[PATCH] Core: fixed uninitialized memory access

Jan Seda nginx-devel at hodor.cz
Thu Sep 29 22:09:12 UTC 2016


On 2016-09-29, 09:02:16, Maxim Dounin wrote:
> Thanks for catching this.
> Dropping the length returned by getsockname() doesn't look like 
> a correct solution though.  Instead, teaching ngx_cmp_sockaddr() to 
> compare sockaddrs with not-completely-filled sun_path - that is, 
> respecting socklen - should be the right way to go.
> Please try the following patch:

Seems to work OK. Cannot reproduce the problem anymore. Thanks.

BTW, wouldn't s/ngx_palloc/ngx_pcalloc/ in ngx_set_inherited_sockets()
be prudent anyway?

Also, I see your patch is prepared for abstract namespace sockets. Is
this feature planned soon? I cobbled up a patch for that (attached) and
such sockets now interop with haproxy (abns@ scheme) and socat (with
unix-tightsocklen=0). But it probably is not production-ready.

Jan Seda
-------------- next part --------------
A non-text attachment was scrubbed...
Name: nginx_abns.diff
Type: text/x-diff
Size: 3691 bytes
Desc: not available
URL: <http://mailman.nginx.org/pipermail/nginx-devel/attachments/20160930/cc7faa25/attachment.bin>

More information about the nginx-devel mailing list