On Sun, Jul 09, 2006 at 05:57:43PM +0200, Benoît Dejean wrote:
> Package: libc6
> Version: 2.3.6-15
> Severity: minor
>
> Hi,
> using getifaddrs i get the following valgrind warning :
>
> ==12949== Syscall param socketcall.sendto(msg) points to uninitialised byte(s)
> ==12949== at 0xE633ACC: sendto (socket.S:106)
> ==12949== by 0xE651F48: __netlink_sendreq (ifaddrs.c:112)
> ==12949== by 0xE6521E4: __netlink_request (ifaddrs.c:131)
> ==12949== by 0xE652AD8: getifaddrs (ifaddrs.c:390)
>
> Indeed, ./glibc-2.3.6/sysdeps/unix/sysv/linux/ifaddrs.c:
>
> 92 struct
> 93 {
> 94 struct nlmsghdr nlh;
> 95 struct rtgenmsg g;
> 96 } req;
> 97 struct sockaddr_nl nladdr;
> ....
> 101
> 102 req.nlh.nlmsg_len = sizeof (req);
> 103 req.nlh.nlmsg_type = type;
> 104 req.nlh.nlmsg_flags = NLM_F_ROOT | NLM_F_MATCH | NLM_F_REQUEST;
> 105 req.nlh.nlmsg_pid = 0;
> 106 req.nlh.nlmsg_seq = h->seq;
> 107 req.g.rtgen_family = AF_UNSPEC;
> ....
> 112 return TEMP_FAILURE_RETRY (__sendto (h->fd, (void *) &req, sizeof
> (req), 0,
> 113 (struct sockaddr *) &nladdr,
> 114 sizeof (nladdr)));
>
> some members are not initialized.that's an interesting assertion, but you should look in the kernel headers to verify (like I did) that rtgenmsg and nlmsghdr members are all initialized in the very code you just cited. Though, in the 2.5 libc the possible paddings have been set to 0, hence that (harmless) valgrind warning should anyway now be gone. -- ·O· Pierre Habouzit ··O [EMAIL PROTECTED] OOO http://www.madism.org
pgpEI5cV8rCIn.pgp
Description: PGP signature
