matthiasgoergens
commented
3 months ago I still have background_compression off, and haven't hit this problem again, since.
Open matthiasgoergens opened 4 months ago
matthiasgoergens
commented
3 months ago I still have background_compression off, and haven't hit this problem again, since.
koverstreet
commented
3 months ago Curious about the 'dirent points to inode that does not point back' subject? It doesn't match the rest of the report.
I just pushed "bcachefs: Convert for_each_btree_node() to lockrestart_do()" to the testing branch to fix the SRCU warning.
matthiasgoergens
commented
3 months ago The subject line is a quote of what's in dmesg.
xarblu
commented
2 months ago Not sure how related this is but I started consistently receiving the same error when starting steam. Specifically it seems to always occur during the steam client update.
Error message:
bcachefs (93879ca1-9ad1-47b1-866e-623b60c911d9): dirent points to inode that does not point back:
u64s 8 type dirent 673049566:1177845408488155637:4294967288 len 0 ver 0: ldconfig -> 1612331176 type reg
inum: 1612331176 mode=100755
flags= (15300000)
journal_seq=10674964
bi_size=991016
bi_sectors=1936
bi_version=0
bi_atime=12537025645715285
bi_ctime=18373447315757940
bi_mtime=6924161156869999
bi_otime=12537025645715285
bi_uid=1000
bi_gid=1000
bi_nlink=0
bi_generation=0
bi_dev=0
bi_data_checksum=0
bi_compression=0
bi_project=0
bi_background_compression=0
bi_data_replicas=0
bi_promote_target=0
bi_foreground_target=0
bi_background_target=0
bi_erasure_code=0
bi_fields_set=0
bi_dir=1479355554
bi_dir_offset=5768648552471826948
bi_subvol=0
bi_parent_subvol=0
bi_nocow=0
bcachefs (93879ca1-9ad1-47b1-866e-623b60c911d9): inconsistency detected - emergency read only at journal seq 10888398
bcachefs (93879ca1-9ad1-47b1-866e-623b60c911d9): unshutdown complete, journal seq 10888398Versions: Kernel: 6.10.8-linux-cachyos Bcachefs-tools: 1.11.0
koverstreet
commented
2 months ago You guys running online fsck?
I just hit it for the first time last night, after running online fsck.
xarblu
commented
2 months ago You guys running online fsck?
Yes
matthiasgoergens
commented
2 months ago I was also running online fsck, during boot.
koverstreet
commented
2 months ago yeah, I see the bug - when we find an unlinked inode we're not checking if it's still open (because if it's not an online fsck, why would it be?)
this'll be tricky to fix, because vfs inodes are indexed by subvolume, and we only have the snapshot id in fsck
matthiasgoergens
commented
2 months ago Btw, is it expected that background compression hits more bugs like these, or was that just a coincidence?
koverstreet
commented
2 months ago background compression wouldn't have anything to do with this, no
RodoMa92
commented
4 weeks ago I've also hit this today, I'll try an offline fsck then. Any progress on fixing this?
koverstreet
commented
4 weeks ago Was online fsck involved?
RodoMa92
commented
4 weeks ago Was online fsck involved?
Yep, it was. I ran that yesterday I think?
matthiasgoergens
commented
4 hours ago After upgrading to Linux 6.12.1-arch1-1 I am getting this problem again.
This makes my system pretty much unusable. Turning off compression doesn't seem to help, this time.
I'm running bcachefs as a root filesystem on an Archlinux desktop. I just updated to kernel
6.10.0-arch1-2. I can boot fine, but after running for a few minutes I get errors like:Digging a bit via
sudo dmesg | grep bcachefsgives:I tried
sudo mount -o fsck,fix_errors,remount,rw /but that doesn't seem to do anything.My bcachefs filesystem lives on
/dev/nvme0n1p3for foreground writes and as a promote target, and an/dev/sdc1(a hard disk) for background writes:Update: I booted from a USB stick. Running fsck multiple times in a row keeps finding problems to fix, weirdly enough.
I'm turning background compression off, just to make the system simpler and reduce the 'surface' of things that can go wrong.
Next update: have booted my normal system again. Running fsck multiple times does not keep finding problems. It's stable so far. Perhaps it randomly fixed itself, or turning off background_compression did the trick? I'll be monitoring.