| Message ID | 1426866170-28739-4-git-send-email-rshearma@brocade.com |
|---|---|
| State | Changes Requested, archived |
| Delegated to: | David Miller |
| Headers | show |
Robert Shearman <rshearma@brocade.com> writes: > The control plane can advertise labels for neighbours that don't have > an outgoing label. RFC 3032 s3.22 states that either the remaining > labels should be popped (if the control plane can determine that it's > safe to do so, which in light of MPLS-VPN, RFC 4364, is never the case > now) or that the packet should be discarded. I can not figure out what you are referring to. There is no section 3.2 in RFC3022. > Therefore, if the peer is unlabeled and the last label wasn't popped > then drop the packet. The peer being unlabeled is signalled by an > empty label stack. However, implicit-null still needs to be supported > (i.e. penultimate hop popping) where the incoming label is popped and > no labels are put on and the packet can still go out labeled with the > unpopped part of the stack. This is achieved by the control plane > specifying a label stack consisting of the single special > implicit-null value. As I understand it you want to handle the case for a label for which there is no next hop, and the packet should be black-holed. In struct mpls_route such routes are currently represented by routes that have no network device. And in rtnetlink should be represented with routes of type RTN_BLACKHOLE which I do not currently support parsing. But that should be simple enough to correc.t With respect to Implicit NULL it should be an error to accept a route that has an RTA_NEWDST that includes an implicit NULL. The rtnetlink is not ldp nor should it have ldp semantics and be made complicated by those semantics. The semantics of RTA_NEWDST are the labels to push on after the top most label has been popped off. I see no reason to include other mechanisms into that processing when it is easy enough to add or tweak other attributes to have those semantics. Certainly it is not something that I think is worth special casing on the fast path in mpls_forward. > Cc: "Eric W. Biederman" <ebiederm@xmission.com> > Signed-off-by: Robert Shearman <rshearma@brocade.com> > --- > net/mpls/af_mpls.c | 21 +++++++++++++++++---- > 1 file changed, 17 insertions(+), 4 deletions(-) > > diff --git a/net/mpls/af_mpls.c b/net/mpls/af_mpls.c > index bf3459a..e3586a7 100644 > --- a/net/mpls/af_mpls.c > +++ b/net/mpls/af_mpls.c > @@ -28,7 +28,8 @@ struct mpls_route { /* next hop label forwarding entry */ > struct rcu_head rt_rcu; > u32 rt_label[MAX_NEW_LABELS]; > u8 rt_protocol; /* routing protocol that set this entry */ > - u8 rt_labels; > + u8 rt_unlabeled : 1; > + u8 rt_labels : 7; > u8 rt_via_alen; > u8 rt_via_table; > u8 rt_via[0]; > @@ -201,6 +202,11 @@ static int mpls_forward(struct sk_buff *skb, struct net_device *dev, > if (unlikely(!new_header_size && dec.bos)) { > if (!mpls_egress(rt, skb, dec)) > goto drop; > + } else if (rt->rt_unlabeled) { > + /* Labeled traffic destined to unlabeled peer should > + * be discarded > + */ > + goto drop; > } else { > bool bos; > int i; > @@ -385,9 +391,16 @@ static int mpls_route_add(struct mpls_route_config *cfg) > if (!rt) > goto errout; > > - rt->rt_labels = cfg->rc_output_labels; > - for (i = 0; i < rt->rt_labels; i++) > - rt->rt_label[i] = cfg->rc_output_label[i]; > + if (cfg->rc_output_labels == 1 && > + cfg->rc_output_label[0] == LABEL_IMPLICIT_NULL) { > + rt->rt_labels = 0; > + } else { > + rt->rt_labels = cfg->rc_output_labels; > + for (i = 0; i < rt->rt_labels; i++) > + rt->rt_label[i] = cfg->rc_output_label[i]; > + if (!rt->rt_labels) > + rt->rt_unlabeled = true; > + } > rt->rt_protocol = cfg->rc_protocol; > RCU_INIT_POINTER(rt->rt_dev, dev); > rt->rt_via_table = cfg->rc_via_table; -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
ebiederm@xmission.com (Eric W. Biederman) writes: > Robert Shearman <rshearma@brocade.com> writes: > >> The control plane can advertise labels for neighbours that don't have >> an outgoing label. RFC 3032 s3.22 states that either the remaining >> labels should be popped (if the control plane can determine that it's >> safe to do so, which in light of MPLS-VPN, RFC 4364, is never the case >> now) or that the packet should be discarded. > > I can not figure out what you are referring to. There is no section 3.2 > in RFC3022. I have found it. That is is RFC3021 Section 3.22. This is something the code already does. If the label can not be looked up with mpls_route_input_rcu the packet is dropped. Beyond that I believe the rest of my comments still stand. If you want to do this explicitly some form of explicit blackhole route needs to be supported. Either just allowing a route to be configured with no output device or an explicit RTN_BLACKHOLE route. >> Therefore, if the peer is unlabeled and the last label wasn't popped >> then drop the packet. The peer being unlabeled is signalled by an >> empty label stack. However, implicit-null still needs to be supported >> (i.e. penultimate hop popping) where the incoming label is popped and >> no labels are put on and the packet can still go out labeled with the >> unpopped part of the stack. This is achieved by the control plane >> specifying a label stack consisting of the single special >> implicit-null value. > > As I understand it you want to handle the case for a label for which > there is no next hop, and the packet should be black-holed. > > In struct mpls_route such routes are currently represented by routes > that have no network device. And in rtnetlink should be represented > with routes of type RTN_BLACKHOLE which I do not currently support > parsing. But that should be simple enough to correc.t > > With respect to Implicit NULL it should be an error to accept a route > that has an RTA_NEWDST that includes an implicit NULL. > > The rtnetlink is not ldp nor should it have ldp semantics and be made > complicated by those semantics. > > The semantics of RTA_NEWDST are the labels to push on after the top most > label has been popped off. I see no reason to include other mechanisms > into that processing when it is easy enough to add or tweak other > attributes to have those semantics. > > Certainly it is not something that I think is worth special casing on > the fast path in mpls_forward. > >> Cc: "Eric W. Biederman" <ebiederm@xmission.com> >> Signed-off-by: Robert Shearman <rshearma@brocade.com> >> --- >> net/mpls/af_mpls.c | 21 +++++++++++++++++---- >> 1 file changed, 17 insertions(+), 4 deletions(-) >> >> diff --git a/net/mpls/af_mpls.c b/net/mpls/af_mpls.c >> index bf3459a..e3586a7 100644 >> --- a/net/mpls/af_mpls.c >> +++ b/net/mpls/af_mpls.c >> @@ -28,7 +28,8 @@ struct mpls_route { /* next hop label forwarding entry */ >> struct rcu_head rt_rcu; >> u32 rt_label[MAX_NEW_LABELS]; >> u8 rt_protocol; /* routing protocol that set this entry */ >> - u8 rt_labels; >> + u8 rt_unlabeled : 1; >> + u8 rt_labels : 7; >> u8 rt_via_alen; >> u8 rt_via_table; >> u8 rt_via[0]; >> @@ -201,6 +202,11 @@ static int mpls_forward(struct sk_buff *skb, struct net_device *dev, >> if (unlikely(!new_header_size && dec.bos)) { >> if (!mpls_egress(rt, skb, dec)) >> goto drop; >> + } else if (rt->rt_unlabeled) { >> + /* Labeled traffic destined to unlabeled peer should >> + * be discarded >> + */ >> + goto drop; >> } else { >> bool bos; >> int i; >> @@ -385,9 +391,16 @@ static int mpls_route_add(struct mpls_route_config *cfg) >> if (!rt) >> goto errout; >> >> - rt->rt_labels = cfg->rc_output_labels; >> - for (i = 0; i < rt->rt_labels; i++) >> - rt->rt_label[i] = cfg->rc_output_label[i]; >> + if (cfg->rc_output_labels == 1 && >> + cfg->rc_output_label[0] == LABEL_IMPLICIT_NULL) { >> + rt->rt_labels = 0; >> + } else { >> + rt->rt_labels = cfg->rc_output_labels; >> + for (i = 0; i < rt->rt_labels; i++) >> + rt->rt_label[i] = cfg->rc_output_label[i]; >> + if (!rt->rt_labels) >> + rt->rt_unlabeled = true; >> + } >> rt->rt_protocol = cfg->rc_protocol; >> RCU_INIT_POINTER(rt->rt_dev, dev); >> rt->rt_via_table = cfg->rc_via_table; -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 22/03/15 21:06, Eric W. Biederman wrote: > ebiederm@xmission.com (Eric W. Biederman) writes: > >> Robert Shearman <rshearma@brocade.com> writes: >> >>> The control plane can advertise labels for neighbours that don't have >>> an outgoing label. RFC 3032 s3.22 states that either the remaining >>> labels should be popped (if the control plane can determine that it's >>> safe to do so, which in light of MPLS-VPN, RFC 4364, is never the case >>> now) or that the packet should be discarded. >> >> I can not figure out what you are referring to. There is no section 3.2 >> in RFC3022. > > I have found it. That is is RFC3021 Section 3.22. This is something > the code already does. If the label can not be looked up with > mpls_route_input_rcu the packet is dropped. No, the existing code handles the lack of an incoming label. s3.22 is stating what should be done with the lack of an outgoing label. > Beyond that I believe the rest of my comments still stand. If you want > to do this explicitly some form of explicit blackhole route needs to be > supported. Either just allowing a route to be configured with no output > device or an explicit RTN_BLACKHOLE route. No, that isn't going to address the problem this patch solves. > >>> Therefore, if the peer is unlabeled and the last label wasn't popped >>> then drop the packet. The peer being unlabeled is signalled by an >>> empty label stack. However, implicit-null still needs to be supported >>> (i.e. penultimate hop popping) where the incoming label is popped and >>> no labels are put on and the packet can still go out labeled with the >>> unpopped part of the stack. This is achieved by the control plane >>> specifying a label stack consisting of the single special >>> implicit-null value. >> >> As I understand it you want to handle the case for a label for which >> there is no next hop, and the packet should be black-holed. >> >> In struct mpls_route such routes are currently represented by routes >> that have no network device. And in rtnetlink should be represented >> with routes of type RTN_BLACKHOLE which I do not currently support >> parsing. But that should be simple enough to correc.t >> >> With respect to Implicit NULL it should be an error to accept a route >> that has an RTA_NEWDST that includes an implicit NULL. >> >> The rtnetlink is not ldp nor should it have ldp semantics and be made >> complicated by those semantics. This isn't specific to LDP - it is used by MP-BGP as well, or indeed would be perfectly valid to be specified in static configuration. As per RFC3031 s4.1.5 (https://tools.ietf.org/html/rfc3031#section-4.1.5) this signals that penultimate hop popping should be done, as opposed to dropping the packet if it would go out as MPLS (s3.22). Thanks, Rob >> The semantics of RTA_NEWDST are the labels to push on after the top most >> label has been popped off. I see no reason to include other mechanisms >> into that processing when it is easy enough to add or tweak other >> attributes to have those semantics. >> >> Certainly it is not something that I think is worth special casing on >> the fast path in mpls_forward. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/net/mpls/af_mpls.c b/net/mpls/af_mpls.c index bf3459a..e3586a7 100644 --- a/net/mpls/af_mpls.c +++ b/net/mpls/af_mpls.c @@ -28,7 +28,8 @@ struct mpls_route { /* next hop label forwarding entry */ struct rcu_head rt_rcu; u32 rt_label[MAX_NEW_LABELS]; u8 rt_protocol; /* routing protocol that set this entry */ - u8 rt_labels; + u8 rt_unlabeled : 1; + u8 rt_labels : 7; u8 rt_via_alen; u8 rt_via_table; u8 rt_via[0]; @@ -201,6 +202,11 @@ static int mpls_forward(struct sk_buff *skb, struct net_device *dev, if (unlikely(!new_header_size && dec.bos)) { if (!mpls_egress(rt, skb, dec)) goto drop; + } else if (rt->rt_unlabeled) { + /* Labeled traffic destined to unlabeled peer should + * be discarded + */ + goto drop; } else { bool bos; int i; @@ -385,9 +391,16 @@ static int mpls_route_add(struct mpls_route_config *cfg) if (!rt) goto errout; - rt->rt_labels = cfg->rc_output_labels; - for (i = 0; i < rt->rt_labels; i++) - rt->rt_label[i] = cfg->rc_output_label[i]; + if (cfg->rc_output_labels == 1 && + cfg->rc_output_label[0] == LABEL_IMPLICIT_NULL) { + rt->rt_labels = 0; + } else { + rt->rt_labels = cfg->rc_output_labels; + for (i = 0; i < rt->rt_labels; i++) + rt->rt_label[i] = cfg->rc_output_label[i]; + if (!rt->rt_labels) + rt->rt_unlabeled = true; + } rt->rt_protocol = cfg->rc_protocol; RCU_INIT_POINTER(rt->rt_dev, dev); rt->rt_via_table = cfg->rc_via_table;
The control plane can advertise labels for neighbours that don't have an outgoing label. RFC 3032 s3.22 states that either the remaining labels should be popped (if the control plane can determine that it's safe to do so, which in light of MPLS-VPN, RFC 4364, is never the case now) or that the packet should be discarded. Therefore, if the peer is unlabeled and the last label wasn't popped then drop the packet. The peer being unlabeled is signalled by an empty label stack. However, implicit-null still needs to be supported (i.e. penultimate hop popping) where the incoming label is popped and no labels are put on and the packet can still go out labeled with the unpopped part of the stack. This is achieved by the control plane specifying a label stack consisting of the single special implicit-null value. Cc: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: Robert Shearman <rshearma@brocade.com> --- net/mpls/af_mpls.c | 21 +++++++++++++++++---- 1 file changed, 17 insertions(+), 4 deletions(-)