You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
tdenetwork/lanbrowsing/lisa
Slávek Banko 5b915e0e82
Fix FTBFS from commit 6d93398d01.
8 months ago
..
AUTHORS Copy the KDE 3.5 branch to branches/trinity for new KDE 3.5 features. 15 years ago
CMakeLists.txt [kdenetwork/lanbrowsing] added cmake support 13 years ago
COPYING Update COPYING file. 12 years ago
ChangeLog Copy the KDE 3.5 branch to branches/trinity for new KDE 3.5 features. 15 years ago
INSTALL Copy the KDE 3.5 branch to branches/trinity for new KDE 3.5 features. 15 years ago
Makefile.am Copy the KDE 3.5 branch to branches/trinity for new KDE 3.5 features. 15 years ago
NEWS Copy the KDE 3.5 branch to branches/trinity for new KDE 3.5 features. 15 years ago
README Rebranding: TDE Control Center --> Trinity Control Center 11 years ago
addressvalidator.cpp rename the following methods: 13 years ago
addressvalidator.h rename the following methods: 13 years ago
client.cpp rename the following methods: 13 years ago
client.h rename the following methods: 13 years ago
configfile.cpp Copy the KDE 3.5 branch to branches/trinity for new KDE 3.5 features. 15 years ago
configfile.h Copy the KDE 3.5 branch to branches/trinity for new KDE 3.5 features. 15 years ago
ipaddress.cpp Copy the KDE 3.5 branch to branches/trinity for new KDE 3.5 features. 15 years ago
ipaddress.h Copy the KDE 3.5 branch to branches/trinity for new KDE 3.5 features. 15 years ago
lisadefines.h Copy the KDE 3.5 branch to branches/trinity for new KDE 3.5 features. 15 years ago
main.cpp Fix FTBFS with stricter C++11 6 years ago
mystring.cpp rename the following methods: 13 years ago
mystring.h rename the following methods: 13 years ago
netmanager.cpp Fix FTBFS with stricter C++11 6 years ago
netmanager.h Fix FTBFS from commit 6d93398d01. 8 months ago
netscanner.cpp rename the following methods: 13 years ago
netscanner.h Fix FTBFS from commit 6d93398d01. 8 months ago
strictmain.cpp Fix FTBFS with stricter C++11 6 years ago
stringlist.cpp Copy the KDE 3.5 branch to branches/trinity for new KDE 3.5 features. 15 years ago
stringlist.h Copy the KDE 3.5 branch to branches/trinity for new KDE 3.5 features. 15 years ago
tcpnode.h Copy the KDE 3.5 branch to branches/trinity for new KDE 3.5 features. 15 years ago

README

This is the readme to the Lan Information Server LISa and the Restricted LAN
Information Server resLISa.

+---------------+
|     LISa      |
+---------------+

LISa is intended to provide a kind of "network neighbourhood" but only
relying on the TCP/IP protocol stack, no smb or whatever.
It is completely independent from KDE/Qt.
The list of running hosts is provided via TCP port 7741.

LISa supports two ways of finding hosts:

1. you give LISa a range of of IP-addresses, then LISa will send 
ICMP echo requests to all given IP addresses and wait for the answers

2. you can say LISa to execute "nmblookup "*"  ", i.e. the command line tool
nmblookup must be installed, it is part of the samba package.
nmblookup "*" sends a broadcast to the attached networks and all hosts
running smb-services will answer this broadcast

lisa and reslisa are distributed under the GNU General Public License.



How it works
-------------

In the configuration file you provide a range of IP-addresses which
LISa should check, wether they are running. In the most simple case
this could be your network address/subnetmask, then LISa would
check every possible host of your network wether it is running.
The hosts are checked using ICMP echo requests. To be able to send and receive
ICMP echo requests and replies the program has to open a so-called
"raw socket". Therefor it needs root privileges. This socket is opened
right after the start of the program, after successfully opening the socket
root privileges are dropped immediately (see main.cpp and strictmain.cpp).
If you configure LISa this way, that it also uses nmblookup, it will 
popen("nmblookup \"*\"") and then parse the results.

Since the ICMP requests and the broadcasts can cause some network traffic 
if there are more than one such server running in one network, the servers
cooperate with each other. Before they start pinging (or nmblookup), 
they send a broadcast on port 7741. 
If somebody answers this broadcast, they will retrieve the complete list 
of running hosts via TCP port 7741 from this host and will not start to 
ping (or nmblookup) theirselves. If nobody answers, the host which sent 
the broadcast will start pinging the hosts (or nmblookup) and then open a 
socket which listens for the mentioned broadcasts. If the host received an 
answer to his broadcast, it won't have the socket for listening to the 
broadcasts open. So usually exactly one of the servers will have this 
socket open and only this one will actually ping (or nmblookup) the hosts. 
In other words, the servers are lazy, they work like "I will only do something
if nobody else can do it for me".

There is another feature which reduces the network load. Let's say you configured LISa 
to update all 10 minutes. Now you don't access your server very often.
If nobody accesses the server for the last update period, the server will
update (either itself or from the one which actually does the work) and then
double its update period, i.e. the next update will happen after 20 minutes.
This will happen 4 times, so if nobody accesses the server with update period 
10 minutes for a long time, its update interval will increase up to
160 minutes, almost three hours. If then somebody accesses the data from the 
server, he will get an old list ( up to 160 minutes old). With accessing 
the server will reset its update interval to its initial value, 
i.e. 10 minutes and immediately start updating if the last update is more 
than these 10 minutes over. This means if you get a very old list, you can try
some seconds later again and you should get a current version.
This will have fast effect for the servers, which don't ping (or nmblookup)
theirselves, since only one user usually accesses them, and it will have
less effect for the server which does the pinging (or nmblookup), since 
this server is accessed from all other servers in the network.


This way it is possible that many hosts in a network run this server, but
the net load will remain low. For the user it is not necessary to know
wether there is a server (i.e. a name server or fileserver or whatever)
in the network which also runs LISa. He can always run LISa locally
and LISa will detect if there is one, transparently to the user.
The first client for LISa is an ioslave for KDE2, so the user
can enter there lan://localhost/ or lan:/, which will both
contact LISa on the own system.
If there is a machine which runs all the time and the user knows
that this machine also runs LISa, he can use his LISa client directly with
this server (would be with the mentioned ioslave lan://the_server_name/).

If you don't want that your LISa takes part in the broadcasting, but always
does the pinging itself, make it use another port with the
command line option --port or -p.
This is not recommended !

If you send SIGHUP to LISa, it will reread its configfile.
If you send SIGUSR1 to LISa, it will print some status information to stdout.

The data provided over the socket has a simple format:

<decimal ip address in network byte order><one space 0x20><full name of the host><a terminating '\0'><newline '\n'>
and the last line
0 succeeded<'\n'>

e.g.

"17302538 some_host.whatever.de
18285834 linux.whatever.de
17827082 nameserver.whatever.de
0 succeeded
"

This should make it easy parseable.

If there are very strict security rules in your network, some people
might consider the pinging a potential attack. If you
have problems with this, try the restricted version, resLISa.

+------------------+
|     resLISa      |
+------------------+

If you hav very strict security rules in your network or you don't want to
have another port open or whatever, you can use resLISa. 

With resLISa you can't ping whole networks and address ranges, you can give
resLISa up to currently 64 hosts by their names in its config file. These
will be pinged. You are still able to use nmblookup.
resLISa will also only provide the information over a unix domain socket, i.e.
not over the network. The name of the socket is "/tmp/resLisa-YourLoginname",
so resLISa can be safely run by more users on one machine.
Since it should also not produce a security risk of any kind it is
safe to install reslisa setuid root. root privileges will be dropped
right after startup (see strictmain.cpp), they are only needed to create a raw socket
for sending the ICMP echo requests..
It will also not send or receive broadcasts.
The first client for this is also an ioslave for KDE2 (makes rlan:/ in e.g. konqy).



Configuration
-------------

Now an example config file:

PingAddresses = 192.168.100.0/255.255.255.0;192.168.100.10-192.168.199.19;192.168.200.1;192-192.168-168.100-199.0-9;
PingNames = bb_mail;
AllowedAddresses = 192.168.0.0/255.255.0.0
BroadcastNetwork = 192.168.100.0/255.255.255.0
SearchUsingNmblookup = 1                #also try nmblookup
FirstWait = 30                          #30 hundredth seconds
SecondWait = -1                         #only one try
#SecondWait = 60                         #try twice, and the second time wait 0.6 seconds
UpdatePeriod = 300                      #update after 300 secs
DeliverUnnamedHosts = 0                 #don't publish hosts without name
MaxPingsAtOnce = 256                    #send up to 256 ICMP echo requests at once


PingAddresses

This is probably the most important entry.
Here you say which addresses will be pinged. You can specify multiple
ranges, they are divided by semicolons.

There are four possible ways to define addresses:
-net address/network mask: 192.168.100.0/255.255.255.0, i.e. an IP address
 and the assigned network mask. This doesn't have the real network address
 and netmask of your machine, it can be less. E.g. if you have
 10.0.0.0/255.0.0.0, you could specify 10.1.2.0/255.255.255.0 if you are only 
 interested in these addresses. The combination IP address-network mask
 must be divided by a slash "/" and the address does not have to be a real
 network address, it can also be a host address of the desired network,
 i.e. 10.12.34.67/255.0.0.0 is the same as 10.0.0.0/255.0.0.0 .

-a range of following IP addresses: 192.168.100.10-192.168.199.19, i.e.
 an IP-address where pinging will start and an IP-address where pinging will end.
 Both addresses must be divided by a "-".
 In this example this would produce 199-100+1=100, 100*256=25.600,
 25.600+(19-10+1)=25.590 addresses
 
-an IP-address can be presented by its four decimal numbers, you can specify 
 ranges four each of these four numbers: 192-192.169-171.100-199.0-9
 In this example all IP addresses with first number 192, second number from
 168 to 168, third number from 100 up to 199 and last number from 0 up
 to 9 will be pinged. This would give 1*1*100*10=1.000 addresses. 
 This is probably only useful in very seldom cases.
 Here you have to provide ranges for every four numbers, always divided 
 by "-".
 
-single IP-addresses: 192.168.200.1
 well, single IP addresses or host names

It is also valid to leave this entry empty.


PingNames
 here you can additionally specify hosts to ping using their names.
 The names have to be divided by semicolons.

It is also valid to leave this entry empty.


AllowedAddresses

 This is also very important. LISa will only ping addresses,
 accept clients and answer broadcasts from addresses, which are covered by the
 addresses given in this line. You can add up to 32 network addresses/network masks
 or single addresses. Divide them by ; and don't put empty space between the
 addresses ! 
 Example: 192.168.0.0/255.255.0.0;192.169.0.0
 -> a complete network and a single address are valid
 Always make this as strict as possible, usually 
 your network address/subnetmask is a good choice.


BroadcastNetwork 

 This entry contains exactly one network address/subnet mask.
 To this network broadcasts will be sent. Usually this should be your
 own network address/subnetmask.
 Example: 192.168.0.0/255.255.0.0


SearchUsingNmblookup

 Here you can give 0 or 1.
 1 means that LISa will execute "nmblookup "*"  " and parse the output
 from this command. This produces less network traffic than the pinging,
 but you will only get hosts which have a smb-service running (Windows
 machines or machines running samba).
 If you enable this option and also give IP addresses to ping, then nmblookup 
 will be executed first and then the pinging will start.
 Then only addresses will be pinged, which were not already delivered
 from nmblookup. This should slightly decrease the network load.


FirstWait
 
 If LISa pings, i.e. if it sends the ICMP echo requests, it sends a bunch
 of requests at once, and the it will wait for the number of hundredth seconds
 you specify here. Usually values from 5 to 50 should be good, the maximum
 is 99 (gives 0.99 seconds, a very long time).
 Try to make this value as small as possible while still finding all 
 running hosts.


SecondWait

 After LISa sent the echo requests the first time, it can be possible
 that some hosts were not found. To improve the results, LISa can ping a
 second time. This time it will only ping hosts, from which it didn't receive
 answers. If you have good results with pinging only once, you can disable
 the second time with setting SecondWait to -1.
 Otherwise it might be a good idea to make this value a little bit bigger
 than the value for FirstWait, since the hosts which were not found
 on the first try, are probably slower or further away so they might take
 some milliseconds longer to answer.
 Usually values from 5 to 50 should be good or -1 to disable the second scan.
 The maximum is 99 (gives 0.99 seconds, a very long time). 
 

UpdatePeriod

 This is the interval after which LISa will update, i.e. ping or nmblookup
 or get the list of hosts from the LISa server which actually does the pinging.
 Valid values are between 30 seconds and 1800 seconds (half an hour).
 If you have a big network, don't make the interval to small (to keep
 network load low). Values from 300 to 900 seconds (5 to 15 minutes) might be
 a good idea. Keep in mind that the update period is doubled 
 if nobody accesses the server, up to 4 times, so the interval will become
 16 times the value given here and will be reseted to the value given here
 if somebody accesses the server.


DeliverUnnamedHosts

 If an answer to an echo request from an IP address was received, were LISa
 could not determine a name, it will be only delivered over the port
 if you set this to 1.
 I am not really sure if this is a useful feature, but maybe
 there are some infrastructure devices in your network without assigned names,
 so they don't have to be published. Set this to 0 if you want to keep them
 secret ;-)
 If unsure, say 0.

MaxPingsAtOnce
 
 When sending the pings (echo requests), LISa sends a bunch of these at once
 and then waits for the answers. By default there are 256 pings sent at once,
 usually you should not need the change this value. If you make it much bigger,
 the internal receive buffers for the answers to the echo requests may become to small,
 if you make it to small, the updating will be slower.


Three different example config files:

You are member of a small network with 24 bit network mask, i.e.
up to 256 hosts:

PingAddresses = 192.168.100.0/255.255.255.0
AllowedAddresses = 192.168.100.0/255.255.255.0
BroadcastNetwork = 192.168.100.0/255.255.255.0
SearchUsingNmblookup = 0                         #don't use nmblookup
FirstWait = 20                                   #20 hundredth seconds
SecondWait = 30                                  #30 hundredth seconds on the seconds try
UpdatePeriod = 300                               #update after 300 secs
DeliverUnnamedHosts = 0                          #don't publish hosts without name


You are only interested in hosts running smb services and you don't have
routers in your network:

AllowedAddresses = 192.168.100.0/255.255.255.0
BroadcastNetwork = 192.168.100.0/255.255.255.0
SearchUsingNmblookup = 1                #use nmblookup
UpdatePeriod = 300                      #update after 300 secs
DeliverUnnamedHosts = 0                 #don't publish hosts without name


The same network, but here both nmblookup and pinging is used.

PingAddresses = 192.168.100.0/255.255.255.0
PingNames = bb_mail
AllowedAddresses = 192.168.0.0/255.255.0.0
BroadcastNetwork = 192.168.100.0/255.255.255.0
SearchUsingNmblookup = 1                #also try nmblookup
FirstWait = 30                          #30 hundredth seconds
SecondWait = -1                         #only one try
#SecondWait = 60                         #try twice, and the second time wait 0.6 seconds
UpdatePeriod = 300                      #update after 300 secs
DeliverUnnamedHosts = 0                 #don't publish hosts without name
MaxPingsAtOnce = 256                    #send up to 256 ICMP echo requests at once

And now a configuration file for resLISa, PingAddresses is not used by resLISa,
neither is BroadcastNetwork.

PingNames = bb_mail;some_host;some_other_host
AllowedAddresses = 192.168.0.0/255.255.0.0
SearchUsingNmblookup = 1                # use nmblookup
FirstWait = 30                          #30 hundredth seconds
SecondWait = -1                         #only one try
#SecondWait = 60                         #try twice, and the second time wait 0.6 seconds
UpdatePeriod = 300                      #update after 300 secs
DeliverUnnamedHosts = 1                 #also publish hosts without name
MaxPingsAtOnce = 256                    #send up to 256 ICMP echo requests at once


+----------------------+
|     Installation     |
+----------------------+

Both reslisa and lisa open a so called raw socket to send and receive
ICMP echo requests (pings). To do this, they need root privileges.

lisa offers a service on TCP port 7741, it should be installed by root
and started when the system comes up, it depends on your distribution
how to do this.

reslisa is intended to be started per user, it doesn't offer anything to
the network. It needs to be installed setuid root.

If you use the rlan-ioslave from KDE2, reslisa can be started automatically
by reslisa.

lisa reads the file lisarc, reslisa reads the file reslisarc.
If you want to be able to configure both from the Trinity Control Center,
you have to start them using the command line switch -K.

For more information where they look for configuration files read
the next chapter.


+--------------------------------------------+
|     Command Line Options and other usage   |
+--------------------------------------------+

The following command line options are supported:

-v, --version      prints a short version info
-h, --help         gives an overview over teh command line options

Some options regarding search order for the configuration files.
For reslisa the file is named reslisarc instead lisarc.

-u, --unix         search at first for $HOME/.lisarc, then
                   for /etc/lisarc, this is the default behaviour
                   
-k, --kde1         search at first for $HOME/.trinity/share/config/lisarc,
                   then for $TDEDIR/share/config/lisarc

-K, --kde2         looks for the file lisarc in every directory
                   returned by running "tde-config --path config"
                   
-c, --config=FILE  read this and no other configuration file


This one is only available for LISa, not for resLISa.

-p, --port PORTNR  start the server om this portnumber
                   if you use this LISa won't be able to
                   cooperate with other LISa's in the network


If you send the Hangup-Signal to lisa or reslisa, it will reread its
configuration file (killall -HUP lisa).

If you send the User1-Signal to lisa or reslisa, it will print some status
information to the standard output (killall -USR1 lisa). You won't see 
anything if the console from which lisa/reslisa was started has terminated.


LISa and resLISa need a libstdc++ (it uses only the string-class from it),
it *doesn't* need neither Qt nor KDE.

So, that's it for now.
If you have suggestions, problems or whatever, contact me.


Have fun
Alexander Neundorf
<neundorf@kde.org>