Changes to Pluto
================

RCSID $Id: CHANGES,v 1.198 2003/06/14 18:51:01 dhr Exp $

- [MCR] changes to support co-terminal connections of the near kind.

- --dontrekey is now a misnomer: the IPsec SA will be rekeyed if we
  are the Initiator and there is outbound traffic in the period
  leading up to the rekeying decision point.  This is a heuristic: a
  mistake doesn't matter much since the connection, once deleted, will
  be rekeyed on demand.

- accept a proposal with compression even if the connection does not
  have the POLICY_COMPRESSION flag.  This is a reversal of a policy
  introduced in 1.6.

- fixed load-triggered deadlock between asynch DNS process and pluto

- add %myid to indirectly refer to current id.  Set from $IPSECmyid on
  startup and via whack --myid

- make whack/pluto version matching less stringent for --status and
  --shutdown.

- remove all code for DODGE_DH_MISSING_ZERO_BUG

- when attempting outbound Opportunism, check our side's DNS
  entries, not just the other side's.  If we don't have them,
  fail: not worth attempting.


Change since 2.00 release by MCR and DHR

- "KEY Restrict": in each place where Pluto used the DNS KEY Resource
  Record, it now uses one of our special TXT Resource Records.  For
  now, we continue to accept KEY records if the TXT record is not
  found.

Changes since 1.99 release by D. Hugh Redelmeier <hugh@mimosa.com

- when Responding with an Opportunistic connection, check our side's
  DNS for proper delegation.  This avoids the promiscuous
  "packetdefault" conn from licensing unexpected access to our clients.

- many changes for "Policy Groups":
  + host=%group and host=%opportunisticgroup signals that
    contents of group files should be substituted
  + an opportunistic connection can be constrained to only
    apply to peers and their clients within a specified subnet
  + Pluto can manage pure shunt connections (involving no IKE).
  + failure-case shunts can now be specified.

- add a listing of pending Phase 2 negotiations to status output

- reorganize impairment mechanism.  Now controlled by unadvertised
  runtime flags.

- [MCR] tolerate Certificate Request Payloads any time

- [MCR] generate Vendor ID Payloads (sent if PLUTO_SENDS_VENDORID defined)

- [MCR] print some of content of Vendor ID Payloads received.

- Corrected design bug in --dontrekey.  If IPsec SA lifetime chosen by
  us is smaller than what other side specified in negotiation, we will
  renegotiate when the shorter lifetime dictates.  This is the only
  sane way to communicate the shorter lifetime.  For ISAKMP SA lifetime,
  we accept the other side's choice.


Changes since 1.98 release by D. Hugh Redelmeier <hugh@mimosa.com>

- Fixed assertion failure in quick_inI1_outR1_continue.  Can happen
  when the result of a DNS TXT query is not needed by the time it
  arrives (due to another negotiation providing the information).
  Can be demonstrated by the test "ipsec-oppo-race-iinr-net".


Changes since 1.97 release by D. Hugh Redelmeier <hugh@mimosa.com>

- added --dnskeyondemand option for connection ends.

- Fixed a bug that provoked PF_KEY errors.  If an updown script fails
  and it was invoked as part of a step that involved installing an
  eroute, any previous eroute should be restored as part of undoing
  that step.  The bug was that any old eroute was lost.

- When representing a long string in a DNS TXT resource record,
  it must be broken up into chunks less than 256.  When reconstitute
  the original string, Pluto now just concatenates them.  It used to
  add a space after a chunk if it was shorter than 255 characters.
  This should make it easier (not easy!) to edit the TXT record
  source by hand.

- [mlafon@arkoon.net] fixed protocol numbers used in delete payload.

- took -DDODGE_DH_MISSING_ZERO_BUG out of Makefile.  Anyone planning
  on talking to FreeS/WAN 1.0 systems should put it back.  And
  check into the Smithsonian Institution.

- in --status and logging output, references to connection instances
  now contain a sequence number for the instance and details of the
  instantiation (i.e. more IP addresses).  Perhaps too bulky.

- reworked initiate_opportunistic to avoid race conditions
  introduced with Asynch DNS.

- added impairment to aid testing: IMPAIR_DELAY_ADNS_ANSWER

- whack --status now displays the bare shunt table.

- Changed Main Mode hash calculation to use the ID Payload as sent
  by peer, rather than reconstituting it from our tables.  This matters
  if we use different case than peer did (eg @example.com vs @Example.Com).

- renamed source files:
  + kernel_comm.[hc] => rcv_whack.[hc]
  + preshared.[hc] => keys.[hc]
  + main.c => plutomain.c

- added POLICY_UP to track whether our admin has requested that this
  connection be up.


Changes since 1.96 release by D. Hugh Redelmeier <hugh@mimosa.com>

- fixed a bug that would make the following (pointless) sequence to
  trigger a failure of Pluto assertion "c->routing == RT_ROUTED_TUNNEL":
	ipsec auto --add conn
	ipsec auto --up conn
	ipsec auto --route conn
	ipsec auto --down conn

- send Delete Notifications when deleting IPSEC SAs.  We don't yet
  understand ones we receive.

- added "keyid" (see ipsec_keyblobtoid(3)) to appropriate messages so
  that RSA key being used is manifest.

- track whether information (security gateway, public key) came from
  DNSsec or unauthenticated DNS.  Untested since normal resolver calls
  can no longer return DNSsec results.  The information isn't used.

- use asynchronous DNS mechanism for all lookups during keying.

- added --interface flag to Pluto to constrain interfaces considered.
  One use of this option is to specify which interface should be used
  when two or more share the same IP address.  Another is to assist
  with test setups.

- small cleanups:
  + fix compile without no -DKLIPS
  + use ttodatav in place of atobytes and ttodata
  + use hosttosubnet
  + define and use close_any
  + define and use USED_BY_KLIPS and USED_BY_DEBUG
  + define and use happy, a kind of assertion macro
  + define and use impossible, a kind of assertion macro
  + when an unknown attribute value is seen in an ISAKMP transform,
    reject only that transform, not the whole proposal.
  + add Hybrid auth methods to table to improve diagnostics

Changes since 1.95 release by D. Hugh Redelmeier <hugh@mimosa.com>

- fixed bug in ipsec_doi.c:main_inR3 with unknown consequences.
  When the Initiator finds out the peer's Phase 1 ID, it might
  change which connection it thinks is being used.  If so,
  this routine used to perform an operation on the old connection
  data structure -- a mistake now corrected.  There are few cases
  where the Responder surprises the Initiator about Phase 1 IDs.

- fixed an error in network error handling that caused a segfault if
  there was a MSG_ERRQUEUE report on the last message of a Quick Mode
  exchange.

- fix leak (detected by assertion failure) triggered by missing
  private key.


Changes since 1.94 release by D. Hugh Redelmeier <hugh@mimosa.com>

- added "whack --deletestate <state_object_number>" as a crude
  tool for deleting instances of connections.

- fix assertion in quick_inI1_outR1 introduced with key rollover
  support in 1.92.  Assertion failed in Responder when Source
  was not the same as Initiator.

- reduce level of routine logging.  Add --debug-lifecycle,
  a temporary hack: it controls some logging, not debugging.

- Reorganize code to prepare for asynchronous DNS.

- Arrange for our file descriptors to be close-on-exec so that
  the updown script won't inherit them.  Some library facilities
  (syslog?) may not be as careful.

- Fix assertion failure in DNS record handling.  Provoked by a
  malformed key in a TXT record.  Introduced in 1.93.

- Change Responder cookie generation algorithm so that a particular
  peer no longer gets the same cookie each time.

- Tidy and correct setting and resetting cur_* variables.  A per-connection
  debug setting could provoke a GLOBALS_ARE_RESET assertion failure.

- Fix handing of pending list in release_dead_interfaces:
  connection must be released before it is removed from host_pair list.

- Ignore IPv6 interface addresses that are of link-local scope.
  We think that they are never relevant.  Trying to bind to these
  addresses without a scope-id causes bind to balk.


Changes since 1.93 release by D. Hugh Redelmeier <hugh@mimosa.com>

- Only run the up-* updown script for tunnel eroutes, not shunt
  eroutes.

- Now only --routed connections may be initiated on demand
  (i.e. triggered by SADB_ACQUIRE triggered by trapped outbound packet).
  Among other things, this eliminates a bug whereby an ACQUIRE
  could be ascribed by Pluto to a Road Warrior template connection.
  Pluto will now refuse to --route a Road Warrior template.

- Correct bug that lost track of a bare shunt table entry.
  Add debugging logging for changes to that table.


Changes since 1.92 release by D. Hugh Redelmeier <hugh@mimosa.com>

- [andreas.steffen@strongsec.com] Bug fix: carrying Phase 1 ID
  over when instantiating a different RW connection due to
  revelation of Phase 2 IDs.  This bug should only have affected
  X509-authenticated negotiations.

- add more context to log and status lines.  If the connection or
  state is an instance, show the peer involved; if Opportunistic, also
  show the client subnet.

- seed rand(3) with bytes from /dev/urandom.

- propose 1536 MODP Group before 1024 MODP Group.  This generates
  more entropy, so should be a little safer.  And slower.

- add --debug-dns flag

- During Phase 1, when an RSA Public Key is require, only query DNS
  for a KEY record for an Opportunistic connection.  (This does not
  change the whack --keyid way of querying for a KEY record during
  preconfiguration.)

- Multiple RSA public keys, if available, will be tried when
  authenticating a signature.  This facilitates key rollover.
  New whack flag --addkey.
  Multiple DNS TXT and KEY records are used if they are found
  (but only one Responder is tried).

- no longer try to figure out --rsasig or --psk if neither specified.
  This would require an extra DNS lookup for a KEY record.

Changes since 1.91 release by D. Hugh Redelmeier <hugh@mimosa.com>

- got rid of caching of DNS results.  They might become stale.

- Added --dontrekey option for a connection.  Crudely useful
  for Road Warrior and Opportunistic connections.

- Ignore the Commit Flag.  This should let us interoperate with
  implementations that insist on it.  The damage should be minor.

- Don't split the topology line of --status output.  The output is now
  more consistent (if wider).

- Improve diagnosis when ipsec.secrets has a non-indented "}"
  at the end of an RSA private key.

- Savatier Sbastien <SebastienSavatier@lyon.msi-sa.fr>:
  fix bug in emitting explicit IP address ID payload.

- Support inbound policy check of source and destination inner
  addresses of tunneled packets.  This will make it possible to
  prevent one peer from successfully spoofing packets from another.

- Use poll(2) to check for MSG_ERRQUEUE messages before recvfrom:
  even though select(2) says that there is a message, a plain
  recvfrom(2) can block if the only message is on MSG_ERRQUEUE.  I
  think that this blocking is a kernel bug, or at least a
  documentation bug.  Also check for MSG_ERRQUEUE messages before
  sendto(2): their presence can cause sendto to fail confusingly (i.e.
  the failure has nothing to do with the actual sendto).


Changes since 1.9 release by D. Hugh Redelmeier <hugh@mimosa.com>

- SIGTERM now causes Pluto to exit cleanly.

- if --nofork is used, Pluto prints the line "Pluto initialized" to
  standard out after creating the lock file and the control socket.

- the startup script relies on the property that whack can be
  used immediately after the pluto command finishes (i.e.
  right after the parent process of the daemon exits).
  In the past, the control socket used by whack was created
  after the fork: not necessarily before the parent exits.
  This race condition has been eliminated.

- refined the code for DODGE_DH_MISSING_ZERO_BUG to ensure
  Responder never drops a negotiation.

- added and exploited builddiag(), a routine make it easier to add
  intermediate context to diagnostics.

- For purposes of IPv4, Pluto will now only consider interfaces that
  are "up".  It has always ignored those configured with address
  0.0.0.0, thus ignoring "down" interfaces in most cases.

- add a list of interface pairs to --status output.

- replace signal(2) calls with sigaction(2) calls: glibc has further
  broken the semantics of signal(2).  We want slow system calls
  to be interruptable and not be restarted.

- improved error message for lack of preshared key by showing
  IDs that were used in lookup.  Collateral tidying.

- documented complex logic of routing and erouting in routing.txt

- When a connection is --routed, a route is installed to direct
  outbound packets destined for the peer's client to go to the ipsec
  interface implicit in the connection.  This is not allowed for Road
  Warrior connections -- they are strictly inbound.  For Opportunistic
  connections, a new default root is installed that points to the
  ipsec interface.  Note that the route does not take into account the
  source address.  In addition, a TRAP shunt eroute is installed to
  catch any traffic sent on the not-yet-initiated connection.

- When an outbound packet is trapped by the TRAP shunt eroute,
  Pluto attempts to initiate the corresponding connection.
  If the attempt fails because an Opportunistic connection was used
  and no gateway information was found, a PASS shunt eroute is
  installed (with no provision for automatic removal).

- if negotiation fails at a later point, a shunt eroute is installed.
  The kind of shunt is a policy decision.  It is specified as part of
  the connection.  (Untested; can not be specified in ipsec.conf.)

- When an inbound negotiation doesn't match a specific connection,
  and there is an Opportunistic connection, Pluto will try to respond
  using that connection.

- Every 2 minutes (SHUNT_SCAN_INTERVAL seconds), Pluto scans for

  + PASS eroutes that it installed that haven't been used recently.
    These will be deleted.

  + HOLD eroutes for which Pluto hasn't received a PFKEY_ACQUIRE message.
    For any it finds, it will attempt an opportunistic initiate.

- We no longer allow Phase 1 negotiations that we initiate to switch
  which connection is being used based on the Peer's Phase 1 ID.
  The fact that we used to do so was probably a bug.

- When Pluto as the Responder receives the Peer's Phase 1 ID, it is
  now willing to use a connection with a different ID and private
  RSA key for our end.  Nothing has committed us to the current guess.

- Changes in internal data structures to aid in better tracking of
  history of attempted and successful communication.  Needed for
  effective Opportunism.
  + struct host_pair represents information about pairs of hosts.
  + "Orientation" is done as soon as possible rather than as late
    as possible.
  + an arbitrary number of Quick Mode negotiations may now queue
    for the completion of a Main Mode negotiation.  Formerly,
    a negotiation could only use a previously completed Keying Channel.


Changes since 1.8 release by D. Hugh Redelmeier <hugh@mimosa.com>

- [Svenning Soerensen] correct check requiring OAKLEY_LIFE_TYPE
  attribute before OAKLEY_LIFE_DURATION.

- Improved whack diagnostics for various cases of failure to connect
  with Pluto.

- Added function state_with_serialno to allow a serial number to be
  used as a safe reference to a state object.  This avoids the risk
  of dangling pointers.

- Add crude support for PF_KEY SADB_ACQUIRE messages from the kernel.
  These cause the initiation of opportunistic negotiation.

- More processing of PF_KEY messages from the kernel.  We check that
  each message we send to the kernel elicits a response.  Other messages
  are queued to be processed at the head of the event loop.

- Fix bug in find_client_connection.  The manifestation is a Pluto
  segfault when a negotiation successfully gets to Phase 2, but there
  is no connection with the appropriate clients AND there is no
  wildcarded connection using our same host interface (not even an
  inappropriate one).

- Purely internal change: simplify EVENT_SO_DISCARD logic.

- Accept GROUP_DESCRIPTION in IPCA (IPcomp SA) proposal.  Although
  this SHOULD NOT appear according to
  draft-shacham-ippcp-rfc2393bis-05.txt 4.1, it should harmlessly help
  interoperation.

- Adjust to whether KLIPS supports IPCOMP.  If it does not, ignore
  --compress policy flag in Connections.  So Pluto would then neither
  propose nor accept IPCOMP transforms.


Changes since 1.7 release by D. Hugh Redelmeier <hugh@mimosa.com>

- Allow IPcomp transformations to use "well known CPIs".  This
  perversion allows better interoperation.

- Added rather experimental code to exploit IP_RECVERR.  This facility
  allows us to report who refuses our packets when they are refused.
  I don't understand the documentation of the mechanism, hence the
  experimental nature of our use of it.  Our code is pretty ugly too.

- [Svenning Soerensen] found a bug in the calculation of the verb for
  the updown script and proposed a fix.  The -host suffix was being
  used when the client subnet contained only one IP address, even if
  that address was not of the host.

- [Svenning Soerensen] Allow Phase 1 ID Payloads of type FQDN and
  user@FQDN to specify UDP/500 explicitly as per RFC2407 (IPSEC DOI)
  4.6.2.

- When responding in Quick mode, after switching to an appropriate
  connection based on the ID payloads, copy the connection's IPSEC
  policy flags into the state object.  Do this before consulting these
  flags.

- --status output now indicates when a connection is an instance
  of another.  "POLICY_" has been removed from the display of each
  policy flag, reducing clutter.  An extra space has been added to
  further indent lines in the --status report.

- Fixed a bug in displaying subnets in the error message
  "cannot respond to IPsec SA request..."

- In SA proposals, accept CPIs that are 4 bytes too, as long as
  the high order 2 bytes are 0.

- bug fix: no longer allocate tunnel SPIs below IPSEC_DOI_SPI_OUR_MIN.

- Added recognition (not support) for new Assigned Numbers for AES
  and friends.


Changes since 1.6 release by D. Hugh Redelmeier <hugh@mimosa.com>

- only accept a proposal with compression if the connection has the
  POLICY_COMPRESSION flag.  Normally, these flags don't affect what we
  propose.  In this case, we need this control since our IPCOMP
  implementation has a history of crashing.

- Fixed nasty bug in connections.c:fc_try().  For a Quick Mode
  exchange, if Pluto picked a connection as a starting point
  that had no peer client, it would not accept any proposed
  peer client.  On the other hand, if it picked a connection
  with a peer client, it would not accept any proposal without
  a peer client.


Changes since 1.5 release by D. Hugh Redelmeier <hugh@mimosa.com>

- added a global pluto option --uniqueids.  If this is enabled,
  whenever a new ISAKMP SA is established, any connections with
  the same Peer Id but different Peer IP address are unoriented,
  blowing away any of their SAs.  This should help eliminate
  dangling Road Warrior connections when the same Road Warrior
  reconnects.  It thus requires that IDs not be shared by hosts
  (a previously legal but probably useless capability).

- introduced clearer notation:
  + --host %any signifying any IP address, for Road Warrior,
    replacing 0.0.0.0 or 0::0
  + --nexthop %direct signifying "same IP as peer", replacing
    0.0.0.0 or 0::0
  + %any and %any6 as indices in ipsec.secrets to match IP addresses of
    Road Warriors (replacing 0.0.0.0 or 0::0)
  + --host %opportunistic signifying that the peer is actually
    to be discovered from the reverse DNS entry for the peer's client.
    This replaces --host 0.0.0.0 --client 0.0.0.0/32 (and IPv6 variants).

- be more strict about Phase 2 ID payloads (the ones that specify clients):
  reject if they specify protocol or port (which we do not support).

- Remove support for Diffie Hellman MODP 768 Group.  Increase
  support for MODP 1536 Group.

- Remove NO_RSA option -- patent expired!!

- Improve support for newer resolvers.

- [initially from Svenning Soerensen <svenning@post5.tele.dk>]
  support IPcomp (compression)

- [mostly Gerhard Gessler <gessler@iabg.de>] initial support for IPv6.

- As part of the IPv6 support, changes were made to the Pluto/updown
  interface.  See pluto(8) for the details.  One oft-requested feature
  is the new PLUTO_*_CLIENT environment variables.  The changes are
  "upward compatible", so the PLUTO_VERSION environment variable
  setting was changed from 1.0 to 1.1.  Unfortunately, this will break
  many customized updown scripts.

- Prototype support for initiating and responding to opportunistic
  negotiation.  A connection is considered for instantiation for
  opportunism if it has a peer of %opportunistic (the connection
  description must not specify a client for the peer).  Currently, the
  only way to provoke an opportunistic initiation is to use whack to
  simulate the interception of an outbound flow (do a "whack --help"
  and look at opportunistic initiation).  These features are not
  documented because they are experimental.  Limitations:  no actual
  interception of packets, DNS query synchronous.

- in ipsec.secrets, if multiple entries are the best match for the
  connection, they must all have the same secret.  In the past
  there was no code to compare RSA keys, so separate RSA entries
  were assumed to be different.  Now they are compared.

- Introduce now() to protect against clock being set backwards.  The
  risk is that events might be delayed a lot.  Still no protection
  against clock being moved forward.

- Don't "discover" interfaces that share IP addresses with others.
  This avoids an assertion failure.  Eventually, Pluto will have to
  ask KLIPS about interfaces.

- prevent infinite loop decoding certain malformed DNS records

- explain "Phase 1", "Phase 2", "Main Mode", and "Quick Mode"

- fiddled with "max number of retransmissions" diagnostic to add a bit
  of explanation.


Changes since 1.4 release by D. Hugh Redelmeier <hugh@mimosa.com>

- reorganized how messages are logged.  More of the serious messages
  are distinguished with the code RC_LOG_SERIOUS and hence should
  make it through ipsec auto's filter.

- Reserve Message IDs only within their ISAKMP SA.  This eliminates
  the unbounded memory requirement when reserving them per peer.

- Pluto's retransmission logic has been improved:
  + the initial retransmission occurs after 10 seconds of silence,
    down from 30 seconds.  The theory is that this will ungum a
    lost-packet situation more quickly
  + the delay after each retransmission is twice the delay before
    it -- exponential backoff
  + In the special case of the first message initiating Main Mode,
    when --keyingtries is 0 (meaning unlimited retries), Pluto
    will attempt more retransmissions at the same rate (no
    exponential backoff).  This cuts down on the pointless
    busywork while a peer isn't responding.

- Pluto will no longer generate SPIs in the range 0x100-0xFFF.
  This has the effect of reserving this range for manual keying.
  Of course Pluto will still allow its peer to use this range.

- Fixed another bug in Road Warrior support.  In responding to Phase 2
  / Quick Mode, once the client subnets (if any) are known, Pluto must
  reselect which connection to use.  If it didn't happen to be using
  the right one already, and no ID was explicitly specified for the
  peer, and the right one is a Road Warrior connection, the right one
  would not be found.


Changes since 1.3 release by D. Hugh Redelmeier <hugh@mimosa.com>

- Pluto can now acquire a public key from DNS.  It must be told
  to do so.  Hint: --pubkeyrsa is optional when specifying keyid.

- On the Responder, if a connection is to be routed, and the peer has
  a client that is a fixed subnet, and that subnet is already routed
  for other connections, and that route conflicts, Pluto will unorient
  the old connections (deleting the SAs that depend on the old route)
  on the theory that they have been superseded.  Too bad we can't
  otherwise tell when a connection is outdated.

- Support for netlink has been removed.  We always use PFKEYV2.
  Pluto no longer #includes any kernel headers!

- Added a TODO file

- Road Warrior support is unconditionally included.  No more need to
  define ROAD_WARRIOR_FUDGE.

- Fixed bug preventing Road Warrior connections being instantiated
  during the connection reselection prompted by receipt of Phase 1 ID
  Payload [Kai Martius <kai@secunet.de>].  Fixed bug that caused Phase
  1 ID to be ignored by connection reselection prompted by receipt of
  Phase 2 client IDs.


Changes since 1.2 release by D. Hugh Redelmeier <hugh@mimosa.com>

- fixed deficiencies in id handling

- changed to use updown script for routing (and firewalling)

- In quick Mode, when Responder, avoid selecting same SPIs as
  initiator.  This prevents KEYMAT being the same in both directions.
  See Ferguson and Schneier: "A Cryptographic Evaluation of IPsec",
  http://www.counterpane.com/ipsec.pdf, 5.6 #2.

- In Quick Mode, when Responder, install inbound IPsec SA one
  message earlier.  This eliminates the chance of a message being
  sent before the SA is established.

- slight complication to RSA private key lookup rules to allow
  match to an entry with multiple identities for the host.

- support per-connection debugging flags

- more use of PFKEY (RGB+DHR)

- inbound SAs are now spigrped and an inbound IPIP SA is created
  if tunneling is used.  This more symmetric with outbound processing
  and it allows KLIPS to check that the correct SAs are all applied.

- The way SA lifetime limits are proposed and accepted is better
  documented.  whack now complains when a specified value exceeds the
  limit.


Changes since 1.1 release by D. Hugh Redelmeier <hugh@mimosa.com>

- Updated constants to track newer IETF drafts

- added support for RSA Signature authentication
  + augmented demux.c to support packet syntax differences
    due to authentication technique.
  + preshared.c now can record RSA private keys
  + whack --keyid --pubkeyrsa records RSA public keys
  + whack --unlisten to allow a sequence of whack operations to be atomic
    with respect to IKE traffic (eg. loading public keys)
  + ipsec_doi.c will now do RSA Signature authentication
  + new policy bits are added to select authentication method (--rsasig, --psk)

- started towards more general ids (@FQDN and user@FQDN,
  in addition to IP addresses).
  + Note: there is *no* meaning attached to the id used beyond
    being an identifier.  Almost no syntax checking is done.
  + these forms of id work in:
    o ipsec.secrets indices
    o whack's --keyid for defining public keys
    o id payloads (generated and accepted)
    o --id option for each side in a connection description
  + the Id may be an IP address that isn't that of one end
    (but it must authenticate)
  + once and ID payload is received, Pluto will reconsider which
    potential connection should be used.  It makes sure that any
    authentication already done would apply to the new connection
    too.  This should make RSASIG + Road Warrior useful.

- [RGB, Peter Onion, and DHR] start of PFKEY2 support


Changes since 1.00 release by D. Hugh Redelmeier <hugh@mimosa.com>

- revamped rekeying:
  + added --rekeyfuzz; defaults to 100% so lifetime must now be
    more than twice rekeymargin
  + added rekeying for responders (but rigged to favour initiators)
  + [BUGFIX] responder of an exchange will not reinitiate
    the exchange if it does not complete

- Renamed --rekeywindow to --rekeymargin to match ipsec.conf.
  The old name will be accepted for a while.

- improved error and debugging messages

- updated list of notification messages (but we still don't support
  them).

- In ID payload, support range representation, but only for a subnet.
  This may improve interoperability

- scatter asterisks in debugging code to support EMACS outline mode.

- many internal changes were made to improve to code.  This should
  make it easier to add new states.  There should be few behaviour
  changes.

- whack --status now shows the SPIs for established SAs.

- [BUGFIX] DH values are now represented with the length specified by
  the group description, not the length actually needed.  About one
  time in 256, this will make a difference.  In those cases, the new
  Pluto won't interoperate with old Plutos.  It looks as if this
  change brings us in line with other IKE daemons.  Added a fudge
  (select with DODGE_DH_MISSING_ZERO_BUG) so that when a problem
  arises, a new replacement exchange is initiated (idea from John
  Gilmore).

- [BUGFIX] whack no longer assumes that UNIX domain sockets preserve
  record boundaries (they don't).  This faulty assumption caused
  whack's exit status to be unreliable

- [BUGFIX] pluto now correctly defaults the client subnet in a
  connection created for a Road Warrior exchange.

- [BUGFIX] Road Warrior code now supports multiple connections
  terminating in a particular Road Warrior node (allowing all
  appropriate combinations of host and subnets to be simultaneously
  connected).

- [BUGFIX] fix various peculiar Road Warrior crashes.

- [BUGFIX] fix spurious deletion of control socket when lock could
  not be acquired (Thomas Bellman <bellman@cendio.se>)

- [BUGFIX] interface discovery properly ignores nonAF_NET interfaces


Changes since .92 release by D. Hugh Redelmeier <hugh@mimosa.com>

- Communication between whack and pluto is now done using UNIX domain
  sockets.  This channel can be secured!

- liberalized ISAKMP SA acceptance.  Now anything up to and including
  16 bytes long is accepted.  How silly.

- All ISAKMP messages in UDP packets generated by pluto are now
  explicitly padded to be a multiple of 4 octets long.  This was wrong
  if certain big numbers (eg. nonces) happened to have leading zero
  octets.

- set socket option SO_REUSEADDR on pluto's whack socket.  This allows
  Pluto to quickly restart.

- Use new, consistent notation for topology:
	client===host---nexthop...nexthop---host===client

- prefix every line of status output with connection name.  This
  allows selection of output using grep.

- Replaced system's assert with passert.  This sends the diagnostic
  to syslog.

- Changed secrets file name processing to support sh-like "globbing"
  for file names.

- Where appropriate, log messages are prefixed by their connection
  name and state object serial numbers.  Connection names are quoted
  with double quotes and serial numbers are prefixed with the number
  sign (#).  Otherwise, where appropriate, log messages are prefixed
  by the IP address and port number from which the current message
  was sent.

- some attempt at making the messages more helpful
  + warnings when authentication (preshared secrets) failure is
    likely cause of the observed symptom
  + status message now highlights which SAs are the most recent
    (those are the ones that are subject to rekeying)and which are
    erouted.
  + state names are slightly improved
  + status message prints the "meaning" of a state after its name.

- the policy options of a connection (--pfs, --authenticate, --encrypt,
  (but not --tunnel)) now apply to negotiations being responded to.
  They continue to apply to negotiations initiated by Pluto.

- The Oakley group used for PFS in Phase 2 is dictated by the initiator.
  We used to dictate one of our choice.  To increase the chances for
  success, we now dictate the same group as was used in Phase 1.

- First, some context.  The "negotiated" lifetime of an SA is actually
  dictated by the initiator.  If the responder doesn't like this
  lifetime, it can tell the initiator in a NOTIFY message.  Pluto
  doesn't do this.  Instead, it will just expire the lifetime sooner
  than negotiated.  In the past, Pluto only initiated rekeying if it
  was the initiator.  Now, a responder Pluto will initiate rekeying if
  it is going to expire the SA earlier than negotiated.  To prevent an
  explosion of SAs, rekeying will only be done if the SA is the newest
  one for its connection.  Rekeying of IPsec SA will respect the
  security properties of the old SA at the level of policy options
  (i.e.  --pfs, --authenticate, --encrypt, --tunnel).

- Replaced --rekeytries with --keyingtries.  This option now applies
  to initial keying as well as rekeying, hence the name change.  Even
  though initial keying will now try more than once, whack logging
  will be stopped after the first attempt.  The value 0 is taken to
  mean, effectively, infinity: don't give up.


Changes since .91 release by D. Hugh Redelmeier <hugh@mimosa.com>

- A hack has been added to support mobile or anonymous initiators.

- The isakmp-secrets file has been renamed ipsec.secrets and the
  format spruced up to aid scalability.  Entries now can be shared
  between relevant machines verbatim.  An include facility was added.
  The file is now only read upon --listen commands.

- If --firewall appears on our end of a connection, Pluto will
  add a firewall rule to enable appropriate forwarding, without
  masquerading for any route it adds.  It will delete the rule
  when it deletes the route.

- When Pluto thinks whack's message is malformed, it now says so
  to whack, not just syslog.

- In addition to the messages traditionally sent back to whack,
  non-debugging messages sent to the log that relate to whack's
  current activity are copied to whack.  Whack's exit status now
  reflects the last message (if any) returned by Pluto.  This should
  allow a script to tell, for example, if an SA was established.

- top-level payload parsing has been centralized.  This should make
  it easier to add new features.  Payload ordering constraints have
  are now just those required by RFC2409 (IKE).  In most cases,
  Pluto will now ignore duplicated packets.  It should recover better
  from the reception of a corrupt packet.

- Interface discovery is more clever.  It notes each configured
  interface with a name ipsec[0-9] as a virtual public interface and
  considers any interface with a different kind of name and the same
  IP address to be the corresponding real public interface.  This is
  only done when Pluto starts, so any interfaces of interest must be
  configured before then.  This feature allows Pluto to support multiple
  public networks.

- Pluto now exploits the fact that eroutes only conflict if their
  local clients AND peer clients are the same.  So we can now support
  multiple subnets behind our security gateway all talking to clients
  behind another security gateway.

- Switched to using ipsec_spi_t to represent SPIs.  In the process
  fixed a related bug found by Peter Onion.


Changes since .9 release by D. Hugh Redelmeier <hugh@mimosa.com>
[incomplete]

- Message IDs are now random, rather than counting up from 1.  This
  should help keep messages in different but simultaneous Phase 2
  exchanges being mixed up.
- syslogged informative (i.e. debugging) messages are now prefixed
  with "| " to make the easier to ignore.
- forbid zero cookies.  Among other things, this prevents feedback
  confusing Pluto.
- Use serial numbers to cause most recent of available ISAKMP SAs
  to be chosen.  Also useful in debugging output.
- Pluto will now only listen to Whack on the loopback interface
  (important security limitation).
- implement rekeying, based on time, for ISAKMP and IPsec SAs
- Whack now talks to Pluto using TCP.  This allows status information
  to be returned to Whack.  For now, not much interesting is sent back.
  The TCP port is the *same* as the IKE port -- no longer 1 greater.
  Pluto closes the socket once the "goal" is established or the
  state object is freed.  All this will evolve.
- For SAs that were initiated by Pluto, Pluto will try to replace
  the SA before it expires.  There is a 10 minute window
  (SA_REPLACEMENT_WINDOW) in which this can occur.
- Support --peer_nexthop for initiator of ISAKMP SA.
- Support --optionsfrom <file>
- be more specific about error conditions: for each STF_FAIL,
  designate which notification message most applies.
- use these results in reporting to whack
- make whack back-talk look like FTP messages
- add and use notion of (potential) connection database.  All scripts change!
- fix handling of the variable form of attribute
- don't allow --initiate before --listen
- use new number for ESP_NULL
- demand each transform include an ENCAPSULATION_MODE attribute
- demand each AH transform include an appropriate AUTH_ALGORITHM attribute
- add not-yet-standardized OAKLEY_GROUP 5 (MODP 1536)
- since KLIPS only allows one IPsec SA to be routed to a particular
  subnet (for a peer's client), detect when a subnet is engaged.
  If we are replacing that SA, OK.  Otherwise, balk.
- [experimental] exploit the new UDP 500 hole to support host mode.
- add --route and --unroute: hysteresis in routing should prevent
  packets flowing in the clear during IPsec SA transitions.
- add --status to display the internal state of Pluto.
- deleted misleading README; other resources fulfill its role
- eliminated EVENT_CLEANUP: using EVENT_RETRANSMIT seemed more correct
- gave special meaning to combination of delete and add
- improved and documented combinations of whack command types
- improved logging
- added and used LEAK_DETECTIVE.  Fixed some leaks.


Changes between .85 and .9 release by D. Hugh Redelmeier <hugh@mimosa.com>
[incomplete]

- change pluto and whack's argument processing to use getopt_long:
  the syntax and expressive power is quite different.
- allow selection of debugging output.  Change pluto to accept
  arguments for specifying this.  Change whack and pluto allow
  settings to change during a run.
- make most controls for debugging run-time rather than compile-time.
  This required the addition of many command line arguments (see README)
- support 3DES encryption of Oakley messages (OAKLEY_3DES_CBC)
- accept modestly long attribute values (32 bits) for
  OAKLEY_LIFE_DURATION and SA_LIFE_DURATION.


Changes between .7alpha and .85 (highlights) by D. Hugh Redelmeier <hugh@mimosa.com>

- support RHL5.0 (glibc): avoid clash between
  <netinet/in.h> and <asm/byteorder.h>
- Lessons from porting to Solaris: endianness, careful typing, alignment,
  correct fd_set bugs in call_server(), correct rnd.c to use sig_atomic_t
- Makefile: add distlist target to put out names of files in distribution
- Makefile: when installing binaries, move old ones to .OLD
- add and exploit pb_stream mechanism for systematically decoding and
  encoding packets
- More flexible security policy, but still hard-wired.
- support new Oakley group (2 -- modp 1024)
- make returned IPsec proposal for acceptance a copy of winning proposal
  (as per spec)
- add and use generic interface to hash functions
- add many comments referencing the draft standards
- change all uses of stdout to use stderr instead (choice between stdout
  and stderr was haphazard)
- fix SPI stuff: Oakley and IPSEC SPIs are different beasts
- generate initial IPSEC SPI as a random number (avoid clashes)
- fix layout bug for struct isakmp_transform
- fix several dangerous memory allocation and buffer overflow errors;
  eliminate all inline uses of calloc (use a wrapper)
- avoid memory leak due to uncleared mpz variables
- general tidying and restructuring; get rid of many "magic" numbers
- de-lint everywhere (add -Wall -Wmissing-prototypes to Makefile)
- switch from BSD b* functions to ANSI mem* functions
- get rid of bitfields
- generate the correct amount of keying material (PRF feedback, if needed)
- improve get_preshared_key (new format too)
- improve handling of informational exchanges.  Still poor.
- improve tracing output
- print version information (whack and pluto)
- wherever an enum-like value is printed, print the name of the value
- make duplicate_state() duplicate the st_myidentity_type field.
- make kernel interface do required route commands
- open and close /dev/ipsec more carefully
- support separate keys for esp encryption and esp authentication