[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [libvirt] Some problem with the save function

Le mercredi 16 septembre 2009 à 21:36 +0100, Daniel P. Berrange a
écrit :
> On Wed, Sep 16, 2009 at 05:05:36PM +0200, Daniel Berteaud wrote:
> > Hi everyone. This is my first post to this list, so excuse me if it's
> > not the right place to poste info like this.
> > 
> > I've found some problems in qemu_driver.c in the latest versions of
> > libvirt:
> > 
> > - the first problem was also present in 0.7.0. If we run qemu guests
> > unprivileged, we cannot use the save function (virsh save
> > guestname /path/to/file). The reason seems to be that libvirt first
> > create the destination file to write the header and the XML definition.
> > The file is owned by root with 600 permission. Then, we ask qemu to
> > append the dump of the memory (via the migrate exec: call) to this file.
> > This part doesn't work because the qemu user cannot write to this file.
> > So we endup with a file which only contains the header and the XML, and
> > the guest is stoped :/
> Doh, that's a nice problem. You can work around it by editing
> /etc/libvirt/qemu.conf and telling qemu to run as root:root
> again.  For a real fix we'll need to make the save method 
> run 'fchown' on the file descriptor after writing the header.
> And then fchown it back to root:root once complete, to stop
> any other guest overwriting it.

As I only use the save function in a script, I work around it by
touching the file and chowning it to qemu:qemu before calling save, I
sleep better when my guests runs with less privileges ;)

> > - the second problem is present since libvirt 0.7.1. Now that the saved
> > file can be compressed, it seems we cannot save in a raw format any
> > more. This is due to this part in the code (qemu_driver.c):
> > 
> > if (STREQ (prog, "raw"))
> >         prog = "cat";
> >     internalret = virAsprintf(&command, "migrate \"exec:"
> >                               "%s -c >> '%s' 2>/dev/null\"", prog,
> > safe_path);
> > 
> > which result in "migrate \"exec cat -c >> safe_path 2>/dev/null\""
> > 
> > But cat doesn't support the -c argument, so once again, the save fails,
> > as we end up with a save file which only contains the header and the XML
> > definition.
> Wierd, I don't know where/when we gained a '-c' arg to cat but it
> looks rather bogus. 

The change is in this patch :

> > - the third problem, is that restore doesn't work any more. I haven't
> > dig it yet, but when I try to restore a guest using a saved file
> > (uncompressed produced under libvirt 0.7.0), I've this error:
> > 
> > virsh restore /tmp/guest.state
> > error: Failed to restore domain from /tmp/guest.state
> > error: internal error unable to start guest: 
> Hmm, bad error message :-(  We might also need todo a chown()
> in the restore path to allow QEMU to read it. NB there is no
> compatability between QEMU version, so if you have upgraded
> your install of QEMU between the time you saved & restored
> it is very likely to crash & burn.

I haven't upgraded qemu, just libvirt. And I must admit that I don't
really know where and how to debug it, I'm very new to libvirt ;) (and I
really don't know anything in C)

Would just like to know if someone could get it working in 0.7.1, or if
there's a big bug somewhere in this release.

Regards, Daniel

> This is something QEMU devs are working on fixing by allowing
> for stable machine ABI (available in Fedora 12), and stable
> pci device addressing (targetted for Fedora 13).
> Regards,
> Daniel
Daniel Berteaud
Société de Services en Logiciels Libres
Technopôle Montesquieu
Tel : 05 56 64 15 32
Fax : 05 56 64 15 32
Mail: daniel firewall-services com
Web : http://www.firewall-services.com

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]