io_uring: fix potential deadlock in io_poll_wake()

commit 7c9e7f0fe0d8abf856a957c150c48778e75154c1 upstream.

We attempt to run the poll completion inline, but we're using trylock to
do so. This avoids a deadlock since we're grabbing the locks in reverse
order at this point, we already hold the poll wq lock and we're trying
to grab the completion lock, while the normal rules are the reverse of
that order.

IO completion for a timeout link will need to grab the completion lock,
but that's not safe from this context. Put the completion under the
completion_lock in io_poll_wake(), and mark the request as entering
the completion with the completion_lock already held.

Fixes: 2665abfd757f ("io_uring: add support for linked SQE timeouts")
Signed-off-by: NJens Axboe <axboe@kernel.dk>
Signed-off-by: NJoseph Qi <joseph.qi@linux.alibaba.com>
Reviewed-by: NXiaoguang Wang <xiaoguang.wang@linux.alibaba.com>

fs/io_uring.c

@@ -342,6 +342,7 @@ struct io_kiocb {
 #define REQ_F_MUST_PUNT		4096	/* must be punted even for NONBLOCK */
 #define REQ_F_TIMEOUT_NOSEQ	8192	/* no timeout sequence */
 #define REQ_F_INFLIGHT		16384	/* on inflight list */
+#define REQ_F_COMP_LOCKED	32768	/* completion under lock */
 	u64			user_data;
 	u32			result;
 	u32			sequence;
@@ -935,14 +936,15 @@ static void io_free_req_find_next(struct io_kiocb *req, struct io_kiocb **nxt)
 	 */
 	if (req->flags & REQ_F_FAIL_LINK) {
 		io_fail_links(req);
-	} else if (req->flags & REQ_F_LINK_TIMEOUT) {
+	} else if ((req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_COMP_LOCKED)) ==
+			REQ_F_LINK_TIMEOUT) {
 		struct io_ring_ctx *ctx = req->ctx;
 		unsigned long flags;
 
 		/*
 		 * If this is a timeout link, we could be racing with the
 		 * timeout timer. Grab the completion lock for this case to
-		 * protection against that.
+		 * protect against that.
 		 */
 		spin_lock_irqsave(&ctx->completion_lock, flags);
 		io_req_link_next(req, nxt);
@@ -2069,13 +2071,20 @@ static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
 
 	list_del_init(&poll->wait.entry);
 
+	/*
+	 * Run completion inline if we can. We're using trylock here because
+	 * we are violating the completion_lock -> poll wq lock ordering.
+	 * If we have a link timeout we're going to need the completion_lock
+	 * for finalizing the request, mark us as having grabbed that already.
+	 */
 	if (mask && spin_trylock_irqsave(&ctx->completion_lock, flags)) {
 		list_del(&req->list);
 		io_poll_complete(req, mask);
+		req->flags |= REQ_F_COMP_LOCKED;
+		io_put_req(req);
 		spin_unlock_irqrestore(&ctx->completion_lock, flags);
 
 		io_cqring_ev_posted(ctx);
-		io_put_req(req);
 	} else {
 		io_queue_async_work(req);
 	}