NetBSD/crypto/dist/kame/racoon/TODO

147 lines
5.9 KiB
Plaintext

$KAME: TODO,v 1.35 2000/10/04 17:40:58 itojun Exp $
Please send any questions or bug reports to snap-users@kame.net.
TODO list
URGENT
o The documents for users convenience.
o split log file based on client. printf-like config directive, i.e.
"logfile racoon.%s.log", should be useful here.
-> beware of possible security issue, don't use sprintf() directly!
make validation before giving a string to sprintf().
o fix flag for logging
o save decrypted IKE packet in tcpdump format
o IPComp SA with wellknown CPI in CPI field. how to handle it?
o hunt for memory leaks
o better rekey
MUST
o Certificate type handling. e.g. CRL, CR, PGP...
o multiple certificate payload handling.
o To consider the use with certificate infrastructure. PXIX ???
o Support scoped (link-local and site-local) IPv6 addresses.
It needs pfkey modification for scope IDs, and advanced API (RFC2292)
related mods in racoon itself.
o kmstat should be improved.
o aggressive mode DH group checks/handling on responder side.
o Informational Exchange processing properly.
o mandatory configuration check.
o require less configuration. phase 2 is easier (as kernel presents racoon
some hints), phase 1 is harder. for example,
- grab phase 2 lifetime and algorith configuration from sadb_comb payloads in
ACQUIRE message.
- give reasonable default behavior when no configuration file is present.
- difficult items:
how to guess a reasonable phase 1 SA lifetime
(hardcoded default? guess from phase 2 lifetime?)
guess what kind of ID payload to use
guess what kind of authentication to be used
guess phase 1 DH group (for aggressive mode, we cannot negotiate it)
guess if we need phase 2 PFS or not (we cannot negotiate it. so
we may need to pick from "no PFS" or "same as phase 1 DH group")
guess how we should negotiate lifetime
(is "strict" a reasonable default?)
guess which mode to use for phase 1 negotiation (is main mode useful?
is base mode popular enough?)
o psk.txt should be a database? (psk.db?) psk_mkdb?
SHOULD
o how to support multi interfaces ?
- perform a dummy connect(2), like ping6.
- bind multi address, like named.
now racoon uses the former way.
o DSA signature authentication method.
o more acceptable check.
o New group mode should be considered.
o Dynamically retry to exchange and resend the packet per nodes.
o Remote maintenance with secure connection.
o To make the list of supported algorithm by sadb_supported payload
in the SADB_REGISTER message which happens asynchronously.
o fix the structure of ph2handle.
We can handle the below case.
node A node B
+--------------SA1----------------+
+--------------SA2----------------+
at node A:
kernel
acquire(A-B) ------> ph2handle(A=B) -----> ph1handle
|
policy
A=B
A=B
But we can not handle the below case because there is no x?handle.
node A node B node C
+--------------SA1----------------+
+------------------------------------------------SA2---------------+
at node A:
kernel
acquire(A-C) ---+---> x?handle ---+---> ph2handle(A=B) -------> ph1handle
| | |
acquire(A-B) ---+ policy +---> ph2handle(A=C) -------> ph1handle
A=B
A=C
o consistency of function name.
o deep copy configuration entry to hander. It's easy to reload configuration.
o don't keep to hold keymat values, do it ?
o local address's field in isakmpsa handler must be kicked out to rmconf.
o responder policy and initiator policy should be separated.
o for lifetime and key length, something like this should be useful.
- propose N
- accept between X and Y
o wildcard "accept any proposal" policy should be allowed.
o replay prevention
- limited total number of session
- limited session per peer
- number of proposal
o full support for variable length SPI. quickhack support for IPComp is done.
MAY
o Effective code.
o interaction between IKE/IPsec and socket layer.
at this moment, IKE/IPsec failure is modeled as total packet loss to other
part of network subsystem, including socket layer. this presents the
following behaviors:
- annoyingly long timeouts on tcp connection attempt, and IKE failure;
need to wait till tcp socket timeouts.
- blackhole if there's mismatching SAs.
we may be able to give socket layer some feedback from IKE/IPsec layer.
still not sure if those make sense or not.
for example:
- send PRU_HOSTDEAD to sockets if IKE negotiation failed
(sys/netkey/key.c:key_acquire2)
to do this, we need to remember which ACQUIRE was caused by which socket,
possibly into larval SAs.
- PRU_QUENCH on "no SA found on output"
- kick tcp retransmission timer on first SA establishment
o IKE daemon should handle situations where peer does not run IKE daemon
(UDP port unreach for port 500) better.
should use connected UDP sockets for sending IKE datagrams.
o rate-limit log messages from kernel IPsec errors, like "no SA found".
TO BE TESTED.
o IKE retransmit behavior
see, draft-*-ipsec-rekeying*.txt
o Reboot recovery (peer reboot losing it's security associations)
see, draft-*-ipsec-rekeying*.txt
o Scenarios
- End-to-End transport long lived security associations
(over night, data transfer >1Gb) with frequent dynamic rekey
- End-to-GW tunnel long lived security associations
(over night, data transfer >1Gb) with frequent dynamic rekey
- Policy change events while under SA load
- End-to-End SA through IPsec tunnels, initiation both ways
- Client End-to-End through client-to-GW tunnel SA, initiate from
client for tunnel, then initiation both ways for end-to-end
- Client-to-GW transport SA for secure management
o behavior to receive multiple auth method proposals and AND proposal
and to be written many many.