Re: [PATCH RFC] btrfs-progs: mkfs: optimize file descriptor usage in mkfs.btrfs

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

On 2024/2/2 05:37, David Sterba wrote:
> On Thu, Feb 01, 2024 at 07:19:15PM +1030, Qu Wenruo wrote:
>> The problem is, even if we change the sequence, it doesn't make much
>> difference.
>>
>> There are several things involved, and most of them are out of our control:
>>
>> - The udev scan is triggered on writable fd close().
>> - The udev scan is always executed on the parent block device
>>     Not the partition device.
>> - The udev scan the whole disk, not just the partition
>>
>> With those involved, changing the nested behavior would not change anything.
>>
>> The write in another partition of the same parent block device can still
>> triggered a scan meanwhile we're making fs on our partition.
>
> So this means that creating ext4 on /dev/sda1 and btrfs on /dev/sda2 can
> trigger the udev events? And when both mkfs utilities would lock the
> device then running them concurrently will fail, right? This could
> happen in installation tools that can create different filesystems at
> the same time.

For ext4, I didn't see any direct flock call, maybe they are using some
helpers to do the same work but I'm not 100% sure.

>
> I wonder if we should add options to assert, skip locking or wait until
> it's free.
> The "udevadm lock" has various options to do that.
I'm pretty sure we can add options like --wait to implement the same
deadline behavior for mkfs.btrfs.

Thanks,
Qu