[PATCH] Bluetooth: check for zapped sk before connecting

From: Archie Pusaka <redacted>

There is a possibility of receiving a zapped sock on
l2cap_sock_connect(). This could lead to interesting crashes, one
such case is tearing down an already tore l2cap_sock as is happened
with this call trace:

__dump_stack lib/dump_stack.c:15 [inline]
dump_stack+0xc4/0x118 lib/dump_stack.c:56
register_lock_class kernel/locking/lockdep.c:792 [inline]
register_lock_class+0x239/0x6f6 kernel/locking/lockdep.c:742
__lock_acquire+0x209/0x1e27 kernel/locking/lockdep.c:3105
lock_acquire+0x29c/0x2fb kernel/locking/lockdep.c:3599
__raw_spin_lock_bh include/linux/spinlock_api_smp.h:137 [inline]
_raw_spin_lock_bh+0x38/0x47 kernel/locking/spinlock.c:175
spin_lock_bh include/linux/spinlock.h:307 [inline]
lock_sock_nested+0x44/0xfa net/core/sock.c:2518
l2cap_sock_teardown_cb+0x88/0x2fb net/bluetooth/l2cap_sock.c:1345
l2cap_chan_del+0xa3/0x383 net/bluetooth/l2cap_core.c:598
l2cap_chan_close+0x537/0x5dd net/bluetooth/l2cap_core.c:756
l2cap_chan_timeout+0x104/0x17e net/bluetooth/l2cap_core.c:429
process_one_work+0x7e3/0xcb0 kernel/workqueue.c:2064
worker_thread+0x5a5/0x773 kernel/workqueue.c:2196
kthread+0x291/0x2a6 kernel/kthread.c:211
ret_from_fork+0x4e/0x80 arch/x86/entry/entry_64.S:604

Signed-off-by: Archie Pusaka <redacted>
Reported-by: syzbot+abfc0f5e668d4099af73@syzkaller.appspotmail.com
Reviewed-by: Alain Michaud <redacted>
Reviewed-by: Abhishek Pandit-Subedi <abhishekpandit@chromium.org>
Reviewed-by: Guenter Roeck <groeck@chromium.org>
---

 net/bluetooth/l2cap_sock.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/net/bluetooth/l2cap_sock.c b/net/bluetooth/l2cap_sock.c
index f1b1edd0b697..b86fd8cc4dc1 100644
--- a/net/bluetooth/l2cap_sock.c
+++ b/net/bluetooth/l2cap_sock.c

@@ -182,6 +182,13 @@ static int l2cap_sock_connect(struct socket *sock, struct sockaddr *addr,
 
 	BT_DBG("sk %p", sk);
 
+	lock_sock(sk);
+	if (sock_flag(sk, SOCK_ZAPPED)) {
+		release_sock(sk);
+		return -EINVAL;
+	}
+	release_sock(sk);
+
 	if (!addr || alen < offsetofend(struct sockaddr, sa_family) ||
 	    addr->sa_family != AF_BLUETOOTH)
 		return -EINVAL;

-- 
2.31.0.rc2.261.g7f71774620-goog

Hi Archie,

hmmm. I wonder if this would look better and easy to see that the locking is done correctly.

	lock_sock(sk);
	zapped = sock_flag(sk, SOCK_ZAPPED);
	release_sock(sk);

	if (zapped)
		return -EINVAL;

Regards

Marcel

Hi Marcel,

Thanks for your suggestion. I implemented it in v2, please take another look.


On Mon, 22 Mar 2021 at 23:53, Marcel Holtmann [off-list ref] wrote:

Hi Archie,

quoted

There is a possibility of receiving a zapped sock on
l2cap_sock_connect(). This could lead to interesting crashes, one
such case is tearing down an already tore l2cap_sock as is happened
with this call trace:

__dump_stack lib/dump_stack.c:15 [inline]
dump_stack+0xc4/0x118 lib/dump_stack.c:56
register_lock_class kernel/locking/lockdep.c:792 [inline]
register_lock_class+0x239/0x6f6 kernel/locking/lockdep.c:742
__lock_acquire+0x209/0x1e27 kernel/locking/lockdep.c:3105
lock_acquire+0x29c/0x2fb kernel/locking/lockdep.c:3599
__raw_spin_lock_bh include/linux/spinlock_api_smp.h:137 [inline]
_raw_spin_lock_bh+0x38/0x47 kernel/locking/spinlock.c:175
spin_lock_bh include/linux/spinlock.h:307 [inline]
lock_sock_nested+0x44/0xfa net/core/sock.c:2518
l2cap_sock_teardown_cb+0x88/0x2fb net/bluetooth/l2cap_sock.c:1345
l2cap_chan_del+0xa3/0x383 net/bluetooth/l2cap_core.c:598
l2cap_chan_close+0x537/0x5dd net/bluetooth/l2cap_core.c:756
l2cap_chan_timeout+0x104/0x17e net/bluetooth/l2cap_core.c:429
process_one_work+0x7e3/0xcb0 kernel/workqueue.c:2064
worker_thread+0x5a5/0x773 kernel/workqueue.c:2196
kthread+0x291/0x2a6 kernel/kthread.c:211
ret_from_fork+0x4e/0x80 arch/x86/entry/entry_64.S:604

Signed-off-by: Archie Pusaka <redacted>
Reported-by: syzbot+abfc0f5e668d4099af73@syzkaller.appspotmail.com
Reviewed-by: Alain Michaud <redacted>
Reviewed-by: Abhishek Pandit-Subedi <abhishekpandit@chromium.org>
Reviewed-by: Guenter Roeck <groeck@chromium.org>
---

net/bluetooth/l2cap_sock.c | 7 +++++++
1 file changed, 7 insertions(+)

diff --git a/net/bluetooth/l2cap_sock.c b/net/bluetooth/l2cap_sock.c
index f1b1edd0b697..b86fd8cc4dc1 100644
--- a/net/bluetooth/l2cap_sock.c
+++ b/net/bluetooth/l2cap_sock.c

@@ -182,6 +182,13 @@ static int l2cap_sock_connect(struct socket *sock, struct sockaddr *addr,

      BT_DBG("sk %p", sk);

+     lock_sock(sk);
+     if (sock_flag(sk, SOCK_ZAPPED)) {
+             release_sock(sk);
+             return -EINVAL;
+     }
+     release_sock(sk);
+

hmmm. I wonder if this would look better and easy to see that the locking is done correctly.

        lock_sock(sk);
        zapped = sock_flag(sk, SOCK_ZAPPED);
        release_sock(sk);

        if (zapped)
                return -EINVAL;

Regards

Marcel