On Thu, Jan 05, 2012 at 08:44:01PM +0100, Michael Biebl wrote:
> @Alex: To unblock the testing migration of libnl3 3.2, I've uploaded a
> fix for ntrack to DELAYED/5 which uses -Wno-error.
> A full debdiff is attached.
> Please let me know, if I should cancel the NMU or delay it further in
> case you want to wait for a fix in libnl3.

That's fine. Thanks!

The ubuntu package has dropped -pedantic only ... which I thought I had done 
upstream, but didn't
(have filed a bug and will fix it). Keeping no-error should be fine.


====

Sidenote on libnl 3.2 trasition: ntrack make check recycle tests revealed a 
crash (not enabled
in package) for libnl > 3.0 due to a dangling ref to the global (auto provided) 
ops cache ...

Ubuntu maintainer has send the libnl patch I ship as part of ntrack source 
upstream:
http://lists.infradead.org/pipermail/libnl/2012-January/000435.html

Not urgent I think as it's unlikely that apps really fully recycle ntrack 
monitors
or all libnl caches, but if you are interested I would be happy to have more 
eyes on
that issue.

The crash happens if you run make check in ntrack and from my investigation it 
was
introduced after libnl 3.0 from this kind of change:

@@ -288,6 +295,17 @@ static int addr_msg_parser(struct nl_cache_ops *ops, 
struct sockaddr_nl *who,
                addr->ce_mask |= ADDR_ATTR_ANYCAST;
        }
 
+       if ((link_cache = __nl_cache_mngt_require("route/link"))) {
+               struct rtnl_link *link;
+
+               if ((link = rtnl_link_get(link_cache, addr->a_ifindex))) {
+                       rtnl_addr_set_link(addr, link);
+
+                       /* rtnl_addr_set_link incs refcnt */
+                       rtnl_link_put(link);
+               }
+       }
+


This means that without my patch, you can free the link cache and then 
addr_msg_parser will
still try to reference the link_cache here as the _free doesn't call unprovide 
and so doesn't
NULL the global cache stored in the ops struct.

> 
> I've also updated the symbols file for libnl-qt4-1 while at it.

The build didn't fail due to this though, right?

 - Alexander



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

Reply via email to