On Sun, Jan 14, 2024 at 06:55:03PM -0500, Alec Moskvin wrote: > On Linux 6.7.0 with the latest bcachefs-tools 1.4.0, I get an error when > trying to run fsck: > > bcachefs fsck -v /dev/sda3 > bch2_parse_mount_opts() Bad mount option read_only > > If I try a newer git revision of bcachefs-tools, it works but does an > upgrade of the fs version: > > bcachefs fsck /dev/sda3 > mounting version 1.3: rebalance_work > opts=ro,compression=zstd:15,degraded,verbose,fsck,fix_errors=ask,read_only > recovering from clean shutdown, journal seq 116268 > Doing compatible version upgrade from 1.3: rebalance_work to 1.4: > member_seq > > and then when I mount it, it downgrades: > > [Jan14 11:06] bcachefs (sda3): mounting version 1.4: (unknown version) > opts=compression=zstd:15 > [ +0.000016] bcachefs (sda3): recovering from clean shutdown, journal > seq 116272 > [ +0.000003] bcachefs (sda3): Version downgrade required: > > [ +0.200171] bcachefs (sda3): alloc_read... done > > Is there a correspondence between the kernel and bcachefs-tools > versions? I.e., should bcachefs-tools 1.3.6 be used until a kernel with > bcachefs 1.4 is released?
No, for now I've been focusing on making the upgrade/downgrade process work smoothly and reliably; and in this case 1.4 just introduced the bch_member->seq field so there's no work to do on upgrade/downgrade - the new version just indicates that the bch_member->seq field can be trusted for split brain detection. So - using mismatched kernel/tools is completely fine.
