kernel/linux.git/include/net/tcp.h, branch v4.14.307

tcp/udp: Make early_demux back namespacified.

2022-11-10T14:47:22+00:00

commit 11052589cf5c0bab3b4884d423d5f60c38fcf25d upstream. Commit e21145a9871a ("ipv4: namespacify ip_early_demux sysctl knob") made it possible to enable/disable early_demux on a per-netns basis. Then, we introduced two knobs, tcp_early_demux and udp_early_demux, to switch it for TCP/UDP in commit dddb64bcb346 ("net: Add sysctl to toggle early demux for tcp and udp"). However, the .proc_handler() was wrong and actually disabled us from changing the behaviour in each netns. We can execute early_demux if net.ipv4.ip_early_demux is on and each proto .early_demux() handler is not NULL. When we toggle (tcp|udp)_early_demux, the change itself is saved in each netns variable, but the .early_demux() handler is a global variable, so the handler is switched based on the init_net's sysctl variable. Thus, netns (tcp|udp)_early_demux knobs have nothing to do with the logic. Whether we CAN execute proto .early_demux() is always decided by init_net's sysctl knob, and whether we DO it or not is by each netns ip_early_demux knob. This patch namespacifies (tcp|udp)_early_demux again. For now, the users of the .early_demux() handler are TCP and UDP only, and they are called directly to avoid retpoline. So, we can remove the .early_demux() handler from inet6?_protos and need not dereference them in ip6?_rcv_finish_core(). If another proto needs .early_demux(), we can restore it at that time. Fixes: dddb64bcb346 ("net: Add sysctl to toggle early demux for tcp and udp") Signed-off-by: Kuniyuki Iwashima Link: https://lore.kernel.org/r/20220713175207.7727-1-kuniyu@amazon.com Signed-off-by: Jakub Kicinski Signed-off-by: Greg Kroah-Hartman

tcp: fix tcp_cwnd_validate() to not forget is_cwnd_limited

2022-10-26T11:17:00+00:00

[ Upstream commit f4ce91ce12a7c6ead19b128ffa8cff6e3ded2a14 ] This commit fixes a bug in the tracking of max_packets_out and is_cwnd_limited. This bug can cause the connection to fail to remember that is_cwnd_limited is true, causing the connection to fail to grow cwnd when it should, causing throughput to be lower than it should be. The following event sequence is an example that triggers the bug: (a) The connection is cwnd_limited, but packets_out is not at its peak due to TSO deferral deciding not to send another skb yet. In such cases the connection can advance max_packets_seq and set tp->is_cwnd_limited to true and max_packets_out to a small number. (b) Then later in the round trip the connection is pacing-limited (not cwnd-limited), and packets_out is larger. In such cases the connection would raise max_packets_out to a bigger number but (unexpectedly) flip tp->is_cwnd_limited from true to false. This commit fixes that bug. One straightforward fix would be to separately track (a) the next window after max_packets_out reaches a maximum, and (b) the next window after tp->is_cwnd_limited is set to true. But this would require consuming an extra u32 sequence number. Instead, to save space we track only the most important information. Specifically, we track the strongest available signal of the degree to which the cwnd is fully utilized: (1) If the connection is cwnd-limited then we remember that fact for the current window. (2) If the connection not cwnd-limited then we track the maximum number of outstanding packets in the current window. In particular, note that the new logic cannot trigger the buggy (a)/(b) sequence above because with the new logic a condition where tp->packets_out > tp->max_packets_out can only trigger an update of tp->is_cwnd_limited if tp->is_cwnd_limited is false. This first showed up in a testing of a BBRv2 dev branch, but this buggy behavior highlighted a general issue with the tcp_cwnd_validate() logic that can cause cwnd to fail to increase at the proper rate for any TCP congestion control, including Reno or CUBIC. Fixes: ca8a22634381 ("tcp: make cwnd-limited checks measurement-based, and gentler") Signed-off-by: Neal Cardwell Signed-off-by: Kevin(Yudong) Yang Signed-off-by: Yuchung Cheng Signed-off-by: Eric Dumazet Signed-off-by: David S. Miller Signed-off-by: Sasha Levin

tcp: Fix a data-race around sysctl_tcp_notsent_lowat.

2022-07-29T15:06:50+00:00

[ Upstream commit 55be873695ed8912eb77ff46d1d1cadf028bd0f3 ] While reading sysctl_tcp_notsent_lowat, it can be changed concurrently. Thus, we need to add READ_ONCE() to its reader. Fixes: c9bee3b7fdec ("tcp: TCP_NOTSENT_LOWAT socket option") Signed-off-by: Kuniyuki Iwashima Signed-off-by: David S. Miller Signed-off-by: Sasha Levin

tcp: fix potential xmit stalls caused by TCP_NOTSENT_LOWAT

2022-05-12T10:17:07+00:00

[ Upstream commit 4bfe744ff1644fbc0a991a2677dc874475dd6776 ] 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 0cbe6a8f089e ("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: c9bee3b7fdec ("tcp: TCP_NOTSENT_LOWAT socket option") Signed-off-by: Eric Dumazet Reported-by: Doug Porter Cc: Soheil Hassas Yeganeh Cc: Neal Cardwell Acked-by: Soheil Hassas Yeganeh Signed-off-by: David S. Miller Signed-off-by: Sasha Levin

tcp: fix TLP timer not set when CA_STATE changes from DISORDER to OPEN

2021-02-03T22:22:23+00:00

commit 62d9f1a6945ba69c125e548e72a36d203b30596e upstream. Upon receiving a cumulative ACK that changes the congestion state from Disorder to Open, the TLP timer is not set. If the sender is app-limited, it can only wait for the RTO timer to expire and retransmit. The reason for this is that the TLP timer is set before the congestion state changes in tcp_ack(), so we delay the time point of calling tcp_set_xmit_timer() until after tcp_fastretrans_alert() returns and remove the FLAG_SET_XMIT_TIMER from ack_flag when the RACK reorder timer is set. This commit has two additional benefits: 1) Make sure to reset RTO according to RFC6298 when receiving ACK, to avoid spurious RTO caused by RTO timer early expires. 2) Reduce the xmit timer reschedule once per ACK when the RACK reorder timer is set. Fixes: df92c8394e6e ("tcp: fix xmit timer to only be reset if data ACKed/SACKed") Link: https://lore.kernel.org/netdev/1611311242-6675-1-git-send-email-yangpc@wangsu.com Signed-off-by: Pengcheng Yang Acked-by: Neal Cardwell Acked-by: Yuchung Cheng Cc: Eric Dumazet Link: https://lore.kernel.org/r/1611464834-23030-1-git-send-email-yangpc@wangsu.com Signed-off-by: Jakub Kicinski Signed-off-by: Greg Kroah-Hartman

tcp: cache line align MAX_TCP_HEADER

2020-05-02T15:24:17+00:00

[ Upstream commit 9bacd256f1354883d3c1402655153367982bba49 ] TCP stack is dumb in how it cooks its output packets. Depending on MAX_HEADER value, we might chose a bad ending point for the headers. If we align the end of TCP headers to cache line boundary, we make sure to always use the smallest number of cache lines, which always help. Signed-off-by: Eric Dumazet Cc: Soheil Hassas Yeganeh Acked-by: Soheil Hassas Yeganeh Signed-off-by: David S. Miller Signed-off-by: Greg Kroah-Hartman

tcp: annotate lockless access to tcp_memory_pressure

2020-01-27T13:46:50+00:00

[ Upstream commit 1f142c17d19a5618d5a633195a46f2c8be9bf232 ] tcp_memory_pressure is read without holding any lock, and its value could be changed on other cpus. Use READ_ONCE() to annotate these lockless reads. The write side is already using atomic ops. Fixes: b8da51ebb1aa ("tcp: introduce tcp_under_memory_pressure()") Signed-off-by: Eric Dumazet Signed-off-by: Jakub Kicinski Signed-off-by: Sasha Levin

tcp: Protect accesses to .ts_recent_stamp with {READ,WRITE}_ONCE()

2019-12-21T09:47:37+00:00

[ Upstream commit 721c8dafad26ccfa90ff659ee19755e3377b829d ] Syncookies borrow the ->rx_opt.ts_recent_stamp field to store the timestamp of the last synflood. Protect them with READ_ONCE() and WRITE_ONCE() since reads and writes aren't serialised. Use of .rx_opt.ts_recent_stamp for storing the synflood timestamp was introduced by a0f82f64e269 ("syncookies: remove last_synq_overflow from struct tcp_sock"). But unprotected accesses were already there when timestamp was stored in .last_synq_overflow. Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2") Signed-off-by: Guillaume Nault Signed-off-by: Eric Dumazet Signed-off-by: David S. Miller Signed-off-by: Greg Kroah-Hartman

tcp: tighten acceptance of ACKs not matching a child socket

2019-12-21T09:47:37+00:00

[ Upstream commit cb44a08f8647fd2e8db5cc9ac27cd8355fa392d8 ] When no synflood occurs, the synflood timestamp isn't updated. Therefore it can be so old that time_after32() can consider it to be in the future. That's a problem for tcp_synq_no_recent_overflow() as it may report that a recent overflow occurred while, in fact, it's just that jiffies has grown past 'last_overflow' + TCP_SYNCOOKIE_VALID + 2^31. Spurious detection of recent overflows lead to extra syncookie verification in cookie_v[46]_check(). At that point, the verification should fail and the packet dropped. But we should have dropped the packet earlier as we didn't even send a syncookie. Let's refine tcp_synq_no_recent_overflow() to report a recent overflow only if jiffies is within the [last_overflow, last_overflow + TCP_SYNCOOKIE_VALID] interval. This way, no spurious recent overflow is reported when jiffies wraps and 'last_overflow' becomes in the future from the point of view of time_after32(). However, if jiffies wraps and enters the [last_overflow, last_overflow + TCP_SYNCOOKIE_VALID] interval (with 'last_overflow' being a stale synflood timestamp), then tcp_synq_no_recent_overflow() still erroneously reports an overflow. In such cases, we have to rely on syncookie verification to drop the packet. We unfortunately have no way to differentiate between a fresh and a stale syncookie timestamp. In practice, using last_overflow as lower bound is problematic. If the synflood timestamp is concurrently updated between the time we read jiffies and the moment we store the timestamp in 'last_overflow', then 'now' becomes smaller than 'last_overflow' and tcp_synq_no_recent_overflow() returns true, potentially dropping a valid syncookie. Reading jiffies after loading the timestamp could fix the problem, but that'd require a memory barrier. Let's just accommodate for potential timestamp growth instead and extend the interval using 'last_overflow - HZ' as lower bound. Signed-off-by: Guillaume Nault Signed-off-by: Eric Dumazet Signed-off-by: David S. Miller Signed-off-by: Greg Kroah-Hartman

tcp: fix rejected syncookies due to stale timestamps

2019-12-21T09:47:36+00:00

[ Upstream commit 04d26e7b159a396372646a480f4caa166d1b6720 ] If no synflood happens for a long enough period of time, then the synflood timestamp isn't refreshed and jiffies can advance so much that time_after32() can't accurately compare them any more. Therefore, we can end up in a situation where time_after32(now, last_overflow + HZ) returns false, just because these two values are too far apart. In that case, the synflood timestamp isn't updated as it should be, which can trick tcp_synq_no_recent_overflow() into rejecting valid syncookies. For example, let's consider the following scenario on a system with HZ=1000: * The synflood timestamp is 0, either because that's the timestamp of the last synflood or, more commonly, because we're working with a freshly created socket. * We receive a new SYN, which triggers synflood protection. Let's say that this happens when jiffies == 2147484649 (that is, 'synflood timestamp' + HZ + 2^31 + 1). * Then tcp_synq_overflow() doesn't update the synflood timestamp, because time_after32(2147484649, 1000) returns false. With: - 2147484649: the value of jiffies, aka. 'now'. - 1000: the value of 'last_overflow' + HZ. * A bit later, we receive the ACK completing the 3WHS. But cookie_v[46]_check() rejects it because tcp_synq_no_recent_overflow() says that we're not under synflood. That's because time_after32(2147484649, 120000) returns false. With: - 2147484649: the value of jiffies, aka. 'now'. - 120000: the value of 'last_overflow' + TCP_SYNCOOKIE_VALID. Of course, in reality jiffies would have increased a bit, but this condition will last for the next 119 seconds, which is far enough to accommodate for jiffie's growth. Fix this by updating the overflow timestamp whenever jiffies isn't within the [last_overflow, last_overflow + HZ] range. That shouldn't have any performance impact since the update still happens at most once per second. Now we're guaranteed to have fresh timestamps while under synflood, so tcp_synq_no_recent_overflow() can safely use it with time_after32() in such situations. Stale timestamps can still make tcp_synq_no_recent_overflow() return the wrong verdict when not under synflood. This will be handled in the next patch. For 64 bits architectures, the problem was introduced with the conversion of ->tw_ts_recent_stamp to 32 bits integer by commit cca9bab1b72c ("tcp: use monotonic timestamps for PAWS"). The problem has always been there on 32 bits architectures. Fixes: cca9bab1b72c ("tcp: use monotonic timestamps for PAWS") Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2") Signed-off-by: Guillaume Nault Signed-off-by: Eric Dumazet Signed-off-by: David S. Miller Signed-off-by: Greg Kroah-Hartman