Re: [PATCH] mm/memory-failure: make sure wait for page writeback in memory_failure

[yangerkun <yangerkun@huawei.com>]

在 2021/5/11 16:46, Jan Kara 写道:
> On Tue 11-05-21 15:03:29, yangerkun wrote:
>> Our syzkaller trigger the "BUG_ON(!list_empty(&inode->i_wb_list))" in
>> clear_inode:
>>
>> [  292.016156] ------------[ cut here ]------------
>> [  292.017144] kernel BUG at fs/inode.c:519!
>> [  292.017860] Internal error: Oops - BUG: 0 [#1] SMP
>> [  292.018741] Dumping ftrace buffer:
>> [  292.019577]    (ftrace buffer empty)
>> [  292.020430] Modules linked in:
>> [  292.021748] Process syz-executor.0 (pid: 249, stack limit =
>> 0x00000000a12409d7)
>> [  292.023719] CPU: 1 PID: 249 Comm: syz-executor.0 Not tainted 4.19.95
>> [  292.025206] Hardware name: linux,dummy-virt (DT)
>> [  292.026176] pstate: 80000005 (Nzcv daif -PAN -UAO)
>> [  292.027244] pc : clear_inode+0x280/0x2a8
>> [  292.028045] lr : clear_inode+0x280/0x2a8
>> [  292.028877] sp : ffff8003366c7950
>> [  292.029582] x29: ffff8003366c7950 x28: 0000000000000000
>> [  292.030570] x27: ffff80032b5f4708 x26: ffff80032b5f4678
>> [  292.031863] x25: ffff80036ae6b300 x24: ffff8003689254d0
>> [  292.032902] x23: ffff80036ae69d80 x22: 0000000000033cc8
>> [  292.033928] x21: 0000000000000000 x20: ffff80032b5f47a0
>> [  292.034941] x19: ffff80032b5f4678 x18: 0000000000000000
>> [  292.035958] x17: 0000000000000000 x16: 0000000000000000
>> [  292.037102] x15: 0000000000000000 x14: 0000000000000000
>> [  292.038103] x13: 0000000000000004 x12: 0000000000000000
>> [  292.039137] x11: 1ffff00066cd8f52 x10: 1ffff00066cd8ec8
>> [  292.040216] x9 : dfff200000000000 x8 : ffff10006ac1e86a
>> [  292.041432] x7 : dfff200000000000 x6 : ffff100066cd8f1e
>> [  292.042516] x5 : dfff200000000000 x4 : ffff80032b5f47a0
>> [  292.043525] x3 : ffff200008000000 x2 : ffff200009867000
>> [  292.044560] x1 : ffff8003366bb000 x0 : 0000000000000000
>> [  292.045569] Call trace:
>> [  292.046083]  clear_inode+0x280/0x2a8
>> [  292.046828]  ext4_clear_inode+0x38/0xe8
>> [  292.047593]  ext4_free_inode+0x130/0xc68
>> [  292.048383]  ext4_evict_inode+0xb20/0xcb8
>> [  292.049162]  evict+0x1a8/0x3c0
>> [  292.049761]  iput+0x344/0x460
>> [  292.050350]  do_unlinkat+0x260/0x410
>> [  292.051042]  __arm64_sys_unlinkat+0x6c/0xc0
>> [  292.051846]  el0_svc_common+0xdc/0x3b0
>> [  292.052570]  el0_svc_handler+0xf8/0x160
>> [  292.053303]  el0_svc+0x10/0x218
>> [  292.053908] Code: 9413f4a9 d503201f f90017b6 97f4d5b1 (d4210000)
>> [  292.055471] ---[ end trace 01b339dd07795f8d ]---
>> [  292.056443] Kernel panic - not syncing: Fatal exception
>> [  292.057488] SMP: stopping secondary CPUs
>> [  292.058419] Dumping ftrace buffer:
>> [  292.059078]    (ftrace buffer empty)
>> [  292.059756] Kernel Offset: disabled
>> [  292.060443] CPU features: 0x10,a1006000
>> [  292.061195] Memory Limit: none
>> [  292.061794] Rebooting in 86400 seconds..
>>
>> Crash of this problem show that someone call __munlock_pagevec to clear
>> page LRU.
>>
>>   #0 [ffff80035f02f4c0] __switch_to at ffff20000808d020
>>   #1 [ffff80035f02f4f0] __schedule at ffff20000985102c
>>   #2 [ffff80035f02f5e0] schedule at ffff200009851d1c
>>   #3 [ffff80035f02f600] io_schedule at ffff2000098525c0
>>   #4 [ffff80035f02f620] __lock_page at ffff20000842d2d4
>>   #5 [ffff80035f02f710] __munlock_pagevec at ffff2000084c4600
>>   #6 [ffff80035f02f870] munlock_vma_pages_range at ffff2000084c5928
>>   #7 [ffff80035f02fa60] do_munmap at ffff2000084cbdf4
>>   #8 [ffff80035f02faf0] mmap_region at ffff2000084ce20c
>>   #9 [ffff80035f02fb90] do_mmap at ffff2000084cf018
>>
>> So memory_failure will call identify_page_state without
>> wait_on_page_writeback. And after generic_truncate_error_page clear the
>                                       ^^^ this seems to be
> truncate_error_page() these days... Or did you mean
> generic_error_remove_page()?

Yeah. You are right. Will change in next version!

> 
>> mapping of this page. end_page_writeback won't call
>> sb_clear_inode_writeback to clear inode->i_wb_list. That will trigger
>> BUG_ON in clear_inode!
> 
> We definitely need to wait for writeback of these pages and the change you
> suggest makes sense to me. I'm just not sure whether the only problem with
> these "pages in the process of being munlocked()" cannot confuse the state
> machinery in memory_failure() also in some other way. Also I'm not sure if
> are really allowed to call wait_on_page_writeback() on just any page that
> hits memory_failure() - there can be slab pages, anon pages, completely
> unknown pages given out by page allocator to device drivers etc. That needs
> someone more familiar with these MM details than me.

Agree with you. The two problem you suggest seems hard to me... Maybe 
someone more familiar with MM can help us...

> 
> 								Honza
> 
>>
>> Fix it by move the wait_on_page_writeback before check of LRU.
>>
>> Signed-off-by: yangerkun <yangerkun@huawei.com>
>> ---
>>   mm/memory-failure.c | 6 +++---
>>   1 file changed, 3 insertions(+), 3 deletions(-)
>>
>> diff --git a/mm/memory-failure.c b/mm/memory-failure.c
>> index bd3945446d47..9870a22800d9 100644
>> --- a/mm/memory-failure.c
>> +++ b/mm/memory-failure.c
>> @@ -1527,15 +1527,15 @@ int memory_failure(unsigned long pfn, int flags)
>>   		return 0;
>>   	}
>>   
>> -	if (!PageTransTail(p) && !PageLRU(p))
>> -		goto identify_page_state;
>> -
>>   	/*
>>   	 * It's very difficult to mess with pages currently under IO
>>   	 * and in many cases impossible, so we just avoid it here.
>>   	 */
>>   	wait_on_page_writeback(p);
>>   
>> +	if (!PageTransTail(p) && !PageLRU(p))
>> +		goto identify_page_state;
>> +
>>   	/*
>>   	 * Now take care of user space mappings.
>>   	 * Abort on fail: __delete_from_page_cache() assumes unmapped page.
>> -- 
>> 2.25.4
>>