Select the network interface in send()

[guedou]

This PR fixes #2355.

[codecov]

Codecov Report

Merging #2356 into master will increase coverage by 0.28%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #2356      +/-   ##
==========================================
+ Coverage    87.5%   87.79%   +0.28%     
==========================================
  Files         243      243              
  Lines       50727    50728       +1     
==========================================
+ Hits        44390    44537     +147     
+ Misses       6337     6191     -146

Impacted Files	Coverage Δ
scapy/sendrecv.py	85.48% <100%> (+0.02%)	⬆️
scapy/pton_ntop.py	89.04% <0%> (-8.22%)	⬇️
scapy/autorun.py	78.51% <0%> (-4.14%)	⬇️
scapy/layers/tls/record_sslv2.py	86.03% <0%> (-1.68%)	⬇️
scapy/layers/tls/record.py	89.59% <0%> (-1.16%)	⬇️
scapy/layers/tls/basefields.py	80.53% <0%> (-0.68%)	⬇️
scapy/automaton.py	87.22% <0%> (-0.4%)	⬇️
scapy/arch/windows/__init__.py	71.81% <0%> (-0.16%)	⬇️
scapy/layers/inet6.py	81.96% <0%> (+0.21%)	⬆️
scapy/layers/tls/handshake.py	86.37% <0%> (+0.28%)	⬆️
... and 6 more

[guedou]

Please don't merge yet. I have some doubts about this PR when multiple packets are sent.

[p-l-]

I do not agree with this PR.

In my mind, send() is supposed to use Scapy's routing table (conf.iface) to send layer 3 packets, while sendp() sends layer 2 packets using conf.iface (or the value for parameter iface=).

Specifying an interface in send() is weird. For instance, what layer 2 address should Scapy use to send a packet outside the LAN?

An option would be to allow specifying an alternative routing table object (or a route).

What do you think?

[guedou]

The current implementation of sr1() behaves the same way as this PR: try to be smart and use the routing table instead of conf.iface.

[p-l-]

OK I misunderstood your PR then, I'll have to read again. Anyway, having send() behavior consistent with sr() and sr1() is what we want.

[p-l-]

But I really don't think that send() (or sr() and sr1()) should have an iface= argument.

[guedou]

The root issue might be related to this line https://github.com/secdev/scapy/blob/master/scapy/arch/linux.py#L478

[guedou]

Can this PR be merged?

[p-l-]

As I said, I don't like the idea of send() having an iface= parameter. send() should rely on a routing table, not an interface (which, in turn, will provide an interface and a gateway). What is the meaning of providing an interface alone? We still need the routes to get the gateway IP address.

[guedou]

This PR aligns sr1() and send() behavior. I think that it is ok to merge it. Bypassing the routing table with these functions is indeed useful for protocols such as IPv6 with link-local addresses when the same prefix exists on different interfaces.

[guedou]

This is indeed related to #2108 and #1975.

As explained several times in this thread, I really think that this PR should be merged.

Please merge or provide a solution to these issues.