On 7/29/24 8:46 PM, Chris Mason wrote:
>
>
> On 7/29/24 2:19 PM, Rik van Riel wrote:
>> In 782f8906f805 the freeing of kfence objects was moved from deep
>> inside do_free_slab to the wrapper functions outside. This is a nice
>> change, but unfortunately it missed one spot in __kmem_cache_free_bulk.
>>
>> This results in a crash like this:
>>
>> BUG skbuff_head_cache (Tainted: G S B E ): Padding overwritten. 0xffff88907fea0f00-0xffff88907fea0fff @offset=3840
>>
>> slab_err (mm/slub.c:1129)
>> free_to_partial_list (mm/slub.c:? mm/slub.c:4036)
>> slab_pad_check (mm/slub.c:864 mm/slub.c:1290)
>> check_slab (mm/slub.c:?)
>> free_to_partial_list (mm/slub.c:3171 mm/slub.c:4036)
>> kmem_cache_alloc_bulk (mm/slub.c:? mm/slub.c:4495 mm/slub.c:4586 mm/slub.c:4635)
>> napi_build_skb (net/core/skbuff.c:348 net/core/skbuff.c:527 net/core/skbuff.c:549)
>>
>> All the other callers to do_free_slab appear to be ok.
changed do_free_slab to do_slab_free in two places.
>>
>> Add a kfence_free check in __kmem_cache_free_bulk to avoid the crash.
>>
>> Reported-by: Chris Mason <clm@xxxxxxxx>
>> Fixes: 782f8906f805 ("mm/slub: free KFENCE objects in slab_free_hook()")
>> Cc: stable@xxxxxxxxxx
>> Signed-off-by: Rik van Riel <riel@xxxxxxxxxxx>
>
> We found this after bisecting a slab corruption down to the kfence
> patch, and with this patch applied we're no longer falling over. So
> thanks Rik!
Indeed thanks and sorry for the trouble! Given that
__kmem_cache_free_bulk is currently only used to unwind a
kmem_cache_bulk_alloc() that runs out of memory in the middle of the
operation, I'm surprised you saw this happen reliably enough to bisect it.
Added to slab/for-6.11-rc1/fixes
> -chris
