/************************************************************************
* 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 3, 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)
