mm: memcontrol: switch to rcu protection in drain_all_stock()
Subsystems (2)
PATCH v2 1/1 mm: memcontrol: switch to rcu protection in drain_all_stock()
Commit 72f0184c8a00 ("mm, memcg: remove hotplug locking from try_charge") introduced css_tryget()/css_put() calls in drain_all_stock(), which are supposed to protect the target memory cgroup from being released during the mem_cgroup_is_descendant() call.
However, it's not completely safe. In theory, memcg can go away between reading stock->cached pointer and calling css_tryget().
This can happen if drain_all_stock() races with drain_local_stock() performed on the remote cpu as a result of a work, scheduled by the previous invocation of drain_all_stock().
The race is a bit theoretical and there are few chances to trigger it, but the current code looks a bit confusing, so it makes sense to fix it anyway. The code looks like as if css_tryget() and css_put() are used to protect stocks drainage. It's not necessary because stocked pages are holding references to the cached cgroup. And it obviously won't work for works, scheduled on other cpus.
So, let's read the stock->cached pointer and evaluate the memory cgroup inside a rcu read section, and get rid of css_tryget()/css_put() calls.
v2: added some explanations to the commit message, no code changes
--- mm/memcontrol.c | 17 +++++++++-------- 1 file changed, 9 insertions(+), 8 deletions(-)
mm/memcontrol.c +9 -8
@@ -2235,21 +2235,22 @@ static void drain_all_stock(struct mem_cgroup *root_memcg)for_each_online_cpu (cpu) {struct memcg_stock_pcp *stock = &per_cpu (memcg_stock, cpu);struct mem_cgroup *memcg; +bool flush = false; +rcu_read_lock (); memcg = stock->cached; -if (!memcg || !stock->nr_pages || !css_tryget (&memcg->css)) -continue ; -if (!mem_cgroup_is_descendant (memcg, root_memcg)) { -css_put (&memcg->css); -continue ; - } -if (!test_and_set_bit (FLUSHING_CACHED_CHARGE , &stock->flags)) { +if (memcg && stock->nr_pages && +mem_cgroup_is_descendant (memcg, root_memcg)) + flush = true; +rcu_read_unlock (); + +if (flush && + !test_and_set_bit (FLUSHING_CACHED_CHARGE , &stock->flags)) {if (cpu == curcpu)drain_local_stock (&stock->work);else schedule_work_on (cpu, &stock->work); } -css_put (&memcg->css); }put_cpu ();mutex_unlock (&percpu_charge_mutex);
Signature · 1 line
2.21.0
Quoted 11 lines · Roman Gushchin
Commit
72f0184c8a00("mm, memcg: remove hotplug locking fromtry_charge") introduced css_tryget()/css_put() calls indrain_all_stock(), which are supposed to protect the target memory cgroup from being released during themem_cgroup_is_descendant()call.However, it's not completely safe. In theory, memcg can go away between reading
stock->cachedpointer and callingcss_tryget().This can happen if
drain_all_stock()races withdrain_local_stock()performed on the remote cpu as a result of a work, scheduled by the previous invocation ofdrain_all_stock().
Maybe I am still missing something but I do not see how 72f0184c8a00 changed the existing race. get_online_cpus doesn't prevent the same race right? If this is the case then it would be great to clarify that. I know that you are mostly after clarifying that css_tryget is insufficient but the above sounds like 72f0184c8a00 has introduced a regression.
Quoted 11 lines
The race is a bit theoretical and there are few chances to trigger it, but the current code looks a bit confusing, so it makes sense to fix it anyway. The code looks like as if
css_tryget()andcss_put()are used to protect stocks drainage. It's not necessary because stocked pages are holding references to the cached cgroup. And it obviously won't work for works, scheduled on other cpus.So, let's read the
stock->cachedpointer and evaluate the memory cgroup inside a rcu read section, and get rid of css_tryget()/css_put() calls.
Quoted 5 lines
v2: added some explanations to the commit message, no code changes
Signed-off-by Roman Gushchin <[email protected]>Cc Michal Hocko <[email protected]>Cc Hillf Danton <[email protected]>
Other than that.
Quoted 41 lines
--- mm/memcontrol.c | 17 +++++++++-------- 1 file changed, 9 insertions(+), 8 deletions(-)mm/memcontrol.c +9
-8@@ -2235,21 +2235,22 @@ static void drain_all_stock(struct mem_cgroup *root_memcg)for_each_online_cpu (cpu) {struct memcg_stock_pcp *stock = &per_cpu (memcg_stock, cpu);struct mem_cgroup *memcg; +bool flush = false; +rcu_read_lock (); memcg = stock->cached; -if (!memcg || !stock->nr_pages || !css_tryget (&memcg->css)) -continue ; -if (!mem_cgroup_is_descendant (memcg, root_memcg)) { -css_put (&memcg->css); -continue ; - } -if (!test_and_set_bit (FLUSHING_CACHED_CHARGE , &stock->flags)) { +if (memcg && stock->nr_pages && +mem_cgroup_is_descendant (memcg, root_memcg)) + flush = true; +rcu_read_unlock (); + +if (flush && + !test_and_set_bit (FLUSHING_CACHED_CHARGE , &stock->flags)) {if (cpu == curcpu)drain_local_stock (&stock->work);else schedule_work_on (cpu, &stock->work); } -css_put (&memcg->css); }put_cpu ();mutex_unlock (&percpu_charge_mutex);Signature · 1 line
2.21.0
Signature · 2 lines
Michal Hocko SUSE Labs
Quoted 13 lines · Michal Hocko
Quoted 11 lines · Roman Gushchin
Commit
72f0184c8a00("mm, memcg: remove hotplug locking fromtry_charge") introduced css_tryget()/css_put() calls indrain_all_stock(), which are supposed to protect the target memory cgroup from being released during themem_cgroup_is_descendant()call.However, it's not completely safe. In theory, memcg can go away between reading
stock->cachedpointer and callingcss_tryget().This can happen if
drain_all_stock()races withdrain_local_stock()performed on the remote cpu as a result of a work, scheduled by the previous invocation ofdrain_all_stock().
Quoted 6 lines
Maybe I am still missing something but I do not see how
72f0184c8a00changed the existing race.get_online_cpusdoesn't prevent the same race right? If this is the case then it would be great to clarify that. I know that you are mostly after clarifying thatcss_trygetis insufficient but the above sounds like72f0184c8a00has introduced a regression.
Yeah, I'm not blaming 72f0184c8a00 for the race, which as I said, is barely reproducible at all. There is no "Fixes" tag, and I don't think we need to backport it to stable.
Let's think about this patch as a refactoring patch, which makes the code cleaner.
Quoted 12 lines
Quoted 12 lines
The race is a bit theoretical and there are few chances to trigger it, but the current code looks a bit confusing, so it makes sense to fix it anyway. The code looks like as if
css_tryget()andcss_put()are used to protect stocks drainage. It's not necessary because stocked pages are holding references to the cached cgroup. And it obviously won't work for works, scheduled on other cpus.So, let's read the
stock->cachedpointer and evaluate the memory cgroup inside a rcu read section, and get rid of css_tryget()/css_put() calls.
Quoted 8 lines
Quoted 5 lines
v2: added some explanations to the commit message, no code changes
Signed-off-by Roman Gushchin <[email protected]>Cc Michal Hocko <[email protected]>Cc Hillf Danton <[email protected]>Other than that.
Acked-by Michal Hocko <[email protected]>
Thanks!