xref: /freebsd/share/doc/IPv6/IMPLEMENTATION (revision 492528c051c93c3c37a5f455115eb8da8a1892ba)
1# NOTE: this is from original KAME distribution.
2# Some portion of this document is not applicable to the code merged into
3# FreeBSD-current (for example, section 5).
4
5	Implementation Note
6
7	KAME Project
8	http://www.kame.net/
9	$KAME: IMPLEMENTATION,v 1.216 2001/05/25 07:43:01 jinmei Exp $
10	$FreeBSD$
11
121. IPv6
13
141.1 Conformance
15
16The KAME kit conforms, or tries to conform, to the latest set of IPv6
17specifications.  For future reference we list some of the relevant documents
18below (NOTE: this is not a complete list - this is too hard to maintain...).
19For details please refer to specific chapter in the document, RFCs, manpages
20come with KAME, or comments in the source code.
21
22Conformance tests have been performed on past and latest KAME STABLE kit,
23at TAHI project.  Results can be viewed at http://www.tahi.org/report/KAME/.
24We also attended Univ. of New Hampshire IOL tests (http://www.iol.unh.edu/)
25in the past, with our past snapshots.
26
27RFC1639: FTP Operation Over Big Address Records (FOOBAR)
28    * RFC2428 is preferred over RFC1639.  ftp clients will first try RFC2428,
29      then RFC1639 if failed.
30RFC1886: DNS Extensions to support IPv6
31RFC1933: (see RFC2893)
32RFC1981: Path MTU Discovery for IPv6
33RFC2080: RIPng for IPv6
34    * KAME-supplied route6d, bgpd and hroute6d support this.
35RFC2283: Multiprotocol Extensions for BGP-4
36    * so-called "BGP4+".
37    * KAME-supplied bgpd supports this.
38RFC2292: Advanced Sockets API for IPv6
39    * For supported library functions/kernel APIs, see sys/netinet6/ADVAPI.
40RFC2362: Protocol Independent Multicast-Sparse Mode (PIM-SM)
41    * RFC2362 defines the packet formats and the protcol of PIM-SM.
42RFC2373: IPv6 Addressing Architecture
43    * KAME supports node required addresses, and conforms to the scope
44      requirement.
45RFC2374: An IPv6 Aggregatable Global Unicast Address Format
46    * KAME supports 64-bit length of Interface ID.
47RFC2375: IPv6 Multicast Address Assignments
48    * Userland applications use the well-known addresses assigned in the RFC.
49RFC2428: FTP Extensions for IPv6 and NATs
50    * RFC2428 is preferred over RFC1639.  ftp clients will first try RFC2428,
51      then RFC1639 if failed.
52RFC2460: IPv6 specification
53RFC2461: Neighbor discovery for IPv6
54    * See 1.2 in this document for details.
55RFC2462: IPv6 Stateless Address Autoconfiguration
56    * See 1.4 in this document for details.
57RFC2463: ICMPv6 for IPv6 specification
58    * See 1.8 in this document for details.
59RFC2464: Transmission of IPv6 Packets over Ethernet Networks
60RFC2465: MIB for IPv6: Textual Conventions and General Group
61    * Necessary statistics are gathered by the kernel.  Actual IPv6 MIB
62      support is provided as patchkit for ucd-snmp.
63RFC2466: MIB for IPv6: ICMPv6 group
64    * Necessary statistics are gathered by the kernel.  Actual IPv6 MIB
65      support is provided as patchkit for ucd-snmp.
66RFC2467: Transmission of IPv6 Packets over FDDI Networks
67RFC2472: IPv6 over PPP
68RFC2492: IPv6 over ATM Networks
69    * only PVC is supported.
70RFC2497: Transmission of IPv6 packet over ARCnet Networks
71RFC2545: Use of BGP-4 Multiprotocol Extensions for IPv6 Inter-Domain Routing
72RFC2553: Basic Socket Interface Extensions for IPv6
73    * IPv4 mapped address (3.7) and special behavior of IPv6 wildcard bind
74      socket (3.8) are,
75	- supported and turned on by default on KAME/FreeBSD[34]x
76	  and KAME/BSDI4,
77	- supported but turned off by default on KAME/NetBSD,
78	- not supported on KAME/FreeBSD228, KAME/OpenBSD and KAME/BSDI3.
79      see 1.12 in this document for details.
80RFC2671: Extension Mechanisms for DNS (EDNS0)
81    * see USAGE for how to use it.
82    * not supported on kame/freebsd4 and kame/bsdi4.
83RFC2673: Binary Labels in the Domain Name System
84    * KAME/bsdi4 supports A6, DNAME and binary label to some extent.
85    * KAME apps/bind8 repository has resolver library with partial A6, DNAME
86      and binary label support.
87RFC2675: IPv6 Jumbograms
88    * See 1.7 in this document for details.
89RFC2710: Multicast Listener Discovery for IPv6
90RFC2711: IPv6 router alert option
91RFC2732: Format for Literal IPv6 Addresses in URL's
92    * The spec is implemented in programs that handle URLs
93      (like freebsd ftpio(3) and fetch(1), or netbsd ftp(1))
94RFC2766: Network Address Translation - Protocol Translation (NAT-PT)
95    * Section 4.2 is implemented by totd (see ports/totd, or pkgsrc/net/totd).
96RFC2874: DNS Extensions to Support IPv6 Address Aggregation and Renumbering
97    * KAME/bsdi4 supports A6, DNAME and binary label to some extent.
98    * KAME apps/bind8 repository has resolver library with partial A6, DNAME
99      and binary label support.
100RFC2893: Transition Mechanisms for IPv6 Hosts and Routers
101    * IPv4 compatible address is not supported.
102    * automatic tunneling (4.3) is not supported.
103    * "gif" interface implements IPv[46]-over-IPv[46] tunnel in a generic way,
104      and it covers "configured tunnel" described in the spec.
105      See 1.5 in this document for details.
106RFC2894: Router renumbering for IPv6
107RFC3041: Privacy Extensions for Stateless Address Autoconfiguration in IPv6
108RFC3056: Connection of IPv6 Domains via IPv4 Clouds
109    * So-called "6to4".
110    * "stf" interface implements it.  Be sure to read
111      draft-itojun-ipv6-transition-abuse-01.txt
112      below before configuring it, there can be security issues.
113draft-ietf-ipngwg-icmp-name-lookups-07: IPv6 Name Lookups Through ICMP
114draft-ietf-dhc-dhcpv6-15.txt: DHCPv6
115draft-ietf-dhc-dhcpv6exts-12.txt: Extensions for DHCPv6
116    * kame/dhcp6 has test implementation, which will not be compiled in
117      default compilation.
118    * 15/12 drafts are not explicit about padding and string termination.
119      at IETF48, the author confirmed that there's no padding/termination
120      (and extensions can appear unaligned).  our code follows the comment.
121draft-itojun-ipv6-tcp-to-anycast-00.txt:
122	Disconnecting TCP connection toward IPv6 anycast address
123draft-ietf-ipngwg-rfc2553bis-03.txt:
124	Basic Socket Interface Extensions for IPv6 (revised)
125draft-ietf-ipngwg-rfc2292bis-02.txt:
126	Advanced Sockets API for IPv6 (revised)
127    * Some of the updates in the draft are not implemented yet.  See
128      TODO.2292bis for more details.
129draft-ietf-mobileip-ipv6-13.txt: Mobility Support in IPv6
130    * See section 6.
131draft-ietf-ngtrans-tcpudp-relay-04.txt:
132	An IPv6-to-IPv4 transport relay translator
133    * FAITH tcp relay translator (faithd) implements this.  See 3.1 for more
134      details.
135draft-ietf-ipngwg-router-selection-01.txt:
136	Default Router Preferences and More-Specific Routes
137    * router-side only.
138draft-ietf-ipngwg-scoping-arch-02.txt:
139	The architecture, text representation, and usage of IPv6
140	scoped addresses.
141    * some part of the documentation (especially about the routing
142      model) is not supported yet.
143draft-ietf-pim-sm-v2-new-02.txt
144	A revised version of RFC2362, which includes the IPv6 specific
145	packet format and protocol descriptions.
146draft-ietf-dnsext-mdns-00.txt: Multicast DNS
147    * kame/mdnsd has test implementation, which will not be built in
148      default compilation.  The draft will experience a major change in the
149      near future, so don't rely upon it.
150draft-itojun-ipv6-transition-abuse-02.txt:
151	Possible abuse against IPv6 transition technologies (expired)
152    * KAME does not implement RFC1933/2893 automatic tunnel.
153    * "stf" interface implements some address filters.  Refer to stf(4)
154      for details.  Since there's no way to make 6to4 interface 100% secure,
155      we do not include "stf" interface into GENERIC.v6 compilation.
156    * kame/openbsd completely disables IPv4 mapped address support.
157    * kame/netbsd makes IPv4 mapped address support off by default.
158    * See section 1.12.6 and 1.14 for more details.
159draft-itojun-ipv6-tclass-api-02.txt: Socket API for IPv6 traffic class field
160draft-itojun-ipv6-flowlabel-api-01.txt: Socket API for IPv6 flow label field
161    * no consideration is made against the use of routing headers and such.
162
1631.2 Neighbor Discovery
164
165Neighbor Discovery is fairly stable.  Currently Address Resolution,
166Duplicated Address Detection, and Neighbor Unreachability Detection
167are supported.  In the near future we will be adding Unsolicited Neighbor
168Advertisement transmission command as admin tool.
169
170Duplicated Address Detection (DAD) will be performed when an IPv6 address
171is assigned to a network interface, or the network interface is enabled
172(ifconfig up).  It is documented in RFC2462 5.4.
173If DAD fails, the address will be marked "duplicated" and message will be
174generated to syslog (and usually to console).  The "duplicated" mark
175can be checked with ifconfig.  It is administrators' responsibility to check
176for and recover from DAD failures.  We may try to improve failure recovery
177in future KAME code.
178DAD procedure may not be effective on certain network interfaces/drivers.
179If a network driver needs long initialization time (with wireless network
180interfaces this situation is popular), and the driver mistakingly raises
181IFF_RUNNING before the driver becomes ready, DAD code will try to transmit
182DAD probes to not-really-ready network driver and the packet will not go out
183from the interface.  In such cases, network drivers should be corrected.
184
185Some of network drivers loop multicast packets back to themselves,
186even if instructed not to do so (especially in promiscuous mode).
187In such cases DAD may fail, because DAD engine sees inbound NS packet
188(actually from the node itself) and considers it as a sign of duplicate.
189In this case, drivers should be corrected to honor IFF_SIMPLEX behavior.
190For example, you may need to check source MAC address on an inbound packet,
191and reject it if it is from the node itself.
192You may also want to look at #if condition marked "heuristics" in
193sys/netinet6/nd6_nbr.c:nd6_dad_timer() as workaround (note that the code
194fragment in "heuristics" section is not spec conformant).
195
196Neighbor Discovery specification (RFC2461) does not talk about neighbor
197cache handling in the following cases:
198(1) when there was no neighbor cache entry, node received unsolicited
199    RS/NS/NA/redirect packet without link-layer address
200(2) neighbor cache handling on medium without link-layer address
201    (we need a neighbor cache entry for IsRouter bit)
202For (1), we implemented workaround based on discussions on IETF ipngwg mailing
203list.  For more details, see the comments in the source code and email
204thread started from (IPng 7155), dated Feb 6 1999.
205
206IPv6 on-link determination rule (RFC2461) is quite different from assumptions
207in BSD IPv4 network code.  To implement behavior in RFC2461 section 5.2
208(when default router list is empty), the kernel needs to know the default
209outgoing interface.  To configure the default outgoing interface, use
210commands like "ndp -I de0" as root.  Note that the spec misuse the word
211"host" and "node" in several places in the section.
212
213To avoid possible DoS attacks and infinite loops, KAME stack will accept
214only 10 options on ND packet.  Therefore, if you have 20 prefix options
215attached to RA, only the first 10 prefixes will be recognized.
216If this troubles you, please contact KAME team and/or modify
217nd6_maxndopt in sys/netinet6/nd6.c.  If there are high demands we may
218provide sysctl knob for the variable.
219
220Proxy Neighbor Advertisement support is implemented in the kernel.
221For instance, you can configure it by using the following command:
222	# ndp -s fe80::1234%ne0 0:1:2:3:4:5 proxy
223where ne0 is the interface which attaches to the same link as the
224proxy target.
225There are certain limitations, though:
226- It does not send unsolicited multicast NA on configuration.  This is MAY
227  behavior in RFC2461.
228- It does not add random delay before transmission of solicited NA.  This is
229  SHOULD behavior in RFC2461.
230- We cannot configure proxy NDP for off-link address.  The target address for
231  proxying must be link-local address, or must be in prefixes configured to
232  node which does proxy NDP.
233- RFC2461 is unclear about if it is legal for a host to perform proxy ND.
234  We do not prohibit hosts from doing proxy ND, but there will be very limited
235  use in it.
236
237Starting mid March 2000, we support Neighbor Unreachability Detection (NUD)
238on p2p interfaces, including tunnel interfaces (gif).  NUD is turned on by
239default.  Before March 2000 KAME stack did not perform NUD on p2p interfaces.
240If the change raises any interoperability issues, you can turn off/on NUD
241by per-interface basis.  Use "ndp -i interface -nud" to turn it off.
242Consult ndp(8) for details.
243
244RFC2461 specifies upper-layer reachability confirmation hint.  Whenever
245upper-layer reachability confirmation hint comes, ND process can use it
246to optimize neighbor discovery process - ND process can omit real ND exchange
247and keep the neighbor cache state in REACHABLE.
248We currently have two sources for hints: (1) setsockopt(IPV6_REACHCONF)
249defined by 2292bis API, and (2) hints from tcp_input.
250It is questionable if they are really trustworthy.  For example, a rogue
251userland program can use IPV6_REACHCONF to confuse ND process.  Neighbor
252cache is a system-wide information pool, and it is bad to allow single process
253to affect others.  Also, tcp_input can be hosed by hijack attempts.  It is
254wrong to allow hijack attempts to affect ND process.
255Starting June 2000, ND code has a protection mechanism against incorrect
256upper-layer reachability confirmation.  ND code counts subsequent upper-layer
257hints.  If the number of hints reaches maximum, ND code will ignore further
258upper-layer hints and run real ND process to confirm reachability to the peer.
259sysctl net.inet6.icmp6.nd6_maxnudhint defines maximum # of subsequent
260upper-layer hints to be accepted.
261(from April 2000 to June 2000, we rejected setsockopt(IPV6_REACHCONF) from
262non-root process - after local discussion, it looks that hints are not
263that trustworthy even if they are from privileged processes)
264
265If inbound ND packets carry invalid values, the KAME kernel will
266drop these packet and increment statistics variable.  See
267"netstat -sn", icmp6 section.  For detailed debugging session, you can
268turn on syslog output from the kernel on errors, by turning on sysctl MIB
269net.inet6.icmp6.nd6_debug.  nd6_debug can be turned on at bootstrap
270time, by defining ND6_DEBUG kernel compilation option (so you can
271debug behavior during bootstrap).  nd6_debug configuration should
272only be used for test/debug purposes - for production environment,
273nd6_debug must be set to 0.  If you leave it to 1, malicious parties
274can inject broken packet and fill up /var/log partition.
275
2761.3 Scope Zone Index
277
278IPv6 uses scoped addresses.  It is therefore very important to
279specify the scope zone index (link index for a link-local address, or
280site index for a site-local address) with an IPv6 address.  Without a
281zone index, a scoped IPv6 address is ambiguous to the kernel, and
282the kernel would not be able to determine the outbound link for a
283packet to the scoped address.  KAME code tries to address the issue in
284several ways.
285
286The entire architecture of scoped addresses is documented in
287draft-ietf-ipngwg-scoping-arch-xx.txt.  One non-trivial point of the
288architecture is that the link scope is (theoretically) larger than the
289interface scope.  That is, two different interfaces can belong to a
290same single link.  However, in a normal operation, we can assume that
291there is 1-to-1 relationship between links and interfaces.  In
292other words, we can usually put links and interfaces in the same scope
293type.  The current KAME implementation assumes the 1-to-1
294relationship.  In particular, we use interface names such as "ne1" as
295unique link identifiers.  This would be much more human-readable and
296intuitive than numeric identifiers, but please keep your mind on the
297theoretical difference between links and interfaces.
298
299Site-local addresses are very vaguely defined in the specs, and both
300the specification and the KAME code need tons of improvements to
301enable its actual use.  For example, it is still very unclear how we
302define a site, or how we resolve host names in a site.  There is work
303underway to define behavior of routers at site border, but, we have
304almost no code for site boundary node support (both forwarding nor
305routing) and we bet almost noone has.  We recommend, at this moment,
306you to use global addresses for experiments - there are way too many
307pitfalls if you use site-local addresses.
308
3091.3.1 Kernel internal
310
311In the kernel, the link index for a link-local scope address is
312embedded into the 2nd 16bit-word (the 3rd and 4th bytes) in the IPv6
313address.
314For example, you may see something like:
315	fe80:1::200:f8ff:fe01:6317
316in the routing table and the interface address structure (struct
317in6_ifaddr).  The address above is a link-local unicast address which
318belongs to a network link whose link identifier is 1 (note that it
319eqauls to the interface index by the assumption of our
320implementation).  The embedded index enables us to identify IPv6
321link-local addresses over multiple links effectively and with only a
322little code change.
323
3241.3.2 Interaction with API
325
326There are several candidates of API to deal with scoped addresses
327without ambiguity.
328
329The IPV6_PKTINFO ancillary data type or socket option defined in the
330advanced API (RFC2292 or draft-ietf-ipngwg-rfc2292bis-xx) can specify
331the outgoing interface of a packet.  Similarly, the IPV6_PKTINFO or
332IPV6_RECVPKTINFO socket options tell kernel to pass the incoming
333interface to user applications.
334
335These options are enough to disambiguate scoped addresses of an
336incoming packet, because we can uniquely identify the corresponding
337zone of the scoped address(es) by the incoming interface.  However,
338they are too strong for outgoing packets.  For example, consider a
339multi-sited node and suppose that more than one interface of the node
340belongs to a same site.  When we want to send a packet to the site,
341we can only specify one of the interfaces for the outgoing packet with
342these options; we cannot just say "send the packet to (one of the
343interfaces of) the site."
344
345Another kind of candidates is to use the sin6_scope_id member in the
346sockaddr_in6 structure, defined in RFC2553 and
347draft-ietf-ipngwg-rfc2553bis-xx.txt.  The KAME kernel interprets the
348sin6_scope_id field properly in order to disambiguate scoped
349addresses.  For example, if an application passes a sockaddr_in6
350structure that has a non-zero sin6_scope_id value to the sendto(2)
351system call, the kernel should send the packet to the appropriate zone
352according to the sin6_scope_id field.  Similarly, when the source or
353the destination address of an incoming packet is a scoped one, the
354kernel should detect the correct zone identifier based on the address
355and the receiving interface, fill the identifier in the sin6_scope_id
356field of a sockaddr_in6 structure, and then pass the packet to an
357application via the recvfrom(2) system call, etc.
358
359However, the semantics of the sin6_scope_id is still vague and on the
360way to standardization.  Additionally, not so many operating systems
361support the behavior above at this moment.
362
363In summary,
364- If your target system is limited to KAME based ones (i.e. BSD
365  variants and KAME snaps), use the sin6_scope_id field assuming the
366  kernel behavior described above.
367- Otherwise, (i.e. if your program should be portable on other systems
368  than BSDs)
369  + Use the advanced API to disambiguate scoped addresses of incoming
370    packets.
371  + To disambiguate scoped addresses of outgoing packets,
372    * if it is okay to just specify the outgoing interface, use the
373      advanced API.  This would be the case, for example, when you
374      should only consider link-local addresses and your system
375      assumes 1-to-1 relationship between links and interfaces.
376    * otherwise, sorry but you lose.  Please rush the IETF IPv6
377      community into standardizing the semantics of the sin6_scope_id
378      field.
379
380Routing daemons and configuration programs, like route6d and ifconfig,
381will need to manipulate the "embedded" zone index.  These programs use
382routing sockets and ioctls (like SIOCGIFADDR_IN6) and the kernel API
383will return IPv6 addresses with the 2nd 16bit-word filled in.  The
384APIs are for manipulating kernel internal structure.  Programs that
385use these APIs have to be prepared about differences in kernels
386anyway.
387
388getaddrinfo(3) and getnameinfo(3) support an extended numeric IPv6
389syntax, as documented in draft-ietf-ipngwg-rfc2553bis-xx.txt.  You can
390specify the outgoing link, by using the name of the outgoing interface
391as the link, like "fe80::1%ne0" (again, note that we assume there is
3921-to-1 relationship between links and interfaces.)  This way you will
393be able to specify a link-local scoped address without much trouble.
394
395Other APIs like inet_pton(3) and inet_ntop(3) are inherently
396unfriendly with scoped addresses, since they are unable to annotate
397addresses with zone identifier.
398
3991.3.3 Interaction with users (command line)
400
401Most of user applications now support the extended numeric IPv6
402syntax.  In this case, you can specify outgoing link, by using the name
403of the outgoing interface like "fe80::1%ne0" (sorry for the duplicated
404notice, but please recall again that we assume 1-to-1 relationship
405between links and interfaces).  This is even the case for some
406management tools such as route(8) or ndp(8).  For example, to install
407the IPv6 default route by hand, you can type like
408	# route add -inet6 default fe80::9876:5432:1234:abcd%ne0
409(Although we suggest you to run dynamic routing instead of static
410routes, in order to avoid configuration mistakes.)
411
412Some applications have command line options for specifying an
413appropriate zone of a scoped address (like "ping6 -I ne0 ff02::1" to
414specify the outgoing interface).  However, you can't always expect such
415options.  Thus, we recommend you to use the extended format described
416above.
417
418In any case, when you specify a scoped address to the command line,
419NEVER write the embedded form (such as ff02:1::1 or fe80:2::fedc),
420which should only be used inside the kernel (see Section 1.3.1), and
421is not supposed to work.
422
4231.4 Plug and Play
424
425The KAME kit implements most of the IPv6 stateless address
426autoconfiguration in the kernel.
427Neighbor Discovery functions are implemented in the kernel as a whole.
428Router Advertisement (RA) input for hosts is implemented in the
429kernel.  Router Solicitation (RS) output for endhosts, RS input
430for routers, and RA output for routers are implemented in the
431userland.
432
4331.4.1 Assignment of link-local, and special addresses
434
435IPv6 link-local address is generated from IEEE802 address (ethernet MAC
436address).  Each of interface is assigned an IPv6 link-local address
437automatically, when the interface becomes up (IFF_UP).  Also, direct route
438for the link-local address is added to routing table.
439
440Here is an output of netstat command:
441
442Internet6:
443Destination                   Gateway                   Flags      Netif Expire
444fe80::%ed0/64                 link#1                    UC           ed0
445fe80::%ep0/64                 link#2                    UC           ep0
446
447Interfaces that has no IEEE802 address (pseudo interfaces like tunnel
448interfaces, or ppp interfaces) will borrow IEEE802 address from other
449interfaces, such as ethernet interfaces, whenever possible.
450If there is no IEEE802 hardware attached, last-resort pseudorandom value,
451which is from MD5(hostname), will be used as source of link-local address.
452If it is not suitable for your usage, you will need to configure the
453link-local address manually.
454
455If an interface is not capable of handling IPv6 (such as lack of multicast
456support), link-local address will not be assigned to that interface.
457See section 2 for details.
458
459Each interface joins the solicited multicast address and the
460link-local all-nodes multicast addresses (e.g.  fe80::1:ff01:6317
461and ff02::1, respectively, on the link the interface is attached).
462In addition to a link-local address, the loopback address (::1) will be
463assigned to the loopback interface.  Also, ::1/128 and ff01::/32 are
464automatically added to routing table, and loopback interface joins
465node-local multicast group ff01::1.
466
4671.4.2 Stateless address autoconfiguration on hosts
468
469In IPv6 specification, nodes are separated into two categories:
470routers and hosts.  Routers forward packets addressed to others, hosts does
471not forward the packets.  net.inet6.ip6.forwarding defines whether this
472node is a router or a host (router if it is 1, host if it is 0).
473
474It is NOT recommended to change net.inet6.ip6.forwarding while the node
475is in operation.   IPv6 specification defines behavior for "host" and "router"
476quite differently, and switching from one to another can cause serious
477troubles.  It is recommended to configure the variable at bootstrap time only.
478
479The first step in stateless address configuration is Duplicated Address
480Detection (DAD).  See 1.2 for more detail on DAD.
481
482When a host hears Router Advertisement from the router, a host may
483autoconfigure itself by stateless address autoconfiguration.
484This behavior can be controlled by net.inet6.ip6.accept_rtadv
485(host autoconfigures itself if it is set to 1).
486By autoconfiguration, network address prefix for the receiving interface
487(usually global address prefix) is added. The default route is also
488configured.
489
490Routers periodically generate Router Advertisement packets.  To
491request an adjacent router to generate RA packet, a host can transmit
492Router Solicitation.  To generate an RS packet at any time, use the
493"rtsol" command. The "rtsold" daemon is also available. "rtsold"
494generates Router Solicitation whenever necessary, and it works great
495for nomadic usage (notebooks/laptops).  If one wishes to ignore Router
496Advertisements, use sysctl to set net.inet6.ip6.accept_rtadv to 0.
497
498To generate Router Advertisement from a router, use the "rtadvd" daemon.
499
500Note that the IPv6 specification assumes the following items and that
501nonconforming cases are left unspecified:
502- Only hosts will listen to router advertisements
503- Hosts have single network interface (except loopback)
504This is therefore unwise to enable net.inet6.ip6.accept_rtadv on routers,
505or multi-interface host.  A misconfigured node can behave strange
506(KAME code allows nonconforming configuration, for those who would like
507to do some experiments).
508
509To summarize the sysctl knob:
510	accept_rtadv	forwarding	role of the node
511	---		---		---
512	0		0		host (to be manually configured)
513	0		1		router
514	1		0		autoconfigured host
515					(spec assumes that host has single
516					interface only, autoconfigred host with
517					multiple interface is out-of-scope)
518	1		1		invalid, or experimental
519					(out-of-scope of spec)
520
521See 1.2 in the document for relationship between DAD and autoconfiguration.
522
5231.4.3 DHCPv6
524
525We supply a tiny DHCPv6 server/client in kame/dhcp6. However, the
526implementation is premature (for example, this does NOT implement
527address lease/release), and it is not in default compilation tree on
528some platforms. If you want to do some experiment, compile it on your
529own.
530
531DHCPv6 and autoconfiguration also needs more work.  "Managed" and "Other"
532bits in RA have no special effect to stateful autoconfiguration procedure
533in DHCPv6 client program ("Managed" bit actually prevents stateless
534autoconfiguration, but no special action will be taken for DHCPv6 client).
535
5361.5 Generic tunnel interface
537
538GIF (Generic InterFace) is a pseudo interface for configured tunnel.
539Details are described in gif(4) manpage.
540Currently
541	v6 in v6
542	v6 in v4
543	v4 in v6
544	v4 in v4
545are available.  Use "gifconfig" to assign physical (outer) source
546and destination address to gif interfaces.
547Configuration that uses same address family for inner and outer IP
548header (v4 in v4, or v6 in v6) is dangerous.  It is very easy to
549configure interfaces and routing tables to perform infinite level
550of tunneling.  Please be warned.
551
552gif can be configured to be ECN-friendly.  See 4.5 for ECN-friendliness
553of tunnels, and gif(4) manpage for how to configure.
554
555If you would like to configure an IPv4-in-IPv6 tunnel with gif interface,
556read gif(4) carefully.  You may need to remove IPv6 link-local address
557automatically assigned to the gif interface.
558
5591.6 Source Address Selection
560
561KAME's source address selection takes care of the following
562conditions:
563- address scope
564- outgoing interface
565- whether an address is deprecated
566- whether an address is temporary (in terms of RFC 3041)
567- prefix matching against the destination
568
569Roughly speaking, the selection policy is as follows:
570- always use an address that belongs to the same scope zone as the
571  destination.
572- addresses that have equal or larger scope than the scope of the
573  destination are preferred.
574- a deprecated address is not used in new communications if an
575  alternate (non-deprecated) address is available and has sufficient
576  scope.
577- a temporary address (in terms of RFC 3041 privacy extension) are
578  preferred to a public address.
579- if none of above conditions tie-breaks, addresses assigned on the
580  outgoing interface are preferred.
581- if none of above conditions tie-breaks, one which is longest prefix
582  matching against the destination is preferred as the last resort.
583
584For instance, ::1 is selected for ff01::1,
585fe80::200:f8ff:fe01:6317%ne0 for fe80::2a0:24ff:feab:839b%ne0.
586To see how longest-matching works, suppose that
5873ffe:501:808:1:200:f8ff:fe01:6317 and 3ffe:2001:9:124:200:f8ff:fe01:6317
588are given on the outgoing interface. Then the former is chosen as the
589source for the destination 3ffe:501:800::1. Note that even if all
590available addresses have smaller scope than the scope of the
591destination, we choose one anyway. For example, if we have link-local
592and site-local addresses only, we choose a site-local addresses for a
593global destination. If the packet is going to break a site boundary,
594the boundary router will return an ICMPv6 destination unreachable
595error with code 2 - beyond scope of source address.
596
597The precise desripction of the algorithm is quite complicated. To
598describe the algorithm, we introduce the following notation:
599
600For a given destination D,
601  samescope(D): The set of addresses that have the same scope as D.
602  largerscope(D): The set of addresses that have a larger scope than D.
603  smallerscope(D): The set of addresses that have a smaller scope than D.
604
605For a given set of addresses A,
606  DEP(A): the set of deprecated addresses in A.
607  nonDEP(A): A - DEP(A).
608
609For a given set of addresses A,
610  tmp(A): the set of preferred temporary-autoconfigured or
611          manually-configure addresses in A.
612
613Also, the algorithm assumes that the outgoing interface for the
614destination D is determined. We call the interface "I".
615
616The algorithm is as follows. Selection proceeds step by step as
617described; For example, if an address is selected by item 1, item 2 and
618later are not considered at all.
619
620  0. If there is no address in the same scope zone as D, just give up;
621     the packet will not be sent.
622  1. If we do not prefer temporary addresses, go to 3.
623     Otherwise, and if tmp(samescope(D)) is not empty,
624     then choose an address that is on the interface I.  If every
625     address is on I, or every address is on a different interface
626     from I, choose an arbitrary one provided that an address longest
627     matching against D is always preferred.
628  2. If tmp(largerscope(D)) is not empty,
629     then choose an address that has the smallest scope. If more than one
630     address has the smallest scope, choose an arbitrary one provided
631     that addresses on I are always preferred.
632  3. If nonDEP(samescope(D)) is not empty,
633     then apply the same logic as of 1.
634  4. If nonDEP(largerscope(D)) is not empty,
635     then apply the same logic as of 2.
636  5. If we do not prefer temporary addresses, go to 7.
637     Otherwise, and if tmp(DEP(samescope(D))) is not empty,
638     then choose an address that is on the interface I.  If every
639     address is on I, or every address is on a different interface
640     from I, choose an arbitrary one provided that an address longest
641     matching against D is always preferred.
642  6. If tmp(DEP(largerscope(D))) is not empty,
643     then choose an address that has the smallest scope. If more than
644     one address has the smallest scope, choose an arbitrary one provided
645     that an address on I is always preferred.
646  7. If DEP(samescope(D)) is not empty,
647     then apply the same logic as of 5.
648  8. If DEP(largerscope(D)) is not empty,
649     then apply the same logic as of 6.
650  9. If we do not prefer temporary addresses, go to 11.
651     Otherwise, and if tmp(nonDEP(smallerscope(D))) is not empty,
652     then choose an address that has the largest scope. If more than
653     one address has the largest scope, choose an arbitrary one provided
654     that an address on I is always preferred.
655 10. If tmp(DEP(smallerscope(D))) is not empty,
656     then choose an address that has the largest scope. If more than
657     one address has the largest scope, choose an arbitrary one provided
658     that an address on I is always preferred.
659 11. If nonDEP(smallerscope(D)) is not empty,
660     then apply the same logic as of 9.
661 12. If DEP(smallerscope(D)) is not empty,
662     then apply the same logic as of 10.
663
664There exists a document about source address selection
665(draft-ietf-ipngwg-default-addr-select-xx.txt). KAME's algorithm
666described above takes a similar approach to the document, but there
667are some differences. See the document for more details.
668
669There are some cases where we do not use the above rule.  One
670example is connected TCP session, and we use the address kept in TCP
671protocol control block (tcb) as the source.
672Another example is source address for Neighbor Advertisement.
673Under the spec (RFC2461 7.2.2) NA's source should be the target
674address of the corresponding NS's target.  In this case we follow
675the spec rather than the above longest-match rule.
676
677If you would like to prohibit the use of deprecated address for some
678reason, configure net.inet6.ip6.use_deprecated to 0.  The issue
679related to deprecated address is described in RFC2462 5.5.4 (NOTE:
680there is some debate underway in IETF ipngwg on how to use
681"deprecated" address).
682
6831.7 Jumbo Payload
684
685KAME supports the Jumbo Payload hop-by-hop option used to send IPv6
686packets with payloads longer than 65,535 octets.  But since currently
687KAME does not support any physical interface whose MTU is more than
68865,535, such payloads can be seen only on the loopback interface(i.e.
689lo0).
690
691If you want to try jumbo payloads, you first have to reconfigure the
692kernel so that the MTU of the loopback interface is more than 65,535
693bytes; add the following to the kernel configuration file:
694	options		"LARGE_LOMTU"		#To test jumbo payload
695and recompile the new kernel.
696
697Then you can test jumbo payloads by the ping6 command with -b and -s
698options.  The -b option must be specified to enlarge the size of the
699socket buffer and the -s option specifies the length of the packet,
700which should be more than 65,535.  For example, type as follows;
701	% ping6 -b 70000 -s 68000 ::1
702
703The IPv6 specification requires that the Jumbo Payload option must not
704be used in a packet that carries a fragment header.  If this condition
705is broken, an ICMPv6 Parameter Problem message must be sent to the
706sender.  KAME kernel follows the specification, but you cannot usually
707see an ICMPv6 error caused by this requirement.
708
709If KAME kernel receives an IPv6 packet, it checks the frame length of
710the packet and compares it to the length specified in the payload
711length field of the IPv6 header or in the value of the Jumbo Payload
712option, if any.  If the former is shorter than the latter, KAME kernel
713discards the packet and increments the statistics. You can see the
714statistics as output of netstat command with `-s -p ip6' option:
715	% netstat -s -p ip6
716	ip6:
717		(snip)
718		1 with data size < data length
719
720So, KAME kernel does not send an ICMPv6 error unless the erroneous
721packet is an actual Jumbo Payload, that is, its packet size is more
722than 65,535 bytes.  As described above, KAME kernel currently does not
723support physical interface with such a huge MTU, so it rarely returns an
724ICMPv6 error.
725
726TCP/UDP over jumbogram is not supported at this moment.  This is because
727we have no medium (other than loopback) to test this.  Contact us if you
728need this.
729
730IPsec does not work on jumbograms.  This is due to some specification twists
731in supporting AH with jumbograms (AH header size influences payload length,
732and this makes it real hard to authenticate inbound packet with jumbo payload
733option as well as AH).
734
735There are fundamental issues in *BSD support for jumbograms.  We would like to
736address those, but we need more time to finalize the task.  To name a few:
737- mbuf pkthdr.len field is typed as "int" in 4.4BSD, so it cannot hold
738  jumbogram with len > 2G on 32bit architecture CPUs.  If we would like to
739  support jumbogram properly, the field must be expanded to hold 4G +
740  IPv6 header + link-layer header.  Therefore, it must be expanded to at least
741  int64_t (u_int32_t is NOT enough).
742- We mistakingly use "int" to hold packet length in many places.  We need
743  to convert them into larger numeric type.  It needs a great care, as we may
744  experience overflow during packet length computation.
745- We mistakingly check for ip6_plen field of IPv6 header for packet payload
746  length in various places.  We should be checking mbuf pkthdr.len instead.
747  ip6_input() will perform sanity check on jumbo payload option on input,
748  and we can safely use mbuf pkthdr.len afterwards.
749- TCP code needs careful updates in bunch of places, of course.
750
7511.8 Loop prevention in header processing
752
753IPv6 specification allows arbitrary number of extension headers to
754be placed onto packets.  If we implement IPv6 packet processing
755code in the way BSD IPv4 code is implemented, kernel stack may
756overflow due to long function call chain.  KAME sys/netinet6 code
757is carefully designed to avoid kernel stack overflow.  Because of
758this, KAME sys/netinet6 code defines its own protocol switch
759structure, as "struct ip6protosw" (see netinet6/ip6protosw.h).
760
761In addition to this, we restrict the number of extension headers
762(including the IPv6 header) in each incoming packet, in order to
763prevent a DoS attack that tries to send packets with a massive number
764of extension headers.  The upper limit can be configured by the sysctl
765value net.inet6.ip6.hdrnestlimit. In particular, if the value is 0,
766the node will allow an arbitrary number of headers. As of writing this
767document, the default value is 50.
768
769IPv4 part (sys/netinet) remains untouched for compatibility.
770Because of this, if you receive IPsec-over-IPv4 packet with massive
771number of IPsec headers, kernel stack may blow up.  IPsec-over-IPv6 is okay.
772
7731.9 ICMPv6
774
775After RFC2463 was published, IETF ipngwg has decided to disallow ICMPv6 error
776packet against ICMPv6 redirect, to prevent ICMPv6 storm on a network medium.
777KAME already implements this into the kernel.
778
779RFC2463 requires rate limitation for ICMPv6 error packets generated by a
780node, to avoid possible DoS attacks.  KAME kernel implements two rate-
781limitation mechanisms, tunable via sysctl:
782- Minimum time interval between ICMPv6 error packets
783	KAME kernel will generate no more than one ICMPv6 error packet,
784	during configured time interval.  net.inet6.icmp6.errratelimit
785	controls the interval (default: disabled).
786- Maximum ICMPv6 error packet-per-second
787	KAME kernel will generate no more than the configured number of
788	packets in one second.  net.inet6.icmp6.errppslimit controls the
789	maximum packet-per-second value (default: 200pps)
790Basically, we need to pick values that are suitable against the bandwidth
791of link layer devices directly attached to the node.  In some cases the
792default values may not fit well.  We are still unsure if the default value
793is sane or not.  Comments are welcome.
794
7951.10 Applications
796
797For userland programming, we support IPv6 socket API as specified in
798RFC2553, RFC2292 and upcoming internet drafts.
799
800TCP/UDP over IPv6 is available and quite stable.  You can enjoy "telnet",
801"ftp", "rlogin", "rsh", "ssh", etc.  These applications are protocol
802independent.  That is, they automatically chooses IPv4 or IPv6
803according to DNS.
804
8051.11 Kernel Internals
806
807 (*) TCP/UDP part is handled differently between operating system platforms.
808     See 1.12 for details.
809
810The current KAME has escaped from the IPv4 netinet logic.  While
811ip_forward() calls ip_output(), ip6_forward() directly calls
812if_output() since routers must not divide IPv6 packets into fragments.
813
814ICMPv6 should contain the original packet as long as possible up to
8151280.  UDP6/IP6 port unreach, for instance, should contain all
816extension headers and the *unchanged* UDP6 and IP6 headers.
817So, all IP6 functions except TCP6 never convert network byte
818order into host byte order, to save the original packet.
819
820tcp6_input(), udp6_input() and icmp6_input() can't assume that IP6
821header is preceding the transport headers due to extension
822headers.  So, in6_cksum() was implemented to handle packets whose IP6
823header and transport header is not continuous.  TCP/IP6 nor UDP/IP6
824header structure don't exist for checksum calculation.
825
826To process IP6 header, extension headers and transport headers easily,
827KAME requires network drivers to store packets in one internal mbuf or
828one or more external mbufs.  A typical old driver prepares two
829internal mbufs for 100 - 208 bytes data, however, KAME's reference
830implementation stores it in one external mbuf.
831
832"netstat -s -p ip6" tells you whether or not your driver conforms
833KAME's requirement.  In the following example, "cce0" violates the
834requirement. (For more information, refer to Section 2.)
835
836        Mbuf statistics:
837                317 one mbuf
838                two or more mbuf::
839                        lo0 = 8
840			cce0 = 10
841                3282 one ext mbuf
842                0 two or more ext mbuf
843
844Each input function calls IP6_EXTHDR_CHECK in the beginning to check
845if the region between IP6 and its header is
846continuous.  IP6_EXTHDR_CHECK calls m_pullup() only if the mbuf has
847M_LOOP flag, that is, the packet comes from the loopback
848interface.  m_pullup() is never called for packets coming from physical
849network interfaces.
850
851TCP6 reassembly makes use of IP6 header to store reassemble
852information.  IP6 is not supposed to be just before TCP6, so
853ip6tcpreass structure has a pointer to TCP6 header.  Of course, it has
854also a pointer back to mbuf to avoid m_pullup().
855
856Like TCP6, both IP and IP6 reassemble functions never call m_pullup().
857
858xxx_ctlinput() calls in_mrejoin() on PRC_IFNEWADDR.  We think this is
859one of 4.4BSD implementation flaws.  Since 4.4BSD keeps ia_multiaddrs
860in in_ifaddr{}, it can't use multicast feature if the interface has no
861unicast address.  So, if an application joins to an interface and then
862all unicast addresses are removed from the interface, the application
863can't send/receive any multicast packets.  Moreover, if a new unicast
864address is assigned to the interface, in_mrejoin() must be called.
865KAME's interfaces, however, have ALWAYS one link-local unicast
866address.  These extensions have thus not been implemented in KAME.
867
8681.12 IPv4 mapped address and IPv6 wildcard socket
869
870RFC2553 describes IPv4 mapped address (3.7) and special behavior
871of IPv6 wildcard bind socket (3.8).  The spec allows you to:
872- Accept IPv4 connections by AF_INET6 wildcard bind socket.
873- Transmit IPv4 packet over AF_INET6 socket by using special form of
874  the address like ::ffff:10.1.1.1.
875but the spec itself is very complicated and does not specify how the
876socket layer should behave.
877Here we call the former one "listening side" and the latter one "initiating
878side", for reference purposes.
879
880Almost all KAME implementations treat tcp/udp port number space separately
881between IPv4 and IPv6.  You can perform wildcard bind on both of the address
882families, on the same port.
883
884There are some OS-platform differences in KAME code, as we use tcp/udp
885code from different origin.  The following table summarizes the behavior.
886
887		listening side		initiating side
888		(AF_INET6 wildcard	(connection to ::ffff:10.1.1.1)
889		socket gets IPv4 conn.)
890		---			---
891KAME/BSDI3	not supported		not supported
892KAME/FreeBSD228	not supported		not supported
893KAME/FreeBSD3x	configurable		supported
894		default: enabled
895KAME/FreeBSD4x	configurable		supported
896		default: enabled
897KAME/NetBSD	configurable		supported
898		default: disabled
899KAME/BSDI4	enabled			supported
900KAME/OpenBSD	not supported		not supported
901
902The following sections will give you more details, and how you can
903configure the behavior.
904
905Comments on listening side:
906
907It looks that RFC2553 talks too little on wildcard bind issue,
908specifically on (1) port space issue, (2) failure mode, (3) relationship
909between AF_INET/INET6 wildcard bind like ordering constraint, and (4) behavior
910when conflicting socket is opened/closed.  There can be several separate
911interpretation for this RFC which conform to it but behaves differently.
912So, to implement portable application you should assume nothing
913about the behavior in the kernel.  Using getaddrinfo() is the safest way.
914Port number space and wildcard bind issues were discussed in detail
915on ipv6imp mailing list, in mid March 1999 and it looks that there's
916no concrete consensus (means, up to implementers).  You may want to
917check the mailing list archives.
918We supply a tool called "bindtest" that explores the behavior of
919kernel bind(2).  The tool will not be compiled by default.
920
921If a server application would like to accept IPv4 and IPv6 connections,
922it should use AF_INET and AF_INET6 socket (you'll need two sockets).
923Use getaddrinfo() with AI_PASSIVE into ai_flags, and socket(2) and bind(2)
924to all the addresses returned.
925By opening multiple sockets, you can accept connections onto the socket with
926proper address family.  IPv4 connections will be accepted by AF_INET socket,
927and IPv6 connections will be accepted by AF_INET6 socket (NOTE: KAME/BSDI4
928kernel sometimes violate this - we will fix it).
929
930If you try to support IPv6 traffic only and would like to reject IPv4
931traffic, always check the peer address when a connection is made toward
932AF_INET6 listening socket.  If the address is IPv4 mapped address, you may
933want to reject the connection.  You can check the condition by using
934IN6_IS_ADDR_V4MAPPED() macro.  This is one of the reasons the author of
935the section (itojun) dislikes special behavior of AF_INET6 wildcard bind.
936
937Comments on initiating side:
938
939Advise to application implementers: to implement a portable IPv6 application
940(which works on multiple IPv6 kernels), we believe that the following
941is the key to the success:
942- NEVER hardcode AF_INET nor AF_INET6.
943- Use getaddrinfo() and getnameinfo() throughout the system.
944  Never use gethostby*(), getaddrby*(), inet_*() or getipnodeby*().
945- If you would like to connect to destination, use getaddrinfo() and try
946  all the destination returned, like telnet does.
947- Some of the IPv6 stack is shipped with buggy getaddrinfo().  Ship a minimal
948  working version with your application and use that as last resort.
949
950If you would like to use AF_INET6 socket for both IPv4 and IPv6 outgoing
951connection, you will need tweaked implementation in DNS support libraries,
952as documented in RFC2553 6.1.  KAME libinet6 includes the tweak in
953getipnodebyname().  Note that getipnodebyname() itself is not recommended as
954it does not handle scoped IPv6 addresses at all.  For IPv6 name resolution
955getaddrinfo() is the preferred API.  getaddrinfo() does not implement the
956tweak.
957
958When writing applications that make outgoing connections, story goes much
959simpler if you treat AF_INET and AF_INET6 as totally separate address family.
960{set,get}sockopt issue goes simpler, DNS issue will be made simpler.  We do
961not recommend you to rely upon IPv4 mapped address.
962
9631.12.1 KAME/BSDI3 and KAME/FreeBSD228
964
965The platforms do not support IPv4 mapped address at all (both listening side
966and initiating side).  AF_INET6 and AF_INET sockets are totally separated.
967
968Port number space is totally separate between AF_INET and AF_INET6 sockets.
969
970It should be noted that KAME/BSDI3 and KAME/FreeBSD228 are not conformant
971to RFC2553 section 3.7 and 3.8.  It is due to code sharing reasons.
972
9731.12.2 KAME/FreeBSD[34]x
974
975KAME/FreeBSD3x and KAME/FreeBSD4x use shared tcp4/6 code (from
976sys/netinet/tcp*) and shared udp4/6 code (from sys/netinet/udp*).
977They use unified inpcb/in6pcb structure.
978
9791.12.2.1 KAME/FreeBSD[34]x, listening side
980
981The platform can be configured to support IPv4 mapped address/special
982AF_INET6 wildcard bind (enabled by default).  There is no kernel compilation
983option to disable it.  You can enable/disable the behavior with sysctl
984(per-node), or setsockopt (per-socket).
985
986Wildcard AF_INET6 socket grabs IPv4 connection if and only if the following
987conditions are satisfied:
988- there's no AF_INET socket that matches the IPv4 connection
989- the AF_INET6 socket is configured to accept IPv4 traffic, i.e.
990  getsockopt(IPV6_V6ONLY) returns 0.
991
992(XXX need checking)
993
9941.12.2.2 KAME/FreeBSD[34]x, initiating side
995
996KAME/FreeBSD3x supports outgoing connection to IPv4 mapped address
997(::ffff:10.1.1.1), if the node is configured to accept IPv4 connections
998by AF_INET6 socket.
999
1000(XXX need checking)
1001
10021.12.3 KAME/NetBSD
1003
1004KAME/NetBSD uses shared tcp4/6 code (from sys/netinet/tcp*) and shared
1005udp4/6 code (from sys/netinet/udp*).  The implementation is made differently
1006from KAME/FreeBSD[34]x.  KAME/NetBSD uses separate inpcb/in6pcb structures,
1007while KAME/FreeBSD[34]x uses merged inpcb structure.
1008
1009It should be noted that the default configuration of KAME/NetBSD is not
1010conformant to RFC2553 section 3.8.  It is intentionally turned off by default
1011for security reasons.
1012
10131.12.3.1 KAME/NetBSD, listening side
1014
1015The platform can be configured to support IPv4 mapped address/special AF_INET6
1016wildcard bind (disabled by default).  Kernel behavior can be summarized as
1017follows:
1018- default: special support code will be compiled in, but is disabled by
1019  default.  It can be controlled by sysctl (net.inet6.ip6.v6only),
1020  or setsockopt(IPV6_V6ONLY).
1021- add "INET6_V6ONLY": No special support code for AF_INET6 wildcard socket
1022  will be compiled in.  AF_INET6 sockets and AF_INET sockets are totally
1023  separate.  The behavior is similar to what described in 1.12.1.
1024
1025sysctl setting will affect per-socket configuration at in6pcb creation time
1026only.  In other words, per-socket configuration will be copied from sysctl
1027configuration at in6pcb creation time.  To change per-socket behavior, you
1028must perform setsockopt or reopen the socket.  Change in sysctl configuration
1029will not change the behavior or sockets that are already opened.
1030
1031Wildcard AF_INET6 socket grabs IPv4 connection if and only if the following
1032conditions are satisfied:
1033- there's no AF_INET socket that matches the IPv4 connection
1034- the AF_INET6 socket is configured to accept IPv4 traffic, i.e.
1035  getsockopt(IPV6_V6ONLY) returns 0.
1036
1037You cannot bind(2) with IPv4 mapped address.  This is a workaround for port
1038number duplicate and other twists.
1039
10401.12.3.2 KAME/NetBSD, initiating side
1041
1042When you initiate a connection, you can always connect to IPv4 destination
1043over AF_INET6 socket, usin IPv4 mapped address destination (::ffff:10.1.1.1).
1044This is enabled independently from the configuration for listening side, and
1045always enabled.
1046
10471.12.4 KAME/BSDI4
1048
1049KAME/BSDI4 uses NRL-based TCP/UDP stack and inpcb source code,
1050which was derived from NRL IPv6/IPsec stack.  We guess it supports IPv4 mapped
1051address and speical AF_INET6 wildcard bind.  The implementation is, again,
1052different from other KAME/*BSDs.
1053
10541.12.4.1 KAME/BSDI4, listening side
1055
1056NRL inpcb layer supports special behavior of AF_INET6 wildcard socket.
1057There is no way to disable the behavior.
1058
1059Wildcard AF_INET6 socket grabs IPv4 connection if and only if the following
1060condition is satisfied:
1061- there's no AF_INET socket that matches the IPv4 connection
1062
10631.12.4.2 KAME/BSDI4, initiating side
1064
1065KAME/BSDi4 supports connection initiation to IPv4 mapped address
1066(like ::ffff:10.1.1.1).
1067
10681.12.5 KAME/OpenBSD
1069
1070KAME/OpenBSD uses NRL-based TCP/UDP stack and inpcb source code,
1071which was derived from NRL IPv6/IPsec stack.
1072
1073It should be noted that KAME/OpenBSD is not conformant to RFC2553 section 3.7
1074and 3.8.  It is intentionally omitted for security reasons.
1075
10761.12.5.1 KAME/OpenBSD, listening side
1077
1078KAME/OpenBSD disables special behavior on AF_INET6 wildcard bind for
1079security reasons (if IPv4 traffic toward AF_INET6 wildcard bind is allowed,
1080access control will become much harder).  KAME/BSDI4 uses NRL-based TCP/UDP
1081stack as well, however, the behavior is different due to OpenBSD's security
1082policy.
1083
1084As a result the behavior of KAME/OpenBSD is similar to KAME/BSDI3 and
1085KAME/FreeBSD228 (see 1.12.1 for more detail).
1086
10871.12.5.2 KAME/OpenBSD, initiating side
1088
1089KAME/OpenBSD does not support connection initiation to IPv4 mapped address
1090(like ::ffff:10.1.1.1).
1091
10921.12.6 More issues
1093
1094IPv4 mapped address support adds a big requirement to EVERY userland codebase.
1095Every userland code should check if an AF_INET6 sockaddr contains IPv4
1096mapped address or not.  This adds many twists:
1097
1098- Access controls code becomes harder to write.
1099  For example, if you would like to reject packets from 10.0.0.0/8,
1100  you need to reject packets to AF_INET socket from 10.0.0.0/8,
1101  and to AF_INET6 socket from ::ffff:10.0.0.0/104.
1102- If a protocol on top of IPv4 is defined differently with IPv6, we need to be
1103  really careful when we determine which protocol to use.
1104  For example, with FTP protocol, we can not simply use sa_family to determine
1105  FTP command sets.  The following example is incorrect:
1106	if (sa_family == AF_INET)
1107		use EPSV/EPRT or PASV/PORT;	/*IPv4*/
1108	else if (sa_family == AF_INET6)
1109		use EPSV/EPRT or LPSV/LPRT;	/*IPv6*/
1110	else
1111		error;
1112  The correct code, with consideration to IPv4 mapped address, would be:
1113	if (sa_family == AF_INET)
1114		use EPSV/EPRT or PASV/PORT;	/*IPv4*/
1115	else if (sa_family == AF_INET6 && IPv4 mapped address)
1116		use EPSV/EPRT or PASV/PORT;	/*IPv4 command set on AF_INET6*/
1117	else if (sa_family == AF_INET6 && !IPv4 mapped address)
1118		use EPSV/EPRT or LPSV/LPRT;	/*IPv6*/
1119	else
1120		error;
1121  It is too much to ask for every body to be careful like this.
1122  The problem is, we are not sure if the above code fragment is perfect for
1123  all situations.
1124- By enabling kernel support for IPv4 mapped address (outgoing direction),
1125  servers on the kernel can be hosed by IPv6 native packet that has IPv4
1126  mapped address in IPv6 header source, and can generate unwanted IPv4 packets.
1127  draft-itojun-ipv6-transition-abuse-01.txt talks more about this scenario.
1128
1129Due to the above twists, some of KAME userland programs has restrictions on
1130the use of IPv4 mapped addresses:
1131- rshd/rlogind do not accept connections from IPv4 mapped address.
1132  This is to avoid malicious use of IPv4 mapped address in IPv6 native
1133  packet, to bypass source-address based authentication.
1134- ftp/ftpd assume that you are on dual stack network.  IPv4 mapped address
1135  will be decoded in userland, and will be passed to AF_INET sockets
1136  (in other words, ftp/ftpd do not support SIIT environment).
1137
11381.12.7 Interaction with SIIT translator
1139
1140SIIT translator is specified in RFC2765.  KAME node cannot become a SIIT
1141translator box, nor SIIT end node (a node in SIIT cloud).
1142
1143To become a SIIT translator box, we need to put additional code for that.
1144We do not have the code in our tree at this moment.
1145
1146There are multiple reasons that we are unable to become SIIT end node.
1147(1) SIIT translators require end nodes in the SIIT cloud to be IPv6-only.
1148Since we are unable to compile INET-less kernel, we are unable to become
1149SIIT end node.  (2) As presented in 1.12.6, some of our userland code assumes
1150dual stack network.  (3) KAME stack filters out IPv6 packets with IPv4
1151mapped address in the header, to secure non-SIIT case (which is much more
1152common).  Effectively KAME node will reject any packets via SIIT translator
1153box.  See section 1.14 for more detail about the last item.
1154
1155There are documentation issues too - SIIT document requires very strange
1156things.  For example, SIIT document asks IPv6-only (meaning no IPv4 code)
1157node to be able to construct IPv4 IPsec headers.  If a node knows how to
1158construct IPv4 IPsec headers, that is not an IPv6-only node, it is a dual-stack
1159node.  The requirements imposed in SIIT document contradict with the other
1160part of the document itself.
1161
11621.13 sockaddr_storage
1163
1164When RFC2553 was about to be finalized, there was discussion on how struct
1165sockaddr_storage members are named.  One proposal is to prepend "__" to the
1166members (like "__ss_len") as they should not be touched.  The other proposal
1167was that don't prepend it (like "ss_len") as we need to touch those members
1168directly.  There was no clear consensus on it.
1169
1170As a result, RFC2553 defines struct sockaddr_storage as follows:
1171	struct sockaddr_storage {
1172		u_char	__ss_len;	/* address length */
1173		u_char	__ss_family;	/* address family */
1174		/* and bunch of padding */
1175	};
1176On the contrary, XNET draft defines as follows:
1177	struct sockaddr_storage {
1178		u_char	ss_len;		/* address length */
1179		u_char	ss_family;	/* address family */
1180		/* and bunch of padding */
1181	};
1182
1183In December 1999, it was agreed that RFC2553bis should pick the latter (XNET)
1184definition.
1185
1186KAME kit prior to December 1999 used RFC2553 definition.  KAME kit after
1187December 1999 (including December) will conform to XNET definition,
1188based on RFC2553bis discussion.
1189
1190If you look at multiple IPv6 implementations, you will be able to see
1191both definitions.  As an userland programmer, the most portable way of
1192dealing with it is to:
1193(1) ensure ss_family and/or ss_len are available on the platform, by using
1194    GNU autoconf,
1195(2) have -Dss_family=__ss_family to unify all occurences (including header
1196    file) into __ss_family, or
1197(3) never touch __ss_family.  cast to sockaddr * and use sa_family like:
1198	struct sockaddr_storage ss;
1199	family = ((struct sockaddr *)&ss)->sa_family
1200
12011.14 Invalid addresses on the wire
1202
1203Some of IPv6 transition technologies embed IPv4 address into IPv6 address.
1204These specifications themselves are fine, however, there can be certain
1205set of attacks enabled by these specifications.  Recent speicifcation
1206documents covers up those issues, however, there are already-published RFCs
1207that does not have protection against those (like using source address of
1208::ffff:127.0.0.1 to bypass "reject packet from remote" filter).
1209
1210To name a few, these address ranges can be used to hose an IPv6 implementation,
1211or bypass security controls:
1212- IPv4 mapped address that embeds unspecified/multicast/loopback/broadcast
1213  IPv4 address (if they are in IPv6 native packet header, they are malicious)
1214	::ffff:0.0.0.0/104	::ffff:127.0.0.0/104
1215	::ffff:224.0.0.0/100	::ffff:255.0.0.0/104
1216- 6to4 (RFC3056) prefix generated from unspecified/multicast/loopback/
1217  broadcast/private IPv4 address
1218	2002:0000::/24		2002:7f00::/24		2002:e000::/24
1219	2002:ff00::/24		2002:0a00::/24		2002:ac10::/28
1220	2002:c0a8::/32
1221- IPv4 compatible address that embeds unspecified/multicast/loopback/broadcast
1222  IPv4 address (if they are in IPv6 native packet header, they are malicious).
1223  Note that, since KAME doe snot support RFC1933/2893 auto tunnels, KAME nodes
1224  are not vulnerable to these packets.
1225	::0.0.0.0/104	::127.0.0.0/104	::224.0.0.0/100	::255.0.0.0/104
1226
1227Also, since KAME does not support RFC1933/2893 auto tunnels, seeing IPv4
1228compatible is very rare.  You should take caution if you see those on the wire.
1229
1230If we see IPv6 packets with IPv4 mapped address (::ffff:0.0.0.0/96) in the
1231header in dual-stack environment (not in SIIT environment), they indicate
1232that someone is trying to inpersonate IPv4 peer.  The packet should be dropped.
1233
1234IPv6 specifications do not talk very much about IPv6 unspecified address (::)
1235in the IPv6 source address field.  Clarification is in progress.
1236Here are couple of comments:
1237- IPv6 unspecified address can be used in IPv6 source address field, if and
1238  only if we have no legal source address for the node.  The legal situations
1239  include, but may not be limited to, (1) MLD while no IPv6 address is assigned
1240  to the node and (2) DAD.
1241- If IPv6 TCP packet has IPv6 unspecified address, it is an attack attempt.
1242  The form can be used as a trigger for TCP DoS attack.  KAME code already
1243  filters them out.
1244- The following examples are seemingly illegal.  It seems that there's general
1245  consensus among ipngwg for those.  (1) mobile-ip6 home address option,
1246  (2) offlink packets (so routers should not forward them).
1247  KAME implmements (2) already.
1248
1249KAME code is carefully written to avoid such incidents.  More specifically,
1250KAME kernel will reject packets with certain source/dstination address in IPv6
1251base header, or IPv6 routing header.  Also, KAME default configuration file
1252is written carefully, to avoid those attacks.
1253
1254draft-itojun-ipv6-transition-abuse-01.txt talks about more about this.
1255
12561.15 Node's required addresses
1257
1258RFC2373 section 2.8 talks about required addresses for an IPv6
1259node.  The section talks about how KAME stack manages those required
1260addresses.
1261
12621.15.1 Host case
1263
1264The following items are automatically assigned to the node (or the node will
1265automatically joins the group), at bootstrap time:
1266- Loopback address
1267- All-nodes multicast addresses (ff01::1)
1268
1269The following items will be automatically handled when the interface becomes
1270IFF_UP:
1271- Its link-local address for each interface
1272- Solicited-node multicast address for link-local addresses
1273- Link-local allnodes multicast address (ff02::1)
1274
1275The following items need to be configured manually by ifconfig(8) or prefix(8).
1276Alternatively, these can be autoconfigured by using stateless address
1277autoconfiguration.
1278- Assigned unicast/anycast addresses
1279- Solicited-Node multicast address for assigned unicast address
1280
1281Users can join groups by using appropriate system calls like setsockopt(2).
1282
12831.15.2 Router case
1284
1285In addition to the above, routers needs to handle the following items.
1286
1287The following items need to be configured manually by using ifconfig(8).
1288o The subnet-router anycast addresses for the interfaces it is configured
1289  to act as a router on (prefix::/64)
1290o All other anycast addresses with which the router has been configured
1291
1292The router will join the following multicast group when rtadvd(8) is available
1293for the interface.
1294o All-Routers Multicast Addresses (ff02::2)
1295
1296Routing daemons will join appropriate multicast groups, as necessary,
1297like ff02::9 for RIPng.
1298
1299Users can join groups by using appropriate system calls like setsockopt(2).
1300
13011.16 Advanced API
1302
1303Current KAME kernel implements 2292bis API, documented in
1304draft-ietf-ipngwg-rfc2292bis-xx.txt.  It also implements RFC2292 API,
1305for backward compatibility purposes with *BSD-integrated codebase.
1306KAME tree ships with 2292bis headers.
1307*BSD-integrated codebase implements either RFC2292, or 2292bis, API.
1308see "COVERAGE" document for detailed implementation status.
1309
1310Here are couple of issues to mention:
1311- *BSD-integrated binaries, compiled for RFC2292, will work on KAME kernel.
1312  For example, OpenBSD 2.7 /sbin/rtsol will work on KAME/openbsd kernel.
1313- KAME binaries, compiled using 2292bis, will not work on *BSD-integrated
1314  kenrel.  For example, KAME /usr/local/v6/sbin/rtsol will not work on
1315  OpenBSD 2.7 kernel.
1316- 2292bis API is not compatible with RFC2292 API.  2292bis #define symbols
1317  conflict with RFC2292 symbols.  Therefore, if you compile programs that
1318  assume RFC2292 API, the compilation itself goes fine, however, the compiled
1319  binary will not work correctly.  The problem is not KAME issue, but API
1320  issue.  For example, Solaris 8 implements 2292bis API.  If you compile
1321  RFC2292-based code on Solaris 8, the binary can behave strange.
1322
1323There are few (or couple of) incompatible behavior in RFC2292 binary backward
1324compatibility support in KAME tree.  To enumerate:
1325- Type 0 routing header lacks support for strict/loose bitmap.
1326  Even if we see packets with "strict" bit set, those bits will not be made
1327  visible to the userland.
1328  Background: RFC2292 document is based on RFC1883 IPv6, and it uses
1329  strict/loose bitmap.  2292bis document is based on RFC2460 IPv6, and it has
1330  no strict/loose bitmap (it was removed from RFC2460).  KAME tree obeys
1331  RFC2460 IPv6, and lacks support for strict/loose bitmap.
1332
13332. Network Drivers
1334
1335KAME requires three items to be added into the standard drivers:
1336
1337(1) mbuf clustering requirement. In this stable release, we changed
1338    MINCLSIZE into MHLEN+1 for all the operating systems in order to make
1339    all the drivers behave as we expect.
1340
1341(2) multicast.  If "ifmcstat" yields no multicast group for a
1342    interface, that interface has to be patched.
1343
1344To avoid troubles, we suggest you to comment out the device drivers
1345for unsupported/unnecessary cards, from the kernel configuration file.
1346If you accidentally enable unsupported drivers, some of the userland
1347tools may not work correctly (routing daemons are typical example).
1348
1349In the following sections, "official support" means that KAME developers
1350are using that ethernet card/driver frequently.
1351
1352(NOTE: In the past we required all pcmcia drivers to have a call to
1353in6_ifattach().  We have no such requirement any more)
1354
13552.1 FreeBSD 2.2.x-RELEASE
1356
1357Here is a list of FreeBSD 2.2.x-RELEASE drivers and its conditions:
1358
1359	driver	mbuf(1)		multicast(2)	official support?
1360	---	---		---		---
1361	(Ethernet)
1362	ar	looks ok	-		-
1363	cnw	ok		ok		yes (*)
1364	ed	ok		ok		yes
1365	ep	ok		ok		yes
1366	fe	ok		ok		yes
1367	sn	looks ok	-		-   (*)
1368	vx	looks ok	-		-
1369	wlp	ok		ok		-   (*)
1370	xl	ok		ok		yes
1371	zp	ok		ok		-
1372	(FDDI)
1373	fpa	looks ok	?		-
1374	(ATM)
1375	en	ok		ok		yes
1376	(Serial)
1377	lp	?		-		not work
1378	sl	?		-		not work
1379	sr	looks ok	ok		-   (**)
1380
1381You may want to add an invocation of "rtsol" in "/etc/pccard_ether",
1382if you are using notebook computers and PCMCIA ethernet card.
1383
1384(*) These drivers are distributed with PAO (http://www.jp.freebsd.org/PAO/).
1385
1386(**) There was some report says that, if you make sr driver up and down and
1387then up, the kernel may hang up.  We have disabled frame-relay support from
1388sr driver and after that this looks to be working fine.  If you need
1389frame-relay support to come back, please contact KAME developers.
1390
13912.2 BSD/OS 3.x
1392
1393The following lists BSD/OS 3.x device drivers and its conditions:
1394
1395	driver	mbuf(1)		multicast(2)	official support?
1396	---	---		---		---
1397	(Ethernet)
1398	cnw	ok		ok		yes
1399	de	ok		ok		-
1400	df	ok		ok		-
1401	eb	ok		ok		-
1402	ef	ok		ok		yes
1403	exp	ok		ok		-
1404	mz	ok		ok		yes
1405	ne	ok		ok		yes
1406	we	ok		ok		-
1407	(FDDI)
1408	fpa	ok		ok		-
1409	(ATM)
1410	en	maybe		ok		-
1411	(Serial)
1412	ntwo	ok		ok		yes
1413	sl	?		-		not work
1414	appp	?		-		not work
1415
1416You may want to use "@insert" directive in /etc/pccard.conf to invoke
1417"rtsol" command right after dynamic insertion of PCMCIA ethernet cards.
1418
14192.3 NetBSD
1420
1421The following table lists the network drivers we have tried so far.
1422
1423	driver		mbuf(1)	multicast(2)	official support?
1424	---		---	---		---
1425	(Ethernet)
1426	awi pcmcia/i386	ok	ok		-
1427	bah zbus/amiga	NG(*)
1428	cnw pcmcia/i386	ok	ok		yes
1429	ep pcmcia/i386	ok	ok		-
1430	le sbus/sparc	ok	ok		yes
1431	ne pci/i386	ok	ok		yes
1432	ne pcmcia/i386	ok	ok		yes
1433	wi pcmcia/i386	ok	ok		yes
1434	(ATM)
1435	en pci/i386	ok	ok		-
1436
1437(*) This may need some fix, but I'm not sure what arcnet interfaces assume...
1438
14392.4 FreeBSD 3.x-RELEASE
1440
1441Here is a list of FreeBSD 3.x-RELEASE drivers and its conditions:
1442
1443	driver	mbuf(1)		multicast(2)	official support?
1444	---	---		---		---
1445	(Ethernet)
1446	cnw	ok		ok		-(*)
1447	ed	?		ok		-
1448	ep	ok		ok		-
1449	fe	ok		ok		yes
1450	fxp	?(**)
1451	lnc	?		ok		-
1452	sn	?		?		-(*)
1453	wi	ok		ok		yes
1454	xl	?		ok		-
1455
1456(*) These drivers are distributed with PAO as PAO3
1457    (http://www.jp.freebsd.org/PAO/).
1458(**) there are trouble reports with multicast filter initialization.
1459
1460More drivers will just simply work on KAME FreeBSD 3.x-RELEASE but have not
1461been checked yet.
1462
14632.5 FreeBSD 4.x-RELEASE
1464
1465Here is a list of FreeBSD 4.x-RELEASE drivers and its conditions:
1466
1467	driver		multicast
1468	---		---
1469	(Ethernet)
1470	lnc/vmware	ok
1471
14722.6 OpenBSD 2.x
1473
1474Here is a list of OpenBSD 2.x drivers and its conditions:
1475
1476	driver		mbuf(1)		multicast(2)	official support?
1477	---		---		---		---
1478	(Ethernet)
1479	de pci/i386	ok		ok		yes
1480	fxp pci/i386	?(*)
1481	le sbus/sparc	ok		ok		yes
1482	ne pci/i386	ok		ok		yes
1483	ne pcmcia/i386	ok		ok		yes
1484	wi pcmcia/i386	ok		ok		yes
1485
1486(*) There seem to be some problem in driver, with multicast filter
1487configuration.  This happens with certain revision of chipset on the card.
1488Should be fixed by now by workaround in sys/net/if.c, but still not sure.
1489
14902.7 BSD/OS 4.x
1491
1492The following lists BSD/OS 4.x device drivers and its conditions:
1493
1494	driver	mbuf(1)		multicast(2)	official support?
1495	---	---		---		---
1496	(Ethernet)
1497	de	ok		ok		yes
1498	exp	(*)
1499
1500You may want to use "@insert" directive in /etc/pccard.conf to invoke
1501"rtsol" command right after dynamic insertion of PCMCIA ethernet cards.
1502
1503(*) exp driver has serious conflict with KAME initialization sequence.
1504A workaround is committed into sys/i386/pci/if_exp.c, and should be okay by now.
1505
15063. Translator
1507
1508We categorize IPv4/IPv6 translator into 4 types.
1509
1510Translator A --- It is used in the early stage of transition to make
1511it possible to establish a connection from an IPv6 host in an IPv6
1512island to an IPv4 host in the IPv4 ocean.
1513
1514Translator B --- It is used in the early stage of transition to make
1515it possible to establish a connection from an IPv4 host in the IPv4
1516ocean to an IPv6 host in an IPv6 island.
1517
1518Translator C --- It is used in the late stage of transition to make it
1519possible to establish a connection from an IPv4 host in an IPv4 island
1520to an IPv6 host in the IPv6 ocean.
1521
1522Translator D --- It is used in the late stage of transition to make it
1523possible to establish a connection from an IPv6 host in the IPv6 ocean
1524to an IPv4 host in an IPv4 island.
1525
1526KAME provides an TCP relay translator for category A.  This is called
1527"FAITH".  We also provide IP header translator for category A.
1528
15293.1 FAITH TCP relay translator
1530
1531FAITH system uses TCP relay daemon called "faithd" helped by the KAME kernel.
1532FAITH will reserve an IPv6 address prefix, and relay TCP connection
1533toward that prefix to IPv4 destination.
1534
1535For example, if the reserved IPv6 prefix is 3ffe:0501:0200:ffff::, and
1536the IPv6 destination for TCP connection is 3ffe:0501:0200:ffff::163.221.202.12,
1537the connection will be relayed toward IPv4 destination 163.221.202.12.
1538
1539	destination IPv4 node (163.221.202.12)
1540	  ^
1541	  | IPv4 tcp toward 163.221.202.12
1542	FAITH-relay dual stack node
1543	  ^
1544	  | IPv6 TCP toward 3ffe:0501:0200:ffff::163.221.202.12
1545	source IPv6 node
1546
1547faithd must be invoked on FAITH-relay dual stack node.
1548
1549For more details, consult kame/kame/faithd/README and
1550draft-ietf-ngtrans-tcpudp-relay-04.txt.
1551
15523.2 IPv6-to-IPv4 header translator
1553
1554(to be written)
1555
15564. IPsec
1557
1558IPsec is implemented as the following three components.
1559
1560(1) Policy Management
1561(2) Key Management
1562(3) AH, ESP and IPComp handling in kernel
1563
1564Note that KAME/OpenBSD does NOT include support for KAME IPsec code,
1565as OpenBSD team has their home-brew IPsec stack and they have no plan
1566to replace it.  IPv6 support for IPsec is, therefore, lacking on KAME/OpenBSD.
1567
1568http://www.netbsd.org/Documentation/network/ipsec/ has more information
1569including usage examples.
1570
15714.1 Policy Management
1572
1573The kernel implements experimental policy management code.  There are two way
1574to manage security policy.  One is to configure per-socket policy using
1575setsockopt(3).  In this cases, policy configuration is described in
1576ipsec_set_policy(3).  The other is to configure kernel packet filter-based
1577policy using PF_KEY interface, via setkey(8).
1578
1579The policy entry will be matched in order.  The order of entries makes
1580difference in behavior.
1581
15824.2 Key Management
1583
1584The key management code implemented in this kit (sys/netkey) is a
1585home-brew PFKEY v2 implementation.  This conforms to RFC2367.
1586
1587The home-brew IKE daemon, "racoon" is included in the kit (kame/kame/racoon,
1588or usr.sbin/racoon).
1589Basically you'll need to run racoon as daemon, then setup a policy
1590to require keys (like ping -P 'out ipsec esp/transport//use').
1591The kernel will contact racoon daemon as necessary to exchange keys.
1592
1593In IKE spec, there's ambiguity about interpretation of "tunnel" proposal.
1594For example, if we would like to propose the use of following packet:
1595	IP AH ESP IP payload
1596some implementation proposes it as "AH transport and ESP tunnel", since
1597this is more logical from packet construction point of view.  Some
1598implementation proposes it as "AH tunnel and ESP tunnel".
1599Racoon follows the former route.
1600This raises real interoperability issue.  We hope this to be resolved quickly.
1601
16024.3 AH and ESP handling
1603
1604IPsec module is implemented as "hooks" to the standard IPv4/IPv6
1605processing.  When sending a packet, ip{,6}_output() checks if ESP/AH
1606processing is required by checking if a matching SPD (Security
1607Policy Database) is found.  If ESP/AH is needed,
1608{esp,ah}{4,6}_output() will be called and mbuf will be updated
1609accordingly.  When a packet is received, {esp,ah}4_input() will be
1610called based on protocol number, i.e. (*inetsw[proto])().
1611{esp,ah}4_input() will decrypt/check authenticity of the packet,
1612and strips off daisy-chained header and padding for ESP/AH.  It is
1613safe to strip off the ESP/AH header on packet reception, since we
1614will never use the received packet in "as is" form.
1615
1616By using ESP/AH, TCP4/6 effective data segment size will be affected by
1617extra daisy-chained headers inserted by ESP/AH.  Our code takes care of
1618the case.
1619
1620Basic crypto functions can be found in directory "sys/crypto".  ESP/AH
1621transform are listed in {esp,ah}_core.c with wrapper functions.  If you
1622wish to add some algorithm, add wrapper function in {esp,ah}_core.c, and
1623add your crypto algorithm code into sys/crypto.
1624
1625Tunnel mode works basically fine, but comes with the following restrictions:
1626- You cannot run routing daemon across IPsec tunnel, since we do not model
1627  IPsec tunnel as pseudo interfaces.
1628- Authentication model for AH tunnel must be revisited.  We'll need to
1629  improve the policy management engine, eventually.
1630- Path MTU discovery does not work across IPv6 IPsec tunnel gateway due to
1631  insufficient code.
1632
1633AH specificaton does not talk much about "multiple AH on a packet" case.
1634We incrementally compute AH checksum, from inside to outside.  Also, we
1635treat inner AH to be immutable.
1636For example, if we are to create the following packet:
1637	IP AH1 AH2 AH3 payload
1638we do it incrementally.  As a result, we get crypto checksums like below:
1639	AH3 has checksum against "IP AH3' payload".
1640		where AH3' = AH3 with checksum field filled with 0.
1641	AH2 has checksum against "IP AH2' AH3 payload".
1642	AH1 has checksum against "IP AH1' AH2 AH3 payload",
1643Also note that AH3 has the smallest sequence number, and AH1 has the largest
1644sequence number.
1645
1646To avoid traffic analysis on shorter packets, ESP output logic supports
1647random length padding.  By setting net.inet.ipsec.esp_randpad (or
1648net.inet6.ipsec6.esp_randpad) to positive value N, you can ask the kernel
1649to randomly pad packets shorter than N bytes, to random length smaller than
1650or equal to N.  Note that N does not include ESP authentication data length.
1651Also note that the random padding is not included in TCP segment
1652size computation.  Negative value will turn off the functionality.
1653Recommeded value for N is like 128, or 256.  If you use a too big number
1654as N, you may experience inefficiency due to fragmented packtes.
1655
16564.4 IPComp handling
1657
1658IPComp stands for IP payload compression protocol.  This is aimed for
1659payload compression, not the header compression like PPP VJ compression.
1660This may be useful when you are using slow serial link (say, cell phone)
1661with powerful CPU (well, recent notebook PCs are really powerful...).
1662The protocol design of IPComp is very similar to IPsec, though it was
1663defined separately from IPsec itself.
1664
1665Here are some points to be noted:
1666- IPComp is treated as part of IPsec protocol suite, and SPI and
1667  CPI space is unified.  Spec says that there's no relationship
1668  between two so they are assumed to be separate in specs.
1669- IPComp association (IPCA) is kept in SAD.
1670- It is possible to use well-known CPI (CPI=2 for DEFLATE for example),
1671  for outbound/inbound packet, but for indexing purposes one element from
1672  SPI/CPI space will be occupied anyway.
1673- pfkey is modified to support IPComp.  However, there's no official
1674  SA type number assignment yet.  Portability with other IPComp
1675  stack is questionable (anyway, who else implement IPComp on UN*X?).
1676- Spec says that IPComp output processing must be performed before AH/ESP
1677  output processing, to achieve better compression ratio and "stir" data
1678  stream before encryption.  The most meaningful processing order is:
1679  (1) compress payload by IPComp, (2) encrypt payload by ESP, then (3) attach
1680  authentication data by AH.
1681  However, with manual SPD setting, you are able to violate the ordering
1682  (KAME code is too generic, maybe).  Also, it is just okay to use IPComp
1683  alone, without AH/ESP.
1684- Though the packet size can be significantly decreased by using IPComp, no
1685  special consideration is made about path MTU (spec talks nothing about MTU
1686  consideration).  IPComp is designed for serial links, not ethernet-like
1687  medium, it seems.
1688- You can change compression ratio on outbound packet, by changing
1689  deflate_policy in sys/netinet6/ipcomp_core.c.  You can also change outbound
1690  history buffer size by changing deflate_window_out in the same source code.
1691  (should it be sysctl accessible, or per-SAD configurable?)
1692- Tunnel mode IPComp is not working right.  KAME box can generate tunnelled
1693  IPComp packet, however, cannot accept tunneled IPComp packet.
1694- You can negotiate IPComp association with racoon IKE daemon.
1695- KAME code does not attach Adler32 checksum to compressed data.
1696  see ipsec wg mailing list discussion in Jan 2000 for details.
1697
16984.5 Conformance to RFCs and IDs
1699
1700The IPsec code in the kernel conforms (or, tries to conform) to the
1701following standards:
1702    "old IPsec" specification documented in rfc182[5-9].txt
1703    "new IPsec" specification documented in:
1704	rfc240[1-6].txt rfc241[01].txt rfc2451.txt
1705	draft-mcdonald-simple-ipsec-api-01.txt
1706		(expired, available in ftp://ftp.kame.net/pub/internet-drafts/)
1707	draft-ietf-ipsec-ciph-aes-cbc-00.txt
1708    IPComp:
1709	RFC2393: IP Payload Compression Protocol (IPComp)
1710IKE specifications (rfc240[7-9].txt) are implemented in userland
1711as "racoon" IKE daemon.
1712
1713Currently supported algorithms are:
1714    old IPsec AH
1715	null crypto checksum (no document, just for debugging)
1716	keyed MD5 with 128bit crypto checksum (rfc1828.txt)
1717	keyed SHA1 with 128bit crypto checksum (no document)
1718	HMAC MD5 with 128bit crypto checksum (rfc2085.txt)
1719	HMAC SHA1 with 128bit crypto checksum (no document)
1720	HMAC RIPEMD160 with 128bit crypto checksum (no document)
1721    old IPsec ESP
1722	null encryption (no document, similar to rfc2410.txt)
1723	DES-CBC mode (rfc1829.txt)
1724    new IPsec AH
1725	null crypto checksum (no document, just for debugging)
1726	keyed MD5 with 96bit crypto checksum (no document)
1727	keyed SHA1 with 96bit crypto checksum (no document)
1728	HMAC MD5 with 96bit crypto checksum (rfc2403.txt
1729	HMAC SHA1 with 96bit crypto checksum (rfc2404.txt)
1730	HMAC SHA2-256 with 96bit crypto checksum (no document)
1731	HMAC SHA2-384 with 96bit crypto checksum (no document)
1732	HMAC SHA2-512 with 96bit crypto checksum (no document)
1733	HMAC RIPEMD160 with 96bit crypto checksum (RFC2857)
1734    new IPsec ESP
1735	null encryption (rfc2410.txt)
1736	DES-CBC with derived IV
1737		(draft-ietf-ipsec-ciph-des-derived-01.txt, draft expired)
1738	DES-CBC with explicit IV (rfc2405.txt)
1739	3DES-CBC with explicit IV (rfc2451.txt)
1740	BLOWFISH CBC (rfc2451.txt)
1741	CAST128 CBC (rfc2451.txt)
1742	RIJNDAEL/AES CBC (draft-ietf-ipsec-ciph-aes-cbc-00.txt,
1743		uses IANA-assigned protocol number)
1744	TWOFISH CBC (draft-ietf-ipsec-ciph-aes-cbc-00.txt)
1745	each of the above can be combined with:
1746	    ESP authentication with HMAC-MD5(96bit)
1747	    ESP authentication with HMAC-SHA1(96bit)
1748    IPComp
1749	RFC2394: IP Payload Compression Using DEFLATE
1750
1751The following algorithms are NOT supported:
1752    old IPsec AH
1753	HMAC MD5 with 128bit crypto checksum + 64bit replay prevention
1754		(rfc2085.txt)
1755	keyed SHA1 with 160bit crypto checksum + 32bit padding (rfc1852.txt)
1756
1757The key/policy management API is based on the following document, with fair
1758amount of extensions:
1759	RFC2367: PF_KEY key management API
1760
17614.6 ECN consideration on IPsec tunnels
1762
1763KAME IPsec implements ECN-friendly IPsec tunnel, described in
1764draft-ietf-ipsec-ecn-02.txt.
1765Normal IPsec tunnel is described in RFC2401.  On encapsulation,
1766IPv4 TOS field (or, IPv6 traffic class field) will be copied from inner
1767IP header to outer IP header.  On decapsulation outer IP header
1768will be simply dropped.  The decapsulation rule is not compatible
1769with ECN, since ECN bit on the outer IP TOS/traffic class field will be
1770lost.
1771To make IPsec tunnel ECN-friendly, we should modify encapsulation
1772and decapsulation procedure.  This is described in
1773draft-ietf-ipsec-ecn-02.txt, chapter 3.3.
1774
1775KAME IPsec tunnel implementation can give you three behaviors, by setting
1776net.inet.ipsec.ecn (or net.inet6.ipsec6.ecn) to some value:
1777- RFC2401: no consideration for ECN (sysctl value -1)
1778- ECN forbidden (sysctl value 0)
1779- ECN allowed (sysctl value 1)
1780Note that the behavior is configurable in per-node manner, not per-SA manner
1781(draft-ietf-ipsec-ecn-02 wants per-SA configuration, but it looks too much
1782for me).
1783
1784The behavior is summarized as follows (see source code for more detail):
1785
1786		encapsulate			decapsulate
1787		---				---
1788RFC2401		copy all TOS bits		drop TOS bits on outer
1789		from inner to outer.		(use inner TOS bits as is)
1790
1791ECN forbidden	copy TOS bits except for ECN	drop TOS bits on outer
1792		(masked with 0xfc) from inner	(use inner TOS bits as is)
1793		to outer.  set ECN bits to 0.
1794
1795ECN allowed	copy TOS bits except for ECN	use inner TOS bits with some
1796		CE (masked with 0xfe) from	change.  if outer ECN CE bit
1797		inner to outer.			is 1, enable ECN CE bit on
1798		set ECN CE bit to 0.		the inner.
1799
1800General strategy for configuration is as follows:
1801- if both IPsec tunnel endpoint are capable of ECN-friendly behavior,
1802  you'd better configure both end to "ECN allowed" (sysctl value 1).
1803- if the other end is very strict about TOS bit, use "RFC2401"
1804  (sysctl value -1).
1805- in other cases, use "ECN forbidden" (sysctl value 0).
1806The default behavior is "ECN forbidden" (sysctl value 0).
1807
1808For more information, please refer to:
1809	draft-ietf-ipsec-ecn-02.txt
1810	RFC2481 (Explicit Congestion Notification)
1811	KAME sys/netinet6/{ah,esp}_input.c
1812
1813(Thanks goes to Kenjiro Cho <kjc@csl.sony.co.jp> for detailed analysis)
1814
18154.7 Interoperability
1816
1817IPsec, IPComp (in kernel) and IKE (in userland as "racoon") has been tested
1818at several interoperability test events, and it is known to interoperate
1819with many other implementations well.  Also, KAME IPsec has quite wide
1820coverage for IPsec crypto algorithms documented in RFC (we do not cover
1821algorithms with intellectual property issues, though).
1822
1823Here are (some of) platforms we have tested IPsec/IKE interoperability
1824in the past, no particular order.  Note that both ends (KAME and
1825others) may have modified their implementation, so use the following
1826list just for reference purposes.
1827	ACC, allied-telesis, Altiga, Ashley-laurent (vpcom.com), BlueSteel,
1828	CISCO IOS, Cryptek, Checkpoint FW-1, Data Fellows (F-Secure),
1829	Ericsson, Fitel, FreeS/WAN, HiFn, HITACHI, IBM AIX, IIJ, Intel Canada,
1830	Intel Packet Protect, MEW NetCocoon, MGCS, Microsoft WinNT/2000,
1831	NAI PGPnet, NetLock, NIST (linux IPsec + plutoplus), NEC IX5000,
1832	Netscreen, NxNetworks, OpenBSD isakmpd, Pivotal, Radguard, RapidStream,
1833	RedCreek, Routerware, RSA, SSH (both IPv4/IPv6), Secure Computing,
1834	Soliton, Sun Solaris8, TIS/NAI Gauntret, Toshiba, VPNet,
1835	Yamaha RT series
1836
1837Here are (some of) platforms we have tested IPComp/IKE interoperability
1838in the past, in no particular order.
1839	IRE, SSH (both IPv4/IPv6), NetLock
1840
1841VPNC (vpnc.org) provides IPsec conformance tests, using KAME and OpenBSD
1842IPsec/IKE implementations.  Their test results are available at
1843http://www.vpnc.org/conformance.html, and it may give you more idea
1844about which implementation interoperates with KAME IPsec/IKE implementation.
1845
18465. ALTQ
1847
1848KAME kit includes ALTQ 2.1 code, which supports FreeBSD2, FreeBSD3,
1849NetBSD and OpenBSD.  For BSD/OS, ALTQ does not work.
1850ALTQ in KAME supports (or tries to support) IPv6.
1851(actually, ALTQ is developed on KAME repository since ALTQ 2.1 - Jan 2000)
1852
1853ALTQ occupies single character device number.  For FreeBSD, it is officially
1854allocated.  For OpenBSD and NetBSD, we use the number which is not
1855currently allocated (will eventually get an official number).
1856The character device is enabled for i386 architecture only.  To enable and
1857compile ALTQ-ready kernel for other archititectures, take the following steps:
1858- assume that your architecture is FOOBAA.
1859- modify sys/arch/FOOBAA/FOOBAA/conf.c (or somewhere that defines cdevsw),
1860  to include a line for ALTQ.  look at sys/arch/i386/i386/conf.c for
1861  example.  The major number must be same as i386 case.
1862- copy kernel configuration file (like ALTQ.v6 or GENERIC.v6) from i386,
1863  and modify accordingly.
1864- build a kernel.
1865- before building userland, change netbsd/{lib,usr.sbin,usr.bin}/Makefile
1866  (or openbsd/foobaa) so that it will visit altq-related sub directories.
1867
18686. mobile-ip6
1869
18706.1 KAME node as correspondent node
1871
1872Default installation recognizes home address option (in destination
1873options header).  No sub-options are supported.  interaction with
1874IPsec, and/or 2292bis API, needs further study.
1875
18766.2 KAME node as home agent/mobile node
1877
1878KAME kit includes Ericsson mobile-ip6 code.  The integration is just started
1879(in Feb 2000), and we will need some more time to integrate it better.
1880
1881See kame/mip6config/{QUICKSTART,README_MIP6.txt} for more details.
1882
1883The Ericsson code implements revision 09 of the mobile-ip6 draft.  There
1884are other implementations available:
1885	NEC: http://www.6bone.nec.co.jp/mipv6/internal-dist/ (-13 draft)
1886	SFC: http://neo.sfc.wide.ad.jp/~mip6/ (-13 draft)
1887
18887. Coding style
1889
1890The KAME developers basically do not make a bother about coding
1891style.  However, there is still some agreement on the style, in order
1892to make the distributed develoment smooth.
1893
1894- the tab character should be 8 columns wide (tabstops are at 8, 16, 24, ...
1895  column).  With vi, use ":set ts=8 sw=8".
1896- each line should be within 80 characters.
1897- keep a single open/close bracket in a comment such as in the following
1898  line:
1899	putchar('(');	/* ) */
1900  without this, some vi users would have a hard time to match a pair of
1901  brackets.  Although this type of bracket seems clumsy and is even
1902  harmful for some other type of vi users and Emacs users, the
1903  agreement in the KAME developers is to allow it.
1904- add the following line to the head of every KAME-derived file:
1905  /*	(dollar)KAME(dollar)	*/
1906  where "(dollar)" is the dollar character ($), and around "$" are tabs.
1907  (this is for C. For other language, you should use its own comment
1908  line.)
1909  Once commited to the CVS repository, this line will contain its
1910  version number (see, for example, at the top of this file).  This
1911  would make it easy to report a bug.
1912- when creating a new file with the WIDE copyright, tap "make copyright.c" at
1913  the top-level, and use copyright.c as a template.  KAME RCS tag will be
1914  included automatically.
1915- when editting a third-party package, keep its own coding style as
1916  much as possible, even if the style does not follow the items above.
1917
1918When you want to contribute something to the KAME project, and if *you
1919do not mind* the agreement, it would be helpful for the project to
1920keep these rules.  Note, however, that we would never intend to force
1921you to adopt our rules.  We would rather regard your own style,
1922especially when you have a policy about the style.
1923
1924						 <end of IMPLEMENTATION>
1925