Codebase list conserver / debian/8.1.14-1 PROTOCOL
debian/8.1.14-1

Tree @debian/8.1.14-1 (Download .tar.gz)

PROTOCOL @debian/8.1.14-1raw · history · blame

			   Conserver Protocol
			   ==================


What Is This?
-------------

The following is an attempt to describe the client/server protocol used
between the server (conserver) and the client (console).  This document
bases it's information on conserver version 8.1.4, as it's the release
currently available.  If there are changes to the client/server
protocol, the INSTALL file should reference them and, ideally, this
document will be updated.

The information is looked at from the point of the server, since it's
the server that controls all information and triggers actions on the
client (like a suspend).  The client's perspective should be obvious
from this information.


SSL
---

The client and server can negotiate an SSL connection.  As far as the
code is concerned, the SSL "layer" is transparent.  Data is sent and
received just as if it was unencrypted.  Therefore, aside bringing up
the SSL connection, the SSL bits are unimportant from a protocol
standpoint.  The client and server still send and receive the same
information - it just happens to be encrypted to everyone else.


"On-The-Wire" Data
------------------

The low-level, "on-the-wire" data is encapsulated similar to the telnet
protocol.  All data is sent "as-is" with the exception of 0xFF.  0xFF is
used as a "command character" and both the client and server expect to
see a predefined option after it.  The possible options are: 0xFF, 'E',
'G', 'Z', and '.'.

The 0xFF option says to use the literal character 0xFF.  So, if there is
a 0xFF character in the data stream to be sent, the code will send two
0xFF characters (it's similar to using '\\' in C strings to embed a
'\').

The other options are used in various contexts, which will be described
in detail below.


Life As A Server
----------------

There are three different interfaces presented to clients by the server.
I'm going to name the three modes "master", "group", and "console".  The
first two are line-based, and the third is character-based.

To understand the differences, I must outline how conserver manages
consoles.  When conserver starts, it reads the configuration file,
listens on the master socket, and, for each group of consoles it must
manage (where the group size is set by -m), it forks off a copy of
itself.  Those child processes are what actually connect to the consoles
and they each listen on a new socket for client connections.  So, you
end up with a parent process (that knows about all consoles) that
manages the child processes (that know only about consoles it manages),
and everyone is listening on an individual socket for connections from
clients.

The parent process interacts with clients in "master" mode.  That mode
expects line-based commands and responds similarly.  Because it's the
master, it understands a certain set of commands that are different than
in "group" mode.

The child processes interact with clients in "group" mode first, and
negotiate a change to "console" mode when a client requests a connection
to a specific console.


"master" Mode
-------------

When parent process gets a connection from a client, it either sends an
"ok" string to signal it's ready or an error message (like "access from
your host is refused") and the connection is dropped.  At this point,
there are a small number of commands recognized by the server, since
most are restricted to "logged in" clients.  Here's the list of
available commands:

    exit   disconnect
    help   this help message
    login  log in
    ssl    start ssl session

An "exit" is sent a "goodbye" response and the connection is dropped.  A
"help" is sent the list above.  A "ssl" is sent an "ok" response and
then the server expects the client to negotiate an ssl connection.  A
"login" requires one argument (the username) and is either sent an "ok",
meaning the client is logged in, or a "passwd?" followed by the local
hostname, asking for the user's password, which it expects next.  If the
client sends a valid password, an "ok" is sent, otherwise an error
message and the connection is dropped.

Upon successful login, the commands available are:

    call      provide port for given console
    exit      disconnect
    groups    provide ports for group leaders
    help      this help message
    master    provide a list of master servers
    newlogs*  close and open all logfiles (SIGUSR2)
    pid       provide pid of master process
    quit*     terminate conserver (SIGTERM)
    restart*  restart conserver (SIGHUP) - deprecated
    reconfig* reread config file (SIGHUP)
    version   provide version info for server
    up*       bring up all downed consoles (SIGUSR1)
    * = requires admin privileges

"exit" and "help" are the same as before the client logged login.

The "call" command expects one argument, the console name to connect to.
The server will respond with either a port number (if it's a locally
managed console), an "@hostname" where hostname is the name of the
remote conserver host managing the console (if it's a remotely managed
console), or an error message (possibly multi-line).  The client is not
disconnected, whatever the response.

The "groups" command responds with a colon-separated list of port
numbers, which correspond to each of the child processes running on the
local host.  The client is not disconnected.

The "master" command responds with a colon-separated list of "@hostname"
names.  The list includes any hosts (including the possibility of the
local host) which have locally managed consoles.  The client is not
disconnected.

The "newlogs" command reopens all logfiles used by conserver, assuming
the user has administrative access.  It responds with a message starting
with "ok" if successful and an error message otherwise (like
"unauthorized command").  The client is disconnected if it's successful.

The "pid" command responds with the pid of the master process (in this
case, the one the client is talking to).  The client is not
disconnected.

The "quit" command will shut down conserver, assuming the user has
administrative access.  It responds with a message starting with "ok" if
successful and an error message otherwise (like "unauthorized command").
The client is disconnected if it's successful.

The "restart" command has been deprecated.  You should use "reconfig".

The "reconfig" command will cause conserver to reread the configuration
file and apply any changes, assuming the user has administrative access.
It responds with a message starting with "ok" if successful and an error
message otherwise (like "unauthorized command").  The client is not
disconnected.

The "version" command responds with the version string.  The client is
not disconnected.

The "up" command tries to "bring up" all disconnected consoles, assuming
the user has administrative access.  It responds with a message starting
with "ok" if successful and an error message otherwise (like
"unauthorized command").  The client is disconnected if it's successful.


"group" Mode
------------

When a child process gets a connection from a client, it either sends an
"ok" string to signal it's ready or an error message (like "access from
your host is refused") and the connection is dropped.  At this point,
"group" mode acts just like "master" mode.  Once the client successfully
logs in, however, "group" mode has the recognizes the following
commands:

    broadcast    send broadcast message
    call         connect to given console
    disconnect*  disconnect the given user(s)
    examine      examine port and baud rates
    exit         disconnect
    group        show users in this group
    help         this help message
    hosts        show host status and user
    info         show console information
    textmsg      send a text message
    * = requires admin privileges

The "exit" and "help" commands are like the others documented above.

The "broadcast" command expects a text string of the message to be sent
to all users connected to this process.  An "ok" is sent as a response.

The "call" command expects one argument, the console name to connect to,
just like in "master" mode.  The difference here is that this requests
the server to attach the client to the console and go into "console"
mode.  If the attachment is successful, the response will begin with a
'[' character.  If not, an error message is returned.  The success
responses are:

    [console is read-only]	 - console is read only
    [read-only -- initializing]	 - console is initializing, and
				   read-only for the time being
    [line to console is down]	 - console is down
    [attached]			 - attached read-write
    [spy]			 - attached read-only


The "disconnect" command expects an argument of the form "user@console"
where either the "user" or "@console" part may be omitted.  Upon
success, a response of the form "ok -- disconnected X users" is sent,
where X is the number of users disconnected.  If a user is unauthorized
or some other problem occurs, an error message (like "unauthorized
command") is sent.

The "examine" command returns a list of console information of the form
that 'console -x' shows.

The "group" command returns a list of console information of the form
that 'console -w' shows.

The "hosts" command returns a list of console information of the form
that 'console -u' shows.

The "info" command returns a list of console information of the form
that 'console -i' shows.

The "textmsg" command expects two arguments, the first being the
recipient of the message in the form "user@console" (again, where the
"user" or "@console" portion may be omitted) and the second being the
string, like the "broadcast" command.  The server returns "ok".


"console" Mode
--------------

As mentioned above, "console" mode is obtained by using the "call"
command when connected to a child processes operating in "group" mode.

"console" mode should look very familiar to a user of conserver, as it's
what the user interacts with when connected to a console.  There's
really nothings special here.  Each character received from the client
is compared to the escape sequence, and if it matches, an action occurs
on the server side.  If it doesn't match the escape sequence, the data
is sent on to the console.  All data received from the console is sent
to the client(s).  Of course, there are certain exceptions to these
rules, based on the state of the console and the state of the client.
And, certain escape sequences cause special behaviors to occur.

Most escape sequences cause the server to send information back to the
user.  Stuff like "^Ecw", "^Eci", and "^Ecu" are examples.  The escape
sequence is absorbed by the server, the server sends the client a
variety of information, and things continue as before.

The more "interesting" escape sequences are the following.

"^Ec;"  The server sends a 0xFF,'G' command sequence to the client, to
	signal a wish to move to a new console.  The client then gets
	put into the same state as the "^Ecz" sequence (paused), which
	gives the client a chance to either resume the connection or
	disconnect.

"^Ec|"  The server sends a 0xFF,'E' command sequence to the client, to
	signal a wish to have the client program interact with a
	program, as opposed to the user.  The server discards all data
	until it receives one of the following command sequences from
	the client:

	    0xFF,'E'	Signals successful redirection of interaction to
			a program.  The server then responds with "[rw]"
			or "[ro]" to tell the client whether or not they
			have read-write access.  If not, the client
			should abort the program and send the abort
			command sequence below, as other data received by
			the server will just get dropped.

	    0xFF,'.'	Abort the operation.  The server assumes the
			redirection didn't happen and returns the client
			to it's normal mode.

	The server keeps the client in the "redirected" state until it
	receives a 0xFF,'.' command sequence from the client (which
	usually occurs when the client command terminates).

	If the client is "bumped" from read-write to read-only by
	another user, the server will send the client a 0xFF,'.' command
	sequence to tell it to abort the redirection and return control
	back to the user.

"^Ecz"  The server sends a 0xFF,'Z' command sequence to the client, to
	signal a wish to suspend to client process.  The client is then
	put into a "paused" state where it receives no more data from
	the server.  When the client is ready to resume receiving data,
	it sends a character of data to the server, at which point the
	server discards the character and sends back a status message of
	the form " -- MSG]".  The current set of possible messages are:

	    " -- line down]"
	    " -- read-only]"
	    " -- attached (nologging)]"
	    " -- attached]"
	    " -- spy mode]"

#
#  $Id: PROTOCOL,v 1.2 2005/09/05 21:54:03 bryan Exp $
#