[PATCH] slab: store tagged freelist for off-slab slabmgmt

[Qian Cai]

The commit 51dedad06b5f ("kasan, slab: make freelist stored without
tags") calls kasan_reset_tag() for off-slab slab management object
leading to freelist being stored non-tagged. However, cache_grow_begin()
-> alloc_slabmgmt() -> kmem_cache_alloc_node() which assigns a tag for
the address and stores in the shadow address. As the result, it causes
endless errors below during boot due to drain_freelist() ->
slab_destroy() -> kasan_slab_free() which compares already untagged
freelist against the stored tag in the shadow address. Since off-slab
slab management object freelist is such a special case, so just store it
tagged. Non-off-slab management object freelist is still stored untagged
which has not been assigned a tag and should not cause any other
troubles with this inconsistency.

BUG: KASAN: double-free or invalid-free in slab_destroy+0x84/0x88
Pointer tag: [ff], memory tag: [99]

CPU: 0 PID: 1376 Comm: kworker/0:4 Tainted: G W
5.1.0-rc3+ #8
Hardware name: HPE Apollo 70 /C01_APACHE_MB , BIOS
L50_5.13_1.0.6 07/10/2018
Workqueue: cgroup_destroy css_killed_work_fn
Call trace:
dump_backtrace+0x0/0x450
show_stack+0x20/0x2c
dump_stack+0xe0/0x16c
print_address_description+0x74/0x2a4
kasan_report_invalid_free+0x80/0xc0
__kasan_slab_free+0x204/0x208
kasan_slab_free+0xc/0x18
kmem_cache_free+0xe4/0x254
slab_destroy+0x84/0x88
drain_freelist+0xd0/0x104
__kmem_cache_shrink+0x1ac/0x224
__kmemcg_cache_deactivate+0x1c/0x28
memcg_deactivate_kmem_caches+0xa0/0xe8
memcg_offline_kmem+0x8c/0x3d4
mem_cgroup_css_offline+0x24c/0x290
css_killed_work_fn+0x154/0x618
process_one_work+0x9cc/0x183c
worker_thread+0x9b0/0xe38
kthread+0x374/0x390
ret_from_fork+0x10/0x18

Allocated by task 1625:
__kasan_kmalloc+0x168/0x240
kasan_slab_alloc+0x18/0x20
kmem_cache_alloc_node+0x1f8/0x3a0
cache_grow_begin+0x4fc/0xa24
cache_alloc_refill+0x2f8/0x3e8
kmem_cache_alloc+0x1bc/0x3bc
sock_alloc_inode+0x58/0x334
alloc_inode+0xb8/0x164
new_inode_pseudo+0x20/0xec
sock_alloc+0x74/0x284
__sock_create+0xb0/0x58c
sock_create+0x98/0xb8
__sys_socket+0x60/0x138
__arm64_sys_socket+0xa4/0x110
el0_svc_handler+0x2c0/0x47c
el0_svc+0x8/0xc

Freed by task 1625:
__kasan_slab_free+0x114/0x208
kasan_slab_free+0xc/0x18
kfree+0x1a8/0x1e0
single_release+0x7c/0x9c
close_pdeo+0x13c/0x43c
proc_reg_release+0xec/0x108
__fput+0x2f8/0x784
____fput+0x1c/0x28
task_work_run+0xc0/0x1b0
do_notify_resume+0xb44/0x1278
work_pending+0x8/0x10

The buggy address belongs to the object at ffff809681b89e00
which belongs to the cache kmalloc-128 of size 128
The buggy address is located 0 bytes inside of
128-byte region [ffff809681b89e00, ffff809681b89e80)
The buggy address belongs to the page:
page:ffff7fe025a06e00 count:1 mapcount:0 mapping:01ff80082000fb00
index:0xffff809681b8fe04
flags: 0x17ffffffc000200(slab)
raw: 017ffffffc000200 ffff7fe025a06d08 ffff7fe022ef7b88 01ff80082000fb00
raw: ffff809681b8fe04 ffff809681b80000 00000001000000e0 0000000000000000
page dumped because: kasan: bad access detected
page allocated via order 0, migratetype Unmovable, gfp_mask
0x2420c0(__GFP_IO|__GFP_FS|__GFP_NOWARN|__GFP_COMP|__GFP_THISNODE)
prep_new_page+0x4e0/0x5e0
get_page_from_freelist+0x4ce8/0x50d4
__alloc_pages_nodemask+0x738/0x38b8
cache_grow_begin+0xd8/0xa24
____cache_alloc_node+0x14c/0x268
__kmalloc+0x1c8/0x3fc
ftrace_free_mem+0x408/0x1284
ftrace_free_init_mem+0x20/0x28
kernel_init+0x24/0x548
ret_from_fork+0x10/0x18

Memory state around the buggy address:
ffff809681b89c00: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe
ffff809681b89d00: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe
>ffff809681b89e00: 99 99 99 99 99 99 99 99 fe fe fe fe fe fe fe fe
^
ffff809681b89f00: 43 43 43 43 43 fe fe fe fe fe fe fe fe fe fe fe
ffff809681b8a000: 6d fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe

Fixes: 51dedad06b5f ("kasan, slab: make freelist stored without tags")
Signed-off-by: Qian Cai <c...@lca.pw>
---
mm/slab.c | 1 -
1 file changed, 1 deletion(-)

diff --git a/mm/slab.c b/mm/slab.c
index 329bfe67f2ca..46a6e084222b 100644
--- a/mm/slab.c
+++ b/mm/slab.c
@@ -2374,7 +2374,6 @@ static void *alloc_slabmgmt(struct kmem_cache *cachep,
/* Slab management obj is off-slab. */
freelist = kmem_cache_alloc_node(cachep->freelist_cache,
local_flags, nodeid);
- freelist = kasan_reset_tag(freelist);
if (!freelist)
return NULL;
} else {
--
2.17.2 (Apple Git-113)

[Andrey Konovalov]

On Wed, Apr 3, 2019 at 4:29 AM Qian Cai <c...@lca.pw> wrote:
>
> The commit 51dedad06b5f ("kasan, slab: make freelist stored without
> tags") calls kasan_reset_tag() for off-slab slab management object
> leading to freelist being stored non-tagged. However, cache_grow_begin()
> -> alloc_slabmgmt() -> kmem_cache_alloc_node() which assigns a tag for
> the address and stores in the shadow address. As the result, it causes
> endless errors below during boot due to drain_freelist() ->
> slab_destroy() -> kasan_slab_free() which compares already untagged
> freelist against the stored tag in the shadow address. Since off-slab
> slab management object freelist is such a special case, so just store it
> tagged. Non-off-slab management object freelist is still stored untagged
> which has not been assigned a tag and should not cause any other
> troubles with this inconsistency.

Hi Qian,

Could you share the config (or other steps) you used to reproduce this?

Thanks!

[Qian Cai]

On Wed, 2019-04-03 at 13:23 +0200, Andrey Konovalov wrote:
> On Wed, Apr 3, 2019 at 4:29 AM Qian Cai <c...@lca.pw> wrote:
> >
> > The commit 51dedad06b5f ("kasan, slab: make freelist stored without
> > tags") calls kasan_reset_tag() for off-slab slab management object
> > leading to freelist being stored non-tagged. However, cache_grow_begin()
> > -> alloc_slabmgmt() -> kmem_cache_alloc_node() which assigns a tag for
> > the address and stores in the shadow address. As the result, it causes
> > endless errors below during boot due to drain_freelist() ->
> > slab_destroy() -> kasan_slab_free() which compares already untagged
> > freelist against the stored tag in the shadow address. Since off-slab
> > slab management object freelist is such a special case, so just store it
> > tagged. Non-off-slab management object freelist is still stored untagged
> > which has not been assigned a tag and should not cause any other
> > troubles with this inconsistency.
>
> Hi Qian,
>
> Could you share the config (or other steps) you used to reproduce this?

https://git.sr.ht/~cai/linux-debug/blob/master/config

Additional command-line option to boot:

page_poison=on crashkernel=768M earlycon page_owner=on numa_balancing=enable
systemd.unified_cgroup_hierarchy=1 debug_guardpage_minorder=1

[Andrey Konovalov]

Reproduced, thanks!

As far as my understanding of how SLAB works goes, this change looks good to me.

Reviewed-by: Andrey Konovalov <andre...@google.com>

Thanks, Qian!

>
> --
> You received this message because you are subscribed to the Google Groups "kasan-dev" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to kasan-dev+...@googlegroups.com.
> To post to this group, send email to kasa...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/kasan-dev/1554296870.26196.32.camel%40lca.pw.
> For more options, visit https://groups.google.com/d/optout.