CVS log for kame/kame/sys/netinet6/in6_proto.c

Up to [KAME snap] / kame / kame / sys / netinet6

Request diff between arbitrary revisions - Display revisions graphically

Keyword substitution: kv
Default branch: MAIN

ANSIfy, to reduce diffs with *BSDs.
based on diff from dunceor at gmail.com.
XXX not tested yet, sorry, we need to ressurect buildlab configs.

random thoughts:
- IGMPv3 support needs a serious cleanup.  argument passing is horrible, style
  does not meet style(9) at all.  i'm afraid there could be a lot of bugs.
- DCCP/SCTP/NAT-PT need someone to maintain them, otherwise we should drop
  them at once.
- we should pick a platform (one of *BSD) and try to get KAME tree in sync
  with more recent version of *BSD.  otherwise, KAME tree gets more and more
  out-of-date.

- Disabled responding to NI queries from a global address by default as
  specified in RFC4620.  A new flag for icmp6_nodeinfo was added to enable the
  feature.
- Also cleaned up the code so that the semantics of the icmp6_nodeinfo
  flags is clearer (i.e., defined specific macro names instead of using
  hard-coded values).

too big default - reduce it

	ipv6 socket should use SOCK_CONN_DGRAM for dccp

- use callout timer mechanism for protocol timer
- so icmp/mld_fasttimo() is no longer necessary

icmp/mld_slowtimeo is no longer used

sync with freebsd 5.4

remove freebsd[234]-specific code, since they are no longer supported in KAME-SNAP

added a knob to enable path MTU discovery for multicast packets.
(by default, it is disabled)

typo

sync w/ 2.0.
needs checking/fixing: igmp3, pf, altq, dccp, sctp
FAST_IPSEC path is not considered, intentionally.

Compile on FreeBSD 5.3 with DCCP config.

New Mobile IPv6 code which also supports NEMO.

a patch for IPPROTO_MH socket so that we can send/recceive MH packets
via a raw socket.

remove support for freebsd[23], simplify #ifdef

clean TCP6 define (i was used for NetBSD + bsdi-modified-tcp6)

typo

simplify #ifdef by de-support bsdi[34] and freebsd[23].

make it compilable on freebsd521+kame

ToDo: reflect changes in freebsd521, test, ...

restructure of header files based on draft-ietf-mip6-mipext-advapi-00.

new patch from randall.

1) Beginning of FreeBSD5 move. It should now compile there,
   not yet sure how stable we are though.

2) Fixes async I/O for 1-1 model.

3) Fix of a rwnd mis-calculation in processing
   packetdrops.

4) Performance enhancment to make it so that always
   the first mbuf can have enough room for the
   outgoing IP header and sctp common header. This
   makes it so that no addition mbuf alloc will be
   needed before output. This adds a special pre-pend
   routine to allow this to happen for SCTP.

5) Beginning work for base freebsd port. This will
   be passed to Michael Tuexen (once it is done) so
   that he can compile into mac X. So far all is
   in but the needed and missing scoping for V6.
   It should work fine right now for V4 and V6 global.
   But not V6 link local .. we need to restore
   old things from the past to make that work.

6) Additional rwnd pre-allocation done on send. We now
   reserve mbuf space for outgoing to. This makes us
   a bit more conservative in our use of the peers
   rwnd, but it prevents us from overrunning the
   peer. I saw this occur on highspeed local loopback
   testing (before the fix).

7) Ifdef'ed out sctp_mprepend and switch back to MPREPEND

8) A kozuka-san missed in his work on 5.x

9) All the k&r changes emacs search/replace could find.

changed mobility header protocol no. and MIP6 related ICMPv6 type no.
based on IANA assignment.

fixed a panic in receiving v6 dccp packet on netbsd.
fixed a bug in receiving linklocal addr.

separate dccp6_usrreq.c from dccp_usrreq.c

removed SOCK_DCCP. use SOCK_DGRAM

included opt_dccp.h on freebsd[45]

dccp from www.dccp.org

added some sysctl switches for mip6

randomize flowlabel/fragment ID.

enforce ipsec on raw wildcard case

there's no need to define SENDREDIRECTS, IMHO

- restructured mip6 code.
  divided cn/ha/mn part to separate files.  this makes it easier to
  merge a paticular part (ex. CN) to original BSD or to integrate
  a paticular part (ex. HA) to routers.

sctp sysctl for openbsd.  cleanup sctp_var.h to expose it to public.

pmtu_probe is used by sys/netinet6/tcp6*.c only.
From: Krister Walfridsson <cato@df.lth.se>

blank lines

compilable when MLDv2 is disabled

added a timer to kick back from MLDv1-compat to MLDv2

compile w/o GIF

latest patch from randall, with tons of style fixes
- space before/after paren/comma
- function prototype must not carry variable name
- indentation

remove PR_ATOMIC from SOCK_STERAM/IPPROTO_SCTP.

>A very small patch to fix the SCTP TCP-mode API (SOCK_STREAM) for
>non-datagram oriented read()/write()'s.

KNF - return is not a function.

move use_deprecated back to 1, after ipngwg discussion

set default value for use_deprecated to 0, it has consequences with ftpd
(ftpd control connections goes successful, then no data connection possible)

new shipment of SCTP code from randall.

>1) A problem with the code in building for OpenBSD and also a issue
>   with the TCP model in NetBSD
>2) Local host specific problem when connecting to a localhost
>communication
>   would fail when the localhost had a private network address only.
>3) Error code correction... E2BIG was the wrong error message.

nd6_drain() to cope with mbuf shortage
XXX locking

whitespace cleanup

style

limit number of fragments (not fragment queues) in kernel.

transition to use ifp->if_afdata.  compiles but not tested

- the first step for mip6 id17 support.
  this commit is for the KAME mip6 developpers.  do not try to use this.

avoid using confusing name, sctp_usrreq*s* for sctp usrreq function.

initial integration of sctp stack (not really ready for actual use)
from randall stewart @ cisco

added a new sysctl knob net.inet6.ip6.pmtu_expire to configure the timer value
of path MTU queue.
as suggested in RFC 1981,
  - values less the 5 minutes (except 0) are invalid
  - 0 is a special value which means no expiration
(avilable in freebsd4 only)

have tcp6_drain

remove opt_inet6.h,mip6_var.h&etc. includes from if_hif.h&mip6.h

move in6_gif_hlim decl to in6_gif.c.  fix hoplimit value for IPvX-over-IPv6
tunnel (has been 0 for some time).

too many space at the end of comment (s,  */, */,)

added sysctl defintion for the new sysctl name

move in6_gif_protosw to in6_gif.c

correct INET6_BINDV6ONLY for netbsd.

flush route cache as necessary on ctlinput.  reported by sumikawa
(IPvX over IPv6 tunnel, IPv6 PMTUD does not get reflected due to cached route)

call encap6_ctlinput.  XXX need testing

added sysctl entry for ip6.prefer_tempaddr

modify the third parameter of mip6_tunnel_protosw from 0 to IPPROTO_IPV6.

netbsd don't need opt_hoge.h. remove it from sources.
freebsd3 need opt_hoge.h, add opt_mip6.h to sys/conf/files.

 Modified Files:
 	freebsd3/sys/conf/options kame/sys/netinet6/dest6.c
 	kame/sys/netinet6/icmp6.c kame/sys/netinet6/in6.c
 	kame/sys/netinet6/in6_proto.c kame/sys/netinet6/in6_src.c
 	kame/sys/netinet6/ip6_forward.c kame/sys/netinet6/ip6_input.c
 	kame/sys/netinet6/ip6_output.c kame/sys/netinet6/mip6.c
 	kame/sys/netinet6/mip6_binding.c kame/sys/netinet6/mip6_ha.c
 	kame/sys/netinet6/mip6_icmp6.c kame/sys/netinet6/mip6_prefix.c
 	kame/sys/netinet6/mip6_subnet.c kame/sys/netinet6/nd6_rtr.c

add mobileip source code.

currently, many parts of spec are not implemented yet.  even
uncompilable under other than freebsd4 now.  also very unstable.
of course, without MIP6 kernel option, there is no problem. all mip6
related code is separated by ifdef MIP6.

this integration is just for KAME mip6 developper's convinience.
the purpose of this early integration is to decrease the maintenance cost
to synchronize KAME main tree and developping mip6 tree.  after this merge,
mip6 code modifications are reflected directly to KAME code.

i will make these codes compilable on the other OSes as soon as possible.

 Modified Files:
	kame/sys/netinet6/dest6.c kame/sys/netinet6/icmp6.c
	kame/sys/netinet6/in6.c kame/sys/netinet6/in6.h
	kame/sys/netinet6/in6_ifattach.c
	kame/sys/netinet6/in6_ifattach.h kame/sys/netinet6/in6_proto.c
	kame/sys/netinet6/in6_src.c kame/sys/netinet6/in6_var.h
	kame/sys/netinet6/ip6_forward.c kame/sys/netinet6/ip6_input.c
	kame/sys/netinet6/ip6_output.c kame/sys/netinet6/mip6.c
	kame/sys/netinet6/mip6.h kame/sys/netinet6/mip6_ha.c
 	kame/sys/netinet6/nd6.h kame/sys/netinet6/nd6_rtr.c
 	freebsd4/sys/net/if_types.h freebsd4/sys/conf/files
 	freebsd4/sys/conf/options
 Added Files:
 	kame/sys/net/if_hif.c kame/sys/net/if_hif.h
 	kame/sys/netinet6/mip6_binding.c
 	kame/sys/netinet6/mip6_icmp6.c kame/sys/netinet6/mip6_prefix.c
 	kame/sys/netinet6/mip6_subnet.c kame/sys/netinet6/mip6_var.h

style: added spaces after /* and before */
(this might be a matter-of-taste issue, though)

retire sys/netinet6/in6_prefix.c altogether.  i'm sure it's now safe
to do this, as there was no operation that added entries into rr_prefix chain.

(equal cost) multipath routing table support for radix tree.
the change has lowest possible impact to users/existing code.
- if you issue multiple RTM_ADD with same rt_key, they will get accepted
  (if multipath is enabled for the family).
- if there's multiple route entry with the same rt_key, rt_gateway is
  required for RTM_DELETE.
- if there's only single route entry for a given rt_key, there's no behavior
  change on RTM_RESOLVE/RTM_DELETE (no rt_gateway required for RTM_DELETE).
- multipath entries are ordered carefully so that the first one get added
  (= the route on old behavior) comes to the top of rn_dupedkey chain.

TODO:
inet and inet6 support for multipath routes.
RTM_GET handling change may be necessary.

create an entry of IPV6CTL_STATS sysctl.
fix the problem that netstat doesn't show raw6 and icmp6 pcblist.
make netstat use sysctl to retreive stats of ipv6 and icmpv6
instead of kread.

netbsd: recover INET6_BINDV6ONLY by protecting ip6_v6only variable.
it is part of netbsd releases, so we need to keep it.
(no #ifdef in in6_pcb.c - which is good)

get rid of INET6_BINDV6ONLY (netbsd).
XXX IPv4 mapped address support cases are way too complicated!
we really really need to clean them up and check if we have made any mistakes
(it looks to me that there there are quite a few).

freebsd[45] make the default value of net.inet6.ip6.maxfragpackets
 =  nmbclusters / 4. from FreeBSD-current.

sync ECONNABORTED processing with openbsd-current.

MIP6 removed.
Ericsson's code is provided as a patch to the plain (without MIP6)
KAME source after this point.

sysctl access to rip6stat (net.inet6.ip6.rip6stats).  for bsdi/freebsd.
not tested.

wrap offsetof()

use protosw flag to determine ECONNABORTED cases.  from thorpej

change net.inet6.icmp6.nodeinfo to bitmap.
1: respond/ignore NI FQDN
2: respond/ignore NI NODEADDR

enforce IPsec policy checking for transport mode input.
fixes *bsd ip6 (modulo openbsd).

slightly reduce #ifdef in in6_proto.c.

freebsd[34]: unify ip6_mapped_addr_on into ip6_v6only.
net.inet6.ip6.mapped_addr is obsoleted.
XXX do we need backward compatibility behavior for sysctl?

supply net.inet6.ip6.v6only on all platforms, to indicate AF_INET6 wildcard
bind behavior.

openbsd/bsdi3/freebsd2: 1, cannot change
netbsd: 1, can change unless INET6_BINDV6ONLY is defined at compilation time
bsdi4: 0, can change (no effect - should be made readonly)
freebsd3/4: 0, cannot change
	(should be made modifyable after conflict with ip6_mapped_addr gets
	fixed)

call encap_init on every protosw entry that uses encap[46]_input.
now encap_init is safe against multiple invocations.

need ATOMIC|ADDR on EON

typo in EON support.

netbsd: ISO over IPv4/v6 on gif*, using EON encapsulation.  from chopps

added sysctl entry for ip6_auto_linklocal.

sysctl knob for nd6_debug

freebsd[234]: on net.inet6.ip6.forwarding change from 0 to 1, do not nuke
prefix database.  the behavior is confuing and is not friendly with the latest
onlink management.

sysctl knobs for addrconf privacy related values on fbsd
(checked on fbsd4 only)

made it compilable.

address/prefix clean-ups:
- completely separate address mainipulation and prefix manipulation.
- handle L and A bits (especially the former) correctly.
- clarification about "a single prefix on multiple interfaces" cases.
TODOs:
  - more clarification on address autoconf.
  - MIP6 addresses and prefixes???

put some #ifdef RADIX_ART glue.  put sctp glue as #if 0.

bring in OpenBSD 2.8 IPv6 ipsec support.  #ifdef hell...

need PR_LISTEN for netbsd

now all OSes need esp6_ctlinput.

call rip6_ctlinput for icmp-against-icmp.  may need more calls

change SYSCTL_HANDLER_ARGS expression for fbsd 4.x-RELEASE

use {ah,esp}6_ctlinput for netbsd

get rid of errratelimit

nuke use of ratelimit(9) for icmp rate limitation.  sync with netbsd/openbsd.

don't call tcp_ini twice if we are dual stack.  from: enami@netbsd

cosmetic

turn off rate limitation for icmp6 error (we still have pps limit)

codeshare for gif.  enable stf on openbsd.

workaround to allow compilation of GENERIC.KAME

add sysctl support for nd6_maxnudhint.

20pps looks too little... maybe 200pps.

change default icmp6 err rate/pps limit to 100us/20pps.
100us: allow "traceroute6 -n ::1" to success on Celeron300 :-)
20pps: 200Kbps on 1280byte packets

* sys/netinet6/icmp6.c: add net.inet6.icmp6.errppslimit sysctl MIB.
  it will let you limit ICMPv6 error messages by pps, not interval.
  still not sure what is the best way to perform rate limitation...

change UDP6CTL_xx symbol name to meet with IPv4 counterpart.

make net.inet6.udp6.* sysctl variable name meet IPv4 variant.

use timeval for rate limit.  use ratecheck(9) where possible - for
netbsd15 and openbsd27.  disallow negative value for net.inet6.icmp6.ratelimit.
(revenge, with freebsd/bsdi consideration)

backout previous for the moment, freebsd/bsdi consideration was lacking

use ratecheck() whenever possible.  change type of icmp6errratelim
from u_int to struct timeval.
it will ease kame/netbsd15 and kame/openbsd27 upgrade.

IPSEC* defines are moved from opt_inet.h to opt_ipsec.h in freebsd3.

INET6 define is moved from opt_inet.h to opt_inet6.h in FreeBSD3,
I believe this helps to share codes with FreeBSD4 easily.

correct encap_init in table.

FreeBSD4 need netinet6/tcp6_var.h

netinet6/udp6.h does not exist on FreeBSD-4.x

remove  MAPPED_ADDR_ENABLED option.
you can turn on/off maped-addr supporting by sysctl instead of it.

nuke spaces end of line.

fix inbound processing - 6to4 outer is IPv4

stf.h should not be included except on netbsd.

add 6to4 pseudo interface.  supports netbsd only at this moment.

kernel config:
	pseudo-device stf
runtime config:
	# ifconfig stf0 inet6 2001:0a01:0101::1 prefixlen 16 alias

TODO: support other platforms.  should be trivial, just change stf_getsrcifa6().
TODO: check router/inbound behavior.  outbound works fine.
TODO: path MTU consideration.
TODO: call to nd6_ifattach() on non-ND6 interface???
TODO: we may want to remove IFF_MULTICAST.

call encap_init.

XXX in "options INET" case, we need to do it in in_proto.c.
we need to call it only once (it if not harmful to call it multiple times,
though).

net.inet6.icmp6.nd6_proxyall was obsoleted.
use ndp -s <v6> <mac> proxy if you need the same functionality.

add $KAME$

sync with latest mip6 from ericsson.
encaptab (ip_encap.[ch]) is now visible to the outside, read-only.

ericsson guys: look at line marked with "BUG ALERT" in mip6_md.c.
we have missing curly brace, or wrong indentation there.

compiled on: freebsd2 (will do on other platforms)

s/in6_gif_input/encap6_input/

implement rip6_ctlinput().  (not tested)

rip6_ctlinput for bsdi4.

revise gif input logic.

- split outer address match logic as encap[46]_input, so that other guys
  (like mobile-ip6, 6to4, 6over4, whatever) can match packets based on src/dst.
  see comment on ip_encap.c for detail - I hate tunnelling.  really.
- allow raw ip{,6} output for proto #4 and #41.
TODO: ordering of packet matching in encap[46] must be prefix length-based.
	(we have src/dst prefix length, so it is not trivial to define total
	order among entries)
TODO: change ipip_input() to conform to this framework?
CAVEAT: IFF_LINK0 for gif is not working with this change, i'll try to cleanup
	afterwards.

changes for other OSes will be committed shortly.

implement rip6_ctlinput().  (at this moment freebsd2, netbsd, bsdi3 only)

allow raw ip6 output for protocol # 4 and 41.

avoid direct reference to netinet6/{ip6,icmp6,in6}.h.

add some question as comment.

sync with change in in6_prefix.c, affects freebsd3 only (not tested)

s/#if defined(\(.*\))/#ifdef \1/

split net.inet6.ip6.bindv6only from net.inet6.ip6.mapped_addr.
- bindv6only is used only on netbsd.  it controls bind(2) side only.
- now mapped_addr is used in freebsd3 only.
  feel free to nuke the flag, or modify it at any time >shin

for netbsd, make connect(2) to mapped addr work at any time.

compiled on: netbsd, freebsd2

add net.inet6.icmp6.nodeinfo sysctl, which enable/disables icmp6 node
information query.  requested by openbsd camp (and I believe
it useful for kame itself).

fix PULLDOWN_TEST code for ni6_input().  add comment on calling convention
for ni6_input().

make MAPPED_ADDR_ENABLE usable on netbsd as well.
- no MAPPED_ADDR_ENABLE option: disabled, can't change
- options MAPPED_ADDR_ENABLE=0: code compiled in, disabled by default
- options MAPPED_ADDR_ENABLE=1: code compiled in, enabled by default
disable/enable status will be set on in6pcb creation time.
(if you would like to change socket behavior, you need to reopen it)

hope this does not break freebsd3...

TODO: net.inet6.ip6.mapped_addr and IPV6_BINDV6ONLY has reversed semantics.
we'd better make them have the same meaning.

typo

fix ICMPv6 on freebsd2 (and perhaps bsdi3).  I made a mistake in enabling
tcp4/udp4 on bsdi4 (I erased wrong #endif).

enable tcp6/udp6.

portability fixes for bsdi4.
tested on: bsdi4, openbsd

do not enable KAME IPSEC code on OpenBSD.
OpenBSD guys will not be switching IPsec stack.

XXX IPv6 IPsec is not supported at all.

make raw socket work for openbsd. (more commit should follow)

openbsd portability #ifdefs.

compiled on: freebsd2, netbsd

OpenBSD portability fixes (not complete).
- OpenBSD needs struct proc on xx_control(), xx_ioctl() and other places.
- do not include if.h twice in in6.c.
- OpenBSD has SIOCSIFMTU defined but does not support it.

added and corrected *_usrreqs for FreeBSD3

fix compilation on FreeBSD2.

cleanup sysctl_ip6_forwarding (oops not tested)

-removed redundant include of proc.h in in6.c
-functionize sysctl of ip6_forwarding

-fix in6_proto.c to make kernel buildable. (also clean-up of FreeBSD3 ifdefs)
-one more addition for mapped_addr

remove ambiguous check for FreeBSD3.

change hard-to-understand ifdef that means "FreeBSD 3 or later".

added ifdefs for FreeBSD3

- non-FreeBSD3: avoid file-global definition of time_second (avoid extra
  sideeffects)
- rewrite hard-to-understand #ifdef like
	#if !defined(__FreeBSD__) || __FreeBSD__ < 3
  which is, actually,
	#if !(defined(__FreeBSD__) && __FreeBSD__ >= 3)

-added some MAPPED_ADDR related ifdefs
-added some inp and in6p ifdefs
-added a sysctl for router renumbering

add net.inet6.ip6.use_deprecated, which allows deprecated address for
responding source, as a last resort.  to conform to RFC2462 5.5.4.
default value is 1 (allow deprecated address as last resort).
turning it to 0 shows the previous behavior (deprecated address will not be
used as source).
NOTE: explicit bind(2) against deprecated address is still prohibited.

merge bug fix:
- allow net.inet6.ip6.forwarding to be changed (bsdi3)
- allow access to net.inet6.ip6.kame_version (freebsd2/3)

fix FreeBSD228 build.  now it can communicate with IPv6 sockets.
(the bug was caused by pmtud code diffs)

sync with today's KAME/FreeBSD228 kernel.
GENERIC.v6 compiles fine but not tested.
need more synchronization.

dropped items: MAPPED_ADDR_ENABLED RADISH
need integration: PM ALTQ

sync with latest KAME/NetBSD.
- defopt INET6/IPSEC/IPSEC_ESP.
- tcp_notify update.
- remove sys/netinet6/in6_systm.h

I think /cvsroot/kame-netbsd/sys can be obsoleted.

bring in portability #ifdef from unified-ipv6.
not really tested and incomplete, incomplete but compiles fine on NetBSD.

KAME/NetBSD 1.4, as of 1999/7/30.
for unified-KAME code.
no cross-OS portability is introduced.
works fine on NetBSD as no change from KAME/NetBSD.

Initial revision