Multipath TCP (RFC 6824) support

[gdetal]

Multipath TCP (MPTCP) is a new extension to TCP that is being standardised at the IETF. We are currently maintaining the reference Linux implementation @multipath-tcp.

The patch allows tcpdump to parse MPTCP options (that are in fact TCP options) in order to print useful information. A sample output can be found at http://pastebin.com/BsJP1w4a.

Signed-off-by: Gregory Detal gregory.detal@uclouvain.be

[infrastation]

It is considered good to complement new decoders with a small .pcap file and relevant text output, that together constitute a basic regression test of the make check command (see a few recent commits for an example).

[fenner]

I know that not all files are consistent, but tcpdump generally uses an 8 space initial indent and 4 or 8 space subsequent indent. 2 space indentation, therefore, ends up looking weird. Can you use something closer to what other files use?

[mcr]

Also, github says that your patch may need to be rebased first. Very happy to accept MPTCP... the next release will go out during the IETF86 meeting.

[gdetal]

Thanks for the feedback, I have adapted my commit accordingly.

@infrastation I've added two regression test @fenner I was using tab and so I switched to 8 spaces indent @mcr I have rebased my branch

[gdetal]

Hello,

Am I missing something ? I've updated the commit and have done a push -f on my branch. Yet the merge is not yet accepted. Is is the right way to do this ?

[infrastation]

Don't worry, sometimes it takes a few days.

[gdetal]

@infrastation Your comment disappeared but thanks for the feedback. I updated my commit to use the PRI output format.

[infrastation]

Functions mp_fail_print() and mp_fast_close_print() should return a value (warning seen when compiling with make CCOPT=-Wall).

[infrastation]

There is some more space for improvement in this code, please see below.

• For consistency of output the option name added to tcp_option_values[] should be lowercase.
• tcp_print() seems to require LENCHECK(datalen) in the newly added case block.
• mptcp_print() should check that the option length is at least 3 before dispatching further.
• mp_capable_print() should report a malformed option for length other than 12 or 20. The function does not print or check the value of the Version field, which is defined in RFC6824 to be 0 (if it is not 0, would the option format be different?). The function converts neither mpc->sender_key nor mpc->receiver_key from network byte order before printing (alternatively, should the keying material be printed as 01:23:45:67:89:ab:cd:ef instead?).
• mp_join_print() should report a malformed option for length other than 12, 16 or 24. The function does not convert the value of mpj->u.syn.token from network byte order before printing and does not print the value of mpj->u.syn.nonce at all. The function handles only the 12-octet (for Initial SYN) format of MP_JOIN option, there are also 16-octet (for Responding SYN/ACK) and 24-octet (for Third ACK) formats. The current code does not interpret whether the format (12/16/24) matches the TCP flags of the current packet (if anyone wants to implement this, the flags would have to be expedited from tcp_print()).
• mp_dss_len() has its second argument equal to 1, should it be omitted (the presence of Checksum field is already covered with the M-bit of the DSS option)?
• mp_dss_print() should report a malformed option for length other than 4, 8, 12, 16, 20, 24 or 28. The function should check if the length matches the flags.
• add_addr_print() should report a malformed option for length other than 8, 10, 20 or 22. The function should check that the length matches the value of IPVer field. The function should not report a malformed option when tcpdump is built without IPv6 support, but rather should print a placeholder string instead of the IPv6 address.
• mp_prio_print() should report a malformed option for length other than 3 or 4.
• mp_fail_print() should report a malformed option for length other than 12.
• mp_fast_close_print() should report a malformed option for length other than 12. It also has the same issue printing the value of receiver's key as mp_capable_print() above.

[guyharris]

Cherry-picked to the 4.4 branch.

[guyharris]

Functions mp_fail_print() and mp_fast_close_print() should return a value (warning seen when compiling with make CCOPT=-Wall).

I've checked in a fix for that and cherry-picked it to the 4.4 branch.