Re: [PATCH] f2fs: fix to relocate f2fs_balance_fs() in mkwrite()

[Chao Yu]

Hi Jaeguek,

On 2019/12/6 12:08, Jaegeuk Kim wrote:
> Hi Chao,
> 
> I was testing this.
> 
> https://github.com/jaegeuk/f2fs/commit/76be33b9f1fce70dd2d3f04f66d0f78b418fe3f5

The patch looks good.

BTW, do you mind adding below call stack into your patch? I guess it
describes this ABBA deadlock with more details. :)

Thanks,

> 
> On 12/06, Chao Yu wrote:
>> As Dinosaur Huang reported, there is a potential deadlock in between
>> GC and mkwrite():
>>
>> Thread A			Thread B
>> - do_page_mkwrite
>>  - f2fs_vm_page_mkwrite
>>   - lock_page
>> 				- f2fs_balance_fs
>>                                  - mutex_lock(gc_mutex)
>> 				 - f2fs_gc
>> 				  - do_garbage_collect
>> 				   - ra_data_block
>> 				    - grab_cache_page
>>   - f2fs_balance_fs
>>    - mutex_lock(gc_mutex)
>>
>> In order to fix this, we just move f2fs_balance_fs() out of page lock's
>> coverage in f2fs_vm_page_mkwrite().
>>
>> Reported-by: Dinosaur Huang <dinosaur.huang@xxxxxxxxxx>
>> Signed-off-by: Chao Yu <yuchao0@xxxxxxxxxx>
>> ---
>>  fs/f2fs/file.c | 4 ++--
>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
>> index c0560d62dbee..ed3290225506 100644
>> --- a/fs/f2fs/file.c
>> +++ b/fs/f2fs/file.c
>> @@ -67,6 +67,8 @@ static vm_fault_t f2fs_vm_page_mkwrite(struct vm_fault *vmf)
>>  
>>  	f2fs_bug_on(sbi, f2fs_has_inline_data(inode));
>>  
>> +	f2fs_balance_fs(sbi, true);
>> +
>>  	file_update_time(vmf->vma->vm_file);
>>  	down_read(&F2FS_I(inode)->i_mmap_sem);
>>  	lock_page(page);
>> @@ -120,8 +122,6 @@ static vm_fault_t f2fs_vm_page_mkwrite(struct vm_fault *vmf)
>>  out_sem:
>>  	up_read(&F2FS_I(inode)->i_mmap_sem);
>>  
>> -	f2fs_balance_fs(sbi, dn.node_changed);
>> -
>>  	sb_end_pagefault(inode->i_sb);
>>  err:
>>  	return block_page_mkwrite_return(err);
>> -- 
>> 2.18.0.rc1
> .
>