| Network Working Group | P. Pfister |
| Internet-Draft | B. Paterson |
| Intended status: Standards Track | Cisco Systems |
| Expires: March 23, 2015 | J. Arkko |
| Ericsson | |
| September 19, 2014 |
Prefix and Address Assignment in a Home Network
draft-ietf-homenet-prefix-assignment-00
This memo describes a home network prefix and address assignment algorithm running on top of any 'flooding protocol' that fulfills the specified requirements. It is expected that home border routers are allocated one or multiple IPv6 prefixes through DHCPv6 Prefix Delegation (PD) or that prefixes are made available through other means. An IPv4 address can also be assigned and private addresses be used with NAT to provide IPv4 connectivity. In both cases, provided prefixes need to be efficiently divided among the multiple links, and routers need to obtain addresses. This document describes a distributed algorithm for IPv4 and IPv6 prefixes division, assignment and router's address assignment, and specifies how hosts can be given addresses and configuration options using DHCP or SLAAC.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on March 23, 2015.
Copyright (c) 2014 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
This memo describes a fully distributed prefix and address assignment algorithm for home networks, running on top of any 'flooding protocol' that fulfills the specified requirements. It is expected that home border routers are allocated one or multiple IPv6 prefixes through DHCPv6 Prefix Delegation (PD) [RFC3633] or that prefixes are made available through other means. When an IPv4 address is assigned, a home private IPv4 prefix may be used with NAT to provide IPv4 connectivity to the whole home, as well as Unique Local Address prefixes [RFC4193] may be used in order to provide internal connectivity whenever global IPv6 connectivity is not available.
Obtained IPv6 or IPv4 prefixes need to be efficiently divided among the multiple links. For the purposes of this document, we refer to this process as prefix assignment. This memo describes an algorithm for such prefix division, assignment and router's address assignment, as well as the way hosts can be given addresses and configuration options using DHCPv4 [RFC2131], DHCPv6 [RFC3315] or SLAAC [RFC4862]. In the rest of this document DHCP refers to both DHCPv4 and DHCPv6.
Although this document recommends the use of 64 bits long prefixes, the algorithm do not require routers to assign prefixes of particular lengths. When a delegated prefix is too small considered the number of links in the home network, higher priority links may be privileged or smaller prefixes can be assigned in order to avoid prefix scarcity.
The rest of this memo is organized as follows. Section 2 defines the usual keywords, Section 3 outlines the algorithms functioning and features, Section 4 describes how a home router behaves when running the prefix and address assignment algorithm. Requirements for the underlying flooding protocol are detailed in Section 5. The prefix assignment algorithm is detailed in Section 6 and Section 7 focuses on the address assignment algorithm. Section 8 explains the hysteresis principles applied to both prefix and address assignments, Section 9 specifies the procedures for automatic generation of ULA and IPv4 prefixes, Section 10 explains what administrative interfaces are useful for advanced users that wish to manually interact with the mechanisms, Section 11 gives values for the constants used in this document, Section 12 discusses the security aspects and finally, Appendix A provides implementation guidelines for the optional scarcity avoidance mechanism.
The Prefix Assignment Algorithm was first detailed in [I-D.arkko-homenet-prefix-assignment]. This document is a continuation and generalization of that draft to any underlying flooding protocol. It also adds support for arbitrary prefix length, IPv4, scarcity avoidance mechanism or manual configuration.
In this document, the key words "MAY", "MUST, "MUST NOT", "OPTIONAL", "RECOMMENDED", "SHOULD", and "SHOULD NOT", are to be interpreted as described in [RFC2119].
Given one or multiple prefixes for the entire network, each prefix is subdivided by the prefix assignment algorithm so that every link is given one assignment per available prefix. Assignments are advertised through the whole network using the underlying flooding protocol, collisions are detected and valid assignments are chosen and applied on every link. Once a prefix is applied, hosts and routers may be given addresses. In summary, the algorithm works in four steps:
This algorithm, which intends to fulfill requirements specified in [I-D.ietf-homenet-arch], has the following features:
All home routers participating in the prefix assignment algorithm MUST fulfill the requirements defined in this document and use a common flooding protocol and routing protocol. Classic CPE routers [RFC7084] are supported as downstream routers and dowstream DHCPv6-PD enabled routers are supported as both downstream and uplink routers, but problems may occur when such router is connected to the home network on both WAN and LAN side. In the later case, finer external interface detection algorithm or static configuration can be used to solve the issue, but these are out of the scope of this document.
Each router MUST maintain a list of all the Delegated Prefixes. These prefixes may be locally generated, as described in Section 4.3, or come from other routers as described in Section 5.4.
Each router MUST maintain a list of all the Assigned Prefixes advertised by other routers. Each AP is learnt through the mechanisms described in Section 5.3 and is defined as a tuple of: Section 5.3.
The AP list is the result of the information provided by the flooding protocol, as specified in
The router MUST maintain a list of all prefixes currently chosen to be applied on connected links. They are Chosen Prefixes (CPs) and described by a tuple of:
Chosen Prefixes that are marked as 'Advertised' are distributed to other routers using the flooding protocol and are therefore considered as Assigned Prefixes by other routers. The goal of the Prefix Assignment Algorithm is to ensure that all routers have a consistent view of Assigned Prefixes on each link.
The router MUST maintain a database of its own address assignments, and address assignments made by other routers on connected links as learnt through means described in Section 5.5.
Each interface MUST either be considered as internal or external. Prefixes and addresses are only assigned to internal interfaces. The criteria to make this distinction are out of the scope of this document.
If an internal interface becomes external, all prefixes and addresses assigned on the considered interface MUST be deleted and no longer announced, and the prefix assignment algorithm MUST be run.
If an external interface becomes internal, the prefix assignment algorithm MUST be run (see Section 6.1).
Whenever two or more interfaces are connected to the same link, all but one of them SHOULD be ignored by the prefix assignment algorithm. A mechanism to detect such situation SHOULD be provided by the flooding algorithm.
A Delegated Prefix can be obtained or generated through different means:
DHCP options MAY be attached to a delegated prefix by the router that either generated the prefix or received it through DHCPv6 PD. IPv6 delegated prefix options MUST be encoded as DHCPv6 options. IPv4 delegated prefix options MUST be encoded as DHCPv4 options.
As DHCP options are numerous and new ones may be defined, specifying routers' behavior regarding each option is out of the scope of this document. In order to avoid misconfiguration, routers must follow the two following general rules:
The mif working group may provide useful inputs concerning the way the home network should handle different prefixes associated with heterogeneous uplinks.
A router considers itself as the Network Leader if and only if its router ID is greater than all other router IDs in received Prefix Assignments and Delegated Prefixes.
On a link where custom host configuration must be provided, or whenever SLAAC cannot be used, a DHCP server must be elected. That router is called designated router and is dynamically chosen by the prefix assignment algorithm.
A router MUST consider itself designated router on a given link if either one of the following conditions holds:
On a given link, the designated router MUST send router advertisements including Prefix Information Options for all the Chosen Prefixes associated to that link. SLAAC SHOULD be enabled when possible, unless the configuration states otherwise. The valid and preferred lifetimes MUST be set to values lower or equal to the associated Delegated Prefix's valid and preferred lifetimes.
On a given link, whenever SLAAC can't be used for all assignments, or DHCP configuration options must be provided to hosts, the designated router MUST act as a DHCP server and serve addresses on the given link. A router MUST stop behaving as a DHCP server whenever it is not the link's designated router anymore.
Routers's addresses pool, specified in Section 7, MUST be excluded from DHCP hosts pools.
The valid and preferred lifetimes MUST be set to values lower or equal to the associated Delegated Prefix's valid and preferred lifetimes.
Once a Chosen Prefix is created, a router first waits some time in order to detect possible collisions (Section 8). Afterwards and if no collision is detected, the prefix is applied as follows:
When a prefix assignment is removed, the previous steps MUST be undone. The router MUST also deprecate the prefix, if it had been advertised in Router Advertisements on an interface. The prefix is deprecated by sending Router Advertisements with the PIO's preferred lifetime set to 0 [RFC4861]. Hosts that support DHCP reconfigure extension ([RFC3203], [RFC3315]) and that have been given leases MUST be reconfigured as well.
DHCP options attached to each delegated prefixes and propagated through the flooding protocol SHOULD contain the DHCP DNS options provided by the ISP (when provided).
Whenever the router knows which DNS server to use, or is acting as a DNS relay, it SHOULD include DNS DHCP options ([RFC3646]) within host's configuration messages and include the Router Advertisement DNS options ([RFC6106]) when sending RAs.
DNS server selection in multi-homed networks is a complex issue that this document doesn't intend to solve. One should look at IETF's mif working-group documents in order to obtain guidelines concerning DNS server selection. It is RECOMMENDED that designated routers turns on a local DNS relay that fetches information from provided DNS servers.
In this document, the Flooding Protocol (FP) refers to a protocol enabling information propagation to the whole network. It was not specified in order to allow the working group to independently decide which routing protocol, configuration protocol, and prefix assignment method to use within the home network. Routing protocol, like OSPFv3 [RFC5340] (With its autoconf extension [I-D.ietf-ospf-ospfv3-autoconfig]) or IS-IS [RFC5308], could be extended in order to fulfill the requirements. An independent protocol, for instance HNCP [I-D.ietf-homenet-hncp], could be used as well.
The specified algorithm can use any protocol that fulfills the requirements specified in this section.
The FP MUST provide a router ID. ID collisions within the network MUST be rare and any conflicts MUST be resolved by the flooding protocol. When the router ID is changed, the FP MUST immediately provide the new ID to the Prefix Assignment Algorithm, which will in turn be run again, without requiring the current state to be flushed.
In the absence of collisions, the router ID MUST NOT be changed, and it SHOULD be stable across reboots, power cycling and router software updates.
The FP MUST provide an approximate upper bound of the time it takes for an update to be propagated to the whole network. This value is referred to as the FLOODING_DELAY. The algorithm ensures that, as long as the upper bound is respected, two identical prefixes will never be applied to different links, and two different prefixes will never be applied to the same link. The algorithm and the network will recover when the upper bound is exceeded, but collisions may appear in the routing protocol and errors may be propagated to upper layers.
If the FP supports link-local flooding, which is used for router's address assignments, it SHOULD provide an approximate upper bound of the time it takes for an update to be propagated to a single link. This value is referred to as the FLOODING_DELAY_LL. If link-local flooding is not available, or the value is not provided, the assignment algorithm MUST use the FLOODING_DELAY value instead.
The FP MUST provide a way to flood Chosen Prefixes marked as advertised and retrieve prefixes assigned by other routers (APs). Retrieved APs MUST contain all the information specified in Section 4.1.
The FP must provide a way to flood Delegated Prefixes and retrieve prefixes delegated to other routers. Retrieved entries must contain the following information.
The FP MUST make sure time values are consistent throughout the network (i.e. differences are small compared to Delegated Prefixes lifetimes). If no time synchronization protocol is used, the FP MUST keep track of prefix age across the network and within its database.
Routers addresses are dynamically allocated, picked from a defined pool, and collisions must be detected using the FP. The FP MUST provide a way to flood routers' addresses. The flooding scope of those values SHOULD be link-local, but as addresses are unique within the home network, this is not mandatory. For each address assignment, the FP SHOULD provide the identifier of the interface connected to the link the address assignment was advertised on.
The Prefix Assignment Algorithm is a distributed algorithm that assigns one prefix from each available Delegated Prefix on every link that is considered to be internal by at least one connected router. The algorithm itself does not distinguish between global IPv6, ULA or IPv4 prefixes. IPv4 prefixes are encoded as their IPv4-mapped IPv6 form, as defined in [RFC4291] (i.e. ::ffff:A.B.C.D/X with X >= 96).
When the Prefix Assignment Algorithm is executed, combinations of Delegated Prefixes and internal interfaces are being considered. For the purpose of this discussion, the Delegated Prefix will be referred to as the current Delegated Prefix, and the interface will be referred to as the current Interface. If a delegated prefix is included inside another delegated prefix, it is ignored. This rule intends to ignore prefixes delegated from non-Homenet routers that previously obtained their larger prefix from one of Homenet's routers.
The algorithm is specified here for the sake of clarity. It can be optimized in some cases. For instance Prefix Assignment deletion might not need to trigger algorithm's execution if all internal interfaces already have assignements associated to the same Delegated Prefix. Similarly, when an ignored Delegated Prefix is deleted, it is not necessary to run the algorithm. An implementation may work differently than specified here as long as the resulting behavior is identical to the behavior a router implementing this exact algorithm would have.
The algorithm MUST be run whenever one of the following event occurs:
It is not required that the algorithm is synchronously run each time such an event occurs. But the delay between the event and the algorithm execution MUST be small compared to FLOODING_DELAY.
An assignment is said to take precedence over another assignment when:
An Assigned Prefix or a Chosen Prefix is said to be valid if all the following conditions are met:
A prefix is said to be available if it does not overlap with any other assignment by any other router in the network.
An AP is said to be accepted when the AP is currently being advertised by a different router on a directly connected link, and will be used by the accepting router as a new Chosen Prefix. When a router accepts a neighbor's assignment, it starts a timer as specified in Section 8. A new CP is created from the AP, with:
When the algorithm decides to make a new assignment, it first needs to specify the desired size of the assigned prefix. Although that choice is completely implementation specific, prefixes of size 64 are RECOMMENDED. The following table MAY be used as default values, where X is the length of the delegated prefix.
When the algorithm decides to make a new assignment, it SHOULD first checks its stable storage for an available assignment that was previously applied on the current interface and is part of the current delegated prefix. If no available assignment can be found that way, the new prefix MUST be randomly selected among a subset of available prefixes (if possible, large enough to avoid collisions). Hardware specific identifiers may be used to seed a pseudo-random generator.
If no available prefix is found, the assignment fails.
The algorithm leaves much room for implementation specific policies. For instance, static prefixes may be configured as specified in Section 10. If implemented, the router MAY also decide to execute the Prefix Scarcity Avoidance mechanisms, as proposed in Appendix A.
If an available prefix is found, a new assignment is made and a new Chosen Prefix entry is created.
A new assignment is always marked as advertised when created and therefore immediately provided to the flooding protocol.
When some authority (Delegating router, system admin, etc...) wants to manually enforce some behavior, it may ask some router to make an Authoritative Prefix Assignment. Such assignments have their Authoritative bit set, CAN NOT be overridden, and will appear in other router's database as Assigned Prefixes with the Authoritative bit set.
There are two kinds of Authoritative Prefix Assignments.
When a delegated prefix is obtained through DHCPv6 PD with a non-empty excluded prefix, as specified in [RFC6603], an Authoritative Prefix Assignment MUST be created with the excluded prefix.
When either a new Prefix Assignment is made, or an Authoritative Prefix Assignment is created, the creating router needs to choose which priority value to use. The assignment priority is kept by the designated router when it starts advertising the assignment, and is useful when not enough prefixes are available.
At the beginning of the algorithm, all assignments that do not have their Authoritative bit set are marked as 'invalid', and the router computes for each connected link whether it is the designated router, as specified in Section 4.5.
The following steps are then executed for every combination of delegated prefixes and interfaces.
In the end all the assignments that are marked as invalid are deleted.
If some host or non-Homenet router asks for Delegated Prefixes, a router MAY assign a set of prefixes and give them to the client. Such assignments MUST be advertised as either not assigned on any link, or assigned on a stub virtual link connected to the router, depending on the Flooding Protocol capabilities. By default assignments priorities MUST be between PRIORITY_AUTO_MIN and PRIORITY_AUTO_MAX, SHOULD be lower than PRIORITY_DEFAULT, and the authoritative bit MUST not be set. Whenever such an assignment becomes invalid, DHCPv6 Reconfigure SHOULD be used in order to remove the prefix from DHCPv6 DP client's lease. If DHCPv6 Reconfigure is not supported, leases lifetimes SHOULD be significantly small.
Provided DPs' valid and preferred lifetimes MUST be lower or equal to their associated Delegated Prefix's lifetimes, and associated DHCPv6 data SHOULD be provided to the DHCPv6 PD client.
By default, an assigned prefix SHOULD NOT be provided to a DHCPv6 PD client before the apply timeout has elapsed. But in order to allow faster response delay, a lease MAY first be provided with a lifetime of 2*FLOODING_DELAY seconds, even if the private assignments' apply timeout has not elapsed yet.
IPv6 routers always get at least one link-local address per link. Routing protocols and link DHCP servers are able to run with these addresses. In some cases though, a router may need to take one or multiple addresses among one or multiple available Delegated Prefixes. For example:
When possible, SLAAC MUST be used. In other cases a different mechanism is necessary for routers to get addresses. This document proposes an Address Assignment Algorithm that extends the Prefix Assignment Algorithm and works as follows. Each prefix assignment is associated with a fixed address pool, reserved for router's addresses assignment. The address pool is a prefix which value is deterministically function of the assigned prefix. A router CAN, at any time, decide to assign itself an address from any of its Chosen Prefixes. Just like prefix assignments, address assignments are advertised to other routers and collisions are detected. Routers MUST keep track of Address Assignments made by other routers on connected links by using information provided by the flooding algorithm, as defined in Section 5.5.
Given an assigned prefix A/X (where all A's latest '128 - X'th bits are set to 0), the routers reserved address pool is defined as follows:
In the case of IPv4 prefixes, the network address (first address of the address pool) MUST not be used.
In this section, we say an address assignment is made by some router when it intends to use, or is using the address specified by this assignment. An assignment, made by some router, MUST be advertised on the link on which the assignment is made. Similarly, an address assignment is said to be applied when the address is pushed to the router's interface configuration. It is unapplied otherwise.
Routers MUST store applied address assignments in their stable storage and reuse the same addresses whenever possible. At least the five previously applied addresses SHOULD be stored for each interface.
For a given prefix assignment, an address is said to be available if it is within the router's address pool associated to the prefix assignment, and it is not being advertised by any other router. If the flooding protocol provides interface identifier in the address assignments, looking for collisions on considered link is enough.
A new address assignment MUST be chosen randomly among available addresses. An address assignment MUST NOT be applied when one of the following condition is true.
An address assignment must be deleted whenever one of the following condition becomes true.
When the flooding protocol is started, the router MUST wait FLOODING_DELAY before executing the prefix assignment algorithm for the first time.
Prefix and address assignment algorithms are distributed. Collisions may occur, but network configuration, routing protocols or upper layers should not suffer from these collisions. For this reason, all assignments that could imply collisions are not immediately applied.
When a router stops advertising a Delegated Prefix, it MUST first deprecate that Delegated Prefix by advertising it for DP_DEPRECATE_FACTOR*FLOODING_DELAY seconds with zero valid and preferred lifetimes.
When a router receives a deprecated Delegated Prefix advertisement, it must remove the Delegated Prefix from its Delegated Prefixes list.
When a router stops receiving a Delegated Prefix from the Flooding Protocol, it SHOULD keep using that delegating prefix up to a period of min(remaining lifetime, DP_KEEP_ALIVE_TIME) seconds.
Although DHCPv6 PD and static configuration are regular means of obtaining IPv6 prefixes, routers MAY, in some cases, autonomously decide to generate a delegated prefix. In this section are specified when and how IPv6 ULA prefixes and IPv4 private prefixes may be autonomously generated.
A router MAY generate a ULA prefix when the two following conditions are met.
A router MUST stop advertising a spontaneously generated ULA prefix whenever another router is advertising a ULA delegated prefix.
The most recently used ULA prefix SHOULD be stored in stable storage by all routers and reused whenever choosing a new ULA delegated prefix. If no ULA prefix can be found in stable storage, it MUST be randomly generated, or generated from hardware specific values.
A router MAY generate an IPv4 prefix when the two following conditions are met.
A router MUST stop advertising an IPv4 prefix whenever another router with an higher router ID is advertising an IPv4 Delegated Prefix.
The IPv4 private prefix must be included in one of the private prefixes defined in [RFC1918]. The prefix 10/8 SHOULD be used by default but it SHOULD be configurable. In the case the address provided by the ISP is already a private address, a different private prefix SHOULD be used. For instance, if the ISP is giving the address 10.1.2.3, 10/8 or any sub-prefix included in 10/8 SHOULD NOT be used. (For instance, 172.16/12 or 192.168/16 can be selected).
The algorithm leaves much room for implementation specific features. For instance, ULA prefix as well IPv4 prefix generation may be disabled whenever a global IPv6 is made available. This section details a few other possible configuration options.
The implementation MAY allow each internal interface to be configured with a custom priority value. The specified priority SHOULD then be used when creating new assignments on the given interface. If not specified, the default priority SHOULD be used.
The implementation SHOULD allow manual assignments on given links. When specified, and whenever such an assignment is valid, it MUST be advertised as Authoritative Assignments on the given interface.
PRIORITY_MIN 0 PRIORITY_AUTHORITY_MIN 4 PRIORITY_AUTO_MIN 6 PRIORITY_DEFAULT 8 PRIORITY_AUTO_MAX 10 PRIORITY_AUTHORITY_MAX 12 PRIORITY_MAX 15 DP_DEPRECATE_FACTOR 3 DP_KEEP_ALIVE_TIME 60 seconds
Prefix assignment algorithm security entirely relies on flooding protocol security features. The flooding protocol SHOULD therefore check for the authenticity of advertised information. Security modes may be classified in three categories.
Whenever a malicious router attacks an unprotected network, or whenever a malicious router is able to authenticate itself to a network as stated in the second case, it may for example:
If a malicious router is able to authenticate itself in a network protected as in the third case, most of the previously listed attacks may still be performed, but traffic could only be redirected toward the origination of the attack, and the source of the attack could be identified.
In any case, in order to protect the network, the routing protocol as well as the way hosts are configured also needs to be protected, hence requiring other link (e.g. WPA) or IP layer (e.g. IPSec-Auth [RFC4302] or SeND [RFC3971]) security solutions.
| [RFC3971] | Arkko, J., Kempf, J., Zill, B. and P. Nikander, "SEcure Neighbor Discovery (SEND)", RFC 3971, March 2005. |
| [RFC4302] | Kent, S., "IP Authentication Header", RFC 4302, December 2005. |
| [RFC5308] | Hopps, C., "Routing IPv6 with IS-IS", RFC 5308, October 2008. |
| [RFC5340] | Coltun, R., Ferguson, D., Moy, J. and A. Lindem, "OSPF for IPv6", RFC 5340, July 2008. |
| [RFC7084] | Singh, H., Beebee, W., Donley, C. and B. Stark, "Basic Requirements for IPv6 Customer Edge Routers", RFC 7084, November 2013. |
| [I-D.ietf-homenet-arch] | Chown, T., Arkko, J., Brandt, A., Troan, O. and J. Weil, "IPv6 Home Networking Architecture Principles", Internet-Draft draft-ietf-homenet-arch-17, July 2014. |
| [I-D.ietf-homenet-hncp] | Stenberg, M. and S. Barth, "Home Networking Control Protocol", Internet-Draft draft-ietf-homenet-hncp-01, June 2014. |
| [I-D.ietf-ospf-ospfv3-autoconfig] | Lindem, A. and J. Arkko, "OSPFv3 Auto-Configuration", Internet-Draft draft-ietf-ospf-ospfv3-autoconfig-09, September 2014. |
| [I-D.arkko-homenet-prefix-assignment] | Arkko, J., Lindem, A. and B. Paterson, "Prefix Assignment in a Home Network", Internet-Draft draft-arkko-homenet-prefix-assignment-04, May 2013. |
| [I-D.bhandari-dhc-class-based-prefix] | Systems, C., Halwasia, G., Gundavelli, S., Deng, H., Thiebaut, L., Korhonen, J. and I. Farrer, "DHCPv6 class based prefix", Internet-Draft draft-bhandari-dhc-class-based-prefix-05, July 2013. |
| [I-D.chelius-router-autoconf] | Chelius, G., Fleury, E. and L. Toutain, "Using OSPFv3 for IPv6 router autoconfiguration", Internet-Draft draft-chelius-router-autoconf-00, June 2002. |
| [I-D.dimitri-zospf] | Dimitrelis, A. and A. Williams, "Autoconfiguration of routers using a link state routing protocol", Internet-Draft draft-dimitri-zospf-00, October 2002. |
Although an ISP should provide enough addresses, an implementation must carefully manage the provided address space. First, when a new assignment is made, the prefix should be selected amongst a set of prefixes so that prefix waste is minimized. Then, a router may decide to execute procedures intended to avoid prefix scarcity. Different approaches are possible. This section intends to provide guidelines for such procedures. They are optional and are compatible with routers that only support basic requirements defined in this document.
Given a Delegated Prefix, different routers may try to assign prefixes of different lengths. Particularly, a non-homenet downstream router may ask for a delegated prefix of significant size, as specified in Section 8.2. Some other routers, like sensors, may also require small prefixes. When randomly selected, a few /80s may easily prevent the assignment of bigger prefixes. Small prefixes should therefore be selected in neighboring areas.
For instance, given a delegated prefix 2001::/56 and an assigned prefix 2001::/64, the best prefix choice in order to reduce prefix space waste is 2001:0:0:1::/64. Other choices are then to be taken in 2001:0:0:2::/63, 2001:0:0:4::/62, 2001:0:0:8::/61, etc...
Creating an efficient prefix selection algorithm may be challenging as it needs to fullfill somehow contradictory requirements:
The following algorithm offers a satisfying tradeoff. Given a Delegated Prefix and the desired prefix length:
If RANDOM_SUBSET_SIZE equals 10, the subset would be {2001:0:0:1::/64, 2 /64s in 2001:0:0:2::/63, 4 /64s in 2001:0:0:4::/62, the 3 first /64s in 2001:0:0:8::/61}.
This algorithm, defined as a sequence of prefix sets computation, may seem algorithmicaly complex, but it can be efficiently implemented. The key element in order to do so is the ability to iterate efficiently over all the available prefixes.
RANDOM_SUBSET_SIZE should provide sufficiently low collision probability. A value of 256 should be enough in most cases. PSEUDO_RANDOM_TENTATIVE is purely implementation dependent, but shouldn't be too high as the probability of finding an available prefix that way quickly decreases with the number of used prefixes. A value of 10 should be sufficient.
When a new assignment can't be created, and if not forbidden by the router's configuration, the router MAY increase the size of the desired prefix. For instance, if an available /64 can't be found, the router may look for a /80. Nevertheless, this implies using DHCPv6 instead of SLAAC, which SHOULD be avoided.
The previously proposed solution may be useful in some particular cases, but won't work when no more prefixes are available. A router MAY try to detect when default length prefixes are becoming rare. In such a situation, it MAY decide to allocate a longer prefix, part of an available shorter prefix. For instance, if A/64 is available, but there are not many other available /64, the router can try to allocate A/80. If the allocation doesn't raise any collision, this procedure will prevent A/64 from being used by other hosts, hence creating a large set of smaller available prefixes to be used.
Such an allocation is considered dynamic. The Authoritative bit MUST NOT be set and the priority MUST be among values authorized as dynamically chosen in Section 6.8.
When different prefixes lengths are being used, the random prefix selection MUST NOT be uniform among all possibilities. Instead, it SHOULD privilege prefixes contained in bigger prefixes that cannot be allocated. For instance, if 2001::/56 is the DP, and 2001:0:0:0:1::/80 is an assigned prefix, other /80 should be randomly chosen in 2001:0:0:0:1::/64 before being chosen in other /64s.
When specifically required by an authority (configuration or DHCP), a router MAY decide to un-assign one of its own assignment, in order to cut it in smaller prefixes, or to send an overriding assignment in order to force the network to stop using a particular prefix. Because such a procedure may imply links reconfiguration, it SHOULD be avoided whenever possible.
Such allocation are considered as required by an authority. The Authoritative bit MAY be set and the priority MUST be among values authorized as specified by an authority in Section 6.8.
As an example, if a router can't find a /64 for a link that, with a high priority, must be given a /64, it chooses a prefix assigned by some other router, to another link, with a lower priority, and creates a new Chosen Prefix with a higher priority. The other router will be forced to remove its own assignment, hence making the new assignment valid.
This document is the continuation of the work being done in [I-D.arkko-homenet-prefix-assignment]. The authors would like to thank all the people that participated in the previous document's development as well as the present one. In particular, the authors would like to thank to Tim Chown, Fred Baker, Mark Townsley, Lorenzo Colitti, Ole Troan, Ray Bellis, Markus Stenberg, Wassim Haddad, Joel Halpern, Samita Chakrabarti, Michael Richardson, Anders Brandt, Erik Nordmark, Laurent Toutain, Ralph Droms, Acee Lindem and Steven Barth for interesting discussions in this problem space. The authors would also like to point out some past work in this space, such as those in [I-D.chelius-router-autoconf] or [I-D.dimitri-zospf].