[OE-core] [PATCH 1/2] wic: When using --use-uuid make sure that we update the fstab with PARTUUID

Wed Nov 8 15:00:54 UTC 2017

On Wed, Nov 08, 2017 at 01:20:15PM +0200, Ed Bartosh wrote:
> On Tue, Nov 07, 2017 at 09:54:57AM -0500, Tom Rini wrote:
> > On Tue, Nov 07, 2017 at 10:11:35AM +0200, Ed Bartosh wrote:
> > > On Mon, Nov 06, 2017 at 07:44:23AM -0500, Tom Rini wrote:
> > > > On Mon, Nov 06, 2017 at 09:36:20AM -0200, Otavio Salvador wrote:
> > > > > On Mon, Nov 6, 2017 at 8:08 AM, Ed Bartosh <ed.bartosh at linux.intel.com> wrote:
> > > > > > On Fri, Nov 03, 2017 at 08:51:50AM -0400, Tom Rini wrote:
> > > > > >> On Fri, Oct 20, 2017 at 09:15:05AM -0400, Tom Rini wrote:
> > > > > >> > On Tue, Oct 10, 2017 at 05:01:49PM -0400, Tom Rini wrote:
> > > > > >> > > On Thu, Sep 21, 2017 at 01:46:16PM -0400, Tom Rini wrote:
> > > > > >> > > > When we have been told to use the UUID we should also update the fstab
> > > > > >> > > > to make use of PARTUUID instead of hard-coding the device in question.
> > > > > >> > > > This will make the resulting image much more portable.
> > > > > >> > > >
> > > > > >> > > > Signed-off-by: Tom Rini <trini at konsulko.com>
> > > > > >> > > > ---
> > > > > >> > > >  scripts/lib/wic/plugins/imager/direct.py | 9 ++++++---
> > > > > >> > > >  1 file changed, 6 insertions(+), 3 deletions(-)
> > > > > >> > >
> > > > > >> > > Where we did we end up with this?  Ed pointed out that you can tell wic
> > > > > >> > > to use a specific UUID, so reproducible images are not a problem.  And
> > > > > >> > > making images that are readily portable is why other distros use
> > > > > >> > > UUID/LABEL and not device names as much as possible.  I personally enjoy
> > > > > >> > > being able to put an image on uSD for minnow and have it work :)
> > > > > >> > > Thanks!
> > > > > >> >
> > > > > >> > ping?
> > > > > >>
> > > > > >> I was just reminded about the real problems this solves (swap isn't
> > > > > >> /dev/sda3, boot is being excessively slow), so, ping?
> > > > > >>
> > > > > >
> > > > > > I'm generally ok with the patchset. The only thing I'm thinking of is if busybox mount
> > > > > > supports PARTUUID syntax in fstab. Can you check this, please?
> > > > > 
> > > > > It does.
> > > > 
> > > > It doesn't, you need util-linux for a mount that figures these out (or,
> > > > it did when I wrote the patch set).
> > 
> > Let me correct myself here.  I think what threw me off for a moment
> > (since I hadn't picked these patches up in a while) was that busybox
> > mount might handle PARTUUID, but busybox swapon/off doesn't.  But that's
> > not quite the point, either.
> > 
> 
> Just checked it out with core-image-minimal:
> 
> root at qemux86-64:~# grep UUID /etc/fstab
> PARTUUID=452e020d-01	/boot	vfat	defaults	0	0
> PARTUUID=452e020d-03	/mnt	ext4	defaults	0	0
> root at qemux86-64:~# mount |grep mnt
> root at qemux86-64:~# mount /mnt
> mount: mounting PARTUUID=452e020d-03 on /mnt failed: No such file or
> directory
> 
> 
> It could be that busybox can be configured to do this, but at least current
> oe-core busybox doesn't understand PARTUUID syntax in fstab.

More clarity required then. :)  busybox mount does support LABEL and
UUID (filesystem UUID, that is) but not PARTUUID.  The kernel doesn't
support LABEL nor UUID (that is to say, root=UUID=xxx works via
initramfs when it's used in other cases).

But the above leaves out one use case, and it's the one that got me
started here, swap.  busybox does not support UUID nor PARTUUID in
swapon/swapoff.  It's an additional feature to have it create swapfiles
with a UUID at all.

> > > This practically means that this patch makes it impossible to use
> > > --use-uuid for busybox-powered images. This is quite a big regression
> > > from my point of view. We need to make this feature optional then.
> > 
> > How does it change anything at all for that case?  Today there's cases
> > where you would want to in some cases use PARTUUID, via --use-uuid, but
> > you can't.  --use-uuid is only valid for the root device as the kernel
> > handles parsing PARTUUID and our fstab only ever has /dev/root listed,
> > and all other entires in the table, even when we say --use-uuid, do not
> > use the UUID but instead use the device name.  Today, we only pass
> > --use-uuid in, in that case.
> > 
> > What 1/2 allows for is passing in --use-uuid for any partition and
> > having it be enforced.  There's no canned wks file changes made at all.
> > 
> > What 2/2 does is update the systemd canned wks (where you will not be
> > able to have busybox swapon/off, you're already forced into util-linux
> > for utils) to make use of --use-uuid.  I would strongly encourage this
> > to be applied as well as it makes systemd-based images much more useful
> > and does not change anything about busybox-only images.  It also allows
> > for external wks files to use UUIDs when they want to.
> 
> This makes sense to me. The patchset doesn't introduce regressions to
> the current functionality.

Agreed.

> It would be nice to at least try to make it
> working with busybox-powered images though.

I'm not sure this would be worth the effort.  Given what I said above,
I don't think it's possible.   We could introduce enough logic to say to
use PARTUUID for root, call out to blkid to get the filesystem UUID and
use that for most filesystems, and then still have the busybox issue for
swap.  PARTUUID is always going to work, with a full util-linux.

> If we can't do that for one or another reason we should at least mention
> it in the documentation and wic help. Please, include this into the
> patch.

OK, I can add a little verbage to help.py.  Is there work that needs
doing to the yocto docs repository as well?

> 
> > > >  That's why 2/2 only updates the
> > > > systemd canned wks as that _will_ have full mount.  But please note that
> > > > you can already make a wks file that doesn't work with busybox mount by
> > > > using the --use-uuid flag for non-root partitions :)
> > >
> > > What's the point of using --use-uuid for non-root partitions? I thought
> > > it only makes sense for root as it triggers update of kernel command
> > > line.
> > 
> > For the same reasons you want to use PARTUUID for the kernel.  You do
> > not know that you're going to be "sda".  You might be "vda" or "mmcblk"
> > or something else.  As a specific example, having "sda3" for swap in the
> > systemd wks file, but your actual device being something else, means
> > that you spend 90s during start-up for systemd trying very hard and
> > being mad that sda3 cannot be swapon'd (it will have swapon'd mmcblk0p3
> > already, automatically).
> 
> My question was about the current functionaily, i.e. without your patch.
> With your patch using --use-uuid for non-root partitions started to make
> sense.

OK.

> P.S. there is a bugzilla item for this:
> https://bugzilla.yoctoproject.org/show_bug.cgi?id=11525
> please, don't hesitate to comment there.

I've grabbed the bugzilla entry since I'm going to push this until it's
done.  Thanks!

-- 
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.openembedded.org/pipermail/openembedded-core/attachments/20171108/c838a7b3/attachment-0002.sig>