Retired by author, looking for new owner :) or some help!
The client appears to be very poor-performing under load. I wrote a basic server that is capable of sending a very large stream of data, "max headroom" demo at telnet 1984.ws, and it totally kills the client. Maybe the client is so busy processing incoming data that it is not allowing for tasks to read it from the other end?
Maybe this is async "backpressure"
outer telnetlib3-server and telnetlib3-client and examples should connect as exit(main(**parse_args(sys.argv))), the _transform_args() function is rather shoe-horned, main() should declare keywords.
this is completed for server, copy to client
We need a way to wish to wait for a state. For example, our client shell might await until local_echo is False, or remote_option[ECHO] is True to change terminal state to reflect it. A function wait_for, receiving a function that returns True when state is met, will be called back continuously after each block of data received containing an IAC command byte, but the boiler code simply returns the waiter.
This should allow us to spray the client with feature requests, and await the completion of their negotiation, especially for things like LINEMODE that might have many state changes, this allows asyncio to solve the complex "awaiting many future states in parallel" event loop easily
-- just accept a future, and on each state change, call an internal function that checks for equality for the parameters given, and when true, set .done()
after some thought, we should hardcode common ones, such as await negotiate_kludge() -> bool, wait_for_naws() -> bool, and, maybe a common wait_for_slc() that just returns some description of the state change that has occurred with maybe values
base_client.py and base_server.py actually share the same ABC base_protocol.py, they are almost mirror images of one another, which is pretty great, actually, so they can be reduced to BaseTelnetProtocol.
How do we write a server which suggests a matrix of preferred linemode negotiated with client, or client negotiated towards server? As a server, we simply honor whatever is requested, which may be wrong for the server shell interface designed.
LINEMODE compliance needs a lot of work. - possibly, we remove LINEMODE support entirely. I only know of one client,
BSD telnet, that is capable of negotiating -- this is the C code from which our implementation was derived!
- callbacks on TelnetServer needed for requesting/replying to mode settings
- the SLC abstractions and 'slc_simul' mode is difficult for the API.
- There are many edge cases of SLC negotiation outlined in the RFC, how comprehensive are our tests, and how well is our SLC working?
- IAC-SB-LINEMODE-DO-FORWARDMASK is unhandled, raises NotImplementedError
feed_byte called by telnet server should be a coroutine receiving data by send. It should yield out-of-bound values, None otherwise? 'is_oob', or 'slc_received', etc.? We're still considering ... the state still requires tracking, but this would turn multiple function calls into a .send() into generator, better for state loops or bandwidth, maybe?
handle_xon resumes writing in a way that is not obvious -- we should be using the true 'pause_writing' and 'resume_writing' methods of our base protocol. The given code was written before these methods became available in asyncio (then, tulip). We need to accommodate the new availabilities.
We've seen everything negotiate fine, but what exactly are we expected to do when the distant end's concept of our negotiation STATUS disagrees with our own? Match theirs, should we re-negotiate or re-affirm misunderstood values? The RFC is not very clear.
- _receive_status(self, buf) response to STATUS does not honor given state
- values. only a non-compliant distant end would cause such a condition. so it is decided to leave it as "conflict report only, no action always"
Also, when the other end says they are capable of STATUS, we unconditionally request them to go ahead and assert their status. We have no need to do this. We should probably make it a toggle option or strictly option of the callback to act on it..
- SLC flushin/flushout attributes are not honored. Not entirely sure how to handle these two values with asyncio yet.
In addition to remote line editing as described below, a pure-python shell, telsh is provided to allow toggling of server options and session parameters. In this way, it provides a suitable interface for testing telnet client capabilities.
It is only in the interest of this project to provide enough shell-like capabilities to demonstrate remote line editing and an extensible environment for session introspection. An example of this is assigning a new value to CHARSET, toggling in and outbinary, thereby enabling UTF8 input/output, etc.
CHARSET (`rfc-2066`_) specifies a codepage, not an encoding. At the time, this was more or less limited to specifying the codepage used to display bytes of the range 127 through 255. Unimplemented in BSD client, and generally found implemented only in recent MUD client (Atlantis) and servers. Most common values are: ASCII, UTF8, BIG5, and LATIN1.
The default preferred encoding for clients that negotiate BINARY but not
CHARSET, such as the BSD client, is defined by the TelnetServer keyword
argument default_encoding ('UTF8' by default).
The example shell telsh allows changing encoding on the fly by setting the 'CHARSET' session environment value at the telsh command prompt by issuing command:
set CHARSET=UTF8
Setting binary for only a single direction ('outbinary' or 'inbinary') is supported. Client support of one does not immediately toggle the other, it must be negotiated both ways for full UTF8 input and output.
Some clients (TinTin++) incorrectly negotiation either directions (WILL, DO/WONT, DONT) as a single option, causing only one reply for a request of either 'outbinary' or 'inbinary' for which it always declines, only once, for either request (Even when configured for UTF8).
Additionally, a contrib.cp437 module is included (authored by tehmaze) which translates output meant to be translated by DOS Emulating programs to their comparable UTF-8 font. This is used by argument --cp437 of the telnet-client_ program.
Some bulletin-board systems will send extended ascii characters (such as those used by
The Telnet protocol is over 40 years old and still in use today. Telnet predates TCP, and was used over a wide array of transports, especially on academic and military systems. Nearly all computer networking that interacted with human interfaces was done using the Telnet protocol prior to the mass-adoption of the World Wide Web in the mid 1990's, when SSH became more commonplace.
Naturally, Telnet as a code project inevitably must handle a wide variety of connecting clients and hosts, due to limitations of their networking Transport , Terminals, their drivers, and host operating systems.
This implementation aims to implement only those capabilities "found in the wild", and includes, or does not include, mechanisms that are suitable only for legacy or vendor-implemented options. It even makes one of its own: the encoding' used in binary mode is the value replied by the CHARSET negotation (`rfc-2066`_).
This project is the only known Server-side implementation of Special Linemode Character (SLC) negotiation and Remote line editing (`rfc-1184`_), other than BSD telnet, which was used as a guide for the bulk of this python implementation.
Remote line editing is a comprehensive approach to providing responsive, low-latency output of characters received over slow network links, allowing incomplete lines to be buffered, while still providing remote editing facilities, such as backspace, kill line, etc.
The Server and Client agree on a series of Special Linemode Character (SLC)
function values, to agree on the keyboard characters used for Backspace,
Interrupt Process (^C), Repaint (^R), Erase Word (^W), etc.
In kludge mode, SLC characters are simulated for remote editing, provide an almost readline-like experience for all telnet clients, except those that perform only local editing, which are unaffected.
The sequence sent by server, WILL-SGA, WILL-ECHO enables "kludge
mode", a form of line mode editing that is compatible with all minimally
implemented telnet clients. This is the most frequent implementation used by
Windows 98 telnet, SyncTerm, netrunner, or TinTin++ to provide
character-at-a-time editing.
Consider that kludge mode provides no way to determine which bytes, received at any indeterminate time, of any indeterminate length, or none at all, are received as the result of which input characters sent.
Accordingly, with Suppress Go-Ahead (SGA) enabled, there can be any
indeterminable state: (1) the remote program is hung, (2) receiving and/or
processing, (3) has responded with output but not yet received by transport,
and (4) has received some, but not yet all output by transport.
This is detrimental to a user experience with character-at-a-time processing, as a user cannot know whether the input was legal, ignored, or not yet replied to, causing some frustration over high latency links.
The IAC-GA signal would seemingly be of little use over today's
bi-directional TCP protocol and virtual terminal emulators -- its original
purpose was to coordinate transmission on half-duplex protocols and terminals.
Only a few 1970-era hosts (AMES-67, UCLA-CON) require the IAC-GA
signal. For this reason, this server takes the modern recommendation of
suppressing the IAC-GA signal (IAC-WILL-SGA) by default; those
clients wishing to make use of the IAC-GA signal must explicitly request
IAC-DONT-SGA to enable the IAC-GA signal.
The IAC-GA signal has been recently restored for character-at-a-time servers,
such as the competition nethack server alt.nethack.org, targeted at client
scripts that play using AI decision-making routines.
Later note: actually this seems to be common with MUD's, will require some experimentation with 'mudlet' client.
Unless otherwise negotiated, the specification describes Telnet's default mode as half-duplex, local line editing. This most basic "dummy" mode is modeled after a Teletype 33, which runs in "half-duplex" mode.
A Telnet implementation attached to 7-bit ASCII teletype may implement the
Telnet protocol by hardware circuit, or by minimal changes to their terminal
line drivers: when the connecting CPU is without MMU or process control, an
IAC interpreter or hardware device could be "interrupted" when the 8th bit is
set high, "Out of band" in regards to 7-bit terminals, the receipt of value
255 indicates that the byte following it Is-A-Command (IAC).
- Each end transmits only 7-bit ASCII, (except as used in the interpreter).
- A server's prompt must be followed by the 'Go-Ahead' (
IAC-GA) command.- Client signals end of input (send) by CR, LF (Carriage Return, Linefeed).
A supervisor connecting a (7-bit) teletype to a telnet (8-bit) data line would simply pipe the streams together by the 7 bits; The teletypist may press 'BREAK' at any time to signal a control line: the supervisor then enters Telnet Synch" mode by sending an "Urgent" mechanism, and ceases printing data received on the transport.
A user could then instruct "Abort Output" (IAC-AO), "Interrupt Process"
(IAC-IP), or others, and then presumably return to normal processing.
Consider the description of a PDP-10 session in `rfc-139`_ (May 1971), presented here as a simple unix session:
Teletype sends command input:
find /usr -name 'telop.c'<CR>Server begins output -- perhaps, after some minutes of pause, many rows of 'Permission Denied'. Meanwhile, the user has already filled his teletype's input buffer, and later deciding to abort the previous program:
ed /usr/local/s^t/tel^t^c
At this point, however, the half-dupex Teletype cannot transmit any input.
The only way to signal the attention of the supervisor, which is currently
blocking the half-duplex transmission with output (having not yet received
IAC-GA), is by a special line signal wired separately from the teletype
keyboard. This is the BREAK or ATTN key.
The terminal driver may then signal the 'supervisor', which then sends INS
(`rfc-139`_). Although the teletype is capable of "flushing" its input buffer,
it does not flush control codes. Remaining control codes from the teletype
(^t^t^c) continues to the remote end, but is discarded by that end, until
the Data-Mark (IAC-DM) is sent by the supervisor.
This ensures the ^t and ^c characters are not received by the remote
program.
In the TCP implementation of telnet, where presumably a half-duplex terminal
may still interconnect, the INS marker referenced in pre-TCP documents is,
instead, marked by sending the TCP Urgent option:
socket.send(IAC, socket.MSG_OOB).
The value of the byte does not seem to matter, can be of any length, and can
continue sending socket.MSG_OOB (presumably, along with the remaining
^t^t^c described previously). The BSD server sends only a single byte:
/* * In 4.2 (and 4.3) systems, there is some question about * what byte in a sendOOB operation is the "OOB" data. * To make ourselves compatible, we only send ONE byte * out of band, the one WE THINK should be OOB (...)
All input is discarded by the IAC interpreter until IAC-DM is received;
including IAC or 8-bit commands. This was used to some abuse to "piggyback"
telnet by breaking out of IAC and into another "protocol" all together, and is
grieved about in `rfc-529`_:
The Telnet SYNCH mechanism is being misused by attempting to give it meaning at two different levels of protocol.
The BSD client may be instructed to send this legacy mechanism by escaping and
using the command send synch:
telnet> send synch
This sends IAC marked MSG_OOB, followed by DM, not marked
MSG_OOB. The BSD server at this point would continue testing whether the
last received byte is still marked urgent, by continuing to test errorfds
(third argument to select select, a modern implementation might rather use
sockatmark(3)).
BSD Telnet Server sets "Packet mode" with the pty driver:
(void) ioctl(p, TIOCPKT, (char *)&on);
And when TIOCPKT_FLUSHWRITE is signaled by the pty driver:
#define TIOCPKT_FLUSHWRITE 0x02 /* flush packet */
Awaiting data buffered on the write transport is cleared; taking care to ensure all IAC commands were sent in the netclear() algorithm, which also sets the neturgent pointer.
There are five supported signaling mechanisms for "send" or "end of line"
received by clients. The default implementation supplies remote line editing
and callback of line_received with all client-supported carriage returns,
but may cause loss of data for implementors wishing to distinguish among them.
Namely, the difference between 'return' and 'enter' or raw file transfers.
Those implementors should directly override data_received, or carefully
deriving their own implementations of editing_received and character_received.
An overview of the primary callbacks and their interaction with carriage returns are described below for those wishing to extend the basic remote line editing or 'character-at-a-time' capabilities.
CR LF(Carriage Return, Linefeed): The Telnet protocol defines the sequenceCR LFto mean "end-of-line". The default implementation strips CL LF, and firesline_receivedon receipt ofCRbyte.CR NUL(Carriage Return, Null): An interpretation of `rfc-854`_ may be thatCR NULshould be sent when only a singleCRis intended on a client and server host capable of distinguishing betweenCRandCR LF(return key vs enter key). The default implementation stripsCL NUL, and firesline_receivedon receipt ofCRbyte.CR(Carriage Return):CRalone may be received, though a client is not RFC-complaint to do so. The default implementation stripsCR, and firesline_received.LF(Linefeed):LFalone may be received, though a client is not RFC-complaint to do so. The default implementation stripsLF, and firesline_received.IAC EOR(Is-A-Command,End-Of-Record): In addition to line-oriented or character-oriented terminals,IAC EORis used to delimit logical records (e.g., "screens") on Data Entry Terminals (DETs), or end of multi-line input on vendor-implemented and some MUD clients, or, together with BINARY, a mechanism to signal vendor-implemented newline outside ofCR LFduring file transfers. MUD clients may readIAC EORas meaning 'Go Ahead', marking the current line to be displayed as a "prompt", optionally not included in the client "history buffer". To register receipt ofIAC EOR, a client must callset_iac_callback(telopt.EOR, func).
It should be said as historical source code, BSD 2.11's telnet source of UCLA and NCSA Telnet client of Univ. of IL for MacOS is most notable. There are also a few modern Telnet servers. Some modern Telnet clients support only kludge mode, with the exception of MUD clients, which are often Linemode only. TinTin++ is the only known client to support both modes.
`rfc-495`_, NIC #15371 "TELNET Protocol Specification." 1 May 1973, A. McKenzie, lists the following attached documents, which are not available:
[...] specifications for TELNET options which allow negotiation of:
o binary transmission
o echoing
o reconnection
o suppression of "Go Ahead"
o approximate message size
o use of a "timing mark"
o discussion of status
o extension of option code set
These specifications have been prepared by Dave Walden (BBN-NET) with
the help of Bernie Cosell, Ray Tomlinson (BBN-TENEX) and Bob Thomas;
by Jerry Burchfiel (BBN-TENEX); and by David Crocker (ULCA-NMC).
If anybody can locate these documents, please forward them along.
for communicating with any telnet server and the keyboard & screen. Most notably, it provides a
--cp437argument that allows connecting to telnet BBS systems from any posix shell, that otherwise would require a DOS Emulating program SyncTerm, mtelnet, netrunner. Instead, these systems may be used with a standard terminal emulator, such as xterm, rxvt, or iTerm2.Some telnet destinations:
- htc.zapto.org: Supports UTF8 or CP437 encoding (enthral).
- 1984.ws: Supports UTF8 or CP437 encoding (`x/84`_).
- nethack.alt.org: Supports latin1, CP437, or UTF8 encoding (dgamelaunch).
- blackflag.acid.org: CP437 encoding only, requires 80x24 window (mystic).
- bbs.pharcyde.org: CP437 encoding only, requires 80x24 window (synchronet).
It is hosted on github. Currently in development stage, feedback is encouraged. Feel free to make use of fork, pull and Issues services to report any bugs, grievances, or enhancements.
xon/xoff is unimplemented, see telnetlib3.stream_writer.TelnetWriter.handle_xon and handle_xoff.
After long-running (~2mo) job of telnetlib3 server on public IP, we ran out of memory ! write test verifying garbage collects!
TelnetReader has no need for declaring server/client=True, it behaves the same either way.
readline(), wow, what a bear of the RFC to provide either CR LF, CR NUL, that LF can happen any time in stream (LF CR is possible/equal), and that CR should never appear alone. What a rule for a bytestream, we wish not to have any stream lookahead beyond the first CR/LF, as this is the end line marker, we would be amiss to do any blocking for subsequent bytes, we most definitely may not receive any. Our implementation so far simply returns up to any first CR/LF discovered, and, if the next call to readline would return a line BEGINNING with either LF or NUL when the previous line ended with CR, we simply discard that byte.
base_client.py and base_server.py actually share the same ABC base_protocol.py, they are almost mirror images of one another, which is pretty great, actually. just reduce.
ValueError is used for many places where, the error is indicating that a negotiation state that was attempted by the remote end is invalid, for example: "received IAC SB LFLOW without first receiving IAC DO LFLOW."
SLC flushin/flushout attributes are not honored. Not entirely sure how to handle these two values with asyncio yet.
LINEMODE compliance needs a lot of work. - possibly, we remove LINEMODE support entirely. I only know of one client,
BSD telnet, that is capable of negotiating -- this is the C code from which our implementation was derived!
- callbacks on TelnetServer needed for requesting/replying to mode settings
- the SLC abstractions and 'slc_simul' mode is difficult for the API.
- There are many edge cases of SLC negotiation outlined in the RFC, how comprehensive are our tests, and how well is our SLC working?
- IAC-SB-LINEMODE-DO-FORWARDMASK is unhandled, raises NotImplementedError
- _receive_status(self, buf) response to STATUS does not honor given state
values. only a non-compliant distant end would cause such a condition. so it is decided to leave it as "conflict report only, no action always"
outer telnetlib3-server and telnetlib3-client and examples should connect as exit(main(**parse_args(sys.argv))), the _transform_args() function is rather shoe-horned, main() should declare keywords
also allow --exec instead of --shell parameter, which uses a pty to allow piping say, /bin/bash to a telnet port.
resources? https://github.com/Stichting-MINIX-Research-Foundation/netbsd/blob/master/usr.bin/telnet
isatty test in client_shell.py is os.path.sameopenfile(0, 1), we usually prefer self._istty = sys.stdin.isatty() ?