Re: the right ircd for Blitzed

While I've not really been part of the blitzed community, I have been an 
avid reader of the lists, and feel I should give some input here.

Andy Smith wrote:
> On Sat, Jul 17, 2004 at 10:52:43PM +0200, Philipp Kern wrote:
<snip>
>>Ok, but still there's no need for many IPv6 servers. I use it by myself,
>>but there are rarely more than eight people online on soylent-green.
> 
> 
> That's true although regardless of how many users there is, you do
> need multiple servers for redundancy.  If IPv6 were something we
> wanted to promote then we should probably have one server at each
> place we can have one.
> 
> I know that you want to run an IPv6 server, and I'm sympathetic to
> this desire, but not when the code is so bad that it could risk
> significant problems for the rest of the network.

The reason there is no IPv6 support in bahamut is that virtually no ISPs 
support it.  After HE.net started blocking IRC on their IPv6 network, 
and after the Foonet raid, IPv6 usage on IRC dropped to almost the point 
that supporting it is a lost cause.

>>>We could be offering SSL client connections but we can't because our
>>>current ircd doesn't support it.
>>
>>SSL support is something to consider. But does Unreal also provide
>>strong encryption between servers? RC4 is the weak point at the moment.
> 
> 
> I'm told that RC4 is no weaker than the protocols SSL uses[1].  I've
> not enough knowledge of cryptology to tell, and to be honest for the
> sake of IRC I think even compressed links are too much trouble for
> the average attacker.  The issue over SSL vs RC4 is for me one of
> standardisation.  SSL is more standard than ad hoc RC4, so the could
> would be cleaner.

Isn't RC4 one of the encryption protocols used by SSL?  Anyway, SSL is 
more standard, as it's on pretty much every machine with sshd already, 
cutting the need for custom encryption systems.  FYI: The new bahamut 
uses SSL for server to server.  Shouldn't be as hard now to implement it 
for client usage, although it wouldn't be a walk in the park.

>>>We could be masking hosts, we could be offering vhosts.  Those
>>>services are controversial and should be voted on, but at the moment
>>>we can't even begin tot hink of offering them because our current
>>>ircd does not support them.
>>
>>Vhosts count as DNS-Spam, and are IMO just another toy which isn't
>>needed.
 >
> By vhosts I mean something like freenode does, i.e.:
> 
> [email protected]
> 
> I don't really see why that is "DNS SPAM", DNS SPAM (to me) is when
> shell compaies dedicate whole /24 to stupid hostnames for their
> irc-using shell customers.

DNS spam is an ideal solution, just like giving a bum on the street a 
few bucks.  Doesn't do squat for the cause. (Don't confuse bum with 
homeless).  Blocking users because of this ideal does absolutely nothing 
but send the users to other networks.

>>Masking hosts is a valid point for protecting our users from
>>attacks, but it wasn't necessary in the past.
> 
> Whether to do it or not is of course a separate question but it
> would be nice to have the option.

Many networks who implemented masks to protect users have come crashing 
down due to warez channel wars.  I've seen it happen too many times for 
it to be just coincidence.  When you take away their ability to packet 
each other, who do you think they will go after next?  I would throw in 
a comment about channel ops not being able to check proxies because the 
monitor of choice for the network sucks, as I do for many networks. 
However, I think we all know BOPM kicks ass, so that point is null.

>>Bahamut isn't bloated at all and had hardly any memory leaks I noticed.
>>It still runs on milk with 190 days uptime at 10M.
>>ircd      1108  0.4  1.7 10116 4420 ?        S    Jan09 1097:44 ./ircd

Google results for unrealircd vulnerability: 9,450
Google results for bahamut vulnerability: 1,000
Compare this with the number of networks using the two IRCds, and it 
makes a huge difference.

> 23:02:39 <Solaris> PID   USERNAME PRI NICE SIZE  RES   STATE  TIME  WCPU  CPU   COMMAND
> 23:02:40 <Solaris> 27978 ircd     28  0    1924K 1244K RUN    0:00  0.51% 0.29% top
> 23:02:40 <Solaris> 185   ircd     2   0    6940K 6440K select 14:06 0.00% 0.00% ircd
> 23:02:41 <Solaris> 186   duden    2   0    3628K 3340K select 1:56 0.00% 0.00% eggdrop-1.6.16
> 23:03:16 <Solaris> on this Server are now 509 Users
> 23:03:55 <grifferz> and on the network as a whole?
> 23:04:54 <Solaris> Now 1600 User online including the +i Users
> 23:12:42 <grifferz> how long has that ircd been up?
> 23:14:14 <Solaris> we update for tow Weeks to 3.2.1
> 23:14:25 <Solaris> but with 3.2 it was up 6 Months
> 23:14:29 <Solaris> without any Crash
> 
> (this discussion wasn't prompted by Solaris asking me why we don't
> use Unreal.  I told him I'd been thinking about Unreal and he said
> he runs it on eu-irc, which is bigger than blitzed)
> 
> 
>>Restarts all the time because of Unreal upgrades would neither be
>>helpful for our userbase.
> 
> 
> Unreal does seem to have a lot of releases.  I agree that because it
> has so many features, bugs in those features require regular
> updates.  A lot of them must be adding more features though, so we
> wouldn't have to restart purely for that.  Also, modules would help.

UnrealIRCd has this disease creeping into many a software: featuritis. 
Many (in fact, most) of these features seem to be built with the intent 
on abuse.  Umode I? Using SAJOIN to force people to join channels? 
Spying on privmsgs?  +u (if that's auditorium mode) to block people from 
seeing who's in a channel?  I've seen the latter used by virtually every 
DDoS botnet writer to hide his bots in the open, so to speak.  Doesn't 
speak to well for the IRCd.  (And yes, I know the creator can't do jack 
about who uses the IRCd, but he could do jack to deter it.)

FYI 2: the new bahamut also supports modules, but in a limited state.  A 
module can be triggered on a 10 second time; a privmsg or notice to any 
client or channel, existent or not, or to the server itself; before 
login (ircd sending nick to other servers); after login; after 
connection motd; and lastly, on signoff.

Another issue is that Unreal, at least for me, has shown an inability to 
host larger networks.  However, at blitzed's size, there should be no 
issues with network size.

In closing, the problems with Unreal lie within its rather large 
vulnerability record, and too many features which can be abused to cause 
network disruption or general discomfort for the network population.

* wshs removes tinfoil hat.

> [1] http://s.dgl.cx/6bc