tcp: connect() race with timewait reuse

Its currently possible that several threads issuing a connect() find
the same timewait socket and try to reuse it, leading to list
corruptions.

Condition for bug is that these threads bound their socket on same
address/port of to-be-find timewait socket, and connected to same
target. (SO_REUSEADDR needed)

To fix this problem, we could unhash timewait socket while holding
ehash lock, to make sure lookups/changes will be serialized. Only
first thread finds the timewait socket, other ones find the
established socket and return an EADDRNOTAVAIL error.

This second version takes into account Evgeniy's review and makes sure
inet_twsk_put() is called outside of locked sections.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

3 files changed, 45 insertions, 18 deletions

diff --git a/net/ipv4/inet_hashtables.c b/net/ipv4/inet_hashtables.c
index 94ef51aa5bc9..21e5e32d8c60 100644
--- a/net/ipv4/inet_hashtables.c
+++ b/net/ipv4/inet_hashtables.c
@@ -286,6 +286,7 @@ static int __inet_check_established(struct inet_timewait_death_row *death_row,
         struct sock *sk2;
         const struct hlist_nulls_node *node;
         struct inet_timewait_sock *tw;
+        int twrefcnt = 0;
 
         spin_lock(lock);
 
@@ -318,20 +319,23 @@ unique:
         sk->sk_hash = hash;
         WARN_ON(!sk_unhashed(sk));
         __sk_nulls_add_node_rcu(sk, &head->chain);
+        if (tw) {
+                twrefcnt = inet_twsk_unhash(tw);
+                NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
+        }
         spin_unlock(lock);
+        if (twrefcnt)
+                inet_twsk_put(tw);
         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 
         if (twp) {
                 *twp = tw;
-                NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
         } else if (tw) {
                 /* Silly. Should hash-dance instead... */
                 inet_twsk_deschedule(tw, death_row);
-                NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
 
                 inet_twsk_put(tw);
         }
-
         return 0;
 
 not_unique:
diff --git a/net/ipv4/inet_timewait_sock.c b/net/ipv4/inet_timewait_sock.c
index 31f931ef3daf..11a107a5af4f 100644
--- a/net/ipv4/inet_timewait_sock.c
+++ b/net/ipv4/inet_timewait_sock.c
@@ -14,22 +14,33 @@
 #include <net/inet_timewait_sock.h>
 #include <net/ip.h>
 
+
+/*
+ * unhash a timewait socket from established hash
+ * lock must be hold by caller
+ */
+int inet_twsk_unhash(struct inet_timewait_sock *tw)
+{
+        if (hlist_nulls_unhashed(&tw->tw_node))
+                return 0;
+
+        hlist_nulls_del_rcu(&tw->tw_node);
+        sk_nulls_node_init(&tw->tw_node);
+        return 1;
+}
+
 /* Must be called with locally disabled BHs. */
 static void __inet_twsk_kill(struct inet_timewait_sock *tw,
                              struct inet_hashinfo *hashinfo)
 {
         struct inet_bind_hashbucket *bhead;
         struct inet_bind_bucket *tb;
+        int refcnt;
         /* Unlink from established hashes. */
         spinlock_t *lock = inet_ehash_lockp(hashinfo, tw->tw_hash);
 
         spin_lock(lock);
-        if (hlist_nulls_unhashed(&tw->tw_node)) {
-                spin_unlock(lock);
-                return;
-        }
-        hlist_nulls_del_rcu(&tw->tw_node);
-        sk_nulls_node_init(&tw->tw_node);
+        refcnt = inet_twsk_unhash(tw);
         spin_unlock(lock);
 
         /* Disassociate with bind bucket. */
@@ -37,9 +48,12 @@ static void __inet_twsk_kill(struct inet_timewait_sock *tw,
                         hashinfo->bhash_size)];
         spin_lock(&bhead->lock);
         tb = tw->tw_tb;
-        __hlist_del(&tw->tw_bind_node);
-        tw->tw_tb = NULL;
-        inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
+        if (tb) {
+                __hlist_del(&tw->tw_bind_node);
+                tw->tw_tb = NULL;
+                inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
+                refcnt++;
+        }
         spin_unlock(&bhead->lock);
 #ifdef SOCK_REFCNT_DEBUG
         if (atomic_read(&tw->tw_refcnt) != 1) {
@@ -47,7 +61,10 @@ static void __inet_twsk_kill(struct inet_timewait_sock *tw,
                        tw->tw_prot->name, tw, atomic_read(&tw->tw_refcnt));
         }
 #endif
-        inet_twsk_put(tw);
+        while (refcnt) {
+                inet_twsk_put(tw);
+                refcnt--;
+        }
 }
 
 static noinline void inet_twsk_free(struct inet_timewait_sock *tw)
@@ -92,6 +109,7 @@ void __inet_twsk_hashdance(struct inet_timewait_sock *tw, struct sock *sk,
         tw->tw_tb = icsk->icsk_bind_hash;
         WARN_ON(!icsk->icsk_bind_hash);
         inet_twsk_add_bind_node(tw, &tw->tw_tb->owners);
+        atomic_inc(&tw->tw_refcnt);
         spin_unlock(&bhead->lock);
 
         spin_lock(lock);
diff --git a/net/ipv6/inet6_hashtables.c b/net/ipv6/inet6_hashtables.c
index 00c6a3e6cddf..c813e294ec0c 100644
--- a/net/ipv6/inet6_hashtables.c
+++ b/net/ipv6/inet6_hashtables.c
@@ -223,6 +223,7 @@ static int __inet6_check_established(struct inet_timewait_death_row *death_row,
         struct sock *sk2;
         const struct hlist_nulls_node *node;
         struct inet_timewait_sock *tw;
+        int twrefcnt = 0;
 
         spin_lock(lock);
 
@@ -250,19 +251,23 @@ unique:
          * in hash table socket with a funny identity. */
         inet->inet_num = lport;
         inet->inet_sport = htons(lport);
+        sk->sk_hash = hash;
         WARN_ON(!sk_unhashed(sk));
         __sk_nulls_add_node_rcu(sk, &head->chain);
-        sk->sk_hash = hash;
+        if (tw) {
+                twrefcnt = inet_twsk_unhash(tw);
+                NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
+        }
         spin_unlock(lock);
+        if (twrefcnt)
+                inet_twsk_put(tw);
         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 
-        if (twp != NULL) {
+        if (twp) {
                 *twp = tw;
-                NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
-        } else if (tw != NULL) {
+        } else if (tw) {
                 /* Silly. Should hash-dance instead... */
                 inet_twsk_deschedule(tw, death_row);
-                NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
 
                 inet_twsk_put(tw);
         }