well (and is a good demonstration of how encapsulating the SSL in a
memory-based state machine can make it easier to apply to different
situations).
The change implements a new command-line switch "-flipped <0|1>" which, if
set to 1, reverses the usual interpretation of a client and server for SSL
tunneling. Normally, an ssl client (ie. "-server 0") accepts "cleartext"
connections and conducts SSL/TLS over a proxied connection acting as an SSL
client. Likewise, an ssl server (ie. "-server 1") accepts connections and
conducts SSL/TLS (as an SSL server) over them and passes "cleartext" over
the proxied connection. With "-flipped 1", an SSL client (specified with
"-server 0") in fact accepts SSL connections and proxies clear, whereas an
SSL server ("-server 1") accepts clear and proxies SSL. NB: most of this
diff is command-line handling, the actual meat of the change is simply the
line or two that plugs "clean" and "dirty" file descriptors into the item
that holds the state-machine - reverse them and you get the desired
behaviour.
This allows a network server to be an SSL client, and a network client to
be an SSL server. Apart from curiosity value, there's a couple of possibly
interesting applications - SSL/TLS is inherently vulnerable to trivial DoS
attacks, because the SSL server usually has to perform a private key
operation first, even if the client is authenticated. With this scenario,
the network client is the SSL server and performs the first private key
operation, whereas the network server serves as the SSL client. Another
possible application is when client-only authentication is required (ie.
the underlying protocol handles (or doesn't care about) authenticating the
server). Eg. an SSL/TLS version of 'ssh' could be concocted where the
client's signed certificate is used to validate login to a server system -
whether or not the client needs to validate who the server is can be
configured at the client end rather than at the server end (ie. a complete
inversion of what happens in normal SSL/TLS).
NB: This is just an experiment/play-thing, using "-flipped 1" probably
creates something that is interoperable with exactly nothing. :-)
Make ca.c correctly initialize the revocation date.
Make ASN1_UTCTIME_set_string() and ASN1_GENERALIZEDTIME_set_string() set the
string type: so they can initialize ASN1_TIME structures properly.
client code certificates to use to only check response signatures.
I'm not entirely sure if the way I just implemented the verification
is the right way to do it, and would be happy if someone would like to
review this.
Bleichenbacher's DSA attack. With this implementation, the expected
number of iterations never exceeds 2.
New semantics for BN_rand_range():
BN_rand_range(r, min, range) now generates r such that
min <= r < min+range.
(Previously, BN_rand_range(r, min, max) generated r such that
min <= r < max.
It is more convenient to have the range; also the previous
prototype was misleading because max was larger than
the actual maximum.)
Bleichenbacher's DSA attack. With this implementation, the expected
number of iterations never exceeds 2.
New semantics for BN_rand_range():
BN_rand_range(r, min, range) now generates r such that
min <= r < min+range.
(Previously, BN_rand_range(r, min, max) generated r such that
min <= r < max.
It is more convenient to have the range; also the previous
prototype was misleading because max was larger than
the actual maximum.)
