On Thu, 2016-09-15 at 15:52 +0000, David Laight wrote: > From: Eric Dumazet > > Sent: 15 September 2016 16:13 > > If a TCP socket gets a large write queue, an overflow can happen > > in a test in __tcp_retransmit_skb() preventing all retransmits. > ... > > if (atomic_read(&sk->sk_wmem_alloc) > > > - min(sk->sk_wmem_queued + (sk->sk_wmem_queued >> 2), sk->sk_sndbuf)) > > + min_t(u32, sk->sk_wmem_queued + (sk->sk_wmem_queued >> 2), > > + sk->sk_sndbuf)) > > return -EAGAIN; > > Might it also be better to split that test to (say): > > u32 wmem_alloc = atomic_read(&sk->sk_wmem_alloc); > if (unlikely((wmem_alloc > sk->sk_sndbuf)) > return -EAGAIN; > if (unlikely(wmem_alloc > sk->sk_wmem_queued + (sk->sk_wmem_queued >> > 2))) > return -EAGAIN;
Well, I find the existing code more readable, but this is just an opinion. Thanks.
