Re: [RFC net-next iproute2 1/2] libnetlink: Add test for error code returned from netlink reply

[Cyrill Gorcunov <gorcunov@gmail.com>]

On Wed, Oct 26, 2016 at 08:52:56PM -0700, Stephen Hemminger wrote:
> On Wed, 26 Oct 2016 22:30:07 +0300
> Cyrill Gorcunov <gorcunov@gmail.com> wrote:
> 
> >  
> >  				if (h->nlmsg_type == NLMSG_DONE) {
> > +					int len;
> > +
> > +					/*
> > +					 * The kernel reports if there is
> > +					 * no inet-diag module present in
> > +					 * the system via negative length
> > +					 * as error code.
> > +					 */
> > +					if (h->nlmsg_len < NLMSG_LENGTH(sizeof(int))) {
> > +						fprintf(stderr, "Truncated length reply\n");
> > +						return -1;
> > +					}
> > +					len = *(int *)NLMSG_DATA(h);
> > +					if (len < 0) {
> > +						errno = -len;
> > +						if (errno == ENOENT ||
> > +						    errno == EOPNOTSUPP)
> > +							return -1;
> > +						perror("RTNETLINK answers");
> > +						return len;
> > +					}
> >  					found_done = 1;
> >  					break; /* process next filter */
> >  				
> 
> This looks like a mistake in how you implemented the functionality in the kernel.
> Despite what it looks like, all netlink request/reply functionality reports
> errors in current implementation by returning error to the sendmsg request.
> 
> What you added implies that the new kernel api is wrong, or many other usages
> are wrong.  Please fix the kernel.

No. This is not my code. This code has been in kernel for the really long time.
I don't know why you've not been doing such test in libnetlink before.

Actually I've hit this problem accidentaly -- I made a patch 2 from this
set and run it on the machine where kernel was unpatched, ie without
raw-diag module, and I found that we can't figure out if kernel notified
us that some diag module simply not present in the system. And here is
the only way to find it out.