Add corresponding changes to the tap interface.

Now I see what you are referring to. Thanks for the clarification. But this is a different issue. The tun/tap interfaces look like an interface and also like a socket. Setting the MTU via the interface way works as expected. The problem is only when changing it via the socket interface. The other tunnelling interfaces only look like interfaces and share a common method for changing the MTU. This works for the tun/tap interfaces but your example shows that there is an issue related to the outer/inner stuff. I might be able to look into this, but this is definitely a different issue.

Why are also other tunnelling interfaces except for if_tap.c affected? Do they have a socket option to change the MTU? Right now, rt_updatemtu() seems only to be called from if.c, so integrating nd6_setmtu() into rt_updatemtu() can be done. But this would be a separate cleanup in my view.
So I can update the patch to cover also if_tap.c, but for the other tunnelling interfaces could you elaborate why they are also affected?

Thanks for the fix. I never tested on a system without SCTP support.

Hi John,
here is the TCP module

In D16794#361490, @jhb wrote:

Michael, I would like to get a copy of your kernel and tcprack.ko when you get some spare time.

I will see how to get this to you. I did a buildworld / installworld yesterday with the result that the system does not boot anymore.
This is NOT related to the change of this review. I posted the error message on the ppc mailing list...

In D16839#358734, @dteske wrote:

I would just like to comment for the record that I wish there was more time before the 12.0 release so that I could have created dwatch profiles to simplify tapping into these new hooks. I don't have enough time to develop that before 12.0 is completely frozen.

In D16839#358690, @markj wrote:

I'll admit to being completely unfamiliar with SCTP, but I can't find any problems with this change. Thanks for adding the tests.

In D16839#358726, @dteske wrote:

Please run igor -Dgpxy on the manual changes. OK to ignore any nits it complains about that you didn't introduce yourself.

Address comments from bz@

With this patch the kernel can load the tcp_rack module, which is not special, I think.

I ran the packetdrill tests from rcv-data-segments and they all pass.

In D16626#353912, @kbowling wrote:

@tuexen latest version of this review as of this comment with your updated packetdrill https://reviews.freebsd.org/P204

In D16626#353912, @kbowling wrote:

@tuexen latest version of this review as of this comment with your updated packetdrill https://reviews.freebsd.org/P204

OK. Then I would like to keep the files separate. This way we don't bother people interested in UDP and not knowing about UDP-Lite with UDP-Lite. I guess these cover most users...

UDP-Lite is a protocol different from UDP. So I just wanted to keep things reflecting that.. However, I can add the contents of udplite.d to udp.d.
I can also add the UDP-Lite stuff to the dtrace-udp man-page. We just should make sure that it shows up, when asking for man dtrace-udplite.

Address final comment from bz@

@bz: I tried to address the issues brought up by you. Regarding the IPv4 related commit: There was no IPv6 support (which is added by this patch) and there was no real checking for IPv4 (which is added by this patch). There was no real IPv4 related commit bringing in a corresponding feature. The checks where added to the connect code path, but where missed in the sendto code path...

Try to address issues reported by bz@.

Update the comments as suggested by jtl@.

In D16377#347692, @dteske wrote:

In D16377#347691, @tuexen wrote:

In D16377#347679, @dteske wrote:

Can you expand your patch to include the creation of a corresponding dwatch module? It's easy ...

cd cddl/usr.sbin/dwatch/libexec
sed -e 's/udp/&lite/g;s/length/coverage/g' udp > udplite
echo "$( awk '(line=$1)||1;NF==2&&sub(/udp/,"&lite");/^LINKS/&&gsub(/udp/,"&lite"){_[n++]=$0}/udplite-send/{for(x=0;x<n;)print _[x++]}' Makefile )" > Makefile

Not that easy:

tuexen@head:~/head/cddl/usr.sbin/dwatch/libexec % echo "$( awk '(line=$1)||1;NF==2&&sub(/udp/,"&lite");/^LINKS/&&gsub(/udp/,"&lite"){_[n++]=$0}/udplite-send/{for(x=0;x<n;)print _[x++]}' Makefile )" > Makefile
Illegal variable name.

Bah, foiled by [t]csh. Use this:

sh -c 'echo "$( awk '\''(line=$1)||1;NF==2&&sub(/udp/,"&lite");/^LINKS/&&gsub(/udp/,"&lite"){_[n++]=$0}/udplite-send/{for(x=0;x<n;)print _[x++]}'\'' Makefile )" > Makefile'

Added dwatch script as provided by dteske@.

In D16377#347679, @dteske wrote:

Can you expand your patch to include the creation of a corresponding dwatch module? It's easy ...

cd cddl/usr.sbin/dwatch/libexec
sed -e 's/udp/&lite/g;s/length/coverage/g' udp > udplite
echo "$( awk '(line=$1)||1;NF==2&&sub(/udp/,"&lite");/^LINKS/&&gsub(/udp/,"&lite"){_[n++]=$0}/udplite-send/{for(x=0;x<n;)print _[x++]}' Makefile )" > Makefile

Not that easy:

tuexen@head:~/head/cddl/usr.sbin/dwatch/libexec % echo "$( awk '(line=$1)||1;NF==2&&sub(/udp/,"&lite");/^LINKS/&&gsub(/udp/,"&lite"){_[n++]=$0}/udplite-send/{for(x=0;x<n;)print _[x++]}' Makefile )" > Makefile
Illegal variable name.

@markj Please let me know your preferred commit split for the files in this review.