Syzbot found a bug "BUG: sleeping function called from invalid context
at kernel/locking/mutex.c:580". It is because hci_link_tx_to holds an
RCU read lock and calls hci_disconnect which would hold a mutex lock
since the commit a13f316e90fd ("Bluetooth: hci_conn: Consolidate code
for aborting connections"). Here's an example call trace:
__dump_stack lib/dump_stack.c:88 [inline]
dump_stack_lvl+0xfc/0x174 lib/dump_stack.c:106
___might_sleep+0x4a9/0x4d3 kernel/sched/core.c:9663
__mutex_lock_common kernel/locking/mutex.c:576 [inline]
__mutex_lock+0xc7/0x6e7 kernel/locking/mutex.c:732
hci_cmd_sync_queue+0x3a/0x287 net/bluetooth/hci_sync.c:388
hci_abort_conn+0x2cd/0x2e4 net/bluetooth/hci_conn.c:1812
hci_disconnect+0x207/0x237 net/bluetooth/hci_conn.c:244
hci_link_tx_to net/bluetooth/hci_core.c:3254 [inline]
__check_timeout net/bluetooth/hci_core.c:3419 [inline]
__check_timeout+0x310/0x361 net/bluetooth/hci_core.c:3399
hci_sched_le net/bluetooth/hci_core.c:3602 [inline]
hci_tx_work+0xe8f/0x12d0 net/bluetooth/hci_core.c:3652
process_one_work+0x75c/0xba1 kernel/workqueue.c:2310
worker_thread+0x5b2/0x73a kernel/workqueue.c:2457
kthread+0x2f7/0x30b kernel/kthread.c:319
ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:298
This patch releases RCU read lock before calling hci_disconnect and
reacquires it afterward to fix the bug.
Fixes: a13f316e90fd ("Bluetooth: hci_conn: Consolidate code for aborting connections")
Signed-off-by: Ying Hsu <yinghsu@xxxxxxxxxxxx>
---
Tested this commit using a C reproducer on qemu-x86_64 for 10 minutes.
net/bluetooth/hci_core.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
index a5992f1b3c9b..db4f28d68d71 100644
--- a/net/bluetooth/hci_core.c
+++ b/net/bluetooth/hci_core.c
@@ -3418,7 +3418,12 @@ static void hci_link_tx_to(struct hci_dev *hdev, __u8 type)
if (c->type == type && c->sent) {
bt_dev_err(hdev, "killing stalled connection %pMR",
&c->dst);
+ /* hci_disconnect might sleep, so, we have to release
+ * the RCU read lock before calling it.
+ */
+ rcu_read_unlock();
hci_disconnect(c, HCI_ERROR_REMOTE_USER_TERM);
+ rcu_read_lock();
}
}
--
2.42.0.283.g2d96d420d3-goog
