Inconsitencies wrt. packetdrill vs. run_all.py script

[Florian Westphal]

Some test cases fail when not run via run_all.py, one example is
tcp/mtu_probe/basic-v4.pkt.

The difference is the remote ip passed to/used by packetdrill.

run_all.py passes: '--remote_ip=192.0.2.1 ', whereas packetdrill
default is '192.0.2.1/24'.

Problem is that basic-v4.pkt does:
'ip route change 192.0.2.1 via 192.168.0.2 dev tun0 advmss 600 mtu lock 640 initcwnd 11'.

This results in an error from kernel when running script via
"../../packetdrill basic-v4.pkt" (s/change/add/ makes it work again).

I'd like to patch run_all.py to use /24 and /32 (ipv6) netmask, just
like the packetdrill default.

I'd fix up the then-failing test scripts as well. Any objections?

Motivation is to make certain MPTCP MP_JOIN test cases (in the MPTCP
packetdrill fork) work with run_all.py. For MP_JOIN, a secondary
address gets added for use by the subflow. Without the netmask the kernel
won't find a route for the syn/ack join response, so those scripts work
when passed to 'packetdrill' directly with default options, but fail
when run via the python script.

Thanks,
Florian

[Neal Cardwell]

Hi Florian,

Thanks for sending this. If I understand the proposal correctly, and it just involves changing the run_all --remote_ip argument and adjusting scripts to pass, then that sounds fine to me. Please feel free to send a pull request.

Thanks!
neal

--
You received this message because you are subscribed to the Google Groups "packetdrill" group.
To unsubscribe from this group and stop receiving emails from it, send an email to packetdrill...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/packetdrill/20201201185652.GA5710%40breakpoint.cc.

[Florian Westphal]

Neal Cardwell <ncar...@google.com> wrote:
> Thanks for sending this. If I understand the proposal correctly, and it
> just involves changing the run_all --remote_ip argument and adjusting
> scripts to pass, then that sounds fine to me. Please feel free to send a
> pull request.

Done, thanks Neal!