libvirt users - Aug 2016 - Re: Libvirt: dynamic ownership did not work

Am Do, 4. Aug, 2016 um 5:07 schrieb Michal Privoznik 
<mprivozn@redhat.com>:
> On 04.08.2016 13:59, Jonatan Schlag wrote:
>> 
>> 
>>  Am Do, 4. Aug, 2016 um 1:38 schrieb Michal Privoznik 
>> <mprivozn@redhat.com>:
>>>  On 04.08.2016 12:12, Jonatan Schlag wrote:
>>>> 
>>>> 
>>>>   Am Do, 4. Aug, 2016 um 11:32 schrieb Michal Privoznik
>>>>   <mprivozn@redhat.com>:
>>>>>   On 03.08.2016 21:17, Jonatan Schlag wrote:
>>>>>>    Hi,
>>>>>>    I have a very strange problem with libvirt. I work
on some
>>>>>> machines
>>>>>>   with
>>>>>>    libvirt (Debian/ Arch Linux) and libvirt set the
ownership of
>>>>>> images
>>>>>>    file automatically to the qemu user / group for
example on
>>>>>> Arch
>>>>>>   Linux to
>>>>>>    nobody:kvm.
>>>>>>    So when I copy an image file with root and use I
then with
>>>>>> qemu,
>>>>>>   libvirt
>>>>>>    change the owner/ group to nobody:kvm.
>>>>>> 
>>>>>>    But I also compiled libvirt for a machine (gcc 4.9.4
glibc
>>>>>> 2.12)
>>>>>>  and on
>>>>>>    this machine libvirt did not change the ownership of
the
>>>>>> image files
>>>>>>    which results in this error:
>>>>>> 
>>>>>>    libvirtError: internal error: process exited while
connecting
>>>>>> to
>>>>>>    monitor: able-ticketing,seamless-migration=on
-device
>>>>>> 
>>>>>> 
>>>>>>  
>>>>>>
qxl-vga,id=video0,ram_size=67108864,vram_size=67108864,vram64_size_mb=0,vgamem_mb=16,bus=pci.0,addr=0x2
>>>>>> 
>>>>>> 
>>>>>>    -device intel-hda,id=sound0,bus=pci.0,addr=0x4
-device
>>>>>>    hda-duplex,id=sound0-codec0,bus=sound0.0,cad=0
-chardev
>>>>>>    spicevmc,id=charredir0,name=usbredir -device
>>>>>>    usb-redir,chardev=charredir0,id=redir0 -device
>>>>>>    virtio-balloon-pci,id=balloon0,bus=pci.0,addr=0x7
-msg
>>>>>> timestamp=on
>>>>>>    2016-08-03T18:19:47.494512Z qemu-system-x86_64:
-drive
>>>>>> 
>>>>>> 
>>>>>>  
>>>>>>
file=/data/hdd1/libvirt/images/test.img,format=raw,if=none,id=drive-virtio-disk0:
>>>>>> 
>>>>>> 
>>>>>>    Could not open
'/data/hdd1/libvirt/images/test.img':
>>>>>> Permission
>>>>>>  denied
>>>>> 
>>>>>   Can you please share the debug logs?
>>>>> 
>>>>>   http://wiki.libvirt.org/page/DebugLogs
>>>>> 
>>>>>   Also, my initial suspect, before diving any deeper is
that
>>>>> usually,
>>>>>  when
>>>>>   users compile libvirt on their own, they forget to set
the
>>>>> correct
>>>>>   prefix, therefore libvirt is looking for its config files
NOT
>>>>> under
>>>>>   /etc/libvirt but /usr/local/etc/ or whatever.
>>>>> 
>>>>>   BTW: is the daemon running under root?
>>>>> 
>>>>>   Michal
>>>> 
>>>>   Hi,
>>>> 
>>>>   The daemon runs under root.
>>>> 
>>>>   I uploaded the debug logs to:
>>>> 
>>>>   http://people.ipfire.org/~jschlag/1363864/1_libvirtd.log
>>>> 
>>>>   The UID of the user nobody is 99, the GID of the group kvm is
>>>> 1011.
>>>> 
>>>>   I added my configure options to the bug report.
>>>> 
>>>>   Following the log the ownership is changed but why is the
file
>>>> still
>>>>   owned by root:root?
>>> 
>>>  Right. the file is set ownership.
>>  But the file ist still owned by root:root and so it is not 
>> accessable by
>>  qemu as nobody:kvm. In the moment the only possible way is that the
>>  change of the ownership fail, but then there should be an error 
>> message,
>>  but there is no error message in the log.
> 
> Then the other option that comes to my mind is a race with somebody 
> else
> on the system. You can attach gdb to the daemon and set breakpoint to
> virSecurityDACSetOwnershipInternal(). In the arguments you should see
> the path eventually among with uid:gid.
> 
> BTW: what's the domain XML?
What did you need the xml file the domain is called test and the image 
file is /data/hdd1/libvirt/images/test.img
> 
> 
> Michal
I did together with Michael Tremer some debugging and Michael posted 
our results in the bugtracker. So it seems that the chown function is 
not executed, because a other function return a wrong value.

Maybe the describtion in the bug report hepls to go furhter, when not 
say what you need (logs) to debug the problem.

Regards Jonatan

On 04.08.2016 20:28, Jonatan Schlag wrote:
> 
> 
>> Then the other option that comes to my mind is a race with somebody
else
>> on the system. You can attach gdb to the daemon and set breakpoint to
>> virSecurityDACSetOwnershipInternal(). In the arguments you should see
>> the path eventually among with uid:gid.
>>
>> BTW: what's the domain XML?
> 
> What did you need the xml file the domain is called test and the image
> file is /data/hdd1/libvirt/images/test.img
Well, in the domain XML. there's <seclabel/> section in the domain XML
that can fine tune relabelling for a domain. Moreover, some devices -
like disks have the <seclabel/> too. And I was wondering whether you
don't have those elements in the XML.

The other reason for me asking domain XML is so that I could try to
reproduce locally on my system.
> 
> I did together with Michael Tremer some debugging and Michael posted our
> results in the bugtracker. So it seems that the chown function is not
> executed, because a other function return a wrong value.
Ah, reading the bz transcript, you are not passing the path directly
into the XML rather than use a volume from a storage pool. This is
supported but the previous case is more tested. Again, this would help
me to narrow down the possible causes.
> 
> Maybe the describtion in the bug report hepls to go furhter, when not
> say what you need (logs) to debug the problem.
Maybe I'm misreading this, but I think I've told you what I need to
debug the problem. Moreover, it's usually better to provide as many
information as possible when debugging a problem. Even a tiny little
thing that user think of as trivial may look crucial in eyes of
experienced developer with insight in the project.


Anyway, I'd like to continue the discussion in the BZ.

Michal

Am Fr, 5. Aug, 2016 um 9:45 schrieb Michal Privoznik 
<mprivozn@redhat.com>:
> On 04.08.2016 20:28, Jonatan Schlag wrote:
>> 
>> 
> 
>>>  Then the other option that comes to my mind is a race with 
>>> somebody else
>>>  on the system. You can attach gdb to the daemon and set breakpoint
>>> to
>>>  virSecurityDACSetOwnershipInternal(). In the arguments you should 
>>> see
>>>  the path eventually among with uid:gid.
>>> 
>>>  BTW: what's the domain XML?
>> 
>>  What did you need the xml file the domain is called test and the 
>> image
>>  file is /data/hdd1/libvirt/images/test.img
> 
> Well, in the domain XML. there's <seclabel/> section in the
domain XML
> that can fine tune relabelling for a domain. Moreover, some devices -
> like disks have the <seclabel/> too. And I was wondering whether you
> don't have those elements in the XML.
> 
> The other reason for me asking domain XML is so that I could try to
> reproduce locally on my system.
> 
>> 
>>  I did together with Michael Tremer some debugging and Michael 
>> posted our
>>  results in the bugtracker. So it seems that the chown function is 
>> not
>>  executed, because a other function return a wrong value.
> 
> Ah, reading the bz transcript, you are not passing the path directly
> into the XML rather than use a volume from a storage pool. This is
> supported but the previous case is more tested. Again, this would help
> me to narrow down the possible causes.
> 
>> 
>>  Maybe the describtion in the bug report hepls to go furhter, when 
>> not
>>  say what you need (logs) to debug the problem.
> Maybe I'm misreading this, but I think I've told you what I need to
> debug the problem. Moreover, it's usually better to provide as many 
> information as possible when debugging a problem. Even a tiny little 
> thing that user think of as trivial may look crucial in eyes of 
> experienced developer with insight in the project.
Hi,
I am sorry this sentence causes more trouble, then he helps.
First, I posted the domain XML and the storage XML file in Bugzilla.
I could not post this files earlier because I had no time to do this.

I will post what I can to help you and I will not hide any information 
which could help, but at the moment I have really no idea what I can 
provide what can help.
So the intention of this sentence was, to say that I will provide what 
I can to help and not that I provide only the bit what you want.

Regards Jonatan
> 
> Anyway, I'd like to continue the discussion in the BZ.
Ok, I will post now everything in the bugtracker.
> 
> 
> Michal