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>
In tcp_create_openreq_child we adjust tcp_header_len for md5 using the
remote address in newsk. But that address is still 0 in newsk at this
point, and it is only set later by the callers (tcp_v[46]_syn_recv_sock).
Use the address from the request socket instead.
Fixes: cfb6eeb4c8 ("[TCP]: MD5 Signature Option (RFC2385) support.")
Signed-off-by: Francesco Ruggeri <fruggeri@arista.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20220421005026.686A45EC01F2@us226.sjc.aristanetworks.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>
A latter patch will postpone the delivery_time clearing until the stack
knows the skb is being delivered locally. That will allow other kernel
forwarding path (e.g. ip[6]_forward) to keep the delivery_time also.
An earlier attempt was to do skb_clear_delivery_time() in
ip_local_deliver() and ip6_input(). The discussion [0] requested
to move it one step later into ip_local_deliver_finish()
and ip6_input_finish() so that the delivery_time can be kept
for the ip_vs forwarding path also.
To do that, this patch also needs to take care of the (rcv) timestamp
usecase in ip_is_fragment(). It needs to expect delivery_time in
the skb->tstamp, so it needs to save the mono_delivery_time bit in
inet_frag_queue such that the delivery_time (if any) can be restored
in the final defragmented skb.
[Note that it will only happen when the locally generated skb is looping
from egress to ingress over a virtual interface (e.g. veth, loopback...),
skb->tstamp may have the delivery time before it is known that it will
be delivered locally and received by another sk.]
[0]: https://lore.kernel.org/netdev/ca728d81-80e8-3767-d5e-d44f6ad96e43@ssi.bg/
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The previous patches handled the delivery_time before sch_handle_ingress().
This patch can now set the skb->mono_delivery_time to flag the skb->tstamp
is used as the mono delivery_time (EDT) instead of the (rcv) timestamp
and also clear it with skb_clear_delivery_time() after
sch_handle_ingress(). This will make the bpf_redirect_*()
to keep the mono delivery_time and used by a qdisc (fq) of
the egress-ing interface.
A latter patch will postpone the skb_clear_delivery_time() until the
stack learns that the skb is being delivered locally and that will
make other kernel forwarding paths (ip[6]_forward) able to keep
the delivery_time also. Thus, like the previous patches on using
the skb->mono_delivery_time bit, calling skb_clear_delivery_time()
is not limited within the CONFIG_NET_INGRESS to avoid too many code
churns among this set.
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Right now, skb->tstamp is reset to 0 whenever the skb is forwarded.
If skb->tstamp has the mono delivery_time, clearing it can hurt
the performance when it finally transmits out to fq@phy-dev.
The earlier patch added a skb->mono_delivery_time bit to
flag the skb->tstamp carrying the mono delivery_time.
This patch adds skb_clear_tstamp() helper which keeps
the mono delivery_time and clears everything else.
The delivery_time clearing will be postponed until the stack knows the
skb will be delivered locally. It will be done in a latter patch.
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
skb->tstamp was first used as the (rcv) timestamp.
The major usage is to report it to the user (e.g. SO_TIMESTAMP).
Later, skb->tstamp is also set as the (future) delivery_time (e.g. EDT in TCP)
during egress and used by the qdisc (e.g. sch_fq) to make decision on when
the skb can be passed to the dev.
Currently, there is no way to tell skb->tstamp having the (rcv) timestamp
or the delivery_time, so it is always reset to 0 whenever forwarded
between egress and ingress.
While it makes sense to always clear the (rcv) timestamp in skb->tstamp
to avoid confusing sch_fq that expects the delivery_time, it is a
performance issue [0] to clear the delivery_time if the skb finally
egress to a fq@phy-dev. For example, when forwarding from egress to
ingress and then finally back to egress:
tcp-sender => veth@netns => veth@hostns => fq@eth0@hostns
^ ^
reset rest
This patch adds one bit skb->mono_delivery_time to flag the skb->tstamp
is storing the mono delivery_time (EDT) instead of the (rcv) timestamp.
The current use case is to keep the TCP mono delivery_time (EDT) and
to be used with sch_fq. A latter patch will also allow tc-bpf@ingress
to read and change the mono delivery_time.
In the future, another bit (e.g. skb->user_delivery_time) can be added
for the SCM_TXTIME where the clock base is tracked by sk->sk_clockid.
[ This patch is a prep work. The following patches will
get the other parts of the stack ready first. Then another patch
after that will finally set the skb->mono_delivery_time. ]
skb_set_delivery_time() function is added. It is used by the tcp_output.c
and during ip[6] fragmentation to assign the delivery_time to
the skb->tstamp and also set the skb->mono_delivery_time.
A note on the change in ip_send_unicast_reply() in ip_output.c.
It is only used by TCP to send reset/ack out of a ctl_sk.
Like the new skb_set_delivery_time(), this patch sets
the skb->mono_delivery_time to 0 for now as a place
holder. It will be enabled in a latter patch.
A similar case in tcp_ipv6 can be done with
skb_set_delivery_time() in tcp_v6_send_response().
[0] (slide 22): https://linuxplumbersconf.org/event/11/contributions/953/attachments/867/1658/LPC_2021_BPF_Datapath_Extensions.pdf
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If recv_actor() returns an incorrect value, tcp_read_sock()
might loop forever.
Instead, issue a one time warning and make sure to make progress.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Jakub Sitnicki <jakub@cloudflare.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/r/20220302161723.3910001-2-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When neighbour become invalid or destroyed, neigh_invalidate() will be
called. neigh->ops->error_report() will be called if the neighbour's
state is NUD_FAILED, and seems here is the only use of error_report().
So we can tell that the reason of skb drops in arp_error_report() is
SKB_DROP_REASON_NEIGH_FAILED.
Replace kfree_skb() used in arp_error_report() with kfree_skb_reason().
Reviewed-by: Mengen Sun <mengensun@tencent.com>
Reviewed-by: Hao Peng <flyingpeng@tencent.com>
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace kfree_skb() which is used in the packet egress path of IP layer
with kfree_skb_reason(). Functions that are involved include:
__ip_queue_xmit()
ip_finish_output()
ip_mc_finish_output()
ip6_output()
ip6_finish_output()
ip6_finish_output2()
Following new drop reasons are introduced:
SKB_DROP_REASON_IP_OUTNOROUTES
SKB_DROP_REASON_BPF_CGROUP_EGRESS
SKB_DROP_REASON_IPV6DISABLED
SKB_DROP_REASON_NEIGH_CREATEFAIL
Reviewed-by: Mengen Sun <mengensun@tencent.com>
Reviewed-by: Hao Peng <flyingpeng@tencent.com>
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Steffen Klassert says:
====================
1) Fix PMTU for IPv6 if the reported MTU minus the ESP overhead is
smaller than 1280. From Jiri Bohac.
2) Fix xfrm interface ID and inter address family tunneling when
migrating xfrm states. From Yan Yan.
3) Add missing xfrm intrerface ID initialization on xfrmi_changelink.
From Antony Antony.
4) Enforce validity of xfrm offload input flags so that userspace can't
send undefined flags to the offload driver.
From Leon Romanovsky.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
The functions do essentially the same work to verify TCP-MD5 sign.
Code can be merged into one family-independent function in order to
reduce copy'n'paste and generated code.
Later with TCP-AO option added, this will allow to create one function
that's responsible for segment verification, that will have all the
different checks for MD5/AO/non-signed packets, which in turn will help
to see checks for all corner-cases in one function, rather than spread
around different families and functions.
Cc: Eric Dumazet <edumazet@google.com>
Cc: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>
Signed-off-by: Dmitry Safonov <dima@arista.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20220223175740.452397-1-dima@arista.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
As Jakub noticed, prints should be avoided on the datapath.
Also, as packets would never come to the else branch in
ping_lookup(), remove pr_err() from ping_lookup().
Fixes: 35a79e64de ("ping: fix the dif and sdif check in ping_lookup")
Reported-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Link: https://lore.kernel.org/r/1ef3f2fcd31bd681a193b1fcf235eee1603819bd.1645674068.git.lucien.xin@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This test is checking if we exited the list via break or not. However
if it did not exit via a break then "node" does not point to a valid
udp_tunnel_nic_shared_node struct. It will work because of the way
the structs are laid out it's the equivalent of
"if (info->shared->udp_tunnel_nic_info != dev)" which will always be
true, but it's not the right way to test.
Fixes: 74cc6d182d ("udp_tunnel: add the ability to share port tables")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In case user space sends a packet destined to a broadcast address when a
matching broadcast route is not configured, the kernel will create a
unicast neighbour entry that will never be resolved [1].
When the broadcast route is configured, the unicast neighbour entry will
not be invalidated and continue to linger, resulting in packets being
dropped.
Solve this by invalidating unresolved neighbour entries for broadcast
addresses after routes for these addresses are internally configured by
the kernel. This allows the kernel to create a broadcast neighbour entry
following the next route lookup.
Another possible solution that is more generic but also more complex is
to have the ARP code register a listener to the FIB notification chain
and invalidate matching neighbour entries upon the addition of broadcast
routes.
It is also possible to wave off the issue as a user space problem, but
it seems a bit excessive to expect user space to be that intimately
familiar with the inner workings of the FIB/neighbour kernel code.
[1] https://lore.kernel.org/netdev/55a04a8f-56f3-f73c-2aea-2195923f09d1@huawei.com/
Reported-by: Wang Hai <wanghai38@huawei.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Tested-by: Wang Hai <wanghai38@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We encounter a tcp drop issue in our cloud environment. Packet GROed in
host forwards to a VM virtio_net nic with net_failover enabled. VM acts
as a IPVS LB with ipip encapsulation. The full path like:
host gro -> vm virtio_net rx -> net_failover rx -> ipvs fullnat
-> ipip encap -> net_failover tx -> virtio_net tx
When net_failover transmits a ipip pkt (gso_type = 0x0103, which means
SKB_GSO_TCPV4, SKB_GSO_DODGY and SKB_GSO_IPXIP4), there is no gso
did because it supports TSO and GSO_IPXIP4. But network_header points to
inner ip header.
Call Trace:
tcp4_gso_segment ------> return NULL
inet_gso_segment ------> inner iph, network_header points to
ipip_gso_segment
inet_gso_segment ------> outer iph
skb_mac_gso_segment
Afterwards virtio_net transmits the pkt, only inner ip header is modified.
And the outer one just keeps unchanged. The pkt will be dropped in remote
host.
Call Trace:
inet_gso_segment ------> inner iph, outer iph is skipped
skb_mac_gso_segment
__skb_gso_segment
validate_xmit_skb
validate_xmit_skb_list
sch_direct_xmit
__qdisc_run
__dev_queue_xmit ------> virtio_net
dev_hard_start_xmit
__dev_queue_xmit ------> net_failover
ip_finish_output2
ip_output
iptunnel_xmit
ip_tunnel_xmit
ipip_tunnel_xmit ------> ipip
dev_hard_start_xmit
__dev_queue_xmit
ip_finish_output2
ip_output
ip_forward
ip_rcv
__netif_receive_skb_one_core
netif_receive_skb_internal
napi_gro_receive
receive_buf
virtnet_poll
net_rx_action
The root cause of this issue is specific with the rare combination of
SKB_GSO_DODGY and a tunnel device that adds an SKB_GSO_ tunnel option.
SKB_GSO_DODGY is set from external virtio_net. We need to reset network
header when callbacks.gso_segment() returns NULL.
This patch also includes ipv6_gso_segment(), considering SIT, etc.
Fixes: cb32f511a7 ("ipip: add GSO/TSO support")
Signed-off-by: Tao Liu <thomas.liu@ucloud.cn>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace tcp_drop() used in tcp_data_queue_ofo with tcp_drop_reason().
Following drop reasons are introduced:
SKB_DROP_REASON_TCP_OFOMERGE
Reviewed-by: Mengen Sun <mengensun@tencent.com>
Reviewed-by: Hao Peng <flyingpeng@tencent.com>
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace tcp_drop() used in tcp_data_queue() with tcp_drop_reason().
Following drop reasons are introduced:
SKB_DROP_REASON_TCP_ZEROWINDOW
SKB_DROP_REASON_TCP_OLD_DATA
SKB_DROP_REASON_TCP_OVERWINDOW
SKB_DROP_REASON_TCP_OLD_DATA is used for the case that end_seq of skb
less than the left edges of receive window. (Maybe there is a better
name?)
Reviewed-by: Mengen Sun <mengensun@tencent.com>
Reviewed-by: Hao Peng <flyingpeng@tencent.com>
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace tcp_drop() used in tcp_rcv_established() with tcp_drop_reason().
Following drop reasons are added:
SKB_DROP_REASON_TCP_FLAGS
Reviewed-by: Mengen Sun <mengensun@tencent.com>
Reviewed-by: Hao Peng <flyingpeng@tencent.com>
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pengcheng Yang