Bug#803531: systemd: timeout mounting /home (btrfs) at boot

[Martin Pitt]

Hey Brian,

Martin Pitt [2015-11-29  0:56 +0100]:
> | P: /devices/pci0000:00/0000:00:1f.2/ata3/host2/target2:0:0/2:0:0:0/block/sdb/sdb1
> | E: ID_FS_UUID=f77d6ce8-12bf-476a-8276-2031ce3e3c42
> | E: ID_BTRFS_READY=1
> |
> | P: /devices/pci0000:00/0000:00:1f.2/ata4/host3/target3:0:0/3:0:0:0/block/sdc/sdc1
> | E: ID_FS_UUID=f77d6ce8-12bf-476a-8276-2031ce3e3c42
> | E: ID_BTRFS_READY=0
> 
> It's indeed very likely that this is due to the "mirrored" mode and
> both partitions have the same UUID. I guess this confuses the kernel
> driver and/or udev somehow?

I tested this with the attached script (in two variants),  and I don't
see anything generally wrong with it; sure, the two devices have the
exact same UUID thus the symlink will randomly point to one or the
other, but both of my devices have ID_BTRFS_READY=1. I tried this some
20 times. However, this is using scsi_debug (i. e. a fake kernel
RAM-backed SCSI drive), not real iron. This could likely be a timing
problem.

So this needs to be reproduced with a full install cycle and adding to
fstab etc.

Brian May [2015-11-29 12:40 +1100]:
> > This ought to be reproducible in a VM. (Won't do it right now myself;
> > too late, sorry.) To clarify, how exactly did you create this
> > "mirrored" btrfs mode, to ensure we try the same thing for
> > reproducing?
> 
> I created it on a single drive, then followed the instructions to mirror
> it in RAID1 mode on the other drive.

"followed the instructions" wasn't quite what I was aiming for :-)
I tried that with

  mkfs.btrfs /dev/sdb1
  mount /dev/sdb1 /mnt
  btrfs device add /dev/sdb2 /mnt

Is that roughly what you did?

Martin

-- 
Martin Pitt                        | http://www.piware.de
Ubuntu Developer (www.ubuntu.com)  | Debian Developer  (www.debian.org)