On 21/10/2021 12:03, Anand Jain wrote:

On 21/10/2021 02:59, David Sterba wrote:

quoted

On Tue, Oct 19, 2021 at 08:22:10AM +0800, Anand Jain wrote:

quoted

In the case of the seed device, the fsid can be different from the 
mounted
sprout fsid.  The userland has to read the device superblock to know the
fsid but, that idea fails if the device is missing. So add a sysfs
interface devinfo/<devid>/fsid to show the fsid of the device.

For example:
  $ cd /sys/fs/btrfs/b10b02a5-f9de-4276-b9e8-2bfd09a578a8

  $ cat devinfo/1/fsid
  c44d771f-639d-4df3-99ec-5bc7ad2af93b
  $ cat  devinfo/3/fsid
  b10b02a5-f9de-4276-b9e8-2bfd09a578a8

How do you create such setup? I can't reproduce it.

The simplest seeding:

   mkfs.btrfs /dev/sdx
   btrfstune -S 1 /dev/sdx
   mount /dev/sdx mnt
   ... the device has the same FSID as is the sysfs directory name

With a new device and removed the seeding one:

   btrfs device add /dev/sdy mnt

At this step, we generate a new fsid for the writeable FS. Let's call
it sprout-fsid. (If you check the sys-fs fsid, you will have two fsid
here).

Also, at this step,
the
     fs_info->fs_devices->fsid
changes from seed-fsid to the sprout-fsid.
So, we should make the <mnt> also a writeable without the need to call
'remount,rw' explicitly IMO. What do you think?

More importantly, the fs_info->super_copy is of sprout device.
So there isn't any reason that we should maintain the new
sprout fs as readonly after the device add.

quoted

   mount -o remount,rw mnt
   btrfs device delete /dev/sdx mnt
   ... both devices have the same fsid as well 

Thanks, Anand