[PATCH] z3fold: limit use of stale list for allocation

[PATCH] z3fold: limit use of stale list for allocation

[Vitaly Wool]

Currently if z3fold couldn't find an unbuddied page it would first
try to pull a page off the stale list. The problem with this
approach is that we can't 100% guarantee that the page is not
processed by the workqueue thread at the same time unless we run
cancel_work_sync() on it, which we can't do if we're in an atomic
context. So let's just limit stale list usage to non-atomic
contexts only.

Signed-off-by: Vitaly Vul <vitaly.vul@sony.com>
---
  mm/z3fold.c | 35 +++++++++++++++++++----------------
  1 file changed, 19 insertions(+), 16 deletions(-)

diff --git a/mm/z3fold.c b/mm/z3fold.c
index 39e1912..9b0d112 100644
--- a/mm/z3fold.c
+++ b/mm/z3fold.c
@@ -620,24 +620,27 @@ static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp,
  		bud = FIRST;
  	}
  
-	spin_lock(&pool->stale_lock);
-	zhdr = list_first_entry_or_null(&pool->stale,
-					struct z3fold_header, buddy);
-	/*
-	 * Before allocating a page, let's see if we can take one from the
-	 * stale pages list. cancel_work_sync() can sleep so we must make
-	 * sure it won't be called in case we're in atomic context.
-	 */
-	if (zhdr && (can_sleep || !work_pending(&zhdr->work))) {
-		list_del(&zhdr->buddy);
-		spin_unlock(&pool->stale_lock);
-		if (can_sleep)
+	page = NULL;
+	if (can_sleep) {
+		spin_lock(&pool->stale_lock);
+		zhdr = list_first_entry_or_null(&pool->stale,
+						struct z3fold_header, buddy);
+		/*
+		 * Before allocating a page, let's see if we can take one from
+		 * the stale pages list. cancel_work_sync() can sleep so we
+		 * limit this case to the contexts where we can sleep
+		 */
+		if (zhdr) {
+			list_del(&zhdr->buddy);
+			spin_unlock(&pool->stale_lock);
  			cancel_work_sync(&zhdr->work);
-		page = virt_to_page(zhdr);
-	} else {
-		spin_unlock(&pool->stale_lock);
-		page = alloc_page(gfp);
+			page = virt_to_page(zhdr);
+		} else {
+			spin_unlock(&pool->stale_lock);
+		}
  	}
+	if (!page)
+		page = alloc_page(gfp);
  
  	if (!page)
  		return -ENOMEM;
-- 
2.7.4


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] z3fold: limit use of stale list for allocation

[Andrew Morton]

On Sat, 10 Feb 2018 12:02:52 +0100 Vitaly Wool <vitalywool@gmail.com> wrote:

> Currently if z3fold couldn't find an unbuddied page it would first
> try to pull a page off the stale list. The problem with this
> approach is that we can't 100% guarantee that the page is not
> processed by the workqueue thread at the same time unless we run
> cancel_work_sync() on it, which we can't do if we're in an atomic
> context. So let's just limit stale list usage to non-atomic
> contexts only.

This smells like a bugfix.  What are the end-user visible effects of
the bug?

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] z3fold: limit use of stale list for allocation

[Vitaly Wool]

[-- Attachment #1: Type: text/plain, Size: 892 bytes --]

[sorry for answering only now, this email slipped through somehow]

2018-02-16 16:28 GMT-08:00 Andrew Morton <akpm@linux-foundation.org>:

> On Sat, 10 Feb 2018 12:02:52 +0100 Vitaly Wool <vitalywool@gmail.com>
> wrote:
>
> > Currently if z3fold couldn't find an unbuddied page it would first
> > try to pull a page off the stale list. The problem with this
> > approach is that we can't 100% guarantee that the page is not
> > processed by the workqueue thread at the same time unless we run
> > cancel_work_sync() on it, which we can't do if we're in an atomic
> > context. So let's just limit stale list usage to non-atomic
> > contexts only.
>
> This smells like a bugfix.  What are the end-user visible effects of
> the bug?
>
>
I have only seen this happening in real life once, and then z3fold ended up
using a page which had been already freed and got blocked on a spinlock.

~Vitaly

[-- Attachment #2: Type: text/html, Size: 1440 bytes --]

[PATCH] z3fold: limit use of stale list for allocation

[Vitaly Vul]

Currently if z3fold couldn't find an unbuddied page it would first
try to pull a page off the stale list. The problem with this
approach is that we can't 100% guarantee that the page is not
processed by the workqueue thread at the same time unless we run
cancel_work_sync() on it, which we can't do if we're in an atomic
context. So let's just limit stale list usage to non-atomic
contexts only.

Signed-off-by: Vitaly Vul <vitaly.vul@sony.com>
Signed-off-by: Oleksiy Avramchenko <Oleksiy.Avramchenko@sony.com>
---
 mm/z3fold.c | 35 +++++++++++++++++++----------------
 1 file changed, 19 insertions(+), 16 deletions(-)

diff --git a/mm/z3fold.c b/mm/z3fold.c
index 39e1912..9b0d112 100644
--- a/mm/z3fold.c
+++ b/mm/z3fold.c
@@ -620,24 +620,27 @@ static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp,
 		bud = FIRST;
 	}
 
-	spin_lock(&pool->stale_lock);
-	zhdr = list_first_entry_or_null(&pool->stale,
-					struct z3fold_header, buddy);
-	/*
-	 * Before allocating a page, let's see if we can take one from the
-	 * stale pages list. cancel_work_sync() can sleep so we must make
-	 * sure it won't be called in case we're in atomic context.
-	 */
-	if (zhdr && (can_sleep || !work_pending(&zhdr->work))) {
-		list_del(&zhdr->buddy);
-		spin_unlock(&pool->stale_lock);
-		if (can_sleep)
+	page = NULL;
+	if (can_sleep) {
+		spin_lock(&pool->stale_lock);
+		zhdr = list_first_entry_or_null(&pool->stale,
+						struct z3fold_header, buddy);
+		/*
+		 * Before allocating a page, let's see if we can take one from
+		 * the stale pages list. cancel_work_sync() can sleep so we
+		 * limit this case to the contexts where we can sleep
+		 */
+		if (zhdr) {
+			list_del(&zhdr->buddy);
+			spin_unlock(&pool->stale_lock);
 			cancel_work_sync(&zhdr->work);
-		page = virt_to_page(zhdr);
-	} else {
-		spin_unlock(&pool->stale_lock);
-		page = alloc_page(gfp);
+			page = virt_to_page(zhdr);
+		} else {
+			spin_unlock(&pool->stale_lock);
+		}
 	}
+	if (!page)
+		page = alloc_page(gfp);
 
 	if (!page)
 		return -ENOMEM;
-- 
2.7.4

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>