io_uring: dont kill fasync under completion_lock

      CPU0                    CPU1
       ----                    ----
  lock(&new->fa_lock);
                               local_irq_disable();
                               lock(&ctx->completion_lock);
                               lock(&new->fa_lock);
  <Interrupt>
    lock(&ctx->completion_lock);

 *** DEADLOCK ***

Move kill_fasync() out of io_commit_cqring() to io_cqring_ev_posted(),
so it doesn't hold completion_lock while doing it. That saves from the
reported deadlock, and it's just nice to shorten the locking time and
untangle nested locks (compl_lock -> wq_head::lock).

Reported-by: syzbot+91ca3f25bd7f795f019c@syzkaller.appspotmail.com
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

fs/io_uring.c

@@ -1345,11 +1345,6 @@ static void __io_commit_cqring(struct io_ring_ctx *ctx)
 
 	/* order cqe stores with ring update */
 	smp_store_release(&rings->cq.tail, ctx->cached_cq_tail);
-
-	if (wq_has_sleeper(&ctx->cq_wait)) {
-		wake_up_interruptible(&ctx->cq_wait);
-		kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
-	}
 }
 
 static void io_put_identity(struct io_uring_task *tctx, struct io_kiocb *req)
@@ -1711,6 +1706,10 @@ static void io_cqring_ev_posted(struct io_ring_ctx *ctx)
 		wake_up(&ctx->sq_data->wait);
 	if (io_should_trigger_evfd(ctx))
 		eventfd_signal(ctx->cq_ev_fd, 1);
+	if (wq_has_sleeper(&ctx->cq_wait)) {
+		wake_up_interruptible(&ctx->cq_wait);
+		kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
+	}
 }
 
 static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx)
@@ -1721,6 +1720,10 @@ static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx)
 	}
 	if (io_should_trigger_evfd(ctx))
 		eventfd_signal(ctx->cq_ev_fd, 1);
+	if (wq_has_sleeper(&ctx->cq_wait)) {
+		wake_up_interruptible(&ctx->cq_wait);
+		kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
+	}
 }
 
 /* Returns true if there are no backlogged entries after the flush */