Network Working Group A. Ayyangar

Network Working Group A. Ayyangar
Request for Comments: 5150 K. Kompella
Category: Standards Track Juniper Networks
JP. Vasseur
Cisco Systems, Inc.
A. Farrel
Old Dog Consulting
February 2008

Label Switched Path Stitching with
Generalized Multiprotocol Label Switching Traffic Engineering (GMPLS TE)

Status of This Memo

This document specifies an Internet standards track protocol for the
Internet community, and requests discussion and suggestions for
improvements. Please refer to the current edition of the "Internet
Official Protocol Standards" (STD 1) for the standardization state
and status of this protocol. Distribution of this memo is unlimited.

Abstract

In certain scenarios, there may be a need to combine several
Generalized Multiprotocol Label Switching (GMPLS) Label Switched
Paths (LSPs) such that a single end-to-end (e2e) LSP is realized and
all traffic from one constituent LSP is switched onto the next LSP.
We will refer to this as "LSP stitching", the key requirement being
that a constituent LSP not be allocated to more than one e2e LSP.
The constituent LSPs will be referred to as "LSP segments" (S-LSPs).

This document describes extensions to the existing GMPLS signaling
protocol (Resource Reservation Protocol-Traffic Engineering (RSVP-
TE)) to establish e2e LSPs created from S-LSPs, and describes how the
LSPs can be managed using the GMPLS signaling and routing protocols.

It may be possible to configure a GMPLS node to switch the traffic
from an LSP for which it is the egress, to another LSP for which it
is the ingress, without requiring any signaling or routing extensions
whatsoever and such that the operation is completely transparent to
other nodes. This will also result in LSP stitching in the data
plane. However, this document does not cover this scenario of LSP
stitching.

Ayyangar, et al. Standards Track [Page 1]

RFC 5150 LSP Stitching with GMPLS TE February 2008

Table of Contents

1. Introduction ....................................................2
1.1. Conventions Used in This Document ..........................3
2. Comparison with LSP Hierarchy ...................................3
3. Usage ...........................................................4
3.1. Triggers for LSP Segment Setup .............................4
3.2. Applications ...............................................5
4. Routing Aspects .................................................5
5. Signaling Aspects ...............................................6
5.1. RSVP-TE Signaling Extensions ...............................7
5.1.1. Creating and Preparing an LSP Segment for
Stitching ...........................................7
5.1.1.1. Steps to Support Penultimate Hop
Popping ....................................8
5.1.2. Stitching the e2e LSP to the LSP Segment ............9
5.1.3. RRO Processing for e2e LSPs ........................10
5.1.4. Teardown of LSP Segments ...........................11
5.1.5. Teardown of e2e LSPs ...............................11
5.2. Summary of LSP Stitching Procedures .......................12
5.2.1. Example Topology ...................................12
5.2.2. LSP Segment Setup ..................................12
5.2.3. Setup of an e2e LSP ................................13
5.2.4. Stitching of an e2e LSP into an LSP Segment ........13
6. Security Considerations ........................................14
7. IANA Considerations ............................................15
7.1. Attribute Flags for LSP_ATTRIBUTES Object .................15
7.2. New Error Codes ...........................................15
8. Acknowledgments ................................................16
9. References .....................................................16
9.1. Normative References ......................................16
9.2. Informative References ....................................17

1. Introduction

A stitched Generalized Multiprotocol Label Switching (GMPLS) Traffic
Engineering (TE) Label Switched Path (LSP) is built from a set of
different "LSP segments" (S-LSPs) that are connected together in the
data plane in such a way that a single end-to-end LSP is realized in
the data plane. In this document, we define the concept of LSP
stitching and detail the control plane mechanisms and procedures
(routing and signaling) to accomplish this. Where applicable,
similarities and differences between LSP hierarchy [RFC4206] and LSP
stitching are highlighted. Signaling extensions required for LSP
stitching are also described here.

Ayyangar, et al. Standards Track [Page 2]

RFC 5150 LSP Stitching with GMPLS TE February 2008

It may be possible to configure a GMPLS node to switch the traffic
from an LSP for which it is the egress, to another LSP for which it
is the ingress, without requiring any signaling or routing extensions
whatsoever and such that the operation is completely transparent to
other nodes. This results in LSP stitching in the data plane, but
requires management intervention at the node where the stitching is
performed. With the mechanism described in this document, the node
performing the stitching does not require configuration of the pair
of S-LSPs to be stitched together. Also, LSP stitching as defined
here results in an end-to-end LSP both in the control and data
planes.

1.1. Conventions Used in This Document

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in RFC 2119 [RFC2119].

2. Comparison with LSP Hierarchy

LSP hierarchy ([RFC4206]) provides signaling and routing procedures
so that:

a. A Hierarchical LSP (H-LSP) can be created. Such an LSP created in
one layer can appear as a data link to LSPs in higher layers. As
such, one or more LSPs in a higher layer can traverse this H-LSP
as a single hop; we call this "nesting".

b. An H-LSP may be managed and advertised (although this is not a
requirement) as a Traffic Engineering (TE) link. Advertising an
H-LSP as a TE link allows other nodes in the TE domain in which it
is advertised to use this H-LSP in path computation. If the H-LSP
TE link is advertised in the same instance of control plane (TE
domain) in which the H-LSP was provisioned, it is then defined as
a forwarding adjacency LSP (FA-LSP) and GMPLS nodes can form a
forwarding adjacency (FA) over this FA-LSP. There is usually no
routing adjacency between end points of an FA. An H-LSP may also
be advertised as a TE link in a different TE domain. In this
case, the end points of the H-LSP are required to have a routing
adjacency between them.

c. RSVP signaling ([RFC3473], [RFC3209]) for LSP setup can occur
between nodes that do not have a routing adjacency.

In case of LSP stitching, instead of an H-LSP, an LSP segment (S-LSP)
is created between two GMPLS nodes. An S-LSP for stitching is
considered to be the moral equivalent of an H-LSP for nesting. An
S-LSP created in one layer, unlike an H-LSP, provides a data link to

Ayyangar, et al. Standards Track [Page 3]

RFC 5150 LSP Stitching with GMPLS TE February 2008

other LSPs in the same layer. Similar to an H-LSP, an S-LSP could be
managed and advertised, although it is not required, as a TE link,
either in the same TE domain as it was provisioned or a different
one. If so advertised, other GMPLS nodes can use the corresponding
S-LSP TE link in path computation. While there is a forwarding
adjacency between end points of an H-LSP TE link, there is no
forwarding adjacency between end points of an S-LSP TE link. In this
aspect, an H-LSP TE link more closely resembles a 'basic' TE link as
compared to an S-LSP TE link.

While LSP hierarchy allows more than one LSP to be mapped to an H-
LSP, in case of LSP stitching, at most one LSP may be associated with
an S-LSP. Thus, if LSP-AB is an H-LSP between nodes A and B, then
multiple LSPs, say LSP1, LSP2, and LSP3, can potentially be 'nested
into' LSP-AB. This is achieved by exchanging a unique label for each
of LSP1..3 over the LSP-AB hop, thereby separating the data
corresponding to each of LSP1..3 while traversing the H-LSP LSP-AB.
Each of LSP1..3 may reserve some bandwidth on LSP-AB. On the other
hand, if LSP-AB is an S-LSP, then at most one LSP, say LSP1, may be
stitched to the S-LSP LSP-AB. LSP-AB is then dedicated to LSP1, and
no other LSPs can be associated with LSP-AB. The entire bandwidth on
S-LSP LSP-AB is allocated to LSP1. However, similar to H-LSPs,
several S-LSPs may be bundled into a TE link ([RFC4201]).

The LSPs LSP1..3 that are either nested or stitched into another LSP
are termed as e2e LSPs in the rest of this document. Routing
procedures specific to LSP stitching are detailed in Section 4.

Targeted (non-adjacent) RSVP signaling defined in [RFC4206] is
required for LSP stitching of an e2e LSP to an S-LSP. Specific
extensions for LSP stitching are described in Section 5.1.
Therefore, in the control plane, there is one RSVP session
corresponding to the e2e LSP as well as one for each S-LSP. The
creation and termination of an S-LSP may be dictated by
administrative control (statically provisioned) or due to another
incoming LSP request (dynamic). Triggers for dynamic creation of an
S-LSP may be different from that of an H-LSP and will be described in
detail in Section 3.1.

3. Usage

3.1. Triggers for LSP Segment Setup

An S-LSP may be created either by administrative control
(configuration trigger) or dynamically due to an incoming LSP
request. LSP hierarchy ([RFC4206]) defines one possible trigger for
dynamic creation of an FA-LSP by introducing the notion of LSP
regions based on Interface Switching Capabilities. As per [RFC4206],

Ayyangar, et al. Standards Track [Page 4]

RFC 5150 LSP Stitching with GMPLS TE February 2008

dynamic FA-LSP creation may be triggered on a node when an incoming
LSP request crosses region boundaries. However, this trigger MUST
NOT be used for creation of an S-LSP for LSP stitching as described
in this document. In case of LSP stitching, the switching
capabilities of the previous hop and the next hop TE links MUST be
the same. Therefore, local policies configured on the node SHOULD be
used for dynamic creation of LSP segments.

Other possible triggers for dynamic creation of both H-LSPs and S-
LSPs include cases where an e2e LSP may cross domain boundaries or
satisfy locally configured policies on the node as described in
[RFC5151].

3.2. Applications

LSP stitching procedures described in this document are applicable to
GMPLS nodes that need to associate an e2e LSP with another S-LSP of
the same switching type and LSP hierarchy procedures do not apply.
For example, if an e2e lambda LSP traverses an LSP segment TE link
that is also lambda-switch capable, then LSP hierarchy is not
possible; in this case, LSP switching may be an option.

LSP stitching procedures can be used for inter-domain TE LSP
signaling to stitch an inter-domain e2e LSP to a local intra-domain
TE S-LSP ([RFC4726] and [RFC5151]).

LSP stitching may also be useful in networks to bypass legacy nodes
that may not have certain new capabilities in the control plane
and/or data plane. For example, one suggested usage in the case of
point-to-multipoint (P2MP) RSVP LSPs ([RFC4875]) is the use of LSP
stitching to stitch a P2MP RSVP LSP to an LSP segment between P2MP-
capable Label Switching Routers (LSRs) in the network. The LSP
segment would traverse legacy LSRs that may be incapable of acting as
P2MP branch points, thereby shielding them from the P2MP control and
data path. Note, however, that such configuration may limit the
attractiveness of RSVP P2MP and should carefully be examined before
deployment.

4. Routing Aspects

An S-LSP is created between two GMPLS nodes, and it may traverse zero
or more intermediate GMPLS nodes. There is no forwarding adjacency
between the end points of an S-LSP TE link. So although in the TE
topology, the end points of an S-LSP TE link are adjacent, in the
data plane, these nodes do not have an adjacency. Hence, any data
plane resource identifier between these nodes is also meaningless.

Ayyangar, et al. Standards Track [Page 5]

RFC 5150 LSP Stitching with GMPLS TE February 2008

The traffic that arrives at the head end of the S-LSP is switched
into the S-LSP contiguously with a label swap, and no label is
associated directly between the end nodes of the S-LSP itself.

An S-LSP MAY be treated and managed as a TE link. This TE link MAY
be numbered or unnumbered. For an unnumbered S-LSP TE link, the
schemes for assignment and handling of the local and remote link
identifiers as specified in [RFC3477] SHOULD be used. When
appropriate, the TE information associated with an S-LSP TE link MAY
be flooded via ISIS-TE [RFC4205] or OSPF-TE [RFC4203]. Mechanisms
similar to that for regular (basic) TE links SHOULD be used to flood
S-LSP TE links. Advertising or flooding the S-LSP TE link is not a
requirement for LSP stitching. If advertised, this TE information
will exist in the TE database (TED) and can then be used for path
computation by other GMPLS nodes in the TE domain in which it is
advertised. When so advertising S-LSPs, one should keep in mind that
these add to the size and complexity of the link-state database.

If an S-LSP is advertised as a TE link in the same TE domain in which
it was provisioned, there is no need for a routing adjacency between
end points of this S-LSP TE link. If an S-LSP TE link is advertised
in a different TE domain, the end points of that TE link SHOULD have
a routing adjacency between them.

The TE parameters defined for an FA in [RFC4206] SHOULD be used for
an S-LSP TE link as well. The switching capability of an S-LSP TE
link MUST be equal to the switching type of the underlying S-LSP;
i.e., an S-LSP TE link provides a data link to other LSPs in the same
layer, so no hierarchy is possible.

An S-LSP MUST NOT admit more than one e2e LSP into it. If an S-LSP
is allocated to an e2e LSP, the unreserved bandwidth SHOULD be set to
zero to prevent further e2e LSPs from being admitted into the S-LSP.

Multiple S-LSPs between the same pair of nodes MAY be bundled using
the concept of Link Bundling ([RFC4201]) into a single TE link. In
this case, each component S-LSP may be allocated to at most one e2e
LSP. When any component S-LSP is allocated for an e2e LSP, the
component's unreserved bandwidth SHOULD be set to zero and the
Minimum and Maximum LSP bandwidth of the TE link SHOULD be
recalculated. This will prevent more than one LSP from being
computed and admitted over an S-LSP.

5. Signaling Aspects

The end nodes of an S-LSP may or may not have a routing adjacency.
However, they SHOULD have a signaling adjacency (RSVP neighbor
relationship) and will exchange RSVP messages with each other. It

Ayyangar, et al. Standards Track [Page 6]

RFC 5150 LSP Stitching with GMPLS TE February 2008

may, in fact, be desirable to exchange RSVP Hellos directly between
the LSP segment end points to allow support for state recovery during
Graceful Restart procedures as described in [RFC3473].

In order to signal an e2e LSP over an LSP segment, signaling
procedures described in Section 8.1.1 of [RFC4206] MUST be used.
Additional signaling extensions for stitching are described in the
next section.

5.1. RSVP-TE Signaling Extensions

The signaling extensions described here MUST be used for stitching an
e2e packet or non-packet GMPLS LSP ([RFC3473]) to an S-LSP.

Stitching an e2e LSP to an LSP segment involves the following two-
step process:

1. Creating and preparing the S-LSP for stitching by signaling the
desire to stitch between end points of the S-LSP; and

2. Stitching the e2e LSP to the S-LSP.

5.1.1. Creating and Preparing an LSP Segment for Stitching

If a GMPLS node desires to create an S-LSP, i.e., one to be used for
stitching, then it MUST indicate this in the Path message for the S-
LSP. This signaling explicitly informs the S-LSP egress node that
the ingress node is planning to perform stitching over the S-LSP.
Since an S-LSP is not conceptually different from any other LSP,
explicitly signaling 'LSP stitching desired' helps clarify the data
plane actions to be carried out when the S-LSP is used by some other
e2e LSP. Also, in the case of packet LSPs, this is what allows the
egress of the S-LSP to carry out label allocation as explained below.
Also, so that the head-end node can ensure that correct stitching
actions will be carried out at the egress node, the egress node MUST
signal this information back to the head-end node in the Resv, as
explained below.

In order to request LSP stitching on the S-LSP, we define a new bit
in the Attributes Flags TLV of the LSP_ATTRIBUTES object defined in
[RFC4420]:

LSP stitching desired bit - This bit SHOULD be set in the Attributes
Flags TLV of the LSP_ATTRIBUTES object in the Path message for the
S-LSP by the head end of the S-LSP that desires LSP stitching. This
bit MUST NOT be modified by any other nodes in the network. Nodes
other than the egress of the S-LSP SHOULD ignore this bit. The bit
number for this flag is defined in Section 7.1.

Ayyangar, et al. Standards Track [Page 7]

RFC 5150 LSP Stitching with GMPLS TE February 2008

An LSP segment can be used for stitching only if the egress node of
the S-LSP is also ready to participate in stitching. In order to
indicate this to the head-end node of the S-LSP, the following new
bit is defined in the Flags field of the Record Route object (RRO)
Attributes subobject: "LSP segment stitching ready". The bit number
for this flag is defined in Section 7.1.

If an egress node of the S-LSP receiving the Path message supports
the LSP_ATTRIBUTES object and the Attributes Flags TLV, and also
recognizes the "LSP stitching desired" bit, but cannot support the
requested stitching behavior, then it MUST send back a PathErr
message with an error code of "Routing Problem" and an error value of
"Stitching unsupported" to the head-end node of the S-LSP. The new
error value is defined in Section 7.2.

If an egress node receiving a Path message with the "LSP stitching
desired" bit set in the Flags field of received LSP_ATTRIBUTES object
recognizes the object, the TLV TLV, and the bit and also supports the
desired stitching behavior, then it MUST allocate a non-NULL label
for that S-LSP in the corresponding Resv message. Also, so that the
head-end node can ensure that the correct label (forwarding) actions
will be carried out by the egress node and that the S-LSP can be used
for stitching, the egress node MUST set the "LSP segment stitching
ready" bit defined in the Flags field of the RRO Attribute subobject.

Finally, if the egress node for the S-LSP supports the LSP_ATTRIBUTES
object but does not recognize the Attributes Flags TLV, or supports
the TLV as well but does not recognize this particular bit, then it
SHOULD simply ignore the above request.

An ingress node requesting LSP stitching MUST examine the RRO
Attributes subobject Flags corresponding to the egress node for the
S-LSP, to make sure that stitching actions are carried out at the
egress node. It MUST NOT use the S-LSP for stitching if the "LSP
segment stitching ready" bit is cleared.

5.1.1.1. Steps to Support Penultimate Hop Popping

Note that this section is only applicable to packet LSPs that use
Penultimate Hop Popping (PHP) at the last hop, where the egress node
distributes the Implicit NULL Label ([RFC3032]) in the Resv Label.
These steps MUST NOT be used for a non-packet LSP and for packet LSPs
where PHP is not desired.

When the egress node of a packet S-LSP receives a Path message for an
e2e LSP that uses the S-LSP, the egress of the S-LSP SHOULD first
check to see if it is also the egress of the e2e LSP. If the egress
node is the egress for both the S-LSP and the e2e TE LSP, and this is

Ayyangar, et al. Standards Track [Page 8]

RFC 5150 LSP Stitching with GMPLS TE February 2008

a packet LSP that requires PHP, then the node MUST send back a Resv
trigger message for the S-LSP with a new label corresponding to the
Implicit or Explicit NULL Label. Note that this operation does not
cause any traffic disruption because the S-LSP is not carrying any
traffic at this time, since the e2e LSP has not yet been established.

If the e2e LSP and the S-LSP are bidirectional, the ingress of the
e2e LSP SHOULD first check whether it is also the ingress of the S-
LSP. If so, it SHOULD re-issue the Path message for the S-LSP with
an Implicit or Explicit NULL Upstream Label, and only then proceed
with the signaling of the e2e LSP.

5.1.2. Stitching the e2e LSP to the LSP Segment

When a GMPLS node receives an e2e LSP request, depending on the
applicable trigger, it may either dynamically create an S-LSP based
on procedures described above or map an e2e LSP to an existing S-LSP.
The switching type in the Generalized Label Request of the e2e LSP
MUST be equal to the switching type of the S-LSP. Other constraints
like the explicit path encoded in the Explicit Route object (ERO),
bandwidth, and local TE policies MUST also be used for S-LSP
selection or signaling. In either case, once an S-LSP has been
selected for an e2e LSP, the following procedures MUST be followed in
order to stitch an e2e LSP to an S-LSP.

The GMPLS node receiving the e2e LSP setup Path message MUST use the
signaling procedures described in [RFC4206] to send the Path message
to the end point of the S-LSP. In this Path message, the node MUST
identify the S-LSP in the RSVP_HOP. An egress node receiving this
RSVP_HOP should also be able to identify the S-LSP TE link based on
the information signaled in the RSVP_HOP. If the S-LSP TE link is
numbered, then the addressing scheme as proposed in [RFC4206] SHOULD
be used to number the S-LSP TE link. If the S-LSP TE link is
unnumbered, then any of the schemes proposed in [RFC3477] SHOULD be
used to exchange S-LSP TE link identifiers between the S-LSP end
points. If the TE link is bundled, the RSVP_HOP SHOULD identify the
component link as defined in [RFC4201].

In case of a bidirectional e2e TE LSP, an Upstream Label MUST be
signaled in the Path message for the e2e LSP over the S-LSP hop.
However, since there is no forwarding adjacency between the S-LSP end
points, any label exchanged between them has no significance. So the
node MAY chose any label value for the Upstream Label. The label
value chosen and signaled by the node in the Upstream Label is out of
the scope of this document and is specific to the implementation on
that node. The egress node receiving this Path message MUST ignore
the Upstream Label in the Path message over the S-LSP hop.

Ayyangar, et al. Standards Track [Page 9]

RFC 5150 LSP Stitching with GMPLS TE February 2008

The egress node receiving this Path message MUST signal a Label in
the Resv message for the e2e TE LSP over the S-LSP hop. Again, since
there is no forwarding adjacency between the egress and ingress S-LSP
nodes, any label exchanged between them is meaningless. So the
egress node MAY choose any label value for the Label. The label
value chosen and signaled by the egress node is out of the scope of
this document and is specific to the implementation on the egress
node. The egress S-LSP node SHOULD also carry out data plane
operations so that traffic coming in on the S-LSP is switched over to
the e2e LSP downstream, if the egress of the e2e LSP is some other
node downstream. If the e2e LSP is bidirectional, this means setting
up label switching in both directions. The Resv message from the
egress S-LSP node is IP routed back to the previous hop (ingress of
the S-LSP). The ingress node stitching an e2e TE LSP to an S-LSP
MUST ignore the Label object received in the Resv for the e2e TE LSP
over the S-LSP hop. The S-LSP ingress node SHOULD also carry out
data plane operations so that traffic coming in on the e2e LSP is
switched into the S-LSP. It should also carry out actions to handle
traffic in the opposite direction if the e2e LSP is bidirectional.

Note that the label exchange procedure for LSP stitching on the S-LSP
hop is similar to that for LSP hierarchy over the H-LSP hop. The
difference is the lack of the significance of this label between the
S-LSP end points in case of stitching. Therefore, in case of
stitching, the recipients of the Label/Upstream Label MUST NOT
process these labels. Also, at most one e2e LSP is associated with
one S-LSP. If a node at the head end of an S-LSP receives a Path
message for an e2e LSP that identifies the S-LSP in the ERO and the
S-LSP bandwidth has already been allocated to some other LSP, then
regular rules of RSVP-TE pre-emption apply to resolve contention for
S-LSP bandwidth. If the LSP request over the S-LSP cannot be
satisfied, then the node SHOULD send back a PathErr with the error
codes as described in [RFC3209].

5.1.3. RRO Processing for e2e LSPs

RRO procedures for the S-LSP specific to LSP stitching are already
described in Section 5.1.1. In this section, we will look at the RRO
processing for the e2e LSP over the S-LSP hop.

An e2e LSP traversing an S-LSP SHOULD record in the RRO for that hop,
an identifier corresponding to the S-LSP TE link. This is applicable
to both Path and Resv messages over the S-LSP hop. If the S-LSP is
numbered, then the IPv4 or IPv6 address subobject ([RFC3209]) SHOULD
be used to record the S-LSP TE link address. If the S-LSP is
unnumbered, then the Unnumbered Interface ID subobject as described
in [RFC3477] SHOULD be used to record the node's Router ID and
Interface ID of the S-LSP TE link. In either case, the RRO subobject

Ayyangar, et al. Standards Track [Page 10]

RFC 5150 LSP Stitching with GMPLS TE February 2008

SHOULD identify the S-LSP TE link end point. Intermediate links or
nodes traversed by the S-LSP itself SHOULD NOT be recorded in the RRO
for the e2e LSP over the S-LSP hop.

5.1.4. Teardown of LSP Segments

S-LSP teardown follows the standard procedures defined in [RFC3209]
and [RFC3473]. This includes procedures without and with setting the
administrative status. Teardown of S-LSP may be initiated by the
ingress, egress, or any other node along the S-LSP path.
Deletion/teardown of the S-LSP SHOULD be treated as a failure event
for the e2e LSP associated with it, and corresponding teardown or
recovery procedures SHOULD be triggered for the e2e LSP. In case of
S-LSP teardown for maintenance purpose, the S-LSP ingress node MAY
treat this to be equivalent to administratively shutting down a TE
link along the e2e LSP path and take corresponding actions to notify
the ingress of this event. The actual signaling procedures to handle
this event is out of the scope of this document.

5.1.5. Teardown of e2e LSPs

e2e LSP teardown also follows standard procedures defined in
[RFC3209] and [RFC3473] either without or with the administrative
status. Note, however, that teardown procedures of e2e LSP and of
S-LSP are independent of each other. So it is possible that while
one LSP follows graceful teardown with administrative status, the
other LSP is torn down without administrative status (using
PathTear/ResvTear/PathErr with state removal).

When an e2e LSP teardown is initiated from the head end, and a
PathTear arrives at the GMPLS stitching node, the PathTear message
like the Path message MUST be IP routed to the LSP segment egress
node with the destination IP address of the Path message set to the
address of the S-LSP end node. Router Alert MUST be off and RSVP
Time to Live (TTL) check MUST be disabled on the receiving node.
PathTear will result in deletion of RSVP states corresponding to the
e2e LSP and freeing of label allocations and bandwidth reservations
on the S-LSP. The unreserved bandwidth on the S-LSP TE link SHOULD
be readjusted.

Similarly, a teardown of the e2e LSP may be initiated from the tail
end either using a ResvTear or a PathErr with state removal. The
egress of the S-LSP MUST propagate the ResvTear/PathErr upstream, and
MUST use IP addressing to target the ingress of the LSP segment.

Graceful LSP teardown using ADMIN_STATUS as described in [RFC3473] is
also applicable to stitched LSPs.

Ayyangar, et al. Standards Track [Page 11]

RFC 5150 LSP Stitching with GMPLS TE February 2008

If the S-LSP was statically provisioned, tearing down of an e2e LSP
MAY not result in tearing down of the S-LSP. If, however, the S-LSP
was dynamically set up due to the e2e LSP setup request, then,
depending on local policy, the S-LSP MAY be torn down if no e2e LSP
is utilizing the S-LSP. Although the S-LSP may be torn down while
the e2e LSP is being torn down, it is RECOMMENDED that a delay be
introduced in tearing down the S-LSP once the e2e LSP teardown is
complete, in order to reduce the simultaneous generation of RSVP
errors and teardown messages due to multiple events. The delay
interval may be set based on local implementation. The RECOMMENDED
interval is 30 seconds.

5.2. Summary of LSP Stitching Procedures

5.2.1. Example Topology

The following topology will be used for the purpose of examples
quoted in the following sections.

e2e LSP
+++++++++++++++++++++++++++++++++++> (LSP1-2)

LSP segment (S-LSP)
====================> (LSP-AB)
C --- E --- G
/|\ | / |\
/ | \ | / | \
R1 ---- A \ | \ | / | / B --- R2
\| \ |/ |/
D --- F --- H

PATH
====================> (LSP stitching desired)
RESV
<==================== (LSP segment stitching ready)

PATH (Upstream Label)
+++++++++++++++++++++
+++++++ ++++++>
<++++++ +++++++
+++++++++++++++++++++
RESV (Label)

5.2.2. LSP Segment Setup

Let us consider an S-LSP LSP-AB being set up between two nodes A and
B that are more than one hop away. Node A sends a Path message for
the LSP-AB with "LSP stitching desired" set in the Flags field of the

Ayyangar, et al. Standards Track [Page 12]

RFC 5150 LSP Stitching with GMPLS TE February 2008

LSP_ATTRIBUTES object. If the egress node B is ready to carry out
stitching procedures, then B will respond with "LSP segment stitching
ready" set in the Flags field of the RRO Attributes subobject, in the
RRO sent in the Resv for the S-LSP. Once A receives the Resv for
LSP-AB and sees this bit set in the RRO, it can then use LSP-AB for
stitching. Node A cannot use LSP-AB for stitching if the bit is
cleared in the RRO.

5.2.3. Setup of an e2e LSP

Let us consider an e2e LSP LSP1-2 starting one hop before A on R1 and
ending on node R2, as shown above. If the S-LSP has been advertised
as a TE link in the TE domain, and R1 and A are in the same domain,
then R1 may compute a path for LSP1-2 over the S-LSP LSP-AB and
identify the LSP-AB hop in the ERO. If not, R1 may compute hops
between A and B and A may use these ERO hops for S-LSP selection or
signaling a new S-LSP. If R1 and A are in different domains, then
LSP1-2 is an inter-domain LSP. In this case, S-LSP LSP-AB, similar
to any other basic TE link in the domain, will not be advertised
outside the domain. R1 would use either per-domain path computation
([RFC5152]) or PCE-based computation ([RFC4655]) for LSP1-2.

5.2.4. Stitching of an e2e LSP into an LSP Segment

When the Path message for the e2e LSP LSP1-2 arrives at node A, A
matches the switching type of LSP1-2 with the S-LSP LSP-AB. If the
switching types are not equal, then LSP-AB cannot be used to stitch
LSP1-2. Once the S-LSP LSP-AB to which LSP1-2 will be stitched has
been determined, the Path message for LSP1-2 is sent (via IP routing,
if needed) to node B with the IF_ID RSVP_HOP identifying the S-LSP
LSP-AB. When B receives this Path message for LSP1-2, if B is also
the egress for LSP1-2, and if this is a packet LSP requiring PHP,
then B will send a Resv refresh for LSP-AB with the NULL Label. In
this case, since B is not the egress, the Path message for LSP1-2 is
propagated to R2. The Resv for LSP1-2 from B is sent back to A with
a Label value chosen by B. B also sets up its data plane to swap the
Label sent to either G or H on the S-LSP with the Label received from
R2. Node A ignores the Label on receipt of the Resv message and then
propagates the Resv to R1. A also sets up its data plane to swap the
Label sent to R1 with the Label received on the S-LSP from C or D.
This stitches the e2e LSP LSP1-2 to an S-LSP LSP-AB between nodes A
and B. In the data plane, this yields a series of label swaps from
R1 to R2 along e2e LSP LSP1-2.

Ayyangar, et al. Standards Track [Page 13]

RFC 5150 LSP Stitching with GMPLS TE February 2008

6. Security Considerations

From a security point of view, the changes introduced in this
document model the changes introduced by [RFC4206]. That is, the
control interface over which RSVP messages are sent or received need
not be the same as the data interface that the message identifies for
switching traffic. But the capability for this function was
introduced in [RFC3473] to support the concept of out-of-fiber
control channels, so there is nothing new in this concept for
signaling or security.

The application of this facility means that the "sending interface"
or "receiving interface" may change as routing changes. So these
interfaces cannot be used to establish security associations between
neighbors, and security associations MUST be bound to the
communicating neighbors themselves.

[RFC2747] provides a solution to this issue: in Section 2.1, under
"Key Identifier", an IP address is a valid identifier for the sending
(and by analogy, receiving) interface. Since RSVP messages for a
given LSP are sent to an IP address that identifies the next/previous
hop for the LSP, one can replace all occurrences of 'sending
[receiving] interface' with 'receiver's [sender's] IP address'
(respectively). For example, in Section 4, third paragraph, instead
of:

"Each sender SHOULD have distinct security associations (and keys)
per secured sending interface (or LIH). ... At the sender,
security association selection is based on the interface through
which the message is sent."

it should read:

"Each sender SHOULD have distinct security associations (and keys)
per secured receiver's IP address. ... At the sender, security
association selection is based on the IP address to which the
message is sent."

Thus, the mechanisms of [RFC2747] can be used unchanged to establish
security associations between control plane neighbors.

This document allows the IP destination address of Path and PathTear
messages to be the IP address of a next hop node (receiver's address)
instead of the RSVP session destination address. This means that the
use of the IPsec Authentication Header (AH) (ruled out in [RFC2747]

Ayyangar, et al. Standards Track [Page 14]

RFC 5150 LSP Stitching with GMPLS TE February 2008

because RSVP messages were encapsulated in IP packets addressed to
the ultimate destination of the Path or PathTear messages) is now
perfectly applicable, and standard IPsec procedures can be used to
secure the message exchanges.

An analysis of GMPLS security issues can be found in [MPLS-SEC].

7. IANA Considerations

IANA has made the following codepoint allocations for this document.

7.1. Attribute Flags for LSP_ATTRIBUTES Object

The "RSVP TE Parameters" registry includes the "Attributes Flags"
sub-registry.

IANA has allocated the following new bit (5) defined for the
Attributes Flags TLV in the LSP_ATTRIBUTES object.

LSP stitching bit - Bit Number 5

This bit is only to be used in the Attributes Flags TLV on a Path
message.

The 'LSP stitching desired' bit has a corresponding 'LSP segment
stitching ready' bit (Bit Number 5) to be used in the RRO Attributes
subobject.

The following text has been includuded in the registry:

Bit | Name | Attribute | Path | RRO | Reference
No | | Flags Path | Flags Resv | |
----+----------------------+------------+------------+-----+----------
5 LSP stitching desired Yes No Yes [RFC5150]

7.2. New Error Codes

The "Resource ReSerVation Protocol (RSVP) Parameters" registry
includes the "Error Codes and Globally-Defined Error Value Sub-Codes"
sub-registry.

IANA has assigned a new error sub-code (30) under the RSVP error-code
"Routing Problem" (24).

This error code (30) is to be used only in an RSVP PathErr.

Ayyangar, et al. Standards Track [Page 15]

RFC 5150 LSP Stitching with GMPLS TE February 2008

The following text has been included in the registry:

24 Routing Problem [RFC3209]

30 = Stitching unsupported [RFC5150]

8. Acknowledgments

The authors would like to thank Dimitri Papadimitriou and Igor
Bryskin for their thorough review of the document and discussions
regarding the same.

9. References

9.1. Normative References

[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.

[RFC2747] Baker, F., Lindell, B., and M. Talwar, "RSVP
Cryptographic Authentication", RFC 2747, January 2000.

[RFC3209] Awduche, D., Berger, L., Gan, D., Li, T., Srinivasan,
V., and G. Swallow, "RSVP-TE: Extensions to RSVP for LSP
Tunnels", RFC 3209, December 2001.

[RFC3473] Berger, L., Ed., "Generalized Multi-Protocol Label
Switching (GMPLS) Signaling Resource ReserVation
Protocol-Traffic Engineering (RSVP-TE) Extensions", RFC
3473, January 2003.

[RFC4206] Kompella, K. and Y. Rekhter, "Label Switched Paths (LSP)
Hierarchy with Generalized Multi-Protocol Label
Switching (GMPLS) Traffic Engineering (TE)", RFC 4206,
October 2005.

[RFC4420] Farrel, A., Ed., Papadimitriou, D., Vasseur, J.-P., and
A. Ayyangar, "Encoding of Attributes for Multiprotocol
Label Switching (MPLS) Label Switched Path (LSP)
Establishment Using Resource ReserVation Protocol-
Traffic Engineering (RSVP-TE)", RFC 4420, February 2006.

Ayyangar, et al. Standards Track [Page 16]

RFC 5150 LSP Stitching with GMPLS TE February 2008

9.2. Informative References

[RFC3032] Rosen, E., Tappan, D., Fedorkow, G., Rekhter, Y.,
Farinacci, D., Li, T., and A. Conta, "MPLS Label Stack
Encoding", RFC 3032, January 2001.

[RFC3477] Kompella, K. and Y. Rekhter, "Signalling Unnumbered
Links in Resource ReSerVation Protocol - Traffic
Engineering (RSVP-TE)", RFC 3477, January 2003.

[RFC4201] Kompella, K., Rekhter, Y., and L. Berger, "Link Bundling
in MPLS Traffic Engineering (TE)", RFC 4201, October
2005.

[RFC4203] Kompella, K., Ed., and Y. Rekhter, Ed., "OSPF Extensions
in Support of Generalized Multi-Protocol Label Switching
(GMPLS)", RFC 4203, October 2005.

[RFC4205] Kompella, K., Ed., and Y. Rekhter, Ed., "Intermediate
System to Intermediate System (IS-IS) Extensions in
Support of Generalized Multi-Protocol Label Switching
(GMPLS)", RFC 4205, October 2005.

[RFC4655] Farrel, A., Vasseur, J.-P., and J. Ash, "A Path
Computation Element (PCE)-Based Architecture", RFC 4655,
August 2006.

[RFC4726] Farrel, A., Vasseur, J.-P., and A. Ayyangar, "A
Framework for Inter-Domain Multiprotocol Label Switching
Traffic Engineering", RFC 4726, November 2006.

[RFC4875] Aggarwal, R., Ed., Papadimitriou, D., Ed., and S.
Yasukawa, Ed., "Extensions to Resource Reservation
Protocol - Traffic Engineering (RSVP-TE) for Point-to-
Multipoint TE Label Switched Paths (LSPs)", RFC 4875,
May 2007.

[RFC5151] Farrel, A., Ed., Ayyangar, A., and JP. Vasseur, "Inter-
Domain MPLS and GMPLS Traffic Engineering -- Resource
Reservation Protocol-Traffic Engineering (RSVP-TE)
Extensions", RFC 5151, February 2008.

[RFC5152] Vasseur, JP., Ed., Ayyangar, A., Ed., and R. Zhang, "A
Per-Domain Path Computation Method for Establishing
Inter-Domain Traffic Engineering (TE) Label Switched
Paths (LSPs)", RFC 5152, February 2008.

Ayyangar, et al. Standards Track [Page 17]

RFC 5150 LSP Stitching with GMPLS TE February 2008

[MPLS-SEC] Fang, L., Ed., Behringer, M., Callon, R., Le Roux, J.
L., Zhang, R., Knight, P., Stein, Y., Bitar, N., and R.
Graveman., "Security Framework for MPLS and GMPLS
Networks", Work in Progress, July 2007.

Authors' Addresses

Arthi Ayyangar
Juniper Networks
1194 N. Mathilda Avenue
Sunnyvale, CA 94089
EMail: [email protected]

Kireeti Kompella
Juniper Networks
1194 N. Mathilda Avenue
Sunnyvale, CA 94089
EMail: [email protected]

JP Vasseur
Cisco Systems, Inc.
300 Beaver Brook Road
Boxborough, MA 01719
EMail: [email protected]

Adrian Farrel
Old Dog Consulting
EMail: [email protected]

Ayyangar, et al. Standards Track [Page 18]

RFC 5150 LSP Stitching with GMPLS TE February 2008

Full Copyright Statement

Copyright (C) The IETF Trust (2008).

This document is subject to the rights, licenses and restrictions
contained in BCP 78, and except as set forth therein, the authors
retain all their rights.

This document and the information contained herein are provided on an
"AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS
OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY, THE IETF TRUST AND
THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS
OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF
THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED
WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.

Intellectual Property

The IETF takes no position regarding the validity or scope of any
Intellectual Property Rights or other rights that might be claimed to
pertain to the implementation or use of the technology described in
this document or the extent to which any license under such rights
might or might not be available; nor does it represent that it has
made any independent effort to identify any such rights. Information
on the procedures with respect to rights in RFC documents can be
found in BCP 78 and BCP 79.

Copies of IPR disclosures made to the IETF Secretariat and any
assurances of licenses to be made available, or the result of an
attempt made to obtain a general license or permission for the use of
such proprietary rights by implementers or users of this
specification can be obtained from the IETF on-line IPR repository at
http://www.ietf.org/ipr.

The IETF invites any interested party to bring to its attention any
copyrights, patents or patent applications, or other proprietary
rights that may cover technology that may be required to implement
this standard. Please address the information to the IETF at
[email protected].

Ayyangar, et al. Standards Track [Page 19]