ircu2.10.12 pk910 fork

[ircu2.10.12-pk.git] / doc / history / 2.4.notes

/************************************************************************
 *   IRC - Internet Relay Chat, 2.4.notes
 *   Copyright (C) 1990   Markku Savela
 *
 *   This program is free software; you can redistribute it and/or modify
 *   it under the terms of the GNU General Public License as published by
 *   the Free Software Foundation; either version 1, or (at your option)
 *   any later version.
 *
 *   This program is distributed in the hope that it will be useful,
 *   but WITHOUT ANY WARRANTY; without even the implied warranty of
 *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *   GNU General Public License for more details.
 *
 *   You should have received a copy of the GNU General Public License
 *   along with this program; if not, write to the Free Software
 *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */

IRC 2.4 release notes  6 May 1990/msa (Markku.Savela@vtt.fi)
============================================================

   This document explains the changes I have done up to this
point. Some additional changes and packaging has been made by
Chelsea (chelsea@earth.cchem.berkeley.edu). This is personal
view of the changes.

CHANGES TO LAST THE OFFICIAL RELEASE (2.2PL1)

   This release of irc2.4 is based to 2.2PL1 release (see the
HISTORY chapter later in this document). Aside from fixing the
bugs, this version is in many ways different from the 2.2PL1.
The purpose of the most changes is to make it easier to run an
IRC server. Normal users benefit from these changes indirectly
by getting a better maintained server.

1. Changes visible to normal users

   Even while mainly fixing bugs, some user visible changes have
crept in too.

1.1 General note on wildcards

   Many commands accept now wildcard matching where applicable. All
compares are case insensitive (e.g. "a" == "A"). The wild cards are

        ?       matches any single character

        *       matches any number of characters, also empty
                string. [PL1 had a bug, which caused "*du*"
                not match "....edu"].

1.2 Server supported wildcards for "/who mask" command.

   Protocol message is "WHO mask", where mask can be

        empty
        0       List all users [No change from PL1]

        *       List all users on the same channel where the user is
                (or all, if user is on 0) [No change from PL1].

        number  List all users on the specified channel [No change
                from PL1]. Note, if the "mask" begings with a digit,
                this form is assumed, and the remainder of mask is
                ignored, e.g. "/who 12*.fi" gives all people from
                channel 12 and ignores the "*.fi" part.

        mask    If the mask is any string, it will be compared
                *separately* to each information field of the user
                and if a match is found in any field, that user
                is included into the list. The fields searched
                are

                        nickname
                        loginname (account name)
                        real name (text shown in parenthesis)
                        hostname (users machine)
                        servername (server he/she is using)

                Note: servername is not usually shown on WHO output,
                but is included in anyway. Example: finding all users
                somehow connected with Finnish sites, can be achieved
                with mask "*.fi".

1.3 Changes to /whois command

   As WHO, also /whois accepts wild cards as a parameters. WHOIS
returns information for all users whose nickname matches the specified
mask.

   WHOIS automaticly calls WHOWAS [see below], if the attempted nickname
is not found.

1.4 Short term "WHOWAS" history

   The server has a short in memory cache of the recent nickname changes
(the current default is set to 200 last changes). The design goal of
this is that it remembers changes in last few minutes, there is no
intention of this to be a long term history. That must be a separate
project, although it could use the hooks provided by this service.

   "WHOWAS nickname" queries this cache and returns about the same
information that WHOIS would do, if the nickname is found. Wildcards
are not accepted here, this is a specifically designed feature. If
the name is not found, WHOWAS doesn't reply anything. This is because
the most useful use of WHOWAS is implicitly through "WHOIS".

   This history is also implicitly utilized by KILL command.

1.5 New SERVER-SERVER/SERVER-CLIENT protocol message WALLOPS

   The message ":source WALLOPS :Message" sends the message text
to all operators that are currently online. Any user can use this
command, it's not restricted. How this function is activated, depends
on the client, but if nothing else works, "/quote wallops text" should.

   NOTE:This function will not be fully operational until *ALL*
        servers have upgraded to version 2.4. Also, operators
        must be using a client that recognizes this command.

   This is really a hasty addition. But, done this way it follows
the general IRC message philosophy, where messages are sent only
to links where they are needed (e.g. WALLOPS goes only to servers
that have opers online--it's not broadcast to every server).

1.6 General use of wildcarding in server queries

   All commands that previously took a servername as a parameter,
now accept also a wildcarded mask. The mask is replaced with
the first matching servername. The following user level commands
are affected

        /admin server   -- administrative info
        /time server    -- local time
        /version server -- the server version
        /motd server    -- "the message of the day"
        /info server    -- info (usually same on same server version)
        /stat f server  -- statistics information
        /users server   -- users logged on server machine

   Note: Remote capability is a new feature for "info" and "stat"
commands. Until all servers have upgraded, these commands may not
reach the intended target and may return the information from some
intermediate server.

1.7 Marking user AWAY

   v2.2PL1 version and earlier showed the AWAY-state (G) only for
the local users of the same server. AWAY status could be queried
only by sending a message to a user. This release (or since msa.4)
broadcasts the away status to every server and the commands /WHO and
/WHOIS give this information reliably.

   A side effect of this change is: when a user marks himself/herself
as AWAY, all pre-msa.4 servers that are reached will send back an
acknowlegde message. Until all servers are upgraged, use of AWAY
is somewhat inconvenient. If you get extra messages from AWAY,
they also contain the server information. Use /admin command and
send a *friendly* request for the admin to upgrage his/her server
to a working version, namely 2.4 :)

1.8 Servers don't restrict characters within messages

   The parameter fields of the messages can now contain any characters
in range 1-255, except '\r', '\n' and '\0'. The client programs should
by default filter away the "dangerous" control characters, but intelligent
clients can utilize this change and allow exchanges with foreign
8-bit (or wider) charactersets. (The actual command parts must still be
represented with the ordinary 7-bit characters.)

2. Changes visible to the server administrator

2.1 Identifying servers

   Servers/clients have now always two names (it was this way in
PL1, but I think this version makes the idea more clear):

        Announced Name:

        The official name of the server (the name you use in
        /time, /quote connect, etc) or users nickname. Servers
        name is usually the hostname, but can actually be almost
        any string of characters resembling hostname. This one
        is given in M-line of ircd.conf.

        Socket hostname:

        Socket hostname of the server or client. This is the hostname
        of the connecting server/client and this is resolved from the
        connection. If resolve cannot be done, ircd defaults to using
        numeric IP-address. *ALL* access checks are based on this
        name, especially noteworthy fact, if your resolver cannot find
        hostnames by IP-address, you must allow the access by IP-numbers
        in your ircd.conf.

   In many places, where servers name is shown, actually both are
shown. The general format of the displayed name is

        AnnouncedName[SocketHostName]

When a connection is yet unkown, there is no AnnouncedName, and if the
AnnouncedName is the same as SocketHostName, the "[..]"-part is omitted.

2.2 Many notices to local operators

   If an oper is signed on the server, he/she will receive many
notices about exceptional conditions and servers actions. When
something goes wrong, it should be much easier to fix the problems.

   Few often occurring, inportant error messages are

   "Write error to SERVERNAME, closing link"

        write() to socket returned with an error. Server is
        closing the link. This means usually network problems
        which you can do nothing about.

   "Max buffering limit exceeded for SERVERNAME"

        This is the situation where old server would have been
        "frozen". The socket buffers in your OS have been filled and
        even servers own predefined internal buffering MAX for a link
        has been exceeded. Exceeding this limit most likely means
        that the link is really dead, so the server closes the link
        and scratches all queued output for it. If the limit is
        set high ( > 20000 bytes), you won't usually see this, but
        just "No responce from SERVERNAME, closing link" as the
        server does not reply to PING as it should.

   "Link SERVERNAME cancelled, server SERVERNAME already exits"

        Two different servers from your net fragment attempted
        to connect same other net fragment about the same time
        and this collision is detected at your server. IRC routing
        does not allow loops, the link causing the loop is closed.
        (Which of the two links gets closed is mostly determined
        by pure chance and timing--you may lose a better link this
        way. Collisions should be rare in normal operation, if
        the timers in "config.h" are not messed up too much...)

        Of course, you get this too, if you try to connect to a
        server that is already connected by some other route. In
        that case your attempted connection is just safely cancelled.

   The notices attempt to be self explaining.

2.3 Links statistics collecting

   IRCD now counts the bytes and messages transmitted to each open
link. This information can be output with a command "/stats l"
("/stats" or "/stat m" will give the old message count statistics).

   Sample output

Link           SendQ  SendM SendBytes  RcveM RcveBytes Open since
oddjob.uchicago    0    203      8067    772     34321 Sun May  6 02:15:45 1990
cs.hut.fi[sauna    0   1916     79798     94      3082 Sun May  6 01:51:25 1990
otax.tky.hut.fi    0   3722    151511    426     22690 Sun May  6 00:25:54 1990
nada.kth.se        0   8775    355811   5333    223853 Sat May  5 14:11:49 1990
vehka.cs.uta.fi    0  23816    882000    901     41156 Fri May  4 22:50:23 1990
lut.fi             0  25145    943765   1068     35020 Fri May  4 22:34:16 1990
kreeta.helsinki    0  24286    899191    957     47085 Fri May  4 22:33:28 1990
naakka.tut.fi      0  27754   1067302   8288    362960 Fri May  4 22:33:14 1990
joyx.joensuu.fi    0  30003   1172949   2300     80053 Fri May  4 22:33:05 1990
tel4.tel.vtt.fi    04083771 167473890 863475  35022755 Mon Apr 23 00:15:17 1990
                   |    |       |       |       |      |
                   |    |       |       |       |      Link established
                   |    |       |       |       The number of bytes received
                   |    |       |       The number protocol messages received
                   |    |       The number of bytes transmitted
                   |    The number of protocol messages transmitted
                   The amount of queued data in bytes (if socket is hung)

   The last row (with the local servername) contains the total
cumulative counts for all connections since the server was started.

   One can query the statistics of a remote server by adding the servers
name to the command "/stat l servername". Of course, this only works,
if all intermediate servers have upgraged. The first "old" server
will stop the propagation and return the message counts by default.

2.4 Connecting servers

   An oper can manually activate a connection phase to any server
defined in ircd.conf C-lines (to successfully complete the connection,
the N-line must be present too). The message achieving this is

        CONNECT servername portnumber

   where servername may be a mask string containing wildcards. This
name is matched against entries in ircd.conf (notice: the testing
is made in reverse order, e.g. the last C-line in ircd.conf is tested
first).  If portnumber is omitted, the ircd uses the one given in the
found C-line. If the C-line does not have the portnumber, the compiled
default will be used (PORTNUM from config.h).

   This release allows also for remote connecting. An oper can send
a connect request to remote server with

        CONNECT servername portnumber remoteserver

This command is passed to the 'remoteserver' and it then tries to
execute it like it was given locally. (If there are opers online on
that server, they will get a notice about this happening.) Note, that
one can remotely connect only what is defined in ircd.conf. Usually
one needs and should use this only for immediate your neighbours. Nobody
should randomly go and give connect requests to distant servers, unless
one knows it's absolutely necessary and is very familiar about the
linking setup there.

2.4 Terminating connections

   The SQUIT command in PL1 was not intended to be used manually and
was very dangerous to use (it also created so called "ghost servers").
Since msa.4, the SQUIT has been safer to use manually.


   "SQUIT z" s                          a
              \                        /
               \                      /
         ------- x ------- y --| |-- z ------- b
                /               ^     \
              /                 |       \
            p                            c

        "SQUIT z" will break the link between "y" and "z" if injected
        into system from "s". After that the net will be in two fragmets,
        broken between "y" and "z". Server "z" never sees the actual
        SQUIT, all it observers is that the link to "y" suddenly closes
        (opers on z would see it as "Server y closed the connection"
        notice. Opers on y would see it as "Received remote SQUIT from
        x", note that the actual source "s" is not identified in the
        current version--for reasons too complicated to be explained
        here).

        *WARNING* *WARNING* If the server "y" is still running pre-msa.4
        (like PL1), don't *EVER* issue a SQUIT for its links (unless the
        link is to a leaf node or verifiably a "ghost server").

        Note, that when the link between "y" and "z" breaks, y will spit
        out SQUIT's for "z", "a", "b" and "c" to "x". At same time "z"
        is sending SQUIT's for "x", "s", "p", etc to "a", "b" and "c".
        SQUIT is normally generated by servers automaticly, it's just
        a later modification (msa.4) that allows an OPER to use this
        same message to "simulate" a link break at certain point.

        *IMPORTANT* If server "z" has configuration "C:y::y:6667", it
        automaticly attempts to reconnect after a short delay (currently
        10 seconds), but only *if* the connection has been up long enough
        reliably (currently set to 10 minutes). If the thus formed link is
        squit another time, it will not attempt to come back immeatedly.
        This gives an oper time to reconfigure the links if that first
        short delay is not enough.
        
    As in all commands, also SQUIT accepts wildcards, but be careful to
give sufficient identification. SQUIT of wrong server is not nice...

2.5 KILL message

    KILL will implicitly use the history database. If a KILL is
issued for a nick that has been changed to another, the server
will automaticly re-issue the kill with the new nickname, if
the change has happened recently (current value should be 90
seconds). If a "terrorist" is clearly distrupting channel by
bombarding it with garbage from negative channels and changing
nick all time, there is no need to consult the "WHOWAS" data
base, just use the nickname that was used to send the garbage
and ircd hunts the culprit down. When this change of target
happens, the oper issuing the kill is notified.

NOTE:   With automatic, kill-proof-reconnecting clients, the
        value of KILL is becoming insignificant...

2.6 Changing the server defaults from the command line

    The servers activation command is now

ircd[ -f configfile][ -h servername][ -p portnumber][ -x debuglevel]

where parameters can be given in any order. If the "configfile"
is defined, it will override the default specified in the file
"config.h". If "servername" is defined, it will override the
one defined in the M-line on the configuration line. "portnumber"
will override the compiled default (from "config.h") or the
one from the M-line of the configuration file. The "debuglevel"
will determine the amout of logging the server does into a
log file that has been define in "config.h". The "debuglevel"
should never be defined for a server running normally, it can
quickly generate megabytes of trace. Usually needed only when
the server is incapable of starting properly at all, then one
run with "-x9" usually is enough to reveal the problem.

3. General cleaning up and commenting the code

   This issue is controversial. My way of fixing bugs is not just
fix them, I also want to program defensively, make it difficult to
make new errors. Thus I have heavily reformatted and reorganized
those files that I have had to touch. Some functions have been
renamed intentionally to catch all uses of those functions [because
the functions semantics or calling sequences have been changed].

   This release (2.4) will be the last IRC version I'm contributing
to. If you have any wishes or complains about the code or functioining
of IRC, use the source or ask whomever it happens to be the current
developer.

HISTORY

   There have been many different versions of IRC and many of those
versions are still in use. The following attempt to bring some
clarification to the versions. This starts from 2.01.6, hopefully
no servers are running older versions...

        ...
        ...
        2.01.6  A version from WiZ in summer 1989
        ...
                2.01t6  A series of releases, which contained minor
                2.01T6  adjustements and bug fixes to the base version.
                2.01u6  Some of those fixes caused extra errors, of
                2.01U6  this series versions 2.01U6 and 2.01v6 are at
                2.01v6  least known to be rather stable.

        2.1.0   Mike Bolotski created these versions from the sources
        2.1.1   of 2.01U6, but unfortunale some devious bug crept in
                and caused a lots of linking problems (the nasty "ghost
                server problem" splintered the net constantly). These
                versions must be deleted on sight :) [Autumn 1989]

        2.2     This version is the 2.01v6 sources repackaged into
                multiple directories by Mike again. Probably nobody
                is running this base version, because is was promptly
                followed by two patch releases [Autumn 1989]

                2.2PL0  These two are the last major "official" releases
                2.2PL1  and most of the servers upgraged to either of
                        these.

        2.2msa  Unfortunately 2.2PL1 version had a tendency to die
                mysteriously very often. So, I started to look into the
                code from March 1990 and that resulted a series of
                patches to the 2.2PL1 server code, but finally
                decided to release full server code releases of which
                few have got wider distribution

                2.2msa.4
                        Has most of the known PL1 bugs fixed and seems
                        to be very reliable. But once servers started
                        staying up, a new problem appeared: socket
                        buffers started getting full and servers tended
                        to freeze very often for long intervals.

        2.3alpha
        2.3     Is an attempt to make an official release from 2.2msa.4
                code, but hassles with changed copyrights make this
                version unacceptable. Besides, 2.3alpha or 2.2msa.4 are
                now obsolete, old versions :)

        2.2msa.x
                To solve the freezing problems, the server code is changed
                to use non-blocking sockets.

                2.2msa.7
                2.2msa.9
                        Are intermediate test versions, of which .9 seems
                        to have most of the problems solved.

                2.2msa.10
                        Never released. This is slightly improved version
                        of msa.9, some new features.

        2.4     Is a release which combines 2.2msa.10 and Chelsea's
                modifications to the server. Also, this release has
                once again reorganized the directories and makefiles.


-- msa (Markku.Savela@vtt.fi)