[Libvir] Re; virDomainBlockStats

[Daniel P. Berrange]

On Thu, Jan 17, 2008 at 04:44:31PM +0000, Gareth Bult wrote:
> Hi, 
> 
> I've a number of servers running XEN and am trying to write some code to monitor them through libvirt. 
> (Xeon/Dual Core/Ubuntu Gutsy/Xen 3.1) 
> 
> All was going well until I got to virDomainBlockStats and it simply refused to accept my coding .. 
> 
> My solution (the code now works) is to change the following; 
> 
> src/stats_linux.c: 
> 
> #ifndef XENVBD_MAJOR 
> #define XENVBD_MAJOR 3 // <--- I changed this! 
> /*#define XENVBD_MAJOR 202*/ 
> #endif 
> 
> device = XENVBD_MAJOR * 256 + 1 + minor /* * 16 */; <-- this too! 
> 
> It now seems to give me some meaningful numbers without feeding my any error messages. 
> Can anyone tell me "why" I needed to do this? 
> 
> I'm using "xvda" as the device path .. tried every other device name 
> I could think of with out any good results ... 

Actually looking at the xenstore-ls output that you sent me (offlist),
you are not using xvda - you're using hda1

     9 = ""
      769 = ""
       domain = "mgm"
       frontend = "/local/domain/9/device/vbd/769"
       uuid = "d6d17ae3-c08c-8a5b-fbc6-3258ea5e06bf"
       dev = "hda1"
       state = "4"
       params = "/cluster/images/domains/mgm/disk.img"
       mode = "w"
       online = "1"
       frontend-id = "9"
       type = "file"
       node = "/dev/loop13"
       physical-device = "7:d"
       hotplug-status = "connected"
       feature-barrier = "1"
       sectors = "4194304"
       info = "0"
       sector-size = "512"

That is a minor detail though. The file: vs tap:aio: is actually a
red herring. Having examined the code for calculating device numbers
I see that it is flawed in multiple ways

 - It doesn't add in the partition number to device number. So 'hdc1'
   gets calculated as 'hdc' which fails
 - It uses completely wrong formula for IDE disks. It is not simply
   HD_MAJOR * 256 + minor * 16. There are in fact 10 IDE major numbers
   each holding 2 disks, with 63 partitions.
 - It doesn't handle SCSI disks at all.
 - It doesn't strip the leading /dev/  if given a fully qualified dev

Basically only xvda -> xvdp with no partitions, and hda with no partitions
was working. This is why it failed for you with 'hda1' and most other
example disks.

Dan.
-- 
|=- Red Hat, Engineering, Emerging Technologies, Boston.  +1 978 392 2496 -=|
|=-           Perl modules: http://search.cpan.org/~danberr/              -=|
|=-               Projects: http://freshmeat.net/~danielpb/               -=|
|=-  GnuPG: 7D3B9505   F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505  -=|