[systemd-devel] fstab, rootfs on btrfs

[Lennart Poettering]

On Wed, 27.11.13 12:15, Chris Murphy (lists at colorremedies.com) wrote:

> 
> 
> On Nov 27, 2013, at 4:53 AM, Kay Sievers <kay at vrfy.org> wrote:
> 
> > On Wed, Nov 27, 2013 at 5:16 AM, Chris Murphy <lists at colorremedies.com> wrote:
> >> 
> >> In Fedora 20, by default anaconda sets fs_passno in fstab to 1 for / on btrfs. During offline updates, this is causing systemd-fstab-generator to freak out not finding fsck.btrfs.
> >> 
> >> https://bugzilla.redhat.com/show_bug.cgi?id=1034563
> > 
> > Right, it should not set 1 for btrfs.
> 
> Since it can be mounted rw from the get go, is it going to far to plan
> for one day dropping the initial ro mount, remount rw sequence?

Hmm, no, not really. If no initrd is used it's already bad enough that
we run fsck while we have the file system mounted, but it would be even
worse if we'd write to the file system.

The recommended logic is: use an initrd, and fsck the rootfs before
mounting, and do so writable. The legacy logic is: use no initrd, and
let the kernel mount read-only, then file system check, then remount
writable.

Or actaully, it's even more complex than that: the remount step after
fsck does more than just mount writable, it will actually apply the
mount options in /etc/fstab to the root file system, whatever those
settings might be. Writability is just one of the.

Lennart

-- 
Lennart Poettering, Red Hat