fragattacks/wpa_supplicant/todo.txt

To do:
- hostap: try other roaming modes
  NOTE: current mode (manual roaming) does not really roam at all..
  Firmware did not notice the current AP disappearing..
- add support for WPA with ap_scan=0 (update selected cipher etc. based on
  AssocInfo; make sure these match with configuration)
- consider closing smart card / PCSC connection when EAP-SIM/EAP-AKA
  authentication has been completed (cache scard data based on serial#(?)
  and try to optimize next connection if the same card is present for next
  auth)
- on disconnect event, could try to associate with another AP if one is
  present in scan results; would need to update scan results periodically..
- if driver/hw is not WPA2 capable, must remove WPA_PROTO_RSN flag from
  ssid->proto fields to avoid detecting downgrade attacks when the driver
  is not reporting RSN IE, but msg 3/4 has one
- Cisco AP and non-zero keyidx for unicast -> map to broadcast
  (actually, this already works with driver_ndis; so maybe just change
  driver_*.c to do the mapping for drivers that cannot handle non-zero keyidx
  for unicast); worked also with Host AP driver and madwifi
- IEEE 802.1X and key update with driver_ndis?? wpa_supplicant did not seem
  to see unencrypted EAPOL-Key frames at all..
- EAP-PAX with PAX_SEC
- EAP (RFC 3748)
  * OTP Extended Responses (Sect. 5.5)
- test what happens if authenticator sends EAP-Success before real EAP
  authentication ("canned" Success); this should be ignored based on
  RFC 3748 Sect. 4.2
- test compilation with gcc -W options (more warnings?)
  (Done once; number of unused function arguments still present)
- add proper support for using dot11RSNAConfigSATimeout
- ctrl_iface: get/set/remove blob
- use doc/docbook/*.sgml and docbook2{txt,html,pdf} to replace README and
  web pages including the same information.. i.e., have this information only
  in one page; how to build a PDF file with all the SGML included?
- EAP-POTP/RSA SecurID profile (RFC 4793)
- document wpa_gui build and consider adding it to 'make install'
- test madwifi with pairwise=TKIP group=WEP104
- possibility to link in WPA Authenticator state machine to wpa_supplicant
  (new PeerKey handshake, WPA2/IEEE 802.11 (RSN) IBSS)
- consider merging hostapd and wpa_supplicant PMKSA cache implementations
- consider redesigning pending EAP requests (identity/password/otp from
  ctrl_iface) by moving the retrying of the previous request into EAP
  state machine so that EAPOL state machine is not needed for this
- rfc4284.txt (network selection for eap)
- www pages about configuring wpa_supplicant:
  * global options (ap_scan, ctrl_interfaces) based on OS/driver
  * network block
  * key_mgmt selection
  * WPA parameters
  * EAP options (one page for each method)
  * "configuration wizard" (step 1: select OS, step 2: select driver, ...) to
    generate example configuration
- error path in rsn_preauth_init: should probably deinit l2_packet handlers
  if something fails; does something else need deinit?
- consider moving SIM card functionality (IMSI fetching) away from eap.c;
  this should likely happen before EAP is initialized for authentication;
  now IMSI is read only after receiving EAP-Identity/Request, but since it is
  really needed for all cases, reading IMSI and generating Identity string
  could very well be done before EAP has been started
- try to work around race in receiving association event and first EAPOL
  message
- helper function to do memcmp(addr, "\x00\x00\x00\x00\x00\x00", ETH_ALEN)
- add wpa_secure_memzero() macro and secure implementation (volatile u8*) to
  clear memory; this would be used to clear temporary buffers containing
  private data (e.g., keys); the macro can be defined to NOP in order to save
  space (i.e., no code should depend on the macro doing something)
- make sure that TLS session cache is not shared between EAP types or if it
  is, that the cache entries are bound to only one EAP type; e.g., cache entry
  created with EAP-TLS must not be allowed to do fast re-auth with EAP-TTLS
- consider moving eap_tls_build_ack() call into eap_tls_process_helper()
  (it seems to be called always if helper returns 1)
  * could need to modify eap_{ttls,peap,fast}_decrypt to do same
- add support for fetching full user cert chain from Windows certificate
  stores even when there are intermediate CA certs that are not in the
  configured ca_cert store (e.g., ROOT) (they could be, e.g., in CA store)


0.6.x branch:
- clean up common.[ch]
- change TLS/crypto library interface to use a structure of function
  pointers and helper inline functions (like driver_ops) instead of
  requiring every TLS wrapper to implement all functions
- add support for encrypted configuration fields (e.g., password, psk,
  passphrase, pin)
- wpa_gui: add support for setting and showing priority, auth_alg
  (open/shared for static WEP)

- cleanup TLS/PEAP/TTLS/FAST fragmentation: both the handshake and Appl. Data
  phases should be able to use the same functions for this;
  the last step in processing sent should be this code and rest of the code
  should not need to care about fragmentation at all
- test EAP-FAST peer with OpenSSL and verify that fallback to full handshake
  (ServerHello followed by something else than ChangeCipherSpec)