Re: [PATCH 3/5] btrfs: extent_io: Kill the BUG_ON() in flush_write_bio()

[Nikolay Borisov <nborisov@suse.com>]

On 17.01.19 г. 10:28 ч., Qu Wenruo wrote:
> 
> 
> On 2019/1/17 下午4:22, Nikolay Borisov wrote:
>>
>>
>> On 17.01.19 г. 9:48 ч., Qu Wenruo wrote:
>>> This BUG_ON() is really just a crappy way to workaround the _must_check
>>> attribute of submit_one_bio().
>>>
>>> Now kill the BUG_ON() and allow flush_write_bio() to return error
>>> number.
>>>
>>> Also add _must_check attribute to flush_write_bio(), and modify all
>>> callers to handle the possible error returned.
>>>
>>> Signed-off-by: Qu Wenruo <wqu@suse.com>
>>> ---
>>>  fs/btrfs/extent_io.c | 64 +++++++++++++++++++++++++++++++-------------
>>>  1 file changed, 46 insertions(+), 18 deletions(-)
>>>
>>> diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
>>> index 8a2335713a2d..a773bc46badc 100644
>>> --- a/fs/btrfs/extent_io.c
>>> +++ b/fs/btrfs/extent_io.c
>>> @@ -169,15 +169,15 @@ static int __must_check submit_one_bio(struct bio *bio, int mirror_num,
>>>  	return blk_status_to_errno(ret);
>>>  }
>>>  
>>> -static void flush_write_bio(struct extent_page_data *epd)
>>> +static int __must_check flush_write_bio(struct extent_page_data *epd)
>>>  {
>>> -	if (epd->bio) {
>>> -		int ret;
>>> +	int ret = 0;
>>>  
>>> +	if (epd->bio) {
>>>  		ret = submit_one_bio(epd->bio, 0, 0);
>>> -		BUG_ON(ret < 0); /* -ENOMEM */
>>>  		epd->bio = NULL;
>>>  	}
>>> +	return ret;
>>>  }
>>>  
>>>  int __init extent_io_init(void)
>>> @@ -3509,8 +3509,10 @@ lock_extent_buffer_for_io(struct extent_buffer *eb,
>>>  	int ret = 0;
>>>  
>>>  	if (!btrfs_try_tree_write_lock(eb)) {
>>> +		ret = flush_write_bio(epd);
>>> +		if (ret < 0)
>>> +			return ret;
>>>  		flush = 1;
>>> -		flush_write_bio(epd);
>>>  		btrfs_tree_lock(eb);
>>>  	}
>>>  
>>> @@ -3519,7 +3521,9 @@ lock_extent_buffer_for_io(struct extent_buffer *eb,
>>>  		if (!epd->sync_io)
>>>  			return 0;
>>>  		if (!flush) {
>>> -			flush_write_bio(epd);
>>> +			ret = flush_write_bio(epd);
>>> +			if (ret < 0)
>>> +				return ret;
>>>  			flush = 1;
>>>  		}
>>>  		while (1) {
>>> @@ -3560,7 +3564,9 @@ lock_extent_buffer_for_io(struct extent_buffer *eb,
>>>  
>>>  		if (!trylock_page(p)) {
>>>  			if (!flush) {
>>> -				flush_write_bio(epd);
>>> +				ret = flush_write_bio(epd);
>>> +				if (ret < 0)
>>> +					return ret;
>>
>> Can't you end up with partially locked pages here? Are you sure that
>> flush_write_bio will ALWAYS be executed when i = 0? If that\'s the case
>> then I think an assert(i == 0) is in order there.
> 
> An ASSERT() indeed makes sense here.
> 
> Although I could also make it better by recording the failed page number
> and unlock those already locked pages.

Unlocking is the correct way to handle it, the assert would be there to
prove that you won't ever need the unlocking code. If it triggers then
this case must be handled.

> 
> I'm OK either way.
> 
> Thanks,
> Qu
> 
>>
>>>  				flush = 1;
>>>  			}
>>>  			lock_page(p);
>>
>> <snip>
>>