Bill Pringlemeir wrote:
> I was looking at the ADNS.c issue that Jeroen posted.  I believe it is
> a race condition.  If the memory that "gethostbyname()" is freed and
> reused would Valingrind report the error, 
 
>  "==14962== Conditional jump or move depends on uninitialised value(s)".

My interpretation of this warning is that it refers to the bytes after
the trailing NUL of the hostname. I don't know what libnss_dns does. The
memcpy() however has been replaced by a strncpy() in the stable branch
as well as in the unstable branch. If this didn't fix it, just say
Bescheid.

> I checked direct logic like large/small names, etc.  I think there
> might be some sort of race condition with the atoms and/or the query
> structs.

There shouldn't be many possible ways to cause race condition WRT to
accessing memory in GTKG because it's mono-threaded. There's surely no
race condition in host_to_ip() and adns_gethostbyname() looks safe to
me as well.

Christian

Attachment: pgpurpb9ih4jn.pgp
Description: PGP signature

Reply via email to