Hi,

This series fixes the error path for adding the first writable device to
a seed filesystem.

With fault injection after the metadata chunk creation in
init_first_rw_device(), the newly created block group is drained by the
transaction abort path. The problem is not a local block group reference
leak, but partially installed device state left behind by
btrfs_init_new_device().

The failed sprout device can remain linked on the transaction device
update list, latest_dev/s_bdev can still point to the failed device, and
the mounted filesystem can remain in the partially initialized sprout
fs_devices state with no open devices.

Before this series, the injected failure triggered:

  - WARN_ON(!list_empty(&device->post_commit_list)) in btrfs_free_device()
  - a NULL pointer dereference through btrfs_show_devname()
  - ASSERT(nr_devices) in btrfs_statfs()

After this series, the same injected failure returns -EIO from the device
add operation, the pending block group is removed by the transaction abort
path, btrfs filesystem show still reports the seed device, and umount
completes without the above warnings or crashes.

Guanghui Yang (3):
  btrfs: detach failed sprout device from transaction update list
  btrfs: restore active device pointers after failed sprout
  btrfs: roll back sprout setup after device add failure

 fs/btrfs/volumes.c | 43 +++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 43 insertions(+)


base-commit: a13c140cc289c0b7b3770bce5b3ad42ab35074aa
-- 
2.52.0.windows.1