[PATCH v3 1/2] kfence: move the size check to the beginning of __kfence_alloc()

[PATCH v3 1/2] kfence: move the size check to the beginning of __kfence_alloc()

[Alexander Potapenko]

Check the allocation size before toggling kfence_allocation_gate.
This way allocations that can't be served by KFENCE will not result in
waiting for another CONFIG_KFENCE_SAMPLE_INTERVAL without allocating
anything.

Suggested-by: Marco Elver <elver@google.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Marco Elver <elver@google.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: stable@vger.kernel.org # 5.12+
Signed-off-by: Alexander Potapenko <glider@google.com>
---
 mm/kfence/core.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/mm/kfence/core.c b/mm/kfence/core.c
index 4d21ac44d5d35..33bb20d91bf6a 100644
--- a/mm/kfence/core.c
+++ b/mm/kfence/core.c
@@ -733,6 +733,13 @@ void kfence_shutdown_cache(struct kmem_cache *s)
 
 void *__kfence_alloc(struct kmem_cache *s, size_t size, gfp_t flags)
 {
+	/*
+	 * Perform size check before switching kfence_allocation_gate, so that
+	 * we don't disable KFENCE without making an allocation.
+	 */
+	if (size > PAGE_SIZE)
+		return NULL;
+
 	/*
 	 * allocation_gate only needs to become non-zero, so it doesn't make
 	 * sense to continue writing to it and pay the associated contention
@@ -757,9 +764,6 @@ void *__kfence_alloc(struct kmem_cache *s, size_t size, gfp_t flags)
 	if (!READ_ONCE(kfence_enabled))
 		return NULL;
 
-	if (size > PAGE_SIZE)
-		return NULL;
-
 	return kfence_guarded_alloc(s, size, flags);
 }
 
-- 
2.32.0.93.g670b81a890-goog

[PATCH v3 2/2] kfence: skip all GFP_ZONEMASK allocations

[Alexander Potapenko]

Allocation requests outside ZONE_NORMAL (MOVABLE, HIGHMEM or DNA) cannot
be fulfilled by KFENCE, because KFENCE memory pool is located in a
zone different from the requested one.

Because callers of kmem_cache_alloc() may actually rely on the
allocation to reside in the requested zone (e.g. memory allocations done
with __GFP_DMA must be DMAable), skip all allocations done with
GFP_ZONEMASK and/or respective SLAB flags (SLAB_CACHE_DMA and
SLAB_CACHE_DMA32).

Fixes: 0ce20dd84089 ("mm: add Kernel Electric-Fence infrastructure")
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Marco Elver <elver@google.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Souptick Joarder <jrdr.linux@gmail.com>
Cc: stable@vger.kernel.org # 5.12+
Signed-off-by: Alexander Potapenko <glider@google.com>

---

v2:
 - added parentheses around the GFP clause, as requested by Marco
v3:
 - ignore GFP_ZONEMASK, which also covers __GFP_HIGHMEM and __GFP_MOVABLE
 - move the flag check at the beginning of the function, as requested by
   Souptick Joarder
---
 mm/kfence/core.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/mm/kfence/core.c b/mm/kfence/core.c
index 33bb20d91bf6a..d51f77329fd3c 100644
--- a/mm/kfence/core.c
+++ b/mm/kfence/core.c
@@ -740,6 +740,14 @@ void *__kfence_alloc(struct kmem_cache *s, size_t size, gfp_t flags)
 	if (size > PAGE_SIZE)
 		return NULL;
 
+	/*
+	 * Skip allocations from non-default zones, including DMA. We cannot guarantee that pages
+	 * in the KFENCE pool will have the requested properties (e.g. reside in DMAable memory).
+	 */
+	if ((flags & GFP_ZONEMASK) ||
+	    (s->flags & (SLAB_CACHE_DMA | SLAB_CACHE_DMA32)))
+		return NULL;
+
 	/*
 	 * allocation_gate only needs to become non-zero, so it doesn't make
 	 * sense to continue writing to it and pay the associated contention
-- 
2.32.0.93.g670b81a890-goog

Re: [PATCH v3 2/2] kfence: skip all GFP_ZONEMASK allocations

[Marco Elver]

On Wed, 30 Jun 2021 at 15:53, Alexander Potapenko <glider@google.com> wrote:
> Allocation requests outside ZONE_NORMAL (MOVABLE, HIGHMEM or DNA) cannot

s/DNA/DMA/
... but probably no need to do v4 just for this (everyone knows we're
not yet in the business of allocating DNA ;-)).

> be fulfilled by KFENCE, because KFENCE memory pool is located in a
> zone different from the requested one.
>
> Because callers of kmem_cache_alloc() may actually rely on the
> allocation to reside in the requested zone (e.g. memory allocations done
> with __GFP_DMA must be DMAable), skip all allocations done with
> GFP_ZONEMASK and/or respective SLAB flags (SLAB_CACHE_DMA and
> SLAB_CACHE_DMA32).
>
> Fixes: 0ce20dd84089 ("mm: add Kernel Electric-Fence infrastructure")
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: Dmitry Vyukov <dvyukov@google.com>
> Cc: Marco Elver <elver@google.com>
> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> Cc: Souptick Joarder <jrdr.linux@gmail.com>
> Cc: stable@vger.kernel.org # 5.12+
> Signed-off-by: Alexander Potapenko <glider@google.com>

With the change below, you can add:

  Reviewed-by: Marco Elver <elver@google.com>


> ---
>
> v2:
>  - added parentheses around the GFP clause, as requested by Marco
> v3:
>  - ignore GFP_ZONEMASK, which also covers __GFP_HIGHMEM and __GFP_MOVABLE
>  - move the flag check at the beginning of the function, as requested by
>    Souptick Joarder
> ---
>  mm/kfence/core.c | 8 ++++++++
>  1 file changed, 8 insertions(+)
>
> diff --git a/mm/kfence/core.c b/mm/kfence/core.c
> index 33bb20d91bf6a..d51f77329fd3c 100644
> --- a/mm/kfence/core.c
> +++ b/mm/kfence/core.c
> @@ -740,6 +740,14 @@ void *__kfence_alloc(struct kmem_cache *s, size_t size, gfp_t flags)
>         if (size > PAGE_SIZE)
>                 return NULL;
>
> +       /*
> +        * Skip allocations from non-default zones, including DMA. We cannot guarantee that pages
> +        * in the KFENCE pool will have the requested properties (e.g. reside in DMAable memory).

Comments should still be 80 cols, like the rest of the file. :-/

> +        */
> +       if ((flags & GFP_ZONEMASK) ||
> +           (s->flags & (SLAB_CACHE_DMA | SLAB_CACHE_DMA32)))
> +               return NULL;
> +
>         /*
>          * allocation_gate only needs to become non-zero, so it doesn't make
>          * sense to continue writing to it and pay the associated contention
> --
> 2.32.0.93.g670b81a890-goog
>

Re: [PATCH v3 1/2] kfence: move the size check to the beginning of __kfence_alloc()

[Marco Elver]

On Wed, 30 Jun 2021 at 15:53, Alexander Potapenko <glider@google.com> wrote:
>
> Check the allocation size before toggling kfence_allocation_gate.
> This way allocations that can't be served by KFENCE will not result in
> waiting for another CONFIG_KFENCE_SAMPLE_INTERVAL without allocating
> anything.
>
> Suggested-by: Marco Elver <elver@google.com>
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: Dmitry Vyukov <dvyukov@google.com>
> Cc: Marco Elver <elver@google.com>
> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> Cc: stable@vger.kernel.org # 5.12+
> Signed-off-by: Alexander Potapenko <glider@google.com>

Reviewed-by: Marco Elver <elver@google.com>


> ---
>  mm/kfence/core.c | 10 +++++++---
>  1 file changed, 7 insertions(+), 3 deletions(-)
>
> diff --git a/mm/kfence/core.c b/mm/kfence/core.c
> index 4d21ac44d5d35..33bb20d91bf6a 100644
> --- a/mm/kfence/core.c
> +++ b/mm/kfence/core.c
> @@ -733,6 +733,13 @@ void kfence_shutdown_cache(struct kmem_cache *s)
>
>  void *__kfence_alloc(struct kmem_cache *s, size_t size, gfp_t flags)
>  {
> +       /*
> +        * Perform size check before switching kfence_allocation_gate, so that
> +        * we don't disable KFENCE without making an allocation.
> +        */
> +       if (size > PAGE_SIZE)
> +               return NULL;
> +
>         /*
>          * allocation_gate only needs to become non-zero, so it doesn't make
>          * sense to continue writing to it and pay the associated contention
> @@ -757,9 +764,6 @@ void *__kfence_alloc(struct kmem_cache *s, size_t size, gfp_t flags)
>         if (!READ_ONCE(kfence_enabled))
>                 return NULL;
>
> -       if (size > PAGE_SIZE)
> -               return NULL;
> -
>         return kfence_guarded_alloc(s, size, flags);
>  }
>
> --
> 2.32.0.93.g670b81a890-goog
>