Re: list corruption in deferred_split_scan()

[Qian Cai <cai@lca.pw>]

> On Aug 5, 2019, at 6:15 PM, Yang Shi <yang.shi@linux.alibaba.com> wrote:
> 
> 
> 
> On 7/25/19 2:46 PM, Yang Shi wrote:
>> 
>> 
>> On 7/24/19 2:13 PM, Qian Cai wrote:
>>> On Wed, 2019-07-10 at 17:43 -0400, Qian Cai wrote:
>>>> Running LTP oom01 test case with swap triggers a crash below. Revert the
>>>> series
>>>> "Make deferred split shrinker memcg aware" [1] seems fix the issue.
>>> You might want to look harder on this commit, as reverted it alone on the top of
>>>   5.2.0-next-20190711 fixed the issue.
>>> 
>>> aefde94195ca mm: thp: make deferred split shrinker memcg aware [1]
>>> 
>>> [1] https://lore.kernel.org/linux-mm/1561507361-59349-5-git-send-email-yang.shi@
>>> linux.alibaba.com/
>> 
>> This is the real meat of the patch series, which converted to memcg deferred split queue actually.
>> 
>>> 
>>> 
>>> list_del corruption. prev->next should be ffffea0022b10098, but was
>>> 0000000000000000
>> 
>> Finally I could reproduce the list corruption issue on my machine with THP swap (swap device is fast device). I should checked this with you at the first place. The problem can't be reproduced with rotate swap device. So, I'm supposed you were using THP swap too.
>> 
>> Actually, I found two issues with THP swap:
>> 1. free_transhuge_page() is called in reclaim path instead of put_page. The mem_cgroup_uncharge() is called before free_transhuge_page() in reclaim path, which causes page->mem_cgroup is NULL so the wrong deferred_split_queue would be used, so the THP was not deleted from the memcg's list at all. Then the page might be split or reused later, page->mapping would be override.
>> 
>> 2. There is a race condition caused by try_to_unmap() with THP swap. The try_to_unmap() just calls page_remove_rmap() to add THP to deferred split queue in reclaim path. This might cause the below race condition to corrupt the list:
>> 
>>                   A                                      B
>> deferred_split_scan
>>     list_move
>>                                                try_to_unmap
>> list_add_tail
>> 
>> list_splice <-- The list might get corrupted here
>> 
>>                                                free_transhuge_page
>>                                                       list_del <-- kernel bug triggered
>> 
>> I hope the below patch would solve your problem (tested locally).
> 
> Hi Qian,
> 
> Did the below patch solve your problem? I would like the fold the fix into the series then target to 5.4 release.

It is going to take a while before I would be able to access that system again. Since you can reproduce this and
test yourself now, I’d say go ahead posting the patch.


> 
> Thanks,
> Yang
> 
>> 
>> 
>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>> index b7f709d..d6612ec 100644
>> --- a/mm/huge_memory.c
>> +++ b/mm/huge_memory.c
>> @@ -2830,6 +2830,19 @@ void deferred_split_huge_page(struct page *page)
>> 
>>         VM_BUG_ON_PAGE(!PageTransHuge(page), page);
>> 
>> +       /*
>> +        * The try_to_unmap() in page reclaim path might reach here too,
>> +        * this may cause a race condition to corrupt deferred split queue.
>> +        * And, if page reclaim is already handling the same page, it is
>> +        * unnecessary to handle it again in shrinker.
>> +        *
>> +        * Check PageSwapCache to determine if the page is being
>> +        * handled by page reclaim since THP swap would add the page into
>> +        * swap cache before reaching try_to_unmap().
>> +        */
>> +       if (PageSwapCache(page))
>> +               return;
>> +
>>         spin_lock_irqsave(&ds_queue->split_queue_lock, flags);
>>         if (list_empty(page_deferred_list(page))) {
>>                 count_vm_event(THP_DEFERRED_SPLIT_PAGE);
>> diff --git a/mm/vmscan.c b/mm/vmscan.c
>> index a0301ed..40c684a 100644
>> --- a/mm/vmscan.c
>> +++ b/mm/vmscan.c
>> @@ -1485,10 +1485,9 @@ static unsigned long shrink_page_list(struct list_head *page_list,
>>                  * Is there need to periodically free_page_list? It would
>>                  * appear not as the counts should be low
>>                  */
>> -               if (unlikely(PageTransHuge(page))) {
>> -                       mem_cgroup_uncharge(page);
>> +               if (unlikely(PageTransHuge(page)))
>>                         (*get_compound_page_dtor(page))(page);
>> -               } else
>> +               else
>>                         list_add(&page->lru, &free_pages);
>>                 continue;
>> 
>> @@ -1909,7 +1908,6 @@ static unsigned noinline_for_stack move_pages_to_lru(struct lruvec *lruvec,
>> 
>>                         if (unlikely(PageCompound(page))) {
>> spin_unlock_irq(&pgdat->lru_lock);
>> -                               mem_cgroup_uncharge(page);
>> (*get_compound_page_dtor(page))(page);
>> spin_lock_irq(&pgdat->lru_lock);
>>                         } else
>> 
>>> [  685.284254][ T3456] ------------[ cut here ]------------
>>> [  685.289616][ T3456] kernel BUG at lib/list_debug.c:53!
>>> [  685.294808][ T3456] invalid opcode: 0000 [#1] SMP DEBUG_PAGEALLOC KASAN NOPTI
>>> [  685.301998][ T3456] CPU: 5 PID: 3456 Comm: oom01 Tainted:
>>> G        W         5.2.0-next-20190711+ #3
>>> [  685.311193][ T3456] Hardware name: HPE ProLiant DL385 Gen10/ProLiant DL385
>>> Gen10, BIOS A40 06/24/2019
>>> [  685.320485][ T3456] RIP: 0010:__list_del_entry_valid+0x8b/0xb6
>>> [  685.326364][ T3456] Code: f1 e0 ff 49 8b 55 08 4c 39 e2 75 2c 5b b8 01 00 00
>>> 00 41 5c 41 5d 5d c3 4c 89 e2 48 89 de 48 c7 c7 c0 5a 73 a3 e8 d9 fa bc ff <0f>
>>> 0b 48 c7 c7 60 a0 e1 a3 e8 13 52 01 00 4c 89 e6 48 c7 c7 20 5b
>>> [  685.345956][ T3456] RSP: 0018:ffff888e0c8a73c0 EFLAGS: 00010082
>>> [  685.351920][ T3456] RAX: 0000000000000054 RBX: ffffea0022b10098 RCX:
>>> ffffffffa2d5d708
>>> [  685.359807][ T3456] RDX: 0000000000000000 RSI: 0000000000000008 RDI:
>>> ffff8888442bd380
>>> [  685.367693][ T3456] RBP: ffff888e0c8a73d8 R08: ffffed1108857a71 R09:
>>> ffffed1108857a70
>>> [  685.375577][ T3456] R10: ffffed1108857a70 R11: ffff8888442bd387 R12:
>>> 0000000000000000
>>> [  685.383462][ T3456] R13: 0000000000000000 R14: ffffea0022b10034 R15:
>>> ffffea0022b10098
>>> [  685.391348][ T3456] FS:  00007fbe26db4700(0000) GS:ffff888844280000(0000)
>>> knlGS:0000000000000000
>>> [  685.400194][ T3456] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>> [  685.406681][ T3456] CR2: 00007fbcabb3f000 CR3: 0000001012e44000 CR4:
>>> 00000000001406a0
>>> [  685.414563][ T3456] Call Trace:
>>> [  685.417736][ T3456]  deferred_split_scan+0x337/0x740
>>> [  685.422741][ T3456]  ? split_huge_page_to_list+0xe10/0xe10
>>> [  685.428272][ T3456]  ? __radix_tree_lookup+0x12d/0x1e0
>>> [  685.433453][ T3456]  ? node_tag_get.part.0.constprop.6+0x40/0x40
>>> [  685.439505][ T3456]  do_shrink_slab+0x244/0x5a0
>>> [  685.444071][ T3456]  shrink_slab+0x253/0x440
>>> [  685.448375][ T3456]  ? unregister_shrinker+0x110/0x110
>>> [  685.453551][ T3456]  ? kasan_check_read+0x11/0x20
>>> [  685.458291][ T3456]  ? mem_cgroup_protected+0x20f/0x260
>>> [  685.463555][ T3456]  shrink_node+0x31e/0xa30
>>> [  685.467858][ T3456]  ? shrink_node_memcg+0x1560/0x1560
>>> [  685.473036][ T3456]  ? ktime_get+0x93/0x110
>>> [  685.477250][ T3456]  do_try_to_free_pages+0x22f/0x820
>>> [  685.482338][ T3456]  ? shrink_node+0xa30/0xa30
>>> [  685.486815][ T3456]  ? kasan_check_read+0x11/0x20
>>> [  685.491556][ T3456]  ? check_chain_key+0x1df/0x2e0
>>> [  685.496383][ T3456]  try_to_free_pages+0x242/0x4d0
>>> [  685.501209][ T3456]  ? do_try_to_free_pages+0x820/0x820
>>> [  685.506476][ T3456]  __alloc_pages_nodemask+0x9ce/0x1bc0
>>> [  685.511826][ T3456]  ? gfp_pfmemalloc_allowed+0xc0/0xc0
>>> [  685.517089][ T3456]  ? kasan_check_read+0x11/0x20
>>> [  685.521826][ T3456]  ? check_chain_key+0x1df/0x2e0
>>> [  685.526657][ T3456]  ? do_anonymous_page+0x343/0xe30
>>> [  685.531658][ T3456]  ? lock_downgrade+0x390/0x390
>>> [  685.536399][ T3456]  ? get_kernel_page+0xa0/0xa0
>>> [  685.541050][ T3456]  ? __lru_cache_add+0x108/0x160
>>> [  685.545879][ T3456]  alloc_pages_vma+0x89/0x2c0
>>> [  685.550444][ T3456]  do_anonymous_page+0x3e1/0xe30
>>> [  685.555271][ T3456]  ? __update_load_avg_cfs_rq+0x2c/0x490
>>> [  685.560796][ T3456]  ? finish_fault+0x120/0x120
>>> [  685.565361][ T3456]  ? alloc_pages_vma+0x21e/0x2c0
>>> [  685.570187][ T3456]  handle_pte_fault+0x457/0x12c0
>>> [  685.575014][ T3456]  __handle_mm_fault+0x79a/0xa50
>>> [  685.579841][ T3456]  ? vmf_insert_mixed_mkwrite+0x20/0x20
>>> [  685.585280][ T3456]  ? kasan_check_read+0x11/0x20
>>> [  685.590021][ T3456]  ? __count_memcg_events+0x8b/0x1c0
>>> [  685.595196][ T3456]  handle_mm_fault+0x17f/0x370
>>> [  685.599850][ T3456]  __do_page_fault+0x25b/0x5d0
>>> [  685.604501][ T3456]  do_page_fault+0x4c/0x2cf
>>> [  685.608892][ T3456]  ? page_fault+0x5/0x20
>>> [  685.613019][ T3456]  page_fault+0x1b/0x20
>>> [  685.617058][ T3456] RIP: 0033:0x410be0
>>> [  685.620840][ T3456] Code: 89 de e8 e3 23 ff ff 48 83 f8 ff 0f 84 86 00 00 00
>>> 48 89 c5 41 83 fc 02 74 28 41 83 fc 03 74 62 e8 95 29 ff ff 31 d2 48 98 90 <c6>
>>> 44 15 00 07 48 01 c2 48 39 d3 7f f3 31 c0 5b 5d 41 5c c3 0f 1f
>>> [  68[  687.120156][ T3456] Shutting down cpus with NMI
>>> [  687.124731][ T3456] Kernel Offset: 0x21800000 from 0xffffffff81000000
>>> (relocation range: 0xffffffff80000000-0xffffffffbfffffff)
>>> [  687.136389][ T3456] ---[ end Kernel panic - not syncing: Fatal exception ]---
>> 
>