diff options
| author | Eric Dumazet <edumazet@google.com> | 2020-05-12 06:54:30 -0700 |
|---|---|---|
| committer | David S. Miller <davem@davemloft.net> | 2020-05-12 12:49:47 -0700 |
| commit | 24adbc1676af4e134e709ddc7f34cf2adc2131e4 (patch) | |
| tree | ab86c8315f4652959c283cb29739dfeb437da15a /net/ipv4 | |
| parent | 92db978f0d686468e527d49268e7c7e8d97d334b (diff) | |
| download | linux-24adbc1676af4e134e709ddc7f34cf2adc2131e4.tar.gz linux-24adbc1676af4e134e709ddc7f34cf2adc2131e4.tar.bz2 linux-24adbc1676af4e134e709ddc7f34cf2adc2131e4.zip | |
tcp: fix SO_RCVLOWAT hangs with fat skbs
We autotune rcvbuf whenever SO_RCVLOWAT is set to account for 100%
overhead in tcp_set_rcvlowat()
This works well when skb->len/skb->truesize ratio is bigger than 0.5
But if we receive packets with small MSS, we can end up in a situation
where not enough bytes are available in the receive queue to satisfy
RCVLOWAT setting.
As our sk_rcvbuf limit is hit, we send zero windows in ACK packets,
preventing remote peer from sending more data.
Even autotuning does not help, because it only triggers at the time
user process drains the queue. If no EPOLLIN is generated, this
can not happen.
Note poll() has a similar issue, after commit
c7004482e8dc ("tcp: Respect SO_RCVLOWAT in tcp_poll().")
Fixes: 03f45c883c6f ("tcp: avoid extra wakeups for SO_RCVLOWAT users")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'net/ipv4')
| -rw-r--r-- | net/ipv4/tcp.c | 14 | ||||
| -rw-r--r-- | net/ipv4/tcp_input.c | 3 |
2 files changed, 13 insertions, 4 deletions
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c index e72bd651d21a..a385fcaaa03b 100644 --- a/net/ipv4/tcp.c +++ b/net/ipv4/tcp.c @@ -476,9 +476,17 @@ static void tcp_tx_timestamp(struct sock *sk, u16 tsflags) static inline bool tcp_stream_is_readable(const struct tcp_sock *tp, int target, struct sock *sk) { - return (READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq) >= target) || - (sk->sk_prot->stream_memory_read ? - sk->sk_prot->stream_memory_read(sk) : false); + int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); + + if (avail > 0) { + if (avail >= target) + return true; + if (tcp_rmem_pressure(sk)) + return true; + } + if (sk->sk_prot->stream_memory_read) + return sk->sk_prot->stream_memory_read(sk); + return false; } /* diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index b996dc1069c5..29c6fc8c7716 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -4757,7 +4757,8 @@ void tcp_data_ready(struct sock *sk) const struct tcp_sock *tp = tcp_sk(sk); int avail = tp->rcv_nxt - tp->copied_seq; - if (avail < sk->sk_rcvlowat && !sock_flag(sk, SOCK_DONE)) + if (avail < sk->sk_rcvlowat && !tcp_rmem_pressure(sk) && + !sock_flag(sk, SOCK_DONE)) return; sk->sk_data_ready(sk); |