Xen devel - Aug 2005 - exception looking up device number f,or hda

Hi Mike,

I have the following line in my config file:

disk = [ ''file:/var/images/min-el3-i386.img,hda,w'' ]

and my host doesn''t have a /dev/hda (it has a SATA disk which shows up 
as /dev/sda).

But I would like the guest to see a /dev/hda. Sounds reasonable?

Now, when I try to create this domain, I get:

[2005-08-05 14:40:53 xend] DEBUG (blkif:24) exception looking up device 
number for hda: [Errno 2] No such file or directory:
''/dev/hda''
2005-08-05 14:40:53 xend] DEBUG (blkif:449) Destroying blkif domain=1
[2005-08-05 14:40:53 xend] DEBUG (blkif:337) Destroying vbd domain=1 id=0

The code in util/blkif.py that tries to convert name to device number 
doesn''t seem to be new, but I don''t know why it started
showing up suddenly.

I worked around by manually creating /dev/hda, but I don''t think
it''s
reasonable to expect that host''s /dev contains entries for all guest 
devices.

Also, the effect of having a bad "disk=" line in xmdefconfig (such as 
pointing to a non-existent file) results in a xend crash. We probably 
need to add a few checks at the python level so that the user sees a 
more useful stack trace.

	-Arun

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Hi Arun,

I just pulled the latest unstable and could not reproduce.  I have an 
identical scenario (no /dev/hda only /dev/sda due to SATA).

Also, if the devno resolution code fails you should also get a vbd: 
Device not found exception in the logs.  The exception you are seeing is
not actually a problem, it just means that stat has failed.

We should probably cleanup that error message.

Can you post a more complete log file?  I think the error is being 
caused somewhere else.

Thanks,

Anthony Liguori

Arun Sharma wrote:
> Hi Mike,
>
> I have the following line in my config file:
>
> disk = [ ''file:/var/images/min-el3-i386.img,hda,w'' ]
>
> and my host doesn''t have a /dev/hda (it has a SATA disk which
shows up
> as /dev/sda).
>
> But I would like the guest to see a /dev/hda. Sounds reasonable?
>
> Now, when I try to create this domain, I get:
>
> [2005-08-05 14:40:53 xend] DEBUG (blkif:24) exception looking up 
> device number for hda: [Errno 2] No such file or directory:
''/dev/hda''
> 2005-08-05 14:40:53 xend] DEBUG (blkif:449) Destroying blkif domain=1
> [2005-08-05 14:40:53 xend] DEBUG (blkif:337) Destroying vbd domain=1 id=0
>
> The code in util/blkif.py that tries to convert name to device number 
> doesn''t seem to be new, but I don''t know why it started
showing up
> suddenly.
>
> I worked around by manually creating /dev/hda, but I don''t think
it''s
> reasonable to expect that host''s /dev contains entries for all
guest
> devices.
>
> Also, the effect of having a bad "disk=" line in xmdefconfig
(such as
> pointing to a non-existent file) results in a xend crash. We probably 
> need to add a few checks at the python level so that the user sees a 
> more useful stack trace.
>
>     -Arun
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
>

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Anthony Liguori wrote:
> Hi Arun,
> 
> I just pulled the latest unstable and could not reproduce.  I have an 
> identical scenario (no /dev/hda only /dev/sda due to SATA).
> 
> Also, if the devno resolution code fails you should also get a vbd: 
> Device not found exception in the logs.  The exception you are seeing is
> not actually a problem, it just means that stat has failed.
> 
> We should probably cleanup that error message.
> 
> Can you post a more complete log file?  I think the error is being 
> caused somewhere else.
> 
Hi Anthony,

You''re probably not able to reproduce the problem because
you''re not
using VMX domains?

We''re overloading the semantics of

disk = [ ''file:/var/images/min-el3-i386.img,hda,w'' ]

to create a IDE hard drive for a VMX domain. So we want to take 
advantage of the config file parsing code from the VBD code path, but we 
don''t want to do other tasks such as connecting the backend/frontend
etc.

The log from a successful VMX domain startup looks like:

[2005-08-05 14:54:41 xend] INFO (XendRoot:126) EVENT> xend.domain.create 
[''fc3'', 2]
[2005-08-05 14:54:42 xend] ERROR (process:37) [block-file] ioctl: 
LOOP_SET_FD: Device or resource busy
[2005-08-05 14:54:42 xend] DEBUG (blkif:81) Connecting blkif 
<BlkifBackend frontend=2 backend=0 id=0>
[2005-08-05 14:54:42 xend] DEBUG (blkif:89) send_be_create <BlkifBackend 
frontend=2 backend=0 id=0>
[2005-08-05 14:54:42 xend] DEBUG (image:293) args: cdrom, val: None
[2005-08-05 14:54:42 xend] DEBUG (image:293) args: boot, val: c
[2005-08-05 14:54:42 xend] DEBUG (image:293) args: fda, val: None
[2005-08-05 14:54:42 xend] DEBUG (image:293) args: fdb, val: None
[2005-08-05 14:54:42 xend] DEBUG (image:293) args: localtime, val: None
[2005-08-05 14:54:42 xend] DEBUG (image:293) args: serial, val: None
[2005-08-05 14:54:42 xend] DEBUG (image:293) args: std-vga, val: None
[2005-08-05 14:54:42 xend] DEBUG (image:293) args: isa, val: None
[2005-08-05 14:54:42 xend] INFO (image:352) spawning device models: 
/usr/lib/xen/bin/qemu-dm [''/usr/lib/xen/bin/qemu-dm'',
''
-vncconnect'', ''143.183.130.223:5502'',
''-d'', ''2'', ''-p'',
''19'', ''-m'',
''128'', ''-boot'', ''c'',
''-hda'', ''/var/images/min-el3-i386.i
mg'', ''-vnc'', ''-k'',
''en-us'', ''-vncport'',
''5902'']

i.e. we were doing things which we really shouldn''t be doing, but it 
didn''t hurt. But it started hurting today :)

In the future, we''re also planning to have a VBD driver for unmodified 
OSes running on VMX domains and I was thinking of a syntax such as:

disk = [ ''file:/var/images/min-el3-i386.img,vhda,w'' ]

Any comments/suggestions on the syntax above to present emulated real 
devices as well as virtual devices to an unmodified OS?

	-Arun

Complete log for the failed case:

[2005-08-05 14:40:33 xend] INFO (SrvDaemon:300) Xend Daemon started
[2005-08-05 14:40:33 xend] INFO (channel:148) Virq 3 on port 13
[2005-08-05 14:40:33 xend] DEBUG (XendDomainInfo:171) 
info={''vcpu_to_cpu'': [0, -1, -1, -1, -1, -1, -1, -1, -1, -1,
-1, -1,
-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 
-1, -1], ''cpu_time'': 20984020436L,
''ssidref'': 0, ''s
hutdown_reason'': 0, ''dying'': 0,
''dom'': 0, ''running'': 1,
''maxmem_kb'': -4,
''paused'': 0, ''crashed'': 0,
''vcpus'': 1, ''shutdown'':
  0, ''mem_kb'': 250880, ''cpumap'': [-1, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
  0, 0], ''blocked'': 0}
[2005-08-05 14:40:33 xend] DEBUG (XendDomainInfo:172)
config=''None''

[2005-08-05 14:40:33 xend] INFO (SrvServer:63) unix 
path=/var/lib/xend/xend-socket
[2005-08-05 14:40:33 xend] INFO (XendRoot:126) EVENT> xend.start 0
[2005-08-05 14:40:33 xend] INFO (process:35) [network] network start 
bridge=xen-br0 netdev=eth0 antispoof=no
[2005-08-05 14:40:53 xend] DEBUG (image:135) initDomain: cpu=-1 
mem_kb=131204 ssidref=-1 dom=1
[2005-08-05 14:40:53 xend] DEBUG (XendDomainInfo:673) init_domain> 
Created domain=1 name=ExampleVMXDomain memory=128
[2005-08-05 14:40:53 xend] INFO (XendRoot:126) EVENT> xend.domain.create 
[''ExampleVMXDomain'', 1]
[2005-08-05 14:40:53 xend] INFO (image:190) buildDomain os=vmx dom=1 vcpus=1
[2005-08-05 14:40:53 xend] INFO (image:259) VMX device model port: 3
[2005-08-05 14:40:53 xend] WARNING (XendDomainInfo:924) Unknown config 
field memmap
[2005-08-05 14:40:53 xend] WARNING (XendDomainInfo:924) Unknown config 
field device_model
[2005-08-05 14:40:53 xend] WARNING (XendDomainInfo:924) Unknown config 
field boot
[2005-08-05 14:40:53 xend] WARNING (XendDomainInfo:924) Unknown config 
field vnc
[2005-08-05 14:40:53 xend] WARNING (XendDomainInfo:924) Unknown config 
field display
[2005-08-05 14:40:53 xend] DEBUG (blkif:24) exception looking up device 
number for hda: [Errno 2] No such file or directory
: ''/dev/hda''
[2005-08-05 14:40:53 xend] DEBUG (blkif:449) Destroying blkif domain=1
[2005-08-05 14:40:53 xend] DEBUG (blkif:337) Destroying vbd domain=1 id=0

 >>>> xend dies here <<<

[2005-08-05 14:43:32 xend] INFO (SrvDaemon:300) Xend Daemon started

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Hi Arun
>disk = [ ''file:/var/images/min-el3-i386.img,hda,w'' ]
Please test the following settings. 
disk = [ ''file:/var/images/min-el3-i386.img,hda1,w'' ]

And, it is necessary to rewrite /etc/fstab of domU. 

Thank,

Akio


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

takebe_akio@jp.fujitsu.com wrote:
> Hi Arun
> 
> 
>>disk = [ ''file:/var/images/min-el3-i386.img,hda,w'' ]
> 
> Please test the following settings. 
> disk = [ ''file:/var/images/min-el3-i386.img,hda1,w'' ]
> 
> And, it is necessary to rewrite /etc/fstab of domU. 
> 
I''m not looking for configuration help for domU.

I was looking for suggestions as to how:

a) IDE hard drives for VMX domains should be configured
b) VBD based devices for VMX domains should be configured

and the bad failure mode for (a) when the user misconfigures it.

	-Arun

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Arun Sharma wrote:
> Hi Mike,
> 
> I have the following line in my config file:
> 
> disk = [ ''file:/var/images/min-el3-i386.img,hda,w'' ]
> 
> and my host doesn''t have a /dev/hda (it has a SATA disk which
shows up
> as /dev/sda).
> 
> But I would like the guest to see a /dev/hda. Sounds reasonable?
> 
> Now, when I try to create this domain, I get:
> 
> [2005-08-05 14:40:53 xend] DEBUG (blkif:24) exception looking up device 
> number for hda: [Errno 2] No such file or directory:
''/dev/hda''
> 2005-08-05 14:40:53 xend] DEBUG (blkif:449) Destroying blkif domain=1
> [2005-08-05 14:40:53 xend] DEBUG (blkif:337) Destroying vbd domain=1 id=0
> 
> The code in util/blkif.py that tries to convert name to device number 
> doesn''t seem to be new, but I don''t know why it started
showing up
> suddenly.
> 
> I worked around by manually creating /dev/hda, but I don''t think
it''s
> reasonable to expect that host''s /dev contains entries for all
guest
> devices.
No, and the blkdev_name_to_number() function should convert names like /dev/hda
without the device existing. This code should be trapping the error:

     try:
         return os.stat(n).st_rdev
     except Exception, ex:
         log.debug("exception looking up device number for %s: %s",
name, ex)
         pass

The code then uses regexps to analyse the device name (though you will see
the debug message in the log).


What xen version are you using?
> 
> Also, the effect of having a bad "disk=" line in xmdefconfig
(such as
> pointing to a non-existent file) results in a xend crash. We probably 
> need to add a few checks at the python level so that the user sees a 
> more useful stack trace.
All errors during domain construction are intended to be trapped and returned
to the caller - so maybe something got broken.
Did you see anything in the xend log?
Might be worth causing the error using xend with tracing on (xend trace_start)
and having a look at the trace (/var/log/xend.trace).

Mike

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

> I was looking for suggestions as to how:
>
> a) IDE hard drives for VMX domains should be configured
> b) VBD based devices for VMX domains should be configured
Ah, I guess we have to distinguish the two...

Hope about an optional: vbd:/emu: prefix (or similar) on the second part of 
the triplet for a disk?  The default (for non VMX domains) could be
"vbd" so
existing behaviour would remain the same.

If / when we move over to using "xvd" for VBDs I guess that could
subsume this
functionality (might still need it for non Linux guests, tho).

Thoughts?

Cheers,
Mark
> and the bad failure mode for (a) when the user misconfigures it.
>
> 	-Arun
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Mark Williamson wrote:
>>I was looking for suggestions as to how:
>>
>>a) IDE hard drives for VMX domains should be configured
>>b) VBD based devices for VMX domains should be configured
> 
> 
> Ah, I guess we have to distinguish the two...
> 
> Hope about an optional: vbd:/emu: prefix (or similar) on the second part of
> the triplet for a disk?  The default (for non VMX domains) could be
"vbd" so
> existing behaviour would remain the same.
Sounds good to me. How about:

hda 		= VBD hda
vbd://hda 	= VBD hda
ioemu://hda	= IDE hda

The VBD patches are not ready yet - so I''d like to propose that we do:

-            self.create_configured_devices()
+           if self.image.ostype != ''vmx'':
+                self.create_configured_devices()

and

-        self.create_blkif()
+       if self.image.ostype != ''vmx'':
+            self.create_blkif()

When our VBD patches are ready, we''ll revert this.

	-Arun

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Arun Sharma wrote:
> Sounds good to me. How about:
> 
> hda         = VBD hda
> vbd://hda     = VBD hda
> ioemu://hda    = IDE hda
In URIs the // syntax indicates a global naming authority such as DNS
(see RFC 2396 section 3.2). Here the scheme-specific-part is in a local
namespace, so it should be just "vbd:hda", for example.

-- 
David Hopwood <david.nospam.hopwood@blueyonder.co.uk>


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

> > Hope about an optional: vbd:/emu: prefix (or similar) on the second
part
> > of the triplet for a disk?  The default (for non VMX domains) could be
> > "vbd" so existing behaviour would remain the same.
>
> Sounds good to me. How about:
>
> hda 		= VBD hda
> vbd://hda 	= VBD hda
> ioemu://hda	= IDE hda
Sounds fine but I think I''d actually ditch the // to fit in with the
phy: and
file: syntax we already have. (and as David said in another post, to avoid 
looking like a URL).
> The VBD patches are not ready yet - so I''d like to propose that we
do:
>
> -            self.create_configured_devices()
> +           if self.image.ostype != ''vmx'':
> +                self.create_configured_devices()
>
> and
>
> -        self.create_blkif()
> +       if self.image.ostype != ''vmx'':
> +            self.create_blkif()
>
> When our VBD patches are ready, we''ll revert this.
Is there still a need for these changes if the ioemu: prefix is implemented?

Cheers,
Mark

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

I''ve already made a patch when I doing para-driver debugging.
see the attach patch.
Does it fit the requirement?
It will skip setupping up the loop device for ioemu image files,
but by xm vbd-list, it can still see the file bind to hda.
so  this command is still useful.


Mark Williamson wrote:
>>>Hope about an optional: vbd:/emu: prefix (or similar) on the second
part
>>>of the triplet for a disk?  The default (for non VMX domains) could
be
>>>"vbd" so existing behaviour would remain the same.
>>
>>Sounds good to me. How about:
>>
>>hda 		= VBD hda
>>vbd://hda 	= VBD hda
>>ioemu://hda	= IDE hda
> 
> 
> Sounds fine but I think I''d actually ditch the // to fit in with
the phy: and
> file: syntax we already have. (and as David said in another post, to avoid 
> looking like a URL).
> 
> 

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Xiaofeng Ling wrote:
> I''ve already made a patch when I doing para-driver debugging.
> see the attach patch.
> Does it fit the requirement?
> It will skip setupping up the loop device for ioemu image files,
> but by xm vbd-list, it can still see the file bind to hda.
> so  this command is still useful.
> -disk = [ ''file:/var/images/min-el3-i386.img,hda,w'' ]
> +disk = [ ''ioemu:/var/images/min-el3-i386.img,hda,w'' ]
The ioemu: prefix should be on the second element i.e:

+disk = [ ''file:/var/images/min-el3-i386.img,ioemu:hda,w'' ]

and it should allow:

+disk = [ ''phy:hdb,ioemu:hda,w'' ]

	-Arun

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Mark Williamson wrote:
>
>>The VBD patches are not ready yet - so I''d like to propose that
we do:
>>
>>-            self.create_configured_devices()
>>+           if self.image.ostype != ''vmx'':
>>+                self.create_configured_devices()
>>
>>and
>>
>>-        self.create_blkif()
>>+       if self.image.ostype != ''vmx'':
>>+            self.create_blkif()
>>
>>When our VBD patches are ready, we''ll revert this.
> 
> 
> Is there still a need for these changes if the ioemu: prefix is
implemented?
We''ll implement the ioemu prefix ASAP - just trying to make sure that 
3.0-testing ships with working VMX support :)

	-Arun


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

> > Is there still a need for these changes if the ioemu: prefix is
> > implemented?
>
> We''ll implement the ioemu prefix ASAP - just trying to make sure
that
> 3.0-testing ships with working VMX support :)
Sure thing :-)

I wouldn''t have thought the ioemu: prefix should be too hard to
implement.  If
it looks like taking a while you could try submitting a patch with your 
workaround before the -testing release.

Cheers,
Mark

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel