By Alexander Barton (2) and DNS777 (1)
* cmode-M:
Add new channel mode "M" to doc/Modes.txt
Remove Can_Send_To_Channel_Identified()
Implement channel mode "M"
Two fixes here: IRC_Send_NAMES was checking the capability of the
wrong client when responding, and it didn't return any prefix for
clients that had either +v or +o but not both.
By Alexander Barton (1) and DNS777 (1)
* better-chan-errors:
Remove unused ERR_CANNOTSENDTOCHAN2_MSG message
Add some more information to channel error numerics
This patch series converts the statically allocated password buffer in the
CLIENT structure into a dynamically (and only when needed) allocated buffer
which is referenced by the CONNECTION structure.
This a) saves memory for clients not using passwords at all and b) allows
for "arbitrarily" long passwords.
By Brett Smith (5) and Alexander Barton (2).
* 'move-connection-password' of git://arthur.barton.de/ngircd-alex:
Login_User(): use "conn" insted of calling Client_Conn(Client)
Free already saved password when storing a new one
Indentation and style fixes.
Connection password is not constant.
Implementation clean-ups.
Dynamically allocate memory for connection password.
Move client password from the Client to the Connection struct.
The "ServiceMask" variable in "Server" blocks now can handle more than
one mask using the new MatchCaseInsensitiveList() function.
This makes marking "service clients" much more specific, which is a
good thing per se, but which is the prerequisite for reasonably
blocking these nick names, too (see commit a6dd2e3 for details).
This patch introduces the new function Conf_NickIsBlocked() which checks
if a given nick name matches with the "service mask" of a configured server.
And Client_CheckNick() uses this information to deny such names for regular
IRC users.
So nick names intended for IRC services are more protected and can't be used
by regular users even when the "services pseudo-server" isn't connected to
the network.
But please note:
Up to now, there can be only one "ServiceMask" pattern per server, which
most probably blocks much more nick names than really required ...
So "ServiceMask" should allow more than one pattern which can be more
specific, and most probably it should be possible to block nick names in
the global server configuration as well.
Nick names introduced by other servers/services are never restricted.
This is a relatively naive implementation, basically doing the bare minimum
necessary to make the switchover go. Subsequent commits can focus on
improving the implementation.
3 new channel user modes have been added.
Half Op: +h(Prefix: %) can set the channel modes +imntvIbek
and kick all +v and normal users.
Admin: +a(Prefix: &) can set channel modes +imntvIbekoRsz and kick all
+o, +h, +v and normal users.
Owner: +q(Prefix: ~) can set channel modes +imntvIbekoRsz and kick all
+a, +o, +h, +v and normal users
Implemented support for hashed hostnames for CloakHost. The admin can
use '%x' in both the CloakHost and CloakHostModeX setting. The config
option CloakHostModeX was renamed to CloakHostSalt. This salt is used
for both cloaking options.
CloakHostModeX can now contain '%x'. It will be replace by the hash of
the original client hostname. The new config option CloakHostModeXSalt
defines the salt for the hash function. When CloakHostModeXSalt is not
set a random salt will be generated after each server restart.
Spelling fix in defines.h
Don't try to establish an outgoing server link after DNS lookup when this
server re-connected on its own in the meantime.
In addition, log a warning message if we try to update the connection
index of an already connected server structure -- and ignore it.
Up to now, both behaviour could lead to a race when the remote server
connects to this daemon while it still prepares the outgoing connection:
- The local server prepares the new outgoing connection ...
- in the meantime the remote server becomes connected and registered.
- Now the new outgoing connection overwrites the (correct) socket handle,
- then the 2nd connection becomes disconnected: "already registered",
- and the 1st connection becomes unhandled ("gets lost") because the
configuration structure is reset because of the wrong socket handle.
This patch hopefully fixes all these problems.
Only alphanumeric characters are allowed in the user name, so ignore
all IDENT replies that would violate this rule and use the one supplied
by the USER command.
Only alphanumeric characters are allowed in the user name, so terminate
the connection if any "strage" characters have been supplied by the user.
This is how other IRC daemons (like ircd2.11 and ircd-seven) behave ...
This fixes the following warnings with GCC 4.4.5 on Linux:
irc-login.c: In function ‘IRC_PASS’:
irc-login.c:92: warning: implicit declaration of function ‘strlen’
irc-login.c:92: warning: incompatible implicit declaration of built-in function ‘strlen’
irc-login.c:113: warning: incompatible implicit declaration of built-in function ‘strlen’
irc-login.c:129: warning: implicit declaration of function ‘strchr’
irc-login.c:129: warning: incompatible implicit declaration of built-in function ‘strchr’
irc-login.c:133: warning: implicit declaration of function ‘strcmp’
irc-login.c: In function ‘IRC_SERVICE’:
irc-login.c:556: warning: incompatible implicit declaration of built-in function ‘strchr’
login.c: In function ‘Login_User’:
login.c:131: warning: implicit declaration of function ‘strcmp’
* capabilities:
"multi-prefix" capability 2/2: adjust NAME and WHO handlers
"multi-prefix" capability 1/2: implement complete CAP infrastructure
IRC_Send_NAMES(): Code cleanup
New function Client_CapSet() in addition to Client_Cap{Add|Del}
"CAP REQ" starts capability negotiation and delays user registration
Correctly handle "CAP END", new client type CLIENT_WAITCAPEND
Implement core IRC capability handling and "CAP" command
New "login" source file
Introduce_Client() => Client_Introduce(), and move it to client.c
Now ngIRCd is able to handle "CAP LS", "CAP REQ", "CAP LIST", and
"CAP CLEAR" commands.
"multi-prefix" can be set/unset, but has no functionality - yet!
This fixes the following warning of clang:
/src/ngircd/lists.c:152:44:
warning: size argument in 'strlcpy' call appears to be size of the
source; expected the size of the destination [-Wstrlcpy-strlcat-size]
But it isn't a real problem, because the size of the source always is the
same than the size of the destination ...
This patch implements the core functions to support "IRC Capabilities"
and the IRC "CAP" command as used by other servers and specified here:
<http://www.leeh.co.uk/draft-mitchell-irc-capabilities-02.html>.
It enables ngIRCd to support the defined handshake, but it doesn't
implement any capabilities, so "CAP LS" and "CAP LIST" always return
the empty set and "CAP REQ ..." always fails with "CAP NAK".
Rename Hello_User[_PostAuth] to Login_User[_PostAuth] and move it to the
new login.c; and move cb_Read_Auth_Result(), too. This will enable further
code to easily call Login_User() when required.
Up to now, ngIRCd silently ignored permission denied errors when trying
to enable a chroot setup: only the "not running chrooted" message became
logged later on.
This patch lets ngIRCd exit with a fatal error when the chroot can't
be enabled on startup -- this is the much safer bevahiour!
This fixes the following warning with gcc 4.6.3.:
irc-mode.c: In function "Channel_Mode":
irc-mode.c:947:26: error: "list" may be used uninitialized
in this function
irc-mode.c:884:25: error: "list" may be used uninitialized
in this function
(The variable has never been used uninitialized, so don't worry)
When
* building the ngircd Debian package (on Linux at least) and
* using the sbuild build system,
the command "ps -af" does not include the commands running inside the
sbuild system. Therefore, start-server.sh will report a fail as getpid.sh
cannot not find the ./T-ngircd1 just started although it's actually
running. This results in a funny build log ...
starting server 1 ... failure!
FAIL: start-server1
running connect-test ... ok.
PASS: connect-test
The self-test of getpid.sh however will likely succeed as it's happy if
it sees any process with "sh" somewhere in the name. Things go downhill
from there.
The confusing things are:
* The alternative cowbuilder/pbuilder does not have this problem.
* The alternative usage "ps ax" does fine.
So, as a quick hack, the patch attached adds another switch to getpid.sh.
Start "regular" logging not until the configuration file has been read in
and "SyslolgFacility" is set, and log all configuration errors using the
generic "daemon" facility.
So if there are no configuration errors, logging starts right after parsing
the configuration and we log the configuration file used _after_ reading it.
But this is no problem because every configuration error message includes
the configuration file name as well.
(The "double hello" has been introduced by commit 3641e51109)
Syslog logging has been initialized before reading the configuraton
file, so ngIRCd always used the default facility and ignored the
"SyslogFacility" configuration option.
Thanks to Patrik Schindler for reporting this issue!
This patch updates the limits for handling commands from a remote server:
- "<user count> / 5 + <min>" using "<min>=10" during normal operation,
- the above count multiplied with 5 while servers are syncing.
The intention is to a) make the limit dependent of the number of users
in the network (the more users, the more commands required to sync) and
b) to significantly rise this limit while servers are joining the network
to make the login and synchronization faster.
At the end of sending all "state" to the remote server, a PING command
is sent to request a PONG reply. Until then, no "regual" PING was sent,
so Conn_LastPing(<connection>) is null and now becomes non-null in the
PONG command handler.
So the servers are still synchronizing when Conn_LastPing(<connection>)
is 0, which could easily be tested.
The numeric RPL_WHOISHOST_MSG(378) returns the DNS hostname (if
available) and the IP address of a client in the WHOIS reply.
Only the user itself and local IRC operators get this numeric.
This allows to use "*!<user>@<host>" or "*!*@<host>" masks to reject
clients even before receiving PASS, NICK and USER commands and before
forking authentication child processes which reduces resource usage.
This allows a channel operator to define exception masks that allow users
to join the channel even when a "ban" would match and prevent them from
joining: the exception list (e) overrides the ban list (b).
If the target user of a PRIVMSG or NOTICE command has the user mode 'C'
set, it is required that both sender and receiver are on the same channel.
This prevents private flooding by completely unknown clients.
- Check correct list for duplicates when adding items.
- Don't generate any messages when adding duplicates or removing
non-existing items (this is how ircd-seven and ircu behave).
- Code cleanup: Add_Ban_Invite(), Del_Ban_Invite().
Commit 565523cb allowed processing of further channel names given to the
JOIN command when a single name was invalid.
After this patch, the JOIN command handler continues to process channel
name lists even after errors like "channel is full", "too many channels",
and the like and generates appropriate error messages for all the
channels given by the client.
Limit the MODE command to handle a maximum of MAX_CMODES_ARG (5) channel
modes that require an argument (+Ibkl) per call.
Please note: Further modes that require arguments are silently ignored
and end the handling of any further modes.
This is similar to the behavior of ircd2.11 (silently ignores but seems
to handle other modes) as well as ircd-seven (silently ignores but handles
some(!) other modes) ...
The assert(client != NULL) got triggered during our tests, so there is
an error path that resulted in the connection being still established
(sock >= 0) but the client structure already freed.
So Conn_Write() should handle it!
This reduces the possibility of flooding channels with commands like
"PRIVMSG/NOTICE #a,#n,#c,... :message" a little bit.
Problem noticed by Cahata -- thanks!
Until now, the penalty time has only been set when longer as the
already set one, so it didn't accumulate.
And add documentation for and clean up code in Conn_SetPenalty() and
Conn_ResetPenalty() functions.
This partly closes bug #118. ngIRCd still starts up even when
Server{UID|GID} is invalid: then the daemon falls back to "nobody"
when running with root(0) privileges (as before).
commit 15fec92ed7
(Update list item, if it already exists) can make ngircd
crash because 'Reason' can be NULL, as reported by
Cahata on the ngircd mailing list.
Doesn't affect any released ngircd versions.
Also, make sure that we do not pass NULL as arguments
to a '%s' printf-like function.
When JOIN is received with more than one channel name, don't stop
processing on the first error (e.g. bad name, wrong channel key, ...)
but report an error and continue with the other given channel names.
Reported by Cahata -- thanks!
When "PAMIsOptional" is set, clients not sending a password are still
allowed to connect: they won't become "identified" and keep the "~"
character prepended to their supplied user name.
This fixes two bugs:
- "WHO <nick>" returned nothing at all if the user was "+i"
(reported by Cahata, thanks).
- "WHO <nick|nickmask>" returned channel names instead of "*"
when the user was member of a (visible) channel.
Clean up code and add documentation as well.
Rename Channel_Count() to Channel_CountVisible() and only count channels
that are visible to the requesting client, so the existence of secret
channels is no longer revealed by using LUSERS.
Reported by Cahata -- thanks!
Unknown user and channel modes no longer stop the mode parser, but are
simply ignored. Therefore modes after the unknown one are now handled.
This is how ircd2.10/ircd2.11/ircd-seven behave, at least.
Reported by Cahata -- thanks!
This fixes:
irc-oper.c: In function ‘IRC_xLINE’:
irc-oper.c:429: warning: ‘class’ may be used uninitialized in this function
irc-oper.c:430: warning: ‘class_c’ may be used uninitialized in this function
If gai_strerror() isn't available, use a macro that simply returns
a static error message (regardless of the real error code).
For example, GNU libc 2.0.7 doesn't implement gai_strerror().
This fixes the problem that ngIRCd can't do any IDENT lookups because
of the socket has already been closed in the child process.
The bug has been introduced starting with ngIRCd 17 ... :-(
(commit ID 6ebb31ab35)
- User mode "R": indicates that the nick name of this user is "registered".
This mode isn't handled by ngIRCd itself, but must be set and unset by
IRC services like Anope.
- Channel mode "R": only registered users (having the user mode "R" set)
are allowed to join this channel.
DEBUG_BUFFER is off by default and therefore disables these messages:
- "Handle_Write() called for connection XX, YY bytes pending ..."
- "Connection XX: ZZ bytes left in read buffer."
* ServerMode:
Handle channel user modes 'a', 'h', and 'q' from remote servers
Handle unknown channel modes on server links
Handle unknown user modes on server links
IRC_MODE(), Client_Mode(): code cleanup [2/2]
Enlarge client user mode buffer, reduce client flags buffer
Infom clients when other servers change their user modes
IRC_MODE(), Client_Mode(): code cleanup [1/2]
These channel user modes aren't used for anything at the moment, but
ngIRCd knows that these three modes are "channel user modes" and not
"channel modes", that is that these modes take an "nick name" argument.
Like unknown user and channel modes, these modes are saved and forwarded,
but ignored otherwise.
We have to enlage our user mode buffer, so we can handle even unknown
user modes in the future; and reduce the client flags buffer, because
I can't imagine why we ever would need ~100 flags!?
Now we support up to 15 user modes (was: 8) and up to 15 flags (was: 99).
So in the end, we even save 99-15+8-15=77 bytes for each client structure!
Use ps(1) flag "-a" (as well as "-f"):
"Select all processes except both session leaders (see getsid(2)) and
processes not associated with a terminal."
Thanks to Götz Hoffart for reporting this problem!
On server-links, spoofed prefixes can happen because of the asynchronous
nature of the IRC protocol. So don't break server-links, only log a message
and ignore the command.
This fixes bug 113, see:
<https://arthur.barton.de/bugzilla/show_bug.cgi?id=113>
Citing an email from Florian to the ngIRCd mailing list:
"I wonder what the expected behaviour is when Conf_MorePrivacy is changed
from 'yes' to 'no' and the config is reloaded.
At the moment, WHOWAS will start giving out information on Users that
were connected during Conf_MorePrivacy=yes period. If this is not
wanted, Client_RegisterWhowas() should be changed to not store a record
when Conf_MorePrivacy is enabled."
And I think it is "not wanted" :-)
For outgoing connections, we use 2048 (DH_BITS) since commit 49b2d0e.
This patch enables ngIRCd to accept incoming connections from other servers
and clients that use at least 1024 bits (and no longer requires 2048 for
incoming connections, too).
Patch proposed by Florian Westphal.
this patch contains:
* Fix for Conf_CloakUserToNick to make it conceal user details
* Adds MorePrivacy-feature
MorePrivacy censors some user information from being reported by the
server. Signon time and idle time is censored. Part and quit messages
are made to look the same. WHOWAS requests are silently dropped. All
of this is useful if one wish to conceal users that access the ngircd
servers from TOR or I2P.
This patch makes it possible to scrub incomming CTCP commands from
other servers and clients alike. The ngircd oper can enable it from
the config file, by adding "ScrubCTCP = yes" under [OPTIONS]. It is
default off.
CTCP can be used to profile IRC users (get user clients name and
version, and also their IP addresses). This is not something we like
to happen when user pseudonymity/secrecy is important.
The server silently drops incomming CTCP requests from both other
servers and from users. The server that scrubs CTCP will not forward
the CTCP requests to other servers in the network either, which can
spell trouble if not every oper knows about the CTCP-scrubbing.
Scrubbing CTCP commands also means that it is not possible to send
files between users.
There is one exception to the CTCP scrubbing performed: ACTION ("/me
commands") requests are not scrubbed. ACTION is not dangerous to users
(unless they use OTR, which does not encrypt CTCP requests) and most
users would be confused if they were just dropped.
A CTCP request looks like this:
ctcp_char, COMMAND, arg0, arg1, arg2, .. argN, ctcp_char
ctcp_char is 0x01. (just like bold is 0x02 and color is 0x03.)
They are sent as part of a message and can be delivered to channels
and users alike.
The intention of this restructuring is to make the [Global] section much
cleaner, so that it only contains variables that most installations must
adjust to the local requirements.
All the optional variables are moved to [Limits], for configurable limits
and timers of ngIRCd, and [Options], for optional features.
The old variables in the [Global] section are deprecated now, but still
recognized.
The io_event_create error handling seems to miss a 'return'
statement.
Fix this by moving io_event_create() call around so we do not
need the Conn_Close/Init calls in the error case.
problem is that some clients refuse to connect to severs that only offer
1024. For interoperability it would be best to just use 4096, but that
takes minutes, even on current hardware.
parse.c:284: warning: suggest parentheses around operand of '!' or
change '&' to '&&' or '!' to '~'
The expression looks dubious, this should probably be
an if-not-set, then... test.
When the write buffer space grows too large, ngircd has to disconnect
the client to avoid wasting too much memory.
ngircd logs this with a scary 'write buffer overflow' message.
Change this to a more descriptive wording.
Not all servers (and services!) using the RFC1459 protocol style send
prefixes on all commands; so don't require them to do so.
This relaxes the requirements introduced by commit 15775e679.
We re-use the same helper function for both forward lookups
(when we want to connect to a peer server) and for validation of reverse
loopups (where we make a lookup on the hostname returned
by a reverse lookup on the IP address that connected).
Problem:
When ConnectIPv6=no, the forward lookup helper sets the adderss family
to AF_INET, and, if out client connected via ipv6, we fail to validate
the result.
Thus move the ConnectIPvX check out of the helper.
we do not need this for cryptographic purposes, but we can do better
than plain srandom(getpid()).
Also, keep in mind that rng state is inherited across fork(), so re-init
it in the child.
When enabled, this configuration option lets ngIRCd send a PING with an
numeric "token" to clients logging in; and it will not become registered
in the network until the client responds with the correct PONG.
This is used by QuakeNet for example (ircu/snircd), and looks like this:
NICK nick
:irc.example.net PING :1858979527
USER user . . :real name
PONG 1858979527
:irc.example.net 001 nick :Welcome to the Internet Relay Network ...
Make sure that all commands received from other servers do have
valid prefixes.
Only exceptions are PING and ERROR commands that can occure without
prefixes when generated by the remote peer itself.
* CloakUserHost:
Add a note not to use a percent sign ("%") in CloakHost variable
Rename ClientHost to CloakHost, and ClientUserNick to CloakUserToNick
Don't use "the.net" in sample-ngircd.conf, use "example.net"
ngircd.conf.5: document "ClientHost" and "ClientUserNick"
Move "ClientHost" and "ClientUserNick" to end of [Global] section
ClientUserNick setting
ClientHost setting
* QuitOnHTTP:
Only "handle" HTTP commands on unregistered connections
Don't use IRC_QUIT_HTTP() if STRICT_RFC is #define'd
IRC_QUIT_HTTP(): enhance error message
Move IRC_QUIT_HTTP() below IRC_QUIT()
quit on HTTP commands: GET & POST
* bug72-WHOIS-List:
Add "whois-test" to testsuite and distribution archive
Add support for up to 3 targets in WHOIS queries.
also allow up to one wildcard query from local hosts.
Follows ircd 2.10 implementation rather than RFC 2812.
At most 10 entries are returned per wildcard expansion.
WHOIS test cases by Dana Dahlstrom.
previously, any client could join in this configuration:
[Channel]
Name = #test
Modes = tnk
KeyFile = /tmp/foobar
fix this by checking for zero-length key before comparing
key to channel key.
This fixes the followin GCC warning on modern Linux systems as well:
irc-login.c: In function ‘Hello_User’:
irc-login.c:876: warning: ignoring return value of ‘write’,
declared with attribute warn_unused_result
This patch
- makes the server write buffer bigger: 64k,
- makes the regular write buffer bigger: 32k,
- tries to flush the write buffer starting at 4K.
Before this patch, a client got disconnected if the buffer flushing at 4k
failed, now regular clients can store up to 32k and servers up 64k even
if flushing is not possible at the moment (e.g. on slow links).
Don't do a NULL-pointer dereference when a remote server using the
IRC+ protocol sends an invalid PASS command without the required
<serverversion> parameter ...
Port number 0 marks remote servers that try to connect to this
daemon, but where this daemon never tries to establis a connection
on its own: only incoming connections are allowed.
Don't use isdigit() function any more, because it only checks the
first character of the variable value and because it doesn't know
about the minus sign which is required e.g. for "Group = -1".
The MOTD file is read in Read_Config(), so don't read it when handling
the "MotdFile" configuration variable. Instead make sure that it is
initialized properly when (re-)reading the configuration.
Fix commit 5a34bb203a:
It is not enough to strip the "No" prefix from "Ident" and "PAM",
but we have to introduce the new [Features] section to fix all
warning messages of ngIRCd.
Variables "Ident" and "PAM" in [Global] are completely wrong :-(
This enables other servers, services and IRC operators to change
channel topics, even when the client is not joined to this channel.
Now the handler for TOPIC behaves like the one for MODE.
This generic function tests if a client is allowed to do administrative
tasks to a specific channel:
- servers and services are always truested ("allowed everything"),
- channel operators are allowed,
- IRC operarors are allowed if OperCanUseMode is set in the config.
libc should know better than us.
Also, this helps debugging with tools like valgrind:
When you allocate an array of size x, and then erronoulsy
use x+1 valgrind cannot detect the bug because due to ALIGN_()
made by array.c we might have allocated more than size x...
ngircd unfortunately uses several options using double-negation, e.g.
NoIdent = No, NoPam = No, etc.
This renames all options by dropping the "No" prefix, e.g.
"NoIdent = no" becomes "Ident = yes".
The old options will continue to work, but will cause a warning
message.
Also update man pages and default config.
To prevent silly
'Ident = yes' from appearing in --configtest output in the
'ident support not compiled in and Ident Option not used' case,
make default value depend on feature availability.
If feature is available, enable by default, otherwise disable.
We might consider moving these options to a new
[Feature]
section, or something like that, because none of these options are
essential.
Another possible improvement:
'Ident = yes' option in ngircd.conf causes a warning if ngircd was
built without ident support.
This does not happen with e.g. zeroconf....
Configuration variables "MotdFile" and "MotdPhrase" are mutually
exclusive; so don't display content in both of them when running
"ngircd --configtest": instead remember which one is beeing used.
The information written to the "error file" (/tmp/ngircd-<PID>.err) when
ngIRCd is compiled with debug code enabled isn't that usefule, so don't
create this file at all.
If a hostname resolves to more than one IP address (round-robin DNS,
IPv4 and IPv6) and an attempt to connect to the first address fails,
ngIRCd should try to connect to the 2nd address, 3rd address etc.
But because of a wrong variable used in the call to New_Server(),
the wrong server structure has been used in further connection attemps
which possibly lead to connection attempts to already connected servers.