2018-01-08 02:00:42

by Jiang Biao

Subject: [PATCH] fs/mbcache: make sure mb_cache_count() not return negative value.

When running ltp stress test for 7*24 hours, vmscan occasionally emits the
following warning continuously:

mb_cache_scan+0x0/0x3f0 negative objects to delete
nr=-9232265467809300450
....

Trace info shows the freeable(mb_cache_count returns) is -1, which causes
the continuous accumulation and overflow of total_scan.

This patch makes sure that mb_cache_count() not return a negative value,
which makes the mbcache shrinker more robust.

Signed-off-by: Jiang Biao <[email protected]>
CC: "Theodore Ts'o" <[email protected]>
CC: Eric Biggers <[email protected]>
CC: Andrew Morton <[email protected]>
CC: Jan Kara <[email protected]>
---
fs/mbcache.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/fs/mbcache.c b/fs/mbcache.c
index b8b8b9c..c758458 100644
--- a/fs/mbcache.c
+++ b/fs/mbcache.c
@@ -238,7 +238,9 @@ void mb_cache_entry_delete(struct mb_cache *cache, u32 key, u64 value)
spin_lock(&cache->c_list_lock);
if (!list_empty(&entry->e_list)) {
list_del_init(&entry->e_list);
- cache->c_entry_count--;
+ /*Make sure c_entry_count is not zero before dec*/
+ if (cache->c_entry_count != 0)
+ cache->c_entry_count--;
atomic_dec(&entry->e_refcnt);
}
spin_unlock(&cache->c_list_lock);
--
2.7.4

2018-01-08 10:44:49

by Jan Kara

Subject: Re: [PATCH] fs/mbcache: make sure mb_cache_count() not return negative value.

On Mon 08-01-18 10:00:42, Jiang Biao wrote:
> When running ltp stress test for 7*24 hours, vmscan occasionally emits the
> following warning continuously:
>
> mb_cache_scan+0x0/0x3f0 negative objects to delete
> nr=-9232265467809300450
> ....
>
> Trace info shows the freeable(mb_cache_count returns) is -1, which causes
> the continuous accumulation and overflow of total_scan.
>
> This patch makes sure that mb_cache_count() not return a negative value,
> which makes the mbcache shrinker more robust.

Thanks for the patch. Couple of comments below.

> Signed-off-by: Jiang Biao <[email protected]>
> CC: "Theodore Ts'o" <[email protected]>
> CC: Eric Biggers <[email protected]>
> CC: Andrew Morton <[email protected]>
> CC: Jan Kara <[email protected]>
> ---
> fs/mbcache.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/fs/mbcache.c b/fs/mbcache.c
> index b8b8b9c..c758458 100644
> --- a/fs/mbcache.c
> +++ b/fs/mbcache.c
> @@ -238,7 +238,9 @@ void mb_cache_entry_delete(struct mb_cache *cache, u32 key, u64 value)
> spin_lock(&cache->c_list_lock);
> if (!list_empty(&entry->e_list)) {
> list_del_init(&entry->e_list);
> - cache->c_entry_count--;
> + /*Make sure c_entry_count is not zero before dec*/

The comment is useless, just delete it.

> + if (cache->c_entry_count != 0)

cache->c_entry_count > 0 would be more logical...

> + cache->c_entry_count--;

OK, but please also add:
else
WARN_ONCE(1, "mbcache: Entry count "
"going negative!");

Also as I said in another email I'd be actually more interested in
debugging how can entry count go to such huge value rather than trying to
paper over it...


Honza
--
Jan Kara <[email protected]>
SUSE Labs, CR