tcp: read multiple skbs in tcp_read_skb()
authorCong Wang <cong.wang@bytedance.com>
Mon, 12 Sep 2022 17:35:53 +0000 (10:35 -0700)
committerPaolo Abeni <pabeni@redhat.com>
Tue, 20 Sep 2022 12:47:21 +0000 (14:47 +0200)
commitdb4192a754ebd52300a28abe1a50dd18eae0eb12
treeb5c8303e4305c06b00f2bc92540299c5cbac07f1
parent90fdd1c1e9c49bcb46cde589dbdee94a6977086a
tcp: read multiple skbs in tcp_read_skb()

Before we switched to ->read_skb(), ->read_sock() was passed with
desc.count=1, which technically indicates we only read one skb per
->sk_data_ready() call. However, for TCP, this is not true.

TCP at least has sk_rcvlowat which intentionally holds skb's in
receive queue until this watermark is reached. This means when
->sk_data_ready() is invoked there could be multiple skb's in the
queue, therefore we have to read multiple skbs in tcp_read_skb()
instead of one.

Fixes: 965b57b469a5 ("net: Introduce a new proto_ops ->read_skb()")
Reported-by: Peilin Ye <peilin.ye@bytedance.com>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jakub Sitnicki <jakub@cloudflare.com>
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Cong Wang <cong.wang@bytedance.com>
Link: https://lore.kernel.org/r/20220912173553.235838-1-xiyou.wangcong@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
net/ipv4/tcp.c