nbd: fix race between timeout and normal completion

[ Upstream commit c9ea57c ]

If request timetout is handled by nbd_requeue_cmd(), normal completion
has to be stopped for avoiding to complete this requeued request, other
use-after-free can be triggered.

Fix the race by clearing NBD_CMD_INFLIGHT in nbd_requeue_cmd(), meantime
make sure that cmd->lock is grabbed for clearing the flag and the
requeue.

Cc: Josef Bacik <[email protected]>
Cc: Yu Kuai <[email protected]>
Fixes: 2895f18 ("nbd: don't clear 'NBD_CMD_INFLIGHT' flag if request is not completed")
Signed-off-by: Ming Lei <[email protected]>
Reviewed-by: Yu Kuai <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jens Axboe <[email protected]>
Signed-off-by: Sasha Levin <[email protected]>

Author: Ming Lei

drivers/block/nbd.c

@@ -181,6 +181,17 @@ static void nbd_requeue_cmd(struct nbd_cmd *cmd)
 {
 	struct request *req = blk_mq_rq_from_pdu(cmd);
 
+	lockdep_assert_held(&cmd->lock);
+
+	/*
+	 * Clear INFLIGHT flag so that this cmd won't be completed in
+	 * normal completion path
+	 *
+	 * INFLIGHT flag will be set when the cmd is queued to nbd next
+	 * time.
+	 */
+	__clear_bit(NBD_CMD_INFLIGHT, &cmd->flags);
+
 	if (!test_and_set_bit(NBD_CMD_REQUEUED, &cmd->flags))
 		blk_mq_requeue_request(req, true);
 }
@@ -461,8 +472,8 @@ static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req)
 					nbd_mark_nsock_dead(nbd, nsock, 1);
 				mutex_unlock(&nsock->tx_lock);
 			}
-			mutex_unlock(&cmd->lock);
 			nbd_requeue_cmd(cmd);
+			mutex_unlock(&cmd->lock);
 			nbd_config_put(nbd);
 			return BLK_EH_DONE;
 		}