Re: [f2fs-dev] [PATCH] f2fs: fix livelock in swapfile writes

From: Chao Yu
Date: Thu Aug 01 2019 - 04:05:09 EST


On 2019/8/1 12:26, Jaegeuk Kim wrote:
> On 08/01, Chao Yu wrote:
>> On 2019/8/1 4:43, Jaegeuk Kim wrote:
>>> This patch fixes livelock in the below call path when writing swap pages.
>>>
>>> [46374.617256] c2 701 __switch_to+0xe4/0x100
>>> [46374.617265] c2 701 __schedule+0x80c/0xbc4
>>> [46374.617273] c2 701 schedule+0x74/0x98
>>> [46374.617281] c2 701 rwsem_down_read_failed+0x190/0x234
>>> [46374.617291] c2 701 down_read+0x58/0x5c
>>> [46374.617300] c2 701 f2fs_map_blocks+0x138/0x9a8
>>> [46374.617310] c2 701 get_data_block_dio_write+0x74/0x104
>>> [46374.617320] c2 701 __blockdev_direct_IO+0x1350/0x3930
>>> [46374.617331] c2 701 f2fs_direct_IO+0x55c/0x8bc
>>> [46374.617341] c2 701 __swap_writepage+0x1d0/0x3e8
>>> [46374.617351] c2 701 swap_writepage+0x44/0x54
>>> [46374.617360] c2 701 shrink_page_list+0x140/0xe80
>>> [46374.617371] c2 701 shrink_inactive_list+0x510/0x918
>>> [46374.617381] c2 701 shrink_node_memcg+0x2d4/0x804
>>> [46374.617391] c2 701 shrink_node+0x10c/0x2f8
>>> [46374.617400] c2 701 do_try_to_free_pages+0x178/0x38c
>>> [46374.617410] c2 701 try_to_free_pages+0x348/0x4b8
>>> [46374.617419] c2 701 __alloc_pages_nodemask+0x7f8/0x1014
>>> [46374.617429] c2 701 pagecache_get_page+0x184/0x2cc
>>> [46374.617438] c2 701 f2fs_new_node_page+0x60/0x41c
>>> [46374.617449] c2 701 f2fs_new_inode_page+0x50/0x7c
>>> [46374.617460] c2 701 f2fs_init_inode_metadata+0x128/0x530
>>> [46374.617472] c2 701 f2fs_add_inline_entry+0x138/0xd64
>>> [46374.617480] c2 701 f2fs_do_add_link+0xf4/0x178
>>> [46374.617488] c2 701 f2fs_create+0x1e4/0x3ac
>>> [46374.617497] c2 701 path_openat+0xdc0/0x1308
>>> [46374.617507] c2 701 do_filp_open+0x78/0x124
>>> [46374.617516] c2 701 do_sys_open+0x134/0x248
>>> [46374.617525] c2 701 SyS_openat+0x14/0x20
>>>
>>> Signed-off-by: Jaegeuk Kim <jaegeuk@xxxxxxxxxx>
>>> ---
>>> fs/f2fs/data.c | 2 +-
>>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
>>> index abbf14e9bd72..f49f243fd54f 100644
>>> --- a/fs/f2fs/data.c
>>> +++ b/fs/f2fs/data.c
>>> @@ -1372,7 +1372,7 @@ static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
>>> return __get_data_block(inode, iblock, bh_result, create,
>>> F2FS_GET_BLOCK_DIO, NULL,
>>> f2fs_rw_hint_to_seg_type(inode->i_write_hint),
>>> - true);
>>> + IS_SWAPFILE(inode) ? false : true);
>>
>> I suspect that we should use node_change for swapfile rather than just changing
>> may_write field to skip lock.
>
> Swap write should not change the node page.

You're right.

Reviewed-by: Chao Yu <yuchao0@xxxxxxxxxx>

Thanks,

>
>>
>> __do_map_lock()
>> if (flag == F2FS_GET_BLOCK_PRE_AIO || IS_SWAPFILE(inode)) {
>> ...
>> } else {
>> ...
>> }
>>
>> Thanks,
>>
>>
>>> }
>>>
>>> static int get_data_block_dio(struct inode *inode, sector_t iblock,
>>>
> .
>