When the packet pointed to by retransmit_skb_hint is unlinked by ACK,
retransmit_skb_hint will be set to NULL in tcp_clean_rtx_queue().
If packet loss is detected at this time, retransmit_skb_hint will be set
to point to the current packet loss in tcp_verify_retransmit_hint(),
then the packets that were previously marked lost but not retransmitted
due to the restriction of cwnd will be skipped and cannot be
retransmitted.
To fix this, when retransmit_skb_hint is NULL, retransmit_skb_hint can
be reset only after all marked lost packets are retransmitted
(retrans_out >= lost_out), otherwise we need to traverse from
tcp_rtx_queue_head in tcp_xmit_retransmit_queue().
Packetdrill to demonstrate:
// Disable RACK and set max_reordering to keep things simple
0 `sysctl -q net.ipv4.tcp_recovery=0`
+0 `sysctl -q net.ipv4.tcp_max_reordering=3`
// Establish a connection
+0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
+0 bind(3, ..., ...) = 0
+0 listen(3, 1) = 0
+.1 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
+0 > S. 0:0(0) ack 1 <...>
+.01 < . 1:1(0) ack 1 win 257
+0 accept(3, ..., ...) = 4
// Send 8 data segments
+0 write(4, ..., 8000) = 8000
+0 > P. 1:8001(8000) ack 1
// Enter recovery and 1:3001 is marked lost
+.01 < . 1:1(0) ack 1 win 257 <sack 3001:4001,nop,nop>
+0 < . 1:1(0) ack 1 win 257 <sack 5001:6001 3001:4001,nop,nop>
+0 < . 1:1(0) ack 1 win 257 <sack 5001:7001 3001:4001,nop,nop>
// Retransmit 1:1001, now retransmit_skb_hint points to 1001:2001
+0 > . 1:1001(1000) ack 1
// 1001:2001 was ACKed causing retransmit_skb_hint to be set to NULL
+.01 < . 1:1(0) ack 2001 win 257 <sack 5001:8001 3001:4001,nop,nop>
// Now retransmit_skb_hint points to 4001:5001 which is now marked lost
// BUG: 2001:3001 was not retransmitted
+0 > . 2001:3001(1000) ack 1
Signed-off-by: Pengcheng Yang <yangpc@wangsu.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Tested-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When we receive a D-SACK, where the sequence number satisfies:
undo_marker <= start_seq < end_seq <= prior_snd_una
we consider this is a valid D-SACK and tcp_is_sackblock_valid()
returns true, then this D-SACK is discarded as "old stuff",
but the variable first_sack_index is not marked as negative
in tcp_sacktag_write_queue().
If this D-SACK also carries a SACK that needs to be processed
(for example, the previous SACK segment was lost), this SACK
will be treated as a D-SACK in the following processing of
tcp_sacktag_write_queue(), which will eventually lead to
incorrect updates of undo_retrans and reordering.
Fixes: fd6dad616d ("[TCP]: Earlier SACK block verification & simplify access to them")
Signed-off-by: Pengcheng Yang <yangpc@wangsu.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The TCPI_OPT_SYN_DATA bit as part of tcpi_options currently reports whether
or not data-in-SYN was ack'd on both the client and server side. We'd like
to gather more information on the client-side in the failure case in order
to indicate the reason for the failure. This can be useful for not only
debugging TFO, but also for creating TFO socket policies. For example, if
a middle box removes the TFO option or drops a data-in-SYN, we can
can detect this case, and turn off TFO for these connections saving the
extra retransmits.
The newly added tcpi_fastopen_client_fail status is 2 bits and has the
following 4 states:
1) TFO_STATUS_UNSPEC
Catch-all state which includes when TFO is disabled via black hole
detection, which is indicated via LINUX_MIB_TCPFASTOPENBLACKHOLE.
2) TFO_COOKIE_UNAVAILABLE
If TFO_CLIENT_NO_COOKIE mode is off, this state indicates that no cookie
is available in the cache.
3) TFO_DATA_NOT_ACKED
Data was sent with SYN, we received a SYN/ACK but it did not cover the data
portion. Cookie is not accepted by server because the cookie may be invalid
or the server may be overloaded.
4) TFO_SYN_RETRANSMITTED
Data was sent with SYN, we received a SYN/ACK which did not cover the data
after at least 1 additional SYN was sent (without data). It may be the case
that a middle-box is dropping data-in-SYN packets. Thus, it would be more
efficient to not use TFO on this connection to avoid extra retransmits
during connection establishment.
These new fields do not cover all the cases where TFO may fail, but other
failures, such as SYN/ACK + data being dropped, will result in the
connection not becoming established. And a connection blackhole after
session establishment shows up as a stalled connection.
Signed-off-by: Jason Baron <jbaron@akamai.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Christoph Paasch <cpaasch@apple.com>
Cc: Yuchung Cheng <ycheng@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the sake of tcp_poll(), there are few places where we fetch
sk->sk_sndbuf while this field can change from IRQ or other cpu.
We need to add READ_ONCE() annotations, and also make sure write
sides use corresponding WRITE_ONCE() to avoid store-tearing.
Note that other transports probably need similar fixes.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the sake of tcp_poll(), there are few places where we fetch
sk->sk_rcvbuf while this field can change from IRQ or other cpu.
We need to add READ_ONCE() annotations, and also make sure write
sides use corresponding WRITE_ONCE() to avoid store-tearing.
Note that other transports probably need similar fixes.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There two places where we fetch tp->urg_seq while
this field can change from IRQ or other cpu.
We need to add READ_ONCE() annotations, and also make
sure write side use corresponding WRITE_ONCE() to avoid
store-tearing.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There are few places where we fetch tp->copied_seq while
this field can change from IRQ or other cpu.
We need to add READ_ONCE() annotations, and also make
sure write sides use corresponding WRITE_ONCE() to avoid
store-tearing.
Note that tcp_inq_hint() was already using READ_ONCE(tp->copied_seq)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There are few places where we fetch tp->rcv_nxt while
this field can change from IRQ or other cpu.
We need to add READ_ONCE() annotations, and also make
sure write sides use corresponding WRITE_ONCE() to avoid
store-tearing.
Note that tcp_inq_hint() was already using READ_ONCE(tp->rcv_nxt)
syzbot reported :
BUG: KCSAN: data-race in tcp_poll / tcp_queue_rcv
write to 0xffff888120425770 of 4 bytes by interrupt on cpu 0:
tcp_rcv_nxt_update net/ipv4/tcp_input.c:3365 [inline]
tcp_queue_rcv+0x180/0x380 net/ipv4/tcp_input.c:4638
tcp_rcv_established+0xbf1/0xf50 net/ipv4/tcp_input.c:5616
tcp_v4_do_rcv+0x381/0x4e0 net/ipv4/tcp_ipv4.c:1542
tcp_v4_rcv+0x1a03/0x1bf0 net/ipv4/tcp_ipv4.c:1923
ip_protocol_deliver_rcu+0x51/0x470 net/ipv4/ip_input.c:204
ip_local_deliver_finish+0x110/0x140 net/ipv4/ip_input.c:231
NF_HOOK include/linux/netfilter.h:305 [inline]
NF_HOOK include/linux/netfilter.h:299 [inline]
ip_local_deliver+0x133/0x210 net/ipv4/ip_input.c:252
dst_input include/net/dst.h:442 [inline]
ip_rcv_finish+0x121/0x160 net/ipv4/ip_input.c:413
NF_HOOK include/linux/netfilter.h:305 [inline]
NF_HOOK include/linux/netfilter.h:299 [inline]
ip_rcv+0x18f/0x1a0 net/ipv4/ip_input.c:523
__netif_receive_skb_one_core+0xa7/0xe0 net/core/dev.c:5004
__netif_receive_skb+0x37/0xf0 net/core/dev.c:5118
netif_receive_skb_internal+0x59/0x190 net/core/dev.c:5208
napi_skb_finish net/core/dev.c:5671 [inline]
napi_gro_receive+0x28f/0x330 net/core/dev.c:5704
receive_buf+0x284/0x30b0 drivers/net/virtio_net.c:1061
read to 0xffff888120425770 of 4 bytes by task 7254 on cpu 1:
tcp_stream_is_readable net/ipv4/tcp.c:480 [inline]
tcp_poll+0x204/0x6b0 net/ipv4/tcp.c:554
sock_poll+0xed/0x250 net/socket.c:1256
vfs_poll include/linux/poll.h:90 [inline]
ep_item_poll.isra.0+0x90/0x190 fs/eventpoll.c:892
ep_send_events_proc+0x113/0x5c0 fs/eventpoll.c:1749
ep_scan_ready_list.constprop.0+0x189/0x500 fs/eventpoll.c:704
ep_send_events fs/eventpoll.c:1793 [inline]
ep_poll+0xe3/0x900 fs/eventpoll.c:1930
do_epoll_wait+0x162/0x180 fs/eventpoll.c:2294
__do_sys_epoll_pwait fs/eventpoll.c:2325 [inline]
__se_sys_epoll_pwait fs/eventpoll.c:2311 [inline]
__x64_sys_epoll_pwait+0xcd/0x170 fs/eventpoll.c:2311
do_syscall_64+0xcf/0x2f0 arch/x86/entry/common.c:296
entry_SYSCALL_64_after_hwframe+0x44/0xa9
Reported by Kernel Concurrency Sanitizer on:
CPU: 1 PID: 7254 Comm: syz-fuzzer Not tainted 5.3.0+ #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Both tcp_v4_err() and tcp_v6_err() do the following operations
while they do not own the socket lock :
fastopen = tp->fastopen_rsk;
snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
The problem is that without appropriate barrier, the compiler
might reload tp->fastopen_rsk and trigger a NULL deref.
request sockets are protected by RCU, we can simply add
the missing annotations and barriers to solve the issue.
Fixes: 168a8f5805 ("tcp: TCP Fast Open Server - main code path")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For receive-heavy cases on the server-side, we want to track the
connection quality for individual client IPs. This counter, similar to
the existing system-wide TCPOFOQueue counter in /proc/net/netstat,
tracks out-of-order packet reception. By providing this counter in
TCP_INFO, it will allow understanding to what degree receive-heavy
sockets are experiencing out-of-order delivery and packet drops
indicating congestion.
Please note that this is similar to the counter in NetBSD TCP_INFO, and
has the same name.
Also note that we avoid increasing the size of the tcp_sock struct by
taking advantage of a hole.
Signed-off-by: Thomas Higdon <tph@fb.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix tcp_ecn_withdraw_cwr() to clear the correct bit:
TCP_ECN_QUEUE_CWR.
Rationale: basically, TCP_ECN_DEMAND_CWR is a bit that is purely about
the behavior of data receivers, and deciding whether to reflect
incoming IP ECN CE marks as outgoing TCP th->ece marks. The
TCP_ECN_QUEUE_CWR bit is purely about the behavior of data senders,
and deciding whether to send CWR. The tcp_ecn_withdraw_cwr() function
is only called from tcp_undo_cwnd_reduction() by data senders during
an undo, so it should zero the sender-side state,
TCP_ECN_QUEUE_CWR. It does not make sense to stop the reflection of
incoming CE bits on incoming data packets just because outgoing
packets were spuriously retransmitted.
The bug has been reproduced with packetdrill to manifest in a scenario
with RFC3168 ECN, with an incoming data packet with CE bit set and
carrying a TCP timestamp value that causes cwnd undo. Before this fix,
the IP CE bit was ignored and not reflected in the TCP ECE header bit,
and sender sent a TCP CWR ('W') bit on the next outgoing data packet,
even though the cwnd reduction had been undone. After this fix, the
sender properly reflects the CE bit and does not set the W bit.
Note: the bug actually predates 2005 git history; this Fixes footer is
chosen to be the oldest SHA1 I have tested (from Sep 2007) for which
the patch applies cleanly (since before this commit the code was in a
.h file).
Fixes: bdf1ee5d3b ("[TCP]: Move code from tcp_ecn.h to tcp*.c and tcp.h & remove it")
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch allows generation of a SYN cookie before an SKB has been
allocated, as is the case at XDP.
Signed-off-by: Petar Penkov <ppenkov@google.com>
Reviewed-by: Lorenz Bauer <lmb@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
This allows us to call this function before an SKB has been
allocated.
Signed-off-by: Petar Penkov <ppenkov@google.com>
Reviewed-by: Lorenz Bauer <lmb@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Performance impact should be minimal because it's under a new
BPF_SOCK_OPS_RTT_CB_FLAG flag that has to be explicitly enabled.
Suggested-by: Eric Dumazet <edumazet@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Priyaranjan Jha <priyarjha@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Jonathan Looney reported that TCP can trigger the following crash
in tcp_shifted_skb() :
BUG_ON(tcp_skb_pcount(skb) < pcount);
This can happen if the remote peer has advertized the smallest
MSS that linux TCP accepts : 48
An skb can hold 17 fragments, and each fragment can hold 32KB
on x86, or 64KB on PowerPC.
This means that the 16bit witdh of TCP_SKB_CB(skb)->tcp_gso_segs
can overflow.
Note that tcp_sendmsg() builds skbs with less than 64KB
of payload, so this problem needs SACK to be enabled.
SACK blocks allow TCP to coalesce multiple skbs in the retransmit
queue, thus filling the 17 fragments to maximal capacity.
CVE-2019-11477 -- u16 overflow of TCP_SKB_CB(skb)->tcp_gso_segs
Fixes: 832d11c5cd ("tcp: Try to restore large SKBs while SACK processing")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Jonathan Looney <jtl@netflix.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Tyler Hicks <tyhicks@canonical.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Bruce Curtis <brucec@netflix.com>
Cc: Jonathan Lemon <jonathan.lemon@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Deferred static key clean_acked_data_enabled uses the deferred
variants of dec and flush. Do the same for inc.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 794200d662 ("tcp: undo cwnd on Fast Open spurious SYNACK
retransmit") may cause tcp_fastretrans_alert() to warn about pending
retransmission in Open state. This is triggered when the Fast Open
server both sends data and has spurious SYNACK retransmission during
the handshake, and the data packets were lost or reordered.
The root cause is a bit complicated:
(1) Upon receiving SYN-data: a full socket is created with
snd_una = ISN + 1 by tcp_create_openreq_child()
(2) On SYNACK timeout the server/sender enters CA_Loss state.
(3) Upon receiving the final ACK to complete the handshake, sender
does not mark FLAG_SND_UNA_ADVANCED since (1)
Sender then calls tcp_process_loss since state is CA_loss by (2)
(4) tcp_process_loss() does not invoke undo operations but instead
mark REXMIT_LOST to force retransmission
(5) tcp_rcv_synrecv_state_fastopen() calls tcp_try_undo_loss(). It
changes state to CA_Open but has positive tp->retrans_out
(6) Next ACK triggers the WARN_ON in tcp_fastretrans_alert()
The step that goes wrong is (4) where the undo operation should
have been invoked because the ACK successfully acknowledged the
SYN sequence. This fixes that by specifically checking undo
when the SYN-ACK sequence is acknowledged. Then after
tcp_process_loss() the state would be further adjusted based
in tcp_fastretrans_alert() to avoid triggering the warning in (6).
Fixes: 794200d662 ("tcp: undo cwnd on Fast Open spurious SYNACK retransmit")
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The TCP option parsing routines in tcp_parse_options function could
read one byte out of the buffer of the TCP options.
1 while (length > 0) {
2 int opcode = *ptr++;
3 int opsize;
4
5 switch (opcode) {
6 case TCPOPT_EOL:
7 return;
8 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
9 length--;
10 continue;
11 default:
12 opsize = *ptr++; //out of bound access
If length = 1, then there is an access in line2.
And another access is occurred in line 12.
This would lead to out-of-bound access.
Therefore, in the patch we check that the available data length is
larger enough to pase both TCP option code and size.
Signed-off-by: Young Xiao <92siuyang@gmail.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit c7d13c8faa ("tcp: properly track retry time on
passive Fast Open") sets the start of SYNACK retransmission
time on passive Fast Open in "retrans_stamp". However the
timestamp is not reset upon the handshake has completed. As a
result, future data packet retransmission may not update it in
tcp_retransmit_skb(). This may lead to socket aborting earlier
unexpectedly by retransmits_timed_out() since retrans_stamp remains
the SYNACK rtx time.
This bug only manifests on passive TFO sender that a) suffered
SYNACK timeout and then b) stalls on very first loss recovery. Any
successful loss recovery would reset the timestamp to avoid this
issue.
Fixes: c7d13c8faa ("tcp: properly track retry time on passive Fast Open")
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
User space can flip the clean_acked_data_enabled static branch
on and off with TLS offload when CONFIG_TLS_DEVICE is enabled.
jump_label.h suggests we use the delayed version in this case.
Deferred branches now also don't take the branch mutex on
decrement, so we avoid potential locking issues.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Relocate the congestion window initialization from tcp_init_metrics()
to tcp_init_transfer() to improve code readability.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use a helper to consolidate two identical code block for passive TFO.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch makes passive Fast Open reverts the cwnd to default
initial cwnd (10 packets) if the SYNACK timeout is spurious.
Passive Fast Open uses a full socket during handshake so it can
use the existing undo logic to detect spurious retransmission
by recording the first SYNACK timeout in key state variable
retrans_stamp. Upon receiving the ACK of the SYNACK, if the socket
has sent some data before the timeout, the spurious timeout
is detected by tcp_try_undo_recovery() in tcp_process_loss()
in tcp_ack().
But if the socket has not send any data yet, tcp_ack() does not
execute the undo code since no data is acknowledged. The fix is to
check such case explicitly after tcp_ack() during the ACK processing
in SYN_RECV state. In addition this is checked in FIN_WAIT_1 state
in case the server closes the socket before handshake completes.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Linux implements RFC6298 and use an initial congestion window
of 1 upon establishing the connection if the SYNACK packet is
retransmitted 2 or more times. In cellular networks SYNACK timeouts
are often spurious if the wireless radio was dormant or idle. Also
some network path is longer than the default SYNACK timeout. In
both cases falsely starting with a minimal cwnd are detrimental
to performance.
This patch avoids doing so when the final ACK's TCP timestamp
indicates the original SYNACK was delivered. It remembers the
original SYNACK timestamp when SYNACK timeout has occurred and
re-uses the function to detect spurious SYN timeout conveniently.
Note that a server may receives multiple SYNs from and immediately
retransmits SYNACKs without any SYNACK timeout. This often happens
on when the client SYNs have timed out due to wireless delay
above. In this case since the server will still use the default
initial congestion (e.g. 10) because tp->undo_marker is reset in
tcp_init_metrics(). This is an intentional design because packets
are not lost but delayed.
This patch only covers regular TCP passive open. Fast Open is
supported in the next patch.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Detecting spurious SYNACK timeout using timestamp option requires
recording the exact SYNACK skb timestamp. Previously the SYNACK
sent timestamp was stamped slightly earlier before the skb
was transmitted. This patch uses the SYNACK skb transmission
timestamp directly.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Linux implements RFC6298 and use an initial congestion window of 1
upon establishing the connection if the SYN packet is retransmitted 2
or more times. In cellular networks SYN timeouts are often spurious
if the wireless radio was dormant or idle. Also some network path
is longer than the default SYN timeout. Having a minimal cwnd on
both cases are detrimental to TCP startup performance.
This patch extends TCP undo feature (RFC3522 aka TCP Eifel) to detect
spurious SYN timeout via TCP timestamps. Since tp->retrans_stamp
records the initial SYN timestamp instead of first retransmission, we
have to implement a different undo code additionally. The detection
also must happen before tcp_ack() as retrans_stamp is reset when
SYN is acknowledged.
Note this patch covers both active regular and fast open.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Previously if an active TCP open has SYN timeout, it always undo the
cwnd upon receiving the SYNACK. This is because tcp_clean_rtx_queue
would reset tp->retrans_stamp when SYN is acked, which fools then
tcp_try_undo_loss and tcp_packet_delayed. Addressing this issue is
required to properly support undo for spurious SYN timeout.
Fixing this is tricky -- for active TCP open tp->retrans_stamp
records the time when the handshake starts, not the first
retransmission time as the name may suggest. The simplest fix is
for tcp_packet_delayed to ensure it is valid before comparing with
other timestamp.
One side effect of this change is active TCP Fast Open that incurred
SYN timeout. Upon receiving a SYN-ACK that only acknowledged
the SYN, it would immediately retransmit unacknowledged data in
tcp_ack() because the data is marked lost after SYN timeout. But
the retransmission would have an incorrect ack sequence number since
rcv_nxt has not been updated yet tcp_rcv_synsent_state_process(), the
retransmission needs to properly handed by tcp_rcv_fastopen_synack()
like before.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For some reason, tcp_grow_window() correctly tests if enough room
is present before attempting to increase tp->rcv_ssthresh,
but does not prevent it to grow past tcp_space()
This is causing hard to debug issues, like failing
the (__tcp_select_window(sk) >= tp->rcv_wnd) test
in __tcp_ack_snd_check(), causing ACK delays and possibly
slow flows.
Depending on tcp_rmem[2], MTU, skb->len/skb->truesize ratio,
we can see the problem happening on "netperf -t TCP_RR -- -r 2000,2000"
after about 60 round trips, when the active side no longer sends
immediate acks.
This bug predates git history.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Wei Wang <weiwan@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Linux currently disable ECN for incoming connections when the SYN
requests ECN and the IP header has ECT(0)/ECT(1) set, as some
networks were reportedly mangling the ToS byte, hence could later
trigger false congestion notifications.
RFC8311 ยง4.3 relaxes RFC3168's requirements such that ECT can be set
one TCP control packets (including SYNs). The main benefit of this
is the decreased probability of losing a SYN in a congested
ECN-capable network (i.e., it avoids the initial 1s timeout).
Additionally, this allows the development of newer TCP extensions,
such as AccECN.
This patch relaxes the previous check, by enabling ECN on incoming
connections using SYN+ECT if at least one bit of the reserved flags
of the TCP header is set. Such bit would indicate that the sender of
the SYN is using a newer TCP feature than what the host implements,
such as AccECN, and is thus implementing RFC8311. This enables
end-hosts not supporting such extensions to still negociate ECN, and
to have some of the benefits of using ECN on control packets.
Signed-off-by: Olivier Tilmans <olivier.tilmans@nokia-bell-labs.com>
Suggested-by: Bob Briscoe <research@bobbriscoe.net>
Cc: Koen De Schepper <koen.de_schepper@nokia-bell-labs.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since the request socket is created locally, it'd make more sense to
use reqsk_free() instead of reqsk_put() in TFO and syncookies' error
path.
However, tcp_get_cookie_sock() may set ->rsk_refcnt before freeing the
socket; tcp_conn_request() may also have non-null ->rsk_refcnt because
of tcp_try_fastopen(). In both cases 'req' hasn't been exposed
to the outside world and is safe to free immediately, but that'd
trigger the WARN_ON_ONCE in reqsk_free().
Define __reqsk_free() for these situations where we know nobody's
referencing the socket, even though ->rsk_refcnt might be non-null.
Now we can consolidate the error path of tcp_get_cookie_sock() and
tcp_conn_request().
Signed-off-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 7716682cc5 ("tcp/dccp: fix another race at listener
dismantle") let inet_csk_reqsk_queue_add() fail, and adjusted
{tcp,dccp}_check_req() accordingly. However, TFO and syncookies
weren't modified, thus leaking allocated resources on error.
Contrary to tcp_check_req(), in both syncookies and TFO cases,
we need to drop the request socket. Also, since the child socket is
created with inet_csk_clone_lock(), we have to unlock it and drop an
extra reference (->sk_refcount is initially set to 2 and
inet_csk_reqsk_queue_add() drops only one ref).
For TFO, we also need to revert the work done by tcp_try_fastopen()
(with reqsk_fastopen_remove()).
Fixes: 7716682cc5 ("tcp/dccp: fix another race at listener dismantle")
Signed-off-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
parameter state in the tcp_sacktag_bsearch() is not used.
So, it can be removed.
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of using pingpong as a single bit information, we refactor the
code to treat it as a counter. When interactive session is detected,
we set pingpong count to TCP_PINGPONG_THRESH. And when pingpong count
is >= TCP_PINGPONG_THRESH, we consider the session in pingpong mode.
This patch is a pure refactor and sets foundation for the next patch.
This patch itself does not change any pingpong logic.
Signed-off-by: Wei Wang <weiwan@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Neal pointed out that non sack flows might suffer from ACK compression
added in the following patch ("tcp: implement coalescing on backlog queue")
Instead of tweaking tcp_add_backlog() we can take into
account how many ACK were coalesced, this information
will be available in skb_shinfo(skb)->gso_segs
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Trivial conflict in net/core/filter.c, a locally computed
'sdif' is now an argument to the function.
Signed-off-by: David S. Miller <davem@davemloft.net>
Only one caller needs to pull TCP headers, so lets
move __skb_pull() to the caller side.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When a qdisc setup including pacing FQ is dismantled and recreated,
some TCP packets are sent earlier than instructed by TCP stack.
TCP can be fooled when ACK comes back, because the following
operation can return a negative value.
tcp_time_stamp(tp) - tp->rx_opt.rcv_tsecr;
Some paths in TCP stack were not dealing properly with this,
this patch addresses four of them.
Fixes: ab408b6dc7 ("tcp: switch tcp and sch_fq to new earliest departure time model")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jean-Louis reported a TCP regression and bisected to recent SACK
compression.
After a loss episode (receiver not able to keep up and dropping
packets because its backlog is full), linux TCP stack is sending
a single SACK (DUPACK).
Sender waits a full RTO timer before recovering losses.
While RFC 6675 says in section 5, "Algorithm Details",
(2) If DupAcks < DupThresh but IsLost (HighACK + 1) returns true --
indicating at least three segments have arrived above the current
cumulative acknowledgment point, which is taken to indicate loss
-- go to step (4).
...
(4) Invoke fast retransmit and enter loss recovery as follows:
there are old TCP stacks not implementing this strategy, and
still counting the dupacks before starting fast retransmit.
While these stacks probably perform poorly when receivers implement
LRO/GRO, we should be a little more gentle to them.
This patch makes sure we do not enable SACK compression unless
3 dupacks have been sent since last rcv_nxt update.
Ideally we should even rearm the timer to send one or two
more DUPACK if no more packets are coming, but that will
be work aiming for linux-4.21.
Many thanks to Jean-Louis for bisecting the issue, providing
packet captures and testing this patch.
Fixes: 5d9f4262b7 ("tcp: add SACK compression")
Reported-by: Jean-Louis Dupond <jean-louis@dupond.be>
Tested-by: Jean-Louis Dupond <jean-louis@dupond.be>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
During tcp coalescing ensure that the skb hardware timestamp refers to the
highest sequence number data.
Previously only the software timestamp was updated during coalescing.
Signed-off-by: Stephen Mallon <stephen.mallon@sydney.edu.au>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Bitwise operation is a little faster.
So I replace after() with using the flag FLAG_SND_UNA_ADVANCED as it is
already set before.
In addtion, there's another similar improvement in tcp_cwnd_reduction().
Cc: Joe Perches <joe@perches.com>
Suggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
With EDT model, SRTT no longer is inflated by pacing delays.
This means that RTO and some other xmit timers might be setup
incorrectly. This is particularly visible with either :
- Very small enforced pacing rates (SO_MAX_PACING_RATE)
- Reduced rto (from the default 200 ms)
This can lead to TCP flows aborts in the worst case,
or spurious retransmits in other cases.
For example, this session gets far more throughput
than the requested 80kbit :
$ netperf -H 127.0.0.2 -l 100 -- -q 10000
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 127.0.0.2 () port 0 AF_INET
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
540000 262144 262144 104.00 2.66
With the fix :
$ netperf -H 127.0.0.2 -l 100 -- -q 10000
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 127.0.0.2 () port 0 AF_INET
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
540000 262144 262144 104.00 0.12
EDT allows for better control of rtx timers, since TCP has
a better idea of the earliest departure time of each skb
in the rtx queue. We only have to eventually add to the
timer the difference of the EDT time with current time.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Minor conflict in net/core/rtnetlink.c, David Ahern's bug fix in 'net'
overlapped the renaming of a netlink attribute in net-next.
Signed-off-by: David S. Miller <davem@davemloft.net>
Previously receiver buffer auto-tuning starts after receiving
one advertised window amount of data. After the initial receiver
buffer was raised by patch a337531b94 ("tcp: up initial rmem to
128KB and SYN rwin to around 64KB"), the reciver buffer may take
too long to start raising. To address this issue, this patch lowers
the initial bytes expected to receive roughly the expected sender's
initial window.
Fixes: a337531b94 ("tcp: up initial rmem to 128KB and SYN rwin to around 64KB")
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Wei Wang <weiwan@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In normal SYN processing, packets are handled without listener
lock and in RCU protected ingress path.
But syzkaller is known to be able to trick us and SYN
packets might be processed in process context, after being
queued into socket backlog.
In commit 06f877d613 ("tcp/dccp: fix other lockdep splats
accessing ireq_opt") I made a very stupid fix, that happened
to work mostly because of the regular path being RCU protected.
Really the thing protecting ireq->ireq_opt is RCU read lock,
and the pseudo request refcnt is not relevant.
This patch extends what I did in commit 449809a66c ("tcp/dccp:
block BH for SYN processing") by adding an extra rcu_read_{lock|unlock}
pair in the paths that might be taken when processing SYN from
socket backlog (thus possibly in process context)
Fixes: 06f877d613 ("tcp/dccp: fix other lockdep splats accessing ireq_opt")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Previously TCP initial receive buffer is ~87KB by default and
the initial receive window is ~29KB (20 MSS). This patch changes
the two numbers to 128KB and ~64KB (rounding down to the multiples
of MSS) respectively. The patch also simplifies the calculations s.t.
the two numbers are directly controlled by sysctl tcp_rmem[1]:
1) Initial receiver buffer budget (sk_rcvbuf): while this should
be configured via sysctl tcp_rmem[1], previously tcp_fixup_rcvbuf()
always override and set a larger size when a new connection
establishes.
2) Initial receive window in SYN: previously it is set to 20
packets if MSS <= 1460. The number 20 was based on the initial
congestion window of 10: the receiver needs twice amount to
avoid being limited by the receive window upon out-of-order
delivery in the first window burst. But since this only
applies if the receiving MSS <= 1460, connection using large MTU
(e.g. to utilize receiver zero-copy) may be limited by the
receive window.
With this patch TCP memory configuration is more straight-forward and
more properly sized to modern high-speed networks by default. Several
popular stacks have been announcing 64KB rwin in SYNs as well.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Wei Wang <weiwan@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pengcheng Yang