Re: [net-next 1/2] seg6: add NEXT-C-SID support for SRv6 End.X behavior

[Andrea Mayer <andrea.mayer@uniroma2.it>]

On Sat, 5 Aug 2023 16:17:03 +0800
Hangbin Liu <liuhangbin@gmail.com> wrote:

> On Fri, Aug 04, 2023 at 02:41:18PM +0200, Andrea Mayer wrote:
> > Hi Hangbin,
> > thanks for your time. Please see below.
> > 
> > On Thu, 3 Aug 2023 17:30:28 +0800
> > Hangbin Liu <liuhangbin@gmail.com> wrote:
> > 
> > > On Mon, Jul 31, 2023 at 07:51:16PM +0200, Andrea Mayer wrote:
> > > > +/* Processing of SRv6 End, End.X, and End.T behaviors can be extended through
> > > > + * the flavors framework. These behaviors must report the subset of (flavor)
> > > > + * operations they currently implement. In this way, if a user specifies a
> > > > + * flavor combination that is not supported by a given End* behavior, the
> > > > + * kernel refuses to instantiate the tunnel reporting the error.
> > > > + */
> > > > +static int seg6_flv_supp_ops_by_action(int action, __u32 *fops)
> > > > +{
> > > > +	switch (action) {
> > > > +	case SEG6_LOCAL_ACTION_END:
> > > > +		*fops = SEG6_LOCAL_END_FLV_SUPP_OPS;
> > > > +		break;
> > > > +	case SEG6_LOCAL_ACTION_END_X:
> > > > +		*fops = SEG6_LOCAL_END_X_FLV_SUPP_OPS;
> > > > +		break;
> > > > +	default:
> > > > +		return -EOPNOTSUPP;
> > > > +	}
> > > > +
> > > > +	return 0;
> > > >  }
> > > >  
> > > 
> > > ...
> > > 
> > > > @@ -2070,7 +2131,8 @@ static int parse_nla_flavors(struct nlattr **attrs, struct seg6_local_lwt *slwt,
> > > >  {
> > > >  	struct seg6_flavors_info *finfo = &slwt->flv_info;
> > > >  	struct nlattr *tb[SEG6_LOCAL_FLV_MAX + 1];
> > > > -	unsigned long fops;
> > > > +	int action = slwt->action;
> > > > +	__u32 fops, supp_fops = 0;
> > > >  	int rc;
> > > >  
> > > >  	rc = nla_parse_nested_deprecated(tb, SEG6_LOCAL_FLV_MAX,
> > > > @@ -2086,7 +2148,8 @@ static int parse_nla_flavors(struct nlattr **attrs, struct seg6_local_lwt *slwt,
> > > >  		return -EINVAL;
> > > >  
> > > >  	fops = nla_get_u32(tb[SEG6_LOCAL_FLV_OPERATION]);
> > > > -	if (fops & ~SEG6_LOCAL_FLV_SUPP_OPS) {
> > > > +	rc = seg6_flv_supp_ops_by_action(action, &supp_fops);
> > > > +	if (rc < 0 || !supp_fops || (fops & ~supp_fops)) {
> > > 
> > > if rc == 0, the supp_fops won't be 0.
> > > 
> > 
> > Yes, you're right.
> > 
> > In this patch, supp_fops is always set properly when rc == 0.
> > Since seg6_flv_supp_ops_by_action() should be extended in the event that other
> > behaviors receive flavors support, I added this check in case the "supp_fops"
> > field was set incorrectly or not set at all.
> > Note that supp_fops == 0 must be considered an inadmissible value.
> > 
> > 
> > So, I think we have two possibilities:
> >   i) remove this "defensive" check, assuming that supp_fops will always be set
> >      correctly by seg6_flv_supp_ops_by_action() (when rc == 0, like in this
> >      patch); 
> >  ii) improve the check by explicitly indicating with a pr_warn_once, for
> >      example, the condition that is occurring is unexpected.
> > 
> > for (ii), something like this:
> > 
> > parse_nla_flavors(...)
> > {
> >     [...]
> >     supp_fops = 0;
> >     [...]
> > 
> >     rc = seg6_flv_supp_ops_by_action(action, &supp_fops);
> >     if (!rc && !supp_fops) {
> >    	 /* supported flavors mask cannot be zero as it is considered to
> >    	  * be invalid.
> >    	  */
> >    	 pr_warn_once("seg6local: invalid Flavor operation(s)");
> >    	 return -EINVAL;
> >     }
> 
> Do you mean there is a possibility *in future* that the supp_fops could be 0
> with rc == 0? If yes, this check would make sense(although we can add this
> check when it's true). If not. I don't see a need to have this check.
> 
> And some static analysis tool would report warn for this code.
> 

Good points, thanks.
There is no possibility at the moment that supp_fops could be 0 with rc == 0.
That check is going to be removed in v2.

> Thanks
> Hangbin

Ciao,
Andrea

> > 
> >     fops = nla_get_u32(tb[SEG6_LOCAL_FLV_OPERATION]);
> >     if (rc < 0 || (fops & ~supp_fops)) {
> >    	 NL_SET_ERR_MSG(extack, "Unsupported Flavor operation(s)");
> >    	 return -EOPNOTSUPP;
> >     }
> > 
> >     finfo->flv_ops = fops;
> > 
> >     [...]
> > }
> > 
> > parse_nla_flavors() is called in the control path so another check would not
> > hit performance. I am more inclined to consider solution (ii).
> > 
> > What do you think?
> > 
> > > Thanks
> > > Hangbin
> > 
> > Ciao,
> > Andrea