[PATCH] add count-only option to DOM0_GETDOMAININFOLIST hypercall

[PATCH] add count-only option to DOM0_GETDOMAININFOLIST hypercall

[Mike D. Day]

signed-off-by: Mike D. Day <ncmike@us.ibm.com>

When retreiving a list of domain info structs it would be helpful
to know the current domain count prior to making the hcall. This 
would make it simpler to allocate a return buffer of the 
appropriate size. 

Adding the six lines to the case for the GETDOMAININFOLIST dom0 op
allows the caller to retrieve only the count of the domains. The 
use model is as follows: 

    int number_domains;
    struct dom0_op info;
    info.cmd = DOM0_GETDOMAININFOLIST;
    info.u.getdomaininfolist.max_domains = 0;

    HYPERCALL_dom0_op(&info);

    number_domains = info.u.getdomaininfolist.max_domains;

Having the number of domains, the caller can now allocate
a buffer of the correct size to get the actual domain 
information.    

# HG changeset patch
# User mdday@dual.silverwood.home
# Node ID cd6869742400411fdaf295cc4e0167361e967619
# Parent  5d3c2cb42ec41984cb1e586d3e47a8692eb8b132
modify domain infolist hypercall to optionally return only a count of domains

diff -r 5d3c2cb42ec4 -r cd6869742400 xen/common/dom0_ops.c
--- a/xen/common/dom0_ops.c	Wed Mar 22 19:05:50 2006 +0100
+++ b/xen/common/dom0_ops.c	Wed Mar 22 19:30:28 2006 -0500
@@ -416,6 +416,12 @@ long do_dom0_op(GUEST_HANDLE(dom0_op_t) 
 
         for_each_domain ( d )
         {
+            if (op->u.getdomaininfolist.max_domains == 0) 
+            {
+                num_domains++;
+                continue;
+            }
+                
             if ( d->domain_id < op->u.getdomaininfolist.first_domain )
                 continue;
             if ( num_domains == op->u.getdomaininfolist.max_domains )

Re: [PATCH] add count-only option to DOM0_GETDOMAININFOLIST hypercall

[Anthony Liguori]

There's a race condition here.  There's no way of knowing that domains 
haven't been destroyed (or created) in between these calls.

Unfortunately, the only way to avoid this is grab them one at a time.

Regards,

Anthony Liguori

Mike D. Day wrote:
> signed-off-by: Mike D. Day <ncmike@us.ibm.com>
>
> When retreiving a list of domain info structs it would be helpful
> to know the current domain count prior to making the hcall. This 
> would make it simpler to allocate a return buffer of the 
> appropriate size. 
>
> Adding the six lines to the case for the GETDOMAININFOLIST dom0 op
> allows the caller to retrieve only the count of the domains. The 
> use model is as follows: 
>
>     int number_domains;
>     struct dom0_op info;
>     info.cmd = DOM0_GETDOMAININFOLIST;
>     info.u.getdomaininfolist.max_domains = 0;
>
>     HYPERCALL_dom0_op(&info);
>
>     number_domains = info.u.getdomaininfolist.max_domains;
>
> Having the number of domains, the caller can now allocate
> a buffer of the correct size to get the actual domain 
> information.    
>
> # HG changeset patch
> # User mdday@dual.silverwood.home
> # Node ID cd6869742400411fdaf295cc4e0167361e967619
> # Parent  5d3c2cb42ec41984cb1e586d3e47a8692eb8b132
> modify domain infolist hypercall to optionally return only a count of domains
>
> diff -r 5d3c2cb42ec4 -r cd6869742400 xen/common/dom0_ops.c
> --- a/xen/common/dom0_ops.c	Wed Mar 22 19:05:50 2006 +0100
> +++ b/xen/common/dom0_ops.c	Wed Mar 22 19:30:28 2006 -0500
> @@ -416,6 +416,12 @@ long do_dom0_op(GUEST_HANDLE(dom0_op_t) 
>  
>          for_each_domain ( d )
>          {
> +            if (op->u.getdomaininfolist.max_domains == 0) 
> +            {
> +                num_domains++;
> +                continue;
> +            }
> +                
>              if ( d->domain_id < op->u.getdomaininfolist.first_domain )
>                  continue;
>              if ( num_domains == op->u.getdomaininfolist.max_domains )
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
>   

Re: [PATCH] add count-only option to DOM0_GETDOMAININFOLIST hypercall

[Mike D. Day]

Anthony Liguori wrote:
> There's a race condition here.  There's no way of knowing that domains 
> haven't been destroyed (or created) in between these calls.
> 
> Unfortunately, the only way to avoid this is grab them one at a time.

The race condition is harmless and the patch is an improvement over the 
current situation where you have to guess the number of domains (and 
hence the size of the buffer you allocate).

Most of the time the caller will be looking for a snapshot of domain 
information, not an up-to-the-instant comprehensive report.

And the pre-existing method of allocating a huge buffer and calling once 
still works.

Mike