aboutsummaryrefslogtreecommitdiffstats
path: root/net
diff options
context:
space:
mode:
authorEric Dumazet <eric.dumazet@gmail.com>2009-12-02 19:49:01 -0500
committerDavid S. Miller <davem@davemloft.net>2009-12-03 19:17:44 -0500
commit47e1c323069bcef0acb8a2b48921688573f5ca63 (patch)
treebebfe86ae36055adae2100200e87727d21a7b762 /net
parent13475a30b66cdb9250a34052c19ac98847373030 (diff)
tcp: fix a timewait refcnt race
After TCP RCU conversion, tw->tw_refcnt should not be set to 1 in inet_twsk_alloc(). It allows a RCU reader to get this timewait socket, while we not yet stabilized it. Only choice we have is to set tw_refcnt to 0 in inet_twsk_alloc(), then atomic_add() it later, once everything is done. Location of this atomic_add() is tricky, because we dont want another writer to find this timewait in ehash, while tw_refcnt is still zero ! Thanks to Kapil Dakhane tests and reports. Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'net')
-rw-r--r--net/ipv4/inet_timewait_sock.c19
1 files changed, 16 insertions, 3 deletions
diff --git a/net/ipv4/inet_timewait_sock.c b/net/ipv4/inet_timewait_sock.c
index 11a107a5af4f..0fdf45e4c90c 100644
--- a/net/ipv4/inet_timewait_sock.c
+++ b/net/ipv4/inet_timewait_sock.c
@@ -109,7 +109,6 @@ void __inet_twsk_hashdance(struct inet_timewait_sock *tw, struct sock *sk,
109 tw->tw_tb = icsk->icsk_bind_hash; 109 tw->tw_tb = icsk->icsk_bind_hash;
110 WARN_ON(!icsk->icsk_bind_hash); 110 WARN_ON(!icsk->icsk_bind_hash);
111 inet_twsk_add_bind_node(tw, &tw->tw_tb->owners); 111 inet_twsk_add_bind_node(tw, &tw->tw_tb->owners);
112 atomic_inc(&tw->tw_refcnt);
113 spin_unlock(&bhead->lock); 112 spin_unlock(&bhead->lock);
114 113
115 spin_lock(lock); 114 spin_lock(lock);
@@ -119,13 +118,22 @@ void __inet_twsk_hashdance(struct inet_timewait_sock *tw, struct sock *sk,
119 * Should be done before removing sk from established chain 118 * Should be done before removing sk from established chain
120 * because readers are lockless and search established first. 119 * because readers are lockless and search established first.
121 */ 120 */
122 atomic_inc(&tw->tw_refcnt);
123 inet_twsk_add_node_rcu(tw, &ehead->twchain); 121 inet_twsk_add_node_rcu(tw, &ehead->twchain);
124 122
125 /* Step 3: Remove SK from established hash. */ 123 /* Step 3: Remove SK from established hash. */
126 if (__sk_nulls_del_node_init_rcu(sk)) 124 if (__sk_nulls_del_node_init_rcu(sk))
127 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 125 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
128 126
127 /*
128 * Notes :
129 * - We initially set tw_refcnt to 0 in inet_twsk_alloc()
130 * - We add one reference for the bhash link
131 * - We add one reference for the ehash link
132 * - We want this refcnt update done before allowing other
133 * threads to find this tw in ehash chain.
134 */
135 atomic_add(1 + 1 + 1, &tw->tw_refcnt);
136
129 spin_unlock(lock); 137 spin_unlock(lock);
130} 138}
131 139
@@ -157,7 +165,12 @@ struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk, const int stat
157 tw->tw_transparent = inet->transparent; 165 tw->tw_transparent = inet->transparent;
158 tw->tw_prot = sk->sk_prot_creator; 166 tw->tw_prot = sk->sk_prot_creator;
159 twsk_net_set(tw, hold_net(sock_net(sk))); 167 twsk_net_set(tw, hold_net(sock_net(sk)));
160 atomic_set(&tw->tw_refcnt, 1); 168 /*
169 * Because we use RCU lookups, we should not set tw_refcnt
170 * to a non null value before everything is setup for this
171 * timewait socket.
172 */
173 atomic_set(&tw->tw_refcnt, 0);
161 inet_twsk_dead_node_init(tw); 174 inet_twsk_dead_node_init(tw);
162 __module_get(tw->tw_prot->owner); 175 __module_get(tw->tw_prot->owner);
163 } 176 }