From: Jens Axboe <axboe@kernel.dk>
Date: Tue, 31 Aug 2021 19:53:00 +0000 (-0600)
Subject: io-wq: fix queue stalling race
X-Git-Url: http://git.maquefel.me/?a=commitdiff_plain;h=0242f6426ea78fbe3933b44f8c55ae93ec37f6cc;p=linux.git

io-wq: fix queue stalling race

We need to set the stalled bit early, before we drop the lock for adding
us to the stall hash queue. If not, then we can race with new work being
queued between adding us to the stall hash and io_worker_handle_work()
marking us stalled.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---

diff --git a/fs/io-wq.c b/fs/io-wq.c
index a94060b72f84f..aa9656eb832e0 100644
--- a/fs/io-wq.c
+++ b/fs/io-wq.c
@@ -436,8 +436,7 @@ static bool io_worker_can_run_work(struct io_worker *worker,
 }
 
 static struct io_wq_work *io_get_next_work(struct io_wqe *wqe,
-					   struct io_worker *worker,
-					   bool *stalled)
+					   struct io_worker *worker)
 	__must_hold(wqe->lock)
 {
 	struct io_wq_work_node *node, *prev;
@@ -475,10 +474,14 @@ static struct io_wq_work *io_get_next_work(struct io_wqe *wqe,
 	}
 
 	if (stall_hash != -1U) {
+		/*
+		 * Set this before dropping the lock to avoid racing with new
+		 * work being added and clearing the stalled bit.
+		 */
+		wqe->flags |= IO_WQE_FLAG_STALLED;
 		raw_spin_unlock(&wqe->lock);
 		io_wait_on_hash(wqe, stall_hash);
 		raw_spin_lock(&wqe->lock);
-		*stalled = true;
 	}
 
 	return NULL;
@@ -518,7 +521,6 @@ static void io_worker_handle_work(struct io_worker *worker)
 
 	do {
 		struct io_wq_work *work;
-		bool stalled;
 get_next:
 		/*
 		 * If we got some work, mark us as busy. If we didn't, but
@@ -527,12 +529,9 @@ get_next:
 		 * can't make progress, any work completion or insertion will
 		 * clear the stalled flag.
 		 */
-		stalled = false;
-		work = io_get_next_work(wqe, worker, &stalled);
+		work = io_get_next_work(wqe, worker);
 		if (work)
 			__io_worker_busy(wqe, worker, work);
-		else if (stalled)
-			wqe->flags |= IO_WQE_FLAG_STALLED;
 
 		raw_spin_unlock(&wqe->lock);
 		if (!work)