Permalink
Browse files

tcp: tcp_probe: use spin_lock_bh()

tcp_rcv_established() can now run in process context.

We need to disable BH while acquiring tcp probe spinlock,
or risk a deadlock.

Fixes: 5413d1b ("net: do not block BH while processing socket backlog")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Ricardo Nabinger Sanchez <rnsanchez@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
  • Loading branch information...
Eric Dumazet authored and davem330 committed Feb 15, 2017
1 parent a725eb1 commit e70ac171658679ecf6bea4bbd9e9325cd6079d2b
Showing with 2 additions and 2 deletions.
  1. +2 −2 net/ipv4/tcp_probe.c
@@ -117,7 +117,7 @@ static void jtcp_rcv_established(struct sock *sk, struct sk_buff *skb,
(fwmark > 0 && skb->mark == fwmark)) &&
(full || tp->snd_cwnd != tcp_probe.lastcwnd)) {

spin_lock(&tcp_probe.lock);
spin_lock_bh(&tcp_probe.lock);
/* If log fills, just silently drop */
if (tcp_probe_avail() > 1) {
struct tcp_log *p = tcp_probe.log + tcp_probe.head;
@@ -157,7 +157,7 @@ static void jtcp_rcv_established(struct sock *sk, struct sk_buff *skb,
tcp_probe.head = (tcp_probe.head + 1) & (bufsize - 1);
}
tcp_probe.lastcwnd = tp->snd_cwnd;
spin_unlock(&tcp_probe.lock);
spin_unlock_bh(&tcp_probe.lock);

wake_up(&tcp_probe.wait);
}

0 comments on commit e70ac17

Please sign in to comment.