On 2019/12/21 下午2:27, Chris Murphy wrote:
> kernel 5.4.5
> btrfs-progs 5.4
> 
> test file is linux.tar (not compressed), it's the only file on the
> file system in each case
> 
> /dev/sda5        53G  2.8G   50G   6% /mnt     none
> /dev/sda5        53G  2.2G   50G   5% /mnt     zstd 1
> /dev/sda5        53G  2.1G   50G   4% /mnt     zstd 15
> 
> 
> mount and dmesg both show the value for the level I've set; but btrfs
> insp dump-t shows extents always have compression 3.
> 
> 
> [47567.500812] BTRFS info (device sda5): use zstd compression, level 1
> 
>     item 14 key (328583 EXTENT_DATA 2060582912) itemoff 15488 itemsize 53
>         generation 51 type 1 (regular)
>         extent data disk byte 6461308928 nr 20480
>         extent data offset 0 nr 131072 ram 131072
>         extent compression 3 (zstd)

This number is not compression level, but compression algorithm.

typedef enum {
        BTRFS_COMPRESS_NONE  = 0,
        BTRFS_COMPRESS_ZLIB  = 1,
        BTRFS_COMPRESS_LZO   = 2,
        BTRFS_COMPRESS_ZSTD  = 3,
        BTRFS_COMPRESS_TYPES = 3,
        BTRFS_COMPRESS_LAST  = 4,
} btrfs_compression_type;

Level is not recorded in that field.

Thanks,
Qu

> 
> 
> ==================
> 
> [47793.161246] BTRFS info (device sda5): use zstd compression, level 15        │
> 
>     item 170 key (328584 EXTENT_DATA 2735341568) itemoff 7220 itemsize 53
>         generation 54 type 1 (regular)
>         extent data disk byte 2168475648 nr 24576
>         extent data offset 0 nr 131072 ram 131072
>         extent compression 3 (zstd)
> 
> 
> I'd expect a bigger difference between level 1 and 15, so I'm a little
> suspicious that it really is always using level 3. But it's also
> possible that it's just a bug with inspect always reporting level 3.
> The rate of the level 15 copy is slower.
>