Marcin

segfault was due to broken images. I didn't track which one triggers bug.

I observed that umount almost always doesn't do umount action when there is a lot of CPU and I/O (on other bcachefs filesystems on the host) pressure.

I still observe that umount often doesn't do full umount. There is no files from fs but there is nomessages indmesg that was unmounted. Also fsck on block device says...

I have couple filesystems on box, but I suspect this is related also to this fs: ``` External UUID: 2fc82775-653c-4995-89f9-5df5f626314c Internal UUID: 1ec747b3-ea96-485b-bdc5-745565da4b69 Device index: 0 Label: media2 Version: 1.3:...

Thanks, i'm not sure if this is in master branch, as for now, at commit #6c5850f4860d issue exists.

Hmm, this happens also in userspace: ``` # bcachefs fsck -n /dev/vg-ZA1C3NKP/burp mounting version 1.3: rebalance_work opts=metadata_replicas=2,compression=lz4,background_compression=zstd:6,degraded,journal_reclaim_delay=1000,fsck,fix_errors=no,nochanges,nodirect_io recovering from clean shutdown, journal seq 23150366 journal read done, replaying entries 23150366-23150366...

The same fs. Now it looks similar, `bch-copygc` is spinning cpu, `bch-rebalance` is almost all the time in D state, from time to time I see it's in R state....

I just run fsck to check how many data fsck reads. fsck from -tools read 119GB of data while doing fsck with "-n". Also, I have a lot of `incorrect...

bcachefs isn't part of mainline kernel, you should compile kernel by your self.

There is also https://github.com/nethappen/blocksync-fast