While reading sysctl_tcp_ecn_fallback, it can be changed concurrently.
Thus, we need to add READ_ONCE() to its reader.

Fixes: 49213555 ("tcp: add rfc3168, section 6.1.1.1. fallback")
Signed-off-by: NKuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

While reading sysctl_tcp_ecn, it can be changed concurrently.
Thus, we need to add READ_ONCE() to its readers.

Fixes: 1da177e4 ("Linux-2.6.12-rc2")
Signed-off-by: NKuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Laurent reported the enclosed report [1]

This bug triggers with following coditions:

0) Kernel built with CONFIG_DEBUG_PREEMPT=y

1) A new passive FastOpen TCP socket is created.
   This FO socket waits for an ACK coming from client to be a complete
   ESTABLISHED one.
2) A socket operation on this socket goes through lock_sock()
   release_sock() dance.
3) While the socket is owned by the user in step 2),
   a retransmit of the SYN is received and stored in socket backlog.
4) At release_sock() time, the socket backlog is processed while
   in process context.
5) A SYNACK packet is cooked in response of the SYN retransmit.
6) -> tcp_rtx_synack() is called in process context.

Before blamed commit, tcp_rtx_synack() was always called from BH handler,
from a timer handler.

Fix this by using TCP_INC_STATS() & NET_INC_STATS()
which do not assume caller is in non preemptible context.

[1]
BUG: using __this_cpu_add() in preemptible [00000000] code: epollpep/2180
caller is tcp_rtx_synack.part.0+0x36/0xc0
CPU: 10 PID: 2180 Comm: epollpep Tainted: G           OE     5.16.0-0.bpo.4-amd64 #1  Debian 5.16.12-1~bpo11+1
Hardware name: Supermicro SYS-5039MC-H8TRF/X11SCD-F, BIOS 1.7 11/23/2021
Call Trace:
 <TASK>
 dump_stack_lvl+0x48/0x5e
 check_preemption_disabled+0xde/0xe0
 tcp_rtx_synack.part.0+0x36/0xc0
 tcp_rtx_synack+0x8d/0xa0
 ? kmem_cache_alloc+0x2e0/0x3e0
 ? apparmor_file_alloc_security+0x3b/0x1f0
 inet_rtx_syn_ack+0x16/0x30
 tcp_check_req+0x367/0x610
 tcp_rcv_state_process+0x91/0xf60
 ? get_nohz_timer_target+0x18/0x1a0
 ? lock_timer_base+0x61/0x80
 ? preempt_count_add+0x68/0xa0
 tcp_v4_do_rcv+0xbd/0x270
 __release_sock+0x6d/0xb0
 release_sock+0x2b/0x90
 sock_setsockopt+0x138/0x1140
 ? __sys_getsockname+0x7e/0xc0
 ? aa_sk_perm+0x3e/0x1a0
 __sys_setsockopt+0x198/0x1e0
 __x64_sys_setsockopt+0x21/0x30
 do_syscall_64+0x38/0xc0
 entry_SYSCALL_64_after_hwframe+0x44/0xae

Fixes: 168a8f58 ("tcp: TCP Fast Open Server - main code path")
Signed-off-by: NEric Dumazet <edumazet@google.com>
Reported-by: NLaurent Fasnacht <laurent.fasnacht@proton.ch>
Acked-by: NNeal Cardwell <ncardwell@google.com>
Link: https://lore.kernel.org/r/20220530213713.601888-1-eric.dumazet@gmail.comSigned-off-by: NJakub Kicinski <kuba@kernel.org>

The code for gso_max_size was added originally to allow for debugging and
workaround of buggy devices that couldn't support TSO with blocks 64K in
size. The original reason for limiting it to 64K was because that was the
existing limits of IPv4 and non-jumbogram IPv6 length fields.

With the addition of Big TCP we can remove this limit and allow the value
to potentially go up to UINT_MAX and instead be limited by the tso_max_size
value.

So in order to support this we need to go through and clean up the
remaining users of the gso_max_size value so that the values will cap at
64K for non-TCPv6 flows. In addition we can clean up the GSO_MAX_SIZE value
so that 64K becomes GSO_LEGACY_MAX_SIZE and UINT_MAX will now be the upper
limit for GSO_MAX_SIZE.

v6: (edumazet) fixed a compile error if CONFIG_IPV6=n,
               in a new sk_trim_gso_size() helper.
               netif_set_tso_max_size() caps the requested TSO size
               with GSO_MAX_SIZE.
Signed-off-by: NAlexander Duyck <alexanderduyck@fb.com>
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The MPTCP RFC requires that the MPTCP-level receive window's
right edge never moves backward. Currently the MPTCP code
enforces such constraint while tracking the right edge, but it
does not reflects it on the wire, as MPTCP lacks a suitable hook
to update accordingly the TCP header.

This change modifies the existing mptcp_write_options() hook,
providing the current packet's TCP header to the MPTCP protocol,
so that the next patch could implement the above mentioned
constraint.

No functional changes intended.
Signed-off-by: NPaolo Abeni <pabeni@redhat.com>
Signed-off-by: NMat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: NJakub 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
0cbe6a8f ("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: c9bee3b7 ("tcp: TCP_NOTSENT_LOWAT socket option")
Signed-off-by: NEric Dumazet <edumazet@google.com>
Reported-by: NDoug Porter <dsp@fb.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Acked-by: NSoheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We had various bugs over the years with code
breaking the assumption that tp->snd_cwnd is greater
than zero.

Lately, syzbot reported the WARN_ON_ONCE(!tp->prior_cwnd) added
in commit 8b8a321f ("tcp: fix zero cwnd in tcp_cwnd_reduction")
can trigger, and without a repro we would have to spend
considerable time finding the bug.

Instead of complaining too late, we want to catch where
and when tp->snd_cwnd is set to an illegal value.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Suggested-by: NYuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Acked-by: NYuchung Cheng <ycheng@google.com>
Link: https://lore.kernel.org/r/20220405233538.947344-1-eric.dumazet@gmail.comSigned-off-by: NJakub 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: c7bb4b89 ("ipv6: tcp: drop silly ICMPv6 packet too big messages")
Reported-by: NAndre Nash <alnash@fb.com>
Reported-by: NNeil Spring <ntspring@fb.com>
Reviewed-by: NWei Wang <weiwan@google.com>
Acked-by: NYuchung Cheng <ycheng@google.com>
Acked-by: NMartin KaFai Lau <kafai@fb.com>
Signed-off-by: NJakub Kicinski <kuba@kernel.org>
Reviewed-by: NEric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20220321165957.1769954-1-kuba@kernel.orgSigned-off-by: NJakub Kicinski <kuba@kernel.org>

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: NEric Dumazet <edumazet@google.com>
Reviewed-by: NNeal Cardwell <ncardwell@google.com>
Link: https://lore.kernel.org/r/20220309015757.2532973-1-eric.dumazet@gmail.comSigned-off-by: NJakub Kicinski <kuba@kernel.org>

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.pdfSigned-off-by: NMartin KaFai Lau <kafai@fb.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Disabling rehash behavior did not affect SYN ACK retransmits because hash
was forcefully changed bypassing the sk_rethink_hash function. This patch
adds a condition which checks for rehash mode before resetting hash.
Signed-off-by: NAkhmat Karakotov <hmukos@yandex-team.ru>
Reviewed-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

sk_gso_max_size is set based on the dst dev. Both users of it
adjust the value by the same offset - (MAX_TCP_HEADER + 1). Rather
than compute the same adjusted value on each call do the adjustment
once when set.
Signed-off-by: NDavid Ahern <dsahern@kernel.org>
Reviewed-by: NEric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20220125024511.27480-1-dsahern@kernel.orgSigned-off-by: NJakub Kicinski <kuba@kernel.org>

This reverts commit aeeecb88.

The new SNMP variable (TCPSmallQueueFailure) can be incremented
for good reasons, even on a 100Gbit single TCP_STREAM flow.

If we really wanted to ease driver debugging [1], this would
require something more sophisticated.

[1] Usually, if a driver is delaying TX completions too much,
this can lead to stalls in TCP output. Various work arounds
have been used in the past, like skb_orphan() in ndo_start_xmit().
Signed-off-by: NEric Dumazet <edumazet@google.com>
Cc: Menglong Dong <imagedong@tencent.com>
Link: https://lore.kernel.org/r/20211201033246.2826224-1-eric.dumazet@gmail.comSigned-off-by: NJakub Kicinski <kuba@kernel.org>

Once tcp small queue check failed in tcp_small_queue_check(), the
throughput of tcp will be limited, and it's hard to distinguish
whether it is out of tcp congestion control.

Add statistics of LINUX_MIB_TCPSMALLQUEUEFAILURE for this scene.
Signed-off-by: NMenglong Dong <imagedong@tencent.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Instead of using a full netdev_features_t, we can use a single bit,
as sk_route_nocaps is only used to remove NETIF_F_GSO_MASK from
sk->sk_route_cap.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Extending these flags using the existing (1 << x) pattern triggers
complaints from checkpatch. Instead of ignoring checkpatch modify the
existing values to use BIT(x) style in a separate commit.
Signed-off-by: NLeonard Crestez <cdleonard@gmail.com>
Reviewed-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Track skbs containing only zerocopy data and avoid charging them to
kernel memory to correctly account the memory utilization for
msg_zerocopy. All of the data in such skbs is held in user pages which
are already accounted to user. Before this change, they are charged
again in kernel in __zerocopy_sg_from_iter. The charging in kernel is
excessive because data is not being copied into skb frags. This
excessive charging can lead to kernel going into memory pressure
state which impacts all sockets in the system adversely. Mark pure
zerocopy skbs with a SKBFL_PURE_ZEROCOPY flag and remove
charge/uncharge for data in such skbs.

Initially, an skb is marked pure zerocopy when it is empty and in
zerocopy path. skb can then change from a pure zerocopy skb to mixed
data skb (zerocopy and copy data) if it is at tail of write queue and
there is room available in it and non-zerocopy data is being sent in
the next sendmsg call. At this time sk_mem_charge is done for the pure
zerocopied data and the pure zerocopy flag is unmarked. We found that
this happens very rarely on workloads that pass MSG_ZEROCOPY.

A pure zerocopy skb can later be coalesced into normal skb if they are
next to each other in queue but this patch prevents coalescing from
happening. This avoids complexity of charging when skb downgrades from
pure zerocopy to mixed. This is also rare.

In sk_wmem_free_skb, if it is a pure zerocopy skb, an sk_mem_uncharge
for SKB_TRUESIZE(skb_end_offset(skb)) is done for sk_mem_charge in
tcp_skb_entail for an skb without data.

Testing with the msg_zerocopy.c benchmark between two hosts(100G nics)
with zerocopy showed that before this patch the 'sock' variable in
memory.stat for cgroup2 that tracks sum of sk_forward_alloc,
sk_rmem_alloc and sk_wmem_queued is around 1822720 and with this
change it is 0. This is due to no charge to sk_forward_alloc for
zerocopy data and shows memory utilization for kernel is lowered.

With this commit we don't see the warning we saw in previous commit
which resulted in commit 84882cf7.
Signed-off-by: NTalal Ahmad <talalahmad@google.com>
Acked-by: NArjun Roy <arjunroy@google.com>
Acked-by: NSoheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: NWillem de Bruijn <willemb@google.com>
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

While commit 097b9146 ("net: fix up truesize of cloned
skb in skb_prepare_for_shift()") fixed immediate issues found
when KFENCE was enabled/tested, there are still similar issues,
when tcp_trim_head() hits KFENCE while the master skb
is cloned.

This happens under heavy networking TX workloads,
when the TX completion might be delayed after incoming ACK.

This patch fixes the WARNING in sk_stream_kill_queues
when sk->sk_mem_queued/sk->sk_forward_alloc are not zero.

Fixes: d3fb45f3 ("mm, kfence: insert KFENCE hooks for SLAB")
Signed-off-by: NEric Dumazet <edumazet@google.com>
Acked-by: NMarco Elver <elver@google.com>
Link: https://lore.kernel.org/r/20211102004555.1359210-1-eric.dumazet@gmail.comSigned-off-by: NJakub Kicinski <kuba@kernel.org>

This reverts commit f1a456f8.

  WARNING: CPU: 1 PID: 6819 at net/core/skbuff.c:5429 skb_try_coalesce+0x78b/0x7e0
  CPU: 1 PID: 6819 Comm: xxxxxxx Kdump: loaded Tainted: G S                5.15.0-04194-gd852503f7711 #16
  RIP: 0010:skb_try_coalesce+0x78b/0x7e0
  Code: e8 2a bf 41 ff 44 8b b3 bc 00 00 00 48 8b 7c 24 30 e8 19 c0 41 ff 44 89 f0 48 03 83 c0 00 00 00 48 89 44 24 40 e9 47 fb ff ff <0f> 0b e9 ca fc ff ff 4c 8d 70 ff 48 83 c0 07 48 89 44 24 38 e9 61
  RSP: 0018:ffff88881f449688 EFLAGS: 00010282
  RAX: 00000000fffffe96 RBX: ffff8881566e4460 RCX: ffffffff82079f7e
  RDX: 0000000000000003 RSI: dffffc0000000000 RDI: ffff8881566e47b0
  RBP: ffff8881566e46e0 R08: ffffed102619235d R09: ffffed102619235d
  R10: ffff888130c91ae3 R11: ffffed102619235c R12: ffff88881f4498a0
  R13: 0000000000000056 R14: 0000000000000009 R15: ffff888130c91ac0
  FS:  00007fec2cbb9700(0000) GS:ffff88881f440000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 00007fec1b060d80 CR3: 00000003acf94005 CR4: 00000000003706e0
  DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
  DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
  Call Trace:
   <IRQ>
   tcp_try_coalesce+0xeb/0x290
   ? tcp_parse_options+0x610/0x610
   ? mark_held_locks+0x79/0xa0
   tcp_queue_rcv+0x69/0x2f0
   tcp_rcv_established+0xa49/0xd40
   ? tcp_data_queue+0x18a0/0x18a0
   tcp_v6_do_rcv+0x1c9/0x880
   ? rt6_mtu_change_route+0x100/0x100
   tcp_v6_rcv+0x1624/0x1830
Signed-off-by: NJakub Kicinski <kuba@kernel.org>

Track skbs with only zerocopy data and avoid charging them to kernel
memory to correctly account the memory utilization for msg_zerocopy.
All of the data in such skbs is held in user pages which are already
accounted to user. Before this change, they are charged again in
kernel in __zerocopy_sg_from_iter. The charging in kernel is
excessive because data is not being copied into skb frags. This
excessive charging can lead to kernel going into memory pressure
state which impacts all sockets in the system adversely. Mark pure
zerocopy skbs with a SKBFL_PURE_ZEROCOPY flag and remove
charge/uncharge for data in such skbs.

Initially, an skb is marked pure zerocopy when it is empty and in
zerocopy path. skb can then change from a pure zerocopy skb to mixed
data skb (zerocopy and copy data) if it is at tail of write queue and
there is room available in it and non-zerocopy data is being sent in
the next sendmsg call. At this time sk_mem_charge is done for the pure
zerocopied data and the pure zerocopy flag is unmarked. We found that
this happens very rarely on workloads that pass MSG_ZEROCOPY.

A pure zerocopy skb can later be coalesced into normal skb if they are
next to each other in queue but this patch prevents coalescing from
happening. This avoids complexity of charging when skb downgrades from
pure zerocopy to mixed. This is also rare.

In sk_wmem_free_skb, if it is a pure zerocopy skb, an sk_mem_uncharge
for SKB_TRUESIZE(MAX_TCP_HEADER) is done for sk_mem_charge in
tcp_skb_entail for an skb without data.

Testing with the msg_zerocopy.c benchmark between two hosts(100G nics)
with zerocopy showed that before this patch the 'sock' variable in
memory.stat for cgroup2 that tracks sum of sk_forward_alloc,
sk_rmem_alloc and sk_wmem_queued is around 1822720 and with this
change it is 0. This is due to no charge to sk_forward_alloc for
zerocopy data and shows memory utilization for kernel is lowered.
Signed-off-by: NTalal Ahmad <talalahmad@google.com>
Acked-by: NArjun Roy <arjunroy@google.com>
Acked-by: NSoheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: NWillem de Bruijn <willemb@google.com>
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NJakub Kicinski <kuba@kernel.org>

sk_wmem_free_skb() is only used by TCP.

Rename it to make this clear, and move its declaration to
include/net/tcp.h
Signed-off-by: NTalal Ahmad <talalahmad@google.com>
Acked-by: NSoheil Hassas Yeganeh <soheil@google.com>
Acked-by: NArjun Roy <arjunroy@google.com>
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NJakub Kicinski <kuba@kernel.org>

Freshly allocated skbs have zero in skb->cb[] already.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Freshly allocated skbs have their csum field cleared already.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Setting skb->ip_summed to CHECKSUM_PARTIAL can be centralized
in tcp_stream_alloc_skb() and __mptcp_do_alloc_tx_skb()
instead of being done multiple times.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

TCP sendmsg() no longer puts payload in skb->head,
remove some dead code from tcp_collapse_retrans().
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

sk_stream_alloc_skb() is only used by TCP.

Rename it to make this clear, and move its declaration
to include/net/tcp.h
Signed-off-by: NEric Dumazet <edumazet@google.com>
Acked-by: NSoheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When user sets SO_RESERVE_MEM socket option, in order to utilize the
reserved memory when in memory pressure state, we adjust rcv_ssthresh
according to the available reserved memory for the socket, instead of
using 4 * advmss always.
Signed-off-by: NWei Wang <weiwan@google.com>
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In order to track CE marks per rate sample (one round trip), TCP needs a
per-skb header field to record the tp->delivered_ce count when the skb
was sent. To make space, we replace the "last_in_flight" field which is
used exclusively for NV congestion control. The stat needed by NV can be
alternatively approximated by existing stats tcp_sock delivered and
mss_cache.

This patch counts the number of packets delivered which have CE marks in
the rate sample, using similar approach of delivery accounting.

Cc: Lawrence Brakmo <brakmo@fb.com>
Signed-off-by: NYuchung Cheng <ycheng@google.com>
Acked-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NLuke Hsiao <lukehsiao@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add gfp_t mask as an input parameter to mem_cgroup_charge_skmem(),
to give more control to the networking stack and enable it to change
memcg charging behavior. In the future, the networking stack may decide
to avoid oom-kills when fallbacks are more appropriate.

One behavior change in mem_cgroup_charge_skmem() by this patch is to
avoid force charging by default and let the caller decide when and if
force charging is needed through the presence or absence of
__GFP_NOFAIL.
Signed-off-by: NWei Wang <weiwan@google.com>
Reviewed-by: NShakeel Butt <shakeelb@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

While TCP stack scales reasonably well, there is still one part that
can be used to DDOS it.

IPv6 Packet too big messages have to lookup/insert a new route,
and if abused by attackers, can easily put hosts under high stress,
with many cpus contending on a spinlock while one is stuck in fib6_run_gc()

ip6_protocol_deliver_rcu()
 icmpv6_rcv()
  icmpv6_notify()
   tcp_v6_err()
    tcp_v6_mtu_reduced()
     inet6_csk_update_pmtu()
      ip6_rt_update_pmtu()
       __ip6_rt_update_pmtu()
        ip6_rt_cache_alloc()
         ip6_dst_alloc()
          dst_alloc()
           ip6_dst_gc()
            fib6_run_gc()
             spin_lock_bh() ...

Some of our servers have been hit by malicious ICMPv6 packets
trying to _increase_ the MTU/MSS of TCP flows.

We believe these ICMPv6 packets are a result of a bug in one ISP stack,
since they were blindly sent back for _every_ (small) packet sent to them.

These packets are for one TCP flow:
09:24:36.266491 IP6 Addr1 > Victim ICMP6, packet too big, mtu 1460, length 1240
09:24:36.266509 IP6 Addr1 > Victim ICMP6, packet too big, mtu 1460, length 1240
09:24:36.316688 IP6 Addr1 > Victim ICMP6, packet too big, mtu 1460, length 1240
09:24:36.316704 IP6 Addr1 > Victim ICMP6, packet too big, mtu 1460, length 1240
09:24:36.608151 IP6 Addr1 > Victim ICMP6, packet too big, mtu 1460, length 1240

TCP stack can filter some silly requests :

1) MTU below IPV6_MIN_MTU can be filtered early in tcp_v6_err()
2) tcp_v6_mtu_reduced() can drop requests trying to increase current MSS.

This tests happen before the IPv6 routing stack is entered, thus
removing the potential contention and route exhaustion.

Note that IPv6 stack was performing these checks, but too late
(ie : after the route has been added, and after the potential
garbage collect war)

v2: fix typo caught by Martin, thanks !
v3: exports tcp_mtu_to_mss(), caught by David, thanks !

Fixes: 1da177e4 ("Linux-2.6.12-rc2")
Signed-off-by: NEric Dumazet <edumazet@google.com>
Reviewed-by: NMaciej Żenczykowski <maze@google.com>
Cc: Martin KaFai Lau <kafai@fb.com>
Acked-by: NMartin KaFai Lau <kafai@fb.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

TSQ provides a nice way to avoid bufferbloat on individual socket,
including retransmit packets. We can get rid of the old
heuristic:

	/* Do not sent more than we queued. 1/4 is reserved for possible
	 * copying overhead: fragmentation, tunneling, mangling etc.
	 */
	if (refcount_read(&sk->sk_wmem_alloc) >
	    min_t(u32, sk->sk_wmem_queued + (sk->sk_wmem_queued >> 2),
		  sk->sk_sndbuf))
		return -EAGAIN;

This heuristic was giving false positives according to Jakub,
whenever TX completions are delayed above RTT. (Ack packets
are processed by TCP stack before clones are orphaned/freed)
Signed-off-by: NEric Dumazet <edumazet@google.com>
Reported-by: NJakub Kicinski <kuba@kernel.org>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Jakub and Neil reported an increase of RTO timers whenever
TX completions are delayed a bit more (by increasing
NIC TX coalescing parameters)

Main issue is that TCP stack has a logic preventing a packet
being retransmit if the prior clone has not yet been
orphaned or freed.

This logic came with commit 1f3279ae ("tcp: avoid
retransmits of TCP packets hanging in host queues")

Thankfully, in the case skb_still_in_host_queue() detects
the initial clone is still in flight, it can use TSQ logic
that will eventually retry later, at the moment the clone
is freed or orphaned.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Reported-by: NNeil Spring <ntspring@fb.com>
Reported-by: NJakub Kicinski <kuba@kernel.org>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The TCP_USER_TIMEOUT is checked by the 0-window probe timer. As the
timer has backoff with a max interval of about two minutes, the
actual timeout for TCP_USER_TIMEOUT can be off by up to two minutes.

In this patch the TCP_USER_TIMEOUT is made more accurate by taking it
into account when computing the timer value for the 0-window probes.

This patch is similar to and builds on top of the one that made
TCP_USER_TIMEOUT accurate for RTOs in commit b701a99e ("tcp: Add
tcp_clamp_rto_to_user_timeout() helper to improve accuracy").

Fixes: 9721e709 ("tcp: simplify window probe aborting on USER_TIMEOUT")
Signed-off-by: NEnke Chen <enchen@paloaltonetworks.com>
Reviewed-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NEric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20210122191306.GA99540@localhost.localdomainSigned-off-by: NJakub Kicinski <kuba@kernel.org>

The TCP session does not terminate with TCP_USER_TIMEOUT when data
remain untransmitted due to zero window.

The number of unanswered zero-window probes (tcp_probes_out) is
reset to zero with incoming acks irrespective of the window size,
as described in tcp_probe_timer():

    RFC 1122 4.2.2.17 requires the sender to stay open indefinitely
    as long as the receiver continues to respond probes. We support
    this by default and reset icsk_probes_out with incoming ACKs.

This counter, however, is the wrong one to be used in calculating the
duration that the window remains closed and data remain untransmitted.
Thanks to Jonathan Maxwell <jmaxwell37@gmail.com> for diagnosing the
actual issue.

In this patch a new timestamp is introduced for the socket in order to
track the elapsed time for the zero-window probes that have not been
answered with any non-zero window ack.

Fixes: 9721e709 ("tcp: simplify window probe aborting on USER_TIMEOUT")
Reported-by: NWilliam McCall <william.mccall@gmail.com>
Co-developed-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NEnke Chen <enchen@paloaltonetworks.com>
Reviewed-by: NYuchung Cheng <ycheng@google.com>
Reviewed-by: NEric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20210115223058.GA39267@localhost.localdomainSigned-off-by: NJakub Kicinski <kuba@kernel.org>

Move skb_set_hash_from_sk s.t. it's called after instead of before
tcp_event_data_sent is called. This enables congestion control
modules to change the socket hash right before restarting from
idle (via the TX_START congestion event).
Signed-off-by: NYuchung Cheng <ycheng@google.com>
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NNeal Cardwell <ncardwell@google.com>
Link: https://lore.kernel.org/r/20210111230552.2704579-1-ycheng@google.comSigned-off-by: NJakub Kicinski <kuba@kernel.org>

When cwnd is not a multiple of the TSO skb size of N*MSS, we can get
into persistent scenarios where we have the following sequence:

(1) ACK for full-sized skb of N*MSS arrives
  -> tcp_write_xmit() transmit full-sized skb with N*MSS
  -> move pacing release time forward
  -> exit tcp_write_xmit() because pacing time is in the future

(2) TSQ callback or TCP internal pacing timer fires
  -> try to transmit next skb, but TSO deferral finds remainder of
     available cwnd is not big enough to trigger an immediate send
     now, so we defer sending until the next ACK.

(3) repeat...

So we can get into a case where we never mark ourselves as
cwnd-limited for many seconds at a time, even with
bulk/infinite-backlog senders, because:

o In case (1) above, every time in tcp_write_xmit() we have enough
cwnd to send a full-sized skb, we are not fully using the cwnd
(because cwnd is not a multiple of the TSO skb size). So every time we
send data, we are not cwnd limited, and so in the cwnd-limited
tracking code in tcp_cwnd_validate() we mark ourselves as not
cwnd-limited.

o In case (2) above, every time in tcp_write_xmit() that we try to
transmit the "remainder" of the cwnd but defer, we set the local
variable is_cwnd_limited to true, but we do not send any packets, so
sent_pkts is zero, so we don't call the cwnd-limited logic to update
tp->is_cwnd_limited.

Fixes: ca8a2263 ("tcp: make cwnd-limited checks measurement-based, and gentler")
Reported-by: NIngemar Johansson <ingemar.s.johansson@ericsson.com>
Signed-off-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NYuchung Cheng <ycheng@google.com>
Acked-by: NSoheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: NEric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20201209035759.1225145-1-ncardwell.kernel@gmail.comSigned-off-by: NJakub Kicinski <kuba@kernel.org>

OoO handling attempts to detect when packet is out-of-window by testing
current ack sequence and remaining space vs. sequence number.

This doesn't work reliably. Store the highest allowed sequence number
that we've announced and use it to detect oow packets.

Do this when mptcp options get written to the packet (wire format).
For this to work we need to move the write_options call until after
stack selected a new tcp window.
Acked-by: NPaolo Abeni <pabeni@redhat.com>
Signed-off-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NMat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: NJakub Kicinski <kuba@kernel.org>

In preparation for unconditionally passing the
struct tasklet_struct pointer to all tasklet
callbacks, switch to using the new tasklet_setup()
and from_tasklet() to pass the tasklet pointer explicitly.
Signed-off-by: NRomain Perier <romain.perier@gmail.com>
Signed-off-by: NAllen Pais <apais@linux.microsoft.com>
Signed-off-by: NJakub Kicinski <kuba@kernel.org>

When the TCP stack splits a packet on the write queue, the tail
half currently lose the associated skb extensions, and will not
carry the DSM on the wire.

The above does not cause functional problems and is allowed by
the RFC, but interact badly with GRO and RX coalescing, as possible
candidates for aggregation will carry different TCP options.

This change tries to improve the MPTCP behavior, propagating the
skb extensions on split.

Additionally, we must prevent the MPTCP stack from updating the
mapping after the split occur: that will both violate the RFC and
fool the reader.
Signed-off-by: NPaolo Abeni <pabeni@redhat.com>
Signed-off-by: NMat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: NJakub Kicinski <kuba@kernel.org>

Whenever host is under very high memory pressure,
__tcp_send_ack() skb allocation fails, and we setup
a 200 ms (TCP_DELACK_MAX) timer before retrying.

On hosts with high number of TCP sockets, we can spend
considerable amount of cpu cycles in these attempts,
add high pressure on various spinlocks in mm-layer,
ultimately blocking threads attempting to free space
from making any progress.

This patch adds standard exponential backoff to avoid
adding fuel to the fire.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Acked-by: NSoheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>