[PATCH] block: don't use blocking queue entered for recursive bio submits

[Jens Axboe <00moses.alexander00@gmail.com>]

commit cd4a4ae4683dc2e09380118e205e057896dcda2b upstream.

If we end up splitting a bio and the queue goes away between
the initial submission and the later split submission, then we
can block forever in blk_queue_enter() waiting for the reference
to drop to zero. This will never happen, since we already hold
a reference.

Mark a split bio as already having entered the queue, so we can
just use the live non-blocking queue enter variant.

Thanks to Tetsuo Handa for the analysis.

We're running fio tests and the tasks get stuck in a D state forever
when systemd-udevd tries to read the partition table. This patch solves
it. Please apply to 4.17 stable.

Reported-by: syzbot+c4f9cebf9d651f6e54de@syzkaller.appspotmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Alexandru Moise <00moses.alexander00@gmail.com>
---
 block/blk-core.c          |  4 +++-
 block/blk-merge.c         | 10 ++++++++++
 include/linux/blk_types.h |  2 ++
 3 files changed, 15 insertions(+), 1 deletion(-)

diff --git a/block/blk-core.c b/block/blk-core.c
index b559b9d4f1a2..47ab2d9d02d9 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -2392,7 +2392,9 @@ blk_qc_t generic_make_request(struct bio *bio)
 
 	if (bio->bi_opf & REQ_NOWAIT)
 		flags = BLK_MQ_REQ_NOWAIT;
-	if (blk_queue_enter(q, flags) < 0) {
+	if (bio_flagged(bio, BIO_QUEUE_ENTERED))
+		blk_queue_enter_live(q);
+	else if (blk_queue_enter(q, flags) < 0) {
 		if (!blk_queue_dying(q) && (bio->bi_opf & REQ_NOWAIT))
 			bio_wouldblock_error(bio);
 		else
diff --git a/block/blk-merge.c b/block/blk-merge.c
index 782940c65d8a..481dc02668f9 100644
--- a/block/blk-merge.c
+++ b/block/blk-merge.c
@@ -210,6 +210,16 @@ void blk_queue_split(struct request_queue *q, struct bio **bio)
 		/* there isn't chance to merge the splitted bio */
 		split->bi_opf |= REQ_NOMERGE;
 
+		/*
+		 * Since we're recursing into make_request here, ensure
+		 * that we mark this bio as already having entered the queue.
+		 * If not, and the queue is going away, we can get stuck
+		 * forever on waiting for the queue reference to drop. But
+		 * that will never happen, as we're already holding a
+		 * reference to it.
+		 */
+		bio_set_flag(*bio, BIO_QUEUE_ENTERED);
+
 		bio_chain(split, *bio);
 		trace_block_split(q, split, (*bio)->bi_iter.bi_sector);
 		generic_make_request(*bio);
diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
index 17b18b91ebac..1602bf4ab4cd 100644
--- a/include/linux/blk_types.h
+++ b/include/linux/blk_types.h
@@ -186,6 +186,8 @@ struct bio {
 				 * throttling rules. Don't do it again. */
 #define BIO_TRACE_COMPLETION 10	/* bio_endio() should trace the final completion
 				 * of this bio. */
+#define BIO_QUEUE_ENTERED 11	/* can use blk_queue_enter_live() */
+
 /* See BVEC_POOL_OFFSET below before adding new flags */
 
 /*
-- 
2.18.0