attempts to
use jitter (but does not block if it can't), the majority of users
should never see those warnings for /dev/urandom at all now, and
the one for in-kernel usage is mainly a debug thing.
- The file_operations struct for /dev/[u]random now implements
.read_iter and .write_iter instead of .read and .write, allowing it
to also implement .splice_read and .splice_write, which makes
splice(2) work again after it was broken here (and in many other
places in the tree) during the set_fs() removal. This was a bit of
a last minute arrival from Jens that hasn't had as much time to
bake, so I'll be keeping my eye on this as well, but it seems
fairly ordinary. Unfortunately, read_iter() is around 3% slower
than read() in my tests, which I'm not thrilled about. But Jens and
Al, spurred by this observation, seem to be making progress in
removing the bottlenecks on the iter paths in the VFS layer in
general, which should remove the performance gap for all drivers.
- Assorted other bug fixes, cleanups, and optimizations.
- A small SipHash cleanup"
* tag 'random-5.19-rc1-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/crng/random: (49 commits)
random: check for signals after page of pool writes
random: wire up fops->splice_{read,write}_iter()
random: convert to using fops->write_iter()
random: convert to using fops->read_iter()
random: unify batched entropy implementations
random: move randomize_page() into mm where it belongs
random: remove mostly unused async readiness notifier
random: remove get_random_bytes_arch() and add rng_has_arch_random()
random: move initialization functions out of hot pages
random: make consistent use of buf and len
random: use proper return types on get_random_{int,long}_wait()
random: remove extern from functions in header
random: use static branch for crng_ready()
random: credit architectural init the exact amount
random: handle latent entropy and command line from random_init()
random: use proper jiffies comparison macro
random: remove ratelimiting for in-kernel unseeded randomness
random: move initialization out of reseeding hot path
random: avoid initializing twice in credit race
random: use symbolic constants for crng_init states
...
Pull io_uring 'more data in socket' support from Jens Axboe:
"To be able to fully utilize the 'poll first' support in the core
io_uring branch, it's advantageous knowing if the socket was empty
after a receive. This adds support for that"
* tag 'for-5.19/io_uring-net-2022-05-22' of git://git.kernel.dk/linux-block:
io_uring: return hint on whether more data is available after receive
tcp: pass back data left in socket after receive
random32.c has two random number generators in it: one that is meant to
be used deterministically, with some predefined seed, and one that does
the same exact thing as random.c, except does it poorly. The first one
has some use cases. The second one no longer does and can be replaced
with calls to random.c's proper random number generator.
The relatively recent siphash-based bad random32.c code was added in
response to concerns that the prior random32.c was too deterministic.
Out of fears that random.c was (at the time) too slow, this code was
anonymously contributed. Then out of that emerged a kind of shadow
entropy gathering system, with its own tentacles throughout various net
code, added willy nilly.
Stop👏making👏bespoke👏random👏number👏generators👏.
Fortunately, recent advances in random.c mean that we can stop playing
with this sketchiness, and just use get_random_u32(), which is now fast
enough. In micro benchmarks using RDPMC, I'm seeing the same median
cycle count between the two functions, with the mean being _slightly_
higher due to batches refilling (which we can optimize further need be).
However, when doing *real* benchmarks of the net functions that actually
use these random numbers, the mean cycles actually *decreased* slightly
(with the median still staying the same), likely because the additional
prandom code means icache misses and complexity, whereas random.c is
generally already being used by something else nearby.
The biggest benefit of this is that there are many users of prandom who
probably should be using cryptographically secure random numbers. This
makes all of those accidental cases become secure by just flipping a
switch. Later on, we can do a tree-wide cleanup to remove the static
inline wrapper functions that this commit adds.
There are also some low-ish hanging fruits for making this even faster
in the future: a get_random_u16() function for use in the networking
stack will give a 2x performance boost there, using SIMD for ChaCha20
will let us compute 4 or 8 or 16 blocks of output in parallel, instead
of just one, giving us large buffers for cheap, and introducing a
get_random_*_bh() function that assumes irqs are already disabled will
shave off a few cycles for ordinary calls. These are things we can chip
away at down the road.
Acked-by: Jakub Kicinski <kuba@kernel.org>
Acked-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Steffen Klassert says:
====================
pull request (net): ipsec 2022-05-18
1) Fix "disable_policy" flag use when arriving from different devices.
From Eyal Birger.
2) Fix error handling of pfkey_broadcast in function pfkey_process.
From Jiasheng Jiang.
3) Check the encryption module availability consistency in pfkey.
From Thomas Bartschies.
Please pull or let me know if there are problems.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
In IPv4 setting the "disable_policy" flag on a device means no policy
should be enforced for traffic originating from the device. This was
implemented by seting the DST_NOPOLICY flag in the dst based on the
originating device.
However, dsts are cached in nexthops regardless of the originating
devices, in which case, the DST_NOPOLICY flag value may be incorrect.
Consider the following setup:
+------------------------------+
| ROUTER |
+-------------+ | +-----------------+ |
| ipsec src |----|-|ipsec0 | |
+-------------+ | |disable_policy=0 | +----+ |
| +-----------------+ |eth1|-|-----
+-------------+ | +-----------------+ +----+ |
| noipsec src |----|-|eth0 | |
+-------------+ | |disable_policy=1 | |
| +-----------------+ |
+------------------------------+
Where ROUTER has a default route towards eth1.
dst entries for traffic arriving from eth0 would have DST_NOPOLICY
and would be cached and therefore can be reused by traffic originating
from ipsec0, skipping policy check.
Fix by setting a IPSKB_NOPOLICY flag in IPCB and observing it instead
of the DST in IN/FWD IPv4 policy checks.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Reported-by: Shmulik Ladkani <shmulik.ladkani@gmail.com>
Signed-off-by: Eyal Birger <eyal.birger@gmail.com>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
This reverts commits:
0dad4087a8 ("tcp/dccp: get rid of inet_twsk_purge()")
d507204d3c ("tcp/dccp: add tw->tw_bslot")
As Leonard pointed out, a newly allocated netns can happen
to reuse a freed 'struct net'.
While TCP TW timers were covered by my patches, other things were not:
1) Lookups in rx path (INET_MATCH() and INET6_MATCH()), as they look
at 4-tuple plus the 'struct net' pointer.
2) /proc/net/tcp[6] and inet_diag, same reason.
3) hashinfo->bhash[], same reason.
Fixing all this seems risky, lets instead revert.
In the future, we might have a per netns tcp hash table, or
a per netns list of timewait sockets...
Fixes: 0dad4087a8 ("tcp/dccp: get rid of inet_twsk_purge()")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Leonard Crestez <cdleonard@gmail.com>
Tested-by: Leonard Crestez <cdleonard@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When ping_group_range is updated, 'ping' uses the DGRAM ICMP socket,
instead of an IP raw socket. In this case, 'ping' is unable to bind its
socket to a local address owned by a vrflite.
Before the patch:
$ sysctl -w net.ipv4.ping_group_range='0 2147483647'
$ ip link add blue type vrf table 10
$ ip link add foo type dummy
$ ip link set foo master blue
$ ip link set foo up
$ ip addr add 192.168.1.1/24 dev foo
$ ip addr add 2001::1/64 dev foo
$ ip vrf exec blue ping -c1 -I 192.168.1.1 192.168.1.2
ping: bind: Cannot assign requested address
$ ip vrf exec blue ping6 -c1 -I 2001::1 2001::2
ping6: bind icmp socket: Cannot assign requested address
CC: stable@vger.kernel.org
Fixes: 1b69c6d0ae ("net: Introduce L3 Master device abstraction")
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In commit 190cc82489 ("tcp: change source port randomizarion at
connect() time"), the table_perturb[] array was introduced and an
index was taken from the port_offset via hash_32(). But it turns
out that hash_32() performs a multiplication while the input here
comes from the output of SipHash in secure_seq, that is well
distributed enough to avoid the need for yet another hash.
Suggested-by: Amit Klein <aksecurity@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Moshe Kol, Amit Klein, and Yossi Gilad reported being able to accurately
identify a client by forcing it to emit only 40 times more connections
than there are entries in the table_perturb[] table. The previous two
improvements consisting in resalting the secret every 10s and adding
randomness to each port selection only slightly improved the situation,
and the current value of 2^8 was too small as it's not very difficult
to make a client emit 10k connections in less than 10 seconds.
Thus we're increasing the perturb table from 2^8 to 2^16 so that the
same precision now requires 2.6M connections, which is more difficult in
this time frame and harder to hide as a background activity. The impact
is that the table now uses 256 kB instead of 1 kB, which could mostly
affect devices making frequent outgoing connections. However such
components usually target a small set of destinations (load balancers,
database clients, perf assessment tools), and in practice only a few
entries will be visited, like before.
A live test at 1 million connections per second showed no performance
difference from the previous value.
Reported-by: Moshe Kol <moshe.kol@mail.huji.ac.il>
Reported-by: Yossi Gilad <yossi.gilad@mail.huji.ac.il>
Reported-by: Amit Klein <aksecurity@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We'll need to further increase the size of this table and it's likely
that at some point its size will not be suitable anymore for a static
table. Let's allocate it on boot from inet_hashinfo2_init(), which is
called from tcp_init().
Cc: Moshe Kol <moshe.kol@mail.huji.ac.il>
Cc: Yossi Gilad <yossi.gilad@mail.huji.ac.il>
Cc: Amit Klein <aksecurity@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Here we're randomly adding between 0 and 7 random increments to the
selected source port in order to add some noise in the source port
selection that will make the next port less predictable.
With the default port range of 32768-60999 this means a worst case
reuse scenario of 14116/8=1764 connections between two consecutive
uses of the same port, with an average of 14116/4.5=3137. This code
was stressed at more than 800000 connections per second to a fixed
target with all connections closed by the client using RSTs (worst
condition) and only 2 connections failed among 13 billion, despite
the hash being reseeded every 10 seconds, indicating a perfectly
safe situation.
Cc: Moshe Kol <moshe.kol@mail.huji.ac.il>
Cc: Yossi Gilad <yossi.gilad@mail.huji.ac.il>
Cc: Amit Klein <aksecurity@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
SipHash replaced MD5 in secure_ipv{4,6}_port_ephemeral() via commit
7cd23e5300 ("secure_seq: use SipHash in place of MD5"), but the output
remained truncated to 32-bit only. In order to exploit more bits from the
hash, let's make the functions return the full 64-bit of siphash_3u32().
We also make sure the port offset calculation in __inet_hash_connect()
remains done on 32-bit to avoid the need for div_u64_rem() and an extra
cost on 32-bit systems.
Cc: Jason A. Donenfeld <Jason@zx2c4.com>
Cc: Moshe Kol <moshe.kol@mail.huji.ac.il>
Cc: Yossi Gilad <yossi.gilad@mail.huji.ac.il>
Cc: Amit Klein <aksecurity@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Merge net branch with the required patch for supporting the io_uring
feature that passes back whether we had more data in the socket or not.
* 'tcp-pass-back-data-left-in-socket-after-receive' of git://git.kernel.org/pub/scm/linux/kernel/git/kuba/linux:
tcp: pass back data left in socket after receive
This is currently done for CMSG_INQ, add an ability to do so via struct
msghdr as well and have CMSG_INQ use that too. If the caller sets
msghdr->msg_get_inq, then we'll pass back the hint in msghdr->msg_inq.
Rearrange struct msghdr a bit so we can add this member while shrinking
it at the same time. On a 64-bit build, it was 96 bytes before this
change and 88 bytes afterwards.
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/650c22ca-cffc-0255-9a05-2413a1e20826@kernel.dk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
I had this bug sitting for too long in my pile, it is time to fix it.
Thanks to Doug Porter for reminding me of it!
We had various attempts in the past, including commit
0cbe6a8f08 ("tcp: remove SOCK_QUEUE_SHRUNK"),
but the issue is that TCP stack currently only generates
EPOLLOUT from input path, when tp->snd_una has advanced
and skb(s) cleaned from rtx queue.
If a flow has a big RTT, and/or receives SACKs, it is possible
that the notsent part (tp->write_seq - tp->snd_nxt) reaches 0
and no more data can be sent until tp->snd_una finally advances.
What is needed is to also check if POLLOUT needs to be generated
whenever tp->snd_nxt is advanced, from output path.
This bug triggers more often after an idle period, as
we do not receive ACK for at least one RTT. tcp_notsent_lowat
could be a fraction of what CWND and pacing rate would allow to
send during this RTT.
In a followup patch, I will remove the bogus call
to tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED)
from tcp_check_space(). Fact that we have decided to generate
an EPOLLOUT does not mean the application has immediately
refilled the transmit queue. This optimistic call
might have been the reason the bug seemed not too serious.
Tested:
200 ms rtt, 1% packet loss, 32 MB tcp_rmem[2] and tcp_wmem[2]
$ echo 500000 >/proc/sys/net/ipv4/tcp_notsent_lowat
$ cat bench_rr.sh
SUM=0
for i in {1..10}
do
V=`netperf -H remote_host -l30 -t TCP_RR -- -r 10000000,10000 -o LOCAL_BYTES_SENT | egrep -v "MIGRATED|Bytes"`
echo $V
SUM=$(($SUM + $V))
done
echo SUM=$SUM
Before patch:
$ bench_rr.sh
130000000
80000000
140000000
140000000
140000000
140000000
130000000
40000000
90000000
110000000
SUM=1140000000
After patch:
$ bench_rr.sh
430000000
590000000
530000000
450000000
450000000
350000000
450000000
490000000
480000000
460000000
SUM=4680000000 # This is 410 % of the value before patch.
Fixes: c9bee3b7fd ("tcp: TCP_NOTSENT_LOWAT socket option")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Doug Porter <dsp@fb.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As pointed out by Jakub Kicinski, currently using TUNNEL_SEQ in
collect_md mode is racy for [IP6]GRE[TAP] devices. Consider the
following sequence of events:
1. An [IP6]GRE[TAP] device is created in collect_md mode using "ip link
add ... external". "ip" ignores "[o]seq" if "external" is specified,
so TUNNEL_SEQ is off, and the device is marked as NETIF_F_LLTX (i.e.
it uses lockless TX);
2. Someone sets TUNNEL_SEQ on outgoing skb's, using e.g.
bpf_skb_set_tunnel_key() in an eBPF program attached to this device;
3. gre_fb_xmit() or __gre6_xmit() processes these skb's:
gre_build_header(skb, tun_hlen,
flags, protocol,
tunnel_id_to_key32(tun_info->key.tun_id),
(flags & TUNNEL_SEQ) ? htonl(tunnel->o_seqno++)
: 0); ^^^^^^^^^^^^^^^^^
Since we are not using the TX lock (&txq->_xmit_lock), multiple CPUs may
try to do this tunnel->o_seqno++ in parallel, which is racy. Fix it by
making o_seqno atomic_t.
As mentioned by Eric Dumazet in commit b790e01aee ("ip_gre: lockless
xmit"), making o_seqno atomic_t increases "chance for packets being out
of order at receiver" when NETIF_F_LLTX is on.
Maybe a better fix would be:
1. Do not ignore "oseq" in external mode. Users MUST specify "oseq" if
they want the kernel to allow sequencing of outgoing packets;
2. Reject all outgoing TUNNEL_SEQ packets if the device was not created
with "oseq".
Unfortunately, that would break userspace.
We could now make [IP6]GRE[TAP] devices always NETIF_F_LLTX, but let us
do it in separate patches to keep this fix minimal.
Suggested-by: Jakub Kicinski <kuba@kernel.org>
Fixes: 77a5196a80 ("gre: add sequence number for collect md mode.")
Signed-off-by: Peilin Ye <peilin.ye@bytedance.com>
Acked-by: William Tu <u9012063@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For GRE and GRETAP devices, currently o_seqno starts from 1 in native
mode. According to RFC 2890 2.2., "The first datagram is sent with a
sequence number of 0." Fix it.
It is worth mentioning that o_seqno already starts from 0 in collect_md
mode, see gre_fb_xmit(), where tunnel->o_seqno is passed to
gre_build_header() before getting incremented.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Peilin Ye <peilin.ye@bytedance.com>
Acked-by: William Tu <u9012063@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pablo Neira Ayuso says:
====================
Netfilter fixes for net
The following patchset contains Netfilter fixes for net:
1) Fix incorrect printing of memory size of IPVS connection hash table,
from Pengcheng Yang.
2) Fix spurious EEXIST errors in nft_set_rbtree.
3) Remove leftover empty flowtable file, from Rongguang Wei.
4) Fix ip6_route_me_harder() with vrf driver, from Martin Willi.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
CONFIG_NF_FLOW_TABLE_IPV4 is already removed and the real user is also
removed(nf_flow_table_ipv4.c is empty).
Fixes: c42ba4290b ("netfilter: flowtable: remove ipv4/ipv6 modules")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
If an ACK (s)acks multiple skbs, we favor the information
from the most recently sent skb by choosing the skb with
the highest prior_delivered count. But in the interval
between receiving ACKs, we send multiple skbs with the same
prior_delivered, because the tp->delivered only changes
when we receive an ACK.
We used RACK's solution, copying tcp_rack_sent_after() as
tcp_skb_sent_after() helper to determine "which packet was
sent last?". Later, we will use tcp_skb_sent_after() instead
in RACK.
Fixes: b9f64820fb ("tcp: track data delivery rate for a TCP connection")
Signed-off-by: Pengcheng Yang <yangpc@wangsu.com>
Cc: Paolo Abeni <pabeni@redhat.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Tested-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/1650422081-22153-1-git-send-email-yangpc@wangsu.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Ido reported that the commit referenced in the Fixes tag broke
a gre use case with dummy devices. Add a check to ip_tunnel_init_flow
to see if the oif is an l3mdev port and if so set the oif to 0 to
avoid the oif comparison in fib_lookup_good_nhc.
Fixes: 40867d74c3 ("net: Add l3mdev index to flow struct and avoid oif reset for port devices")
Reported-by: Ido Schimmel <idosch@idosch.org>
Signed-off-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit ebe48d368e ("esp: Fix possible buffer overflow in ESP
transformation") tried to fix skb_page_frag_refill usage in ESP by
capping allocsize to 32k, but that doesn't completely solve the issue,
as skb_page_frag_refill may return a single page. If that happens, we
will write out of bounds, despite the check introduced in the previous
patch.
This patch forces COW in cases where we would end up calling
skb_page_frag_refill with a size larger than a page (first in
esp_output_head with tailen, then in esp_output_tail with
skb->data_len).
Fixes: cac2661c53 ("esp4: Avoid skb_cow_data whenever possible")
Fixes: 03e2a30f6a ("esp6: Avoid skb_cow_data whenever possible")
Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Alexei Starovoitov says:
====================
pull-request: bpf-next 2022-03-21 v2
We've added 137 non-merge commits during the last 17 day(s) which contain
a total of 143 files changed, 7123 insertions(+), 1092 deletions(-).
The main changes are:
1) Custom SEC() handling in libbpf, from Andrii.
2) subskeleton support, from Delyan.
3) Use btf_tag to recognize __percpu pointers in the verifier, from Hao.
4) Fix net.core.bpf_jit_harden race, from Hou.
5) Fix bpf_sk_lookup remote_port on big-endian, from Jakub.
6) Introduce fprobe (multi kprobe) _without_ arch bits, from Masami.
The arch specific bits will come later.
7) Introduce multi_kprobe bpf programs on top of fprobe, from Jiri.
8) Enable non-atomic allocations in local storage, from Joanne.
9) Various var_off ptr_to_btf_id fixed, from Kumar.
10) bpf_ima_file_hash helper, from Roberto.
11) Add "live packet" mode for XDP in BPF_PROG_RUN, from Toke.
* https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next: (137 commits)
selftests/bpf: Fix kprobe_multi test.
Revert "rethook: x86: Add rethook x86 implementation"
Revert "arm64: rethook: Add arm64 rethook implementation"
Revert "powerpc: Add rethook support"
Revert "ARM: rethook: Add rethook arm implementation"
bpftool: Fix a bug in subskeleton code generation
bpf: Fix bpf_prog_pack when PMU_SIZE is not defined
bpf: Fix bpf_prog_pack for multi-node setup
bpf: Fix warning for cast from restricted gfp_t in verifier
bpf, arm: Fix various typos in comments
libbpf: Close fd in bpf_object__reuse_map
bpftool: Fix print error when show bpf map
bpf: Fix kprobe_multi return probe backtrace
Revert "bpf: Add support to inline bpf_get_func_ip helper on x86"
bpf: Simplify check in btf_parse_hdr()
selftests/bpf/test_lirc_mode2.sh: Exit with proper code
bpf: Check for NULL return from bpf_get_btf_vmlinux
selftests/bpf: Test skipping stacktrace
bpf: Adjust BPF stack helper functions to accommodate skip > 0
bpf: Select proper size for bpf_prog_pack
...
====================
Link: https://lore.kernel.org/r/20220322050159.5507-1-alexei.starovoitov@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tp->rx_opt.mss_clamp is not populated, yet, during TFO send so we
rise it to the local MSS. tp->mss_cache is not updated, however:
tcp_v6_connect():
tp->rx_opt.mss_clamp = IPV6_MIN_MTU - headers;
tcp_connect():
tcp_connect_init():
tp->mss_cache = min(mtu, tp->rx_opt.mss_clamp)
tcp_send_syn_data():
tp->rx_opt.mss_clamp = tp->advmss
After recent fixes to ICMPv6 PTB handling we started dropping
PMTU updates higher than tp->mss_cache. Because of the stale
tp->mss_cache value PMTU updates during TFO are always dropped.
Thanks to Wei for helping zero in on the problem and the fix!
Fixes: c7bb4b8903 ("ipv6: tcp: drop silly ICMPv6 packet too big messages")
Reported-by: Andre Nash <alnash@fb.com>
Reported-by: Neil Spring <ntspring@fb.com>
Reviewed-by: Wei Wang <weiwan@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20220321165957.1769954-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Eliminate anonymous module_init() and module_exit(), which can lead to
confusion or ambiguity when reading System.map, crashes/oops/bugs,
or an initcall_debug log.
Give each of these init and exit functions unique driver-specific
names to eliminate the anonymous names.
Example 1: (System.map)
ffffffff832fc78c t init
ffffffff832fc79e t init
ffffffff832fc8f8 t init
Example 2: (initcall_debug log)
calling init+0x0/0x12 @ 1
initcall init+0x0/0x12 returned 0 after 15 usecs
calling init+0x0/0x60 @ 1
initcall init+0x0/0x60 returned 0 after 2 usecs
calling init+0x0/0x9a @ 1
initcall init+0x0/0x9a returned 0 after 74 usecs
Fixes: f587de0e2f ("[NETFILTER]: nf_conntrack/nf_nat: add H.323 helper port")
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Acked-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
The fib expression stores to a register, so we can't add empty stub.
Check that the register that is being written is in fact redundant.
In most cases, this is expected to cancel tracking as re-use is
unlikely.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Skip register tracking for expressions that perform read-only operations
on the registers. Define and use a cookie pointer NFT_REDUCE_READONLY to
avoid defining stubs for these expressions.
This patch re-enables register tracking which was disabled in ed5f85d422
("netfilter: nf_tables: disable register tracking"). Follow up patches
add remaining register tracking for existing expressions.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
The PMTU update and ICMP redirect helper functions initialise their fl4
variable with either __build_flow_key() or build_sk_flow_key(). These
initialisation functions always set ->flowi4_scope with
RT_SCOPE_UNIVERSE and might set the ECN bits of ->flowi4_tos. This is
not a problem when the route lookup is later done via
ip_route_output_key_hash(), which properly clears the ECN bits from
->flowi4_tos and initialises ->flowi4_scope based on the RTO_ONLINK
flag. However, some helpers call fib_lookup() directly, without
sanitising the tos and scope fields, so the route lookup can fail and,
as a result, the ICMP redirect or PMTU update aren't taken into
account.
Fix this by extracting the ->flowi4_tos and ->flowi4_scope sanitisation
code into ip_rt_fix_tos(), then use this function in handlers that call
fib_lookup() directly.
Note 1: We can't sanitise ->flowi4_tos and ->flowi4_scope in a central
place (like __build_flow_key() or flowi4_init_output()), because
ip_route_output_key_hash() expects non-sanitised values. When called
with sanitised values, it can erroneously overwrite RT_SCOPE_LINK with
RT_SCOPE_UNIVERSE in ->flowi4_scope. Therefore we have to be careful to
sanitise the values only for those paths that don't call
ip_route_output_key_hash().
Note 2: The problem is mostly about sanitising ->flowi4_tos. Having
->flowi4_scope initialised with RT_SCOPE_UNIVERSE instead of
RT_SCOPE_LINK probably wasn't really a problem: sockets with the
SOCK_LOCALROUTE flag set (those that'd result in RTO_ONLINK being set)
normally shouldn't receive ICMP redirects or PMTU updates.
Fixes: 4895c771c7 ("ipv4: Add FIB nexthop exceptions.")
Signed-off-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The fundamental premise of VRF and l3mdev core code is binding a socket
to a device (l3mdev or netdev with an L3 domain) to indicate L3 scope.
Legacy code resets flowi_oif to the l3mdev losing any original port
device binding. Ben (among others) has demonstrated use cases where the
original port device binding is important and needs to be retained.
This patch handles that by adding a new entry to the common flow struct
that can indicate the l3mdev index for later rule and table matching
avoiding the need to reset flowi_oif.
In addition to allowing more use cases that require port device binds,
this patch brings a few datapath simplications:
1. l3mdev_fib_rule_match is only called when walking fib rules and
always after l3mdev_update_flow. That allows an optimization to bail
early for non-VRF type uses cases when flowi_l3mdev is not set. Also,
only that index needs to be checked for the FIB table id.
2. l3mdev_update_flow can be called with flowi_oif set to a l3mdev
(e.g., VRF) device. By resetting flowi_oif only for this case the
FLOWI_FLAG_SKIP_NH_OIF flag is not longer needed and can be removed,
removing several checks in the datapath. The flowi_iif path can be
simplified to only be called if the it is not loopback (loopback can
not be assigned to an L3 domain) and the l3mdev index is not already
set.
3. Avoid another device lookup in the output path when the fib lookup
returns a reject failure.
Note: 2 functional tests for local traffic with reject fib rules are
updated to reflect the new direct failure at FIB lookup time for ping
rather than the failure on packet path. The current code fails like this:
HINT: Fails since address on vrf device is out of device scope
COMMAND: ip netns exec ns-A ping -c1 -w1 -I eth1 172.16.3.1
ping: Warning: source address might be selected on device other than: eth1
PING 172.16.3.1 (172.16.3.1) from 172.16.3.1 eth1: 56(84) bytes of data.
--- 172.16.3.1 ping statistics ---
1 packets transmitted, 0 received, 100% packet loss, time 0ms
where the test now directly fails:
HINT: Fails since address on vrf device is out of device scope
COMMAND: ip netns exec ns-A ping -c1 -w1 -I eth1 172.16.3.1
ping: connect: No route to host
Signed-off-by: David Ahern <dsahern@kernel.org>
Tested-by: Ben Greear <greearb@candelatech.com>
Link: https://lore.kernel.org/r/20220314204551.16369-1-dsahern@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
If tcp_bpf_sendmsg is running during a tear down operation, psock may be
freed.
tcp_bpf_sendmsg()
tcp_bpf_send_verdict()
sk_msg_return()
tcp_bpf_sendmsg_redir()
unlikely(!psock))
sk_msg_free()
The mem of msg has been uncharged in tcp_bpf_send_verdict() by
sk_msg_return(), and would be uncharged by sk_msg_free() again. When psock
is null, we can simply returning an error code, this would then trigger
the sk_msg_free_nocharge in the error path of __SK_REDIRECT and would have
the side effect of throwing an error up to user space. This would be a
slight change in behavior from user side but would look the same as an
error if the redirect on the socket threw an error.
This issue can cause the following info:
WARNING: CPU: 0 PID: 2136 at net/ipv4/af_inet.c:155 inet_sock_destruct+0x13c/0x260
Call Trace:
<TASK>
__sk_destruct+0x24/0x1f0
sk_psock_destroy+0x19b/0x1c0
process_one_work+0x1b3/0x3c0
worker_thread+0x30/0x350
? process_one_work+0x3c0/0x3c0
kthread+0xe6/0x110
? kthread_complete_and_exit+0x20/0x20
ret_from_fork+0x22/0x30
</TASK>
Fixes: 604326b41a ("bpf, sockmap: convert to generic sk_msg interface")
Signed-off-by: Wang Yufen <wangyufen@huawei.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20220304081145.2037182-5-wangyufen@huawei.com
In tcp_bpf_send_verdict(), if msg has more data after
tcp_bpf_sendmsg_redir():
tcp_bpf_send_verdict()
tosend = msg->sg.size //msg->sg.size = 22220
case __SK_REDIRECT:
sk_msg_return() //uncharged msg->sg.size(22220) sk->sk_forward_alloc
tcp_bpf_sendmsg_redir() //after tcp_bpf_sendmsg_redir, msg->sg.size=11000
goto more_data;
tosend = msg->sg.size //msg->sg.size = 11000
case __SK_REDIRECT:
sk_msg_return() //uncharged msg->sg.size(11000) to sk->sk_forward_alloc
The msg->sg.size(11000) has been uncharged twice, to fix we can charge the
remaining msg->sg.size before goto more data.
This issue can cause the following info:
WARNING: CPU: 0 PID: 9860 at net/core/stream.c:208 sk_stream_kill_queues+0xd4/0x1a0
Call Trace:
<TASK>
inet_csk_destroy_sock+0x55/0x110
__tcp_close+0x279/0x470
tcp_close+0x1f/0x60
inet_release+0x3f/0x80
__sock_release+0x3d/0xb0
sock_close+0x11/0x20
__fput+0x92/0x250
task_work_run+0x6a/0xa0
do_exit+0x33b/0xb60
do_group_exit+0x2f/0xa0
get_signal+0xb6/0x950
arch_do_signal_or_restart+0xac/0x2a0
? vfs_write+0x237/0x290
exit_to_user_mode_prepare+0xa9/0x200
syscall_exit_to_user_mode+0x12/0x30
do_syscall_64+0x46/0x80
entry_SYSCALL_64_after_hwframe+0x44/0xae
</TASK>
WARNING: CPU: 0 PID: 2136 at net/ipv4/af_inet.c:155 inet_sock_destruct+0x13c/0x260
Call Trace:
<TASK>
__sk_destruct+0x24/0x1f0
sk_psock_destroy+0x19b/0x1c0
process_one_work+0x1b3/0x3c0
worker_thread+0x30/0x350
? process_one_work+0x3c0/0x3c0
kthread+0xe6/0x110
? kthread_complete_and_exit+0x20/0x20
ret_from_fork+0x22/0x30
</TASK>
Fixes: 604326b41a ("bpf, sockmap: convert to generic sk_msg interface")
Signed-off-by: Wang Yufen <wangyufen@huawei.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20220304081145.2037182-4-wangyufen@huawei.com
Back when tcp_tso_autosize() and TCP pacing were introduced,
our focus was really to reduce burst sizes for long distance
flows.
The simple heuristic of using sk_pacing_rate/1024 has worked
well, but can lead to too small packets for hosts in the same
rack/cluster, when thousands of flows compete for the bottleneck.
Neal Cardwell had the idea of making the TSO burst size
a function of both sk_pacing_rate and tcp_min_rtt()
Indeed, for local flows, sending bigger bursts is better
to reduce cpu costs, as occasional losses can be repaired
quite fast.
This patch is based on Neal Cardwell implementation
done more than two years ago.
bbr is adjusting max_pacing_rate based on measured bandwidth,
while cubic would over estimate max_pacing_rate.
/proc/sys/net/ipv4/tcp_tso_rtt_log can be used to tune or disable
this new feature, in logarithmic steps.
Tested:
100Gbit NIC, two hosts in the same rack, 4K MTU.
600 flows rate-limited to 20000000 bytes per second.
Before patch: (TSO sizes would be limited to 20000000/1024/4096 -> 4 segments per TSO)
~# echo 0 >/proc/sys/net/ipv4/tcp_tso_rtt_log
~# nstat -n;perf stat ./super_netperf 600 -H otrv6 -l 20 -- -K dctcp -q 20000000;nstat|egrep "TcpInSegs|TcpOutSegs|TcpRetransSegs|Delivered"
96005
Performance counter stats for './super_netperf 600 -H otrv6 -l 20 -- -K dctcp -q 20000000':
65,945.29 msec task-clock # 2.845 CPUs utilized
1,314,632 context-switches # 19935.279 M/sec
5,292 cpu-migrations # 80.249 M/sec
940,641 page-faults # 14264.023 M/sec
201,117,030,926 cycles # 3049769.216 GHz (83.45%)
17,699,435,405 stalled-cycles-frontend # 8.80% frontend cycles idle (83.48%)
136,584,015,071 stalled-cycles-backend # 67.91% backend cycles idle (83.44%)
53,809,530,436 instructions # 0.27 insn per cycle
# 2.54 stalled cycles per insn (83.36%)
9,062,315,523 branches # 137422329.563 M/sec (83.22%)
153,008,621 branch-misses # 1.69% of all branches (83.32%)
23.182970846 seconds time elapsed
TcpInSegs 15648792 0.0
TcpOutSegs 58659110 0.0 # Average of 3.7 4K segments per TSO packet
TcpExtTCPDelivered 58654791 0.0
TcpExtTCPDeliveredCE 19 0.0
After patch:
~# echo 9 >/proc/sys/net/ipv4/tcp_tso_rtt_log
~# nstat -n;perf stat ./super_netperf 600 -H otrv6 -l 20 -- -K dctcp -q 20000000;nstat|egrep "TcpInSegs|TcpOutSegs|TcpRetransSegs|Delivered"
96046
Performance counter stats for './super_netperf 600 -H otrv6 -l 20 -- -K dctcp -q 20000000':
48,982.58 msec task-clock # 2.104 CPUs utilized
186,014 context-switches # 3797.599 M/sec
3,109 cpu-migrations # 63.472 M/sec
941,180 page-faults # 19214.814 M/sec
153,459,763,868 cycles # 3132982.807 GHz (83.56%)
12,069,861,356 stalled-cycles-frontend # 7.87% frontend cycles idle (83.32%)
120,485,917,953 stalled-cycles-backend # 78.51% backend cycles idle (83.24%)
36,803,672,106 instructions # 0.24 insn per cycle
# 3.27 stalled cycles per insn (83.18%)
5,947,266,275 branches # 121417383.427 M/sec (83.64%)
87,984,616 branch-misses # 1.48% of all branches (83.43%)
23.281200256 seconds time elapsed
TcpInSegs 1434706 0.0
TcpOutSegs 58883378 0.0 # Average of 41 4K segments per TSO packet
TcpExtTCPDelivered 58878971 0.0
TcpExtTCPDeliveredCE 9664 0.0
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Link: https://lore.kernel.org/r/20220309015757.2532973-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tcp_should_autocork() is evaluating if it makes senses
to not immediately send current skb, hoping that
user space will add more payload on it by the
time TCP stack reacts to upcoming TX completions.
If current skb got MSG_EOR mark, then we know
that no further data will be added, it is therefore
futile to wait.
SOF_TIMESTAMPING_TX_ACK will become a bit more accurate,
if prior packets are still in qdisc/device queues.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Martin KaFai Lau <kafai@fb.com>
Cc: Willem de Bruijn <willemb@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Link: https://lore.kernel.org/r/20220309054706.2857266-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Steffen Klassert says:
====================
pull request (net): ipsec 2022-03-09
1) Fix IPv6 PMTU discovery for xfrm interfaces.
From Lina Wang.
2) Revert failing for policies and states that are
configured with XFRMA_IF_ID 0. It broke a
user configuration. From Kai Lueke.
3) Fix a possible buffer overflow in the ESP output path.
4) Fix ESP GSO for tunnel and BEET mode on inter address
family tunnels.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
We have a number of cases where function returns drop/no drop
decision as a boolean. Now that we want to report the reason
code as well we have to pass extra output arguments.
We can make the reason code evaluate correctly as bool.
I believe we're good to reorder the reasons as they are
reported to user space as strings.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The esp tunnel GSO handlers use skb_mac_gso_segment to
push the inner packet to the segmentation handlers.
However, skb_mac_gso_segment takes the Ethernet Protocol
ID from 'skb->protocol' which is wrong for inter address
family tunnels. We fix this by introducing a new
skb_eth_gso_segment function.
This function can be used if it is necessary to pass the
Ethernet Protocol ID directly to the segmentation handler.
First users of this function will be the esp4 and esp6
tunnel segmentation handlers.
Fixes: c35fe4106b ("xfrm: Add mode handlers for IPsec on layer 2")
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
The xfrm{4,6}_beet_gso_segment() functions did not correctly set the
SKB_GSO_IPXIP4 and SKB_GSO_IPXIP6 gso types for the address family
tunneling case. Fix this by setting these gso types.
Fixes: 384a46ea7b ("esp4: add gso_segment for esp4 beet mode")
Fixes: 7f9e40eb18 ("esp6: add gso_segment for esp6 beet mode")
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
The maximum message size that can be send is bigger than
the maximum site that skb_page_frag_refill can allocate.
So it is possible to write beyond the allocated buffer.
Fix this by doing a fallback to COW in that case.
v2:
Avoid get get_order() costs as suggested by Linus Torvalds.
Fixes: cac2661c53 ("esp4: Avoid skb_cow_data whenever possible")
Fixes: 03e2a30f6a ("esp6: Avoid skb_cow_data whenever possible")
Reported-by: valis <sec@valis.email>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
The previous patches handled the delivery_time in the ingress path
before the routing decision is made. This patch can postpone clearing
delivery_time in a skb until knowing it is delivered locally and also
set the (rcv) timestamp if needed. This patch moves the
skb_clear_delivery_time() from dev.c to ip_local_deliver_finish()
and ip6_input_finish().
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
