Re: [BUGFIX][PATCH] memcg: fix memory migration of shmem swapcache

[Daisuke Nishimura]

On Wed, 5 Jan 2011 12:58:40 +0100
Johannes Weiner <hannes@xxxxxxxxxxx> wrote:

> On Wed, Jan 05, 2011 at 01:00:20PM +0900, Daisuke Nishimura wrote:
> > In current implimentation, mem_cgroup_end_migration() decides whether the page
> > migration has succeeded or not by checking "oldpage->mapping".
> > 
> > But if we are tring to migrate a shmem swapcache, the page->mapping of it is
> > NULL from the begining, so the check would be invalid.
> > As a result, mem_cgroup_end_migration() assumes the migration has succeeded
> > even if it's not, so "newpage" would be freed while it's not uncharged.
> > 
> > This patch fixes it by passing mem_cgroup_end_migration() the result of the
> > page migration.
> 
> Are there other users that rely on unused->mapping being NULL after
> migration?
> 
As long as I can see, no.

> If so, aren't they prone to misinterpreting this for shmem swapcache
> as well?
> 
> If not, wouldn't it be better to remove that page->mapping = NULL from
> migrate_page_copy() altogether?  I think it's an ugly exception where
> the outcome of PageAnon() is not meaningful for an LRU page.
> 
IIUC, oldpage will be freed on success of page migration, so we hit bad_page
check at freeing the page unless we clear oldpage->mapping, 

> To your patch:
> 
> > --- a/mm/memcontrol.c
> > +++ b/mm/memcontrol.c
> > @@ -2856,7 +2856,7 @@ int mem_cgroup_prepare_migration(struct page *page,
> >  
> >  /* remove redundant charge if migration failed*/
> >  void mem_cgroup_end_migration(struct mem_cgroup *mem,
> > -	struct page *oldpage, struct page *newpage)
> > +	struct page *oldpage, struct page *newpage, int result)
> >  {
> >  	struct page *used, *unused;
> >  	struct page_cgroup *pc;
> > @@ -2865,8 +2865,7 @@ void mem_cgroup_end_migration(struct mem_cgroup *mem,
> >  		return;
> >  	/* blocks rmdir() */
> >  	cgroup_exclude_rmdir(&mem->css);
> > -	/* at migration success, oldpage->mapping is NULL. */
> > -	if (oldpage->mapping) {
> > +	if (result) {
> 
> Since this function does not really need more than a boolean value,
> wouldn't it make the code more obvious if the parameter was `bool
> success'?
> 
> 	if (!success) {
> >  		used = oldpage;
> >  		unused = newpage;
> >  	} else {
> 
> Minor nit, though.  I agree with the patch in general.
> 
Thank you for your review.
How about this ?

===
From: Daisuke Nishimura <nishimura@xxxxxxxxxxxxxxxxx>

In current implimentation, mem_cgroup_end_migration() decides whether the page
migration has succeeded or not by checking "oldpage->mapping".

But if we are tring to migrate a shmem swapcache, the page->mapping of it is
NULL from the begining, so the check would be invalid.
As a result, mem_cgroup_end_migration() assumes the migration has succeeded
even if it's not, so "newpage" would be freed while it's not uncharged.

This patch fixes it by passing mem_cgroup_end_migration() the result of the
page migration.

Signed-off-by: Daisuke Nishimura <nishimura@xxxxxxxxxxxxxxxxx>
---
 include/linux/memcontrol.h |    5 ++---
 mm/memcontrol.c            |    5 ++---
 mm/migrate.c               |    2 +-
 3 files changed, 5 insertions(+), 7 deletions(-)

diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
index 159a076..cc5a8fd 100644
--- a/include/linux/memcontrol.h
+++ b/include/linux/memcontrol.h
@@ -93,7 +93,7 @@ extern int
 mem_cgroup_prepare_migration(struct page *page,
 	struct page *newpage, struct mem_cgroup **ptr);
 extern void mem_cgroup_end_migration(struct mem_cgroup *mem,
-	struct page *oldpage, struct page *newpage);
+	struct page *oldpage, struct page *newpage, bool success);
 
 /*
  * For memory reclaim.
@@ -231,8 +231,7 @@ mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
 }
 
 static inline void mem_cgroup_end_migration(struct mem_cgroup *mem,
-					struct page *oldpage,
-					struct page *newpage)
+		struct page *oldpage, struct page *newpage, bool success)
 {
 }
 
diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 61678be..fbecd02 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -2856,7 +2856,7 @@ int mem_cgroup_prepare_migration(struct page *page,
 
 /* remove redundant charge if migration failed*/
 void mem_cgroup_end_migration(struct mem_cgroup *mem,
-	struct page *oldpage, struct page *newpage)
+	struct page *oldpage, struct page *newpage, bool success)
 {
 	struct page *used, *unused;
 	struct page_cgroup *pc;
@@ -2865,8 +2865,7 @@ void mem_cgroup_end_migration(struct mem_cgroup *mem,
 		return;
 	/* blocks rmdir() */
 	cgroup_exclude_rmdir(&mem->css);
-	/* at migration success, oldpage->mapping is NULL. */
-	if (oldpage->mapping) {
+	if (!success) {
 		used = oldpage;
 		unused = newpage;
 	} else {
diff --git a/mm/migrate.c b/mm/migrate.c
index 6ae8a66..be66b23 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -756,7 +756,7 @@ rcu_unlock:
 		rcu_read_unlock();
 uncharge:
 	if (!charge)
-		mem_cgroup_end_migration(mem, page, newpage);
+		mem_cgroup_end_migration(mem, page, newpage, rc == 0);
 unlock:
 	unlock_page(page);
 
-- 
1.7.1

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/