The kasan quarantine is designed to delay freeing slab objects to catch
use-after-free. The quarantine can be large (several percent of machine
memory size). When kmem_caches are deleted related objects are flushed
from the quarantine but this requires scanning the entire quarantine
which can be very slow. We have seen the kernel busily working on this
while holding slab_mutex and badly affecting cache_reaper, slabinfo
readers and memcg kmem cache creations.
It can easily reproduced by following script:
yes . | head -1000000 | xargs stat > /dev/null
for i in `seq 1 10`; do
seq 500 | (cd /cg/memory && xargs mkdir)
seq 500 | xargs -I{} sh -c 'echo $BASHPID > \
/cg/memory/{}/tasks && exec stat .' > /dev/null
seq 500 | (cd /cg/memory && xargs rmdir)
done
The busy stack:
kasan_cache_shutdown
shutdown_cache
memcg_destroy_kmem_caches
mem_cgroup_css_free
css_free_rwork_fn
process_one_work
worker_thread
kthread
ret_from_fork
This patch is based on the observation that if the kmem_cache to be
destroyed is empty then there should not be any objects of this cache in
the quarantine.
Without the patch the script got stuck for couple of hours. With the
patch the script completed within a second.
Signed-off-by: Shakeel Butt <[email protected]>
---
mm/kasan/kasan.c | 3 ++-
mm/slab.c | 12 ++++++++++++
mm/slab.h | 1 +
mm/slub.c | 11 +++++++++++
4 files changed, 26 insertions(+), 1 deletion(-)
diff --git a/mm/kasan/kasan.c b/mm/kasan/kasan.c
index 49fffb0ca83b..135ce2838c89 100644
--- a/mm/kasan/kasan.c
+++ b/mm/kasan/kasan.c
@@ -382,7 +382,8 @@ void kasan_cache_shrink(struct kmem_cache *cache)
void kasan_cache_shutdown(struct kmem_cache *cache)
{
- quarantine_remove_cache(cache);
+ if (!__kmem_cache_empty(cache))
+ quarantine_remove_cache(cache);
}
size_t kasan_metadata_size(struct kmem_cache *cache)
diff --git a/mm/slab.c b/mm/slab.c
index 9212c64bb705..b59f2cdf28d1 100644
--- a/mm/slab.c
+++ b/mm/slab.c
@@ -2291,6 +2291,18 @@ static int drain_freelist(struct kmem_cache *cache,
return nr_freed;
}
+bool __kmem_cache_empty(struct kmem_cache *s)
+{
+ int node;
+ struct kmem_cache_node *n;
+
+ for_each_kmem_cache_node(s, node, n)
+ if (!list_empty(&n->slabs_full) ||
+ !list_empty(&n->slabs_partial))
+ return false;
+ return true;
+}
+
int __kmem_cache_shrink(struct kmem_cache *cachep)
{
int ret = 0;
diff --git a/mm/slab.h b/mm/slab.h
index e8981e811c45..68bdf498da3b 100644
--- a/mm/slab.h
+++ b/mm/slab.h
@@ -166,6 +166,7 @@ static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
SLAB_TEMPORARY | \
SLAB_ACCOUNT)
+bool __kmem_cache_empty(struct kmem_cache *);
int __kmem_cache_shutdown(struct kmem_cache *);
void __kmem_cache_release(struct kmem_cache *);
int __kmem_cache_shrink(struct kmem_cache *);
diff --git a/mm/slub.c b/mm/slub.c
index 1edc8d97c862..44aa7847324a 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -3707,6 +3707,17 @@ static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
discard_slab(s, page);
}
+bool __kmem_cache_empty(struct kmem_cache *s)
+{
+ int node;
+ struct kmem_cache_node *n;
+
+ for_each_kmem_cache_node(s, node, n)
+ if (n->nr_partial || slabs_node(s, node))
+ return false;
+ return true;
+}
+
/*
* Release all resources used by a slab cache.
*/
--
2.17.0.rc1.321.gba9d0f2565-goog
On Tue, 27 Mar 2018, Shakeel Butt wrote:
> diff --git a/mm/kasan/kasan.c b/mm/kasan/kasan.c
> index 49fffb0ca83b..135ce2838c89 100644
> --- a/mm/kasan/kasan.c
> +++ b/mm/kasan/kasan.c
> @@ -382,7 +382,8 @@ void kasan_cache_shrink(struct kmem_cache *cache)
>
> void kasan_cache_shutdown(struct kmem_cache *cache)
> {
> - quarantine_remove_cache(cache);
> + if (!__kmem_cache_empty(cache))
> + quarantine_remove_cache(cache);
> }
>
> size_t kasan_metadata_size(struct kmem_cache *cache)
> diff --git a/mm/slab.c b/mm/slab.c
> index 9212c64bb705..b59f2cdf28d1 100644
> --- a/mm/slab.c
> +++ b/mm/slab.c
> @@ -2291,6 +2291,18 @@ static int drain_freelist(struct kmem_cache *cache,
> return nr_freed;
> }
>
> +bool __kmem_cache_empty(struct kmem_cache *s)
> +{
> + int node;
> + struct kmem_cache_node *n;
> +
> + for_each_kmem_cache_node(s, node, n)
> + if (!list_empty(&n->slabs_full) ||
> + !list_empty(&n->slabs_partial))
> + return false;
> + return true;
> +}
> +
> int __kmem_cache_shrink(struct kmem_cache *cachep)
> {
> int ret = 0;
> diff --git a/mm/slab.h b/mm/slab.h
> index e8981e811c45..68bdf498da3b 100644
> --- a/mm/slab.h
> +++ b/mm/slab.h
> @@ -166,6 +166,7 @@ static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
> SLAB_TEMPORARY | \
> SLAB_ACCOUNT)
>
> +bool __kmem_cache_empty(struct kmem_cache *);
> int __kmem_cache_shutdown(struct kmem_cache *);
> void __kmem_cache_release(struct kmem_cache *);
> int __kmem_cache_shrink(struct kmem_cache *);
> diff --git a/mm/slub.c b/mm/slub.c
> index 1edc8d97c862..44aa7847324a 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -3707,6 +3707,17 @@ static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
> discard_slab(s, page);
> }
>
> +bool __kmem_cache_empty(struct kmem_cache *s)
> +{
> + int node;
> + struct kmem_cache_node *n;
> +
> + for_each_kmem_cache_node(s, node, n)
> + if (n->nr_partial || slabs_node(s, node))
> + return false;
> + return true;
> +}
> +
> /*
> * Release all resources used by a slab cache.
> */
Any reason not to just make quarantine_remove_cache() part of
__kmem_cache_shutdown() instead of duplicating its logic?
On Tue, Mar 27, 2018 at 5:16 PM, David Rientjes <[email protected]> wrote:
> On Tue, 27 Mar 2018, Shakeel Butt wrote:
>
>> diff --git a/mm/kasan/kasan.c b/mm/kasan/kasan.c
>> index 49fffb0ca83b..135ce2838c89 100644
>> --- a/mm/kasan/kasan.c
>> +++ b/mm/kasan/kasan.c
>> @@ -382,7 +382,8 @@ void kasan_cache_shrink(struct kmem_cache *cache)
>>
>> void kasan_cache_shutdown(struct kmem_cache *cache)
>> {
>> - quarantine_remove_cache(cache);
>> + if (!__kmem_cache_empty(cache))
>> + quarantine_remove_cache(cache);
>> }
>>
>> size_t kasan_metadata_size(struct kmem_cache *cache)
>> diff --git a/mm/slab.c b/mm/slab.c
>> index 9212c64bb705..b59f2cdf28d1 100644
>> --- a/mm/slab.c
>> +++ b/mm/slab.c
>> @@ -2291,6 +2291,18 @@ static int drain_freelist(struct kmem_cache *cache,
>> return nr_freed;
>> }
>>
>> +bool __kmem_cache_empty(struct kmem_cache *s)
>> +{
>> + int node;
>> + struct kmem_cache_node *n;
>> +
>> + for_each_kmem_cache_node(s, node, n)
>> + if (!list_empty(&n->slabs_full) ||
>> + !list_empty(&n->slabs_partial))
>> + return false;
>> + return true;
>> +}
>> +
>> int __kmem_cache_shrink(struct kmem_cache *cachep)
>> {
>> int ret = 0;
>> diff --git a/mm/slab.h b/mm/slab.h
>> index e8981e811c45..68bdf498da3b 100644
>> --- a/mm/slab.h
>> +++ b/mm/slab.h
>> @@ -166,6 +166,7 @@ static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
>> SLAB_TEMPORARY | \
>> SLAB_ACCOUNT)
>>
>> +bool __kmem_cache_empty(struct kmem_cache *);
>> int __kmem_cache_shutdown(struct kmem_cache *);
>> void __kmem_cache_release(struct kmem_cache *);
>> int __kmem_cache_shrink(struct kmem_cache *);
>> diff --git a/mm/slub.c b/mm/slub.c
>> index 1edc8d97c862..44aa7847324a 100644
>> --- a/mm/slub.c
>> +++ b/mm/slub.c
>> @@ -3707,6 +3707,17 @@ static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
>> discard_slab(s, page);
>> }
>>
>> +bool __kmem_cache_empty(struct kmem_cache *s)
>> +{
>> + int node;
>> + struct kmem_cache_node *n;
>> +
>> + for_each_kmem_cache_node(s, node, n)
>> + if (n->nr_partial || slabs_node(s, node))
>> + return false;
>> + return true;
>> +}
>> +
>> /*
>> * Release all resources used by a slab cache.
>> */
>
> Any reason not to just make quarantine_remove_cache() part of
> __kmem_cache_shutdown() instead of duplicating its logic?
Can you please explain what you mean by making
quarantine_remove_cache() part of __kmem_cache_shutdown()? Do you mean
calling quarantine_remove_cache() inside __kmem_cache_shutdown()? The
__kmem_cache_shutdown() of both SLAB & SLUB does per-cpu
draining/flushing and we want the free the quarantined objects before
that. So, I am not sure how to incorporate quarantine_remove_cache()
into __kmem_cache_shutdown() without duplicating the for-loop &
if-condition.
On 03/28/2018 02:06 AM, Shakeel Butt wrote:
> The kasan quarantine is designed to delay freeing slab objects to catch
> use-after-free. The quarantine can be large (several percent of machine
> memory size). When kmem_caches are deleted related objects are flushed
> from the quarantine but this requires scanning the entire quarantine
> which can be very slow. We have seen the kernel busily working on this
> while holding slab_mutex and badly affecting cache_reaper, slabinfo
> readers and memcg kmem cache creations.
>
> It can easily reproduced by following script:
>
> yes . | head -1000000 | xargs stat > /dev/null
> for i in `seq 1 10`; do
> seq 500 | (cd /cg/memory && xargs mkdir)
> seq 500 | xargs -I{} sh -c 'echo $BASHPID > \
> /cg/memory/{}/tasks && exec stat .' > /dev/null
> seq 500 | (cd /cg/memory && xargs rmdir)
> done
>
> The busy stack:
> kasan_cache_shutdown
> shutdown_cache
> memcg_destroy_kmem_caches
> mem_cgroup_css_free
> css_free_rwork_fn
> process_one_work
> worker_thread
> kthread
> ret_from_fork
>
> This patch is based on the observation that if the kmem_cache to be
> destroyed is empty then there should not be any objects of this cache in
> the quarantine.
>
> Without the patch the script got stuck for couple of hours. With the
> patch the script completed within a second.
>
> Signed-off-by: Shakeel Butt <[email protected]>
>
Acked-by: Andrey Ryabinin <[email protected]>
On Tue, 27 Mar 2018, Shakeel Butt wrote:
> The kasan quarantine is designed to delay freeing slab objects to catch
> use-after-free. The quarantine can be large (several percent of machine
> memory size). When kmem_caches are deleted related objects are flushed
> from the quarantine but this requires scanning the entire quarantine
> which can be very slow. We have seen the kernel busily working on this
> while holding slab_mutex and badly affecting cache_reaper, slabinfo
> readers and memcg kmem cache creations.
Acked-by: Christoph Lameter <[email protected]>
On Tue, 27 Mar 2018, Shakeel Butt wrote:
> On Tue, Mar 27, 2018 at 5:16 PM, David Rientjes <[email protected]> wrote:
> > On Tue, 27 Mar 2018, Shakeel Butt wrote:
> >
> >> diff --git a/mm/kasan/kasan.c b/mm/kasan/kasan.c
> >> index 49fffb0ca83b..135ce2838c89 100644
> >> --- a/mm/kasan/kasan.c
> >> +++ b/mm/kasan/kasan.c
> >> @@ -382,7 +382,8 @@ void kasan_cache_shrink(struct kmem_cache *cache)
> >>
> >> void kasan_cache_shutdown(struct kmem_cache *cache)
> >> {
> >> - quarantine_remove_cache(cache);
> >> + if (!__kmem_cache_empty(cache))
> >> + quarantine_remove_cache(cache);
> >> }
> >>
> >> size_t kasan_metadata_size(struct kmem_cache *cache)
> >> diff --git a/mm/slab.c b/mm/slab.c
> >> index 9212c64bb705..b59f2cdf28d1 100644
> >> --- a/mm/slab.c
> >> +++ b/mm/slab.c
> >> @@ -2291,6 +2291,18 @@ static int drain_freelist(struct kmem_cache *cache,
> >> return nr_freed;
> >> }
> >>
> >> +bool __kmem_cache_empty(struct kmem_cache *s)
> >> +{
> >> + int node;
> >> + struct kmem_cache_node *n;
> >> +
> >> + for_each_kmem_cache_node(s, node, n)
> >> + if (!list_empty(&n->slabs_full) ||
> >> + !list_empty(&n->slabs_partial))
> >> + return false;
> >> + return true;
> >> +}
> >> +
> >> int __kmem_cache_shrink(struct kmem_cache *cachep)
> >> {
> >> int ret = 0;
> >> diff --git a/mm/slab.h b/mm/slab.h
> >> index e8981e811c45..68bdf498da3b 100644
> >> --- a/mm/slab.h
> >> +++ b/mm/slab.h
> >> @@ -166,6 +166,7 @@ static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
> >> SLAB_TEMPORARY | \
> >> SLAB_ACCOUNT)
> >>
> >> +bool __kmem_cache_empty(struct kmem_cache *);
> >> int __kmem_cache_shutdown(struct kmem_cache *);
> >> void __kmem_cache_release(struct kmem_cache *);
> >> int __kmem_cache_shrink(struct kmem_cache *);
> >> diff --git a/mm/slub.c b/mm/slub.c
> >> index 1edc8d97c862..44aa7847324a 100644
> >> --- a/mm/slub.c
> >> +++ b/mm/slub.c
> >> @@ -3707,6 +3707,17 @@ static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
> >> discard_slab(s, page);
> >> }
> >>
> >> +bool __kmem_cache_empty(struct kmem_cache *s)
> >> +{
> >> + int node;
> >> + struct kmem_cache_node *n;
> >> +
> >> + for_each_kmem_cache_node(s, node, n)
> >> + if (n->nr_partial || slabs_node(s, node))
> >> + return false;
> >> + return true;
> >> +}
> >> +
> >> /*
> >> * Release all resources used by a slab cache.
> >> */
> >
> > Any reason not to just make quarantine_remove_cache() part of
> > __kmem_cache_shutdown() instead of duplicating its logic?
>
> Can you please explain what you mean by making
> quarantine_remove_cache() part of __kmem_cache_shutdown()? Do you mean
> calling quarantine_remove_cache() inside __kmem_cache_shutdown()? The
> __kmem_cache_shutdown() of both SLAB & SLUB does per-cpu
> draining/flushing and we want the free the quarantined objects before
> that. So, I am not sure how to incorporate quarantine_remove_cache()
> into __kmem_cache_shutdown() without duplicating the for-loop &
> if-condition.
>
__kmem_cache_empty() is largely a copy and paste of
__kmem_cache_shutdown() logic, is there no way to simplify this? I was
thinking of generalizing the iteration (for_each_kmem_cach_node_nonempty?)
that eliminates the need for __kmem_cache_empty().
kasan_cache_shutdown() would do
for_each_kmem_cache_node_nonempty(cache, node, n) {
quarantine_remove_cache(cache);
break;
}
and __kmem_cache_shutdown() would use it for the iteration over
kmem_cache_node's to drain.