[OE-core] systemd/busybox fsck fail to run

[Alex Kiernan]

On Wed, Sep 5, 2018 at 9:25 PM Khem Raj <raj.khem at gmail.com> wrote:
>
>
>
> On 9/5/18 5:46 AM, Andrej Valek wrote:
> > Hi Alex,
> >
> > There could be multiple options, how to handle this.
> >
> > My 2cents about this problem:
> >  1. Use update-alternatives to handle this
> >  2. Rework busybox itself to handle this option
> >  3. Just create a patch to ignore this option
> >
>
> I am not sure if systemd-fsck is supposed to work with anything other
> than util-linux and e2progfs tools. So lets add a hard dependency in
> systemd for this.
>

I started prepping this change and then realised why I'm seeing
this... I've been chopping away at image size so am running with
NO_RECOMMENDATIONS in my image recipe and we currently have
util-linux-fsck as a RRECOMMENDS not an RDEPEND, so I end up not
taking it.

But... my feeling is we should probably still make it a hard
dependency as it doesn't seem to me that it meets the definition of
"packages that extends the usability of a package being built".

> > Cheers,
> > Andrej
> >
> > On 09/05/18 10:37, Alex Kiernan wrote:
> >> I've just come across a box with a corrupt filesystem which wasn't
> >> being cleaned (using systemd and busybox):
> >>
> >> Aug 29 16:03:22 localhost systemd-fsck[143]: read_bad_blocks_file: No
> >> such file or directory while trying to open -M
> >> Aug 29 16:03:22 localhost systemd-fsck[143]: Warning: fsck.ext4
> >> /dev/mmcblk0p4 terminated by signal 8
> >> Aug 29 16:03:22 localhost systemd-fsck[143]: fsck failed with exit status 8.
> >> Aug 29 16:03:22 localhost systemd-fsck[143]: Ignoring error.
> >> Aug 29 16:03:22 localhost systemd[1]:
> >> systemd-fsck at dev-mmcblk0p4.service: cgroup is empty
> >> Aug 29 16:03:22 localhost systemd[1]:
> >> systemd-fsck at dev-mmcblk0p4.service: Child 143 belongs to
> >> systemd-fsck at dev-mmcblk0p4.service.
> >> Aug 29 16:03:22 localhost systemd[1]:
> >> systemd-fsck at dev-mmcblk0p4.service: Main process exited, code=exited,
> >> status=0/SUCCESS
> >> Aug 29 16:03:22 localhost systemd[1]:
> >> systemd-fsck at dev-mmcblk0p4.service: Changed start -> exited
> >> Aug 29 16:03:22 localhost systemd[1]:
> >> systemd-fsck at dev-mmcblk0p4.service: Job
> >> systemd-fsck at dev-mmcblk0p4.service/start finished, result=done
> >> Aug 29 16:03:22 localhost systemd[1]: Started File System Check on
> >> /dev/mmcblk0p4.
> >>
> >> It turns out systemd-fsck unconditionally passes `-l` to lock the filesystem:
> >>
> >> https://github.com/systemd/systemd/blame/master/src/fsck/fsck.c#L408
> >>
> >> which busybox fsck doesn't interpret and so passes on to fsck.ext4,
> >> which then interprets it as a badblocks file...
> >>
> >> What's the right answer here? Add a patch to busybox fsck so it just
> >> ignores `-l`, or do we need something more sophisticated?
> >>
> >> Obviously swapping in util-linux-fsck works around the problem.
> >>
>


-- 
Alex Kiernan