Re: [PATCH v4 2/4] kasan: record and print the free track

From: Dmitry Vyukov
Date: Wed May 20 2020 - 05:44:29 EST


On Wed, May 20, 2020 at 11:17 AM Walter Wu <walter-zh.wu@xxxxxxxxxxxx> wrote:
> > > On Wed, 2020-05-20 at 13:14 +0800, Walter Wu wrote:
> > > > > On Wed, May 20, 2020 at 6:03 AM Walter Wu <walter-zh.wu@xxxxxxxxxxxx> wrote:
> > > > > >
> > > > > > > On Tue, May 19, 2020 at 4:25 AM Walter Wu <walter-zh.wu@xxxxxxxxxxxx> wrote:
> > > > > > > >
> > > > > > > > Move free track from slub alloc meta-data to slub free meta-data in
> > > > > > > > order to make struct kasan_free_meta size is 16 bytes. It is a good
> > > > > > > > size because it is the minimal redzone size and a good number of
> > > > > > > > alignment.
> > > > > > > >
> > > > > > > > For free track in generic KASAN, we do the modification in struct
> > > > > > > > kasan_alloc_meta and kasan_free_meta:
> > > > > > > > - remove free track from kasan_alloc_meta.
> > > > > > > > - add free track into kasan_free_meta.
> > > > > > > >
> > > > > > > > [1]https://bugzilla.kernel.org/show_bug.cgi?id=198437
> > > > > > > >
> > > > > > > > Signed-off-by: Walter Wu <walter-zh.wu@xxxxxxxxxxxx>
> > > > > > > > Suggested-by: Dmitry Vyukov <dvyukov@xxxxxxxxxx>
> > > > > > > > Cc: Andrey Ryabinin <aryabinin@xxxxxxxxxxxxx>
> > > > > > > > Cc: Dmitry Vyukov <dvyukov@xxxxxxxxxx>
> > > > > > > > Cc: Alexander Potapenko <glider@xxxxxxxxxx>
> > > > > > > > ---
> > > > > > > > mm/kasan/common.c | 22 ++--------------------
> > > > > > > > mm/kasan/generic.c | 18 ++++++++++++++++++
> > > > > > > > mm/kasan/kasan.h | 7 +++++++
> > > > > > > > mm/kasan/report.c | 20 --------------------
> > > > > > > > mm/kasan/tags.c | 37 +++++++++++++++++++++++++++++++++++++
> > > > > > > > 5 files changed, 64 insertions(+), 40 deletions(-)
> > > > > > > >
> > > > > > > > diff --git a/mm/kasan/common.c b/mm/kasan/common.c
> > > > > > > > index 8bc618289bb1..47b53912f322 100644
> > > > > > > > --- a/mm/kasan/common.c
> > > > > > > > +++ b/mm/kasan/common.c
> > > > > > > > @@ -51,7 +51,7 @@ depot_stack_handle_t kasan_save_stack(gfp_t flags)
> > > > > > > > return stack_depot_save(entries, nr_entries, flags);
> > > > > > > > }
> > > > > > > >
> > > > > > > > -static inline void set_track(struct kasan_track *track, gfp_t flags)
> > > > > > > > +void kasan_set_track(struct kasan_track *track, gfp_t flags)
> > > > > > > > {
> > > > > > > > track->pid = current->pid;
> > > > > > > > track->stack = kasan_save_stack(flags);
> > > > > > > > @@ -299,24 +299,6 @@ struct kasan_free_meta *get_free_info(struct kmem_cache *cache,
> > > > > > > > return (void *)object + cache->kasan_info.free_meta_offset;
> > > > > > > > }
> > > > > > > >
> > > > > > > > -
> > > > > > > > -static void kasan_set_free_info(struct kmem_cache *cache,
> > > > > > > > - void *object, u8 tag)
> > > > > > > > -{
> > > > > > > > - struct kasan_alloc_meta *alloc_meta;
> > > > > > > > - u8 idx = 0;
> > > > > > > > -
> > > > > > > > - alloc_meta = get_alloc_info(cache, object);
> > > > > > > > -
> > > > > > > > -#ifdef CONFIG_KASAN_SW_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
> > > > > > > > -
> > > > > > > > - set_track(&alloc_meta->free_track[idx], GFP_NOWAIT);
> > > > > > > > -}
> > > > > > > > -
> > > > > > > > void kasan_poison_slab(struct page *page)
> > > > > > > > {
> > > > > > > > unsigned long i;
> > > > > > > > @@ -492,7 +474,7 @@ static void *__kasan_kmalloc(struct kmem_cache *cache, const void *object,
> > > > > > > > KASAN_KMALLOC_REDZONE);
> > > > > > > >
> > > > > > > > if (cache->flags & SLAB_KASAN)
> > > > > > > > - set_track(&get_alloc_info(cache, object)->alloc_track, flags);
> > > > > > > > + kasan_set_track(&get_alloc_info(cache, object)->alloc_track, flags);
> > > > > > > >
> > > > > > > > return set_tag(object, tag);
> > > > > > > > }
> > > > > > > > diff --git a/mm/kasan/generic.c b/mm/kasan/generic.c
> > > > > > > > index 3372bdcaf92a..763d8a13e0ac 100644
> > > > > > > > --- a/mm/kasan/generic.c
> > > > > > > > +++ b/mm/kasan/generic.c
> > > > > > > > @@ -344,3 +344,21 @@ void kasan_record_aux_stack(void *addr)
> > > > > > > > alloc_info->aux_stack[1] = alloc_info->aux_stack[0];
> > > > > > > > alloc_info->aux_stack[0] = kasan_save_stack(GFP_NOWAIT);
> > > > > > > > }
> > > > > > > > +
> > > > > > > > +void kasan_set_free_info(struct kmem_cache *cache,
> > > > > > > > + void *object, u8 tag)
> > > > > > > > +{
> > > > > > > > + struct kasan_free_meta *free_meta;
> > > > > > > > +
> > > > > > > > + free_meta = get_free_info(cache, object);
> > > > > > > > + kasan_set_track(&free_meta->free_track, GFP_NOWAIT);
> > > > > > > > +}
> > > > > > > > +
> > > > > > > > +struct kasan_track *kasan_get_free_track(struct kmem_cache *cache,
> > > > > > > > + void *object, u8 tag)
> > > > > > > > +{
> > > > > > > > + struct kasan_free_meta *free_meta;
> > > > > > > > +
> > > > > > > > + free_meta = get_free_info(cache, object);
> > > > > > > > + return &free_meta->free_track;
> > > > > > > > +}
> > > > > > > > diff --git a/mm/kasan/kasan.h b/mm/kasan/kasan.h
> > > > > > > > index a7391bc83070..ad897ec36545 100644
> > > > > > > > --- a/mm/kasan/kasan.h
> > > > > > > > +++ b/mm/kasan/kasan.h
> > > > > > > > @@ -127,6 +127,9 @@ struct kasan_free_meta {
> > > > > > > > * Otherwise it might be used for the allocator freelist.
> > > > > > > > */
> > > > > > > > struct qlist_node quarantine_link;
> > > > > > > > +#ifdef CONFIG_KASAN_GENERIC
> > > > > > > > + struct kasan_track free_track;
> > > > > > > > +#endif
> > > > > > > > };
> > > > > > > >
> > > > > > > > struct kasan_alloc_meta *get_alloc_info(struct kmem_cache *cache,
> > > > > > > > @@ -168,6 +171,10 @@ void kasan_report_invalid_free(void *object, unsigned long ip);
> > > > > > > > struct page *kasan_addr_to_page(const void *addr);
> > > > > > > >
> > > > > > > > depot_stack_handle_t kasan_save_stack(gfp_t flags);
> > > > > > > > +void kasan_set_track(struct kasan_track *track, gfp_t flags);
> > > > > > > > +void kasan_set_free_info(struct kmem_cache *cache, void *object, u8 tag);
> > > > > > > > +struct kasan_track *kasan_get_free_track(struct kmem_cache *cache,
> > > > > > > > + void *object, u8 tag);
> > > > > > > >
> > > > > > > > #if defined(CONFIG_KASAN_GENERIC) && \
> > > > > > > > (defined(CONFIG_SLAB) || defined(CONFIG_SLUB))
> > > > > > > > diff --git a/mm/kasan/report.c b/mm/kasan/report.c
> > > > > > > > index 6f8f2bf8f53b..96d2657fe70f 100644
> > > > > > > > --- a/mm/kasan/report.c
> > > > > > > > +++ b/mm/kasan/report.c
> > > > > > > > @@ -159,26 +159,6 @@ static void describe_object_addr(struct kmem_cache *cache, void *object,
> > > > > > > > (void *)(object_addr + cache->object_size));
> > > > > > > > }
> > > > > > > >
> > > > > > > > -static 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 = get_alloc_info(cache, object);
> > > > > > > > -
> > > > > > > > -#ifdef CONFIG_KASAN_SW_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];
> > > > > > > > -}
> > > > > > > > -
> > > > > > > > #ifdef CONFIG_KASAN_GENERIC
> > > > > > > > static void print_stack(depot_stack_handle_t stack)
> > > > > > > > {
> > > > > > > > diff --git a/mm/kasan/tags.c b/mm/kasan/tags.c
> > > > > > > > index 25b7734e7013..201dee5d6ae0 100644
> > > > > > > > --- a/mm/kasan/tags.c
> > > > > > > > +++ b/mm/kasan/tags.c
> > > > > > > > @@ -162,3 +162,40 @@ void __hwasan_tag_memory(unsigned long addr, u8 tag, unsigned long size)
> > > > > > > > kasan_poison_shadow((void *)addr, size, tag);
> > > > > > > > }
> > > > > > > > EXPORT_SYMBOL(__hwasan_tag_memory);
> > > > > > > > +
> > > > > > > > +void kasan_set_free_info(struct kmem_cache *cache,
> > > > > > > > + void *object, u8 tag)
> > > > > > > > +{
> > > > > > > > + struct kasan_alloc_meta *alloc_meta;
> > > > > > > > + u8 idx = 0;
> > > > > > > > +
> > > > > > > > + alloc_meta = get_alloc_info(cache, object);
> > > > > > > > +
> > > > > > > > +#ifdef CONFIG_KASAN_SW_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 = get_alloc_info(cache, object);
> > > > > > > > +
> > > > > > > > +#ifdef CONFIG_KASAN_SW_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];
> > > > > > > > +}
> > > > > > >
> > > > > > > Hi Walter,
> > > > > > >
> > > > > > > FTR I've uploaded this for review purposes here:
> > > > > > > https://linux-review.googlesource.com/c/linux/kernel/git/torvalds/linux/+/2458
> > > > > > >
> > > > > > > Diff from the previous version is available as:
> > > > > > > https://linux-review.googlesource.com/c/linux/kernel/git/torvalds/linux/+/2458/1..2
> > > > > > >
> > > > > > > I've tested this locally and with syzkaller. This is :
> > > > > > >
> > > > > > > [ 80.583021][ C3] Freed by task 0:
> > > > > > > [ 80.583480][ C3] kasan_save_stack+0x1b/0x40 mm/kasan/common.c:49
> > > > > > > [ 80.584056][ C3] kasan_set_track+0x1c/0x30 mm/kasan/common.c:57
> > > > > > > [ 80.584617][ C3] kasan_set_free_info+0x1b/0x30 mm/kasan/generic.c:354
> > > > > > > [ 80.585221][ C3] __kasan_slab_free+0xd8/0x120 mm/kasan/common.c:438
> > > > > > > [ 80.585814][ C3] __cache_free mm/slab.c:3426 [inline]
> > > > > > > [ 80.585814][ C3] kfree+0x10b/0x2b0 mm/slab.c:3757
> > > > > > > [ 80.586291][ C3] kasan_rcu_reclaim+0x16/0x43 [test_kasan]
> > > > > > > [ 80.587009][ C3] rcu_do_batch kernel/rcu/tree.c:2207 [inline]
> > > > > > > [ 80.587009][ C3] rcu_core+0x59f/0x1370 kernel/rcu/tree.c:2434
> > > > > > > [ 80.587537][ C3] __do_softirq+0x26c/0x9fa kernel/softirq.c:292
> > > > > > > [ 80.588085][ C3]
> > > > > > > [ 80.588367][ C3] Last one call_rcu() call stack:
> > > > > > > [ 80.589052][ C3] kasan_save_stack+0x1b/0x40 mm/kasan/common.c:49
> > > > > > > [ 80.589622][ C3] kasan_record_aux_stack+0x82/0xb0 mm/kasan/generic.c:345
> > > > > > > [ 80.590254][ C3] __call_rcu kernel/rcu/tree.c:2672 [inline]
> > > > > > > [ 80.590254][ C3] call_rcu+0x14f/0x7f0 kernel/rcu/tree.c:2746
> > > > > > > [ 80.590782][ C3] kasan_rcu_uaf+0xe4/0xeb [test_kasan]
> > > > > > > [ 80.591697][ C3] kmalloc_tests_init+0xbc/0x1097 [test_kasan]
> > > > > > > [ 80.592900][ C3] do_one_initcall+0x10a/0x7d0 init/main.c:1196
> > > > > > > [ 80.593494][ C3] do_init_module+0x1e6/0x6d0 kernel/module.c:3539
> > > > > > > [ 80.594066][ C3] load_module+0x7464/0x9450 kernel/module.c:3890
> > > > > > > [ 80.594626][ C3] __do_sys_init_module+0x1e3/0x220 kernel/module.c:3953
> > > > > > > [ 80.595265][ C3] do_syscall_64+0xf6/0x7d0 arch/x86/entry/common.c:295
> > > > > > > [ 80.595822][ C3] entry_SYSCALL_64_after_hwframe+0x49/0xb3
> > > > > > >
> > > > > > >
> > > > > > > Overall this looks very good to me.
> > > > > > > But there is one aspect that bothers me. In the previous patch you had
> > > > > > > code that returned NULL from kasan_get_free_track() if the object is
> > > > > > > live (which means free meta is not available, it's occupied by object
> > > > > > > data). Now you dropped that code, but I think we still need it.
> > > > > > > Otherwise we cast user object data to free meta and print the free
> > > > > > > stack/pid from whatever garbage is there. This may lead to very
> > > > > > > confusing output and potentially to crashes in stackdepot.
> > > > > > >
> > > > > >
> > > > > > Yes, I totally agree with you. In the previous email I thought that
> > > > > > there is a problem with free track, but I did not point it out. Thank
> > > > > > you for pointing this problem. As you mentioned, we should fix it.
> > > > > >
> > > > > > > What do you think about this patch on top of your patches?
> > > > > > > https://linux-review.googlesource.com/c/linux/kernel/git/torvalds/linux/+/2478
> > > > > > > This way we very precisely mark the period of time when the object has
> > > > > > > free track live and set.
> > > > > > > If it looks good to you, feel free to incorporate it into your series.
> > > > > > >
> > > > > >
> > > > > > Thank you for providing good idea solution.
> > > > > >
> > > > > > I saw this patch, that is a great patch. I think it can fix the issue
> > > > > > which has garbage stack. it should work as described below.
> > > > > >
> > > > > > 1). When object is live, then don't print free stack.
> > > > > > 2). When object is NOT alive, after free object:
> > > > > > 2a). when object is in quarantine, then it can print free stack
> > > > > > 2b). when object is NOT in quarantine, then it can NOT print free stack.
> > > > > >
> > > > > > I have a question about 2), why we don't directly use
> > > > > > KASAN_KMALLOC_FREE? if we directly use it, then 2b) can print free
> > > > > > stack? 2b) may has use-after-free? so that it may need free stack.
> > > > >
> > >
> > > About 2b), I see another question. When do qlink_free(), it will be
> > > written KASAN_KMALLOC_FREE from KASAN_KMALLOC_FREETRACK? if we don't
> > > write shadow memory, it is still KASAN_KMALLOC_FREETRACK, then 2b) will
> > > have free stack? Because I see you add KASAN_KMALLOC_FREETRACK to get
> > > use-after-free in get_shadow_bug_type(). so should it not write
> > > KASAN_KMALLOC_FREE?
> >
> > It may or may not work.
> > The potential problem is that when qlink_free calls ___cache_free,
> > slab/slub may start using object memory for its own purposes, e.g.
> > store the next link. This next link may overwrite part of free meta.
> > It actually may work because the slab/slib next link is likely to
> > overlap with kasan_free_meta.quarantine_link only. And we may have
> > kasan_free_meta.free_track intact while KASAN_KMALLOC_FREE is set. But
> > this needs careful checking for both slab and slub and if they may use
> > more than 1 word in some configurations.
> >
>
> This problem looks like existing, even without this change? currently
> KASAN may get wrong free stack?

No, we should not have this problem now. Currently free track is
stored in alloc meta. Alloc meta does not overlap with the object.
It's only free meta that overlaps with the object and slab metadata at
different periods of the block lifetime. Schematically what we have
is:

struct block_t {
alloc_meta kasan_alloc_meta;
union {
user_data char[N];
slab_meta slab_meta;
free_meta kasan_free_meta;
};
}

free_meta shared storage space with 2 other things.

> Regardless of whether the shadow memory content is
> KASAN_KMALLOC_FREETRACK or KASAN_KMALLOC_FREE, it may have this problem?

KASAN_KMALLOC_FREETRACK is set only when nobody else uses the storage.

> But because of kasan_get_free_track() have conditions to get free track,
> so that if shadow memory content is KASAN_KMALLOC_FREE, then it will
> avoid this problem and always print right free stack.



> > > > > We can't use KASAN_KMALLOC_FREE because of this part:
> > > > >
> > > > > static bool __kasan_slab_free(struct kmem_cache *cache, void *object,
> > > > > unsigned long ip, bool quarantine)
> > > > > {
> > > > > ...
> > > > > kasan_poison_shadow(object, rounded_up_size, KASAN_KMALLOC_FREE);
> > > > >
> > > > > if ((IS_ENABLED(CONFIG_KASAN_GENERIC) && !quarantine) ||
> > > > > unlikely(!(cache->flags & SLAB_KASAN)))
> > > > > return false;
> > > > >
> > > > > kasan_set_free_info(cache, object, tag);
> > > > > ...
> > > > >
> > > >
> > > > Ok, I see. When return false, then the shadow memory content has
> > > > KASAN_KMALLOC_FREE, but it doesn't set free stack, so that we need to
> > > > avoid this situation. Thank for you reminder.
> > > >
> > > > >
> > > > > We may set KASAN_KMALLOC_FREE, but not set the track (or even have
> > > > > memory for the track!).
> > > > > The object may not have free meta allocated at all, e.g. very large
> > > > > object with ctor (no place to store meta), or it may be in a mempool:
> > > > > https://elixir.bootlin.com/linux/v5.7-rc6/source/mm/mempool.c#L109
> > > > > and mempool may be using the object memory itself (for its own next
> > > > > link or something).
> > > > >
> > > > > KASAN_KMALLOC_FREETRACK very explicitly tracks the exact condition we
> > > > > want: we have meta info live now and we have free track set.