Re: Request for comments about support btrfs in lvresize

On 3/11/25 09:47, Glass Su wrote:
> 
> 
>> On Mar 11, 2025, at 00:24, David Teigland <[email protected]> wrote:
>>
>> On Mon, Mar 10, 2025 at 04:35:54PM +0800, Glass Su wrote:
>>> Hi
>>> I am drafting patches for btrfs support in lvresize command. All things went smoothly until I realized
>>> multi-devices btrfs handling is a little complicated.
>>
>> btrfs using multiple devices sounds like an alternative to putting btrfs
>> on lvm, so is this actually done?
> 
> Yeah. Btrfs don’t care about the backing device is lv or not. It just treats devices as how ext* and xfs do.
> For those people don’t trust btrfs raid stability especially raid56, the use case indeed exits.

I feel that few users use btrfs on lvm2 with multiple LVs. because I have
never received any issue related to btrfs on lvm2. And from what I remember,
there are ​not too many users using LVM2 RAID mode either. People ​tend to
prefer using ​mdadm to manage RAID disks (or using hardware raid card).

Anyway, it is ​possible to set up btrfs on LVM2 with multiple LVs, so we
should support this scenario.

- Heming

>>
>>> For one device btrfs, fslastblock * fsblocksize/FSSIZE is the correct value like ext* and xfs
>>> But for multi-devices btrfs, the two values are whole fs size. The used size of device can be
>>> known by btrfs-progs or superblock read and parse.
>>
>> These values are used to calculate fs_last_byte, and that is used:
>>
>> - by lvextend as a sanity check after the fs resize is done.  So, it could
>>   be made optional for lvextend.
>>
>> - by lvreduce to check if the fs has already been sufficiently shrunk, so
>>   the fs shrink command can be skipped.  Perhaps for btrfs, always require
>>   the fs shrink command to be run, and always make fs_last_byte large enough
>>   to do that.
> 
> Right. I’ve considered the way making fs_last_byte zero/large enough as `btrfs filesystem resize ` can
> handle the newsize rightly. For me, it’s more like a quirk. If it’s acceptable to you I can provide a patchset to implement
> the logic, and a testcase, of course.
> 
> —
> Su
>>
>> Dave
>>
>