netfilter.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Eric Dumazet <eric.dumazet@gmail.com>
To: Evgeniy Polyakov <zbr@ioremap.net>, David Miller <davem@davemloft.net>
Cc: kdakhane@gmail.com, netdev@vger.kernel.org, netfilter@vger.kernel.org
Subject: Re: [PATCH net-next-2.6] tcp: connect() race with timewait reuse
Date: Thu, 03 Dec 2009 09:31:19 +0100	[thread overview]
Message-ID: <4B177757.3010407@gmail.com> (raw)
In-Reply-To: <4B175E65.3070905@gmail.com>

Eric Dumazet a écrit :
>
> 
> You are probably right, we could defer the inet_twsk_put(tw) out of locked
> section, you or I will submit another patch to correct this.
> 

Here is an updated patch, tested on my dev machine.

I found another problem about tw refcnt I am going to address ASAP.

Thanks !

[PATCH net-next-2.6] 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>
---
 include/net/inet_timewait_sock.h |    2 +
 net/ipv4/inet_hashtables.c       |   10 +++++--
 net/ipv4/inet_timewait_sock.c    |   38 +++++++++++++++++++++--------
 net/ipv6/inet6_hashtables.c      |   15 +++++++----
 4 files changed, 47 insertions(+), 18 deletions(-)

diff --git a/include/net/inet_timewait_sock.h b/include/net/inet_timewait_sock.h
index 773b10f..cb7d93b 100644
--- a/include/net/inet_timewait_sock.h
+++ b/include/net/inet_timewait_sock.h
@@ -199,6 +199,8 @@ static inline __be32 inet_rcv_saddr(const struct sock *sk)
 
 extern void inet_twsk_put(struct inet_timewait_sock *tw);
 
+extern int inet_twsk_unhash(struct inet_timewait_sock *tw);
+
 extern struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
 						  const int state);
 
diff --git a/net/ipv4/inet_hashtables.c b/net/ipv4/inet_hashtables.c
index 94ef51a..30e73c5 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 1f5d508..11380e6 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 00c6a3e..7207801 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);
 	}

  reply	other threads:[~2009-12-03  8:31 UTC|newest]

Thread overview: 31+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2009-12-01  2:02 soft lockup in inet_csk_get_port kapil dakhane
2009-12-01  6:10 ` Eric Dumazet
2009-12-01 15:00 ` [PATCH] tcp: Fix a connect() race with timewait sockets Eric Dumazet
2009-12-02  8:59   ` David Miller
2009-12-02  9:23     ` Eric Dumazet
2009-12-02 10:33       ` Eric Dumazet
2009-12-02 11:32         ` Evgeniy Polyakov
2009-12-02 19:18           ` kapil dakhane
2009-12-03  2:43             ` kapil dakhane
2009-12-03 10:49               ` [PATCH] tcp: fix a timewait refcnt race Eric Dumazet
2009-12-04  0:19                 ` David Miller
2009-12-04  3:20                 ` kapil dakhane
2009-12-04  6:29                   ` Eric Dumazet
2009-12-04  6:39                     ` David Miller
2009-12-02 15:08         ` [PATCH net-next-2.6] tcp: connect() race with timewait reuse Eric Dumazet
2009-12-02 22:15           ` Evgeniy Polyakov
2009-12-03  6:44             ` Eric Dumazet
2009-12-03  8:31               ` Eric Dumazet [this message]
2009-12-03 23:22                 ` Evgeniy Polyakov
2009-12-04  0:18                 ` David Miller
2009-12-02 16:05     ` [PATCH] tcp: Fix a connect() race with timewait sockets Ashwani Wason
2009-12-03  6:38       ` David Miller
2009-12-04 13:45   ` [PATCH 0/2] tcp: Fix connect() races " Eric Dumazet
2009-12-04 13:46   ` [PATCH 1/2] tcp: Fix a connect() race " Eric Dumazet
2009-12-05 21:21     ` Evgeniy Polyakov
2009-12-07  9:59       ` [PATCH] tcp: documents timewait refcnt tricks Eric Dumazet
2009-12-07 16:06         ` Randy Dunlap
2009-12-09  4:20           ` David Miller
2009-12-09  4:18     ` [PATCH 1/2] tcp: Fix a connect() race with timewait sockets David Miller
2009-12-04 13:47   ` [PATCH 2/2] " Eric Dumazet
2009-12-09  4:19     ` David Miller

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=4B177757.3010407@gmail.com \
    --to=eric.dumazet@gmail.com \
    --cc=davem@davemloft.net \
    --cc=kdakhane@gmail.com \
    --cc=netdev@vger.kernel.org \
    --cc=netfilter@vger.kernel.org \
    --cc=zbr@ioremap.net \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).