Layer 2 VPN Architectures: Understanding Any Transport over MPLS
This chapter covers the following topics:
Label Distribution Protocol (LDP)
To provide Layer 2 VPN services over an IP/Multiprotocol Label Switching (MPLS) net-work infrastructure, the Internet Engineering Task Force (IETF) developed a series of solution and protocol specifications for various Layer 2 VPN applications, including pseudowire emulation. Based on the pseudowire emulation specifications, Any Transport over MPLS (AToM) is implemented as part of the Cisco Unified VPN Suite Solution. The Cisco solution also includes alternative pseudowire emulation using Layer 2 Tunnel Protocol Version 3 (L2TPv3). Chapter 3, "Layer 2 VPN Architectures," outlines the benefits and implications of using each technology and highlights some important factors that help network planners and operators determine the appropriate technology.
This chapter starts with an overview of LDP used by pseudowire emulation over MPLS, followed by an explanation of the protocol specifications and operations of AToM. You learn the general properties of the pseudowire emulation over MPLS networks specified in IETF documents. Additional features that AToM supports are also highlighted in this chapter.
Introducing the Label Distribution Protocol
One of the fundamental tasks in the MPLS architecture is to exchange labels between label switch routers (LSR) and define the semantics of these labels. LSRs follow a set of proce-dures, known as label distribution protocol, to accomplish this task. A label distribution protocol can be an existing protocol with MPLS label extensions or a new protocol that is specifically designed for this purpose. Although the MPLS architecture allows different label distribution protocols, only LDP is used as the signaling protocol for AToM.
The next few sections review some fundamental LDP specifications and operations that are relevant to AToM:
LDP protocol components
Label distribution and management
LDP Protocol Components
To have a firm understanding of the protocol operations of LDP, you need to be familiar with the key terminology and protocol entities that are defined in LDP.
LDP peers are two LSRs that use LDP to exchange label information. An LSR might have more than one LDP peer, and it establishes an LDP session with each LDP peer. An LDP session is always bidirectional, which allows both LDP peers to exchange label information. However, using a bidirectional signaling session does not make the label-switched path (LSP) bidirection-al. As described in Chapter 3, an LSP is unidirectional, and a pseudowire consists of two LSPs of the opposite directions. Besides directly connected LSRs, LDP sessions can be established between non-directly connected LSRs, which are further explained in the later section titled "LDP Extended Discovery."
Label space specifies the label assignment. The two types of label space are as follows:
Per-interface label space—Assigns labels from an interface-specific pool of labels. This space typically uses interface resources for labels. For example, a label-controlled ATM interface uses virtual path identifiers (VPI) and virtual circuit identifiers (VCI) as labels.
Per-platform label space—Assigns labels from a platform-wide pool of labels and typically uses resources that are shared across the platform. Hop-by-hop best-effort IP/MPLS forwarding is an example of using the per-platform label space.
In Chapter 3, the AToM overview explains the use of label stacking. To recap, the label stack of AToM typically consists of two labels: tunnel label and pseudowire label. Tunnel labels can be from either per-interface label space or per-platform label space depending on whether the LSRs perform IP/MPLS forwarding in cell mode or frame mode. Pseudowire labels are always allocated from the general-purpose per-platform label space.
LDP uses User Datagram Protocol (UDP) and TCP to transport the protocol data unit (PDU) that carries LDP messages. Figure 6-1 illustrates the structure of an LDP packet. Each LDP PDU is an LDP header followed by one or more LDP messages. All LDP messages have a common LDP message header followed by one or more structured parameters that use a type, length, value (TLV) encoding scheme. The Value field of a TLV might consist of one or more sub-TLVs.
Figure 6-1 LDP Packet Structure
The LDP header consists of the following fields, as depicted in Figure 6-2:
Version—The Version field is 2 octets containing the version number of the protocol. The current LDP version is version 1.
PDU Length—The PDU Length field is a 2-octet integer specifying the total length of this PDU in octets, excluding the Version and PDU Length fields. The maximum PDU Length can be negotiated during LDP session initialization.
LDP Identifier—An LDP Identifier consists of 6 octets and identifies an LSR label space. The first 4 octets are a globally unique value that identifies the LSR. The globally unique value is usually the 32-bit router ID of the LSR. The last 2 octets identify a specific label space within the LSR. A zero value of the last 2 octets represents the platform-wide label space. When an LSR uses LDP to advertise more than one label space to another LSR, it creates a separate LDP session for each label space.
Figure 6-2 LDP Header Format
Four categories exist for LDP messages:
Discovery messages—Provide a mechanism in which LSRs indicate their presence in a network by sending Hello messages periodically. Discovery messages include the LDP Link Hello message and the LDP Targeted Hello message. You learn more about discovery messages in the next section "Discovery Mechanisms."
Session messages—Establish, maintain, and disconnect sessions between LDP peers. Session messages are LDP Initialization messages and Keepalive messages. You learn more about session messages in the section "Session Establishment" later in this chapter.
Advertisement messages—Create, update, and delete label mappings. All LDP Address messages and LDP Label messages belong to advertisement messages.
Notification messages—Provide advisory information and signal error information to LDP peers.
Except for discovery messages that use UDP as the underlying transport, LDP messages rely on TCP to ensure reliable and in-order delivery of messages. All LDP messages have the format that is depicted in Figure 6-3.
Figure 6-3 LDP Message Format
The following fields make up the LDP message format:
Unknown Message Bit (U-Bit)—The U-bit tells the receiver of the message what action to take if he does not understand the message. If the U-bit is set to 0, the receiver needs to respond to the originator of the message with a notification message. Otherwise, the receiver should silently ignore this unknown message.
Message Type—The Message Type field identifies the type of message.
Message Length—The Message Length field specifies the total number of octets of the Message ID, Mandatory Parameters, and Optional Parameters.
Message ID—The Message ID field is a 4-octet value that identifies individual messages.
Mandatory Parameters—The Mandatory Parameters field is a set of required para-meters with variable lengths that pertain to this message. Some messages do not have mandatory parameters.
Optional Parameters—The Optional Parameters field is a set of optional parameters that have variable lengths. Many messages do not have optional parameters.
Most information that is carried in an LDP message is encoded in TLVs. TLV provides a generic and extensible encoding scheme for existing and future applications that use LDP signaling. An LDP TLV consists of a 2-bit Flag field, a 14-bit Type field, and a 2-octet Length field, followed by a variable length Value field. Figure 6-4 shows the common TLV encoding scheme.
Figure 6-4 LDP TLV Encoding
Like the unknown message bit, the unknown TLV bit (U-bit) tells the receiver whether it should send a notification message to the originator if the receiver does not understand the TLV. If the U-bit is set to 0, the receiver must respond with a notification message and discard the entire message. Otherwise, the unknown TLV is silently ignored and the rest of the message is pro-cessed as if the unknown TLV does not exist.
The forward unknown TLV bit (F-bit) applies only when the U-bit is set to 1 and the TLV is unknown to the receiver. If the F-bit is set to 0, the unknown TLV is not forwarded. Otherwise, it is forwarded with the containing message.
LSRs use LDP discovery procedures to locate possible LDP peers. The basic discovery mechanism identifies directly connected LDP peers. The extended discovery mechanism identifies non-directly connected LDP peers. LSRs discover LDP peers by exchanging LDP Hello messages. As you learned in the previous section, two types of LDP Hello messages exist. LDP Link Hellos are used for LDP basic discovery, and LDP Targeted Hellos are used for LDP extended discovery. Figure 6-5 illustrates where LDP basic discovery and LDP extended discovery occur in an MPLS network.
Figure 6-5 LDP Basic and Extended Discovery
LDP Basic Discovery
With LDP basic discovery enabled on an interface, an LSR periodically sends LDP Link Hello messages out the interface. LDP Link Hellos are encapsulated in UDP packets and sent to the well-known LDP discovery port 646 with the destination address set to the multicast group address 188.8.131.52. This multicast address represents all routers on this subnet.
An LDP Link Hello message that an LSR sends carries the LDP identifier for the label space that the LSR intends to use for the interface and other information, such as Hello hold time. When the LSR receives an LDP Link Hello on an interface, it creates a Hello adjacency to keep track of a potential LDP peer reachable at the link level on the interface and learns the label space that the peer intends to use for the interface.
LDP Extended Discovery
For some MPLS applications such as AToM, exchanging label information between non-directly connected LSRs is necessary. Before establishing LDP sessions between non-directly connected LSRs, the LSRs engage in LDP extended discovery by periodically sending Targeted Hello messages to a specific address. LDP Targeted Hello messages are encapsulated in UDP packets and sent to the well-known LDP discovery port 646 with a specific unicast address.
An LDP Targeted Hello message that an LSR sends carries the LDP Identifier for the label space that the LSR intends to use and other information. When the receiving LSR receives an LDP Targeted Hello, it creates a Hello adjacency with a potential LDP peer reachable at the network level and learns the label space that the peer intends to use.
When an LSR sends LDP a Targeted Hello to a receiving LSR, the receiving LSR can either accept the Targeted Hello or ignore it. The receiving LSR accepts the Targeted Hello by creating a Hello adjacency with the originating LSR and periodically sending Targeted Hellos to it.
After two LSRs exchange LDP discovery Hello messages, they start the process of session establishment, which proceeds in two sequential phases:
Transport connection establishment
The objective of the transport connection establishment phase is to establish a reliable TCP connection between two LDP peers. If both LDP peers initiate an LDP TCP connection, it might result in two concurrent TCP connections. To avoid this situation, an LSR first determines whether it should play the active or passive role in session establishment by comparing its own transport address with the transport address it obtains through the exchange of LDP Hellos. If its address has a higher value, it assumes the active role. Otherwise, it is passive. When an LSR plays the active role, it initiates a TCP connection to the LDP peer on the well-known LDP TCP port 646.
After the LSR establishes the TCP connection, session establishment proceeds to the session initialization phase. In this phase, LDP peers exchange and negotiate session parameters such as the protocol version, label distribution methods, timer values, label ranges, and so on.
If an LSR plays the active role, it starts the negotiation of session parameters by sending an Initialization message to its LDP peer. The Initialization message carries both the LDP Ident-ifier for the label space of the active LSR and the LDP Identifier of the passive LSR. The receiver compares the LDP Identifier with the Hello adjacencies created during LDP discovery. If the receiver finds a match and the session parameters are acceptable, it replies with an Initialization message with its own session parameters and a Keepalive message to acknow-ledge the sender's parameters. When the sender receives an Initialization message with acceptable session parameters, it responds with a Keepalive message.
When both LDP peers exchange Initialization and Keepalive messages with each other, the session initialization phase is completed successfully and the LDP session is considered operational.
Label Distribution and Management
Label distribution and management consist of different control, retention, and advertisement modes. Even though it is possible to use an arbitrary permutation for an MPLS application, a certain combination of control, retention, and advertisement modes is usually more preferable or appropriate for a particular MPLS application.
The next few sections explain the following aspects in label distribution and management:
Label advertisement message
Label advertisement mode
Label distribution control mode
Label retention mode
The main focus of an MPLS application is the distribution and management of label bindings. Label bindings are always the centerpiece of information in LDP signaling.
LDP associates a Forwarding Equivalence Class (FEC) with each LSP that it creates. An FEC specifies which packets should be forwarded through the associated LSP. Each FEC is defined as a collection of one or more FEC elements. Each FEC element identifies a set of packets that are mapped to the corresponding LSP. For those who are familiar with IP routing, you can consider an FEC as a set of IP routes following a common forwarding path, and an FEC element as a specific IP route prefix.
A label binding is the association between an FEC and a label that represents a specific LSP. The association is created by placing an FEC TLV and a Label TLV in a label advertisement message. Figure 6-6 depicts the FEC TLV encoding.
Figure 6-6 FEC TLV Encoding
For FEC element 1 to FEC element n, the first octet in the FEC element indicates the FEC element type. The encoding scheme of the FEC element varies depending on the FEC element type, such as address prefix and host address. MPLS pseudowire emulation applications such as AToM use the Pseudowire ID FEC element.
Several different types of Label TLV encodings are available, including the following:
Generic Label TLV
ATM Label TLV
Frame Relay Label TLV
Generic Label TLV carries a label from the platform-wide label space and is the most common encoding among MPLS applications (see Figure 6-7).
Figure 6-7 Generic Label TLV Encoding
Generic Label TLV has a type of 0x0200. A label is a 20-bit label value in a 4-octet Label field.
LDP Advertisement Message
Label bindings are exchanged through LDP advertisement messages. The advertisement messages that are most relevant to pseudowire emulation over MPLS application are these
Label Mapping messages advertise label bindings to LDP peers. A Label Mapping message contains one FEC TLV and one Label TLV. Each FEC TLV might have one or more FEC elements depending on the type of application, and each Label TLV has one label.
When an LSR needs a label binding for a specific FEC but does not already have it, it can ex-plicitly request this label binding from its LDP peer by sending a Label Request message. A Label Request message contains the FEC for which a label is being requested. The receiving LSR then responds to a Label Request message with a Label Mapping message for the requested FEC if it has such a binding. Otherwise, it responds with a Notification message indicating why it cannot satisfy the request.
Whereas Label Mapping messages create the bindings between FECs and labels, Label With-draw messages break them. An LSR sends a Label Withdraw message to an LDP peer to signal that the peer should not continue to use specified label bindings that the LSR previously adver-tised. A Label Withdraw message contains the FEC for which the label binding is being withdrawn and optionally the originally advertised label. If no Label TLV is included in a Label Withdraw message, all labels that are associated with the FEC are to be withdrawn. Otherwise, only the label that is specified in the Label TLV is to be withdrawn.
An LSR that receives a Label Withdraw message must acknowledge it with a Label Release message. The LSR also uses Label Release messages to indicate that it no longer needs specific label bindings previously requested of or advertised by its LDP peer. A Label Release message contains the FEC for which the label binding is being released and optionally the originally advertised label. If no Label TLV is included in a Label Release message, all labels that are associated with the FEC are to be released. Otherwise, only the label that is specified in the Label TLV is to be released.
Label Advertisement Mode
The MPLS architecture specifies two label advertisement modes. If an LSR explicitly requests a label binding for a particular FEC from the next-hop LSR of this FEC, it uses downstream on- demand label advertisement mode. If an LSR advertises label bindings to its LDP peers that have not explicitly requested them, it uses downstream unsolicited advertisement mode.
Choosing which label advertisement mode to use depends on the characteristics of a particular MPLS implementation and application. Between each pair of LDP peers, they must have the same label advertisement mode.
Label Distribution Control Mode
Label distribution control determines how LSPs are established initially, and it has two modes: independent and ordered label distribution control.
With independent label distribution control, each LSR advertises label bindings to its peers at any time. It does not wait for the downstream or next-hop LSR to advertise the label binding for the FEC that is being distributed in the upstream direction. A consequence of using inde-pendent mode is that an upstream label can be advertised before a downstream label is received.
When an LSR is using ordered label distribution control, it cannot advertise a label binding for an FEC unless it has a label binding for the FEC from the downstream or next-hop LSR. It has to wait for the downstream LSR to advertise the label binding for the FEC that is being distrib-uted in the upstream direction. As a result, ordered control makes the label distribution of a given LSP occur sequentially from the last hop of the LSP toward the first hop of the LSP.
Label Retention Mode
When an LSR receives a label binding for an FEC from a peer that is not the next hop for the FEC, it has the option to either store or discard the label binding based on the label retention mode in use.
Conservative label retention keeps only the label bindings that will be used to forward packets. The main advantage is that only the labels that are required for data forwarding are allocated and maintained. Because downstream on-demand advertisement mode is mainly employed when the label space is limited, it is normally used with the conservation label retention mode.
With liberal label retention, an LSR keeps every label binding it receives from its LDP peers regardless of whether the peers are the next-hop LSRs for the advertised label binding. The main advantage is that an LSP can be updated quickly when the label forwarding information is changed. Liberal label retention is mainly used where the label space is considered an inexpensive resource. When it is used with downstream unsolicited advertisement mode, liberal label retention reduces the total number of label advertisement messages required to set up LSPs. If an LSR is using conservative retention mode in this scenario, it has to send Label Request messages to the peer for the label bindings that it has discarded during the initial label advertisement if that peer becomes the next-hop LSR for the FECs that are being requested.
LDP uses TCP for transport of LDP messages. The LDP specification does not provide its own security measures but leverages the existing TCP MD5 authentication mechanism defined in RFC 1321 and also used by BGP in RFC 2385. MD5 authentication uses a message digest to validate the authenticity and integrity of an LDP message.
A message digest is calculated with the MD5 hash algorithm that uses a shared secret key and the contents of the TCP segment. Unlike clear-text passwords, message digest prevents the shared secret from being snooped. In addition to protecting against spoofing, MD5 authent-ication provides good protection against denial of service (DoS) and man-in-the-middle attacks.