[Devel] [PATCH hci-8.1] dm-ploop: do not lock the md pages of a delta read for a BAT walk

[Konstantin Khorenko]

notify_delta_merged() and ploop_check_delta_before_flip() walk the live
BAT together with the BAT of a delta that ploop_read_delta_metadata()
read into a local md tree. For each page they take the md_lock of the
live page and then the md_lock of the local one.

Every md_lock is initialized in ploop_alloc_md_page(), so both locks
share one lockdep class. With CONFIG_PROVE_LOCKING the first walk
reports "possible recursive locking detected", and lockdep turns itself
off for the rest of the boot.

Only the cmd that read the local tree can reach it, and it frees the
tree after the walk, so its md_lock protects nothing. Stop taking it.

https://virtuozzo.atlassian.net/browse/VSTOR-153709

Fixes: 0d4b4d0a20a9 ("dm-ploop: introduce per-md page locking")
Feature: dm-ploop: ploop target driver
Signed-off-by: Konstantin Khorenko <khorenko at virtuozzo.com>
---
This obsoletes "[PATCH vz10 10/24] dm-ploop: annotate nested md_lock in
ploop_check_delta_before_flip()" of the VSTOR-137234 selftests series.
Pavel pointed out that the SINGLE_DEPTH_NESTING annotation works around
a lock that is not needed at all, and that notify_delta_merged() has
the same problem. This patch stops taking that lock in both walks.

Assisted-by: LLM
---
 drivers/md/dm-ploop-cmd.c | 4 ----
 1 file changed, 4 deletions(-)

diff --git a/drivers/md/dm-ploop-cmd.c b/drivers/md/dm-ploop-cmd.c
index 2d1fa449b052d..f6135e07c2737 100644
--- a/drivers/md/dm-ploop-cmd.c
+++ b/drivers/md/dm-ploop-cmd.c
@@ -812,7 +812,6 @@ static void notify_delta_merged(struct ploop *ploop, u8 level,
 		d_bat_entries = d_md->kmpage;
 
 		spin_lock_irqsave(&md->md_lock, flags); /* write */
-		spin_lock(&d_md->md_lock); /* write */
 		for (; i <= end; i++) {
 			clu = ploop_page_clu_idx_to_bat_clu(md->id, i);
 			if (clu == nr_be - 1)
@@ -846,7 +845,6 @@ static void notify_delta_merged(struct ploop *ploop, u8 level,
 				WRITE_ONCE(md->bat_levels[i], level);
 		}
 
-		spin_unlock(&d_md->md_lock);
 		spin_unlock_irqrestore(&md->md_lock, flags);
 
 		if (stop)
@@ -1147,7 +1145,6 @@ static int ploop_check_delta_before_flip(struct ploop *ploop, struct file *file)
 		d_bat_entries = d_md->kmpage;
 
 		spin_lock_irq(&md->md_lock); /* read */
-		spin_lock(&d_md->md_lock);
 		for (; i <= end; i++) {
 			if (ploop_md_page_cluster_is_in_top_delta(ploop, md, i) &&
 			    d_bat_entries[i] != BAT_ENTRY_NONE) {
@@ -1156,7 +1153,6 @@ static int ploop_check_delta_before_flip(struct ploop *ploop, struct file *file)
 				break;
 			}
 		}
-		spin_unlock(&d_md->md_lock);
 		spin_unlock_irq(&md->md_lock);
 		if (stop)
 			break;

---
base-commit: f0cff9cebdb0d481d05b02bdb7bdafe864adca1c
change-id: 20261105-b4-prep-khorenko-vstor-153709-dm-ploop-drop-the-lock-of-local-md-pages-4f25ac778924

Best regards,
--  
Konstantin Khorenko,
Virtuozzo Linux Kernel Team