Re: [PATCHv3 iproute2-next] rtnetlink: add new function rtnl_echo_talk()

[Ido Schimmel <idosch@idosch.org>]

On Tue, Nov 08, 2022 at 05:09:05PM +0800, Hangbin Liu wrote:
> On Tue, Nov 08, 2022 at 10:40:45AM +0200, Ido Schimmel wrote:
> > > +	return rtnl_talk(&rth, &req.n, NULL);
> > >  }
> > 
> > Hangbin,
> > 
> > This change breaks the nexthop selftest:
> > tools/testing/selftests/net/fib_nexthops.sh
> > 
> > Which is specifically checking for "2" as the error code. Example:
> 
> Hi Ido,
> 
> Thanks for the report.
> 
> > 
> > # attempt to create nh without a device or gw - fails
> > run_cmd "$IP nexthop add id 1"
> > log_test $? 2 "Nexthop with no device or gateway"
> > 
> > I think it's better to restore the original error code than "fixing" all
> > the tests / applications that rely on it.
> 
> I can fix this either in iproute2 or in the selftests.
> I'd perfer ask David's opinion.

Sure, but note that:

1. Other than the 4 selftests that we know about and can easily patch,
there might be a lot of other applications that invoke iproute2 and
expect this return code. It is used by iproute2 since at least 2004.

2. There is already precedence for restoring the original code. See
commit d58ba4ba2a53 ("ip: return correct exit code on route failure").

> 
> > 
> > The return code of other subcommands was also changed by this patch, but
> > so far all the failures I have seen are related to "nexthop" subcommand.
> 
> I grep "log_test \$? 2" in selftest/net folder and found the following tests
> would use it
> 
> fib_tests.sh
> test_vxlan_vnifiltering.sh
> fcnal-test.sh
> fib_nexthops.sh
> 
> Thanks
> Hangbin