Re: [PATCH net-next] xfrm: policy: Remove obsolete WARN while xfrm policy inserting

[Steffen Klassert <steffen.klassert@secunet.com>]

On Fri, Apr 17, 2020 at 07:01:52PM +0800, Yuehaibing wrote:
> On 2020/4/15 15:14, Steffen Klassert wrote:
> > On Thu, Apr 09, 2020 at 04:19:37PM +0800, Yuehaibing wrote:
> >>
> >>
> >> On 2020/4/6 17:03, Steffen Klassert wrote:
> >>> On Mon, Mar 30, 2020 at 10:05:32PM +0800, Yuehaibing wrote:
> >>>> On 2020/3/28 19:23, Steffen Klassert wrote:
> >>>>> On Fri, Mar 27, 2020 at 08:34:43PM +0800, YueHaibing wrote:
> >>>>>> Since commit 7cb8a93968e3 ("xfrm: Allow inserting policies with matching
> >>>>>> mark and different priorities"), we allow duplicate policies with
> >>>>>> different priority, this WARN is not needed any more.
> >>>>>
> >>>>> Can you please describe a bit more detailed why this warning
> >>>>> can't trigger anymore?
> >>>>
> >>>> No, this warning is triggered while detect a duplicate entry in the policy list
> >>>>
> >>>> regardless of the priority. If we insert policy like this:
> >>>>
> >>>> policy A (mark.v = 3475289, mark.m = 0, priority = 1)	//A is inserted
> >>>> policy B (mark.v = 0, mark.m = 0, priority = 0) 	//B is inserted
> >>>> policy C (mark.v = 3475289, mark.m = 0, priority = 0)	//C is inserted and B is deleted
> >>>
> >>> The codepath that replaces a policy by another should just trigger
> >>> on policy updates (XFRM_MSG_UPDPOLICY). Is that the case in your
> >>> test?
> >>
> >> Yes, this is triggered by XFRM_MSG_UPDPOLICY
> >>
> >>>
> >>> It should not be possible to add policy C with XFRM_MSG_NEWPOLICY
> >>> as long as you have policy B inserted.
> >>>
> >>> The update replaces an old policy by a new one, the lookup keys of
> >>> the old policy must match the lookup keys of the new one. But policy
> >>> B has not the same lookup keys as C, the mark is different. So B should
> >>> not be replaced with C.
> >>
> >> 1436 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
> >> 1437                                    struct xfrm_policy *pol)
> >> 1438 {
> >> 1439         u32 mark = policy->mark.v & policy->mark.m;
> >> 1440
> >> 1441         if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
> >> 1442                 return true;
> >> 1443
> >> 1444         if ((mark & pol->mark.m) == pol->mark.v &&    //policy is C, pol is B, so mark is 0, pol->mark.m is 0, pol->mark.v is 0
> >> 1445             policy->priority == pol->priority)	   //priority is same zero, so return true, B is replaced with C
> >> 1446                 return true;
> >> 1447
> >> 1448         return false;
> >> 1449 }
> >>
> >> Should xfrm_policy_mark_match be fixed？
> > 
> > Yes, xfrm_policy_mark_match should only replace if the found
> > policy has the same lookup keys.
> 
> I'm wonder that lookup keys means association of mark.v and mark.m, or the mark (mark.v & mark.m).

Good point. I'd say the lookup lookup keys are identical if the policy
lookup can't distinguish between the policies. So (mark.v & mark.m)
should be it.