[Fedora-livecd-list] possible bug alert: device-mapper: snapshots: invalid or corrupt snapshot

Douglas McClendon dmc.fedora at filteredperception.org
Wed Oct 10 05:29:10 UTC 2007


I've been running into a bug, but can't yet rule out my own code, even 
though I really really don't think I could have caused it.

The behaviour would be obvious, i.e. boot fails real early and drops you 
into the emergency initramfs shell, with the subject line error message 
either visible obviously, or at the end of dmesg output.

One bizarre behavior I've noticed while trying to pin this down, is that 
the sparse overlay file used in the dm-snapshot, created with dd 
of=/overlay bs=1024 count=1 seek=$((512*1024)), seems to often be 
512*1024*1024 bytes long, instead of 512*1025*1024 bytes long like it 
should.  I.e. if you remove the '2> /dev/null' from the dd call in 
mayflower's init, you may see 0 records written in&out, when you should 
see 1 record written in&out.

Now, why count=1 is used instead of count=0 is an entirely separate 
question, to which I'm not sure enough of the answer to suggest using 
count=0.  (reason would I guess seem to be that dd might be reasonably 
be entitled to just do _absolutely_ nothing if you say count=0).

Anyway...  If you hit this, you'll know it.  And probably would have 
posted it anyway, but now you'll know immediately that you aren't the 
only one hitting this bug.

The nasty part of it is that it is not deterministic.  I rerun the exact 
same qemu command on the same iso, and sometimes it hits it, sometimes 
it just works fine.

Though if anybody is curious enough to remove the dd '2> /dev/null' and 
see if they get the 0 records in/out result, and hopefully can explain 
why in the world that is happening...  I'd be greatful...

-dmc




More information about the livecd mailing list