net: af_packet: fix race in PACKET_{R|T}X_RING

In order to remove the race caught by syzbot [1], we need to lock the socket before using po->tp_version as this could change under us otherwise.

This means lock_sock() and release_sock() must be done by packet_set_ring() callers.

[1] :

BUG: KMSAN: uninit-value in packet_set_ring+0x1254/0x3870 net/packet/af_packet.c:4249
CPU: 0 PID: 20195 Comm: syzkaller707632 Not tainted 4.16.0+ #83
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 __dump_stack lib/dump_stack.c:17 [inline]
 dump_stack+0x185/0x1d0 lib/dump_stack.c:53
 kmsan_report+0x142/0x240 mm/kmsan/kmsan.c:1067
 __msan_warning_32+0x6c/0xb0 mm/kmsan/kmsan_instr.c:676
 packet_set_ring+0x1254/0x3870 net/packet/af_packet.c:4249
 packet_setsockopt+0x12c6/0x5a90 net/packet/af_packet.c:3662
 SYSC_setsockopt+0x4b8/0x570 net/socket.c:1849
 SyS_setsockopt+0x76/0xa0 net/socket.c:1828
 do_syscall_64+0x309/0x430 arch/x86/entry/common.c:287
 entry_SYSCALL_64_after_hwframe+0x3d/0xa2
RIP: 0033:0x449099
RSP: 002b:00007f42b5307ce8 EFLAGS: 00000246 ORIG_RAX: 0000000000000036
RAX: ffffffffffffffda RBX: 000000000070003c RCX: 0000000000449099
RDX: 0000000000000005 RSI: 0000000000000107 RDI: 0000000000000003
RBP: 0000000000700038 R08: 000000000000001c R09: 0000000000000000
R10: 00000000200000c0 R11: 0000000000000246 R12: 0000000000000000
R13: 000000000080eecf R14: 00007f42b53089c0 R15: 0000000000000001

Local variable description: ----req_u@packet_setsockopt Variable was created at:

 packet_setsockopt+0x13f/0x5a90 net/packet/af_packet.c:3612
 SYSC_setsockopt+0x4b8/0x570 net/socket.c:1849

---
 net/packet/af_packet.c | 23 ++++++++++++++---------
 1 file changed, 14 insertions(+), 9 deletions(-)

index 616cb9c18f88edd759dfb461051670c225978afa..c31b0687396a6ef45413f06efcc7c3f923e91d01 100644
@@ -3008,6 +3008,7 @@ static int packet_release(struct socket *sock)
 
 	packet_flush_mclist(sk);
 
+	lock_sock(sk);
 	if (po->rx_ring.pg_vec) {
 		memset(&req_u, 0, sizeof(req_u));
 		packet_set_ring(sk, &req_u, 1, 0);

@@ -3017,6 +3018,7 @@ static int packet_release(struct socket *sock)
 		memset(&req_u, 0, sizeof(req_u));
 		packet_set_ring(sk, &req_u, 1, 1);
 	}
+	release_sock(sk);
 
 	f = fanout_release(sk);
 

@@ -3643,6 +3645,7 @@ packet_setsockopt(struct socket *sock, int level, int optname, char __user *optv
 		union tpacket_req_u req_u;
 		int len;
 
+		lock_sock(sk);
 		switch (po->tp_version) {
 		case TPACKET_V1:
 		case TPACKET_V2:

@@ -3653,12 +3656,17 @@ packet_setsockopt(struct socket *sock, int level, int optname, char __user *optv
 			len = sizeof(req_u.req3);
 			break;
 		}
-		if (optlen < len)
-			return -EINVAL;
-		if (copy_from_user(&req_u.req, optval, len))
-			return -EFAULT;
-		return packet_set_ring(sk, &req_u, 0,
-			optname == PACKET_TX_RING);
+		if (optlen < len) {
+			ret = -EINVAL;
+		} else {
+			if (copy_from_user(&req_u.req, optval, len))
+				ret = -EFAULT;
+			else
+				ret = packet_set_ring(sk, &req_u, 0,
+						    optname == PACKET_TX_RING);
+		}
+		release_sock(sk);
+		return ret;
 	}
 	case PACKET_COPY_THRESH:
 	{

@@ -4208,8 +4216,6 @@ static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 	/* Added to avoid minimal code churn */
 	struct tpacket_req *req = &req_u->req;
 
-	lock_sock(sk);
-
 	rb = tx_ring ? &po->tx_ring : &po->rx_ring;
 	rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
 

@@ -4347,7 +4353,6 @@ static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 	if (pg_vec)
 		free_pg_vec(pg_vec, order, req->tp_block_nr);
 out:
-	release_sock(sk);
 	return err;
 }
 

2.17.0.484.g0c8726318c-goog

From: Eric Dumazet <[email protected]>
Date: Sun, 15 Apr 2018 17:52:04 -0700

...

The locking in AF_PACKET is very unkind, as has been discussed before. Good thing syzbot found this one.

The only other place we access po->tp_version asynchronously is in the getsockopt() for statistics, and I guess that case is OK.

Applied and queued up for -stable, thanks Eric.