While working on documenting the nwam service properties, I
came across some inconsistencies in what's defined vs. what's
used:
Defined in network-physical.xml:
debug
upgraded
use_net_svc
autoconf
dhcp_wait_time
scan_interval
scan_level
strict_bssid
active_ncp
value_authorization
Actually consumed by nwamd:
debug
upgraded
autoconf
scan_interval
scan_level
strict_bssid
active_ncp
There are three properties that appear to be ignored by nwamd. One
is value_authorization; I think that's something that's needed by
the smf framework (I need to verify that; but it was there in
phase 0/0.5, and did not appear to be used by nwamd then, either).
The remaining two defined-but-not-used properties are use_net_svc
and dhcp_wait_time.
use_net_svc was added specifically for punchin. People were having
problems where nwam would get them connected somewhere, they would
punchin, and punchin would set up a swan-specific resolv.conf; some
time later, a network interruption would cause nwam to tear down and
immediately rebuild the connection, reacquire the lease, and re-run
the network/service start method, resulting in the swan-specific
resolv.conf being overwritten by the user's local resolv.conf. But
if the address didn't actually change, the punchin session would
survive this interruption, only name services would no longer work.
So the use_net_svc property was created, with a default value of
true. When punching in, the punchin code would set it to false,
and back to true when punching out. This property thus indicated
whether or not the network/service start method should be run to
re-create the resolv.conf file on lease acquisition.
We clearly no longer need this in its current state. We probably
do need to think about how this scenario will play out in the
phase 1 world; but I suspect this property will no longer be the
right answer, no matter what. So I recommend removing it.
dhcp_wait_time specified how long to wait for a DHCP server response
before moving on to the next available llp. I initially thought we
didn't need this one anymore, either; but as I've started thinking
about user expectations and different dependencies/priorities, I'm
starting to think we do need it, or something like it. But that's
another can of worms, I'll do a separate e-mail for it.
So, bottom line here: any disagreement with my recommendation to
remove the use_net_svc property for phase 1?
Thanks,
renee
