Re: [PATCH 1/2] btrfs-progs: check: detect and warn about tree blocks cross 64K page boundary

[Qu Wenruo <quwenruo.btrfs@gmx.com>]

On 2021/3/2 下午6:36, Wang Yugui wrote:
> Hi,
>
> This is a full mkfs.btrtfs log
>
> [root@T7610 ~]# mkfs.btrfs -O no-holes -R free-space-tree /dev/sdb1 -f
> btrfs-progs v5.10.1
> See http://btrfs.wiki.kernel.org for more information.
>
> Detected a SSD, turning off metadata duplication.  Mkfs with -m dup if you want to force metadata duplication.
> Label:              (null)
> UUID:               8c745f77-3cdb-45f0-9b67-c69a9bd491a1
> Node size:          16384
> Sector size:        4096
> Filesystem size:    60.00GiB
> Block group profiles:
>    Data:             single            8.00MiB
>    Metadata:         single            8.00MiB
>    System:           single            4.00MiB
> SSD detected:       yes
> Incompat features:  extref, skinny-metadata, no-holes
> Runtime features:   free-space-tree
> Checksum:           crc32c
> Number of devices:  1
> Devices:
>     ID        SIZE  PATH
>      1    60.00GiB  /dev/sdb1
>
> [root@T7610 ~]# btrfs check /dev/sdb1
> Opening filesystem to check...
> Checking filesystem on /dev/sdb1
> UUID: 8c745f77-3cdb-45f0-9b67-c69a9bd491a1
> [1/7] checking root items
> [2/7] checking extents
> WARNING: tree block [5292032, 5308416) crosses 64K page boudnary, may cause problem for 64K page system
> WARNING: tree block [5357568, 5373952) crosses 64K page boudnary, may cause problem for 64K page system

This seems to be a false alert.

5292032 % 65536 = 49152, which is the last slot for 16K, and it doesn't
cross 64K page boundary.

The same goes for bytenr 5357568.

I'll fix it soon.

Thanks,
Qu
> [3/7] checking free space tree
> [4/7] checking fs roots
> [5/7] checking only csums items (without verifying data)
> [6/7] checking root refs
> [7/7] checking quota groups skipped (not enabled on this FS)
> found 147456 bytes used, no error found
> total csum bytes: 0
> total tree bytes: 147456
> total fs tree bytes: 32768
> total extent tree bytes: 16384
> btree space waste bytes: 140356
> file data blocks allocated: 0
>   referenced 0
> [root@T7610 ~]# mount /dev/sdb1 /mnt/scratch/
> [root@T7610 ~]# btrfs filesystem usage /mnt/scratch
> Overall:
>      Device size:                  60.00GiB
>      Device allocated:             20.00MiB
>      Device unallocated:           59.98GiB
>      Device missing:                  0.00B
>      Used:                        144.00KiB
>      Free (estimated):             59.99GiB      (min: 59.99GiB)
>      Free (statfs, df):            59.99GiB
>      Data ratio:                       1.00
>      Metadata ratio:                   1.00
>      Global reserve:                3.25MiB      (used: 0.00B)
>      Multiple profiles:                  no
>
> Data,single: Size:8.00MiB, Used:0.00B (0.00%)
>     /dev/sdb1       8.00MiB
>
> Metadata,single: Size:8.00MiB, Used:128.00KiB (1.56%)
>     /dev/sdb1       8.00MiB
>
> System,single: Size:4.00MiB, Used:16.00KiB (0.39%)
>     /dev/sdb1       4.00MiB
>
> Unallocated:
>     /dev/sdb1      59.98GiB
>
>
> [root@T7610 ~]# parted /dev/sdb unit s print
> Model: TOSHIBA PX05SMQ040 (scsi)
> Disk /dev/sdb: 97677846s
> Sector size (logical/physical): 4096B/4096B
> Partition Table: gpt
> Disk Flags:
>
> Number  Start      End        Size       File system  Name     Flags
>   1      262144s    15990783s  15728640s  btrfs        primary
>   2      15990784s  31719423s  15728640s               primary
>   3      31719424s  47448063s  15728640s               primary
>   4      47448064s  63176703s  15728640s               primary
>   5      63176704s  78905343s  15728640s               primary
>
> Best Regards
> Wang Yugui (wangyugui@e16-tech.com)
> 2021/03/02
>
>>
>>
>> On 2021/3/2 下午4:48, Wang Yugui wrote:
>>> Hi, Qu Wenruo
>>>
>>> This warning message happen even in new created filesystem on amd64
>>> system.
>>>
>>> Should we slicent it before mkfs.btrfs is ready for  64K page system?
>>
>> Nope.
>>
>> If your fs reports such problem, it means your metadata chunk is not properly aligned.
>>
>> The behavior of chunk allocator alignment has been there for a long long time, thus most metadata chunks should already been properly aligned to 64K.
>>
>> Either btrfs kernel module or mkfs.btrfs has something wrong.
>>
>>
>>>
>>> The paration is aligned in 1GiB
>>>
>>> btrfs-progs: v5.10.x branch
>>>
>>> # mkfs.btrfs /dev/sdb1 -f
>>>
>> And running v5.10.1 I can't reproduce it.
>>
>>> # btrfs check /dev/sdb1
>>> Opening filesystem to check...
>>> Checking filesystem on /dev/sdb1
>>> UUID: b298271d-6d1d-4792-a579-fb93653aa811
>>> [1/7] checking root items
>>> [2/7] checking extents
>>> WARNING: tree block [5292032, 5308416) crosses 64K page boudnary, may cause problem for 64K page system
>>> WARNING: tree block [5357568, 5373952) crosses 64K page boudnary, may cause problem for 64K page system
>>
>> I doubt if you're really using v5.10.x mkfs.btrfs.
>>
>> As for default btrfs, the metadata chunk is after DATA and SYS chunks, this means metadata chunks should only exist after bytenr 16M, but here your metadata is only at around 5M.
>>
>> I strongly doubt your mkfs parameters.
>>
>> Please attach the full mkfs output.
>>
>> Thanks,
>> Qu
>>
>>> [3/7] checking free space tree
>>> [4/7] checking fs roots
>>> [5/7] checking only csums items (without verifying data)
>>> [6/7] checking root refs
>>> [7/7] checking quota groups skipped (not enabled on this FS)
>>> found 147456 bytes used, no error found
>>> total csum bytes: 0
>>> total tree bytes: 147456
>>> total fs tree bytes: 32768
>>> total extent tree bytes: 16384
>>> btree space waste bytes: 140356
>>> file data blocks allocated: 0
>>>    referenced 0
>>>
>>> # parted /dev/sdb unit KiB print
>>> Model: TOSHIBA PX05SMQ040 (scsi)
>>> Disk /dev/sdb: 390711384kiB
>>> Sector size (logical/physical): 4096B/4096B
>>> Partition Table: gpt
>>> Disk Flags:
>>>
>>> Number  Start         End           Size         File system  Name     Flags
>>>    1      1048576kiB    63963136kiB   62914560kiB  btrfs        primary
>>>
>>>
>>> Best Regards
>>> Wang Yugui (wangyugui@e16-tech.com)
>>> 2021/03/02
>>>
>>>> For the incoming subpage support, there is a new requirement for tree
>>>> blocks.
>>>> Tree blocks should not cross 64K page boudnary.
>>>>
>>>> For current btrfs-progs and kernel, there shouldn't be any causes to
>>>> create such tree blocks.
>>>>
>>>> But still, we want to detect such tree blocks in the wild before subpage
>>>> support fully lands in upstream.
>>>>
>>>> This patch will add such check for both lowmem and original mode.
>>>> Currently it's just a warning, since there aren't many users using 64K
>>>> page size yet.
>>>>
>>>> Signed-off-by: Qu Wenruo <wqu@suse.com>
>>>> ---
>>>>    check/main.c        |  2 ++
>>>>    check/mode-common.h | 18 ++++++++++++++++++
>>>>    check/mode-lowmem.c |  2 ++
>>>>    3 files changed, 22 insertions(+)
>>>>
>>>> diff --git a/check/main.c b/check/main.c
>>>> index e7996b7c8c0e..0ce9c2f334b4 100644
>>>> --- a/check/main.c
>>>> +++ b/check/main.c
>>>> @@ -5375,6 +5375,8 @@ static int process_extent_item(struct btrfs_root *root,
>>>>    		      num_bytes, gfs_info->sectorsize);
>>>>    		return -EIO;
>>>>    	}
>>>> +	if (metadata)
>>>> +		btrfs_check_subpage_eb_alignment(key.objectid, num_bytes);
>>>>   >>   	memset(&tmpl, 0, sizeof(tmpl));
>>>>    	tmpl.start = key.objectid;
>>>> diff --git a/check/mode-common.h b/check/mode-common.h
>>>> index 4efc07a4f44d..bcda0f53e2c4 100644
>>>> --- a/check/mode-common.h
>>>> +++ b/check/mode-common.h
>>>> @@ -171,4 +171,22 @@ static inline u32 btrfs_type_to_imode(u8 type)
>>>>   >>   	return imode_by_btrfs_type[(type)];
>>>>    }
>>>> +
>>>> +/*
>>>> + * Check tree block alignement for future subpage support on 64K page system.
>>>> + *
>>>> + * Subpage support on 64K page size require one eb to be completely contained
>>>> + * by a page. Not allowing a tree block to cross 64K page boudanry.
>>>> + *
>>>> + * Since subpage support is still under development, this check only provides
>>>> + * warning.
>>>> + */
>>>> +static inline void btrfs_check_subpage_eb_alignment(u64 start, u32 len)
>>>> +{
>>>> +	if (start / BTRFS_MAX_METADATA_BLOCKSIZE !=
>>>> +	    (start + len) / BTRFS_MAX_METADATA_BLOCKSIZE)
>>>> +		warning(
>>>> +"tree block [%llu, %llu) crosses 64K page boudnary, may cause problem for 64K page system",
>>>> +			start, start + len);
>>>> +}
>>>>    #endif
>>>> diff --git a/check/mode-lowmem.c b/check/mode-lowmem.c
>>>> index 2b689b2abf63..6dbfe829bb7c 100644
>>>> --- a/check/mode-lowmem.c
>>>> +++ b/check/mode-lowmem.c
>>>> @@ -4206,6 +4206,8 @@ static int check_extent_item(struct btrfs_path *path)
>>>>    		      key.objectid, key.objectid + nodesize);
>>>>    		err |= CROSSING_STRIPE_BOUNDARY;
>>>>    	}
>>>> +	if (metadata)
>>>> +		btrfs_check_subpage_eb_alignment(key.objectid, nodesize);
>>>>   >>   	ptr = (unsigned long)(ei + 1);
>>>>   >> --
>>>> 2.29.2
>>>
>>>
>
>