Re: [PATCH net 4/4] tcp: various missing rcu_read_lock around __sk_dst_get
From: Eric Dumazet <hidden>
Date: 2016-04-01 03:13:06
Subsystem:
networking [general], networking [sockets], the rest · Maintainers:
"David S. Miller", Eric Dumazet, Jakub Kicinski, Paolo Abeni, Kuniyuki Iwashima, Willem de Bruijn, Linus Torvalds
On Fri, 2016-04-01 at 04:01 +0200, Hannes Frederic Sowa wrote:
I thought so first, as well. But given the double check for the spin_lock and the "mutex" we end up with the same result for the lockdep_sock_is_held check. Do you see other consequences?
Well, we release the spinlock in __release_sock() So another thread could come and acquire the socket, then call mutex_acquire() while the first thread did not call yet mutex_release() So maybe lockdep will complain (but I do not know lockdep enough to tell) So maybe the following would be better : (Absolutely untested, really I need to take a break)
diff --git a/include/net/sock.h b/include/net/sock.h
index 255d3e03727b..7d5dfa7e1918 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h@@ -1327,7 +1327,13 @@ static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) static inline void sock_release_ownership(struct sock *sk) { - sk->sk_lock.owned = 0; + if (sk->sk_lock.owned) { + /* + * The sk_lock has mutex_unlock() semantics: + */ + mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); + sk->sk_lock.owned = 0; + } } /*
diff --git a/net/core/sock.c b/net/core/sock.c
index b67b9aedb230..c7ab98e72346 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c@@ -2429,10 +2429,6 @@ EXPORT_SYMBOL(lock_sock_nested); void release_sock(struct sock *sk) { - /* - * The sk_lock has mutex_unlock() semantics: - */ - mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); spin_lock_bh(&sk->sk_lock.slock); if (sk->sk_backlog.tail)