Skip to contentSkip to conversation

net/tcp_fastopen: fix data races around tfo_active_disable_stamp

PATCH Applied Eric Dumazet 3 participants 3 messages Jul 19, 2021

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 net/tcp_fastopen: fix data races around tfo_active_disable_stamp

From: Eric Dumazet <[email protected]>

tfo_active_disable_stamp is read and written locklessly. We need to annotate these accesses appropriately.

Then, we need to perform the atomic_inc(tfo_active_disable_times) after the timestamp has been updated, and thus add barriers to make sure tcp_fastopen_active_should_disable() wont read a stale timestamp.

Fixes cf1ef3f0719b ("net/tcp_fastopen: Disable active side TFO in certain scenarios")
Cc Wei Wang <[email protected]>
Cc Yuchung Cheng <[email protected]>
Cc Neal Cardwell <[email protected]>
---
 net/ipv4/tcp_fastopen.c | 19 ++++++++++++++++---
 1 file changed, 16 insertions(+), 3 deletions(-)
net/ipv4/tcp_fastopen.c +16 -3
index 47c32604d38fca960d2cd56f3588bfd2e390b789..b32af76e21325373126b51423496e3b8d47d97ff 100644
@@ -507,8 +507,15 @@ void tcp_fastopen_active_disable(struct sock *sk)
 {
 	struct net *net = sock_net(sk);
 
+	/* Paired with READ_ONCE() in tcp_fastopen_active_should_disable() */
+	WRITE_ONCE(net->ipv4.tfo_active_disable_stamp, jiffies);
+
+	/* Paired with smp_rmb() in tcp_fastopen_active_should_disable().
+	 * We want net->ipv4.tfo_active_disable_stamp to be updated first.
+	 */
+	smp_mb__before_atomic();
 	atomic_inc(&net->ipv4.tfo_active_disable_times);
-	net->ipv4.tfo_active_disable_stamp = jiffies;
+
 	NET_INC_STATS(net, LINUX_MIB_TCPFASTOPENBLACKHOLE);
 }
 
@@ -526,10 +533,16 @@ bool tcp_fastopen_active_should_disable(struct sock *sk)
 	if (!tfo_da_times)
 		return false;
 
+	/* Paired with smp_mb__before_atomic() in tcp_fastopen_active_disable() */
+	smp_rmb();
+
 	/* Limit timeout to max: 2^6 * initial timeout */
 	multiplier = 1 << min(tfo_da_times - 1, 6);
-	timeout = multiplier * tfo_bh_timeout * HZ;
-	if (time_before(jiffies, sock_net(sk)->ipv4.tfo_active_disable_stamp + timeout))
+
+	/* Paired with the WRITE_ONCE() in tcp_fastopen_active_disable(). */
+	timeout = READ_ONCE(sock_net(sk)->ipv4.tfo_active_disable_stamp) +
+		  multiplier * tfo_bh_timeout * HZ;
+	if (time_before(jiffies, timeout))
 		return true;
 
 	/* Mark check bit so we can check for successful active TFO
Signature · 1 line
2.32.0.402.g57bb445576-goog
Acked-by on v1 1/1
#1 lore raw
Quoted 11 lines · Eric Dumazet

From: Eric Dumazet <[email protected]>

tfo_active_disable_stamp is read and written locklessly. We need to annotate these accesses appropriately.

Then, we need to perform the atomic_inc(tfo_active_disable_times) after the timestamp has been updated, and thus add barriers to make sure tcp_fastopen_active_should_disable() wont read a stale timestamp.

Quoted 6 lines
Fixes cf1ef3f0719b ("net/tcp_fastopen: Disable active side TFO in certain scenarios")
Cc Wei Wang <[email protected]>
Cc Yuchung Cheng <[email protected]>
Cc Neal Cardwell <[email protected]>

---

Thanks Eric!

Quoted 46 lines
 net/ipv4/tcp_fastopen.c | 19 ++++++++++++++++---
 1 file changed, 16 insertions(+), 3 deletions(-)
net/ipv4/tcp_fastopen.c +16 -4
index 47c32604d38fca960d2cd56f3588bfd2e390b789..b32af76e21325373126b51423496e3b8d47d97ff 100644
@@ -507,8 +507,15 @@ void tcp_fastopen_active_disable(struct sock *sk)
 {
        struct net *net = sock_net(sk);
 
+       /* Paired with READ_ONCE() in tcp_fastopen_active_should_disable() */
+       WRITE_ONCE(net->ipv4.tfo_active_disable_stamp, jiffies);
+
+       /* Paired with smp_rmb() in tcp_fastopen_active_should_disable().
+        * We want net->ipv4.tfo_active_disable_stamp to be updated first.
+        */
+       smp_mb__before_atomic();
        atomic_inc(&net->ipv4.tfo_active_disable_times);
-       net->ipv4.tfo_active_disable_stamp = jiffies;
+
        NET_INC_STATS(net, LINUX_MIB_TCPFASTOPENBLACKHOLE);
 }
 
@@ -526,10 +533,16 @@ bool tcp_fastopen_active_should_disable(struct sock *sk)
        if (!tfo_da_times)
                return false;
 
+       /* Paired with smp_mb__before_atomic() in tcp_fastopen_active_disable() */
+       smp_rmb();
+
        /* Limit timeout to max: 2^6 * initial timeout */
        multiplier = 1 << min(tfo_da_times - 1, 6);
-       timeout = multiplier * tfo_bh_timeout * HZ;
-       if (time_before(jiffies, sock_net(sk)->ipv4.tfo_active_disable_stamp + timeout))
+
+       /* Paired with the WRITE_ONCE() in tcp_fastopen_active_disable(). */
+       timeout = READ_ONCE(sock_net(sk)->ipv4.tfo_active_disable_stamp) +
+                 multiplier * tfo_bh_timeout * HZ;
+       if (time_before(jiffies, timeout))
                return true;
 
        /* Mark check bit so we can check for successful active TFO
--

2.32.0.402.g57bb445576-goog

#2 lore raw
In reply to Eric Dumazet
This patch was applied to netdev/net.git (refs/heads/master):

Hello:

This patch was applied to netdev/net.git (refs/heads/master):

Quoted 11 lines · On Mon, 19 Jul 2021 02:12:18 -0700 you wrote:

From: Eric Dumazet <[email protected]>

tfo_active_disable_stamp is read and written locklessly. We need to annotate these accesses appropriately.

Then, we need to perform the atomic_inc(tfo_active_disable_times) after the timestamp has been updated, and thus add barriers to make sure tcp_fastopen_active_should_disable() wont read a stale timestamp.

[...]

Here is the summary with links:
- [net] net/tcp_fastopen: fix data races around tfo_active_disable_stamp

    https://git.kernel.org/netdev/net/c/6f20c8adb181

You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html