|
InspIRCd Developer
Join Date: Apr 2004
Posts: 532
|
Full changelog:
- Patch to not increment nickchange counter for nick changes that would be blocked elsewhere, fixes bug #874. Patch by dz, but he was too slow to commit it
 - Don't apply bans to E:Lined connections. Should fix bug #896.
- Check that the <connect> block actually specifies a host before checking whether the user matches it. Allows for <connect> blocks on other criteria, like ports - fixes bug #797.
- Reconnect to LDAP server if connection goes away, fixes bug #818.
- Rehash message fixes from Ankit, thanks
- Send a failure numeric instead of QuitUser when STARTTLS is used after registration, because it could be used accidentally by users. Suggested by dKingston
- Don't send out global snotices from every server on NICKUNLOCK, fixes bug #892 reported by Taros
- Use correct behavior (as per C++) for 'delete NULL' on windows; found by Adam and CyberBotX
- Remove bad check for channel op in m_auditorium, fixes bug #886
- Remove m_ssl_oper_cert and clarify how to use it in m_sslinfo. Patch by jdhore.
- Use server name for source of remote TIME rather than SID
- Fix m_callerid message routing for remote users, destination got no notice of attempted message
- Ask the user if they want to generate SSL certificates before generating one
- Fix symlink creation of openssl if both it and gnutls are selected
- Fix excessive snomask sending on fitler add/removal
- Fix double-send of SSL metadata in openssl, possible segfault in gnutls
- Updates to example configuration [by jdhore]
- Merge m_ssl_oper_cert and m_ssl_data into m_sslinfo to reduce the number of SSL modules
- Send ssl_cert metadata on signon
- Remove memory-wasting map in ssl_cert
- Migrate SSL metadata and visible information (/whois line) to single module
- Fix backport compile error: m_ssl_* uses ServerInstance->Config->ports which is now ServerInstance->ports
- Fix error reporting on rehash
- Make rehash generate a new ServerInstance->Config object
- Move list of open ports out of Config object
- Check for oper max chans before user max chans, fixes code weirdness and bug #876.
- Fix +g messages being borked with multiple servers, fixes bug #880
- Add BS/BotServ to default atheme aliases.
- Roll back mysterious DNS change from 2007 which broke multiple-answer DNS requests (CNAME, A). It looks like this was to be related to bitmask dnsbl support, but it would never have worked anyway (resultnum was always 0..).
- m_sslinfo does not depend on external libraries, it should not be in extra
- Make pkg-config failure error clearer
- Rebuild modeclasses.a when configure -update is run
- Fix incorrect attempted inline of irc::sockets::sa_size
- Fix sockaddr length argument, BSD will complain if it doesn't exactly match the expected size
- Fix ENCAP forwarding with multiword final parameter, noticed by Namegduf
- Improve valdebug-unattended: increase timeout to 90 seconds, and generate/use suppresions
- Backport BufferedSocket:
oConnect rewrite to fix valgrind warning - These commands do not need to be VF_COMMON
- Don't globally route SETNAME, the FNAME is sufficient. Fixes bug #875
- Makefile updates: clean up some installation assumptions
- Allow SSL fingerprint-based server authentication
- Don't display locally bound SSL ports in SSL= (for hidden services and such)
- Makefile updates: Add "make help" and "make M=m_modname" to build an individual module
- Add daniel's third-party module repository
- Use pkg-config to search for gnutls rather than the now-removed libgnutls-config
- Change m_override to send snomasks globally
- Ignore and remove {BSD,GNU}makefile
- Makefile updates: Fix BSD make conditionals
- Makefile updates: Allow run-cc wrapper to be disabled by "make V=1"
- Makefile updates: clean up command lines
- This time it's actually compile tested
- OK, it's more useful if this compiles...
- Fix unsafe iteration pattern in m_timedbans - vector::erase invalidates all iterators following the elements to be erased.
- Makefile updates: remove MAKEPROG/MAKEORDER
- Remove the ServerConfig:
irValid check that breaks symlinks - Make timestamp output in /CHECK consistent
- OnRehash changes: split to multiple hooks to clarify use and prevent explosion of rehashes on a /REHASH *
- m_dnsbl: Add default duration and error message on invalid duration
- Fix UID collision on reload of m_spanningtree.so
- Fix some incorrect STL iterations (using < end() instead of != end())
- Fix r11381 checking HideWhoisServer incorrectly
- Fix incomple memset() in irc::sockets::aptosa, add support for converting empty addresses
- Make User::ServerInstance public so that FakeUser can access it. There is no reason to make this private, really
- Hide server sources if HideWhoisServer is set, noticed by Brain
- Show correct server source to users on FTOPIC and FMODE
- Use Utils->ServerUser instead of ServerInstance->FakeClient in m_spanningtree
- Replace loopCall detection in OnUserKick with a check for the m_spanningtree fake user
- Use a FakeUser source for server-sourced commands in spanningtree
- Create subclass of User for FakeClient, to allow for use as command source
- Fix for some modules sending global snotices at each individual server in a server path.
- Translate UIDs properly in m_sakick
- Revert some files accidentally committed in r11371. I'm apparently doing a very bad job of multitasking.
- Fix an issue in r11370 spotted by danieldg
- Fix m_sakick to properly send local/global snomasks like other modules, instead of repeating a global snomask once per linked server. Found by SnoFox
- Remove handshake timer on server sockets that die before completing handshake
- Compile fix: openssl does not have an ISSL_CLOSING state, unlike gnutls
- Fix ListenSocketBase::GetPort() returning incorrect value on ports not bound to a specific address
- Fix two compile errors found by Shawn
- Set errno in openssl when handshake fails
- Remove FindDescriptor wrapper for GetRef, it doesn't do anything useful and in fact is a bit harmful in some ways as we've seen..
- Set errno properly on GnuTLS sessions that die during the SSL handshake
- m_filter: Do not attempt to filter messages from remote users
- Canonicalize bind addresses when creating listen sockets
- Fix unnecessary heap allocation on rehash: stack allocated ConfigReader is fine.
- Some general tidyup in here (also some possible desync fixes, check IS_LOCAL..), also check that the target channel exists and sender is opped before allowing +b redirects from a remote sender. Fixes bug #851, reported by SnoFox.
- (for now), comment out opercount from individual servers in XML stats. This can be addressed in a future release. Bandaid fix for bug #846, reported by SnoFox.
- Allow E:Lines to exempt targets from SHUN and RLINE. Fixes bug #852 reported by Taros.
- Change making callerid significantly less expensive for quitting large amounts of users at once. This is *not* runtime tested yet. Testnet, please update this module.
- Minor cleanup: move RemoveFromAllAccepts() to a private class method
- Globally route "a" snotices for more modules. Patch from Milliways.
- Remove commands makefile also on make distclean.
- Fix two stupid mistakes here
- Patch from eLement notifying the user who sent a NICKLOCK that the target was locked successfully, fixes bug #854.
- Remove an old debug log, and fix compile.
- Fix: Channel aliases should work a bit more like user aliases. This is a BC change: all existing channel aliases will likely be BROKEN, and will need to be changed. Like user aliases, $1 is now the fantasy command that was invoked, $2 is the first real param of the command. Reported by Namegduf.
- Replace $chan to "" for a non-channel command.
- Send rehash message to global sno, fixes bug #855. Also only send message when the rehash happens.
- Fix SILENCE command in m_silence being reported as being in m_slience_ext, noticed by SnoFox
- Fix bug found by Jeremy from Teranova, thanks: if an explicit bind to an IP is not made, don't assume all <bind> tags are for ipv6 IPs on an IPv6 compile (WTF).
- Fix m_redirect to work properly with SAMODE
- How about we make this compile?
- Normalize IPv6 addresses when producing SSL port list, fixes bug #853
- Add IP address conversion utility functions, irc::sockets::aptosa and irc::sockets::satoap
- Overloading SetMode seems to cause the compiler to interpret "" as a bool and not a string. Fixes bug #849
- Fix OnSetAway hook denying remote users' change rather than local users
- Remove sending +q from OnPostJoin, fixes bug #847. See removed comment for why it existed
- Correct 005 output of m_ssl_gnutls when used for STARTTLS with no client SSL ports configured
- Enable lag-check pings on active servers - if bandwidth waste is an issue, there should be an option to disable lag pings
- Remove lag reporting in /map from servers that do not yet have a valid value; <1ms is misleading.
- Fix OnUserRegister being triggered twice if someone sent a NICK while waiting on an ident timeout
- Fix crash from unregistered users joining a channel, discovered by Namegduf
- Addendum to previous ping patch: NextPingTime was not being set at endburst, and servers without ENDBURST were no longer being detected.
- Fix PingFlag not being set correctly for servers whose burst takes longer than PingFreq
- Abort link early when mismatched modules are detected so that the error message is more useful
- Actually use irc::modestacker correctly
- Use loopCall to protect OnUserKick rather than the triggerevents hack
- Initialize loopCall on construction
- Add loopCall to prevent double-broadcasting new X:lines
- Send snotices when m_dnsbl adds an xline; fixes bug #837
- Change a few more modules to send global a/A snotices
- Use server source for RemoveMode, fixes bug #838
- Convert m_dnsbl to send globally routed snomasks. Fixes bug #836
- Add global-routing snomask functions
- Don't send out UID for users that are quitting, because the QUIT won't be sent and they become a ghost
- Fix CullList to not use O(n^2) version of vector clear
- Fix realhost leak when users on different servers get +W notifications. Noticed by mooncup.
- Remove CullList rate limiting to fix UID collisions on large networks
- Fix memory leak on rehash: opertype used char* instead of std::string
- Fix memory leak in m_filter rehash
Fix SetModeParam to use std::string and handle edge cases.
__________________
Master of all trades; jack of none.
|