non specified port for IPRSocket results in assertion failure about pyroute2 HOT 15 CLOSED

I thought that missing call to bind() is the reason but I see the same problem with the following code, too:

import os
import re
import socket
import subprocess
from pyroute2 import IPRSocket
from pyroute2.netlink import NLM_F_REQUEST
from pyroute2.netlink import NLM_F_DUMP
from pyroute2.netlink import NLMSG_DONE
from pyroute2.netlink import NLMSG_ERROR
from pyroute2.netlink.iproute import RTM_GETLINK
from pyroute2.netlink.iproute import RTM_NEWLINK
from pyroute2.netlink.rtnl.ifinfmsg import ifinfmsg

def normalize_hwaddr(hwaddr):
    return hwaddr.upper().rstrip("\n")

scan = []
nl_socket = IPRSocket()
msg = ifinfmsg()
msg["family"] = socket.AF_UNSPEC
msg["header"]["type"] = RTM_GETLINK
msg["header"]["flags"] = NLM_F_REQUEST | NLM_F_DUMP
msg["header"]["pid"] = os.getpid()
msg["header"]["sequence_number"] = 1
msg.encode()

nl_socket.sendto(msg.buf.getvalue(), (0,0))

finished = False
while not finished:
    parts = nl_socket.get()
    for part in parts:
        if part["header"]["type"] in [NLMSG_DONE, NLMSG_ERROR]:
            finished = True
            continue
        if part["header"]["sequence_number"] != 1:
            continue

        if part["header"]["type"] == RTM_NEWLINK:
            new_link = {}
            new_link["netlink_msg"] = part
            new_link["index"] = part["index"]
            new_link["name"] = part.get_attr("IFLA_IFNAME")
            print new_link["name"]
            hwaddr = part.get_attr("IFLA_ADDRESS")
            new_link["hwaddr"] = normalize_hwaddr(hwaddr)
            scan.append(new_link)

nl_socket.close()

from pyroute2.

Thanks, got it. Fixing.

from pyroute2.

Peet, it seems that adding bind() before sendto() actually fixed the problem.

from pyroute2.

Yep. Just wondering, how it worked before w/o bind? o_O

from pyroute2.

Another question, should we keep the old behaviour (w/o bind? how it come?), or just move to the new with bind? I would prefer the latter, but what you'll say?

from pyroute2.

I'm ok with the latter, it's one-line fix in the LNST code for me.

from pyroute2.

Ok, thanks.

from pyroute2.

Hi, I've just come across this problem. What is the reason behind the assert? The example that @jtluka posted uses the socket like a client would (to send requests to the kernel) - no need to bind it to anything. But such sockets should still be close-able.

from pyroute2.

The reason was in data integrity, when you have several sockets, and wanna properly release resources. But you're right, it effectively make impossible to close the socket w/o binding it before.

Since w/o binding the socket doesn't allocate any resource that should be tracked, it would be better to make more smart close() routine. Wait a minute.

from pyroute2.

Yep, so looks better and still passes test cycle. Thanks!

from pyroute2.

Thanks for the quick fix!

from pyroute2.

You're always welcome :)

from pyroute2.

btw, when can I expect a new version that will contain this fix?

from pyroute2.

By the end of the week, I suppose. I need to finish some test code.

from pyroute2.

great, thanks again!

from pyroute2.