Re: [PATCH 13/17] btrfs: extent_io: only require sector size alignment for page read

From: Qu Wenruo <wqu@suse.com>
To: Nikolay Borisov <nborisov@suse.com>, linux-btrfs@vger.kernel.org
Subject: Re: [PATCH 13/17] btrfs: extent_io: only require sector size alignment for page read
Date: Tue, 15 Sep 2020 09:54:48 +0800	[thread overview]
Message-ID: <6a515ae8-7357-7383-9501-e876b905248f@suse.com> (raw)
In-Reply-To: <89fd9545-c539-3f58-e48a-218a9b111edd@suse.com>

On 2020/9/11 下午9:55, Nikolay Borisov wrote:
> 
> 
> On 8.09.20 г. 10:52 ч., Qu Wenruo wrote:
>> If we're reading partial page, btrfs will warn about this as our
>> read/write are always done in sector size, which equals page size.
>>
>> But for the incoming subpage RO support, our data read is only aligned
>> to sectorsize, which can be smaller than page size.
>>
>> Thus here we change the warning condition to check it against
>> sectorsize, thus the behavior is not changed for regular sectorsize ==
>> PAGE_SIZE case, while won't report error for subpage read.
>>
>> Also, pass the proper start/end with bv_offset for check_data_csum() to
>> handle.
>>
>> Signed-off-by: Qu Wenruo <wqu@suse.com>
>> ---
>>  fs/btrfs/extent_io.c | 19 ++++++++++++-------
>>  1 file changed, 12 insertions(+), 7 deletions(-)
>>
>> diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
>> index 81e43d99feda..a83b63ecc5f8 100644
>> --- a/fs/btrfs/extent_io.c
>> +++ b/fs/btrfs/extent_io.c
>> @@ -2819,6 +2819,7 @@ static void end_bio_extent_readpage(struct bio *bio)
>>  		struct page *page = bvec->bv_page;
>>  		struct inode *inode = page->mapping->host;
>>  		struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
>> +		u32 sectorsize = fs_info->sectorsize;
>>  		bool data_inode = btrfs_ino(BTRFS_I(inode))
>>  			!= BTRFS_BTREE_INODE_OBJECTID;
>>  
>> @@ -2829,13 +2830,17 @@ static void end_bio_extent_readpage(struct bio *bio)
>>  		tree = &BTRFS_I(inode)->io_tree;
>>  		failure_tree = &BTRFS_I(inode)->io_failure_tree;
>>  
>> -		/* We always issue full-page reads, but if some block
>> +		/*
>> +		 * We always issue full-sector reads, but if some block
>>  		 * in a page fails to read, blk_update_request() will
>>  		 * advance bv_offset and adjust bv_len to compensate.
>> -		 * Print a warning for nonzero offsets, and an error
>> -		 * if they don't add up to a full page.  */
>> -		if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
>> -			if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
>> +		 * Print a warning for unaligned offsets, and an error
>> +		 * if they don't add up to a full sector.
>> +		 */
>> +		if (!IS_ALIGNED(bvec->bv_offset, sectorsize) ||
>> +		    !IS_ALIGNED(bvec->bv_offset + bvec->bv_len, sectorsize)) {
>> +			if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len,
>> +					sectorsize))
> 
> Duplicated check ...

BTW, this is not duplicated, it's to distinguish two different error
patterns...
One for read request which doesn't end at sector boundary, and the other
one for which doesn't start at sector boundary.

> 
>>  				btrfs_err(fs_info,
>>  					"partial page read in btrfs with offset %u and length %u",
>>  					bvec->bv_offset, bvec->bv_len);
>> @@ -2845,8 +2850,8 @@ static void end_bio_extent_readpage(struct bio *bio)
>>  					bvec->bv_offset, bvec->bv_len);
>>  		}
>>  
>> -		start = page_offset(page);
>> -		end = start + bvec->bv_offset + bvec->bv_len - 1;
>> +		start = page_offset(page) + bvec->bv_offset;
>> +		end = start + bvec->bv_len - 1;
> 
> nit: 'start' and 'end' must really be renamed - to file_offset and
> file_end because they represent values in the logical namespace of the
> file. And given the context they are used i.e endio handler where we
> also deal with extent starts and physical offsets such a rename is long
> over due. Perhaps you can create a separate patch when  you are
> resending the series alternatively I'll make a sweep across those
> low-level functions to clean that up.

I guess we could do that in another patchset.

The naming is really aweful, but there are tons of other similar
situations across the code base.

It may be a big batch of work to properly unify the naming.

And the naming itself will take some time to mature.

We have a lot of different terms which share the similar meanings but
still slightly different:
- bytenr
  btrfs logical bytenr

- file_offset
  the offset inside a file

And in this particular case, for btree inode, bytenr == file_offset,
which may make things more complex.

While for regualr file inodes, file_offset is different from the extent
bytenr.

So we really need to come out with a proper term table for this...

Thanks,
Qu

> 
>>  		len = bvec->bv_len;
>>  
>>  		mirror = io_bio->mirror_num;
>>
> 