Hello Kent et al.

My machine running 6.12-rc1 has a failing data-only (no btree) drive.

The failure seems to occur while seeking while appearing to run smoothly on 
linear reads, so after listening to some awful noises while running the 
`evacuate` subcommand for a few minutes I decided it would be wise to instead 
mount the rest of the filesystem as very_degraded without the failing drive, 
and image the disk into a file on the filesystem (allowing me to later loop 
this image for a non-degraded mount+evacuate).

During this imaging process, the machine crashed and rebooted of its own accord 
(journalctl recorded `watchdog: Watchdog detected hard LOCKUP on cpu 6`; 
unclear if this is related to bcachefs), and since the reboot, the filesystem 
cannot mount (neither as `-o very_degraded,fix_errors` without the bad drive, 
nor `-o fix_errors` with the failing drive):

bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): starting version 1.12: 
rebalance_work_acct_fix 
opts=metadata_replicas=2,data_replicas=2,metadata_target=optane,foreground_target=optane,background_target=hdd,promote_target=optane,nopromote_whole_extents,ver>
bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): recovering from unclean 
shutdown
bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): journal read done, replaying 
entries 9096592-9098738
bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): dropped unflushed entries 
9098739-9098745
bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): accounting_read...
accounting not marked in superblock replicas
  replicas btree: 1/2 [2 6], shutting down
bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): inconsistency detected - 
emergency read only at journal seq 9098753
bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): bch2_accounting_read(): error 
fsck_errors_not_fixed
bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): bch2_fs_recovery(): error 
fsck_errors_not_fixed
bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): bch2_fs_start(): error 
starting filesystem fsck_errors_not_fixed
bcachefs: bch2_fs_get_tree() error: fsck_errors_not_fixed 

I updated to the latest torvalds/linux.git to catch the recent `bcachefs: Disk 
accounting device validation fixes` you sent in for -rc4 which looked 
potentially relevant, but no improvement:

bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): starting version 1.13: 
inode_has_child_snapshots 
opts=metadata_replicas=2,data_replicas=2,metadata_target=optane,foreground_target=optane,background_target=hdd,promote_target=optane,nopromote_whole_extents,very_degraded,journal_flush_disabled,journal_reclaim_delay=200,fix_errors=yes
bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): recovering from unclean 
shutdown
bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): superblock requires following 
recovery passes to be run:
  check_inodes
bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): Version upgrade from 1.12: 
rebalance_work_acct_fix to 1.13: inode_has_child_snapshots incomplete
Doing compatible version upgrade from 1.12: rebalance_work_acct_fix to 1.13: 
inode_has_child_snapshots

bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): journal read done, replaying 
entries 9096592-9098738
bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): dropped unflushed entries 
9098739-9098745
bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): accounting_read...
accounting not marked in superblock replicas
  replicas btree: 1/2 [2 6], shutting down
bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): inconsistency detected - 
emergency read only at journal seq 9098753
bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): bch2_accounting_read(): error 
fsck_errors_not_fixed
bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): bch2_fs_recovery(): error 
fsck_errors_not_fixed
bcachefs (b7f3ad1a-1a17-487d-a670-333e0aaee6e2): bch2_fs_start(): error 
starting filesystem fsck_errors_not_fixed
bcachefs: bch2_fs_get_tree() error: fsck_errors_not_fixed

Any ideas?

Thanks
- Jamie McClymont

Reply via email to