Re: [PATCH v5 2/7] btrfs: use the page-cache for super block reading

[Nikolay Borisov <nborisov@suse.com>]

On 7.02.20 г. 9:20 ч., Johannes Thumshirn wrote:
> Super-block reading in BTRFS is done using buffer_heads. Buffer_heads have
> some drawbacks, like not being able to propagate errors from the lower
> layers.
> 
> Directly use the page cache for reading the super-blocks from disk or
> invalidating an on-disk super-block. We have to use the page-cache so to
> avoid races between mkfs and udev. See also 6f60cbd3ae44 ("btrfs: access
> superblock via pagecache in scan_one_device").
> 
> Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
> 
> ---
> Changes to v4:
> - Remove mapping_gfp_constraint() and GFP_NOFAIL (hch)
> 
> Changes to v3:
> - Use read_cache_pages() and write_one_page() for IO (hch)
> - Changed subject (David)
> - Dropped Josef's R-b due to change
> 
> Changes to v2:
> - open-code kunmap() + put_page() (David)
> - fix double kunmap() (David)
> - don't use bi_set_op_attrs() (David)
> 
> Changes to v1:
> - move 'super_page' into for-loop in btrfs_scratch_superblocks() (Nikolay)
> - switch to using pagecahce instead of alloc_pages() (Nikolay, David)
> ---
>  fs/btrfs/disk-io.c | 76 +++++++++++++++++++++++++---------------------
>  fs/btrfs/disk-io.h |  4 +--
>  fs/btrfs/volumes.c | 57 ++++++++++++++++++----------------
>  fs/btrfs/volumes.h |  2 --
>  4 files changed, 74 insertions(+), 65 deletions(-)
> 

<snip>

>  
> @@ -3355,40 +3363,38 @@ static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
>  }
>  
>  int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
> -			struct buffer_head **bh_ret)
> +			struct page **super_page)
>  {
> -	struct buffer_head *bh;
>  	struct btrfs_super_block *super;
> +	struct page *page;
>  	u64 bytenr;
> +	struct address_space *mapping = bdev->bd_inode->i_mapping;
>  
>  	bytenr = btrfs_sb_offset(copy_num);
>  	if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
>  		return -EINVAL;

You don't use page_offset(bytenr) here but you do it in
btrfs_scratch_superblocks. I'm aware that this could be omitted entirely
since the sb is always aligned on a 4k. But in this case either you omit
it everywhere or you use it everywhere for the sake of consistency.

>  
> -	bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
> -	/*
> -	 * If we fail to read from the underlying devices, as of now
> -	 * the best option we have is to mark it EIO.
> -	 */
> -	if (!bh)
> -		return -EIO;
> +	page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
> +	if (IS_ERR_OR_NULL(page))
> +		return -ENOMEM;

read_cache_page can return an error from ->readpage as well. Also
looking at do_read_cache_page it doesn't seem like it can return a NULL
pointer.

>  
> -	super = (struct btrfs_super_block *)bh->b_data;
> +	super = kmap(page);
>  	if (btrfs_super_bytenr(super) != bytenr ||
>  		    btrfs_super_magic(super) != BTRFS_MAGIC) {
> -		brelse(bh);
> +		kunmap(page);
> +		put_page(page);
>  		return -EINVAL;
>  	}
> +	kunmap(page);
>  
> -	*bh_ret = bh;
> +	*super_page = page;
>  	return 0;
>  }
>  
>  
> -struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
> +int btrfs_read_dev_super(struct block_device *bdev, struct page **page)
>  {
> -	struct buffer_head *bh;
> -	struct buffer_head *latest = NULL;
> +	struct page *latest = NULL;
>  	struct btrfs_super_block *super;
>  	int i;
>  	u64 transid = 0;

<snip>