Re: "netstat -ranl" and MTU

[Mouse <mouse%Rodents-Montreal.ORG@localhost>]

>> I see that "netstat -ranl" only show MTU for the lo0 address(es).

ITYPM "route(s)" here...?

>> Isn't it suppose to show the local ethernet interface MTU as well?

In my experience, not recently.

>> And NetBSD doesn't seem to store the MTU for all the host routes it
>> learned?

This is at least partially version-dependent.  The versions I run
differ on this point; 1.4T reports an MTU with each route, whereas 2.0
and 4.0.1 just prints a dash for most MTUs.  Looking at 4.0.1's netstat
code, the dash means the MTU fields of their data structures are zero;
looking at the kernel code, I think this means that the route does not
restrict packet size, letting the destination interface, if anything,
do that.

> It does.  I'm seeing one after a test with traceroute -D.

The ICMPs that drive PMTU-D (and traceroute -D) update route MTUs, so
this is perfectly reasonable.

However, if PMTU-D is disabled (net.inet.ip.mtudisc=0), or if the route
has never run into MTU limits, then the route's MTU may have never been
set.  This is the case for most routes for me.

So, yes, NetBSD is capable of storing route MTUs, but it does not
necessarily actually do so for any particular route.  As for "host
routes it learned", I speculate that that depends on how it learned the
route.  For example, host redirects don't carry MTUs, so a route
learned that way might well not have one set.

/~\ The ASCII                             Mouse
\ / Ribbon Campaign
 X  Against HTML                mouse%rodents-montreal.org@localhost
/ \ Email!           7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B