Change to simulator arguments for readability

[drzaeus77]

• Let one argument be optional
• Disable ipv6 for whole namespace once
• Python 2/3 compatibility fix

Signed-off-by: Brenden Blanco [email protected]

[yonghong-song]

To disable ipv6, do we need to disable both veth end ifc, or just the ifc inside netns?

[drzaeus77]

Don't all of the ifcs end up inside a netns?

[yonghong-song]

out_ifc is not in the namespace.

On Mon, Jul 13, 2015 at 9:57 PM, Brenden Blanco [email protected]
wrote:

In examples/simulation.py
#85 (comment):

     if name in self.ipdbs:
         ns_ipdb = self.ipdbs[name]
     else:
         ns_ipdb = IPDB(nl=NetNS(name))

•        if disable_ipv6:
  

•            cmd = ["sysctl", "-q", "-w", "net.ipv6.conf.default.disable_ipv6=1"]
  

•            nsp = NSPopen(ns_ipdb.nl.netns, cmd)
  

•            nsp.wait(); nsp.release()
  

•        ns_ipdb.interfaces.lo.up().commit()
  

Don't all of the ifcs end up inside a netns?

—
Reply to this email directly or view it on GitHub
https://github.com/iovisor/bcc/pull/85/files#r34537240.

[yonghong-song]

I just tried the patch (git checkout bblanco_dev) and tried on my local box:

9: F
9: ======================================================================
9: FAIL: test_brb2 (main.TestBPFSocket)
9: ----------------------------------------------------------------------
9: Traceback (most recent call last):
9: File "/home/plumgrid/iovisor/bcc/tests/cc/test_brb2.py", line 165, in
test_brb2
9: self.assertEqual(self.pem_stats[c_uint(0)].value, 12)
9: AssertionError: 23L != 12

So it does seem we also need to disable ipv6 for outer ifc as well.

On Mon, Jul 13, 2015 at 9:57 PM, Brenden Blanco [email protected]
wrote:

In examples/simulation.py
#85 (comment):

     if name in self.ipdbs:
         ns_ipdb = self.ipdbs[name]
     else:
         ns_ipdb = IPDB(nl=NetNS(name))

•        if disable_ipv6:
  

•            cmd = ["sysctl", "-q", "-w", "net.ipv6.conf.default.disable_ipv6=1"]
  

•            nsp = NSPopen(ns_ipdb.nl.netns, cmd)
  

•            nsp.wait(); nsp.release()
  

•        ns_ipdb.interfaces.lo.up().commit()
  

Don't all of the ifcs end up inside a netns?

—
Reply to this email directly or view it on GitHub
https://github.com/iovisor/bcc/pull/85/files#r34537240.

[drzaeus77]

I see. You are using a bridge to link the two veth pairs together. It could be done by inserting one end of a veth pair into the other namespace directly, such that neither of them show up in the abstract namespace. In fact, the support for this is exactly what this function is supposed to achieve, but in a different set of code example (in a private codebase, so you don't see it here). I will amend the patch to use that approach, and eliminate the extra bridges.

[yonghong-song]

This will be fine too. Please do add some comments in simulation.py though
to warn that if users do have one end of veth not in the namespace, they
will be responsible to disable ipv6 if it is desirable to them.

On Mon, Jul 13, 2015 at 10:14 PM, Brenden Blanco [email protected]
wrote:

In examples/simulation.py
#85 (comment):

     if name in self.ipdbs:
         ns_ipdb = self.ipdbs[name]
     else:
         ns_ipdb = IPDB(nl=NetNS(name))

•        if disable_ipv6:
  

•            cmd = ["sysctl", "-q", "-w", "net.ipv6.conf.default.disable_ipv6=1"]
  

•            nsp = NSPopen(ns_ipdb.nl.netns, cmd)
  

•            nsp.wait(); nsp.release()
  

•        ns_ipdb.interfaces.lo.up().commit()
  

I see. You are using a bridge to link the two veth pairs together. It
could be done by inserting one end of a veth pair into the other namespace
directly, such that neither of them show up in the abstract namespace. In
fact, the support for this is exactly what this function is supposed to
achieve, but in a different set of code example (in a private codebase, so
you don't see it here). I will amend the patch to use that approach, and
eliminate the extra bridges.

—
Reply to this email directly or view it on GitHub
https://github.com/iovisor/bcc/pull/85/files#r34537754.

[drzaeus77]

I went back to the original, since I realized that you actually wanted the outer bridges to exist as part of the test scenario.