Salvatore.

The problem occurs if the dnsmasq binary is compiled against libc
headers which #define SO_REUSEPORT and then run on a kernel which
doesn't support that option. I guess the security builds have picked up
SO_REUSEPORT from a libc backport.

The fix applied at the time was:

http://thekelleys.org.uk/gitweb/?p=dnsmasq.git;a=commit;h=56a1142f033234e3ee3b6361e9a1bcdbe606f816


Cheers,

Simon.



On 06/05/15 15:16, Salvatore Bonaccorso wrote:
> Hi Ian and Luca,
> 
> On Wed, May 06, 2015 at 12:59:03PM +0200, Luca Olivetti wrote:
>> On Wed, 06 May 2015 11:30:35 +0100 Ian Campbell <i...@debian.org> wrote:
>>
>>
>>> I've just noticed that running kernel on the machine is 3.2.57-3+deb7u1
>>> which is quite out of date wrt point releases etc. Looking at the
>>> changelog there have been dozens of stable update fixes, one of which
>>> might be relevant here.
>>>
>>> I'll reboot when I get home and see if perhaps that fixes the issue.
>>
>> I have the same problem, but my machine uses 3.2.68:
>>
>>
>> #uname 
>> -ahttp://thekelleys.org.uk/gitweb/?p=dnsmasq.git;a=commit;h=56a1142f033234e3ee3b6361e9a1bcdbe606f816
>> Linux lacie 3.2.0-4-kirkwood #1 Debian 3.2.68-1+deb7u1 armv5tel GNU/Linux
>>
>>
>> I had to comment the "bind-interfaces" line from /etc/dnsmasq.conf in
>> order to make it start.
> 
> Thanks to both for feedback. Might it be a problem with the armel
> build?  If you look at the build log (which I have uploaded to [1]) on
> the armel chroot to build the package on the buildd there installed
> linux-libc-dev which is not from stable but from backports, and indeed
> for linux >= 3.9. Thus the armel build will have (dhcp.c):
> 
> [...]
>       int rc = setsockopt(fd, 1, 15, &oneopt, sizeof(oneopt));
> 
>       if (rc == -1)
>  die(dcgettext (((void *)0), "failed to set SO_REUSE{ADDR|PORT} on DHCP 
> socket: %s", __LC_MESSAGES), ((void *)0), 2);
>     }
> [...]
> 
> Could either of you try to rebuild dnsmasq in a clean chroot and see
> if the problem resolves? That the buildd have 3.14.13-2~bpo70+1
> installed is odd and should not be.
> 
>  [1] 
> https://people.debian.org/~carnil/tmp/dnsmasq/dnsmasq_2.62-3+deb7u2_armel-20150505-1143.gz
>  [2] https://lwn.net/Articles/542629/
> 
> Regards,
> Salvatore
> 


-- 
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org

Reply via email to