[tor-bugs] #13338 [Tor]: Rewrite tor-fw-helper in Go (or another memory-safe language)

[Tor Bug Tracker & Wiki]

#13338: Rewrite tor-fw-helper in Go (or another memory-safe language)
-----------------------------+------------------------------
     Reporter:  arma         |      Owner:  yawning
         Type:  enhancement  |     Status:  assigned
     Priority:  minor        |  Milestone:  Tor: unspecified
    Component:  Tor          |    Version:
   Resolution:               |   Keywords:  flashproxy
Actual Points:               |  Parent ID:  #5213
       Points:               |
-----------------------------+------------------------------
Changes (by yawning):

 * milestone:  Tor: very long term => Tor: unspecified


Comment:

 Ok, I have a router that supports NAT-PMP, so I'm writing the client code
 now.  I also found out some interesting things about miniupnpd, and fixed
 "compatibility with broken non-RFC compliant HTTP servers in existing
 routers" in the uPNP code[0].

 The NAT-PMP RFC makes threatening noises about uPNP implementations that
 crash if the lease duration isn't set to 0, so I'll probably go change the
 code to always issue "indefinite" leases when talking uPNP 1.0, even
 though using it like how flashproxy wants to (randomized port) will clog
 up the uPNP lease table and cause catastrophic failure modes in other uPNP
 implementations[1].  I figure eventual failure is better than immediate,
 but there is no good answer here apart from (only use one port, pray)
 given the current architecture.

 I'm starting to think that the whole "call a helper once in a while to
 extend the lease" isn't that great of a design in view of how broken all
 the various uPNP routers out there are (especially given that the original
 tor-fw-helper has no method of cleaning up existing leases), and that this
 would be much better off as a daemon, but going to such an architecture
 makes the code considerably more complex (as it needs to be resilient to
 router reboots), but fixing that is a longer term thing that can wait till
 after feature parity has been reached.

 [0]: The embedded webserver shipped in all versions including the master
 in their repository does not handle chunked transfer coding despite
 claiming to be HTTP 1.1.
 (https://github.com/miniupnp/miniupnp/blob/master/miniupnpd/upnphttp.c)
 [1]: http://www.upnp-hacks.org/annoyances.html

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/13338#comment:8>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online