MPLS Forwarding Engine Design in Linux
========================================
-Pranjal Kumar Dutta
Here are a few notes that I have scribbled when I started designing the
forwarding engine components for MPLS Ingress LSR, Transit or Core LSR and
Egress LSR. Don't assume that this document is "everything that includes
all". I would utter the usual GPL developer quote "read the source code".
These notes are written in 2002 and after that Linux kernel has gone through
lots of changes till 2.6.15.6. Thse notes are mostly platform independent
design, that is not dependent on any Linux features.
Linux User Space
+----------+ +------------+ +----------+
| RSVP-TE |LSP_ADD/LSP_DEL | zMPLS | LSP_ADD/DEL |LDP/CR-LDP|
| |--------------->| |<-------------| |
+----------+ +------------+ +----------+
-----------------------------------|------------------------------------
Linux Kernel Space |
|MPLS Forwarding Services APIs
+----------------+
| |
| NETLINK_MPLS |
| |
+----------------+
|
|
MPLS Forwarding Functional APIs
|
+---------------------------------------------------+-------------------------+
| | | | | | | |
mpls_fec.c mpls_ilm.c mpls_nhlfe.c mpls_input.c mpls_output.c mpls_forward.c |
mpls_lsp.c
MicroACE Blocks
Currently I am working on improving the MPLS forwarding for linux2.6.15.6
kernel.
The MPLS Fowarding Engine in Linux Kernel is comprised of following three main
modules.
1. MPLS Forwarding Services APIs
==============================
This module is extension of Linux NETLINK. This is the common interface for
LDP/CR-LDP and RSVP-TE. It provides the message based interface for MPLS
Forwarding Components. The scope of this API is restricted to LSPs
associated with Ethernet (Generic), ATM, FR and POS (Generic interfaces).
2. MPLS Forwarding Functional APIs
====================================
This component provides the set of APIs for configuring individual MPLS
forwarding components. NETLINK module talks to each forwarding engine
component through these set of APIs.Each forwarding engine module exports
its own API(s) to this set.
3. MPLS Forwarding Engine Components
====================================
The MPLS forwarding Engine is divided into following subcomponents.
3.1 MPLS FEC mpls_fec.c
=====================
This module is the LER component and classifies the data being forwarded
into FECs. This module provides L2/L3 FEC (policy based as wll as
non-policy based.)
For L3 there are two FEC Tables.
1) Best Effort LSPs (Programmed by LDP component of the control plane).
2) TE-LSPs ( Programmed by RSVP Component of the control plane).
I am yet to sort out how to deal with LSPs installed by CR-LDP.
Design Considerations :
3.1.1. FTN Entries are organized in a system specific FEC table. An FTN
Entry is uniquely identified based on FEC Parameters. A FTN Entry can
be IPV4/IPV4 Prefix, IPV4/IPV6 Host L2FEC (PWID Element) etc.
The FEC Table is populated by Control Plane. The FEC Table points to
FEC Type specific sub-Tables. So each sub-table will have FTN entry
of a specific type.
3.1.2. FTN Entries can be stored in traditional linux IP/L2 forwarders. For
example, for L3 LSP, the fib_entry can have pointer to FTN Entry.
The same entry can be populated to the ip flow caches in wheh a flow
fast path is programmed.
3.2 MPLS ILM mpls_ilm.c
====================
This module maintains abstractions for ILM.
Design Considerations :
3.2.1 ILM Entries are maintained in a system wide table (Not interface
specific).
3.2.2 An ILM entry is uniquely identified based on incoming label and
incoming interface identifier.
3.2.3
3.3 MPLS NHLFE Module mpls_nhlfe.c
==============================
This module maintains abstractions for NHLFE.
Design Considerations:
3.3.1 NHLFE are kept in a system wide table.
3.3.2.An NHLFE Entry is uniquely identfied based on outgoing interface
identifier, a next hop IP address and outgoing label Stack. The
outgoing label will be the top label in the label stack information
associated with the NHLFE.
3.3.3 An NHLFE/NHLFE Set can be created first in the NHLFE/NHLFE Set Table
database. The creation will result in a unique NHLFE/NHLFRE Set
Handle. The NHLFE handle value will be passed along with the LSP
creation as part of the NHLFE Set information.
3.3.4 An NHLFE/NHLFE Set is created along with the LSP, when the LSP is
created in kernel. An NHLFE Set will be created even if i is just
one NHLFE.
3.3.5 An NHLFE/NHLFE Set can be shared by multiple LSPs, if the system can
support label merging. This will arise when a label space MAY be
partitioned into multiple non-overlapping signaling protocol
partitions.
3.4 MPLS LSP Table/Database
============================
This module maintains the LSP database.
Design considerations:
This module contains the LSPs programmed by NETLINK into the Kernel.
3.4.1 For an LSP there can be multiple NHLFEs.For example, P2MP LSP or PWs.
A policy may be defined to select between the NHLFEs. The policy
information and the associated NHLFEs are contained within a NHLFE
set.
3.5 MPLS Input Module mpls_input.c
==============================
3.6 MPLS output module mpls_output.c
=================================