Because we miss to wipe the remainder of i->addr[] in packet_mc_add(),
pdiag_put_mclist() leaks uninitialized heap bytes via the
PACKET_DIAG_MCLIST netlink attribute.
Fix this by explicitly memset(0)ing the remaining bytes in i->addr[].
Fixes: eea68e2f1a ("packet: Report socket mclist info via diag module")
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Acked-by: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Trinity and other fuzzers can hit this WARN on far too easily,
resulting in a tainted kernel that hinders automated fuzzing.
Replace it with a rate-limited printk.
Signed-off-by: Dave Jones <davej@codemonkey.org.uk>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace link layer header validation check ll_header_truncate with
more generic dev_validate_header.
Validation based on hard_header_len incorrectly drops valid packets
in variable length protocols, such as AX25. dev_validate_header
calls header_ops.validate for such protocols to ensure correctness
below hard_header_len.
See also http://comments.gmane.org/gmane.linux.network/401064
Fixes 9c7077622d ("packet: make packet_snd fail on len smaller than l2 header")
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Support socket option PACKET_VNET_HDR together with PACKET_TX_RING.
When enabled, a struct virtio_net_hdr is expected to precede the data
in the ring. The vnet option must be set before the ring is created.
The implementation reuses the existing skb_copy_bits code that is used
when dev->hard_header_len is non-zero. Move this ll_header check to
before the skb alloc and combine it with a test for vnet_hdr->hdr_len.
Allocate and copy the max of the two.
Verified with test program at
github.com/wdebruij/kerneltools/blob/master/tests/psock_txring_vnet.c
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
GSO packet headers must be stored in the linear skb segment.
Move tpacket header parsing before sock_alloc_send_skb. The GSO
follow-on patch will later increase the skb linear argument to
sock_alloc_send_skb if needed for large packets.
The header parsing code does not require an allocated skb, so is
safe to move. Later pass to tpacket_fill_skb the computed data
start and length.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Support socket option PACKET_VNET_HDR together with PACKET_RX_RING.
When enabled, a struct virtio_net_hdr will precede the data in the
packet ring slots.
Verified with test program at
github.com/wdebruij/kerneltools/blob/master/tests/psock_rxring_vnet.c
pkt: 1454269209.798420 len=5066
vnet: gso_type=tcpv4 gso_size=1448 hlen=66 ecn=off
csum: start=34 off=16
eth: proto=0x800
ip: src=<masked> dst=<masked> proto=6 len=5052
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
packet_snd and packet_rcv support virtio net headers for GSO.
Move this logic into helper functions to be able to reuse it in
tpacket_snd and tpacket_rcv.
This is a straighforward code move with one exception. Instead of
creating and passing a separate gso_type variable, reuse
vnet_hdr.gso_type after conversion from virtio to kernel gso type.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 9c7077622d ("packet: make packet_snd fail on len smaller
than l2 header") added validation for the packet size in packet_snd.
This change enforces that every packet needs a header (with at least
hard_header_len bytes) plus a payload with at least one byte. Before
this change the payload was optional.
This fixes PPPoE connections which do not have a "Service" or
"Host-Uniq" configured (which is violating the spec, but is still
widely used in real-world setups). Those are currently failing with the
following message: "pppd: packet size is too short (24 <= 24)"
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use PAGE_ALIGNED(...) instead of open-coding it.
Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
rb->frames_per_block is an unsigned int, thus can never be negative.
Also fix spacing in the calculation of frames_per_block.
Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since it's introduction in commit 69e3c75f4d ("net: TX_RING and
packet mmap"), TX_RING could be used from SOCK_DGRAM and SOCK_RAW
side. When used with SOCK_DGRAM only, the size_max > dev->mtu +
reserve check should have reserve as 0, but currently, this is
unconditionally set (in it's original form as dev->hard_header_len).
I think this is not correct since tpacket_fill_skb() would then
take dev->mtu and dev->hard_header_len into account for SOCK_DGRAM,
the extra VLAN_HLEN could be possible in both cases. Presumably, the
reserve code was copied from packet_snd(), but later on missed the
check. Make it similar as we have it in packet_snd().
Fixes: 69e3c75f4d ("net: TX_RING and packet mmap")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In case no struct sockaddr_ll has been passed to packet
socket's sendmsg() when doing a TX_RING flush run, then
skb->protocol is set to po->num instead, which is the protocol
passed via socket(2)/bind(2).
Applications only xmitting can go the path of allocating the
socket as socket(PF_PACKET, <mode>, 0) and do a bind(2) on the
TX_RING with sll_protocol of 0. That way, register_prot_hook()
is neither called on creation nor on bind time, which saves
cycles when there's no interest in capturing anyway.
That leaves us however with po->num 0 instead and therefore
the TX_RING flush run sets skb->protocol to 0 as well. Eric
reported that this leads to problems when using tools like
trafgen over bonding device. I.e. the bonding's hash function
could invoke the kernel's flow dissector, which depends on
skb->protocol being properly set. In the current situation, all
the traffic is then directed to a single slave.
Fix it up by inferring skb->protocol from the Ethernet header
when not set and we have ARPHRD_ETHER device type. This is only
done in case of SOCK_RAW and where we have a dev->hard_header_len
length. In case of ARPHRD_ETHER devices, this is guaranteed to
cover ETH_HLEN, and therefore being accessed on the skb after
the skb_store_bits().
Reported-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Packet sockets can be used by various net devices and are not
really restricted to ARPHRD_ETHER device types. However, when
currently checking for the extra 4 bytes that can be transmitted
in VLAN case, our assumption is that we generally probe on
ARPHRD_ETHER devices. Therefore, before looking into Ethernet
header, check the device type first.
This also fixes the issue where non-ARPHRD_ETHER devices could
have no dev->hard_header_len in TX_RING SOCK_RAW case, and thus
the check would test unfilled linear part of the skb (instead
of non-linear).
Fixes: 57f89bfa21 ("network: Allow af_packet to transmit +4 bytes for VLAN packets.")
Fixes: 52f1454f62 ("packet: allow to transmit +4 byte in TX_RING slot for VLAN case")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We concluded that the skb_probe_transport_header() should better be
called unconditionally. Avoiding the call into the flow dissector has
also not really much to do with the direct xmit mode.
While it seems that only virtio_net code makes use of GSO from non
RX/TX ring packet socket paths, we should probe for a transport header
nevertheless before they hit devices.
Reference: http://thread.gmane.org/gmane.linux.network/386173/
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In tpacket_fill_skb() commit c1aad275b0 ("packet: set transport
header before doing xmit") and later on 40893fd0fd ("net: switch
to use skb_probe_transport_header()") was probing for a transport
header on the skb from a ring buffer slot, but at a time, where
the skb has _not even_ been filled with data yet. So that call into
the flow dissector is pretty useless. Lets do it after we've set
up the skb frags.
Fixes: c1aad275b0 ("packet: set transport header before doing xmit")
Reported-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is a race conditions between packet_notifier and packet_bind{_spkt}.
It happens if packet_notifier(NETDEV_UNREGISTER) executes between the
time packet_bind{_spkt} takes a reference on the new netdevice and the
time packet_do_bind sets po->ifindex.
In this case the notification can be missed.
If this happens during a dev_change_net_namespace this can result in the
netdevice to be moved to the new namespace while the packet_sock in the
old namespace still holds a reference on it. When the netdevice is later
deleted in the new namespace the deletion hangs since the packet_sock
is not found in the new namespace' &net->packet.sklist.
It can be reproduced with the script below.
This patch makes packet_do_bind check again for the presence of the
netdevice in the packet_sock's namespace after the synchronize_net
in unregister_prot_hook.
More in general it also uses the rcu lock for the duration of the bind
to stop dev_change_net_namespace/rollback_registered_many from
going past the synchronize_net following unlist_netdevice, so that
no NETDEV_UNREGISTER notifications can happen on the new netdevice
while the bind is executing. In order to do this some code from
packet_bind{_spkt} is consolidated into packet_do_dev.
import socket, os, time, sys
proto=7
realDev='em1'
vlanId=400
if len(sys.argv) > 1:
vlanId=int(sys.argv[1])
dev='vlan%d' % vlanId
os.system('taskset -p 0x10 %d' % os.getpid())
s = socket.socket(socket.PF_PACKET, socket.SOCK_RAW, proto)
os.system('ip link add link %s name %s type vlan id %d' %
(realDev, dev, vlanId))
os.system('ip netns add dummy')
pid=os.fork()
if pid == 0:
# dev should be moved while packet_do_bind is in synchronize net
os.system('taskset -p 0x20000 %d' % os.getpid())
os.system('ip link set %s netns dummy' % dev)
os.system('ip netns exec dummy ip link del %s' % dev)
s.close()
sys.exit(0)
time.sleep(.004)
try:
s.bind(('%s' % dev, proto+1))
except:
print 'Could not bind socket'
s.close()
os.system('ip netns del dummy')
sys.exit(0)
os.waitpid(pid, 0)
s.close()
os.system('ip netns del dummy')
sys.exit(0)
Signed-off-by: Francesco Ruggeri <fruggeri@arista.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The function ip_defrag is called on both the input and the output
paths of the networking stack. In particular conntrack when it is
tracking outbound packets from the local machine calls ip_defrag.
So add a struct net parameter and stop making ip_defrag guess which
network namespace it needs to defragment packets in.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Acked-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Recent TCP listener patches exposed a prior af_packet bug :
match_fanout_group() blindly assumes it is always safe
to cast sk to a packet socket to compare fanout with af_packet_priv
But SYNACK packets can be sent while attached to request_sock, which
are smaller than a "struct sock".
We can read non existent memory and crash.
Fixes: c0de08d042 ("af_packet: don't emit packet on orig fanout group")
Fixes: ca6fb06518 ("tcp: attach SYNACK messages to request sockets instead of listener")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Cc: Eric Leblond <eric@regit.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Edward Hyunkoo Jee <edjee@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
eBPF socket filter programs may see junk in 'u32 cb[5]' area,
since it could have been used by protocol layers earlier.
For socket filter programs used in af_packet we need to clean
20 bytes of skb->cb area if it could be used by the program.
For programs attached to TCP/UDP sockets we need to save/restore
these 20 bytes, since it's used by protocol layers.
Remove SK_RUN_FILTER macro, since it's no longer used.
Long term we may move this bpf cb area to per-cpu scratch, but that
requires addition of new 'per-cpu load/store' instructions,
so not suitable as a short term fix.
Fixes: d691f9e8d4 ("bpf: allow programs to write to certain skb fields")
Reported-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The current ongoing effort to dump existing cBPF seccomp filters back
to user space requires to hold the pre-transformed instructions like
we do in case of socket filters from sk_attach_filter() side, so they
can be reloaded in original form at a later point in time by utilities
such as criu.
To prepare for this, simply extend the bpf_prog_create_from_user()
API to hold a flag that tells whether we should store the original
or not. Also, fanout filters could make use of that in future for
things like diag. While fanout filters already use bpf_prog_destroy(),
move seccomp over to them as well to handle original programs when
present.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: Tycho Andersen <tycho.andersen@canonical.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Tested-by: Tycho Andersen <tycho.andersen@canonical.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 7d82410950 ("virtio: add explicit big-endian support to memory
accessors") accidentally changed the virtio_net header used by
AF_PACKET with PACKET_VNET_HDR from host-endian to big-endian.
Since virtio_legacy_is_little_endian() is a very long identifier,
define a vio_le macro and use that throughout the code instead of the
hard-coded 'false' for little-endian.
This restores the ABI to match 4.1 and earlier kernels, and makes my
test program work again.
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add fanout mode PACKET_FANOUT_EBPF that accepts an en extended BPF
program to select a socket.
Update the internal eBPF program by passing to socket option
SOL_PACKET/PACKET_FANOUT_DATA a file descriptor returned by bpf().
Signed-off-by: Willem de Bruijn <willemb@google.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add fanout mode PACKET_FANOUT_CBPF that accepts a classic BPF program
to select a socket.
This avoids having to keep adding special case fanout modes. One
example use case is application layer load balancing. The QUIC
protocol, for instance, encodes a connection ID in UDP payload.
Also add socket option SOL_PACKET/PACKET_FANOUT_DATA that updates data
associated with the socket group. Fanout mode PACKET_FANOUT_CBPF is the
only user so far.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
arch/s390/net/bpf_jit_comp.c
drivers/net/ethernet/ti/netcp_ethss.c
net/bridge/br_multicast.c
net/ipv4/ip_fragment.c
All four conflicts were cases of simple overlapping
changes.
Signed-off-by: David S. Miller <davem@davemloft.net>
Follow e8e85cc5eb ("packet: remove handling of tx_ring") and remove
the tx_ring parameter from prb_shutdown_retire_blk_timer() as it is only
called with tx_ring = 0.
Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
tpacket_fill_skb() can return a negative value (-errno) which
is stored in tp_len variable. In that case the following
condition will be (but shouldn't be) true:
tp_len > dev->mtu + dev->hard_header_len
as dev->mtu and dev->hard_header_len are both unsigned.
That may lead to just returning an incorrect EMSGSIZE errno
to the user.
Fixes: 52f1454f62 ("packet: allow to transmit +4 byte in TX_RING slot for VLAN case")
Signed-off-by: Alexander Drozdov <al.drozdov@gmail.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
When binding a PF_PACKET socket, the use count of the bound interface is
always increased with dev_hold in dev_get_by_{index,name}. However,
when rebound with the same protocol and device as in the previous bind
the use count of the interface was not decreased. Ultimately, this
caused the deletion of the interface to fail with the following message:
unregister_netdevice: waiting for dummy0 to become free. Usage count = 1
This patch moves the dev_put out of the conditional part that was only
executed when either the protocol or device changed on a bind.
Fixes: 902fefb82e ('packet: improve socket create/bind latency in some cases')
Signed-off-by: Lars Westerhoff <lars.westerhoff@newtec.eu>
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
drivers/net/ethernet/mellanox/mlx4/main.c
net/packet/af_packet.c
Both conflicts were cases of simple overlapping changes.
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove handling of tx_ring in prb_setup_retire_blk_timer
for TPACKET_V3 because init_prb_bdqc is called only for zero tx_ring
and thus prb_setup_retire_blk_timer for zero tx_ring only.
And also in functon init_prb_bdqc there is no usage of tx_ring.
Thus removing tx_ring from init_prb_bdqc.
Signed-off-by: Maninder Singh <maninder1.s@samsung.com>
Suggested-by: Frans Klaver <fransklaver@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
PACKET_FANOUT_LB computes f->rr_cur such that it is modulo
f->num_members. It returns the old value unconditionally, but
f->num_members may have changed since the last store. Ensure
that the return value is always < num.
When modifying the logic, simplify it further by replacing the loop
with an unconditional atomic increment.
Fixes: dc99f60069 ("packet: Add fanout support.")
Suggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Willem de Bruijn <willemb@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Destruction of the po->rollover must be delayed until there are no
more packets in flight that can access it. The field is destroyed in
packet_release, before synchronize_net. Delay using rcu.
Fixes: 0648ab70af ("packet: rollover prepare: per-socket state")
Suggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Willem de Bruijn <willemb@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We need to tell compiler it must not read f->num_members multiple
times. Otherwise testing if num is not zero is flaky, and we could
attempt an invalid divide by 0 in fanout_demux_cpu()
Note bug was present in packet_rcv_fanout_hash() and
packet_rcv_fanout_lb() but final 3.1 had a simple location
after commit 95ec3eb417 ("packet: Add 'cpu' fanout policy.")
Fixes: dc99f60069 ("packet: Add fanout support.")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Rollover can be enabled as flag or mode. Allocate state in both cases.
This solves a NULL pointer exception in fanout_demux_rollover on
referencing po->rollover if using mode rollover.
Also make sure that in rollover mode each silo is tried (contrary
to rollover flag, where the main socket is excluded after an initial
try_self).
Tested:
Passes tools/testing/net/psock_fanout.c, which tests both modes and
flag. My previous tests were limited to bench_rollover, which only
stresses the flag. The test now completes safely. it still gives an
error for mode rollover, because it does not expect the new headroom
(ROOM_NORMAL) requirement. I will send a separate patch to the test.
Fixes: 0648ab70af ("packet: rollover prepare: per-socket state")
Signed-off-by: Willem de Bruijn <willemb@google.com>
----
I should have run this test and caught this before submission, of
course. Apologies for the oversight.
Signed-off-by: David S. Miller <davem@davemloft.net>
Avoid two xchg calls whose return values were unused, causing a
warning on some architectures.
The relevant variable is a hint and read without mutual exclusion.
This fix makes all writers hold the receive_queue lock.
Suggested-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Rollover indicates exceptional conditions. Export a counter to inform
socket owners of this state.
If no socket with sufficient room is found, rollover fails. Also count
these events.
Finally, also count when flows are rolled over early thanks to huge
flow detection, to validate its correctness.
Tested:
Read counters in bench_rollover on all other tests in the patchset
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Migrate flows from a socket to another socket in the fanout group not
only when the socket is full. Start migrating huge flows early, to
divert possible 4-tuple attacks without affecting normal traffic.
Introduce fanout_flow_is_huge(). This detects huge flows, which are
defined as taking up more than half the load. It does so cheaply, by
storing the rxhashes of the N most recent packets. If over half of
these are the same rxhash as the current packet, then drop it. This
only protects against 4-tuple attacks. N is chosen to fit all data in
a single cache line.
Tested:
Ran bench_rollover for 10 sec with 1.5 Mpps of single flow input.
lpbb5:/export/hda3/willemb# ./bench_rollover -l 1000 -r -s
cpu rx rx.k drop.k rollover r.huge r.failed
0 14 14 0 0 0 0
1 20 20 0 0 0 0
2 16 16 0 0 0 0
3 6168824 6168824 0 4867721 4867721 0
4 4867741 4867741 0 0 0 0
5 12 12 0 0 0 0
6 15 15 0 0 0 0
7 17 17 0 0 0 0
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Rollover has to call packet_rcv_has_room on sockets in the fanout
group to find a socket to migrate to. This operation is expensive
especially if the packet sockets use rings, when a lock has to be
acquired.
Avoid pounding on the lock by all sockets by temporarily marking a
socket as "under memory pressure" when such pressure is detected.
While set, only the socket owner may call packet_rcv_has_room on the
socket. Once it detects normal conditions, it clears the flag. The
socket is not used as a victim by any other socket in the meantime.
Under reasonably balanced load, each socket writer frequently calls
packet_rcv_has_room and clears its own pressure field. As a backup
for when the socket is rarely written to, also clear the flag on
reading (packet_recvmsg, packet_poll) if this can be done cheaply
(i.e., without calling packet_rcv_has_room). This is only for
edge cases.
Tested:
Ran bench_rollover: a process with 8 sockets in a single fanout
group, each pinned to a single cpu that receives one nic recv
interrupt. RPS and RFS are disabled. The benchmark uses packet
rx_ring, which has to take a lock when determining whether a
socket has room.
Sent 3.5 Mpps of UDP traffic with sufficient entropy to spread
uniformly across the packet sockets (and inserted an iptables
rule to drop in PREROUTING to avoid protocol stack processing).
Without this patch, all sockets try to migrate traffic to
neighbors, causing lock contention when searching for a non-
empty neighbor. The lock is the top 9 entries.
perf record -a -g sleep 5
- 17.82% bench_rollover [kernel.kallsyms] [k] _raw_spin_lock
- _raw_spin_lock
- 99.00% spin_lock
+ 81.77% packet_rcv_has_room.isra.41
+ 18.23% tpacket_rcv
+ 0.84% packet_rcv_has_room.isra.41
+ 5.20% ksoftirqd/6 [kernel.kallsyms] [k] _raw_spin_lock
+ 5.15% ksoftirqd/1 [kernel.kallsyms] [k] _raw_spin_lock
+ 5.14% ksoftirqd/2 [kernel.kallsyms] [k] _raw_spin_lock
+ 5.12% ksoftirqd/7 [kernel.kallsyms] [k] _raw_spin_lock
+ 5.12% ksoftirqd/5 [kernel.kallsyms] [k] _raw_spin_lock
+ 5.10% ksoftirqd/4 [kernel.kallsyms] [k] _raw_spin_lock
+ 4.66% ksoftirqd/0 [kernel.kallsyms] [k] _raw_spin_lock
+ 4.45% ksoftirqd/3 [kernel.kallsyms] [k] _raw_spin_lock
+ 1.55% bench_rollover [kernel.kallsyms] [k] packet_rcv_has_room.isra.41
On net-next with this patch, this lock contention is no longer a
top entry. Most time is spent in the actual read function. Next up
are other locks:
+ 15.52% bench_rollover bench_rollover [.] reader
+ 4.68% swapper [kernel.kallsyms] [k] memcpy_erms
+ 2.77% swapper [kernel.kallsyms] [k] packet_lookup_frame.isra.51
+ 2.56% ksoftirqd/1 [kernel.kallsyms] [k] memcpy_erms
+ 2.16% swapper [kernel.kallsyms] [k] tpacket_rcv
+ 1.93% swapper [kernel.kallsyms] [k] mlx4_en_process_rx_cq
Looking closer at the remaining _raw_spin_lock, the cost of probing
in rollover is now comparable to the cost of taking the lock later
in tpacket_rcv.
- 1.51% swapper [kernel.kallsyms] [k] _raw_spin_lock
- _raw_spin_lock
+ 33.41% packet_rcv_has_room
+ 28.15% tpacket_rcv
+ 19.54% enqueue_to_backlog
+ 6.45% __free_pages_ok
+ 2.78% packet_rcv_fanout
+ 2.13% fanout_demux_rollover
+ 2.01% netif_receive_skb_internal
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Only migrate flows to sockets that have sufficient headroom, where
sufficient is defined as having at least 25% empty space.
The kernel has three different buffer types: a regular socket, a ring
with frames (TPACKET_V[12]) or a ring with blocks (TPACKET_V3). The
latter two do not expose a read pointer to the kernel, so headroom is
not computed easily. All three needs a different implementation to
estimate free space.
Tested:
Ran bench_rollover for 10 sec with 1.5 Mpps of single flow input.
bench_rollover has as many sockets as there are NIC receive queues
in the system. Each socket is owned by a process that is pinned to
one of the receive cpus. RFS is disabled. RPS is enabled with an
identity mapping (cpu x -> cpu x), to count drops with softnettop.
lpbb5:/export/hda3/willemb# ./bench_rollover -r -l 1000 -s
Press [Enter] to exit
cpu rx rx.k drop.k rollover r.huge r.failed
0 16 16 0 0 0 0
1 21 21 0 0 0 0
2 5227502 5227502 0 0 0 0
3 18 18 0 0 0 0
4 6083289 6083289 0 5227496 0 0
5 22 22 0 0 0 0
6 21 21 0 0 0 0
7 9 9 0 0 0 0
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace rollover state per fanout group with state per socket. Future
patches will add fields to the new structure.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
packet_rcv_fanout calls fanout_demux_rollover twice. Move all rollover
logic into the callee to simplify these callsites, especially with
upcoming changes.
The main differences between the two callsites is that the FLAG
variant tests whether the socket previously selected by another
mode (RR, RND, HASH, ..) has room before migrating flows, whereas the
rollover mode has no original socket to test.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Four minor merge conflicts:
1) qca_spi.c renamed the local variable used for the SPI device
from spi_device to spi, meanwhile the spi_set_drvdata() call
got moved further up in the probe function.
2) Two changes were both adding new members to codel params
structure, and thus we had overlapping changes to the
initializer function.
3) 'net' was making a fix to sk_release_kernel() which is
completely removed in 'net-next'.
4) In net_namespace.c, the rtnl_net_fill() call for GET operations
had the command value fixed, meanwhile 'net-next' adjusted the
argument signature a bit.
This also matches example merge resolutions posted by Stephen
Rothwell over the past two days.
Signed-off-by: David S. Miller <davem@davemloft.net>
In preparation for changing how struct net is refcounted
on kernel sockets pass the knowledge that we are creating
a kernel socket from sock_create_kern through to sk_alloc.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch fixes an issue where the send(MSG_DONTWAIT) call
on a TX_RING is not fully non-blocking in cases where the device's sndBuf is
full. We pass nonblock=true to sock_alloc_send_skb() and return any possibly
occuring error code (most likely EGAIN) to the caller. As the fast-path stays
as it is, we keep the unlikely() around skb == NULL.
Signed-off-by: Mathias Kretschmer <mathias.kretschmer@fokus.fraunhofer.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Introduce TP_STATUS_CSUM_VALID tp_status flag to tell the
af_packet user that at least the transport header checksum
has been already validated.
For now, the flag may be set for incoming packets only.
Signed-off-by: Alexander Drozdov <al.drozdov@gmail.com>
Cc: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It is just an optimization. We don't need the value of status variable
if the packet is filtered.
Signed-off-by: Alexander Drozdov <al.drozdov@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Having to say
> #ifdef CONFIG_NET_NS
> struct net *net;
> #endif
in structures is a little bit wordy and a little bit error prone.
Instead it is possible to say:
> typedef struct {
> #ifdef CONFIG_NET_NS
> struct net *net;
> #endif
> } possible_net_t;
And then in a header say:
> possible_net_t net;
Which is cleaner and easier to use and easier to test, as the
possible_net_t is always there no matter what the compile options.
Further this allows read_pnet and write_pnet to be functions in all
cases which is better at catching typos.
This change adds possible_net_t, updates the definitions of read_pnet
and write_pnet, updates optional struct net * variables that
write_pnet uses on to have the type possible_net_t, and finally fixes
up the b0rked users of read_pnet and write_pnet.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
drivers/net/ethernet/cadence/macb.c
Overlapping changes in macb driver, mostly fixes and cleanups
in 'net' overlapping with the integration of at91_ether into
macb in 'net-next'.
Signed-off-by: David S. Miller <davem@davemloft.net>
When an interface is deleted from a net namespace the ifindex in the
corresponding entries in PF_PACKET sockets' mclists becomes stale.
This can create inconsistencies if later an interface with the same ifindex
is moved from a different namespace (not that unlikely since ifindexes are
per-namespace).
In particular we saw problems with dev->promiscuity, resulting
in "promiscuity touches roof, set promiscuity failed. promiscuity
feature of device might be broken" warnings and EOVERFLOW failures of
setsockopt(PACKET_ADD_MEMBERSHIP).
This patch deletes the mclist entries for interfaces that are deleted.
Since this now causes setsockopt(PACKET_DROP_MEMBERSHIP) to fail with
EADDRNOTAVAIL if called after the interface is deleted, also make
packet_mc_drop not fail.
Signed-off-by: Francesco Ruggeri <fruggeri@arista.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
