Bugzilla Bug 87091

dnsmasq-2.21 causes the following issues on my servers

1) Any client matched in dhcp-host for DHCP requests cause dnsmasq to stop responding to DNS and DHCP requests

2) when mx-host, mx-target or localmx options are set, dnsmasq will go into a infinite loop and not fork to the background when it is started

Reverting back to stable 2.15 or unstable 2.20 fixes the above issues.

found the --mx-target= bug, waiting for Uberlord to test my fix.

option.c has:

      if ((daemon->mxtarget || (daemon->options & OPT_LOCALMX)) && !mx)
        {
          daemon->mxnames = safe_malloc(sizeof(struct mx_srv_record));
          daemon->mxnames->next = daemon->mxnames;
          daemon->mxnames->issrv = 0;
          daemon->mxnames->target = NULL;
          daemon->mxnames->name = safe_string_alloc(buff);
        }

and later:

      for (mx = daemon->mxnames; mx; mx = mx->next)
        if (!mx->issrv && !mx->target)
          mx->target = daemon->mxtarget;

This loops forever because mx->next = mx = mx->next as there is no sanity check
for loops (in this case, next=self).

Not quite sure what was meant to happen here, but it seems maybe it should be:

daemon->mxnames->next =  NULL;

instead of

daemon->mxnames->next = daemon->mxnames;

We don't have anyone here who can test this functionality easily, so mailing upstream.

I can't easily pinpoint the dhcp-host bug because I can't test locally and gdb
wasn't playing nice on Uberlord's machine.

I'll let Simon know about this bug.

author mailed.

dnsmasq should probably be masked until resolution

I mean 2.21 should be masked (or keywords reset) until resolution as 2.21 is
the current stable.

uberlord: please test http://www.thekelleys.org.uk/test

It should definitely fix the mx-host thing and hopefully the dhcp-host stuff as well.

That test version does NOT fix the dhcp-host problem :(
If it helps, here is a line that causes the problem

dhcp-host=00:09:5b:84:88:7d,uberpc,192.168.2.10

Removing that line makes uberpc get a DHCP address correctly and not cause dnsmasq to eat CPU cycles like it's going out of fashion.

Bug is more harmful than potential security problems in <=2.20. I've masked
2.21 for now.

The 2.22 test version in comment #7 fixed the mx bug

Just the dhcp-host bug to fix now!

the dhc-host bug manifests itself at dhcp.c:422

  for (; context; context = context->current)

loops forever. This is another context == context->current thing, same as last one.

Not sure of the fix but I've let the author know.

test fix version from simon at:

http://thekelleys.org.uk/test/dnsmasq-2.22test7.tar.gz

fix confirmed to work :)

I've let the author know.

22 released with the fix.

2.22 committed. Thanks Rob!