Re: [PATCH 1/1] kasan: add memory corruption identification for hardware tag-based mode

From: Kuan-Ying Lee
Date: Mon May 31 2021 - 13:24:41 EST


On Mon, May 31, 2021 at 10:50:24AM +0200, Marco Elver wrote:
> On Sun, May 30, 2021 at 12:47PM +0800, Kuan-Ying Lee wrote:
> > Add memory corruption identification at bug report for hardware tag-based
> > mode. The report shows whether it is "use-after-free" or "out-of-bound"
> > error instead of "invalid-access" error. This will make it easier for
> > programmers to see the memory corruption problem.
> >
> > We extend the slab to store five old free pointer tag and free backtrace,
> > we can check if the tagged address is in the slab record and make a good
> > guess if the object is more like "use-after-free" or "out-of-bound".
> > therefore every slab memory corruption can be identified whether it's
> > "use-after-free" or "out-of-bound".
> >
> > Signed-off-by: Kuan-Ying Lee <kylee0686026@xxxxxxxxx>
>

> On a whole this makes sense because SW_TAGS mode supports this, too.
>
> My main complaints are the copy-paste of the SW_TAGS code.
>
> Does it make sense to refactor per my suggestions below?

Thanks for your suggestions.
I will refactor them in v2.
>
> This is also a question to KASAN maintainers (Andrey, any preference?).
>
> > ---
> > lib/Kconfig.kasan | 8 ++++++++
> > mm/kasan/hw_tags.c | 25 ++++++++++++++++++++++---
> > mm/kasan/kasan.h | 4 ++--
> > mm/kasan/report_hw_tags.c | 28 ++++++++++++++++++++++++++++
> > 4 files changed, 60 insertions(+), 5 deletions(-)
> >
> > diff --git a/lib/Kconfig.kasan b/lib/Kconfig.kasan
> > index cffc2ebbf185..f7e666b23058 100644
> > --- a/lib/Kconfig.kasan
> > +++ b/lib/Kconfig.kasan
> > @@ -163,6 +163,14 @@ config KASAN_SW_TAGS_IDENTIFY
> > (use-after-free or out-of-bounds) at the cost of increased
> > memory consumption.
> >
> > +config KASAN_HW_TAGS_IDENTIFY
> > + bool "Enable memory corruption identification"
> > + depends on KASAN_HW_TAGS
> > + help
> > + This option enables best-effort identification of bug type
> > + (use-after-free or out-of-bounds) at the cost of increased
> > + memory consumption.
>
> Can we rename KASAN_SW_TAGS_IDENTIFY -> KASAN_TAGS_IDENTIFY in a
> separate patch and then use that?
>
> Or do we have a problem renaming this options if there are existing
> users of it?

I tend to keep KASAN_SW_TAGS_IDENTIFY and KASAN_HW_TAGS_IDENTIFY
separately.

We need these two configs to decide how many stacks we will store.

If we store as many stacks as SW tag-based kasan does(5 stacks), we might
mistake out-of-bound issues for use-after-free sometime. Becuase HW
tag-based kasan only has 16 kinds of tags. When Out-of-bound issues happened, it might
find the same tag in the stack we just stored and mistake happened.
There is high probability that this mistake will happen.
>
> > config KASAN_VMALLOC
> > bool "Back mappings in vmalloc space with real shadow memory"
> > depends on KASAN_GENERIC && HAVE_ARCH_KASAN_VMALLOC
> > diff --git a/mm/kasan/hw_tags.c b/mm/kasan/hw_tags.c
> > index 4004388b4e4b..b1c6bb116600 100644
> > --- a/mm/kasan/hw_tags.c
> > +++ b/mm/kasan/hw_tags.c
> > @@ -220,22 +220,41 @@ void kasan_set_free_info(struct kmem_cache *cache,
> > void *object, u8 tag)
> > {
> > struct kasan_alloc_meta *alloc_meta;
> > + u8 idx = 0;
> >
> > alloc_meta = kasan_get_alloc_meta(cache, object);
> > - if (alloc_meta)
> > - kasan_set_track(&alloc_meta->free_track[0], GFP_NOWAIT);
> > + if (!alloc_meta)
> > + return;
> > +
> > +#ifdef CONFIG_KASAN_HW_TAGS_IDENTIFY
> > + idx = alloc_meta->free_track_idx;
> > + alloc_meta->free_pointer_tag[idx] = tag;
> > + alloc_meta->free_track_idx = (idx + 1) % KASAN_NR_FREE_STACKS;
> > +#endif
> > +
> > + kasan_set_track(&alloc_meta->free_track[idx], GFP_NOWAIT);
> > }
> >
> > struct kasan_track *kasan_get_free_track(struct kmem_cache *cache,
> > void *object, u8 tag)
> > {
> > struct kasan_alloc_meta *alloc_meta;
> > + int i = 0;
> >
> > alloc_meta = kasan_get_alloc_meta(cache, object);
> > if (!alloc_meta)
> > return NULL;
> >
> > - return &alloc_meta->free_track[0];
> > +#ifdef CONFIG_KASAN_HW_TAGS_IDENTIFY
> > + for (i = 0; i < KASAN_NR_FREE_STACKS; i++) {
> > + if (alloc_meta->free_pointer_tag[i] == tag)
> > + break;
> > + }
> > + if (i == KASAN_NR_FREE_STACKS)
> > + i = alloc_meta->free_track_idx;
> > +#endif
> > +
> > + return &alloc_meta->free_track[i];
> > }
>
> Again, we now have code duplication. These functions are now identical
> to the sw_tags.c ones?
>
> Does it make sense to also move them in a preparatory patch to a new
> 'tags.c'?
>
Yes, moving them into tags.c will be better.
I will refactor in v2.
> > #if IS_ENABLED(CONFIG_KASAN_KUNIT_TEST)
> > diff --git a/mm/kasan/kasan.h b/mm/kasan/kasan.h
> > index 8f450bc28045..41b47f456130 100644
> > --- a/mm/kasan/kasan.h
> > +++ b/mm/kasan/kasan.h
> > @@ -153,7 +153,7 @@ struct kasan_track {
> > depot_stack_handle_t stack;
> > };
> >

I think my v1 patch sets KASAN_NR_FREE_STACKS to 5 is not suitable.
The same reason as above.

I am thinking to store 2 or 1 stacks is acceptable in HW tag-based kasan mode.
Does it make sense?
Any suggetions are appreciated.

> > -#ifdef CONFIG_KASAN_SW_TAGS_IDENTIFY
> > +#if defined(CONFIG_KASAN_SW_TAGS_IDENTIFY) || defined(CONFIG_KASAN_HW_TAGS_IDENTIFY)
> > #define KASAN_NR_FREE_STACKS 5
> > #else
> > #define KASAN_NR_FREE_STACKS 1
> > @@ -170,7 +170,7 @@ struct kasan_alloc_meta {
> > #else
> > struct kasan_track free_track[KASAN_NR_FREE_STACKS];
> > #endif
> > -#ifdef CONFIG_KASAN_SW_TAGS_IDENTIFY
> > +#if defined(CONFIG_KASAN_SW_TAGS_IDENTIFY) || defined(CONFIG_KASAN_HW_TAGS_IDENTIFY)
> > u8 free_pointer_tag[KASAN_NR_FREE_STACKS];
> > u8 free_track_idx;
> > #endif
> > diff --git a/mm/kasan/report_hw_tags.c b/mm/kasan/report_hw_tags.c
> > index 42b2168755d6..d77109b85a09 100644
> > --- a/mm/kasan/report_hw_tags.c
> > +++ b/mm/kasan/report_hw_tags.c
> > @@ -14,9 +14,37 @@
> > #include <linux/types.h>
> >
> > #include "kasan.h"
> > +#include "../slab.h"
> >
> > const char *kasan_get_bug_type(struct kasan_access_info *info)
> > {
> > +#ifdef CONFIG_KASAN_HW_TAGS_IDENTIFY
> > + struct kasan_alloc_meta *alloc_meta;
> > + struct kmem_cache *cache;
> > + struct page *page;
> > + const void *addr;
> > + void *object;
> > + u8 tag;
> > + int i;
> > +
> > + tag = get_tag(info->access_addr);
> > + addr = kasan_reset_tag(info->access_addr);
> > + page = kasan_addr_to_page(addr);
> > + if (page && PageSlab(page)) {
> > + cache = page->slab_cache;
> > + object = nearest_obj(cache, page, (void *)addr);
> > + alloc_meta = kasan_get_alloc_meta(cache, object);
> > +
> > + if (alloc_meta) {
> > + for (i = 0; i < KASAN_NR_FREE_STACKS; i++) {
> > + if (alloc_meta->free_pointer_tag[i] == tag)
> > + return "use-after-free";
> > + }
> > + }
> > + return "out-of-bounds";
> > + }
> > +
> > +#endif
> > return "invalid-access";
> > }
>
> This function is an almost copy-paste of what we have in
> report_sw_tags.c. Does it make sense to try and share this code or would
> it complicate things?
>

I got your point.
I will refactor them in v2.

Thanks,
Kuan-Ying Lee

> I imagine we could have a header report_tags.h, which defines a static
> const char *kasan_try_get_bug_type(..), and simply returns NULL if it
> couldn't identify it:
>
> #if defined(CONFIG_KASAN_SW_TAGS_IDENTIFY) || defined(CONFIG_KASAN_HW_TAGS_IDENTIFY)
> static const char *kasan_try_get_bug_type(struct kasan_access_info *info)
> {
> ... the code above ...
>
> return NULL;
> }
> #else
> static const char *kasan_try_get_bug_type(struct kasan_access_info *info) { return NULL; }
> #endif
>
>
> Thanks,
> -- Marco