[PATCH v2 3/7] tests: check decoding of netlink protocol

Fabien Siron fabien.siron at epita.fr
Wed Jun 15 09:23:11 UTC 2016


Quoting Dmitry V. Levin (2016-06-14 12:03:47)
> On Mon, Jun 13, 2016 at 02:37:23PM +0000, Fabien Siron wrote:
> [...]
> > +static void
> > +send_query(const int fd)
> > +{
> > +     struct {
> > +             struct nlmsghdr nlh;
> > +             char magic[4];
> > +     } req = {
> > +             .nlh = {
> > +                     .nlmsg_len = sizeof(req),
> > +                     .nlmsg_type = SOCK_DIAG_BY_FAMILY,
> > +                     .nlmsg_flags = NLM_F_DUMP | NLM_F_REQUEST
> > +             },
> > +             .magic = "abcd"
> > +     };
> > +
> > +     if (sendto(fd, &req, sizeof(req), MSG_DONTWAIT, NULL, 0) <= 0)
> > +             perror_msg_and_skip("sendto");
> 
> Why <= 0?  The next statement assumes == (unsigned) sizeof(req).

Indeed.

> 
> > +
> > +     printf("sendto(%d, {{len=%u, type=20, flags=NLM_F_REQUEST|0x300, "
> > +            "seq=0, pid=0}, \"abcd\"}, 20, MSG_DONTWAIT, NULL, 0) = %u\n",
> > +            fd, (unsigned) sizeof(req), (unsigned) sizeof(req));
> 
> Please do not encode values of constants into output strings verbatim.

Well, the size of the magic must be aligned on 4 characters to avoid '\0' stuff.
So it means that print req.magic will fail because of no null byte. An easy
way to do that is to use a define.
But maybe do you prefer a variable with a memcpy?

e.g.:
static void
send_query(const int fd)
{
	const char magic[] = "abcd";
	struct {
	     struct nlmsghdr nlh;
	     char magic[4];
	} req = {
	     .nlh = {
             	  .nlmsg_len = sizeof(req),
                  .nlmsg_type = SOCK_DIAG_BY_FAMILY,
                  .nlmsg_flags = NLM_F_DUMP | NLM_F_REQUEST
             },
	};
	memcpy(req.magic, magic, 4);
	...
> 
> -- 
> ldv
> 

Regards,
--
Fabien Siron




More information about the Strace-devel mailing list