--- gvpe/NEWS 2013/07/17 04:36:09 1.115 +++ gvpe/NEWS 2015/07/30 19:03:44 1.127 @@ -5,39 +5,56 @@ TODO: replace ripemd160 as the only authentication hash. TODO: increase rsa size. TODO: replace transport bits by transport endpoint structs? -TODO: ecdh to avoid session replay attacks TODO: http://incog-izick.blogspot.de/2011/08/using-openssl-aes-gcm.html TODO: http://stackoverflow.com/questions/12153009/openssl-c-example-of-aes-gcm-using-evp-interfaces -proposed: 3 types, req, resreq, res - -req (hmac1) rsa(seqno1 hmac1 aes1 seqno2 hmac2 aes2 auth) ecdh1 -res (hmac1) hash(rsa-contents) ecdh2 - -req hmac0(*) rsa(seqno hmac0 hmac aes auth) hkdf-salt ecdh1 -res hmac0(rsa-contents ecdh2) - hmac_key = hkdf(hkdf-salt, hmac | ecdh) - aes_key = hkdf(hkdf-salt, aes | ecdh) - -TODO: very much larger intiial seed -TODO: don't use RAND_pseudo_bytes :/ +TODO: verify +TODO: make sense of overhead calculation +TODO: if-up &c should not be scripts? +TODO: ipv6 +TODO: gvpectrl should not use default privatekey,. or maybe document it better - INCOMPATIBLE CHANGE: core protocol version 1.0. + - INCOMPATIBLE CHANGE: node sections are now introduced + with "node nodename", not "node = nodename". - while individual packets couldn't be replayed, a whole session could be replayed - this has been fixed by an extra key exchange. + - fix a delete vs. delete [] mismatch in the central logging function. - in addition to rsa key exchange and authentication, the handshake now adds a diffie-hellman key exchange (using curve25119) for perfect forward secrecy. mac and cipher keys are derived using HKDF. - rsa key sizes are now configurable and larger (default is 3072). correspondingly, the minimum mtu is no longer 296 but 576. - fixed a potential (unverified) buffer overrun on rsa decryption. + - new per-node low-power setting that tries to reduce cpu/network usage. + - router reconnects could cause excessive rekeying on other connections. - gvpectrl no longer generates all missing public keys, but only missing private keys. private keys are also put into the configured location. - the pid-file now accepts %s as nodename as elsewhere. + - switch to counter mode (only aes supported at the moment in + openssl). this gets rid of the need to generate a random iv, + is likely more secure (and, as a side effect, gets rid of + slow randomness generation. counter mode is often faster + then cbc mode as well, and packets are smaller). - no longer use RAND_bytes to generate session keys - you NEED a real source of entropy now (e.g. egd or /dev/random - see the openssl documentation). - + - multiple node statements for the same node are now supported + and will be merged. + - a new directive "global" switches back to the global section + of the config file. + - if-up scripts can now be specified with absolute paths. + - new global option: serial, to detect configuration mismatches. + - use HKDF as authentication proof, not HMAC or a plain hash + (hint by Ilmari Karonen). + - during rekeying or conenction establishments, hmac authentication + errors could occur and reset the connection. Transient hmac + authentication errors are now being ignored for 3 seconds. + - log the reason for a conneciton loss. + - use a (hopefully) constant time memcmp to compare internal secrets. + - fix a (harmless) errornous out of bounds stack read that would trigger + gcc's -fsanitize=address. + 2.25 Sat Jul 13 06:42:33 CEST 2013 - INCOMPATIBLE CHANGE: no longer enable udp protocol if no other protocols are enabled - this is necessary when you have nodes with @@ -56,7 +73,7 @@ to chroot to a specified or anonymous new root, and change user id. - new global configuration options seed_device and seed_interval, to configure another device than /dev/urandom for random seeds, - and to configure a regular imterval to reseed the rng. + and to configure a regular interval to reseed the rng. - prefer inet_aton over gethostbyname, as the latter is not guaranteed to "resolve" literal ip addresses. - configure didn't detect openssl 1.0 because SHA1_version became private @@ -71,7 +88,7 @@ - major, but incremental, dns transport improvements: - do not simply abort in some error cases in the dns transport, but try to recover. - - allow lowercase/uppercase alises for base-n encodings that do + - allow lowercase/uppercase aliases for base-n encodings that do not rely on case. - use base26 instead of base22 encoding for dns syn's, and base36 instead of base22 for headers (saves one byte/packet).