io_uring/poll: serialize poll linked timer start with poll removal

We selectively grab the ctx->uring_lock for poll update/removal, but
we really should grab it from the start to fully synchronize with
linked timeouts. Normally this is indeed the case, but if requests
are forced async by the application, we don't fully cover removal
and timer disarm within the uring_lock.

Make this simpler by having consistent locking state for poll removal.

Cc: [email protected] # 6.1+
Reported-by: Querijn Voet <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>

io_uring/poll.c

@@ -977,8 +977,9 @@ int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags)
 	struct io_hash_bucket *bucket;
 	struct io_kiocb *preq;
 	int ret2, ret = 0;
-	struct io_tw_state ts = {};
+	struct io_tw_state ts = { .locked = true };
 
+	io_ring_submit_lock(ctx, issue_flags);
 	preq = io_poll_find(ctx, true, &cd, &ctx->cancel_table, &bucket);
 	ret2 = io_poll_disarm(preq);
 	if (bucket)
@@ -990,12 +991,10 @@ int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags)
 		goto out;
 	}
 
-	io_ring_submit_lock(ctx, issue_flags);
 	preq = io_poll_find(ctx, true, &cd, &ctx->cancel_table_locked, &bucket);
 	ret2 = io_poll_disarm(preq);
 	if (bucket)
 		spin_unlock(&bucket->lock);
-	io_ring_submit_unlock(ctx, issue_flags);
 	if (ret2) {
 		ret = ret2;
 		goto out;
@@ -1019,17 +1018,17 @@ int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags)
 		if (poll_update->update_user_data)
 			preq->cqe.user_data = poll_update->new_user_data;
 
-		ret2 = io_poll_add(preq, issue_flags);
+		ret2 = io_poll_add(preq, issue_flags & ~IO_URING_F_UNLOCKED);
 		/* successfully updated, don't complete poll request */
 		if (!ret2 || ret2 == -EIOCBQUEUED)
 			goto out;
 	}
 
 	req_set_fail(preq);
 	io_req_set_res(preq, -ECANCELED, 0);
-	ts.locked = !(issue_flags & IO_URING_F_UNLOCKED);
 	io_req_task_complete(preq, &ts);
 out:
+	io_ring_submit_unlock(ctx, issue_flags);
 	if (ret < 0) {
 		req_set_fail(req);
 		return ret;