responses to Dave's code review comments follow...

> ndp.xml:
>
> 34: I think this should be ndp, not in.ndpd; at least that's the
> convention in a couple of other manifests I checked. This comment also
> applies to the other manifests.

accepted.

> 67: After looking at svc-ndp, why isn't ":kill" sufficient? And there
> is certainly no reason to pass the arguments.
>
>
> 77: Similarly, ":kill -HUP" seems like it would do the job.

yep, :kill and :kill -HUP are indeed sufficient, i'll make the the changes.

> 92: please say "group" instead of "pg"; would not "reserved" be more
> correct than "accessible"? This comment also applies to the other
> manifests.

accepted.

> 105: perhaps "neighbor discovery" would be better than "autoconfiguration"?

sure.

> svc-ndp:
>
> This whole file would be superfluous if SMF provided a way to tag
> services as global-zone-only; I believe this RFE exists already, but if
> not, please file it; at least add a comment noting it would be useful
> for these services, too. Anyway, related to my comments above, you only
> need the "start" portion here, the rest would be replaced by the
> contract signal functionality.

couldn't find a similar bug/RFE, as a result i've filed:

CR6431298: smf could provide a way of tying exec methods to global/non-global 
zones

and added you to the interest list.


> svc-ripng:
>
> 42: So why isn't it better to replace the /etc/.dynamic_routing file
> used here and in other method scripts with a boolean property on the
> network/routing service?

it's much better - i've made this change and excised references to 
.dynamic_routing as appropriate, replacing the file with a boolean 
dynamic_routing
application property in network/routing which is set to true by the start 
methods 
of routing daemons (it's set to false in network/routing if global routing 
services ipv4(6)-routing are both disabled, since this ensures no routing 
daemons can run).

> in.rdisc.c: this file shouldn't have been changed at all.
>

it was just moved, i updated the copyright info since wx pbchk was complaining.
maybe this is wx ignorance on my part - it still needs to be in the active list
due to the move, right?

> routing.xml: could you add comments as to the reason for each dependency?

sure.

> svc-routeadm:
>
> 44: this whole nawk script is sufficient evidence that routeadm -p ought
> to be enhanced. At the very least please file an RFE to add
> functionality sufficient to eliminate this, if you don't have the
> ability to add it in this project.

good idea - how about supporting:

# routeadm -p current/option_name
# routeadm -p default/option_name
# routeadm -p option_name (default gives current option value)

in addition to routeadm -p with no arguments? this may require
a PSARC fasttrack, i'll look into it.

> 82: wouldn't it be advisable to use "cp -p" here?
> 90: s/invokation/invocation/

both accepted.

> 311: why do we need a pid file? and is /var/tmp the best place for it?

i was following legacy routeadm practice here - it monitored the
running state of such daemons using pidfiles in /var/tmp. would /tmp
be better, to ensure pidfiles don't hang around post-reboot?

> net-init:
>
> 32-37: I'd suggest just referencing the service name, not the script
> name, in these comments; yeah, I know you're following what's there. It
> was wrong.

sure.

> network-initial.xml:
>
> 68: I don't think the dependency is necessary - see comments later about
> seed.

accepted.

> net-loopback:
>
> 52: I don't get why the service enables that are done in net-routing
> based on this aren't just put here.

if i can remember i was trying to keep all routing-related configuration
in network/routing, though introducing another /etc file is a pretty
lame way of doing it. also, thinking about it i suspect that on first boot, 
since manifest-import might not have run at this point, these service state 
changes would fail (whether they were made by an in-seed network/routing
service or by network/loopback). maybe the right answer here is adding
an initialize_routing property to network/loopback that network/routing 
(no longer in the seed as you suggest below) can pick up after manifest-import
runs?

> network-routing.xml
>
> Comments explaining the reason for each dependency would be appreciated.

accepted.

> svc/seed/Makefile
>
> I'm not sure network/routing needs to be in the seed - I can't think of
> a reason why it's needed to get any supported configuration to boot to
> the point that manifest-import runs. I realize that it's because you
> placed a dependency on this in network/initial, but it doesn't seem to
> me that the dependency is necessary - routes come and go, nobody should
> be depending on whether they're set up or not.

understood. i'll excise network/routing from the seed.

and thanks again for your help! i'll follow up with an updated webrev shortly,
once i've tested the above changes.

alan
 
 
This message posted from opensolaris.org
_______________________________________________
networking-discuss mailing list
[email protected]

Reply via email to