[Devel] [PATCH RHEL8 COMMIT] ploop: Update cached bat_entries before bio completion

[Konstantin Khorenko]

The commit is pushed to "branch-rh8-4.18.0-80.1.2.vz8.3.x-ovz" and will appear at https://src.openvz.org/scm/ovz/vzkernel.git
after rh8-4.18.0-80.1.2.vz8.3.2
------>
commit 1c743b026e999dc34f411b80e6fc69d4132a5daf
Author: Kirill Tkhai <ktkhai at virtuozzo.com>
Date:   Tue Mar 10 18:05:26 2020 +0300

    ploop: Update cached bat_entries before bio completion
    
    Right after data_bio->end_io is called, it's expected
    a subsequent read returns written data. But there is
    a race between ploop_bat_write_complete() and ploop_map():
    
    WRITE                                   READ
    ploop_bat_write_complete()
      data_bio->end_io()
                                            write submitter sees write completion
                                            next submittion is READ
                                            ploop_map()
                                              check BAT --- sees zero BAT entry
                                              zero_fill_bio()
      update BAT
    
    So, the order has to be vise versa: first is to update BAT cache,
    second is to call data_bio->end_io().
    
    https://jira.sw.ru/browse/PSBM-101499
    
    Signed-off-by: Kirill Tkhai <ktkhai at virtuozzo.com>
---
 drivers/md/dm-ploop-map.c | 21 ++++++++++++---------
 1 file changed, 12 insertions(+), 9 deletions(-)

diff --git a/drivers/md/dm-ploop-map.c b/drivers/md/dm-ploop-map.c
index a67656b2b01e..4ca0f8e83400 100644
--- a/drivers/md/dm-ploop-map.c
+++ b/drivers/md/dm-ploop-map.c
@@ -623,6 +623,18 @@ static void ploop_bat_write_complete(struct bio *bio)
 
 	track_bio(ploop, bio);
 
+	if (!bio->bi_status) {
+		/*
+		 * Success: now update local BAT copy. We could do this
+		 * from our delayed work, but we want to publish new
+		 * mapping in the fastest way. This must be done before
+		 * data bios completion, since right after we complete
+		 * a bio, subsequent read wants to see written data
+		 * (ploop_map() wants to see not zero bat_entries[.]).
+		 */
+		ploop_advance_local_after_bat_wb(ploop, piwb, true);
+	}
+
 	spin_lock_irqsave(&piwb->lock, flags);
 	piwb->completed = true;
 	piwb->bi_status = bio->bi_status;
@@ -644,15 +656,6 @@ static void ploop_bat_write_complete(struct bio *bio)
 		complete_cow(cow, bio->bi_status);
 	}
 
-	if (!piwb->bi_status) {
-		/*
-		 * Success: now update local BAT copy. We could do this
-		 * from our delayed work, but we want to publish new
-		 * mapping in the fastest way.
-		 */
-		ploop_advance_local_after_bat_wb(ploop, piwb, true);
-	}
-
 	/*
 	 * In case of update BAT is failed, dst_clusters will be
 	 * set back to holes_bitmap on last put_piwb().