On Fri, 2009-08-07 at 19:09 +0200, Erik Nordmark wrote:
> Some more review:
> 
> net/if_types.h: OK
> os/policy.c: OK
> os/priv_defs: OK
> 
> os/zone.c: There is a comment
> 6034  * Add an data link name for the zone. Does not check for duplicates.
> but the code now checks for duplicates for the same zone.
> Makes sense to update the comment.

ACCEPT: Yes, the comment needs fixing.  It previously didn't check for
duplicates.

> I assume that there is some user-space logic so that bge3 can be 
> assigned to the zone for both IPv4 and IPv6 (which is what the old 
> comments say is the need for allowing duplicates for the same zone.) Or 
> was that logic just unnecessary even before your changes?

That comment and the justification for not checking for duplicates
didn't make sense before.  Assigning a link to a zone involves setting
the link's zone property.  When a link belongs to a zone, it can be
plumbed for IPv4 or IPv6, but the concept of assigning IPv4 or IPv6
separately doesn't make sense.  The link is the thing that is assigned
to the zone.

> You've made zone_add_datalink and zone_remove_datalink no longer be 
> static. But they are a bit hard to use from the kernel in general since 
> they assume a syscall called them (the use set_errno). Do you use them 
> from elsewhere in the kernel? If so it probably make sense to have them 
> return an errno just like zone_check_datalink. Otherwise keep them as 
> static functions.

ACCEPT: Only zone_check_datalink() is meant to be used from the kernel.
All other functions are only used from syscall context and need to be
static.  Will fix.

> 
> KM_SLEEP allocation while holding zone_lock isn't a good idea at
> 6225         idarray = kmem_alloc(sizeof (datalink_id_t) * idcount, 
> KM_SLEEP);

ACCEPT: Indeed, KM_NOSLEEP is more appropriate here.

> usr/src/lib/brand/native/zone/platform.xml: OK
> usr/src/lib/brand/native/zone/config.xml: OK
> 
> Post integration you need to make sure those two file changes also 
> appear in the ipkg brand for opensolaris.

ACCEPT: Yes, I'll be in contact with Dan Price to make sure that gets
done.

> vplat.c: Can we remove the dladm_setzid() from the system as a result of 
> vplat.c no longer using it?

ACCEPT: Yes, in fact, I've removed it. ;-)  (see changes to libdllink.h
and libdllink.c)

Thanks again,
-Seb


_______________________________________________
networking-discuss mailing list
[email protected]

Reply via email to