On 07/29/2016 08:32 PM, Andrey Ryabinin wrote:
> 
> 
> On 07/29/2016 06:49 PM, Dave Jones wrote:
>> On Fri, Jul 29, 2016 at 06:21:12PM +0300, Andrey Ryabinin wrote:
>>  > 2016-07-29 18:19 GMT+03:00 Dave Jones <[email protected]>:
>>  > > On Fri, Jul 29, 2016 at 11:05:14AM -0400, Dave Jones wrote:
>>  > >  > I've just gotten back into running trinity on daily pulls of master, 
>> and it seems pretty horrific
>>  > >  > right now.  I can reproduce some kind of memory corruption within a 
>> couple minutes runtime.
>>  > >  >
>>  > >  > ,,,
>>  > >  >
>>  > >  > I'll work on narrowing down the exact syscalls needed to trigger 
>> this.
>>  > >
>>  > > Even limiting it to do just a simple syscall like execve (which fails 
>> most the time in trinity)
>>  > > triggers it, suggesting it's not syscall related, but the fact that 
>> trinity is forking/killing
>>  > > tons of processes at high rate is stressing something more fundamental.
>>  > >
>>  > > Given how easy this reproduces, I'll see if bisecting gives up 
>> something useful.
>>  > 
>>  > I suspect this is false positives due to changes in KASAN.
>>  > Bisection probably will point to
>>  > 80a9201a5965f4715d5c09790862e0df84ce0614 ("mm, kasan: switch SLUB to
>>  > stackdepot, enable memory quarantine for SLUB)"
>>
>> good call. reverting that changeset seems to have solved it.
>>
> 
> Unfortunately, I wasn't able to reproduce it.
> 
> Could you please try with this?
> 
> ---
>  mm/kasan/kasan.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/mm/kasan/kasan.c b/mm/kasan/kasan.c
> index b6f99e8..bf25340 100644
> --- a/mm/kasan/kasan.c
> +++ b/mm/kasan/kasan.c
> @@ -543,8 +543,8 @@ bool kasan_slab_free(struct kmem_cache *cache, void 
> *object)
>               switch (alloc_info->state) {
>               case KASAN_STATE_ALLOC:
>                       alloc_info->state = KASAN_STATE_QUARANTINE;
> -                     quarantine_put(free_info, cache);
>                       set_track(&free_info->track, GFP_NOWAIT);
> +                     quarantine_put(free_info, cache);
>                       kasan_poison_slab_free(cache, object);
>                       return true;
>               case KASAN_STATE_QUARANTINE:
> 

Actually, this is not quite right, it should be like this:

---
 mm/kasan/kasan.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/kasan/kasan.c b/mm/kasan/kasan.c
index b6f99e8..3019cec 100644
--- a/mm/kasan/kasan.c
+++ b/mm/kasan/kasan.c
@@ -543,9 +543,9 @@ bool kasan_slab_free(struct kmem_cache *cache, void *object)
                switch (alloc_info->state) {
                case KASAN_STATE_ALLOC:
                        alloc_info->state = KASAN_STATE_QUARANTINE;
-                       quarantine_put(free_info, cache);
                        set_track(&free_info->track, GFP_NOWAIT);
                        kasan_poison_slab_free(cache, object);
+                       quarantine_put(free_info, cache);
                        return true;
                case KASAN_STATE_QUARANTINE:
                case KASAN_STATE_FREE:
-- 
2.7.3

Reply via email to