* [PATCH v2 0/3] dm thin: stop waiting for grace periods when activating and removing thins
@ 2026-11-19 5:45 Qiliang Yuan
2026-11-19 5:45 ` [PATCH v2 1/3] dm thin: drop the grace period from thin_ctr() Qiliang Yuan
` (3 more replies)
0 siblings, 4 replies; 5+ messages in thread
From: Qiliang Yuan @ 2026-11-19 5:45 UTC (permalink / raw)
To: Alasdair Kergon, Mike Snitzer, Mikulas Patocka, Benjamin Marzinski
Cc: dm-devel, linux-kernel, Qiliang Yuan, stable
Activating a thin device waits for an RCU grace period in thin_ctr(),
and removing one waits for another in thin_dtr(). Tools that create and
remove thin snapshots for every container, such as the containerd
devmapper snapshotter, pay for both each time.
Patch 1 drops the grace period from thin_ctr(), which isn't needed.
Patch 2 fixes the pool worker continuing its walk of active_thins from a
thin that thin_dtr() already took off the list, whose successor may have
been freed: a thin now stays on the list until its last reference is
dropped. Patch 3 then lets thin_dtr() free the thin with kfree_rcu()
instead of waiting. Activating a thin device gets 51% faster and
removing one 19%.
Signed-off-by: Qiliang Yuan <qiliang.yuan@linux.dev>
---
V1 -> V2:
- Add patch 2: keep a thin on active_thins until its last reference is
dropped, so that the worker never continues from a thin whose
successor was freed (Benjamin Marzinski)
- Patch 3: rebase onto patch 2, which already skips dying thins, so it
only replaces the grace period with kfree_rcu()
- Patch 1: add Benjamin's Reviewed-by
- Rebase onto current mainline and remeasure
v1: https://lore.kernel.org/r/20261116-bug-dm-thin-rcu-gp-v1-0-14ed3567da3e@linux.dev
---
Qiliang Yuan (3):
dm thin: drop the grace period from thin_ctr()
dm thin: keep a thin on active_thins until its last reference is dropped
dm thin: free thins with kfree_rcu() instead of waiting in thin_dtr()
drivers/md/dm-thin.c | 60 +++++++++++++++++++++++++++++-----------------------
1 file changed, 34 insertions(+), 26 deletions(-)
---
base-commit: 63fdb511c27afcebce797e3db580b3534ddfa639
change-id: 20261114-bug-dm-thin-rcu-gp-d29d4b74a0b8
Best regards,
--
Qiliang Yuan <qiliang.yuan@linux.dev>
^ permalink raw reply [flat|nested] 5+ messages in thread
* [PATCH v2 1/3] dm thin: drop the grace period from thin_ctr()
2026-11-19 5:45 [PATCH v2 0/3] dm thin: stop waiting for grace periods when activating and removing thins Qiliang Yuan
@ 2026-11-19 5:45 ` Qiliang Yuan
2026-11-19 5:45 ` [PATCH v2 2/3] dm thin: keep a thin on active_thins until its last reference is dropped Qiliang Yuan
` (2 subsequent siblings)
3 siblings, 0 replies; 5+ messages in thread
From: Qiliang Yuan @ 2026-11-19 5:45 UTC (permalink / raw)
To: Alasdair Kergon, Mike Snitzer, Mikulas Patocka, Benjamin Marzinski
Cc: dm-devel, linux-kernel, Qiliang Yuan
thin_ctr() adds the new thin to pool->active_thins with
list_add_tail_rcu() and then calls synchronize_rcu(), so that a
wake_worker() call can't miss the bios of a thin that isn't visible yet.
The grace period doesn't make the entry any more visible. A worker that
walks the list after list_add_tail_rcu() finds the thin, and the thin
gets no bios before its table is resumed. A walk already in progress may
miss it, but the wake_worker() for the first bio queues the worker
again. Still, activating a thin device waits for a full grace period,
which takes about half of the activation.
Remove the synchronize_rcu() call.
Activating 100 thin devices of an idle pool with dmsetup create takes
per device (mean of five rounds, min-max of the rounds in parentheses):
before (ms) after (ms) change
15.3 (14.8-15.7) 7.5 (7.3-7.7) -51%
Signed-off-by: Qiliang Yuan <qiliang.yuan@linux.dev>
Reviewed-by: Benjamin Marzinski <bmarzins@redhat.com>
---
drivers/md/dm-thin.c | 7 -------
1 file changed, 7 deletions(-)
diff --git a/drivers/md/dm-thin.c b/drivers/md/dm-thin.c
index 0b65c1b09f47b..bdd250d1eb2e1 100644
--- a/drivers/md/dm-thin.c
+++ b/drivers/md/dm-thin.c
@@ -4292,13 +4292,6 @@ static int thin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
init_completion(&tc->can_destroy);
list_add_tail_rcu(&tc->list, &tc->pool->active_thins);
spin_unlock_irq(&tc->pool->lock);
- /*
- * This synchronize_rcu() call is needed here otherwise we risk a
- * wake_worker() call finding no bios to process (because the newly
- * added tc isn't yet visible). So this reduces latency since we
- * aren't then dependent on the periodic commit to wake_worker().
- */
- synchronize_rcu();
dm_put(pool_md);
--
2.43.0
^ permalink raw reply [flat|nested] 5+ messages in thread
* [PATCH v2 2/3] dm thin: keep a thin on active_thins until its last reference is dropped
2026-11-19 5:45 [PATCH v2 0/3] dm thin: stop waiting for grace periods when activating and removing thins Qiliang Yuan
2026-11-19 5:45 ` [PATCH v2 1/3] dm thin: drop the grace period from thin_ctr() Qiliang Yuan
@ 2026-11-19 5:45 ` Qiliang Yuan
2026-11-19 5:45 ` [PATCH v2 3/3] dm thin: free thins with kfree_rcu() instead of waiting in thin_dtr() Qiliang Yuan
2026-11-19 6:25 ` [PATCH v2 0/3] dm thin: stop waiting for grace periods when activating and removing thins Qiliang Yuan
3 siblings, 0 replies; 5+ messages in thread
From: Qiliang Yuan @ 2026-11-19 5:45 UTC (permalink / raw)
To: Alasdair Kergon, Mike Snitzer, Mikulas Patocka, Benjamin Marzinski
Cc: dm-devel, linux-kernel, Qiliang Yuan, stable
get_first_thin() and get_next_thin() let the pool worker process the
thins on pool->active_thins without holding rcu_read_lock() while it
does: they take a reference on a thin under the lock, and
get_next_thin() continues from that thin in a new read-side section.
thin_dtr() takes the thin off the list, waits for a grace period and
then for the references still held.
Once thin_dtr() has removed a thin that the worker holds, nothing
updates its list.next pointer any more. If the thin after it is removed
and freed before the worker calls get_next_thin(), the walk continues
from freed memory.
Leave a thin on the list until its last reference is dropped and unlink
it then, under the pool lock, so that removing its neighbours keeps its
list.next valid. Have thin_dtr() mark the thin dead, so that no new
walker takes a reference, and make the walkers skip dead thins and the
ones whose last reference is already gone. thin_dtr() waits for the
grace period after the thin is unlinked, before freeing it.
With a test-only 3 s msleep() added to the worker after each thin it
processes, thins 1 and 2 of a pool with three thins are suspended and
then removed while the worker holds thin 1. Trials in which KASAN reports
a slab-use-after-free in get_next_thin():
before (trials) after (trials) change
KASAN report 3 of 3 0 of 5 -100%
Fixes: b10ebd34cccae ("dm thin: fix rcu_read_lock being held in code that can sleep")
Cc: stable@vger.kernel.org
Reported-by: Benjamin Marzinski <bmarzins@redhat.com>
Closes: https://lore.kernel.org/r/avz7qDWd2g-uubmL@redhat.com
Suggested-by: Benjamin Marzinski <bmarzins@redhat.com>
Signed-off-by: Qiliang Yuan <qiliang.yuan@linux.dev>
---
drivers/md/dm-thin.c | 51 +++++++++++++++++++++++++++++++++------------------
1 file changed, 33 insertions(+), 18 deletions(-)
diff --git a/drivers/md/dm-thin.c b/drivers/md/dm-thin.c
index bdd250d1eb2e1..f79f5982ae36c 100644
--- a/drivers/md/dm-thin.c
+++ b/drivers/md/dm-thin.c
@@ -356,10 +356,12 @@ struct thin_c {
/*
* Ensures the thin is not destroyed until the worker has finished
- * iterating the active_thins list.
+ * iterating the active_thins list. Set dead once thin_dtr() starts,
+ * so that no new walker picks the thin up.
*/
refcount_t refcount;
struct completion can_destroy;
+ bool dead;
};
/*----------------------------------------------------------------*/
@@ -2318,25 +2320,28 @@ static void process_thin_deferred_cells(struct thin_c *tc)
} while (!list_empty(&cells));
}
-static void thin_get(struct thin_c *tc);
+static bool thin_tryget(struct thin_c *tc);
static void thin_put(struct thin_c *tc);
/*
* We can't hold rcu_read_lock() around code that can block. So we
* find a thin with the rcu lock held; bump a refcount; then drop
- * the lock.
+ * the lock. Skip the thins that thin_dtr() is destroying.
*/
static struct thin_c *get_first_thin(struct pool *pool)
{
- struct thin_c *tc = NULL;
+ struct thin_c *tc;
rcu_read_lock();
- tc = list_first_or_null_rcu(&pool->active_thins, struct thin_c, list);
- if (tc)
- thin_get(tc);
+ list_for_each_entry_rcu(tc, &pool->active_thins, list) {
+ if (thin_tryget(tc)) {
+ rcu_read_unlock();
+ return tc;
+ }
+ }
rcu_read_unlock();
- return tc;
+ return NULL;
}
static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc)
@@ -2345,7 +2350,8 @@ static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc)
rcu_read_lock();
list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) {
- thin_get(tc);
+ if (!thin_tryget(tc))
+ continue;
thin_put(old_tc);
rcu_read_unlock();
return tc;
@@ -4131,28 +4137,37 @@ static struct target_type pool_target = {
* Thin target methods
*--------------------------------------------------------------
*/
-static void thin_get(struct thin_c *tc)
+static bool thin_tryget(struct thin_c *tc)
{
- refcount_inc(&tc->refcount);
+ return !READ_ONCE(tc->dead) && refcount_inc_not_zero(&tc->refcount);
}
static void thin_put(struct thin_c *tc)
{
- if (refcount_dec_and_test(&tc->refcount))
- complete(&tc->can_destroy);
+ unsigned long flags;
+
+ if (!refcount_dec_and_test(&tc->refcount))
+ return;
+
+ /*
+ * Unlink tc only once nobody holds it: a worker holding it continues
+ * its walk from tc->list.next.
+ */
+ spin_lock_irqsave(&tc->pool->lock, flags);
+ list_del_rcu(&tc->list);
+ spin_unlock_irqrestore(&tc->pool->lock, flags);
+ complete(&tc->can_destroy);
}
static void thin_dtr(struct dm_target *ti)
{
struct thin_c *tc = ti->private;
- spin_lock_irq(&tc->pool->lock);
- list_del_rcu(&tc->list);
- spin_unlock_irq(&tc->pool->lock);
- synchronize_rcu();
-
+ WRITE_ONCE(tc->dead, true);
thin_put(tc);
wait_for_completion(&tc->can_destroy);
+ /* Let RCU readers of active_thins finish with tc */
+ synchronize_rcu();
mutex_lock(&dm_thin_pool_table.mutex);
--
2.43.0
^ permalink raw reply [flat|nested] 5+ messages in thread
* [PATCH v2 3/3] dm thin: free thins with kfree_rcu() instead of waiting in thin_dtr()
2026-11-19 5:45 [PATCH v2 0/3] dm thin: stop waiting for grace periods when activating and removing thins Qiliang Yuan
2026-11-19 5:45 ` [PATCH v2 1/3] dm thin: drop the grace period from thin_ctr() Qiliang Yuan
2026-11-19 5:45 ` [PATCH v2 2/3] dm thin: keep a thin on active_thins until its last reference is dropped Qiliang Yuan
@ 2026-11-19 5:45 ` Qiliang Yuan
2026-11-19 6:25 ` [PATCH v2 0/3] dm thin: stop waiting for grace periods when activating and removing thins Qiliang Yuan
3 siblings, 0 replies; 5+ messages in thread
From: Qiliang Yuan @ 2026-11-19 5:45 UTC (permalink / raw)
To: Alasdair Kergon, Mike Snitzer, Mikulas Patocka, Benjamin Marzinski
Cc: dm-devel, linux-kernel, Qiliang Yuan
thin_dtr() waits until the last reference on the thin is dropped, which
also takes it off pool->active_thins, and then for a grace period, so
that requeue_bios() and error_retry_list_with_code(), which walk the
list under rcu_read_lock(), are done with the thin before it is freed.
Removing a thin device thus waits for a full grace period in thin_dtr(),
although those readers only need the memory of the thin to stay around.
Free the thin with kfree_rcu() instead of waiting for the grace period.
Removing 100 thin devices of an idle pool with dmsetup remove takes per
device (mean of five rounds, min-max of the rounds in parentheses). The
queue freeze in del_gendisk() waits for a grace period of its own, which
the one in thin_dtr() used to overlap, so only part of it is saved:
before (ms) after (ms) change
16.6 (16.2-17.0) 13.5 (13.2-14.0) -19%
Signed-off-by: Qiliang Yuan <qiliang.yuan@linux.dev>
---
drivers/md/dm-thin.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/drivers/md/dm-thin.c b/drivers/md/dm-thin.c
index f79f5982ae36c..9d9b3cd817f7a 100644
--- a/drivers/md/dm-thin.c
+++ b/drivers/md/dm-thin.c
@@ -362,6 +362,7 @@ struct thin_c {
refcount_t refcount;
struct completion can_destroy;
bool dead;
+ struct rcu_head rcu;
};
/*----------------------------------------------------------------*/
@@ -4166,8 +4167,6 @@ static void thin_dtr(struct dm_target *ti)
WRITE_ONCE(tc->dead, true);
thin_put(tc);
wait_for_completion(&tc->can_destroy);
- /* Let RCU readers of active_thins finish with tc */
- synchronize_rcu();
mutex_lock(&dm_thin_pool_table.mutex);
@@ -4176,7 +4175,8 @@ static void thin_dtr(struct dm_target *ti)
dm_put_device(ti, tc->pool_dev);
if (tc->origin_dev)
dm_put_device(ti, tc->origin_dev);
- kfree(tc);
+ /* Let RCU readers of active_thins finish with tc before freeing it */
+ kfree_rcu(tc, rcu);
mutex_unlock(&dm_thin_pool_table.mutex);
}
--
2.43.0
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH v2 0/3] dm thin: stop waiting for grace periods when activating and removing thins
2026-11-19 5:45 [PATCH v2 0/3] dm thin: stop waiting for grace periods when activating and removing thins Qiliang Yuan
` (2 preceding siblings ...)
2026-11-19 5:45 ` [PATCH v2 3/3] dm thin: free thins with kfree_rcu() instead of waiting in thin_dtr() Qiliang Yuan
@ 2026-11-19 6:25 ` Qiliang Yuan
3 siblings, 0 replies; 5+ messages in thread
From: Qiliang Yuan @ 2026-11-19 6:25 UTC (permalink / raw)
To: Benjamin Marzinski
Cc: Qiliang Yuan, Alasdair Kergon, Mike Snitzer, Mikulas Patocka,
dm-devel, linux-kernel, stable
On Thu, Nov 19, 2026 at 12:24:21AM -0500, Benjamin Marzinski wrote:
> I posted an alternate version of your [PATCH 2/2] that fixes the issue
> I mentioned. Can you verify that it works for you?
>
> I squashed it with your patch, since it your original patch did most of
> the work of fixing the issue, and it seemed easier to just have one
> patch to backport to fix it. I could refactor it into two patches: one
> the fixes the bug, and a second that just swaps the synchronize_rcu()
> for kfree_rcu(), if people would rather not mix the bug fix and feature.
Ben, this v2 crossed with your patch [1], and its 2/3 carries the wrong
Fixes tag. I'll send a v3 built on your version, still as three patches,
so that the fix can go to stable without the kfree_rcu() change:
- 1/3 "dm thin: drop the grace period from thin_ctr()": unchanged, and
it keeps your Reviewed-by.
- 2/3: the fix, taken from your patch. thin_dtr() marks the thin dying,
the walkers take references with refcount_inc_not_zero() and skip
dying thins, requeue_bios() and error_retry_list_with_code() skip
them too, and the thin leaves active_thins only after its refcount
drops to zero, followed by synchronize_rcu() before it is freed.
Fixes: 17181fb7a0c3, Cc: stable, with your Co-developed-by and
Signed-off-by.
- 3/3: replace that synchronize_rcu() with kfree_rcu().
Before sending it I'll rerun the KASAN reproducer on 2/3 alone and
remeasure each patch.
[1] https://lore.kernel.org/r/20261119051620.507189-1-bmarzins@redhat.com
Thanks,
Qiliang
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2026-11-19 6:25 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-19 5:45 [PATCH v2 0/3] dm thin: stop waiting for grace periods when activating and removing thins Qiliang Yuan
2026-11-19 5:45 ` [PATCH v2 1/3] dm thin: drop the grace period from thin_ctr() Qiliang Yuan
2026-11-19 5:45 ` [PATCH v2 2/3] dm thin: keep a thin on active_thins until its last reference is dropped Qiliang Yuan
2026-11-19 5:45 ` [PATCH v2 3/3] dm thin: free thins with kfree_rcu() instead of waiting in thin_dtr() Qiliang Yuan
2026-11-19 6:25 ` [PATCH v2 0/3] dm thin: stop waiting for grace periods when activating and removing thins Qiliang Yuan
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®