Re: [PATCH V4] btrfs-progs: fix btrfs resize failed.

[David Sterba <dsterba@suse.cz>]

On Tue, Jul 26, 2022 at 01:04:39PM +0800, Qu Wenruo wrote:
> On 2022/7/26 12:56, li zhang wrote:
> > Hi, I have another question, issue 471, extending resizing to support
> > the parameter all to represent all devices, I agree this should
> > iterate over all devices in userspace as it works on older kernel
> > versions. But what happens if the command fails?
> 
> Just error out at the failed device, and print a proper error message.

Agreed, what device (id, maybe path), original size and target size so
the operation can be repeated or the other devices resized back.

If it's a one message per line like we have now then the error could
repeat some key information and that should be sufficient.

> 
> > For example, if the
> > filesystem contains 3 devices, 2 of them resize successfully and the
> > last one fails, in this case I mean the whole command should fail,
> 
> Yes, the command should fail, even if some devices have been resized
> successfully.
> 
> > which means it's better to put these 3 resize subcommands in a
> > transaction. So I'm stuck in a dilemma whether to implement all device
> > resizing in user space or kernel space.
> 
> I still prefer to do it inside user space.
> 
> I don't think there is anything wrong if we resized some devices but not
> all, we still return an error.
> 
> Sure, it's better to output some messages to indicate that some devices
> have been resized, other than that, I see no problem at all.

Agreed.