Restarting interfaces and routing table stickiness

[Hiroki Sato]

Garrett Cooper <yanegomi at gmail.com> wrote
  in <CAGH67wSemVK2FtGHgHX2tMicHd7fEs27M6_RCxdKiJxFPveRCQ at mail.gmail.com>:

ya> Hi -net!
ya>     I've been doing some IPv6 testing lately, and one of the issues
ya> that I've run into in the past (since at least 7.0-CURRENT) is that if
ya> I do `service netif restart <foo>`, e.g. `service netif restart`
ya> multiple times, and have addresses statically configured in rc.conf,
ya> more often than not the routing table doesn't properly get flushed and
ya> my addresses become unreachable (even after I call `service routing
ya> restart`). Recently, I ran into an issue with 9.1-RC1 where had to
ya> kill ntpd before it would allow me to reach one of the addresses I had
ya> assigned on a test server (server has a total of 4 physical NICs and 1
ya> vlan'ed interface). The only surefire method to get things back to a
ya> sane state is to reboot the box (of course, please keep in mind that
ya> I'm using the netif rc script and not using other commands like route
ya> flush, route delete, etc).
ya>     This behavior can be simply triggered like so on 8.x and below like so:
ya>
ya>     /etc/rc.d/network_ipv6 restart
ya>     /etc/rc.d/network_ipv6 restart

 Do you have a procedure which can reproduce the problem on 9.x and/or
 output with rc_debug="YES"?  I guess the cause was that
 initialization of the routing table and interface configuration are
 separated, and it ended up being some incomplete configuration when
 restarting.

-- Hiroki
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 196 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-net/attachments/20120913/f459613e/attachment.pgp