Re: [PATCH 4/4] btrfs: show temp_fsid feature in sysfs

[Anand Jain <anand.jain@oracle.com>]

On 10/7/23 17:12, Anand Jain wrote:
> 
> 
> On 10/6/23 22:55, David Sterba wrote:
>> On Wed, Oct 04, 2023 at 11:00:27PM +0800, Anand Jain wrote:
>>> This adds sysfs objects to indicate temp_fsid feature support and
>>> its status.
>>>
>>>    /sys/fs/btrfs/features/temp_fsid
>>>    /sys/fs/btrfs/<UUID>/temp_fsid
>>>
>>>    For example:
>>>
>>>       Consider two cloned and mounted devices.
>>>
>>>     $ blkid /dev/sdc[1-2]
>>>     /dev/sdc1: UUID="509ad44b-ad2a-4a8a-bc8d-fe69db7220d5" ..
>>>     /dev/sdc2: UUID="509ad44b-ad2a-4a8a-bc8d-fe69db7220d5" ..
>>>
>>>       One gets actual fsid, and the other gets the temp_fsid when
>>>       mounted.
>>>
>>>     $ btrfs filesystem show -m
>>>     Label: none  uuid: 509ad44b-ad2a-4a8a-bc8d-fe69db7220d5
>>>         Total devices 1 FS bytes used 54.14MiB
>>>         devid    1 size 300.00MiB used 144.00MiB path /dev/sdc1
>>>
>>>     Label: none  uuid: 33bad74e-c91b-43a5-aef8-b3cab97ae63a
>>>         Total devices 1 FS bytes used 54.14MiB
>>>         devid    1 size 300.00MiB used 144.00MiB path /dev/sdc2
>>>
>>>       Their sysfs as below.
>>>
>>>     $ cat /sys/fs/btrfs/features/temp_fsid
>>>     0
>>>
>>>     $ cat /sys/fs/btrfs/509ad44b-ad2a-4a8a-bc8d-fe69db7220d5/temp_fsid
>>>     0
>>>
>>>     $ cat /sys/fs/btrfs/33bad74e-c91b-43a5-aef8-b3cab97ae63a/temp_fsid
>>>     1
>>
>> So the fsid used for the directory is always the new one, is there a way
>> to read which is the original filesystem's fsid? In this case it would
>> be the 509ad44b-... We could print it in that file instead of '1',
>> though it could be confusing that it's not the temp_fsid but the
>> original one, file name mismatches the contents on first look.
> 
> Instead, can we emit 'fsid' in another kobject altogether?
> Furthermore, we also have a 'metadata_uuid' kobject. Here
> is how they relate.
> 
> 
> 1. normally:
> 
>   $ cat /sys/fs/btrfs/<meta-fsid>/fsid
>   <meta-fsid>
>   $ cat /sys/fs/btrfs/<meta-fsid>/metadata_uuid
>   <meta-fsid>
> 
> 
> 2. metadata-uuid flag is set:
> 
>   $ cat /sys/fs/btrfs/<sb-fsid>/fsid
>   <sb-fsid>
>   $ cat /sys/fs/btrfs/<sb-fsid>/metadata_uuid
>   <meta-fsid>
> 
> 
> 3. normal + temp-fsid:
> 
>   $ cat /sys/fs/btrfs/<temp-fsid>/fsid
>   <meta-fsid>
>   $ cat /sys/fs/btrfs/<temp-fsid>/metadata_uuid
>   <meta-fsid>
> 
> 
> 4. metadata-uuid flag is set + temp-fsid:
> 
>   $ cat /sys/fs/btrfs/<temp-fsid>/fsid
>   <sb-fsid>
>   $ cat /sys/fs/btrfs/<temp-fsid>/metadata_uuid
>   <meta-fsid>



Let's consider this thought:

The proposal above implies that /sys/fs/btrfs/<temp-fsid>/temp_fsid
will continue to function as described in this patch. Otherwise,
determining whether temp_fsid is enabled becomes challenging.

As part of the temp-fsid feature, we plan to introduce two additional
kobjects. However, if this results in an excessive number of kobjects,
then...

We have the option to rename /sys/fs/btrfs/<temp-fsid>/temp_fsid
to /sys/fs/btrfs/<temp-fsid>/temp-fsid_enabled_super_fsid to display
the actual sb::fsid.

Thanks, Anand