Now ngIRCd catches more errors on the server-to-server (S2S) protocol
that could crash the daemon before. This hasn't been a real problem
because the IRC S2S protocol is "trusted" by design, but the behavior
is much better now.
Thanks to wowaname on #ngircd for pointing this out!
At the moment ngircd fails the tests for reproducible builds in Debian
since it uses the __DATE__ and __TIME__ macros for the INFO command.
Instead of patching this out I decided to implement an optional
constant BIRTHTIME that allows you to set a time stamp for the "Birth
Date" information, in seconds since the epoch, like in
export CFLAGS += -DBIRTHTIME=$(shell date +%s --date="2015/08/15 23:42:22")
In the future, Debian will provide a SOURCE_DATE_EPOCH environment
variable, dealing with the situation until then will be my job.
The time format was taken from the NGIRCd_StartStr formatting in
ngircd.c so the "Birth Date" and "On-line since" lines in the INFO
output look similar:
:irc.example.net 371 nick :ngIRCd 22.1-IDENT+IPv6+IRCPLUS+PAM+SSL+SYSLOG+ZLIB-x86_64/pc/linux-gnu
:irc.example.net 371 nick :Birth Date: Tue Aug 25 2015 at 18:11:11 (CEST)
:irc.example.net 371 nick :On-line since Tue Aug 25 2015 at 18:11:33 (CEST)
:irc.example.net 374 nick :End of INFO list
The format of the time stamped is changed, but as far as I can tell, there's no
rule that is violated by that.
Bonus level: Reformat the messages so the time stamps are aligned.
AUTH is a valid nickname so sending notices to it is probably not
a good idea. Use * as the target instead as done with numerics
when the nick is not available.
This mimics the behaviour in Charybdis, IRCD-Hybrid, InspIRCd 2.2,
Plexus 4, etc.
Reconnecting to ngIRCd 22.1 built with OpenSSL with some OpenSSL
clients, including Pidgin and stunnel 5.06, attempts to reuse a session
and fails due to the absence of this line.
The error message in syslog from ngIRCd is:
> SSL protocol error: SSL_accept (error:140D9115:SSL
> routines:SSL_GET_PREV_SESSION:session id context uninitialized)
This patch appears to fix the problem for both Pidgin and stunnel; it
may work for other OpenSSL clients that attempt to re-use sessions.
* <https://github.com/ngircd/ngircd/issues/182>
* <https://developer.pidgin.im/ticket/11568>
* <https://www.openssl.org/docs/ssl/SSL_CTX_set_session_id_context.html>
* LucentW/master:
Fix with oneshot invites
Fixed building issues\
Implement timestamp tracking of invites
Keep track of who placed bans/invites/excepts
IRC operators w/OperCanMode can kick anyone [already cherry-picked]
Closes#203, Closes#205.
For example, Interix is missing this function, which prevented
ngIRCd to build on this platform. When setgroups(3) isn't available,
a warning message is issued when ngIRCd starts up.
lightIRC and other clients expecting RPL_LISTSTART should now behave correctly.
Closes#207.
(cherry picked from commit 0680ce5fd99bc643651d1433bcdaf271aeb73c46)
Up to now, only the the string ("haystack") became lowercased and was
the compared to the pattern ("needle") -- which failed, when the pattern
itself wasn't all lowercase ...
- Update documentation in ngircd.conf(5)
- LIST: Don't hide channels for IRC Ops when "MorePrivacy" is in effect
- WHOIS: Don't hide IP addresses/hostnames when "MorePrivacy" is in effect
Closes#198
Now lists nodes also have the "onlyonce" field, since the valid_until is
used to keep the timestamp of placing. Found no references to onlyonce
or about valid_until being == 1, though, so it might be unused, but
still available for other enhancements.
Implement ERR_USERNOTONSERV(504) numeric and make sure that the
target user is on the same server when inviting other users to
local ("&") channels.
ircd-ratbox uses the ERR_USERNOTONSERV(504) numeric for this, and I
think this is a good idea -- other IRC daemons (like ircu) silently
drop such impossible invites, but thats not a big benefit ...
Idea by Cahata, thanks! Closes#183.
Up to now when receiving a MODE command, ngIRCd only reported the channel
creation time to clients that were members of the channel. This patch
reports the channel creation time to all clients, regardless if they are
joined to that channel or not.
At least ircd-seven behaves like this.
This closes#188. Thanks Cahata!
No other IRC daemon seems to do this (today?), don't remember why
ngIRCd did it in the first place ...
Closes#185.
Reported by Cahata in #ngircd, thanks!
The logic is reversed ...
This bug has been introduced by commit c74115f2, "Simplify mode checking on
channels and users within a channel", ngIRCd releases 21, 21.1, and 22 are
affected :-(
Problem reported by Cahata in #ngircd, Thanks!
Up to now, ban, invite, and G-Line lists have been synced between servers
while linking -- but obviously nobody noticed that except list have been
missing ever since. Until now.
Thanks to "j4jackj", who reported this issue in #ngircd.
David Binderman <dcb314@hotmail.com> reported the following compiler warning,
which is a real bug in ngIRCd, thanks!
conn.c:2077:55: warning: logical not is only applied to the left hand
side of comparison [-Wlogical-not-parentheses]
The "SYSLOG" #define isn't related to "Conf_ScrubCTCP" at all, so
initialize the latter even when "SYSLOG" isn't #define'd.
Pointed out by wowaname on #ngircd, thanks!
There are installations out there that would like to configure more
than 16 links per server, so increase this limit. Best would be to
get rid of MAX_SERVERS altogether and make if fully dynamic, but
start with this quick and dirty hack ...
Different operating systems do behave quite differently when doing DNS
lookups, for example "127.0.0.1" sometimes resolves to "localhost" and
sometimes to "localhost.localdomain" (for example OpenBSD). And other
systems resolve "localhost" to the real host name (for example Cygwin).
So not using DNS at all makes the test site much more portable.
Let IRC_MODE() detect that the "fake" MODE command originated on the local
sever, which enables all modes to be settable using "DefaultUserModes"
that can be set by regular MODE commands, including modes only settable by
IRC Operators.
ngIRCd relaxes its flood protection for users having the user mode "F" set
and allows them to rapidly send data to the daemon. This mode is only
settable by IRC Operators and can cause problems in the network -- so be
careful and only set it on "trusted" clients!
User mode "F" is used by Bahamut for this purpose, for example, see
<http://docs.dal.net/docs/modes.html#4.9>.
ngIRCd uses "command throttling" and "bps throttling" (bytes per second).
The states are detected in different functions, Conn_Handler() and
Read_Request(), but handle the actual "throttling" in a common function:
this enables us to guarantee consistent behavior and to disable throttling
for special connections in only one place, eventually.
Change all #define's to follow the form
#define DEBUG_xxx {0|1}
to disable (0, default) or enable (1) additional debug messages.
And somewhat enhance some DEBUG_BUFFER messages.
This partially reverts commit b130b35f4, "Update #include's: remove
unused and add missing ones", but fixes the following compiler and
analyzer warnings of Apple Xcode 5:
"Semantic issue: No previous prototype for function 'yyy'"
The "deheader" tool (<http://www.catb.org/~esr/deheader/>) has been
used to find unused #include directives as well as missing ones.
Tested on:
- A/UX 3.1.1
- ArchLinux (2014-03-17)
- Debian GNU/Hurd
- Debian GNU/Linux 6.0.9
- Debian GNU/Linux 7.4
- Fedora 20
- FreeBSD 9.2
- OpenBSD 4.8
- OpenBSD 5.1
- OS X 10.9
- Solaris 11
Fix the following two errors emitted by the Apple K&R C compiler
on Apple A/UX:
"./class.c", line 47: no automatic aggregate initialization
"./class.c", line 47: illegal lhs of assignment operator
"./conf.c", line 1052: syntax error
Tested on A/UX 3.1.1.
The cloaked IRC mask of a user is his visible mask, so the daemon has
to use it for generating the "one time" entries for the invite list of
the given channel.
Without this patch, ngIRCd records the real IRC mask which will never
match while the target client is "+x", and even worse, will disclose
the real mask on "MODE #channel +I" commands :-/
Bug reported by Cahata on #ngircd, thanks!
This fixes commit 2e168c78 ("Return ISUPPORT(005) numerics on
'VERSION'") and make sure, that the 005 numerics are correctly
routed back to the sender.
Don't support "range marching" in our pattern matching code using
the "[...]" syntax, because [ and ] are valid characters in nick
names and one has to quote them currently using the "\" character,
which is quite unexpected. For example:
Nick "te[st" => "MODE #channel +b te\[st"
And remove quoting altogether, too, because "*" and "?" don't need
to be quoted because these characters are not allowed in IRC masks,
nicks, and hostnames.
Reported by "hifi" (Toni Spets) on IRC, thanks!
According to RFC 2812 3.2.3 "Channel mode message" and the examples
there, it looks like clients should use "MODE -k <key>" to unset channel
keys; and that's how other servers and services behave and do expect it.
(But please note that this is NOT the case for "MODE -l"!)
In the end, it doesn't make sense to specify a key when UNsetting it at
all, and different services behave diffrently when clients do not send
the currently set key to unset it - some ignore such calls, for example!
But this implementation is quite relaxed, it accepts any key when
unsetting channel mode "k" and even accepts no key at all. But the reply
will always include an "*" character for every "-k" parameter.
C99 states that vsnprintf() "returns the number of characters that
would have been printed if the n were unlimited"; but according to the
Linux manual page "glibc until 2.0.6 would return -1 when the output
was truncated" -- so we have to handle both cases ...
Test functions snprintf(), strlcpy(), strlcat(), vsnprintf() for
correctness, not only existance (which was quite useless, because
if they weren't available, the program could not have been linked
at all ...).
The new configuration variable "Network" is used to set the (completely
optional) "network name", to which this instance of the daemon belongs.
When set, this name is used in the ISUPPORT(005) numeric which is sent to
all clients connecting to the server after logging in.
Closes bug #165.
When DNS lookups are disabled, don't set the hostname received by the
WEBIRC command, but use the IP address instead.
Reported by Toni Spets <toni.spets@iki.fi>, thanks!
Closes bug #167.
At least AIX 4.3.3 and 5.1 have a broken implementation of getaddrinfo()
which doesn't handle "0" as numeric service correctly. This patch adds
a configure check for this case and changes all calling functions to only
use getaddrinfo() if it "works".
See <http://www.stacken.kth.se/lists/heimdal-discuss/2004-05/msg00059.html>
Increase the penalty for a command before checking its arguments. This
makes the handling more consistent and allow us to move more penalties to
Handle_Request().
This fixes the following error when compiling without zlib support:
irc.c: In function ‘Option_String’:
irc.c:487: error: ‘options’ undeclared (first use in this function)
Reported by "der_baer" on #ngircd.
FreeBSD prior to 10.0 does not automatically stir on fork(). Same with
current NetBSD. If arc4random_stir() is present assume is needed and
call it instead of srand().
This fixes the following warning on OpenBSD 5.3:
ngircd.o(.text+0xeb4): In function `main':
src/ngircd/ngircd.c:300: warning: strcat() is almost always misused,
please use strlcat()
Thanks to Götz Hoffart for reporting this!
Kill all clients that match a new GLINE/KLINE mask and genrate apropriate
KILL commands. These KILL commands can be superfluous, but are required
when the IRC Operator isn't allowd to set remote G-Lines or if there are
older servers in the network that don't kill clients on GLINE/KLINE.
Closes bug #156.
The old local function Kill_Nick() in irc.c has been an ugly hack. This
patch implements a generic function for killing clients.
Adjust all callers of Kill_Nick() and respect the return code!
Now invalid prefixes aren't logged no more when originating from an other
server (besides in debug mode), and spoofed prefixes are correctly logged
using LOG_WARNING (from an other server) or LOG_ERR (from a client) levels.
In addition, the log message texts have been adjusted to better reflect
what will happen: commands with invalid prefixes are ignored and logged,
commands with spoofed prefixes will result in the client being disconncted
(regular users) or the command being ignored (other servers).
This cleans up logging of commands related to already KILL'ed clients.
All places where Client_OperByMe() is used can either be converted to
Client_HasMode(Client, 'o') or Op_Check().
And Op_Check() itself can use the connection handle for deciding whether
the IRC Operator is a local user or not.
* alex/bug162-SSLCipherList:
Cipher list selection for GnuTLS
ConnSSL_Init_SSL(): correctly set CONN_SSL flag
Cipher list selection for OpenSSL
ConnSSL_InitLibrary(): Code cleanup
In some error cases conn_id will be left as SERVER_WAIT and
subsequently ignored in Check_Servers(). Ensure conn_id is set to
NONE before returning from New_Server() if we couldn't establish
the connection.
Prompted by a report from gabrielgi-at-gmail-dot-com.
Without this patch, ngIRCd ignores SSL-related messages and continues
to start up but only listens on plain text communication ports -- and
this most probably isn't what the administrator wanted ...
Closes bug #163.
This patch introduces the possibility to arbitrarily select ciphers which
should be promoted resp. declined when establishing a SSL connection
with a client by implementing the new configuration option "CipherList".
By default, OpenSSL would accept low and medium strength and RC-4 ciphers,
which nowadays are known to be broken.
This patch only implements the feature for OpenSSL. A GnuTLS counterpart
has to be implemented in another patch ...
Original patch by Bastian <bastian-ngircd@t6l.de>.
Closes bug #162.
As per RFC whois should return 431 if no nick is provided. While
here convert upper check to use irc-macros. As a bonus we get to set
the penalty for free.
The configuration option "AllowedChannelTypes" must only be enforced for
regular clients and not for remote servers. Channels created by other
servres are always allowed, because they already exist and the daemon
must stay in sync with the network.
Without this patch, ngIRCd logged the "IDENT lookup for connection X:
no result"-message even when IDENT lookups have been disabled using the
"Ident = no" configuration option, which is a little bit misleading.
Reported by "btwe" in #ngircd.
This patch introduces a new field in the CLIENT structure, "ipa_text",
which points to an optional textual representation of the client IP
address (or NULL) which can be used to store the "real" IP address
information of a client using the "WEBIRC" protocol.
Without this patch, ngIRCd ignored the <ip-address> paramater ...
In addition, the functions Client_SetIPAText() and Client_IPAText()
have been introduced to set and get the textual representation of the
client IP address.
Client_IPAText() can be used even when no "IP address text" has been
set before, it then returns the real IP address of the connection.
Closes bug #159.
The new configuration option "DefaultUserModes" lists user modes that
become automatically set on new local clients right after login.
Please note that only modes can be set that the client could set on
itself, you can't set "a" (away) or "o" (IRC Op), for example! User
modes "i" (invisible) or "x" (cloaked) etc. are "interesting", though.
Default: set no modes (like without this patch).
Closes bug #160.
There have been code paths that ignored the return code of Handle_Write()
when sending "notice auth" messages to new clients connecting to the
server. But because Handle_Write() would have closed the client connection
again if an error occurred, this would have resulted in new errors and
assert()'s later on that could have crashed the server (denial of service).
Only setups having the configuration option "NoticeAuth" enabled are
affected, which is not the default.
CVE-2013-5580.
Channel_Modes() and Channel_Topic() return static memory that is zeroed at
creation time. Fix checking for empty modes and topic so they're propagated
correctly.
This function is used to send "error messages", including numerics,
back to clients and to automatically enforce a 2 second penalty. With
this patch, all error results enforces a delay for the client.
All callers of IRC_WriteStrClient(ERR_xxx) have been converted.
Please note that this patch prolongs the time "make check" needs
to complete its tests, because of lots of new enforced penalties ...
These functions usually write to more than one target, and therefore
can't return value indicating success or failure in a sane way. And, even
more omportant, these functions should send to as much as possible clients
and not stop at the first client which isn't reachable any more!
The new configuration variable "AllowedChannelTypes" lists all allowed
channel types (channel prefixes) for newly created channels on the local
server. By default, all supported channel types are allowed.
If set to the empty string, local clients can't create new channels at
all, which equals the old "PredefChannelsOnly = yes" setting.
This patch deprecates the "PredefChannelsOnly" variable, too, but it is
still supported and translated to the apropriate "AllowedChannelTypes"
setting. When "PredefChannelsOnly" is processed, a warning message is
logged.
Closes bug #152.
Move Announce_User() to client.c and rename it to Client_Announce().
Use this in cb_introduceClient() instead of duplicating the code.
This fix the certificate fingerprint announcement for new clients.
Also ensure the certificate fingerprint is only announced if the
client supports it (`M' flag).
