cleaned up server code, renamed some files & classes
renamed demo-server.py and demo-host-key to demo_server.py and
demo_host_key, just to be consistent.
renamed SSHException -> SecshException.
generalized the mechanism where Channel decides whether to allow
different channel requests: 4 of the main ones (pty, window-change,
shell, and subsystem) go through easily override-able methods now.
you could probably make an actual ssh shell server.
gave ChannelFile a repr().
turned off ultra debugging in the demos. demo_server creates a
subclass of Channel to allow pty/shell and sets an event when the
shell request is made, so that it knows when it can start sending
the fake bbs.
renamed to charmander and updated some of the distutils files.
notes about the exported api
just wrote some quick notes (for a few of the classes) about which
methods are intended to be the exported API. python has no decent
way of distinguishing private vs public.
big chunk of work which makes server code 95% done
fixed auth check methods to return just a result (failed, succeeded,
partially succeeded) and always use get_allowed_auths to determine the
list of allowed auth methods to return.
channel's internal API changed a bit to allow for client-side vs.
server-side channels. we now honor the "want-reply" bit from channel
requests. in server mode (for now), we automatically allow pty-req
and shell requests without doing anything.
ChannelFile was fixed up a bit to support universal newlines. readline
got rewritten: the old way used the "greedy" read call from ChannelFile,
which won't work if the socket doesn't have that much data buffered and
ready. now it uses recv directly, and tracks the different newlines.
demo-server.py now answers to a single shell request (like a CLI ssh
tool will make) and does a very simple demo pretending to be a BBS.
transport: fixed a bug with parsing the remote side's banner. channel
requests are passed to another method in server mode, to determine if
we should allow it. new allowed channels are added to an accept queue,
and a new method 'accept' (with timeout) will block until the next
incoming channel is ready.
change kex-gex server code to generate primes by hand
added a util function "generate_prime" to compare to the incredibly slow C
version, but it's no faster of course. i think kex-gex from the server is
just not going to be feasible without having a separate thread generate some
primes in the background to have handy when a request comes in. so in short,
this still doesn't work.
also i put bit_length into util and a tb_strings function which gets stack
traceback info and splits it into a list of strings.
Robey Pointer