Replace netutils.py with from-scratch implementation under Apache 2.0 license

[jafingerhut]

No description provided.

[jafingerhut]

Intended to address issue #210

I will update with another comment after I have used this version to run ptf tests in all of these places, with my test results:

• https://github.com/jafingerhut/p4-guide
• https://github.com/p4lang/tutorials
• https://github.com/p4lang/p4c - full make check test suite

I plan to test on at least Ubuntu 20.04 and Ubuntu 24.04, combined with {x86_64, aarch64} processor architectures.

[jafingerhut]

Updates:

PTF tests are passing with these changes on these combos:

• aarch64 Ubuntu 20.04
• x86_64 Ubuntu 24.04

It does require doing pip3 install netifaces in addition to installing ptf. If someone knows how to change any Python setup/install files in this repo so that installing ptf will automatically cause the netifaces package to be installed as a dependency, I would appreciate hints there.

Except for that, I think this is ready to review, and merge if approved.

[fruffy]

Not sure we should need an unmaintained dependency for a simple task like this. I rather we write these functions from scratch.

[jafingerhut]

As far as I can tell, no p4lang repository code is actually calling the get_mac function anywhere. If you prefer, I can just delete it, and then there is no reason to import netifaces package.

[jafingerhut]

Or I could copy this Apache 2.0 implementation in this repo for get_mac into the file netutils.py, just in case someone does call get_mac in other repos that are not in the p4lang Github org: https://github.com/p4lang/ptf/blob/main/ptf_nn/ptf_nn_agent.py#L51-L65

Let me try that.

[fruffy]

Yeah that looks good. For testing, ideally you want to run this on the P4C tests.
This is what Gemini gives me. I think LLMs are a good use case for straightforward cases like this one.

import socket
import struct
import fcntl

# Given 'iff', the name of a network interface (e.g. 'veth0') as a
# string, return a string of the form '00:00:00:00:00:00' containing
# the MAC address of that interface, where all digits are hexadecimal.
def get_mac(iff: str) -> str:
    try:
        s = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
        info = fcntl.ioctl(s.fileno(), 0x8927,  struct.pack('256s', iff[:15].encode()))
        mac = ':'.join(['{:02x}'.format(x) for x in info[18:24]])
        return mac
    except Exception as e:
        print(f"Error getting MAC address: {e}")
        return "00:00:00:00:00:00"


# Given 'iff', the name of a network interface (e.g. 'veth0') as a
# string, and 's', a SOCK_RAW socket bound to that interface, set the
# interface in promiscuous mode if parameter 'val' != 0, or into
# non-promiscuous mode if 'val' == 0.
def set_promisc(s, iff: str, val=1):
    try:
        if val:
            # Set promiscuous mode
            ifr = struct.pack('16sH', iff.encode(), socket.htons(socket.IF_PROMISC))
            fcntl.ioctl(s.fileno(), socket.SIOCSIFFLAGS, ifr)
        else:
            # Set non-promiscuous mode
            ifr = struct.pack('16sH', iff.encode(), socket.htons(0))
            fcntl.ioctl(s.fileno(), socket.SIOCSIFFLAGS, ifr)
    except Exception as e:
        print(f"Error setting promiscuous mode: {e}")

[jafingerhut]

Your first try with an LLM gave me failures, so I looked elsewhere.

[jafingerhut]

I have tested this implementation of set_promisc with the full p4c test suite, on my local system.

[jafingerhut]

I have tested this implementation of set_promisc with the full p4c test suite, on my local system.

by "this implementation" I mean not the one produced by the LLM, but the one that is in this PR now.

[fruffy]

The way we ideally should test this is to set the PTF dependency to a Github URL here:
https://github.com/p4lang/p4c/blob/main/requirements.txt#L5

Using this approach: https://stackoverflow.com/a/35998253

Then we can update and test PTF directly on the P4C repository. Let me look into how to set this up.

[jafingerhut]

I certainly have no objections if you want to test this in a different way than I already have. There is no rush on merging this. Just trying to keep the ball rolling on letting us license ptf as Apache-2.0 (or at least not copyleft) with a straight face.

[fruffy]

Approving considering a lot of tests pass. I am not sure how these files are exactly exercised in the PTF framework.