[RFC/PATCH 3/3] SLOB: make ksize() more strict for page allocator pass-through

[RFC/PATCH 3/3] SLOB: make ksize() more strict for page allocator pass-through

[Pekka J Enberg]

From: Pekka Enberg <penberg@cs.helsinki.fi>

This patch re-uses the PG_slab flag for marking SLOB bigblock pages so we can
check that the pointer passed ksize() was really allocated by SLOB.

Cc: Christoph Lameter <clameter@sgi.com>
Cc: Matt Mackall <mpm@selenic.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: David Howells <dhowells@redhat.com>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
---
 mm/slob.c |   34 ++++++++++++++++++++++++++++++----
 1 files changed, 30 insertions(+), 4 deletions(-)

diff --git a/mm/slob.c b/mm/slob.c
index a3ad667..4dc077e 100644
--- a/mm/slob.c
+++ b/mm/slob.c
@@ -144,6 +144,24 @@ static inline void clear_slob_page(struct slob_page *sp)
 }
 
 /*
+ * slob_bigpage: True for slob bigblock pages
+ */
+static inline int slob_bigpage(struct page *page)
+{
+	return test_bit(PG_slab, &page->flags);
+}
+
+static inline void set_slob_bigpage(struct page *page)
+{
+	__set_bit(PG_slab, &page->flags);
+}
+
+static inline void clear_slob_bigpage(struct page *page)
+{
+	__clear_bit(PG_slab, &page->flags);
+}
+
+/*
  * slob_page_free: true for pages on free_slob_pages list.
  */
 static inline int slob_page_free(struct slob_page *sp)
@@ -481,6 +499,7 @@ void *__kmalloc_node(size_t size, gfp_t gfp, int node)
 			struct page *page;
 			page = virt_to_page(ret);
 			page->private = size;
+			set_slob_bigpage(page);
 		}
 		return ret;
 	}
@@ -499,8 +518,12 @@ void kfree(const void *block)
 		int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
 		unsigned int *m = (unsigned int *)(block - align);
 		slob_free(m, *m + align);
-	} else
-		put_page(&sp->page);
+	} else {
+		struct page *page = &sp->page;
+
+		clear_slob_bigpage(page);
+		put_page(page);
+	}
 }
 EXPORT_SYMBOL(kfree);
 
@@ -508,6 +531,7 @@ EXPORT_SYMBOL(kfree);
 size_t ksize(const void *block)
 {
 	struct slob_page *sp;
+	struct page *page;
 
 	BUG_ON(!block);
 	if (unlikely(block == ZERO_SIZE_PTR))
@@ -516,8 +540,10 @@ size_t ksize(const void *block)
 	sp = (struct slob_page *)virt_to_page(block);
 	if (slob_page(sp))
 		return ((slob_t *)block - 1)->units + SLOB_UNIT;
-	else
-		return sp->page.private;
+
+	page = &sp->page;
+	BUG_ON(!slob_bigpage(page));
+	return page->private;
 }
 EXPORT_SYMBOL(ksize);
 
-- 
1.5.2.5

Re: [RFC/PATCH 3/3] SLOB: make ksize() more strict for page allocator pass-through

[Matt Mackall]

On Wed, 2008-05-21 at 21:26 +0300, Pekka J Enberg wrote:
> From: Pekka Enberg <penberg@cs.helsinki.fi>
> 
> This patch re-uses the PG_slab flag for marking SLOB bigblock pages so we can
> check that the pointer passed ksize() was really allocated by SLOB.

Ok, first, BUG_ON is wrong.

We already know users (ie nommu memory usage reporting) are abusing this
interface. Making them blow up is unnecessary. Just give them wrong
answers. Issuing a warning might be a good idea for debugging.

Second, I'd prefer to keep this sort of debugging code out of SLOB.
SLUB is much better suited to debugging allocator abuse.

-- 
Mathematics is the supreme nostalgia of our time.

Re: [RFC/PATCH 3/3] SLOB: make ksize() more strict for page allocator pass-through

[Pekka Enberg]

Matt Mackall wrote:
> We already know users (ie nommu memory usage reporting) are abusing this
> interface. Making them blow up is unnecessary. Just give them wrong
> answers. Issuing a warning might be a good idea for debugging.

But kobjsize() never calls ksize() with SLOB as it doesn't set PageSlab. 
Or are you talking about something else here?

Matt Mackall wrote:
> Second, I'd prefer to keep this sort of debugging code out of SLOB.
> SLUB is much better suited to debugging allocator abuse.

Fair enough. I'll drop the patch.

Re: [RFC/PATCH 3/3] SLOB: make ksize() more strict for page allocator pass-through

[Christoph Lameter]

On Wed, 21 May 2008, Pekka Enberg wrote:

> Matt Mackall wrote:
> > We already know users (ie nommu memory usage reporting) are abusing this
> > interface. Making them blow up is unnecessary. Just give them wrong
> > answers. Issuing a warning might be a good idea for debugging.
> 
> But kobjsize() never calls ksize() with SLOB as it doesn't set PageSlab. Or
> are you talking about something else here?
> 
> Matt Mackall wrote:
> > Second, I'd prefer to keep this sort of debugging code out of SLOB.
> > SLUB is much better suited to debugging allocator abuse.
> 
> Fair enough. I'll drop the patch.

I also would prefer to keep this out of SLUB. Set/Clear PageSlab dirties 
the cacheline and it forces the unlining of kmalloc_large.

Re: [RFC/PATCH 3/3] SLOB: make ksize() more strict for page allocator pass-through

[Pekka Enberg]

Christoph Lameter wrote:
> I also would prefer to keep this out of SLUB. Set/Clear PageSlab dirties 
> the cacheline and it forces the unlining of kmalloc_large.

Hmm, well, that leaves ksize() open for more abuse. I guess we'll just 
have to live with that then.

Re: [RFC/PATCH 3/3] SLOB: make ksize() more strict for page allocator pass-through

[Christoph Lameter]

On Thu, 22 May 2008, Pekka Enberg wrote:

> Christoph Lameter wrote:
> > I also would prefer to keep this out of SLUB. Set/Clear PageSlab dirties the
> > cacheline and it forces the unlining of kmalloc_large.
> 
> Hmm, well, that leaves ksize() open for more abuse. I guess we'll just have to
> live with that then.

SLUB could add a BUG_ON(!PageCompound ) since we added buffering for order 
0 allocs. So only order 1 and higher go through kmalloc large. That should 
catch the general case.