On Fri, Sep 09, 2016 at 11:30:06AM +0200, Marcin Mirosław wrote:

W dniu 09.09.2016 o 11:15, Kent Overstreet pisze:

quoted

On Fri, Sep 09, 2016 at 09:36:59AM +0200, Marcin Mirosław wrote:

quoted

Hi!
Today I launched host at my job. When I tried to mount bcachefs I got:
bcache: bch_open_as_blockdevs() register_cache_set err Bad checksum

This is simple, not tieres fs. I don't have idea what to do next to help
debug this. Data on this fs are completely unimportant.

The superblock checksum is bad (I'm going to improve that error message)... what
in the heck...

Anything remotely unusual about the machine or how you were using it?

It's a desktop with KDE, I used this fs for keeping portage tree & i put
compilation tmp dir on it (with lz4 compression). Stack is:
lvm->raid10->4 x HDD

That is the kind of setup that would make torn writes much more likely.

This is physical machine, I turned it off gracefully. Does image of
block device be usefull for you?

I don't think so. Regardless of what it is, I'm gonna need to work on some kind
of logged superblock or backup superblocks.

Assuming it is just a torn write (which I think is pretty likely), you should be
able to just disable that check and it should work fine - a torn write would
mostl likely only cause a superblock that's actually invalid if the size of the
superblock was changing (which is pretty much just when we're allocating the
journal, or when we're adding or removing devices).

The check is on super.c line 415 - just change the goto err to a printk and try
it, I bet it'll work.