alloc_tag: Fix boot failure due to NULL pointer dereference

[PATCH linux-next] alloc_tag: Fix boot failure due to NULL pointer dereference

Posted by ranxiaokai627@163.com 4 months, 2 weeks ago

From: Ran Xiaokai <ran.xiaokai@zte.com.cn>

There is a boot failure when both CONFIG_DEBUG_KMEMLEAK and
CONFIG_MEM_ALLOC_PROFILING are enabled.

BUG: kernel NULL pointer dereference, address: 0000000000000000
RIP: 0010:__alloc_tagging_slab_alloc_hook+0x181/0x2f0
Call Trace:
 kmem_cache_alloc_noprof+0x1c8/0x5c0
 __alloc_object+0x2f/0x290
 __create_object+0x22/0x80
 kmemleak_init+0x122/0x190
 mm_core_init+0xb6/0x160
 start_kernel+0x39f/0x920
 x86_64_start_reservations+0x18/0x30
 x86_64_start_kernel+0x104/0x120
 common_startup_64+0x12c/0x138

In kmemleak, mem_pool_alloc() directly calls kmem_cache_alloc_noprof(),
as a result, the alloc_tag structure associated with object_cache is not
defined neither initialized. So current->alloc_tag is NULL,
leading to a null pointer dereference.

Move the checks for SLAB_NO_OBJ_EXT, SLAB_NOLEAKTRACE, and
__GFP_NO_OBJ_EXT to the parent function __alloc_tagging_slab_alloc_hook()
to fix this.

Also this distinguishes the SLAB_NOLEAKTRACE case between the actual memory
allocation failures case, make CODETAG_FLAG_INACCURATE more accurate.

Fixes: b9e2f58ffb84 ("alloc_tag: mark inaccurate allocation counters in /proc/allocinfo output")
Signed-off-by: Ran Xiaokai <ran.xiaokai@zte.com.cn>
---
 mm/slub.c | 18 +++++++++---------
 1 file changed, 9 insertions(+), 9 deletions(-)

diff --git a/mm/slub.c b/mm/slub.c
index 867a07260acf..09cbe580842c 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -2197,15 +2197,6 @@ prepare_slab_obj_exts_hook(struct kmem_cache *s, gfp_t flags, void *p)
 {
 	struct slab *slab;
 
-	if (!p)
-		return NULL;
-
-	if (s->flags & (SLAB_NO_OBJ_EXT | SLAB_NOLEAKTRACE))
-		return NULL;
-
-	if (flags & __GFP_NO_OBJ_EXT)
-		return NULL;
-
 	slab = virt_to_slab(p);
 	if (!slab_obj_exts(slab) &&
 	    alloc_slab_obj_exts(slab, s, flags, false)) {
@@ -2223,6 +2214,15 @@ __alloc_tagging_slab_alloc_hook(struct kmem_cache *s, void *object, gfp_t flags)
 {
 	struct slabobj_ext *obj_exts;
 
+	if (!object)
+		return;
+
+	if (s->flags & (SLAB_NO_OBJ_EXT | SLAB_NOLEAKTRACE))
+		return;
+
+	if (flags & __GFP_NO_OBJ_EXT)
+		return;
+
 	obj_exts = prepare_slab_obj_exts_hook(s, flags, object);
 	/*
 	 * Currently obj_exts is used only for allocation profiling.
-- 
2.25.1

Re: [PATCH linux-next] alloc_tag: Fix boot failure due to NULL pointer dereference

Posted by Vlastimil Babka 4 months, 2 weeks ago

On 9/26/25 10:06, ranxiaokai627@163.com wrote:
> From: Ran Xiaokai <ran.xiaokai@zte.com.cn>
> 
> There is a boot failure when both CONFIG_DEBUG_KMEMLEAK and
> CONFIG_MEM_ALLOC_PROFILING are enabled.
> 
> BUG: kernel NULL pointer dereference, address: 0000000000000000
> RIP: 0010:__alloc_tagging_slab_alloc_hook+0x181/0x2f0
> Call Trace:
>  kmem_cache_alloc_noprof+0x1c8/0x5c0
>  __alloc_object+0x2f/0x290
>  __create_object+0x22/0x80
>  kmemleak_init+0x122/0x190
>  mm_core_init+0xb6/0x160
>  start_kernel+0x39f/0x920
>  x86_64_start_reservations+0x18/0x30
>  x86_64_start_kernel+0x104/0x120
>  common_startup_64+0x12c/0x138
> 
> In kmemleak, mem_pool_alloc() directly calls kmem_cache_alloc_noprof(),
> as a result, the alloc_tag structure associated with object_cache is not
> defined neither initialized. So current->alloc_tag is NULL,
> leading to a null pointer dereference.

Agree with Harry. This should be enough:

"as a result, current->alloc_tag is NULL, leading to a null pointer
dereference."

> Move the checks for SLAB_NO_OBJ_EXT, SLAB_NOLEAKTRACE, and
> __GFP_NO_OBJ_EXT to the parent function __alloc_tagging_slab_alloc_hook()
> to fix this.
> 
> Also this distinguishes the SLAB_NOLEAKTRACE case between the actual memory
> allocation failures case, make CODETAG_FLAG_INACCURATE more accurate.

Good point.

> Fixes: b9e2f58ffb84 ("alloc_tag: mark inaccurate allocation counters in /proc/allocinfo output")

That's in mm-stable so the fix should go there (probably too late to fold
now) if it's to be in the merge window PR.

> Signed-off-by: Ran Xiaokai <ran.xiaokai@zte.com.cn>

Acked-by: Vlastimil Babka <vbabka@suse.cz>

> ---
>  mm/slub.c | 18 +++++++++---------
>  1 file changed, 9 insertions(+), 9 deletions(-)
> 
> diff --git a/mm/slub.c b/mm/slub.c
> index 867a07260acf..09cbe580842c 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -2197,15 +2197,6 @@ prepare_slab_obj_exts_hook(struct kmem_cache *s, gfp_t flags, void *p)
>  {
>  	struct slab *slab;
>  
> -	if (!p)
> -		return NULL;
> -
> -	if (s->flags & (SLAB_NO_OBJ_EXT | SLAB_NOLEAKTRACE))
> -		return NULL;
> -
> -	if (flags & __GFP_NO_OBJ_EXT)
> -		return NULL;
> -
>  	slab = virt_to_slab(p);
>  	if (!slab_obj_exts(slab) &&
>  	    alloc_slab_obj_exts(slab, s, flags, false)) {
> @@ -2223,6 +2214,15 @@ __alloc_tagging_slab_alloc_hook(struct kmem_cache *s, void *object, gfp_t flags)
>  {
>  	struct slabobj_ext *obj_exts;
>  
> +	if (!object)
> +		return;
> +
> +	if (s->flags & (SLAB_NO_OBJ_EXT | SLAB_NOLEAKTRACE))
> +		return;
> +
> +	if (flags & __GFP_NO_OBJ_EXT)
> +		return;
> +
>  	obj_exts = prepare_slab_obj_exts_hook(s, flags, object);
>  	/*
>  	 * Currently obj_exts is used only for allocation profiling.

Re: [PATCH linux-next] alloc_tag: Fix boot failure due to NULL pointer dereference

Posted by Suren Baghdasaryan 4 months, 2 weeks ago

On Fri, Sep 26, 2025 at 5:48 AM Vlastimil Babka <vbabka@suse.cz> wrote:
>
> On 9/26/25 10:06, ranxiaokai627@163.com wrote:
> > From: Ran Xiaokai <ran.xiaokai@zte.com.cn>
> >
> > There is a boot failure when both CONFIG_DEBUG_KMEMLEAK and
> > CONFIG_MEM_ALLOC_PROFILING are enabled.
> >
> > BUG: kernel NULL pointer dereference, address: 0000000000000000
> > RIP: 0010:__alloc_tagging_slab_alloc_hook+0x181/0x2f0
> > Call Trace:
> >  kmem_cache_alloc_noprof+0x1c8/0x5c0
> >  __alloc_object+0x2f/0x290
> >  __create_object+0x22/0x80
> >  kmemleak_init+0x122/0x190
> >  mm_core_init+0xb6/0x160
> >  start_kernel+0x39f/0x920
> >  x86_64_start_reservations+0x18/0x30
> >  x86_64_start_kernel+0x104/0x120
> >  common_startup_64+0x12c/0x138
> >
> > In kmemleak, mem_pool_alloc() directly calls kmem_cache_alloc_noprof(),
> > as a result, the alloc_tag structure associated with object_cache is not
> > defined neither initialized. So current->alloc_tag is NULL,
> > leading to a null pointer dereference.
>
> Agree with Harry. This should be enough:
>
> "as a result, current->alloc_tag is NULL, leading to a null pointer
> dereference."

Yes, that's much more clear.

>
> > Move the checks for SLAB_NO_OBJ_EXT, SLAB_NOLEAKTRACE, and
> > __GFP_NO_OBJ_EXT to the parent function __alloc_tagging_slab_alloc_hook()
> > to fix this.
> >
> > Also this distinguishes the SLAB_NOLEAKTRACE case between the actual memory
> > allocation failures case, make CODETAG_FLAG_INACCURATE more accurate.

Awsome!

>
> Good point.
>
> > Fixes: b9e2f58ffb84 ("alloc_tag: mark inaccurate allocation counters in /proc/allocinfo output")
>
> That's in mm-stable so the fix should go there (probably too late to fold
> now) if it's to be in the merge window PR.
>
> > Signed-off-by: Ran Xiaokai <ran.xiaokai@zte.com.cn>
>
> Acked-by: Vlastimil Babka <vbabka@suse.cz>

Reviewed-by: Suren Baghdasaryan <surenb@google.com>

Thanks for the fix!

>
> > ---
> >  mm/slub.c | 18 +++++++++---------
> >  1 file changed, 9 insertions(+), 9 deletions(-)
> >
> > diff --git a/mm/slub.c b/mm/slub.c
> > index 867a07260acf..09cbe580842c 100644
> > --- a/mm/slub.c
> > +++ b/mm/slub.c
> > @@ -2197,15 +2197,6 @@ prepare_slab_obj_exts_hook(struct kmem_cache *s, gfp_t flags, void *p)
> >  {
> >       struct slab *slab;
> >
> > -     if (!p)
> > -             return NULL;
> > -
> > -     if (s->flags & (SLAB_NO_OBJ_EXT | SLAB_NOLEAKTRACE))
> > -             return NULL;
> > -
> > -     if (flags & __GFP_NO_OBJ_EXT)
> > -             return NULL;
> > -
> >       slab = virt_to_slab(p);
> >       if (!slab_obj_exts(slab) &&
> >           alloc_slab_obj_exts(slab, s, flags, false)) {
> > @@ -2223,6 +2214,15 @@ __alloc_tagging_slab_alloc_hook(struct kmem_cache *s, void *object, gfp_t flags)
> >  {
> >       struct slabobj_ext *obj_exts;
> >
> > +     if (!object)
> > +             return;
> > +
> > +     if (s->flags & (SLAB_NO_OBJ_EXT | SLAB_NOLEAKTRACE))
> > +             return;
> > +
> > +     if (flags & __GFP_NO_OBJ_EXT)
> > +             return;
> > +
> >       obj_exts = prepare_slab_obj_exts_hook(s, flags, object);
> >       /*
> >        * Currently obj_exts is used only for allocation profiling.
>

Re: [PATCH linux-next] alloc_tag: Fix boot failure due to NULL pointer dereference

Posted by Harry Yoo 4 months, 2 weeks ago

On Fri, Sep 26, 2025 at 08:06:59AM +0000, ranxiaokai627@163.com wrote:
> From: Ran Xiaokai <ran.xiaokai@zte.com.cn>
> 
> There is a boot failure when both CONFIG_DEBUG_KMEMLEAK and
> CONFIG_MEM_ALLOC_PROFILING are enabled.
> 
> BUG: kernel NULL pointer dereference, address: 0000000000000000
> RIP: 0010:__alloc_tagging_slab_alloc_hook+0x181/0x2f0
> Call Trace:
>  kmem_cache_alloc_noprof+0x1c8/0x5c0
>  __alloc_object+0x2f/0x290
>  __create_object+0x22/0x80
>  kmemleak_init+0x122/0x190
>  mm_core_init+0xb6/0x160
>  start_kernel+0x39f/0x920
>  x86_64_start_reservations+0x18/0x30
>  x86_64_start_kernel+0x104/0x120
>  common_startup_64+0x12c/0x138
> 
> In kmemleak, mem_pool_alloc() directly calls kmem_cache_alloc_noprof(),
> as a result, the alloc_tag structure associated with object_cache is not
> defined neither initialized.

nit: "the alloc_tag structure associated with object_cache is not
defined neither initialized" sounds weird because here we're talking
about current->alloc_tag being NULL, and it is not something we
associate with a slab cache.

> So current->alloc_tag is NULL,
> leading to a null pointer dereference.

> Move the checks for SLAB_NO_OBJ_EXT, SLAB_NOLEAKTRACE, and
> __GFP_NO_OBJ_EXT to the parent function __alloc_tagging_slab_alloc_hook()
> to fix this.
> 
> Also this distinguishes the SLAB_NOLEAKTRACE case between the actual memory
> allocation failures case, make CODETAG_FLAG_INACCURATE more accurate.
> 
> Fixes: b9e2f58ffb84 ("alloc_tag: mark inaccurate allocation counters in /proc/allocinfo output")
> Signed-off-by: Ran Xiaokai <ran.xiaokai@zte.com.cn>
> ---

In general, the fix makes sense and looks good to me.

Reviewed-by: Harry Yoo <harry.yoo@oracle.com>

-- 
Cheers,
Harry / Hyeonggon

>  mm/slub.c | 18 +++++++++---------
>  1 file changed, 9 insertions(+), 9 deletions(-)
> 
> diff --git a/mm/slub.c b/mm/slub.c
> index 867a07260acf..09cbe580842c 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -2197,15 +2197,6 @@ prepare_slab_obj_exts_hook(struct kmem_cache *s, gfp_t flags, void *p)
>  {
>  	struct slab *slab;
>  
> -	if (!p)
> -		return NULL;
> -
> -	if (s->flags & (SLAB_NO_OBJ_EXT | SLAB_NOLEAKTRACE))
> -		return NULL;
> -
> -	if (flags & __GFP_NO_OBJ_EXT)
> -		return NULL;
> -
>  	slab = virt_to_slab(p);
>  	if (!slab_obj_exts(slab) &&
>  	    alloc_slab_obj_exts(slab, s, flags, false)) {
> @@ -2223,6 +2214,15 @@ __alloc_tagging_slab_alloc_hook(struct kmem_cache *s, void *object, gfp_t flags)
>  {
>  	struct slabobj_ext *obj_exts;
>  
> +	if (!object)
> +		return;
> +
> +	if (s->flags & (SLAB_NO_OBJ_EXT | SLAB_NOLEAKTRACE))
> +		return;
> +
> +	if (flags & __GFP_NO_OBJ_EXT)
> +		return;
> +
>  	obj_exts = prepare_slab_obj_exts_hook(s, flags, object);
>  	/*
>  	 * Currently obj_exts is used only for allocation profiling.
> -- 
> 2.25.1
> 
>