Re: CVS commit: src/sys

[Robert Elz <kre%munnari.OZ.AU@localhost>]

    Date:        Fri, 11 May 2018 12:49:11 +0100
    From:        Roy Marples <roy%marples.name@localhost>
    Message-ID:  <99536ae5-2bcf-c8d5-7c39-f0436c6d18b2%marples.name@localhost>

[Taking your reply out of order...]

  | The cvs commit increases the default size of the ICMP6 receive socket, 
  | so it's not incidental at all.

If that were true, the problem would not be fixed, as even with the bigger buffer,
the problem could still happen.

  | The address could also have been lost due to dhcpcd not seeing the RA 
  | the IPv6 router sent and thus timing out.

Yes.

  | This could be due to the ICMP6 receive socket overflowing and thus the 
  | kernel discards the RA.

Yes - and I have always agreed that for dhcpd learning that routing
messages have been lost, so it can react is a good idea.   It could be
for other routing daemons too, though most of those have other
recovery mechanisms.

That the problem was fixed by the dhcpcd changes was indicated in
the PR by this fragment ...

   The patch has the desired effect.  I see a couple of the
   "ipv6nd_handledata: No buffer space available" messages in the log
   from this morning, which at this point is about 5 hours ago.  Since
   then, I've also seen the usual dance where the default router expires
   and it settles on the right one again.  I still have an IPv6 address
   and my IPv6 connections are still good.

That is, even with the lost messages still being reported, everything
was good.  At that point the problem was fixed.

Also note I don't object to making the receive buffers bigger, so it is
less likely that messages will be lost - just to calling that a "fix" for
anything at all.   It isn't.

kre