Advertising MPLS labels in IS-ISJuniper Networks, Inc.1194 N. Mathilda Ave.SunnyvaleCA94089UShannes@juniper.netLevel 3 Communications, Inc.1025 Eldorado BlvdBroomfieldCO80021USshane@level3.netAmazonSeattleWNUStscholl@amazon.comVerizon1201 E Arapaho Rd.RichardsonTX75081USluay.jalil@verizon.com
Routing
IS-IS for IP InternetsMPLSIGPIS-ISLabel advertisementHistorically MPLS label distribution was driven by
protocols like LDP, RSVP and LBGP. All of those protocols are session
oriented. In order to obtain a label binding for a given destination FEC from
a given router one needs first to establish an LDP/RSVP/LBGP session
with that router.
Advertising MPLS labels in IGPs advertisement
describes several use cases where utilizing the flooding machinery
of link-state protocols for MPLS label distribution allows
to obtain the binding without requiring to establish an LDP/RSVP/LBGP
session with that router.
This document describes the protocol extension to distribute
MPLS labels by the IS-IS protocol.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.MPLS label allocations are predominantly distributed by using
the LDP , RSVP or
labeled BGP protocol. All of those protocols
have in common that they are session oriented, which means that in
order to obtain label binding for a given destination FEC from
a given router one needs first to establish a direct
control plane (LDP/RSVP/LBGP) session with that router.
There are a couple of
use cases
where the consumer of a MPLS label binding may not
be adjacent to the router that performs the binding. Bringing up an
explicit session using the existing label distribution protocols between
the non-adjacent router that bind the label and the router that
acts as a consumer of this binding is the existing remedy for
this dilemma. This document describes a single IS-IS protocol extension which allows
routers to advertise MPLS label bindings within and beyond an IGP domain, and
controlling inter-area distribution.
One possible way of distributing MPLS labels using IS-IS has been described in
Segment
Routing. The authors propose to re-use the IS-Reach TLVs (22,
23, 222) and Extended IP Prefix TLVs (135, 236) for carrying the label
information. While retrofitting existing protocol machinery for new
purposes is generally a good thing, Segment
Routing falls short of addressing some use-cases defined in
.
The dominant issue around re-using IS-Reach TLVs and the extended IP Prefix TLVs is that
both family of TLVs have existing protocol semantics, which might not be well suitable
to advertising MPLS label switched paths in a generic fashion. These are specifically:Bi-directionality semanticsIP path semanticsLack of 'path' notion
'Bi-directionality semantics', affects the complexity around advertisement of unidirectional LSPs.
Label advertisement of per-link labels or 'Adj-SIDs'
is done using IS-reach TLVs. Usually implementations need to have
an adjacency in 'Up' state prior to advertising this adjacency
as IS-reach TLV in its Link State PDUs (LSPs).
In order to advertise e.g. one-hop MPLS LSP in a given link an implementation first needs to have
an adjacency, which only transitions to 'Up' state after passing the 3-way
check. This implies bi-directionality. If an implementation wants to advertise
per-link LSPs to e.g. outside the IGP domain then it would need
to fake-up an adjacency. Changing existing IGP Adjacency code to support
such cases defeats the purpose of re-using existing functionality as there is
not much common functionality to be shared.
LSPs pointing to a Node are advertised as 'Node-SIDs'
using the family of extended IP Reach TLVs. That means that in order
to advertise a MPLS LSP, one is inheriting the semantics of advertising an
IP path. Consider router A has got existing MPLS LSPs to its entire one-hop
neighborhood and is re-advertising those MPLS LSPs using IP reachability semantics.
Now we have two exact matching IP advertisements. One from the owning router
(router B) which advertises its stable transport loopback address and another one
from router A re-advertising a MPLS LSP path to router B. Existing routing software
may get confused now as the 'stable transport' address shows up from multiple places
in the network and more worse the IP forwarding path for control-plane protocols may get
mingled with the MPLS data plane.
Both IS-Reach TLVs and IP Prefix Reachability TLVs have a
limited semantics describing MPLS label-switched paths in the sense of
a 'path'. Both encoding formats allow to specify a pointer to some
specific router, but not to describe a MPLS label switched path
containing all of its path segments. allows to define
'Forwarding Adjacencies' as per . The way to
describe a path of a given forwarding adjacency is to carry a list of
"Segment IDs". That implies that nodes which do not yet participate in
'Segment routing' or are outside of a 'Segment routing' domain can not
be expressed using those path semantics.
A protocol for advertising MPLS label switched paths, should
be generic enough to express paths sourced by existing MPLS LSPs,
such that ingress routers can flexibly combine them according to
application needs.
IGP advertisement of MPLS label switched paths requires a new set of
protocol semantics (path paradigm), which hardly can be expressed using
the existing IS-IS protocol. This document describes IS-IS protocol extensions
which allows generic advertisement of MPLS label switched paths in IS-IS.
The Protocol extensions described in this document are equally applicable to
IPv4 and IPv6 carried over MPLS. Furthermore the proposed use of distributing
MPLS Labels using IGP prototocols adheres to the architectural principles laid
out in .
The MPLS Label TLV may be originated by any Traffic Engineering
capable router in an IS-IS domain. A router may advertise MPLS labels along with so called 'ERO'
path segments describing the label switched path. Since ERO style path notation allows to
express pointers to link and node IP addresses any label switched path, sourced by any protocol,
can be described.
Due to the limited size of subTLV space (See section 4.5 for details), The MPLS Label TLV has
cumulative rather than canceling semantics. If a router originates
more than one MPLS Label TLV with the same Label value, then the
subTLVs of the second, third, etc. TLV are accumulated. Since some
subTLVs represent an ordered set (e.g. ERO subTLVs) allocation and
ordering of TLV space inside particular IS-IS LSP fragment is
significant.
The MPLS Label TLV has type 149 and has the following format:4 bits of flags, consisting of:
1 bit of up/down information (U bit)3 bits are reserved for future use20 bits of MPLS label information0-252 octets of sub-TLVs, where each sub-TLV consists of a
sequence of:
1 octet of sub-TLV type1 octet of length of the value field of the sub-TLV0-250 octets of valueFlags
Up/Down Bit: A router may flood MPLS label information across level boundaries.
In order to prevent flooding loops, a router will Set the Up/Down (U-Bit)
when propagating from Level 2 down to Level 1. This is done as per the
procedures for IP Prefixes lined out in .
An originating router MAY want to attach one or more
subTLVs to the MPLS label TLV. SubTLVs presence is inferred from the
length of the MPLS Label TLV. If the MPLS Label TLV Length field is >
3 octets then one or more subTLVs may be present.
The IPv4 ERO subTLV (Type 1) describes a path segment using IPv4 Prefix style of encoding.
Its appearance and semantics have been borrowed from
Section 4.3.3.2.
The 'Prefix Length' field contains the length of the prefix in bits.
Only the most significant octets of the prefix are encoded. I.e. 1
octet for prefix length 1 up to 8, 2 octets for prefix length 9 to
16, 3 octets for prefix length 17 up to 24 and 4 octets for prefix
length 25 up to 32, etc.
The 'L' bit in the subTLV is a one-bit attribute. If the L bit is
set, then the value of the attribute is 'loose.' Otherwise, the
value of the attribute is 'strict.'
The IPv6 ERO subTLV (Type 2) describes a path segment using IPv6 Prefix style of encoding.
Its appearance and semantics have been borrowed from
Section 4.3.3.3.
The 'Prefix Length' field contains the length of the prefix in bits.
Only the most significant octets of the prefix are encoded. I.e. 1
octet for prefix length 1 up to 8, 2 octets for prefix length 9 to
16, 3 octets for prefix length 17 up to 24 and 4 octets for prefix
length 25 up to 32, ...., 16 octets for prefix length 113 up to 128.
The 'L' bit in the subTLV is a one-bit attribute. If the L bit is
set, then the value of the attribute is 'loose.' Otherwise, the
value of the attribute is 'strict.'
The IPv4 Bypass ERO subTLV (Type 3) describes a Bypass LSP path segment using IPv4 Prefix style of encoding.
Its appearance and semantics have been borrowed from
Section 4.3.3.2.
The 'Prefix Length' field contains the length of the prefix in bits.
Only the most significant octets of the prefix are encoded, i.e. 1
octet for prefix length 1 up to 8, 2 octets for prefix length 9 to
16, 3 octets for prefix length 17 up to 24 and 4 octets for prefix
length 25 up to 32, etc.
The 'L' bit in the subTLV is a one-bit attribute. If the L bit is
set, then the value of the attribute is 'loose.' Otherwise, the
value of the attribute is 'strict.'
The IPv6 ERO subTLV (Type 4) describes a Bypass LSP path segment using IPv6 Prefix style of encoding.
Its appearance and semantics have been borrowed from
Section 4.3.3.3.
The 'Prefix Length' field contains the length of the prefix in bits.
Only the most significant octets of the prefix are encoded, i.e. 1
octet for prefix length 1 up to 8, 2 octets for prefix length 9 to
16, 3 octets for prefix length 17 up to 24 and 4 octets for prefix
length 25 up to 32, ...., 16 octets for prefix length 113 up to 128.
The 'L' bit in the subTLV is a one-bit attribute. If the L bit is
set, then the value of the attribute is 'loose.' Otherwise, the
value of the attribute is 'strict.'
All 'Prefix ERO' and 'Prefix Bypass ERO' information represents an ordered set
which describes the segments of a label-switched path. The last Prefix
ERO subTLV describes the segment closest to the egress point of the
LSP. Contrary the first Prefix ERO subTLV describes the first segment
of a label switched path. If a router extends or stitches a label
switched path it MUST prepend the new segments path information to the
Prefix ERO list.
The 'All Router Block' subTLV (Type 6) denominates the label block size of
an MPLS Label advertisement and its semantics to connect to all routers in
a given IS-IS domain using a local assigned
label range. Note that the actual mapping of a router within the label range
is done using the subTLVs described in
and . Since generation of
an 'All Router ID IPv4 Map' or 'All Router ID IPv6 Map' subTLV
is a local policy decision, it might be the case
that connectivity is provided not to 'All' but rather a subset of 'All'
routers. Keeping policy decisions aside, for simplicity reasons,
assume that All Routers in a domain do generate either the
'All Router ID IPv4 Map' or
'All Router ID IPv6 Map' subTLVs and therefore all routers
desire construction of a Label switched path from every source router
in the network.
The basic concept of using label blocks to provide connectivity
to a set of routers has been borrowed from
which allows to advertise labels from multiple end-points using a single control-plane
message. The difference to is that rather than advertising
where a particular packet came from (=source semantics), destination semantics
(where a particular packet will be going to) is advertised.
Along with each label block a router advertises one for more 'IDs'.
The 'ID' must be unique within a given domain.
The 'ID' serves as ordinal to determine the actual label value
inside the set of all advertised label ranges of a given router.
A receiving router uses the ordinal to determine the actual label value
in order to construct forwarding state to
a particular destination router.
The 'ID' is separately advertised using the subTLVs described in
and
.
The ability to advertise more than one label block eases operational
procedures for increasing the number of supported routers within a domain.
For example consider a given domain has got support for <M> routers and runs
out of ID space. It simply advertises one more label block to cover additional
ordinals outside the range of the first label block.
An example of this is described in more detail in
The 'Block Size' value contains the size of the label advertisement.
The 'value determines the amount of reachable router endpoints
within a given domain.
It MUST contain a value greater or equal than two.
Note that the label base is inferred from the Label Value in the
carrying MPLS Label TLV.
For example if a router wants to advertise a label range of 5000-5099 then
it would need to generate a MPLS Label TLV with a Label value of 5000
and a Block Size of 100.
The 'Algo' value denominates the path computation algorithm in order
to calculate the forwarding topology. The basic SPF algorithm
has an assigned 'Algo' code point of zero. The purpose of the
'Algo' field is to extend the notion of Label Block Signaling
to arbitrary algorithms like for example 'MRT'
(.
Advertised Label Blocks with an unknown, unsupported or non-configured
algorithm MUST be silently ignored.
The 'Reserved' bits are for future use. They should be zero on
transmission and ignored on receipt.
The 'Topology-ID' field contains the Multi Topology ID
() for which the advertised Label Block
does apply. The basic IPv4 unicast Topology has an assigned
'Topology-ID' code point of zero. The basic IPv6 unicast Topology has
an assigned 'Topology=ID' code point of 2.
Advertised Label Blocks with an unknown, unsupported or non-configured
Topology-ID MUST be silently ignored.
A MPLS Label TLV containing the 'All Router Block' subTLV MUST only contain
the 'All Router IPv4 Map' subTLV
()
or the 'All Router IPv6 Map' subTLV
().
The 'All Router ID IPv4 Map' TLV (Type 7) maps an 'ID' to a given
stable transport IPv4 address. Its purpose is to associate
a given transport IPv4 IP address to the ordinal inside a label range
as described in .
A router MAY advertise more than one 'ID' to 'IPv4 address'
mapping pair, in case it has more than one stable transport
IPv4 address.
The 'IPv4 address' contains stable IPv4 transport address of a given router.
The 'ID' contains the ordinal value of an advertising router inside the
set of all advertised label blocks of a given router.
The 'All Router ID IPv6 Map' TLV (Type 8) maps an 'ID' to a given
stable transport IPv6 address. Its purpose is to associate
a given transport IPv6 IP address to the ordinal inside a label range
as described in .
A router MAY advertise more than one 'ID' to 'IPv6 address'
mapping pair, in case it has more than one stable transport
IPv6 address.
The 'IPv6 address' contains the stable IPv6 transport address
of a given router.
The 'ID' contains the ordinal value of an advertising router inside the
set of all advertised label blocks of a given router.
The following topology and IP addresses shall be used throughout
the Label advertisement examples.
R1: 192.168.1.1R2: 192.168.1.2R3: 192.168.1.3R4: 192.168.1.4R5: 192.168.1.5R6: 192.168.1.6R7: 192.168.1.7R1 to R2 link: 10.0.0.1, 10.0.0.2R1 to R4 link: 10.0.0.3, 10.0.0.4R2 to R3 link #1: 10.0.0.5, 10.0.0.6R2 to R3 link #2: 10.0.0.7, 10.0.0.8R2 to R5 link: 10.0.0.9, 10.0.0.10R3 to R6 link: 10.0.0.13, 10.0.0.14R3 to R7 link: 10.0.0.15, 10.0.0.16R4 to R5 link: 10.0.0.17, 10.0.0.18R5 to R6 link: 10.0.0.11, 10.0.0.12R6 to R7 link: 10.0.0.19, 10.0.0.20
The IGP link metrics are displayed in the middle of the link.
All of them are assumed to be bi-directional.
If R1 would advertise a label <N> bound to a one-hop LSP
from R1 to R2 it would encode as follows:TLV 149: MPLS label <N>, Flags {}:
IPv4 Prefix ERO subTLV: 192.168.1.2/32, StrictIf R2 would advertise a label <N> bound to a one-hop LSP
from R2 to R3, using the link #2 it would encode as followsTLV 149: MPLS label <N>, Flags {}:
IPv4 Prefix ERO subTLV: 10.0.0.8/32, StrictR2 may advertise a one-hop LSP from R2 to R3, along with a Link
Protection Bypass for the directly adjacent links between those
two nodes. The Link Protection Bypass would use the path: {R2,
R5, R6, R3}. R2 would encode both the primary LSP and Link
Protection Bypass LSP as follows:TLV 149: MPLS label <N>, Flags {}:
IPv4 Prefix ERO subTLV: 192.168.1.3/32, StrictIPv4 Prefix Bypass ERO subTLV: 192.168.1.5/32, StrictIPv4 Prefix Bypass ERO subTLV: 192.168.1.6/32, StrictIPv4 Prefix Bypass ERO subTLV: 192.168.1.3/32, StrictConsider a RSVP LSP name "R2-to-R6" traversing (R2 to R3 using link #1, R6):If R2 would advertise a label <N> bound to the RSVP LSP named 'R2-to-R6',
it would encode as followsTLV 149: MPLS label <N>, Flags {}:
IPv4 Prefix ERO subTLV: 10.0.0.6/32, StrictIPv4 Prefix ERO subTLV: 192.168.1.6/32, StrictConsider R2 that creates a LDP label binding for FEC 172.16.0.0/12
using label <N>.If R2 would re-advertise this binding in IS-IS it would encode as followsTLV 149: MPLS label <N>, Flags {}:
IPv4 Prefix ERO subTLV: 172.16.0.0/12, LooseConsider two R2->R6 paths: {R2, R3, R6} and {R2, R5, R6}Consider two R5->R3 paths: {R5, R2, R3} and {R5, R6, R3}R2 encodes its two paths to R6 as follows:TLV 149: MPLS label <N1>, Flags {}:
IPv4 Prefix ERO subTLV: 192.168.1.3, StrictIPv4 Prefix ERO subTLV: 192.168.1.6, StrictTLV 149: MPLS label <N2>, Flags {}:
IPv4 Prefix ERO subTLV: 192.168.1.5, StrictIPv4 Prefix ERO subTLV: 192.168.1.6, StrictR5 encodes its two paths to R3 as follows:TLV 149: MPLS label <N1>, Flags {}:
IPv4 Prefix ERO subTLV: 192.168.1.2, StrictIPv4 Prefix ERO subTLV: 192.168.1.3, StrictTLV 149: MPLS label <N2>, Flags {}:
IPv4 Prefix ERO subTLV: 192.168.1.6, StrictIPv4 Prefix ERO subTLV: 192.168.1.3, Strict
A receiving L1 router does see now all 4 paths and may decide
to load-balance across all or a subset of them.
All routers within a given area MUST advertise their Label Blocks
along with an 'ID'.
If R2 would advertise a label block <N1> with a size of 10,
declaring SPT label forwarding support to all routers within a given domain,
it would encode as follows:TLV 149: MPLS Label <N1>, Flags {}:
All Router Block subTLV: Block Size 10 All Router ID IPv4 Map subTLV: ID 2, 192.168.1.2If R3 would advertise a label block <N2> with a size of 10,
declaring SPT label forwarding support to all routers within a given domain,
it would encode as follows:TLV 149, MPLS Label <N2>, Flags {}:
All Router Block subTLV: Block Size 10 All Router ID IPv4 Map subTLV: ID 3, 192.168.1.3If R5 would advertise a label block <N3> with a size of 10,
declaring SPT label forwarding support to all routers within a given domain,
it would encode as follows:TLV 149, MPLS Label <N3>, Flags {}:
All Router Block subTLV: Block Size 10 All Router ID IPv4 Map subTLV: ID 5, 192.168.1.5If R6 would advertise a label block <N4> with a size of 10,
declaring SPT label forwarding support to all routers within a given domain,
it would encode as follows:TLV 149, MPLS Label <N4>, Flags {}:
All Router Block subTLV: Block Size 10 All Router ID IPv4 Map subTLV: ID 6, 192.168.1.6
Consider now R2 constructing a SPT label for R6. R2s SPT to R6 is
{R2, IP4, R3, R6}. R2 first determines if its downstream router
(R3) has advertised a label-block. Since R3 has advertised a label block
'N2' and it has received R6 'ID' of 6 it will be picking the 6th label
value inside the advertised range of its downstream neighbor.
Specifically R2 MUST be program a
MPLS SWAP for its own label range Label(N1+6) to Label(N2+6),
NH 10.0.0.4 into its MPLS transit RIB.
Furthermore R2 MAY program a MPLS PUSH operation for IP 192.168.1.6 to
Label (N2+6), NH 10.0.0.04 into its IPv4 tunnel RIB.
Next walk down to R3, which is
the next router on the SPT tree towards R6.
R3s SPT to R6 is {R3, R6}.
R3 determines if its downstream router
(R6) has advertised a label-block. Since R6 has advertised a label block
'N4' and it has received R6 'ID' of 6 it will be picking the 6th label
value inside the advertised range of its downstream neighbor.
Since R3 is the penultimate router to R6 it MUST program a
MPLS POP for its own label range Label(N2+6) NH 10.0.0.14 into
its MPLS transit RIB.
Furthermore R3 MAY program a MPLS NOP for IP 192.168.1.6,
NH 10.0.0.14 into its IPv4 tunnel RIB.
Propagation of a MPLS LSP across a level boundary is a local policy decision.
If local policy dictates that a given L1L2 router needs to
re-advertise a MPLS LSPs from one Level to another then it MUST
allocate a new label and program its label forwarding table to connect
the new label to the path in the respective other level. Depending on
how to reach the re-advertised LSP, this is typically done using
a MPLS 'SWAP' or 'SWAP/PUSH' data plane operation.
If local policy dictates that a given L1L2 router
re-advertises a MPLS LSPs into another Level then it
MUST prepend its "Traffic-Engineering-ID" as a loose hop in the
Prefix ERO subTLV list. If the LSP is propagated from a higher Level
to a lower Level then the 'Down' bit MUST be set.
If local policy dictates that a given L1L2 router advertises
its 'All Router Block' into another Level, then it also
MUST re-advertise all known 'ID' ordinals (again gated by policy)
to the respective other Level. Without knowledge of all 'ID's
in the network no router is able to construct SPT label switched paths.
If a Label Block and its ID mappings are propagated from a
higher Level to a lower Level then the 'Down' bit MUST be set.
Many thanks to Yakov Rekhter for his useful comments.
This documents request allocation for the following TLVs and subTLVs.
PDUTLVsubTLVTypesubType#OccurenceLSPMPLS Label149>=0IPv4 Prefix ERO1>=0IPv6 Prefix ERO2>=0IPv4 Prefix Bypass ERO3>=0IPv6 Prefix Bypass ERO4>=0All Router Block6>=0All Router ID IPv4 Map7>=0All Router ID IPv6 Map8>=0
The MPLS Label TLV requires a new sub-registry.
Type value 149 has been assigned, with a starting sub-TLV
value of 1, range from 1-127, and managed by Expert Review.
This document does not introduce any change in terms of IS-IS
security. It simply proposes to flood MPLS label information via the IGP.
All existing procedures to ensure message integrity do apply here.