Quoting Eric Spakman <[EMAIL PROTECTED]>:

> Hello,
>
> dnsmasq-2.31 doesn't build without a lot of tweaking, but before
> going that road can you tell me on what occasion 2.27 is segfaulting?
> If you have problems with 2.27 I don't think 2.31 will solve that. We
> applied a fix to dnsmasq that solves a segfault problem which is also
> applied upstream.
>
> Eric

Simon Kelley was able to determine the problem from a core dump I sent him. Here
is his reply:

Simon Kelley wrote:
> Looking at the core dump, it's a different problem than the one fixed
> in the 2.27 patch, but it's already fixed in 2.29 and later versions.
> For reference, it's a memory overwrite which happens if the DNS
> system gets presented with a query with an empty (zero-length) name.
>
> The malloc implementation in uclibc seems to fall over reliably with a
> 1-byte overrun at either end of allocated memory, whilst the glibc one
> seems to survive. It looks like glibc has some unused or nearly unsed
> stuff there, whilst uclibc packs allocated blocks right next to each
> other, so that real data gets overwritten. This is why both of these
> bugs hit uclibc but nobody noticed then under glibc.
>
> I don't seem to have Eric's address, so Chris please could you  pass
> this on to him?
>
> I'm happy to work on making dnsmasq releases easier to put into
>   OpenWRT, new releases get tested by being linked against uclibc on
> Debian, but
> I don't currently have a WRT box to do testing on. Maybe we can work
> something out? At very least I'm happy to get patches back.
>
> I'd also like to encourage openWRT to move the leases file onto flash
> storage, and use the HAVE_BROKEN_RTC compile flag. That would fix most
> of the "my local machine lose their names when I reboot openWRT"
> problems which I see reported.
>
> Cheers,
>
> Simon.


------------------------------------------------------------------------
leaf-user mailing list: leaf-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/leaf-user
Support Request -- http://leaf-project.org/

Reply via email to