Skip to contentSkip to conversation

net/smc: add fallback check to connect()

PATCH Applied Karsten Graul 2 participants 2 messages Dec 12, 2019 – Dec 15, 2019

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 net/smc: add fallback check to connect()

From: Ursula Braun <[email protected]>

FASTOPEN setsockopt() or sendmsg() may switch the SMC socket to fallback mode. Once fallback mode is active, the native TCP socket functions are called. Nevertheless there is a small race window, when FASTOPEN setsockopt/sendmsg runs in parallel to a connect(), and switch the socket into fallback mode before connect() takes the sock lock. Make sure the SMC-specific connect setup is omitted in this case.

This way a syzbot-reported refcount problem is fixed, triggered by different threads running non-blocking connect() and FASTOPEN_KEY setsockopt.

Reported-by [email protected]
Fixes 6d6dd528d5af ("net/smc: fix refcount non-blocking connect() -part 2")
---
 net/smc/af_smc.c | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)
net/smc/af_smc.c +8 -6
index b997072c72e5..cee5bf4a9bb9 100644
@@ -857,6 +857,8 @@ static int smc_connect(struct socket *sock, struct sockaddr *addr,
 		goto out;
 
 	sock_hold(&smc->sk); /* sock put in passive closing */
+	if (smc->use_fallback)
+		goto out;
 	if (flags & O_NONBLOCK) {
 		if (schedule_work(&smc->connect_work))
 			smc->connect_nonblock = 1;
@@ -1721,8 +1723,6 @@ static int smc_setsockopt(struct socket *sock, int level, int optname,
 		sk->sk_err = smc->clcsock->sk->sk_err;
 		sk->sk_error_report(sk);
 	}
-	if (rc)
-		return rc;
 
 	if (optlen < sizeof(int))
 		return -EINVAL;
@@ -1730,6 +1730,8 @@ static int smc_setsockopt(struct socket *sock, int level, int optname,
 		return -EFAULT;
 
 	lock_sock(sk);
+	if (rc || smc->use_fallback)
+		goto out;
 	switch (optname) {
 	case TCP_ULP:
 	case TCP_FASTOPEN:
@@ -1741,15 +1743,14 @@ static int smc_setsockopt(struct socket *sock, int level, int optname,
 			smc_switch_to_fallback(smc);
 			smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP;
 		} else {
-			if (!smc->use_fallback)
-				rc = -EINVAL;
+			rc = -EINVAL;
 		}
 		break;
 	case TCP_NODELAY:
 		if (sk->sk_state != SMC_INIT &&
 		    sk->sk_state != SMC_LISTEN &&
 		    sk->sk_state != SMC_CLOSED) {
-			if (val && !smc->use_fallback)
+			if (val)
 				mod_delayed_work(system_wq, &smc->conn.tx_work,
 						 0);
 		}
@@ -1758,7 +1759,7 @@ static int smc_setsockopt(struct socket *sock, int level, int optname,
 		if (sk->sk_state != SMC_INIT &&
 		    sk->sk_state != SMC_LISTEN &&
 		    sk->sk_state != SMC_CLOSED) {
-			if (!val && !smc->use_fallback)
+			if (!val)
 				mod_delayed_work(system_wq, &smc->conn.tx_work,
 						 0);
 		}
@@ -1769,6 +1770,7 @@ static int smc_setsockopt(struct socket *sock, int level, int optname,
 	default:
 		break;
 	}
+out:
 	release_sock(sk);
 
 	return rc;
Signature · 1 line
2.21.0
#1 lore raw
Quoted 9 lines · Karsten Graul

From: Ursula Braun <[email protected]>

FASTOPEN setsockopt() or sendmsg() may switch the SMC socket to fallback mode. Once fallback mode is active, the native TCP socket functions are called. Nevertheless there is a small race window, when FASTOPEN setsockopt/sendmsg runs in parallel to a connect(), and switch the socket into fallback mode before connect() takes the sock lock. Make sure the SMC-specific connect setup is omitted in this case.

Quoted 8 lines

This way a syzbot-reported refcount problem is fixed, triggered by different threads running non-blocking connect() and FASTOPEN_KEY setsockopt.

Reported-by [email protected]
Fixes 6d6dd528d5af ("net/smc: fix refcount non-blocking connect() -part 2")

Applied, and queued for stable, thank you!