bcachefs format can overwrite existing bcachefs filesystem with no warning

[iskarian]

Not sure if this is a documentation or implementation error.

I initially formatted one device with bcachefs

bcachefs format \
--label=bulk.hdd1 /dev/sdb1 \
--compression=lz4

transferred some data do it, then (naively) attempted to add a cache layer in front of it

bcachefs format \
--label=cache.nvme1 /dev/nvme/cache \
--label=bulk.hdd1 /dev/sdb1 \
--compression=lz4 \
--foreground_target=cache \
--promote_target=cache \
--metadata_target=cache \
--background_target=bulk \
--metadata_replicas=2

which succeeded without errors, silently overwriting the data on /dev/sdb1. I would argue that this is a common enough use case that warning would be a good idea. In fact, in the manpage, there is an option

            -f, --force
                     Force the filesystem to be created, even if the device already contains a filesystem.

so I would assume that the default is to warn before overwriting.

# bcachefs version
1.12.0

Also, if anyone has suggestions for how I might recover said overwritten data...

[g2p]

The issue was that bcachefs superblock detection can silently fail due to an outdated blkid dependency. That's now detected at format time (#347) but it needs to be in a release. Am also proposing #348 and #336 to make packaging less error prone.

For recovery, I have no idea if it's possible, but I would hold on to that superblock dump you mentioned, as that has various filesystem UUIDs and parameters that might turn out useful to detect where the previous data is.