The system will immediate fill up stack and crash when both
CONFIG_DEBUG_KMEMLEAK and CONFIG_MEM_ALLOC_PROFILING are enabled.
Avoid allocation tagging of kmemleak caches, otherwise recursive
allocation tracking occurs.
Fixes: 279bb991b4d9 ("mm/slab: add allocation accounting into slab allocation and free paths")
Signed-off-by: Kees Cook <[email protected]>
---
Cc: Suren Baghdasaryan <[email protected]>
Cc: Kent Overstreet <[email protected]>
Cc: Catalin Marinas <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Christoph Lameter <[email protected]>
Cc: Pekka Enberg <[email protected]>
Cc: David Rientjes <[email protected]>
Cc: Joonsoo Kim <[email protected]>
Cc: Vlastimil Babka <[email protected]>
Cc: Roman Gushchin <[email protected]>
Cc: Hyeonggon Yoo <[email protected]>
Cc: [email protected]
---
mm/kmemleak.c | 4 ++--
mm/slub.c | 2 +-
2 files changed, 3 insertions(+), 3 deletions(-)
diff --git a/mm/kmemleak.c b/mm/kmemleak.c
index c55c2cbb6837..fdcf01f62202 100644
--- a/mm/kmemleak.c
+++ b/mm/kmemleak.c
@@ -463,7 +463,7 @@ static struct kmemleak_object *mem_pool_alloc(gfp_t gfp)
/* try the slab allocator first */
if (object_cache) {
- object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
+ object = kmem_cache_alloc_noprof(object_cache, gfp_kmemleak_mask(gfp));
if (object)
return object;
}
@@ -947,7 +947,7 @@ static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
untagged_objp = (unsigned long)kasan_reset_tag((void *)object->pointer);
if (scan_area_cache)
- area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp));
+ area = kmem_cache_alloc_noprof(scan_area_cache, gfp_kmemleak_mask(gfp));
raw_spin_lock_irqsave(&object->lock, flags);
if (!area) {
diff --git a/mm/slub.c b/mm/slub.c
index a94a0507e19c..9ae032ed17ed 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -2016,7 +2016,7 @@ prepare_slab_obj_exts_hook(struct kmem_cache *s, gfp_t flags, void *p)
if (!p)
return NULL;
- if (s->flags & SLAB_NO_OBJ_EXT)
+ if (s->flags & (SLAB_NO_OBJ_EXT | SLAB_NOLEAKTRACE))
return NULL;
if (flags & __GFP_NO_OBJ_EXT)
--
2.34.1
On Thu, Apr 25, 2024 at 01:55:23PM -0700, Kees Cook wrote:
> The system will immediate fill up stack and crash when both
> CONFIG_DEBUG_KMEMLEAK and CONFIG_MEM_ALLOC_PROFILING are enabled.
> Avoid allocation tagging of kmemleak caches, otherwise recursive
> allocation tracking occurs.
>
> Fixes: 279bb991b4d9 ("mm/slab: add allocation accounting into slab allocation and free paths")
> Signed-off-by: Kees Cook <[email protected]>
> ---
> Cc: Suren Baghdasaryan <[email protected]>
> Cc: Kent Overstreet <[email protected]>
> Cc: Catalin Marinas <[email protected]>
> Cc: Andrew Morton <[email protected]>
> Cc: Christoph Lameter <[email protected]>
> Cc: Pekka Enberg <[email protected]>
> Cc: David Rientjes <[email protected]>
> Cc: Joonsoo Kim <[email protected]>
> Cc: Vlastimil Babka <[email protected]>
> Cc: Roman Gushchin <[email protected]>
> Cc: Hyeonggon Yoo <[email protected]>
> Cc: [email protected]
> ---
> mm/kmemleak.c | 4 ++--
> mm/slub.c | 2 +-
> 2 files changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/mm/kmemleak.c b/mm/kmemleak.c
> index c55c2cbb6837..fdcf01f62202 100644
> --- a/mm/kmemleak.c
> +++ b/mm/kmemleak.c
> @@ -463,7 +463,7 @@ static struct kmemleak_object *mem_pool_alloc(gfp_t gfp)
>
> /* try the slab allocator first */
> if (object_cache) {
> - object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
> + object = kmem_cache_alloc_noprof(object_cache, gfp_kmemleak_mask(gfp));
What do these get accounted to, or does this now pop a warning with
CONFIG_MEM_ALLOC_PROFILING_DEBUG?
> if (object)
> return object;
> }
> @@ -947,7 +947,7 @@ static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
> untagged_objp = (unsigned long)kasan_reset_tag((void *)object->pointer);
>
> if (scan_area_cache)
> - area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp));
> + area = kmem_cache_alloc_noprof(scan_area_cache, gfp_kmemleak_mask(gfp));
>
> raw_spin_lock_irqsave(&object->lock, flags);
> if (!area) {
> diff --git a/mm/slub.c b/mm/slub.c
> index a94a0507e19c..9ae032ed17ed 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -2016,7 +2016,7 @@ prepare_slab_obj_exts_hook(struct kmem_cache *s, gfp_t flags, void *p)
> if (!p)
> return NULL;
>
> - if (s->flags & SLAB_NO_OBJ_EXT)
> + if (s->flags & (SLAB_NO_OBJ_EXT | SLAB_NOLEAKTRACE))
> return NULL;
>
> if (flags & __GFP_NO_OBJ_EXT)
> --
> 2.34.1
>
On Thu, Apr 25, 2024 at 2:09 PM Kent Overstreet
<[email protected]> wrote:
>
> On Thu, Apr 25, 2024 at 01:55:23PM -0700, Kees Cook wrote:
> > The system will immediate fill up stack and crash when both
> > CONFIG_DEBUG_KMEMLEAK and CONFIG_MEM_ALLOC_PROFILING are enabled.
> > Avoid allocation tagging of kmemleak caches, otherwise recursive
> > allocation tracking occurs.
> >
> > Fixes: 279bb991b4d9 ("mm/slab: add allocation accounting into slab allocation and free paths")
> > Signed-off-by: Kees Cook <[email protected]>
> > ---
> > Cc: Suren Baghdasaryan <[email protected]>
> > Cc: Kent Overstreet <[email protected]>
> > Cc: Catalin Marinas <[email protected]>
> > Cc: Andrew Morton <[email protected]>
> > Cc: Christoph Lameter <[email protected]>
> > Cc: Pekka Enberg <[email protected]>
> > Cc: David Rientjes <[email protected]>
> > Cc: Joonsoo Kim <[email protected]>
> > Cc: Vlastimil Babka <[email protected]>
> > Cc: Roman Gushchin <[email protected]>
> > Cc: Hyeonggon Yoo <[email protected]>
> > Cc: [email protected]
> > ---
> > mm/kmemleak.c | 4 ++--
> > mm/slub.c | 2 +-
> > 2 files changed, 3 insertions(+), 3 deletions(-)
> >
> > diff --git a/mm/kmemleak.c b/mm/kmemleak.c
> > index c55c2cbb6837..fdcf01f62202 100644
> > --- a/mm/kmemleak.c
> > +++ b/mm/kmemleak.c
> > @@ -463,7 +463,7 @@ static struct kmemleak_object *mem_pool_alloc(gfp_t gfp)
> >
> > /* try the slab allocator first */
> > if (object_cache) {
> > - object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
> > + object = kmem_cache_alloc_noprof(object_cache, gfp_kmemleak_mask(gfp));
>
> What do these get accounted to, or does this now pop a warning with
> CONFIG_MEM_ALLOC_PROFILING_DEBUG?
Thanks for the fix, Kees!
I'll look into this recursion more closely to see if there is a better
way to break it. As a stopgap measure seems ok to me. I also think
it's unlikely that one would use both tracking mechanisms on the same
system.
>
> > if (object)
> > return object;
> > }
> > @@ -947,7 +947,7 @@ static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
> > untagged_objp = (unsigned long)kasan_reset_tag((void *)object->pointer);
> >
> > if (scan_area_cache)
> > - area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp));
> > + area = kmem_cache_alloc_noprof(scan_area_cache, gfp_kmemleak_mask(gfp));
> >
> > raw_spin_lock_irqsave(&object->lock, flags);
> > if (!area) {
> > diff --git a/mm/slub.c b/mm/slub.c
> > index a94a0507e19c..9ae032ed17ed 100644
> > --- a/mm/slub.c
> > +++ b/mm/slub.c
> > @@ -2016,7 +2016,7 @@ prepare_slab_obj_exts_hook(struct kmem_cache *s, gfp_t flags, void *p)
> > if (!p)
> > return NULL;
> >
> > - if (s->flags & SLAB_NO_OBJ_EXT)
> > + if (s->flags & (SLAB_NO_OBJ_EXT | SLAB_NOLEAKTRACE))
> > return NULL;
> >
> > if (flags & __GFP_NO_OBJ_EXT)
> > --
> > 2.34.1
> >
On Thu, 25 Apr 2024 14:30:55 -0700 Suren Baghdasaryan <[email protected]> wrote:
> > > --- a/mm/kmemleak.c
> > > +++ b/mm/kmemleak.c
> > > @@ -463,7 +463,7 @@ static struct kmemleak_object *mem_pool_alloc(gfp_t gfp)
> > >
> > > /* try the slab allocator first */
> > > if (object_cache) {
> > > - object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
> > > + object = kmem_cache_alloc_noprof(object_cache, gfp_kmemleak_mask(gfp));
> >
> > What do these get accounted to, or does this now pop a warning with
> > CONFIG_MEM_ALLOC_PROFILING_DEBUG?
>
> Thanks for the fix, Kees!
> I'll look into this recursion more closely to see if there is a better
> way to break it. As a stopgap measure seems ok to me. I also think
> it's unlikely that one would use both tracking mechanisms on the same
> system.
I'd really like to start building mm-stable without having to route
around memprofiling. How about I include Kees's patch in that for now?
On Thu, Apr 25, 2024 at 04:49:17PM -0700, Andrew Morton wrote:
> On Thu, 25 Apr 2024 14:30:55 -0700 Suren Baghdasaryan <[email protected]> wrote:
>
> > > > --- a/mm/kmemleak.c
> > > > +++ b/mm/kmemleak.c
> > > > @@ -463,7 +463,7 @@ static struct kmemleak_object *mem_pool_alloc(gfp_t gfp)
> > > >
> > > > /* try the slab allocator first */
> > > > if (object_cache) {
> > > > - object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
> > > > + object = kmem_cache_alloc_noprof(object_cache, gfp_kmemleak_mask(gfp));
> > >
> > > What do these get accounted to, or does this now pop a warning with
> > > CONFIG_MEM_ALLOC_PROFILING_DEBUG?
> >
> > Thanks for the fix, Kees!
> > I'll look into this recursion more closely to see if there is a better
> > way to break it. As a stopgap measure seems ok to me. I also think
> > it's unlikely that one would use both tracking mechanisms on the same
> > system.
>
> I'd really like to start building mm-stable without having to route
> around memprofiling. How about I include Kees's patch in that for now?
Agreed
On Thu, Apr 25, 2024 at 5:19 PM Kent Overstreet
<[email protected]> wrote:
>
> On Thu, Apr 25, 2024 at 04:49:17PM -0700, Andrew Morton wrote:
> > On Thu, 25 Apr 2024 14:30:55 -0700 Suren Baghdasaryan <[email protected]> wrote:
> >
> > > > > --- a/mm/kmemleak.c
> > > > > +++ b/mm/kmemleak.c
> > > > > @@ -463,7 +463,7 @@ static struct kmemleak_object *mem_pool_alloc(gfp_t gfp)
> > > > >
> > > > > /* try the slab allocator first */
> > > > > if (object_cache) {
> > > > > - object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
> > > > > + object = kmem_cache_alloc_noprof(object_cache, gfp_kmemleak_mask(gfp));
> > > >
> > > > What do these get accounted to, or does this now pop a warning with
> > > > CONFIG_MEM_ALLOC_PROFILING_DEBUG?
> > >
> > > Thanks for the fix, Kees!
> > > I'll look into this recursion more closely to see if there is a better
> > > way to break it. As a stopgap measure seems ok to me. I also think
> > > it's unlikely that one would use both tracking mechanisms on the same
> > > system.
> >
> > I'd really like to start building mm-stable without having to route
> > around memprofiling. How about I include Kees's patch in that for now?
>
> Agreed
Yes, please. When I figure out a better way, I'll post a separate patch. Thanks!
On Thu, Apr 25, 2024 at 01:55:23PM -0700, Kees Cook wrote:
> The system will immediate fill up stack and crash when both
> CONFIG_DEBUG_KMEMLEAK and CONFIG_MEM_ALLOC_PROFILING are enabled.
> Avoid allocation tagging of kmemleak caches, otherwise recursive
> allocation tracking occurs.
>
> Fixes: 279bb991b4d9 ("mm/slab: add allocation accounting into slab allocation and free paths")
> Signed-off-by: Kees Cook <[email protected]>
For the kmemleak bits:
Acked-by: Catalin Marinas <[email protected]>
On Fri, Apr 26, 2024 at 03:52:24PM +0100, Catalin Marinas wrote:
> On Thu, Apr 25, 2024 at 01:55:23PM -0700, Kees Cook wrote:
> > The system will immediate fill up stack and crash when both
Oops, typo from me: "immediately". You'd never guess I'm a native English
speaker! :)
> > CONFIG_DEBUG_KMEMLEAK and CONFIG_MEM_ALLOC_PROFILING are enabled.
> > Avoid allocation tagging of kmemleak caches, otherwise recursive
> > allocation tracking occurs.
> >
> > Fixes: 279bb991b4d9 ("mm/slab: add allocation accounting into slab allocation and free paths")
> > Signed-off-by: Kees Cook <[email protected]>
>
> For the kmemleak bits:
>
> Acked-by: Catalin Marinas <[email protected]>
Thanks!
-Kees
--
Kees Cook