Re: [PATCH ipsec] xfrm: state: match with both mark and mask on user interfaces

[Xin Long <lucien.xin@gmail.com>]

On Tue, Jul 7, 2020 at 2:17 PM Steffen Klassert
<steffen.klassert@secunet.com> wrote:
>
> On Mon, Jul 06, 2020 at 04:33:53PM +0800, Xin Long wrote:
> > >
> > > > @@ -1051,7 +1061,6 @@ xfrm_state_find(const xfrm_address_t *daddr, const xfrm_address_t *saddr,
> > > >       int acquire_in_progress = 0;
> > > >       int error = 0;
> > > >       struct xfrm_state *best = NULL;
> > > > -     u32 mark = pol->mark.v & pol->mark.m;
> > > >       unsigned short encap_family = tmpl->encap_family;
> > > >       unsigned int sequence;
> > > >       struct km_event c;
> > > > @@ -1065,7 +1074,7 @@ xfrm_state_find(const xfrm_address_t *daddr, const xfrm_address_t *saddr,
> > > >       hlist_for_each_entry_rcu(x, net->xfrm.state_bydst + h, bydst) {
> > > >               if (x->props.family == encap_family &&
> > > >                   x->props.reqid == tmpl->reqid &&
> > > > -                 (mark & x->mark.m) == x->mark.v &&
> > > > +                 (pol->mark.v == x->mark.v && pol->mark.m == x->mark.m) &&
> > > >                   x->if_id == if_id &&
> > > >                   !(x->props.flags & XFRM_STATE_WILDRECV) &&
> > > >                   xfrm_state_addr_check(x, daddr, saddr, encap_family) &&
> > > > @@ -1082,7 +1091,7 @@ xfrm_state_find(const xfrm_address_t *daddr, const xfrm_address_t *saddr,
> > > >       hlist_for_each_entry_rcu(x, net->xfrm.state_bydst + h_wildcard, bydst) {
> > > >               if (x->props.family == encap_family &&
> > > >                   x->props.reqid == tmpl->reqid &&
> > > > -                 (mark & x->mark.m) == x->mark.v &&
> > > > +                 (pol->mark.v == x->mark.v && pol->mark.m == x->mark.m) &&
> > > >                   x->if_id == if_id &&
> > > >                   !(x->props.flags & XFRM_STATE_WILDRECV) &&
> > > >                   xfrm_addr_equal(&x->id.daddr, daddr, encap_family) &&
> > >
> > > While this should usually not be a problem for strongSwan, as we set the
> > > same mark/value on both states and corresponding policies (although the
> > > latter can be disabled as users may want to install policies themselves
> > > or via another daemon e.g. for MIPv6), it might be a limitation for some
> > > use cases.  The current code allows sharing states with multiple
> > > policies whose mark/mask doesn't match exactly (i.e. depended on the
> > > masks of both).  I wonder if anybody uses it like this, and how others
> > > think about it.
> > IMHO, the non-exact match "(mark & x->mark.m) == x->mark.v" should be only
> > for packet flow. "sharing states with multiple policies" should not be the
> > purpose of xfrm_mark. (Add Jamal to the CC list)
> >
> > "(((pol->mark.v & pol->mark.m) & x->mark.m) == x->mark.v)" is just strange.
> > We could do either:
> >  (pol->mark.v == x->mark.v && pol->mark.m == x->mark.m), like this patch.
> > Or use fl->flowi_mark in xfrm_state_find():
> >  (fl->flowi_mark & x->mark.m) == x->mark.v)
> >
> > The 1st one will require userland to configure policy and state with the
> > same xfrm_mark, like strongswan.
> >
> > The 2nd one will match state with tx packet flow's mark, it's more like
> > rx packet flow path.
> >
> > But you're right, either of these may cause slight differences, let's see
> > how other userland cases use it. (also Add Libreswan's developer, Paul Wouters)
>
> We should care that everything that worked before, will still work after
> this change. Otherwise we might break someones usecase, even if the
> usecase seems to be odd and nobody complains now.
Hi, Steffen, You're right.

But what if the current one may cause problems?

policyA: mark(value: 0x11, mask: 0xff)
stateA: mark(value: 0x1, mask: 0xf)
stateB: mark(value: 0x11, mask: 0xff)

stateA is added later than stateB, and ahead of stateB in the list.

So one packet flow with mask 0x11, will get stateA through policyA, like on tx.
But one packet flow with mask 0x11, will get stateB by
xfrm_state_lookup(), like on rx.

I'm not sure if this may be an inconsistency issue.