I discovered that we can overflow stack if CONFIG_SLUB_DEBUG=y and use
slabs with many objects, since list_slab_objects() and process_slab()
use DECLARE_BITMAP(map, page->objects);

With 65535 bits, we use 8192 bytes of stack ...

A possible fix is to lower MAX_OBJS_PER_PAGE so that these bitmaps dont
use more than a third of THREAD_SIZE. I suspect plain memory allocation
in these functions is not an option.

Using non dynamic stack allocation makes the problem more obvious if
somebody runs checkstack.pl

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
---
diff --git a/mm/slub.c b/mm/slub.c
index b364844..adf04c1 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -167,7 +167,13 @@
 
 #define OO_SHIFT	16
 #define OO_MASK		((1 << OO_SHIFT) - 1)
+
+#ifdef CONFIG_SLUB_DEBUG
+/* We use an onstack BITMAP while debugging, make sure this wont be too big */
+#define MAX_OBJS_PER_PAGE	min_t(int, 65535, 8*(THREAD_SIZE/3))
+#else
 #define MAX_OBJS_PER_PAGE	65535 /* since page.objects is u16 */
+#endif
 
 /* Internal SLUB flags */
 #define __OBJECT_POISON		0x80000000 /* Poison object */
@@ -2426,7 +2432,7 @@ static void list_slab_objects(struct kmem_cache *s, struct page *page,
 #ifdef CONFIG_SLUB_DEBUG
 	void *addr = page_address(page);
 	void *p;
-	DECLARE_BITMAP(map, page->objects);
+	DECLARE_BITMAP(map, MAX_OBJS_PER_PAGE);
 
 	bitmap_zero(map, page->objects);
 	slab_err(s, page, "%s", text);
@@ -3651,7 +3657,7 @@ static void process_slab(struct loc_track *t, struct kmem_cache *s,
 		struct page *page, enum track_item alloc)
 {
 	void *addr = page_address(page);
-	DECLARE_BITMAP(map, page->objects);
+	DECLARE_BITMAP(map, MAX_OBJS_PER_PAGE);
 	void *p;
 
 	bitmap_zero(map, page->objects);


--

[ message continues ]

Maybe we better allocate the bitmap via kmalloc then.

--

[ message continues ]

Hmm...

Are we allowed to nest in these two functions ?

GFP_KERNEL, GFP_ATOMIC ?

These are debugging functions, what happens if kmalloc() returns NULL ?



--

[ message continues ]

Then you return ENOMEM and the user gets an error. We already do that in
validate_slab_cache().

Hmmm... In this case we called from list_slab_objects() which gets called
from free_partial() (which took a spinlock!) which gets called from
kmem_cache_close().

Its just a debugging aid so no problem if it fails. GFP_ATOMIC?


--

[ message continues ]

OK, here is second version of the patch, thanks !


[PATCH] slub: Potential stack overflow

I discovered that we can overflow stack if CONFIG_SLUB_DEBUG=y and use
slabs with many objects, since list_slab_objects() and process_slab()
use DECLARE_BITMAP(map, page->objects);

With 65535 bits, we use 8192 bytes of stack ...

A possible solution is to allocate memory, using GFP_ATOMIC, and do
nothing if allocation fails.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
---
diff --git a/mm/slub.c b/mm/slub.c
index b364844..5ee857a 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -2426,9 +2426,11 @@ static void list_slab_objects(struct kmem_cache *s, struct page *page,
 #ifdef CONFIG_SLUB_DEBUG
 	void *addr = page_address(page);
 	void *p;
-	DECLARE_BITMAP(map, page->objects);
+	long *map = kzalloc(BITS_TO_LONGS(page->objects) * sizeof(long),
+			    GFP_ATOMIC);
 
-	bitmap_zero(map, page->objects);
+	if (!map)
+		return;
 	slab_err(s, page, "%s", text);
 	slab_lock(page);
 	for_each_free_object(p, s, page->freelist)
@@ -2443,6 +2445,7 @@ static void list_slab_objects(struct kmem_cache *s, struct page *page,
 		}
 	}
 	slab_unlock(page);
+	kfree(map);
 #endif
 }
 
@@ -3651,16 +3654,19 @@ static void process_slab(struct loc_track *t, struct kmem_cache *s,
 		struct page *page, enum track_item alloc)
 {
 	void *addr = page_address(page);
-	DECLARE_BITMAP(map, page->objects);
+	long *map = kzalloc(BITS_TO_LONGS(page->objects) * sizeof(long),
+			    GFP_ATOMIC);
 	void *p;
 
-	bitmap_zero(map, page->objects);
+	if (!map)
+		return;
 	for_each_free_object(p, s, page->freelist)
 		set_bit(slab_index(p, s, addr), map);
 
 	for_each_object(p, s, addr, page->objects)
 		if (!test_bit(slab_index(p, s, addr), map))
 			add_location(t, s, get_track(s, p, alloc));
+	kfree(map);
 }
 
 static int list_locations(struct kmem_cache *s, char *buf,


--

[ message continues ]

Hmmm... Thats another case. We should alloate the map higher up there I
guess and pass the address in so that one allocation can be used for all
slabs. validate_slab_cache() does that.

--

[ message continues ]

Here is a patch for the second case. I think its better since it results
in an error display and it avoids the alloc for each slab. Add this piece
to your patch?

Signed-off-by: Christoph Lameter <cl@linux-foundation.org>

---
 mm/slub.c |   12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

Index: linux-2.6/mm/slub.c
===================================================================
--- linux-2.6.orig/mm/slub.c	2010-03-24 16:10:32.000000000 -0500
+++ linux-2.6/mm/slub.c	2010-03-24 16:13:06.000000000 -0500
@@ -3648,10 +3648,10 @@ static int add_location(struct loc_track
 }

 static void process_slab(struct loc_track *t, struct kmem_cache *s,
-		struct page *page, enum track_item alloc)
+		struct page *page, enum track_item alloc,
+		unsigned long *map)
 {
 	void *addr = page_address(page);
-	DECLARE_BITMAP(map, page->objects);
 	void *p;

 	bitmap_zero(map, page->objects);
@@ -3670,8 +3670,10 @@ static int list_locations(struct kmem_ca
 	unsigned long i;
 	struct loc_track t = { 0, 0, NULL };
 	int node;
+	unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
+				sizeof(unsigned long), GFP_KERNEL);

-	if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
+	if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
 			GFP_TEMPORARY))
 		return sprintf(buf, "Out of memory\n");

@@ -3688,9 +3690,9 @@ static int list_locations(struct kmem_ca

 		spin_lock_irqsave(&n->list_lock, flags);
 		list_for_each_entry(page, &n->partial, lru)
-			process_slab(&t, s, page, alloc);
+			process_slab(&t, s, page, alloc, map);
 		list_for_each_entry(page, &n->full, lru)
-			process_slab(&t, s, page, alloc);
+			process_slab(&t, s, page, alloc, map);
 		spin_unlock_irqrestore(&n->list_lock, flags);
 	}

--

[ message continues ]

Here is a patch for the second case. I think its better since it results
in an error display and it avoids the alloc for each slab. Add this piece
to your patch?

V1->V2 Fix missing kfree

Signed-off-by: Christoph Lameter <cl@linux-foundation.org>

---
 mm/slub.c |   13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

Index: linux-2.6/mm/slub.c
===================================================================
--- linux-2.6.orig/mm/slub.c	2010-03-24 16:23:19.000000000 -0500
+++ linux-2.6/mm/slub.c	2010-03-24 16:24:21.000000000 -0500
@@ -3648,10 +3648,10 @@ static int add_location(struct loc_track
 }

 static void process_slab(struct loc_track *t, struct kmem_cache *s,
-		struct page *page, enum track_item alloc)
+		struct page *page, enum track_item alloc,
+		unsigned long *map)
 {
 	void *addr = page_address(page);
-	DECLARE_BITMAP(map, page->objects);
 	void *p;

 	bitmap_zero(map, page->objects);
@@ -3670,8 +3670,10 @@ static int list_locations(struct kmem_ca
 	unsigned long i;
 	struct loc_track t = { 0, 0, NULL };
 	int node;
+	unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
+				sizeof(unsigned long), GFP_KERNEL);

-	if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
+	if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
 			GFP_TEMPORARY))
 		return sprintf(buf, "Out of memory\n");

@@ -3688,11 +3690,12 @@ static int list_locations(struct kmem_ca

 		spin_lock_irqsave(&n->list_lock, flags);
 		list_for_each_entry(page, &n->partial, lru)
-			process_slab(&t, s, page, alloc);
+			process_slab(&t, s, page, alloc, map);
 		list_for_each_entry(page, &n->full, lru)
-			process_slab(&t, s, page, alloc);
+			process_slab(&t, s, page, alloc, map);
 		spin_unlock_irqrestore(&n->list_lock, flags);
 	}
+	kfree(map);

 	for (i = 0; i < t.count; i++) {
 		struct location *l = &t.loc[i];
--

[ message continues ]

--

[ message continues ]

Sure, here is third version :

Thanks

[PATCH] slub: Potential stack overflow

I discovered that we can overflow stack if CONFIG_SLUB_DEBUG=y and use
slabs with many objects, since list_slab_objects() and process_slab()
use DECLARE_BITMAP(map, page->objects);

With 65535 bits, we use 8192 bytes of stack ...

Switch these allocations to dynamic allocations.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Christoph Lameter <cl@linux-foundation.org>
---
 mm/slub.c |   25 ++++++++++++++++---------
 1 file changed, 16 insertions(+), 9 deletions(-)
diff --git a/mm/slub.c b/mm/slub.c
index b364844..7dc8e73 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -2426,9 +2426,11 @@ static void list_slab_objects(struct kmem_cache *s, struct page *page,
 #ifdef CONFIG_SLUB_DEBUG
 	void *addr = page_address(page);
 	void *p;
-	DECLARE_BITMAP(map, page->objects);
+	long *map = kzalloc(BITS_TO_LONGS(page->objects) * sizeof(long),
+			    GFP_ATOMIC);
 
-	bitmap_zero(map, page->objects);
+	if (!map)
+		return;
 	slab_err(s, page, "%s", text);
 	slab_lock(page);
 	for_each_free_object(p, s, page->freelist)
@@ -2443,6 +2445,7 @@ static void list_slab_objects(struct kmem_cache *s, struct page *page,
 		}
 	}
 	slab_unlock(page);
+	kfree(map);
 #endif
 }
 
@@ -3648,10 +3651,10 @@ static int add_location(struct loc_track *t, struct kmem_cache *s,
 }
 
 static void process_slab(struct loc_track *t, struct kmem_cache *s,
-		struct page *page, enum track_item alloc)
+		struct page *page, enum track_item alloc,
+		long *map)
 {
 	void *addr = page_address(page);
-	DECLARE_BITMAP(map, page->objects);
 	void *p;
 
 	bitmap_zero(map, page->objects);
@@ -3670,11 +3673,14 @@ static int list_locations(struct kmem_cache *s, char *buf,
 	unsigned long i;
 	struct loc_track t = { 0, 0, NULL };
 	int node;
+	unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
+				     sizeof(unsigned long), GFP_KERNEL);
 
-	if (!alloc_loc_track(&t, PAGE_SIZE / ...
[ message continues ]

Christoph, does this look OK to you? I think Eric has all your later 

--

[ message continues ]

Yes just dont know how to add an Ack given that there is already a
signoff.
--

[ message continues ]

Applied.
--

[ message continues ]