[PATCH 1/4] ublk: don't cancel commands in QUIESCE_DEV on a device that isn't live

[Josef Bacik <josef@toxicpanda.com>]

QUIESCE_DEV on a device which is not LIVE returns 0 as "already in
expected state", but ublk_cancel_dev() still runs after ub->mutex is
dropped. A device which is QUIESCED because its server died may have a
new server fetching its commands for recovery at that point, and the
cancel completes them without marking anything:

    new server                       QUIESCE_DEV
    START_USER_RECOVERY
    FETCH part of the queue
                                     device is QUIESCED, ret = 0
                                     ublk_cancel_dev()
                                       io->cmd = NULL, command done
    FETCH the rest of the queue
    END_USER_RECOVERY
      device goes LIVE
                                     ublk_queue_rq()
                                       NULL io->cmd

Skip the cancel when the device was not LIVE. Its server is gone, so
there is nothing of the server QUIESCE_DEV is meant for left to cancel.

Fixes: b465ae7b2524 ("ublk: add feature UBLK_F_QUIESCE")
Assisted-by: LLM
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 drivers/block/ublk_drv.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/drivers/block/ublk_drv.c b/drivers/block/ublk_drv.c
index 39eb7775a351..0bd0b95b3217 100644
--- a/drivers/block/ublk_drv.c
+++ b/drivers/block/ublk_drv.c
@@ -5413,6 +5413,7 @@ static int ublk_ctrl_quiesce_dev(struct ublk_device *ub,
 	/* zero means wait forever */
 	u64 timeout_ms = header->data[0];
 	struct gendisk *disk;
+	bool live = true;
 	int ret = -ENODEV;
 
 	if (!(ub->dev_info.flags & UBLK_F_QUIESCE))
@@ -5427,8 +5428,15 @@ static int ublk_ctrl_quiesce_dev(struct ublk_device *ub,
 
 	ret = 0;
 	/* already in expected state */
-	if (ub->dev_info.state != UBLK_S_DEV_LIVE)
+	if (ub->dev_info.state != UBLK_S_DEV_LIVE) {
+		/*
+		 * Nothing to cancel either: the server this was meant for is
+		 * gone, and the commands of the next one, which may be fetching
+		 * them for recovery right now, are not ours to cancel.
+		 */
+		live = false;
 		goto put_disk;
+	}
 
 	/* Mark the device as canceling */
 	mutex_lock(&ub->cancel_mutex);
@@ -5447,7 +5455,7 @@ static int ublk_ctrl_quiesce_dev(struct ublk_device *ub,
 	mutex_unlock(&ub->mutex);
 
 	/* Cancel pending uring_cmd */
-	if (!ret)
+	if (!ret && live)
 		ublk_cancel_dev(ub);
 	return ret;
 }
-- 
2.55.0