4. The POST-REGISTRATION Handoff Method
The POST-REGISTRATION handoff method uses bidirectional edge tunnels (BETs) or unidirectional tunnels to perform low-latency change in the L2 point of attachment for the MN without requiring any involvement by the MN. Figure 5 illustrates the basic POST-REGISTRATION handoff. +------+ | CN | +------+ | ... | +------+ BET +------+ | aFA |==========| nFA | +------+ +------+ | wireless link | movement +------+ ---------> | MN | +------+ Figure 5 - POST-REGISTRATION Concept Following a successful Mobile IPv4 Registration between MN and oFA, the oFA becomes the mobility anchor point for the MN, called the anchor FA (aFA). When the MN moves from oFA to nFA, rather than performing signaling over the wireless link to register with the nFA, the MN can defer the L3 handoff and continue to use its aFA (i.e., oFA in this case). If the MN moves to a third FA before registering with the nFA, in certain cases described later, the third FA signals aFA to move the wireless link end of the BET from nFA to it. The network end of the BET remains anchored at aFA until the MN performs the Mobile IPv4 Registration. Messages between oFA/aFA and nFA MUST be authenticated. The minimal requirement is that all FAs involved in low-latency handoffs MUST support manual pre-configuration of security associations with other neighboring FAs, involving shared keys and the default algorithms of [1]. POST-REGISTRATION FAs MUST implement the inter-FA authentication extension (FA-FA authentication extension) specified in [11] and MAY additionally use other security mechanisms.
4.1. Two-Party Handoff
Two-party handoff occurs when the MN moves from oFA to nFA. Normally, this movement would result in a new Mobile IPv4 Registration at nFA. However, in POST-REGISTRATION, the MN and nFA MAY delay this but maintain connectivity using the BET (or alternatively unidirectional tunnel) between oFA and nFA. The protocol is shown in Figure 6. 1a) L2-ST ~~~~> +------+ 2) HRqst +------+ <~~~ 1b) L2-TT | oFA |<-------->| nFA | 4a) L2-LD~> +------+ 3) HRply +------+ <~~~ 4b) L2-LU ^ ^ old L2 | | new L2 +-------+ +-----+ | | | | V V +------+ movement 4c) L2-LU ---> | MN | ---------> +------+ Figure 6 - Two-Party Handoff (POST-REGISTRATION) The following describes the progress of a two-party handoff. The numbered items refer to steps in Figure 6. The source-triggered HRqst/HRply message for tunnel creation, the target-triggered HRqst/HRply message for tunnel creation, and the HRqst/HRply to extend or terminate a BET (or unidirectional tunnel) are identified using the suffixes (s), (t), and (r), respectively. 1) Either the oFA or nFA receives an L2 trigger informing it that a certain MN is about to move from oFA to nFA. The two cases are: a) The L2 trigger is a source trigger (L2-ST) at oFA. The trigger contains the MN's L2 address and an identifier for the nFA (the IPv4 address itself or an L2 address that can be resolved to the IPv4 address of the nFA). b) The L2 trigger is a target trigger (L2-TT) at nFA. The trigger contains the MN's L2 address and an identifier for the oFA (the IPv4 address itself or an L2 address that can be resolved to the IPv4 address of the oFA). 2) The FA receiving the trigger sends a Handoff Request (HRqst) to the other FA. There are two cases:
a) If oFA is sending the HRqst, the H bit is set and the N bit is unset, indicating it is an HRqst(s). The HRqst(s) contains the lifetime of the tunnel the oFA is willing to support, the MN's IPv4 home address, the MN's HA address, and an LLA option with the MN's L2 address. If the lifetime is zero and the T bit is not set, the oFA is not willing to tunnel any packets for MN. A positive lifetime and a set T bit indicate that the oFA is willing to tunnel for the indicated time. Section 4.5 describes the HRqst(s) and Section 9 describes the LLA option. b) If nFA is sending the HRqst, the N bit is set and the H bit is unset, indicating that it is an HRqst(t). If the T bit is set, nFA has requested a reverse tunnel and the HRqst(t) contains the lifetime of the tunnel the nFA is requesting. The HRqst(t) also contains an LLA option with the MN's L2 address. The MN's IPv4 home address and HA address are not sent, unless they are discovered by some means outside the scope of this document (for example, as part of the L2-TT). Section 4.5 describes the HRqst(t). 3) The FA receiving the HRqst sends a Handoff Reply (HRply) to the other FA. There are two cases: a) If oFA is sending the HRply, the N bit is set and the H and R bits are unset, indicating that the reply is in response to a HRqst(t), i.e., it is an HRply(t). If the T bit is set, the HRply(t) contains the tunnel lifetime the oFA is willing to provide; otherwise, the tunnel lifetime is set to zero indicating that the oFA is not willing to provide tunnel service. If both HRply(t) and HRqst(t) have the T bit set and non-zero lifetime, a BET is established. The HRply(t) also contains the MN's home subnet IPv4 address, the MN's HA address, and an LLA option containing the MN's L2 address. Section 4.6 describes the HRply(t). b) If nFA sends the HRply, the H bit is set and the N and R bits are unset, indicating that this is a response to HRqst(s), i.e., it is an HRply(s). If the T bit is set, the nFA indicates that it requests a reverse tunnel, and the lifetime field is set with the requested tunnel lifetime. The T bit can be set in HRply only if the oFA had set the T bit in the corresponding HRqst or if the nFA is required to reverse tunnel incoming packets to oFA because ingress filtering is enabled on its network. This establishes a BET. The tunnel lifetime requested by the nFA must be less than or equal to the tunnel lifetime offered by oFA in the HRqst(s). Section 4.6 describes the HRply(s).
4) The point during the L2 handoff in which the MN is no longer connected on a given link is signaled by an L2-LD trigger at oFA and MN. Completion of L2 handoff is signaled by an L2-LU trigger at nFA and MN. The trigger is handled as follows: a) When oFA receives the L2-LD trigger, it begins forwarding MN-bound packets through the forward tunnel to nFA. b) When the nFA receives the L2-LU trigger, it begins delivering packets tunneled from oFA to MN and forwards outbound packets from MN using normal routing mechanisms or through a reverse tunnel to oFA or HA. The nFA at this point may not yet be the default router of the MN (see Section 4.4); therefore, to receive all outbound packets from the MN the nFA must send a unicast proxy ARP message (used in [1]) to the MN upon receiving an L2-LU trigger. This proxy ARP message is an ARP Reply [5] sent by the nFA on behalf of oFA, therefore supplying the nFA link-layer address in the Sender Hardware Address field and the oFA IPv4 address in the Target Protocol Address field. c) When the MN receives the L2-LU, it MAY initiate the Mobile IPv4 Registration process by soliciting an Agent Advertisement as described in [1]. If the registration is successful, the nFA takes over the role of anchor FA (aFA) from the oFA. Alternatively, the MN MAY defer the Mobile IPv4 Registration (see Section 4.4). 5) The oFA becomes an aFA if the MN moves to a third FA before having performed a Mobile IPv4 Registration with nFA. 6) Should L2 handoff fail in Step 4 (due to L2 reasons) and a ping-pong situation arise, the oFA may be able to determine this case through the trigger mechanism (i.e., FA sees successive L2-ST/L2-TT followed by L2-LD and then L2-LU). The FA that originated the HRqst can in this case cancel the tunnel by sending an HRqst(r) to the other FA with lifetime zero. It will then simply continue delivering packets to MN exactly as if no handoff had been pending. Section 4.5 describes the HRqst(r). If the oFA sets the B bit in HRqst/HRply and the nFA has not requested a reverse tunnel by setting the T bit, the nFA SHOULD tunnel outgoing packets from the MN to the HA because the MN has requested this service from the oFA. The nFA SHOULD offer this service only if no security between the nFA and the MN's HA is required, or if there is an existing nFA-HA security association.
The actual timing of BET or unidirectional tunnel placement depends on the available L2 triggers. The forward tunnel from oFA to nFA is constructed using one of the tunneling procedures described in [1] for the HA to FA tunnel with the difference that the ends of the tunnel are at the oFA and nFA, respectively. The reverse tunnel from nFA to oFA is constructed as described in [3] with the difference that the network end of the tunnel is at the oFA instead of the HA. If both forward and reverse tunnels are established, then a BET has been established. With optimal L2 trigger information, as described above, the FAs can set up the BET immediately when the L2 handoff is initiated, start tunneling MN-bound data when the link to the MN goes down, and the nFA can use the link-up trigger to start delivering packets. In the absence of optimal L2 trigger information, the HRply can act as the trigger to start tunneling MN-bound data, but in this case, the period of packet delivery disruption to the MN could still be present and additional measures may be required to provide uninterrupted service. Particular implementation and deployment scenarios could require techniques to smooth the handoff by providing a means to convey packets arriving during the L2 handoff. The exact techniques are outside the scope of this document. Figures 7 and 8 show timing diagrams for source trigger (L2-ST) and target trigger (L2-TT) two-party handoffs, respectively. MN nFA oFA | | | | | HRqst(s) |<~~~ L2-ST | |<------------------| | | HRply(s) | | |------------------>| | | | --------------------------------------------<~~~ L2-LD L2 Handoff --------------------------------------------<~~~ L2-LU | | | |<------------------->| | | MN's traffic | | Figure 7 - Two-Party Source Trigger Handoff Timing
MN nFA oFA | | | | L2-TT ~~~>| HRqst(t) | | |------------------>| | | HRply(t) | | |<------------------| | | | --------------------------------------------<~~~ L2-LD L2 Handoff --------------------------------------------<~~~ L2-LU | | | |<------------------->| | | MN's traffic | | Figure 8 - Two-Party Target Trigger Handoff Timing Once the tunnel between aFA and the current FA is in place, it is torn down by one of the following events: 1) The aFA decides to stop tunneling because the lifetime it sent expires and was not renewed, or the aFA or current FA decide to terminate tunnel service prematurely for some other reason (refer to Section 4.3). 2) The MN completes the process by performing a Mobile IPv4 Registration with the current FA. This may be initiated by the FA that sends an Agent Advertisement or by the MN that solicits for an Agent Advertisement as in [1]. 3) The MN moves to a third FA (see Section 4.2)4.2. Three-Party Handoff
Three-party handoff is applicable when an MN, which has already established an aFA and is receiving tunneled packets through its current FA, moves to a new FA without performing a Mobile IPv4 Registration. The need for the three-party handoff function depends on the wireless system in which POST-REGISTRATION is being implemented. For radio L2 protocols in which it is possible for the MN to move so rapidly from one FA to another such that a probability exists that the Mobile IPv4 Registration with nFA will not complete before the MN moves on, HTT (Handoff to Third) SHOULD be implemented. Certain wireless systems and implementations do not allow such fast movement between FAs and may force the Mobile IPv4 Registration to occur soon after L2 handoff, in which case three-party handoff is not applicable. If this three-party handoff feature is not implemented, the FA SHOULD
send an Agent Advertisement to the MN after L2 handoff has completed (L2-LU at nFA) and/or the MN SHOULD solicit an Agent Advertisement after L2 handoff (L2-LU at MN). +------+ +--->| aFA |<---+ | +------+ | 4b) HRqst(r) | | 3) HRqst(t) HRply(r) | | HRply(t) | | v 2a) HRqst v 1a) L2-ST ~~~> +------+ HTT +------+ <~~~ 1b) L2-TT | oFA |<-------->| nFA | 4a) L2-LD ~~~> +------+ 2b) HTT +------+ <~~~ 5a) L2-LU ^ HRply ^ old L2 | | new L2 +-------+ +-----+ | | | | V V +------+ movement 5b) L2-LU ~~~> | MN | ---------> +------+ Figure 9 - Three-Party Handoff The L3 handoff can be deferred either because of a decision by the MN/FA (i.e., MN does not send Agent Solicitations and FA does not send Agent Advertisements), or it may result from rapid movement between oFA and nFA that does not allow enough time for the registration to complete. This scenario is shown in Figure 9. In this case, oFA must inform nFA (i.e., the third FA) to contact aFA about moving the radio end of the tunnel. This is performed with the HTT message. The general idea behind the three-party handoff procedure is that the oFA supplies nFA with the same information it would have obtained via an L2-TT if handoff had occurred from aFA to nFA; then, the nFA performs an HRqst(t)/HRply(t) sequence with aFA in order to move the BET to nFA. When the L2 handoff is complete, oFA sends an HRqst(r) to aFA to terminate the previous BET. The following describes the progress of a three-party handoff. The numbered items refer to steps in Figure 9. 1) Either the oFA or nFA receives an L2 trigger informing it that a certain MN is about to be moved. The two cases are:
a) The L2 trigger is a source trigger (L2-ST) at oFA. The trigger contains the MN's L2 address and an identifier for the nFA (the IPv4 address itself or an L2 address that can be mapped to the IPv4 address of the nFA). b) The L2 trigger is a target trigger (L2-TT) at nFA. The trigger contains the MN's L2 address and an identifier for the oFA (the IPv4 address itself or an L2 address that can be resolved to the IPv4 address of the oFA). 2) The oFA and nFA exchange an HTT/HRply or HRqst/HTT pair. HTT is indicated by setting both the H and N bits in the HRqst or HRply. The HTT message MUST NOT have any tunnel flag bits set, because the tunnel is negotiated between the aFA and nFA, not oFA and nFA. There are two cases: a) The L2 trigger is an L2-ST. The oFA sends HTT to nFA containing the MN's home IPv4 address, the MN's HA address, an LLA containing the aFA's IPv4 address, and an LLA containing the L2 address of the MN. This is enough information for nFA to perform a target-triggered handoff with aFA. The nFA responds with an HRply(s). Section 4.7 describes the HTT. b) The L2 trigger is an L2-TT. The nFA sends HRqst(t) to oFA, exactly as if a two-party handoff were occurring. The oFA responds with HTT containing the same information as in a) above. This is enough information for nFA to perform a target-triggered handoff with aFA. 3) Upon receipt of the HTT, the nFA first checks its Visitor Cache to see whether it is already tunneling for MN. If so, Step 6 is performed. If not, nFA performs a target-triggered handoff with aFA, exactly as in Section 4.1, exchanging an HRqst(t)/HRply(t) pair. Because aFA receives no L2 trigger indicating when L2 handoff starts, it may start tunneling to nFA upon transmission of the HRply(t). 4) Once the L2 handoff is under way and the MN gets disconnected at L2, aFA and oFA exchange messages canceling tunnel service between aFA and oFA and allowing aFA to start the tunnel with nFA. a) The point in the L2 handoff process where the MN gets disconnected from oFA is signaled at oFA by L2-LD.
b) The oFA exchanges an HRqst(r)/HRply(r) pair having lifetime zero with aFA. This cancels tunnel service between oFA and aFA. If aFA has not already established a tunnel to nFA, it must do so immediately upon receipt of the HRqst(r). The aFA provides tunneling service exactly as described in Section 4.1, Step 4a. 5) Completion of L2 handoff is signaled by an L2-LU trigger at nFA and/or MN. The nFA and MN handle the trigger as follows: a) The nFA provides packet delivery service to the MN exactly as described in Section 4.1, Step 4b. b) The MN either defers or initiates Mobile IPv4 Registration when it receives the L2-LU, as in Section 4.1. 6) In the special case where nFA and aFA are the same (i.e., the MN is moving back to the original anchor FA), aFA recognizes that it is tunneling to oFA when it checks its Visitor Cache in Step 3. In this case, there is no need for aFA to send the HRqst(t)/HRply(t) in Step 3. Upon receipt of the L2-LU trigger on handoff completion, the aFA begins routing packets to MN and the tunnel to nFA is torn down. The oFA still exchanges the HRqst(r)/HRply(r) with aFA in Step 4b because oFA cannot know a priori that aFA and nFA are the same, but they are redundant.
Figures 10 and 11 show timing diagrams for source trigger (L2-ST) and target trigger (L2-TT) three-party handoff, respectively. MN nFA oFA aFA | | L2-ST ~~~> | | | | | | | |<-------------| | | | HTT | | | |------------->| | | | HRply(s) | | | |------------------------------>| | | HRqst(t) | | | |<------------------------------| | | HRply(t) | | | | | | ----------------------------------<~~~ L2-LD | |--------------->| L2 Handoff | HRqst(r) | | | |<---------------| | HRply(r) | | | ----------------------------------<~~~ L2-LU | | MN's traffic | | | |<-------------->| | | Figure 10 - Three-Party Source Trigger Handoff Timing
MN nFA oFA aFA | | | | | |<~~~ L2-TT | | | |------------->| | | | HRqst(t) | | | |<-------------| | | | HTT | | | |------------------------------>| | | HRqst(t) | | | |<------------------------------| | | HRply(t) | | | | | | ----------------------------------<~~~ L2-LD | |--------------->| L2 Handoff | HRqst(r) | | | |<---------------| | HRply(r) | | | ----------------------------------<~~~ L2-LU | | MN's traffic | | | |<-------------->| | | Figure 11 - Three-Party Target Trigger Handoff Timing Unlike two-party handoff, the timing of BET establishment between aFA and nFA cannot fully depend on the availability of L2 trigger information because aFA does not receive an L2 trigger signaling L2 handoff. The two timing extremes at which aFA can place the BET with nFA are: 1) At the earliest, aFA MAY start tunneling packets using the BET to nFA after sending the HRply(t) to nFA in response to the request for target-triggered handoff. 2) At the latest, aFA MAY start tunneling packets using the BET to nFA and tear down the BET with oFA when receiving the HRqst(r) from oFA indicating that the MN has disconnected. In addition, aFA MAY continue tunneling to oFA if 1) is selected, until the HRqst(r) is received. In this case, the result may be duplicated packets at the MN because the MN will receive packets through oFA on the old L2 until it disconnects (L2-LD). If 2) is selected, the additional latency will add to the MN's L3 service disruption period. Of course, aFA can choose to place the BET sometime between 1) and 2) if reliable bounds are available on the duration of time between L2-ST/L2-TT and the MN's disconnection (L2- LD). The exact selection of when to establish the BET is likely to
be influenced by network engineering and implementation considerations, including whether a handoff smoothing solution is used, and is beyond the scope of this specification.4.3. Renewal or Termination of Tunneling Service
To prevent a BET from expiring when its lifetime runs out, the MN's current FA signals the aFA to either renew or terminate the BET. This may be the case when the MN defers Mobile IPv4 Registration. If no such signal is received, the aFA will terminate the BET when the lifetime expires. In addition, the current FA or aFA may need to terminate the BET prior to the lifetime expiring. In order to avoid error conditions in which tunnels do not expire even though the MN to which they apply is no longer reachable, FAs SHOULD set the tunnel lifetime field to some value other that 0xffff, which indicates "good until canceled". Figure 12 illustrates the message exchange that occurs between the FA needing to terminate or extend the tunnel (designated FA(1) in the figure) and the other FA (designated FA(2) in the figure). The HRqst(r)/HRply(r) is indicated by setting the R bit in the HRqst/HRply messages. If the HRqst(r) is renewing a BET, then it contains a non-zero lifetime; otherwise, if the lifetime is set to zero, it indicates tunnel termination. The aFA has complete control over whether a tunnel is extended or terminated, and it MAY reply to a request for extension with a shorter lifetime than was requested. HRqst(r) +------+ <-------- +------+ | FA(2)| ---------> | FA(1)| +------+ HRply(r) +------+ Figure 12 - BET Renewal or Termination4.4. When Will the MN Perform a Mobile IPv4 Registration?
The MN/FA have control over when to perform the Mobile IPv4 Registration. Although the MN/FA may decide to defer Mobile IPv4 Registration for a certain period, three possible events can lead to the need to terminate tunneling service. If this occurs, the MN MUST perform the Mobile IPv4 Registration. These events are: 1) The end of life for the BET is pending and a request by the current FA to aFA for renewal has been denied, or alternatively the current FA or aFA needs to terminate the BET prematurely. The FA in this case MUST initiate the Mobile IPv4 Registration by sending an Agent Advertisement to the MN as in [1].
2) The MN itself decides to perform a Mobile IPv4 Registration and initiates it by sending an Agent Solicitation as in [1]. 3) During a source-triggered handoff, the oFA attempts to perform BET handoff but nFA is not capable of performing it. The FA in this case MUST initiate the Mobile IPv4 Registration by sending the MN an Agent Advertisement as in [1]. Note that this situation will never arise during target-triggered handoff because an HRqst(t) will not be sent to oFA by an nFA that doesn't support POST-REGISTRATION. Some detailed scenarios relating to case 2) will be described hereafter. According to [1], when using an FA care-of address, the MN MAY use the FA as its default router. Otherwise, it MUST choose its default router from those advertised in the ICMP Router Advertisement portion of the Agent Advertisement. Here we assume that the FA router is also the MN's default router. In POST- REGISTRATION, when a tunnel is established between oFA and nFA and the MN has moved to nFA, the oFA MUST NOT send Agent Advertisements to the MN. In this case, it is possible that the MN will not receive Agent Advertisements for extended periods of time. According to [8], hosts will remove default router entries if the lifetime of the Router Advertisement expires and no further advertisements are received. Note that the ICMP Router Advertisement lifetime is not related to the Registration Lifetime in the Mobility Agent Advertisement extension [1]. To avoid this disruption, the MN MUST solicit the default router (i.e., FA) before the lifetime of its active default router entry runs out, or alternatively, the FA MUST advertise as soon as the MN-nFA link is up (L2-LU). This effectively means that the MN will at most be able to defer Mobile IPv4 Registration for as long as the remaining lifetime of the active default router, as configured in the ICMP Router Advertisements. The MN MUST perform a Mobile IPv4 Registration [1] when it receives an Agent Advertisement following a POST-REGISTRATION handoff.
4.5. Handoff Request (HRqst) Message Format
This is a new Mobile IPv4 message carried on UDP (destination port 434) [1]. The UDP header is followed by the fields below. 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Type |H|N|R|M|G|T|B| Reserved | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Lifetime | Reserved | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | MN Home Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | HA Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | + Identification + | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Extensions ... +-+-+-+-+-+-+-+- Type 16 (Handoff Request) H Source-triggered handoff request. When set and the N bit is unset, indicates that the request was the result of an L2-ST at oFA. N Target triggered handoff request. When set and the H bit is unset, indicates that the request was the result of an L2-TT at nFA. R Set if the request is an HRqst(r) (i.e., a request to renew the tunnel, H and N bits must be unset). M The FA issuing the HRqst will use Minimal Encapsulation as defined in [1,5] for the tunnel. G The FA issuing the HRqst will use Generic Routing Encapsulation (GRE) [4] as defined in [1,5] for the tunnel. Extensions of HRqst containing GRE type and key Fields are outside the scope of this document.
T For an HRqst(s), indicates that the oFA is willing to support both forward and reverse tunnel service. For an HRqst(t), indicates that the nFA is requesting reverse tunnel service. B When sent in an HRqst(s), indicates that the MN has requested a reverse tunnel to the HA and that the nFA SHOULD use a reverse tunnel to the HA if it will not be reverse tunneling to the oFA. Lifetime The lifetime of the tunnel in seconds. If this is an HRqst(t), then the lifetime represents a request by nFA for a reverse tunnel. If this is an HRqst(s), then the lifetime represents the maximum amount of time that oFA is willing to maintain both forward and reverse tunnels. If this is an HRqst(r), then the lifetime represents a request for the amount of time to renew the tunnel's lifetime. A value of 0 on an HRqst(s) indicates that the oFA is unwilling to grant tunnel service. A value of 0 on an HRqst(t) indicates that the nFA does not require reverse tunnel service. A value of 0 on an HRqst(r) indicates that the tunnel should be terminated. A value of 0xffff indicates infinity. MN Home Address For HRqst(s), the home address of the MN. HA Addr For HRqst(s), the HA address of the mobile node. Identification As defined in [1]. Extensions The message MUST include an LLA (see Section 9) containing the MN's L2 address and an L2 address that can be mapped to an IPv4 address for the FA. This message MUST contain the FA-FA Authentication Extension [11] that is used to secure the HRqst message.
4.6. Handoff Reply (HRply) Message Format
This is a new Mobile IPv4 message carried on UDP (destination port 434) [1]. The UDP header is followed by the fields below. 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Type |H|N|R|M|G|T|B| Reserved | Code | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Lifetime | Reserved | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | MN Home Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | HA Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | + Identification + | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Extensions ... +-+-+-+-+-+-+-+- Type 17 (Handoff Reply) Code A value indicating the result of the Handoff Request. Only two codes are currently supported, 0, indicating success, and 1, indicating that the handoff cannot be performed. The remaining values are for future use. Lifetime The lifetime, in seconds, for which the bidirectional tunnel for the MN will be maintained. If this is an HRply(s), then the lifetime represents a request by nFA, and it can be any value up to the maximum value sent in the HRqst(s). Larger values are assumed to default to oFA's maximum. If this is an HRply(t), then the lifetime represents the maximum amount of time that the oFA will grant to the nFA. If this is an HRply(r), then the lifetime represents the amount of time by which the tunnel life will be extended. If the Code field indicates that handoff failed, the Lifetime field will be ignored and SHOULD be set to zero. A value of 0 on an HRply(t) indicates that the oFA is unwilling to grant service. A value of 0 on an HRply(s) indicates that the nFA does not
require service. A value of 0 on an HRply(r) indicates that the tunnel lifetime will be terminated. A value of 0xffff indicates an infinite lifetime. H Source-triggered handoff reply. When set and the N bit is unset, indicates that the reply is in response to an HRqst(s). N Target-triggered handoff reply. When set and the H bit is unset, indicates that the reply is in response to an HRqst(t). R Set if the reply is an HRply(r). Neither the H nor the N bit are set. M The FA issuing the HRqst will use Minimal Encapsulation as defined in [1,5] for the tunnel. G The FA issuing the HRqst will use GRE [4] Encapsulation as defined in [1,5] for the tunnel. When this flag bit is set, the HRply may require extensions containing the GRE type and key fields, but they are outside the scope of this document. T For an HRply(s), indicates that the nFA is requesting to reverse tunnel service. For an HRply(t), indicates that the oFA is willing to provide both forward and reverse tunnel service. B When sent in an HRply(t), indicates that the MN has requested a reverse tunnel to the HA and that the nFA SHOULD use a reverse tunnel to the HA if it will not be reverse tunneling to the oFA. It can be set in HRply(t) only if the T bit was unset in the corresponding HRqst(t). MN Home Address For HRply(t), the home IPv4 address of the MN. HA Addr For HRply(t), the HA IPv4 address of the MN. Identification As defined in [1]. Extensions This Message MUST contain the FA-FA Authentication Extension [11] that is used to secure the HRply message.
4.7. Handoff to Third (HTT) Message Format
The Handoff to Third message has the same format as the Handoff Request and Handoff Reply messages, except both the H and N bits are set. If the HTT message is in response to an L2-ST and is sent to initiate a handoff, then, with the exception of the H and N bits, the message has the same fields set and includes the same extensions as an HRqst(s). If the HTT message is sent in response to an HRqst(t), then, with the exception of the H and N bits, the message has the same fields set and includes the same extensions as an HRply(t). The tunnel bits MUST NOT be set in the HTT message because BET construction is not negotiated between oFA and nFA; it is negotiated between nFA and aFA in the ensuing HRqst(t)/HRply(t). In addition, the HTT MUST contain the following extensions in the specified order: Solicited IPv4 Address Option: containing aFA's Address LLA Option: containing the L2 address of the MN.4.8. Applicability of POST-REGISTRATION Handoff Method
The POST-REGISTRATION handoff approach allows FAs to communicate directly about a pending handoff, and does not require any IPv4-layer messages to be sent to or from an MN prior to the L2 handoff event. Therefore, it eliminates a possible source of handoff latency. This may be required when the link layer imposes hard deadlines on the time at which a handoff must occur, such as when an MN is rapidly moving out of a radio coverage area. Consequently, POST-REGISTRATION is primarily of interest in handoff between FAs that support the same radio access technology. Handoff between heterogeneous radio technologies will, of necessity, require interaction between the MN and the network, and so is not a domain of applicability for POST- REGISTRATION. Because a POST-REGISTRATION handoff is triggered by an unspecified mechanism that informs the oFA or nFA that an L2 handoff is pending, the POST-REGISTRATION approach is only applicable to networks where such a mechanism is available. For example, an L2 may provide indications of radio signal quality that cause the oFA or nFA to send the POST-REGISTRATION handoff messages. Any such indications must also provide each FA involved in the handoff with the identity of the other, so that messages can be sent to the right place. This may involve mapping L2 information onto FA IPv4 addresses. Also, the FAs involved in a handoff must have pre-provisioned security arrangements so that the POST-REGISTRATION messages can be authenticated. If a handoff is to be completed as a result of the POST-REGISTRATION
messaging, any L2 handoff indications must also be securely authenticated so that traffic to the old point of attachment is not improperly halted. POST-REGISTRATION handoff is appropriate in the following cases: - L2 triggers are available on the network to indicate that L2 handoff is pending. - Pre-provisioned security mechanisms are in place to allow fast and secure messaging between the FAs and between the MN and an FA. - Access point choice by the MN is not a concern or the choice requires user intervention and therefore is not on the critical path for handoff.5. Combined Handoff Method
The combined method uses both PRE-REGISTRATION and POST-REGISTRATION handoff. If PRE-REGISTRATION does not complete prior to the expiration of a timer on the nFA, the POST-REGISTRATION mechanism is used to create the tunnel between oFA and nFA. This protects the MN from delays caused by errors such as loss of the Mobile IPv4 Registration Reply message involved in PRE-REGISTRATION for the mobile-initiated and network-initiated source-triggered cases. It also protects the MN from delays caused by errors or the loss of any of the Mobile IPv4 messages involved in PRE-REGISTRATION for the network-initiated target-triggered case. When the nFA receives a target trigger, it will follow the PRE- REGISTRATION procedure. When the combined method is used, the nFA MUST also start a timer when it receives a target trigger. The timer should be set to a small value (default for target trigger case: 1 second). According to PRE-REGISTRATION, the nFA will receive the Registration Request from the MN. When the combined method is used, this Registration Request sent by the MN MUST contain the IPv4 address of the oFA in an FA IPv4 address LLA extension (see Section 9.7). This same Registration Request message will contain multiple LLA extensions, since it will also contain the MN's layer 2 address in an LLA extension as described for PRE-REGISTRATION (see Sections 3.7 and 9). When the nFA has not started the handoff procedure using a target trigger (i.e., mobile-initiated or network-initiated target- triggered cases), the nFA MUST start a timer as soon as it receives the low-latency Registration Request from the MN. This timer should be set to a small value (default: 1 second).
In all cases, the timer MUST be reset when the Registration Reply message is received by nFA. If the timer expires before the Registration Reply is received, the nFA MUST initiate the POST- REGISTRATION procedure. The nFA utilizes the oFA IPv4 address (previously received in the extension to the Registration Request message) as the destination of the POST-REGISTRATION HRqst message to create the tunnel between nFA and oFA. The nFA MAY tear down this tunnel when it receives and forwards a successful Registration Reply for that MN.6. Layer 2 and Layer 3 Handoff Timing Considerations
In the optimal cases considered in the PRE-REGISTRATION and POST- REGISTRATION handoffs, it was assumed that a timely L2 trigger would be received in such a way that packets could be delivered to the MN via its nFA immediately upon connection. In this way, the MN does not suffer disruption due to the L3 handoff. However, such precise timing of the L2 trigger and handoff mechanism with respect to the actual L2 handoff event will not be possible in all wireless systems and may depend on particular implementation techniques. Therefore, some uncertainty may exist at L3 as to exactly when the L2 connection between the MN and the nFA becomes fully established and can be used for L3 traffic. It is possible that in certain implementations traffic will be re-routed too early or too late with respect to the moment when the connection between the MN and the nFA becomes fully established. The techniques that may solve this problem and allow the MN to receive traffic independently of the timing of the L2 handoff event include buffering and simultaneous bindings (i.e., bicasting: setting the S bit [1] in Registration Requests). However, these are optional and are not mandated.7. Reverse Tunneling Support
The handoff methods all support reverse tunneling. The MN may request reverse tunneling [3] by setting the T bit in its Registration Request. In the case of POST-REGISTRATION, if the MN had requested reverse tunneling previously at oFA, the handoff message from oFA (see Section 4) includes the T bit enabled to inform nFA to establish a BET for the visitor entry. Typically, the T bit will always be set to ensure that any delays in the MN receiving its new care-of address do not result in any delay in uplink packet transmission from the MN, but local policies and particular L2 technologies may allow the reverse tunnel to be turned off.
8. Handoff Signaling Failure Recovery
In general and to a greater extent in wireless networks, packets carrying handoff signaling may be dropped or lost due to errors on the link. In this section, we consider mechanisms for recovery from handoff signaling failures.8.1. PRE-REGISTRATION Signaling Failure Recovery
Failure of PRE-REGISTRATION signaling breaks down into three cases: 1) Loss of messages PrRtSol and PrRtAdv on the air link. 2) Loss of the solicitation by an FA to obtain another neighboring FA's Advertisement or loss of the neighboring FA's advertisement. 3) Failure of the standard Mobile IPv4 Registration. Of these, case 3) is handled by standard Mobile IPv4 mechanisms described in [1]. Case 2) is expected to be a rare event because spontaneous packet drop rates on the fixed network are caused by congestion or router failure. Since bit error rates on wireless links are higher than on fixed links, case 1) is more likely to occur. In the following subsections, cases 1) and 2) are considered.8.1.1. Failure of PrRtSol and PrRtAdv
PRE-REGISTRATION handoff can fail in network-initiated handoff when the PrRtAdv sent by oFA in response to the source trigger (L2-ST) or the advertisement sent by nFA in response to the target trigger (L2- TT) fails to reach the MN. PRE-REGISTRATION handoff can also fail in mobile-initiated handoff when either the PrRtSol sent from the MN or return PrRtAdv sent from the oFA is dropped. To reduce the probability that PrRtAdv and PrRtSol are lost, the MN and FA MAY transmit multiple copies of these messages. Should these messages fail anyway, in both cases the MN connects to the nFA without having received any prior signaling. In this case, the MN solicits an FA Advertisement when it connects to nFA at L2 (L2-LU), as described in Section 3.6, and performs a standard Mobile IPv4 Registration with the nFA as specified in [1].
8.1.2. Failure of Inter-FA Solicitation and Advertisement
The solicitation from an FA to another neighboring FA may fail or the corresponding advertisement from the neighboring FA may be lost. To reduce the probability that these messages are lost, the FAs MAY transmit multiple copies of these messages. If a failure occurs anyway, the FA soliciting the Agent Advertisement is unable to send a PrRtAdv in response to a source trigger or to a mobile-initiated PrRtSol. In these cases, when the MN does not receive a notification or confirmation of a PRE-REGISTRATION handoff, the MN MUST perform a standard Mobile IPv4 Registration as soon as it connects to the nFA (L2-LU) as described in Section 8.1.1.8.2. POST-REGISTRATION Signaling Failure Recovery
Failure occurs in POST-REGISTRATION when either the HRqst or HRply message is dropped. The effects of the failure and the recovery procedure depend on which message is dropped, and whether the handoff is source or target triggered. Since all of the POST-REGISTRATION signaling is going over the fixed network, it can be expected that spontaneous dropping of packets in the absence of congestion and router failure should be a relatively rare event. Nevertheless, failure recovery mechanisms SHOULD be implemented.8.2.1. HRqst Message Dropped
If the HRqst message is dropped, the effect is the same for both source- and target-triggered handoffs. In either case, the FA to which the HRqst was destined will never respond with an HRply message. If the handoff is source triggered, then the nFA never learns of the handoff, and the oFA never receives confirmation. If the handoff is target-triggered, then the oFA never learns of the handoff, and the nFA never receives confirmation. The recovery procedure in this case is as follows: the oFA MUST NOT construct a forward tunnel when the MN moves off-link (L2-LD) if the handoff is source-triggered, and the nFA MUST NOT construct a reverse tunnel if the handoff is target triggered. If the nFA was not informed of the handoff by an HRqst message (corresponding to failure of source-triggered handoff) or if the handoff was not confirmed by an HRply message (corresponding to failure of target-triggered handoff), the nFA MUST unicast an Agent Advertisement to the MN as soon as its L2 connection is established (L2-LU at nFA).
8.2.2. HRply Message Dropped
If the HRply message is dropped, the FA sending the HRply will assume that the handoff has been confirmed, but the FA that is expecting to receive the HRply does not receive confirmation. In this case, the failure recovery procedure is different for source-triggered and target-triggered handoffs. In a target-triggered handoff, the oFA assumes that the handoff is confirmed because it has sent the HRply, but the nFA has not received it so it does not have confirmation. The oFA starts tunneling packets to the nFA when the MN moves from its link (L2-LD). The nFA MUST send an FA Advertisement to the MN as soon as its L2 link is up (L2-LU at nFA) and MAY drop the tunneled packets. The nFA SHOULD send an ICMP Destination Unreachable [9] message to the oFA. When the oFA receives this message, it will terminate the tunnel and stop forwarding packets. If reverse tunneling was requested, the nFA MUST NOT reverse tunnel because it has not received handoff confirmation. In source-triggered handoff, the nFA assumes that the handoff is confirmed because it has sent the HRply, but the oFA has not received it so it does not have confirmation. Without failure recovery, the MN could move to the nFA without the oFA being able to start the forward tunnel for the MN's packets, and the MN would not be able to initiate a Mobile IPv4 Registration because it does not know that the handoff has failed. In this situation, the oFA MUST send out an HRqst message to the nFA with lifetime zero as soon as the MN leaves its link (L2-LD). The oFA SHOULD continue to retransmit the HRqst message, with exponential backoff for CONFIG-HFAIL seconds or until it receives an HRply acknowledging the request to cancel the tunnel. The default value for CONFIG-HFAIL is 10 seconds. When the nFA receives the HRqst, it MUST immediately send an Agent Advertisement to the MN, as is the case whenever a tunnel is canceled. In addition, the oFA MUST also drop any packets received through the reverse tunnel from the nFA. The oFA SHOULD NOT send the ICMP Destination Unreachable message to the nFA because the nFA has been informed by the HRqst message to cancel the tunnel. However, if the nFA receives an ICMP Destination Unreachable message for the tunnel prior to receiving the HRqst canceling the tunnel, it MUST send an FA Advertisement to the MN and cancel the tunnel.