From: m...@TC.Cornell.EDU (Matt Welsh)
Subject: [comp.os.linux] Linux Networking FAQ (Part 1)
Keywords: Linux TCP/IP Networking
Organization: Cornell Theory Center
Date: Thu, 18 Feb 1993 20:28:53 GMT
Approved: linux-annou...@tc.cornell.edu (Matt Welsh)
Last-modified: 17 Feb 1993
This is part 1/2 of the Linux Networking FAQ.
Note that this is still Phil's doc! I just helped with it. I will be uploading
this to tsx-11 and sunsite in a few minutes. -mdw
This is the LINUX NETWORKING FAQ
by Phil Copeland (p_cop...@csd.uwe.ac.uk).
Last revision: 17 Feb 1993
quick disclaimer: I must appologize for my luck of a spoll checkr
0. About this NET-FAQ
This is the Linux Networking FAQ, which covers all of the details
of setting up TCP/IP under Linux, either for a network or only
for loopback mode. It's maintained by Phil Copeland, but this revision
is by Matt Welsh (m...@tc.cornell.edu). New versions of this doc
will be posted to comp.os.linux.announce and can be found on the
major Linux FTP sites such as sunsite.unc.edu and tsx-11.mit.edu.
The version of the kernel used in this NET-FAQ is 0.99.5 and the
GCC compiler is 2.3.2. Thus, some of these things may or may not
work depending on your kernel and compiler setup.
My personal setup is a 486dx-25, 8Mb mem, 105 Mb Scsi Disk,
Adaptec 1542B Scsi Controller, Generic Scsi Tape (60Mb),
1200 Baud Hayes Modem (HP on com2), Inmos B004 transputer board (2Mb),
Western Digital 8013 16Bit network card, 2 serial ports (com1/4),
Single printer port and Paradise Pro Designer II SVGA card.
This mountain of equipment co-exists happily with each other and
works in harmony with each other. (I only include it here so that
people realize that such setups can exist)
0.1. A Request
If you find some text I've written which no longer applies or is a
complete load of rubarb, please tell me and include a reason or
corrective text (patch file/ context diff/ off the top of your head
formats are very welcome)
This NET-FAQ has grown quite large (~70k) and the past few versions
contained so much information and were downright confusing. So,
we revamped it and added a "Quick Start Guide", a quick overview of
setting up TCP/IP under Linux. It's really quite easy to get
everything going. There is a lot of reference information here so
don't be scared off.
Hello and welcome to the wonderful world of Linux networking.
Networking has always been one of the most exciting things that you can
coax a computer to take advantage of. It allows you to store/retrieve
files from remote machines (some of which are probably located in
countries which you'll never get to visit).
Networking also allows computers to interactively communicate with
other processes or users on these remote machines allowing a new
social aspect of computing to be approached (mainly in the form of
talk or MUD (Multi User Dungeon) sessions).
Networking also has many stumbling blocks for the administrator to
fall over, most notably the initial setting up of a system network
which can send the most sane person to eating the proverbial hat
through the hell of trying to coax their machines into networking life.
This FAQ is designed to help you start into networking in a positive
direction by leading you simply through the network configuration that
best suits you, whether you have a single machine with no network
attachment (silly I know) or a multi billion credit networking
computer for your country's local stock exchange. Please note that
this FAQ does not follow the 'normal' format of other FAQ's as it's
designed to teach you networking and its idiosyncacies.
As of 21 Jan 93 there is a Linux Networking Quickstart guide (in the
next section) by Matt Welsh to help review the process of getting it
1.1. Linux Networking Support
The Linux kernel is now distributed with the TCP/IP code in it.
Basically, Linux's network support is for either UNIX (local)
domain sockets or INET (TCP/IP) domain. This FAQ specifically
covers configuring TCP/IP for Linux. You can either configure it
in "loopback mode" (which allows you to telnet, ftp, etc. only
to your own machine) or, if you have an Ethernet card, for use
on a network such as the Internet.
1.2. Supported Ethernet Cards
To put your machine on a network you need an Ethernet connection
of some kind and an Ethernet card in your machine. Linux supports
a number of Ethernet cards, although only the WD8003 and WD8013 (aka
SMC Elite) cards come with the standard kernel. Donald Becker wrote
up the following information regarding Ethernet cards, prices, etc.
The least expensive 8 bit ethercards start at $70 and are usually
NE1000 clones. It's definitely worth it to pay an extra $10 and go
for a 16 bit bus interface NE2000 clone. For another $10-$15 you can
get a shared memory 8013 clone, which will give you somewhat higher
You should expect to pay more than the price I listed above unless you
do careful shopping from the back of Computer Shopper and (even
better) LAN magazine. I've gotten network things from both MCW
Distributors in Gaitherburg MD (good prices, sort of local, advertises
in Comp.Shop.) and Network Express (a little more expensive).
You'll also have to decide the kind of interface you need. "Thinnet"
is RG58A 50ohm cable with BNC connectors. 10BaseT is twisted pair
("TP") to a central "hub". There is also traditional thick 50ohm
cable, but it has no advantage in most installations. An "AUI" port
is a 15 pin D-shell connector that can be hooked to an external
transceiver (ca. $50 for 10BaseT or thinnet), usually for thicknet
(in which case it's $100+). Cards typically have an AUI connector and
either a thinnet or twisted pair transceiver. You'll pay about $20
more and give up the AUI to get both thinnet and 10BaseT.
Some ethercards advertise status LED. These are most useful for
10BaseT connections, which are easy to mix up.
IMHO, thinnet with on-card transceivers results in a _much_ cheaper
system. You only need to buy T connectors($3ea.), cables ($6/12ft at
RS), and two terminators ($2ea.), leading to a per-node cost of under
$100. At these price levels it's definitely cheap enough to put on a
home system! With twisted pair you'll need a hub which can easily
double your per-node code. TP is only cost-effective if the wiring is
already there and its expensive to run more.
These drivers support all common 8390-based ethernet boards. Currently
"common" is defined as:
2. What you need to get started
To configure TCP/IP under Linux you need:
1) A linux machine with linux kernel 0.98.5 although I'd
recommend going all the way to 0.99.5 as many tcp/ip errors
have been stomped out (although not all).
2) Version 4.2 of the jump table library image (/lib/libc.so.4.2).
This is needed for the various network binaries and so on.
The most recent version is on sunsite.unc.edu:/pub/Linux/GCC.
2) If you're going to use TCP/IP over the network (i.e. not just
loopback mode), then you need one of the following Ethernet
SMC Elite 16
Alta Combo (ne2000 clone)
Aritsoft LANtastic AE-2 (ne2000 clone w/ extra memory)
D-Link Ethernet II
3Com 3c503 EtherlinkII
Cabletron E1010, E1010-x, E2010, E2010-x
various HP 8390-based boards such as the HP27245, HP27247A,
The wd8003, wd8013, and SMC Elite 16 are all included in the
standard Linux kernel. The ne2000, ne1000, 3c503, Cabletron,
HP, and and other 8390 card drivers are available for beta
testing. This will be covered later.
3) If you are only going to use 'loopback' mode, you won't need
a card! A special loopback device is used to communicate
*** NOTE when talking of ethernet devices, it should
be noted that /dev/eth0 does NOT exist, the kernel
knows about it and thats all you need to know, /dev/eth0
and /dev/loopback are fictionous (FS speaking)
4) The tcpip-0.8 networking package. This is the old, original
release of the TCP/IP software. The only things you need
from this package are the 'config' program and the network
installation scripts (such as rc.net, install.net, and so on).
Everything else in the tcpip-0.8 package (the kernel code,
diffs, binaries, etc.) is obsolete.
You also need the tcpip-0.8-fixes package. You need more or
less everything from this package: the exact files you need
are covered later.
NOTE: If you have SLS you should have everything you need in
It's available from all of the major Linux FTP sites, in the
file tcpip-0.8.tar.Z. The fixes are in tcpip-0.8-fixes.tar.Z.
They should both be in the same place.
5) The net-bin-0.2 package. It's on sunsite and tsx-11 in the
This file contains all of the TCP/IP clients and daemons that
you'll need, including: telnet, telnetd, ftp, ftpd, inetd,
named, rcp, rlogin, rsh, talk, ping, nslookup, and more.
6) You don't need the net-lib-1.1 package. The libraries have
now been added to the most recent libc.so.4.2, so if you have
that you're set.
7) If you want NFS support, Linux 0.99 now contains NFS as a
of mount which lets you NFS mount a filesystem (i.e. mount a
filesystem on another machine). Look on nic.funet.fi in
8) Know the IRQ's of your internal cards. This is to avoid
conflicts and allow the 'drivers' to communicate with your
9) Also, If you do have ethernet cable, both coax (thin and thick)
as well as twisted pair will work, the cable is only there to
carry signals, your ethernet board works out how and the linux
'drivers' simply stuff data onto the card.
10) A lot of coffee and one of those stress relieving
gadgets you can get in the local market. [Ed. note: I had
about 3 Dr. Peppers and I was okay. -mdw]
3. Quick Start Guide to setting up Linux TCP/IP
This is a rundown of what you need to do to setup TCP/IP. Read it
through and then keep it all in mind as you're cleaning up all of
the details below. It's not difficult if you do everything correctly.
It's not as quick as I wanted it to be. Basically I get all of the
installation stuff straight and then let Phil explain the details
of setting up named, etc. later in the NET-FAQ. This section was
written by Matt Welsh.
- NOTE: In this discussion, the directory /usr/etc/inet is used
to hold the tcp/ip daemons, configuration files, and so on.
You can use ANY directory you want, as long as you're consistent.
Two popular alternatives are /etc/inet or just /etc. I like to
keep all of my tcp/ip stuff in /usr/etc/inet just to keep it
seperate from my other /etc files (because I toy with it a lot).
This is mostly personal taste.
TCP/IP clients (such as telnet, ftp, and so on) can go anywhere
on your user's path. The canonical place is /usr/bin. It doesn't
really matter; here I install clients in /usr/bin.
- (Another) NOTE: Some programs, like fingerd, expect certain files
to be in certain places. For example, fingerd won't work if
finger is not in /usr/bin. The easiest solution is to make a
symbolic link if you put your clients, etc. elsewhere. If
something doesn't seem to be working, make sure everything's
in the right place and has correct permissions. One way to
find out where a program expects companion programs or files
to be is to use 'strings'. For example,
strings fingerd | more
will show you all of the printable strings in the fingerd
binary; you can use this information to find out where fingerd
expects finger to be, and so on.
- First things first: Get all of the files, etc. listed above in
section 2.0. When unpacking the tcpip-0.8, tcpip-0.8-fixes, and
net-bin packages, it's helpful to unpack them in separate directories,
because we'll be moving the files around to the right places. For
example, unpack tcpip-0.8.tar.Z in /usr/src/tcpip-0.8 and
net-bin-0.2.tar.Z in /usr/src/net-bin (or something like that).
NOTE: The current version of SLS (0.99.2 and up) already have
pretty much everything you need to get networking going. The
configuration files all live in /etc/inet, with /usr/etc/inet
being a logical link to this location. So if you have SLS you
probably don't need to get all of these files.
- Most of the files in tcpip-0.8 you don't need. After you've unpacked
it somewhere, take inet.tar and unpack it in /usr/etc/inet (which you
may need to create). You can delete the following files in
(Don't worry; later we replace them with newer versions).
- The rest of the files from tcpip-0.8.tar.Z you can delete.
- Unpack tcpip-0.8-fixes.tar.Z in /usr/etc/inet. You can delete
the file 'config' from it.
- Take the config.c (from tcpip-0.8-fixes) and compile it in
/usr/etc/inet with the command
gcc -o config config.c
NOTE: If you do not recompile config, you will probably get an
ioctl error when you reboot with networking installed. To avoid
problems, you should recompile the program with the above command.
- Having unpacked net-bin-0.2.tar.Z in /usr/src somewhere, you
can install these binaries. The following files are copied to
ping (must be setuid root; i.e. do 'chmod 4755 /usr/bin/ping')
rsh (must be setuid root)
rcp (must be setuid root)
rlogin (must be setuid root)
The following files are copied to /usr/etc/inet:
The man pages are copied to /usr/man... for example, all *.1 are
copied to /usr/man/man1 and *.8 are copied to /usr/man/man8.
- Now you've got all the software installed, you need to recompile
your kernel with TCP/IP enabled. This is easy unless you have an
old kernel (pre-0.99) or need to install the ne2000/3c503/ne1000
drivers. Here's how.
IF you're installing the 8390/n2000/3c503/ne1000 drivers (from
super.org, directory /pub/linux/newether), follow the directions
below for installing the driver. If you're NOT installing the
8390 driver (or only want to use loopback), just skip down to
compiling the kernel.
Get the files that you need. See the README's there for full details.
Basically you need:
one or more of ne.c, wd.c, 3c503.c/3c503reg.h, and so on,
depending on the card you have.
Note that if you have 0.99.pl5 or above you need to get the
8390.c from /pub/linux/ether-995 instead (as a lot of
kernel TCP/IP code changed/got better with 0.99.pl5).
Just follow the directions found in the file INSTALL on super.org.
It's easy. Just:
- Put the files above in /usr/src/linux/net/tcp.
- Edit the GNUmakefile to define which card you have, your
base address, and your IRQ. Note that with these new
drivers if EI8390 (the base address) and EI8390_IRQ (the
IRQ) are defined to be 0, they will be automatically
detected at bootup time.
- Edit Space.c (if needed),
- If you changed the GNUmakefile to use "eth_if" instead of
"eth0" (note that the newest 8390 drivers use "eth0" like
everyone else, they previously used "eth_if"), then you need to
edit /usr/etc/inet/rc.net to run $CONFIG on "eth_if" instead of
"eth0". If not you'll get an ioctl error from config.
If you have problems with the 8390 driver, contact bec...@super.org.
- If you're NOT installing the 8390 driver (i.e. just using the wd8003
driver with the standard kernel), then you need to edit
/usr/src/linux/net/tcp/Space.c to reflect your card's IRQ, base
address, and so on. If you're only using loopback you can skip
this step, too.
Anyway for those who are flexible, the standard kernel parameters
for this are :
IRQ: 5 (card interrupt)
mem: D0000 (where in memory to buffer data)
i/o addr: 280 (low level address of card)
mem start: D0000 (nearly all boards have a jumper to
mem end: D2000 (for wd8013, make this D4000)
NOTE: If you have problems with the memory start addr for the
WD80[0/1]3, please get in touch with b...@leland.stanford.edu.
- Now you're all set to compile the kernel. I really suggest that
you use version 0.99.pl4 or newer (probably 1.0 by the time this
is out). If you don't have at least 0.99 you can't run 'make config'
to autoconfigure the kernel and you'll have to do some stuff by
In any case, it's easy. If you have 0.99 or newer, just cd to
/usr/src/linux and do a 'make config'. Make sure you answer 'yes'
to the question on configuring TCP/IP. The rest of the options are
up to you. Also make sure you edit /usr/src/linux/Makefile to fix
your root device, keyboard, and so on.
Then do a 'make dep' to fix your dependencies--- THIS
STEP IS VERY IMPORTANT. Then (if you've already compiled this
version of the kernel) do a 'make clean'. FINALLY you're ready to
just do 'make' to compile the kernel.
When you're done you'll have the new kernel in /usr/src/linux/Image.
Copy it to a floppy or install it in /etc for use with LILO, or
whatever. Reboot with your new kernel.
- Once you're rebooted you can configure the stuff in /usr/etc/inet.
Run the script 'install.net' there, and answer the questions to
set your IP address, net address, router, domain name, and
nameserver. This is covered later in the NET-FAQ.
NOTE: If you have SLS then the "install.net" file isn't used. Instead
you need to edit hosts, resolv.conf, rc.net, and so on by hand to
set up the various addresses. It's very straightforward; just make
sure that the various configuration files (discussed below) in
/etc/inet have the correct information.
NOTE 2: If you're only using loopback, then your IP address is
"127.0.0.1", and you don't have a router, network address, or net
mask (these are things prompted for by install.net). For SLS,
which doesn't have install.net, you just edit the config files
in /etc/inet to reflect this.
- I had to edit resolv.conf there to make sure that the hostname and
domain names were right. No big deal. Under SLS you need to set
your hostname in the file /etc/inet/host (not 'hosts') and set
the domain name in /etc/inet/domain in addition to this step.
- Set up your named configuration files. Named is the service that
allows your machine to act as a nameserver. If you have a real
nameserver already, you probably don't want to run named (wastes
memory). If you're on loopback, you don't need it either (just put
all of your hostnames and ip addresses in /usr/etc/inet/hosts).
Named is nice if you have a LAN setup and want your Linux box to be
the name server. This is covered in detail later in the NET-FAQ as
In general you don't need to run named unless you really like
hacking with DNS. I don't see any need for it, since you can put
all of your hostnames in /usr/etc/inet/hosts and/or consult
- Create the file /usr/etc/inet/host.conf. This file tells the
name-binding libraries how to look up names: in this case, we're
going to tell the libraries to check first /usr/etc/inet/hosts
and THEN ask the nameserver (if any). So, create
/usr/etc/inet/host.conf. It should contain only these 2 lines:
This is VERY IMPORTANT. If you don't create this file then you
probably won't be able to look up names as expected.
- Set up inetd.conf to include lines for all of the tcp/ip daemons
(such as telnetd, fingerd, etc.) that you have in /usr/etc/inet.
This is covered later.
- Make sure that /usr/etc/rc.net is run from your /etc/rc.local.
- Edit rc.net to make sure it's getting your IP address right. As
it stands now it tries to grep for it in /usr/etc/inet/hosts,
and this doesn't always work. I just hardcode my IP address in
rc.net since my IP address isn't going to change much. :)
SLS also tries to look up your net and router address from
/etc/inet/hosts. I just hardcode these in as well as I don't
FOR LOOPBACK ONLY: If you're only using loopback, then edit
rc.net to make your IP address 127.0.0.1, and you can ignore
the netowkr and router addresses. In rc.net, you should only be
running the config commands for "loopback", and no others, so
comment out the lines which run config on "eth0".
If you're using the 8390 driver (see above) make sure you've
changed 'eth0' to 'eth_if' on the config commands in rc.net.
- If you're not running named, you can comment out the lines which
start it in rc.net. This will save memory and CPU time.
- If you're not going to run NFS, you can comment out the lines in
rc.net which run nfsd, mountd, portmapper, and routed.
- If you want to use NFS (network file system), you're on your
own. It should suffice to say that you need the nfs-client
stuff from tsx-11 and nfs enabled in your kernel. Should be easy,
I haven't played with it yet.
- If you didn't already, read all of the README files that come
with net-bin-0.2 and all that. They contain more up-to-date
info. NOTE that the info in tcpip-0.8's README file is mostly
out-of-date, follow the directions above and you'll be okay.
- At this point you should be able to reboot your system, rc.net will
run, and you'll see something like
loomer -> 220.127.116.11
which is output from rc.net. If you don't see this (or if there are
errors) then there's a problem; the best way to fix this is to
edit rc.net and the other files in /usr/etc/inet and make sure you
have your IP addresses and everything set right.
Okay, that's about it for this so-called "Quick Start" guide. the
rest of the NET-FAQ will fill in the gaps and talk more about
networking than how to install the softs and configure the kernel.
4. Running install.net
As mentioned above, to set the various network numbers, etc.
for your system you need to run the install.net script, which sets
lots of things in /usr/etc/inetd (mostly in hosts, resolv.conf, and
NOTE: If you're running SLS you don't have the install.net script.
Just edit the files discussed in sections 5 and 6 of this net-faq by
hand, it's not very difficult. All install.net does is put default
values in these files for you.
NOTE: If you're only on loopback, the only IP address you should
be using is '127.0.0.1' which stands for loopback. You will
be your own nameserver (either running named or just using
/usr/etc/inet/hosts), and you don't need to worry about the router
and subnetwork addresses.
When running install.net you'll have to answer these questions:
Enter IP Address for (your host) (aaa.bbb.ccc.ddd)
Here you are being asked what network address you would like to be known
as. Ip address are unique numbers so as to identify your machine from
another on a multiuser network. Normally if you reside in the Internet
you will have a network address assigned by the NIC or your local
network controller and you really must stick to it since there is no
room for you to bugger up the network by using someone elses ip
address. If you do not have a connection to the Internet, you will
have less of a problem although it would still be a good idea to apply
for a internet class c/d network number depending on your setup.
There is a convention being used that allows people who are completely
bemused by all the ip registration stuff that allocates a band of ip
numbers (192.0.2.xxx) which are encouraged to be safely ignored by the
rest of the internet. So if you don't know what ip you'll be assigned
or (naughty) can't be bothered, please use that range to avoid
bringing sections of the internet around your ears.
IP numbers are typically of the 0-255.0-255.0-255.0-255 range so
valid answers are 243.123.4.23 or 18.104.22.168, etc. 324.234.545.2
is completely wrong.
Enter Net Address for (your hostname) (aaa.bbb.ccc.0)
Here you are being asked for your subnetwork address. This requires a
bit of explaination. Subnets are a "unit" of connectivity which depict
how many possible hosts 'live' on the same piece of cable as you do
(typically this never exceeds 253 on one piece on cable) a quick way
of getting the question right is to type in whatever you have for your
ip address but make the last number 0 eg if my ip address were
22.214.171.124, my 'safe' Net address would be 126.96.36.199. 0.0.0.0
means the whole world and is probably what slip people should use.
Enter Router Address for (your hostname) (aaa.bbb.ccc.ddd)
Wibble! Ok here what is being asked is if you have a gateway machine
through which IP traffic can be passed to the great blue yonder. We
are sneekily getting the routeing machine to do some hard work for
us. Routers tend to have 2 ethernet boards in them with differing
network numbers for them so that they can 'bridge' between different
numbered networks, eg you could not talk directly to a ip address of
188.8.131.52 from an ip address of 184.108.40.206 but a machine in the
middle with two ip address 220.127.116.11 and 18.104.22.168 can 'collect'
the data from the 192.35.173.xxx network and transfer it to the
192.35.175.xxx network. All we have to do here is stick in the ip
address of the local router. You need to find this out from your local
network admin types. If you don't have a router use 0.0.0.0 meaning
don't route anything.
Enter Domain name for (your host)
This isn't too bad, domain names are 'convenient' labels eg uwe.ac.uk is
the domain name that appends to all the machines on site so that a sun
called csd would be known as csd.uwe.ac.uk This is so that you don't
have to know the full ip number of the host, it's more convenient to
call out a semi inteligable name eg 22.214.171.124 = csd.uwe.ac.uk but
the 192.35.175 is aliased to uwe.ac.uk (University in the West of
England, academic community, United Kingdom). Again this should be
given to you with a registered ip address but for now you could put
in 'at.linux.net' it can be changed later.
mdw: In short the domain name is the name of your ENTIRE domain.
For instance, my machine is loomer.ithaca.ny.us. The full hostname
of the machine is 'loomer.ithaca.ny.us', and the DOMAIN name is just
'ithaca.ny.us'. Here you're being asked for the DOMAIN name only.
Name Server for Domain (aaa.bbb.ccc.ddd)
If you're on a University or business network, you'll probably have
a nameserver. A name server just looks up machine names for you.
For example, if you want to telnet to 'shoop.vpizza.com', you don't
have to tell your machine what shoop.vpizza.com's IP address is; your
machine can ask the nameserver instead.
Ask your local network people what the nameserver for your network is.
Here you're being asked for the IP address (number) of the machine,
not the name. If you don't have a nameserver, then just put in your
own IP address, and you can either run named or go without a nameserver
(putting all of your names/IP addrs in /usr/etc/inet/hosts).
5. Other /usr/etc/inet configuration files
Ok time for a quick check of what you minimally *SHOULD* have in
config - This sets up the ethernet ip tables.
inetd - Daemon process that invokes other network daemons
inetd.conf - Configuration file for inetd about the other daemons
install.net - The semi automatic script I just talked about
named-xfer - Used for updating the nameserver records
named.reload - used to load in the named
named.restart - user to stop and restart the named process
rc.net - a network rc file called from /etc/rc.local
services - a file specifying what 'port' numbers certain
services are available on
telnetd - daemon for accpting incoming telnet requests
named - the nameservice daemon
Other daemons, such as fingerd, tftpd, and so on.
Time for some explainations I think...
'config' is a general do it all 'fix your ethernet board to your
local setup' command. It was configured when you ran the install.net
script and if you look at the rc.net file you'll see where it plugged
in all the IP stuff that you fed the script with... a bit technical
but otherwise nothing to worry too much about provided that your
original information was correct. One thing though, I have found that
it is best to edit the rc.net file and 'hard wire' the ip addresses
directly in rather than relying on the grep search from /etc/hosts but
you may disagree (personal preferance).
You'll have to create this file yourself if you don't have it.
With the new net-libs being made available by Mitch, you will find
that it is possible to set up how ip addresses are looked up using the
file /usr/etc/inet/host.conf with the entries:
which tells it in what order it should attempt to resolve an IP/domain
name. In this case, when trying to match hostnames & ip addresses,
the name binding libraries will search /etc/hosts and if no match is
found then query the nameserver).
If you run named then this is moot; you're your own nameserver. See
below about named.
'inetd' is a daemon process that wait's for certain events to happen
upon which it will select which process to run eg if no network
communication is happening, only inetd will be running but if a telnet
session is requested by a remote machine, inetd will start running
telnetd for that incoming call to connect to.
Of much more interest is 'inetd.conf' which has information about what
services to run and where to find them. Here's an example:
# Serv type packet wait/nowait run as program to run invoke as
telnet stream tcp nowait root /usr/etc/inet/telnetd telnetd
talk dgram udp wait root /usr/etc/inet/ntalkd talkd
echo dgram tcp nowait root internal
ftp stream tcp nowait root /usr/etc/inet/ftpd ftpd -l
The net-bin-0.2 README file has a list of entries which you may add
to inetd.conf. NOTE that inetd.conf cannot have any blank lines in it.
This is a bug which will be fixed soon. Also, don't start services
you don't need or don't understand, like tftpd. They will only waste
resources and may have security implications.
Now another file that comes to mind at this stage is /etc/protocols or
rather /usr/etc/inet/protocols (I've made the symlink
/etc/protocols -> /usr/etc/inet/protocols)
This file contain's information on what protocol is to be used
when the datagram packet arrives ie how it is to be treated.
Here's an example /usr/etc/inet/protocols file:
# protocols - standard well defined IP protocols
ip 0 # internet protcol, pseudo protocol number
icmp 1 # internet control message protocol
igmp 2 # internet group multicast protocol
ggp 3 # gateway -> gateway protocol
tcp 6 # transmission control protocol
pup 12 # PARC universal packet protocol
udp 17 # user datagram protocol
raw 255 # raw
There are others but these are normally never needed.
(NOTE: the /etc/protocols from the tcpip-0.8 distribution defines ggp
to be 2 which isn't the case)
If this file is missing or empty, you will never get any transports
(ftp/telnet) to work and will be told that there isn't any such
'services' is a file which informs the tcp/ip code what port number a
particular program will run on for example if you telnetted to port 7
on a sun you would be connected to an echo service which would send
back a carbon copy of what you typed in but that service has a
specially allocated port number referenced in the /etc/services file
of both machines.
There is a complete standardized services file in circulation from Ross
Biro; it is included in the tcpip-0.8-fixes.tar.Z package.
Ross: This is the one I made from the relevant rfc. It has some
typos and such here, but it is probably ok for most use.
Here's a *small* excerpt (not the entire file):
tcpmux 1/tcp # TCP Port Service Multiplexer
discard 9/tcp sink null
discard 9/udp sink null
systat 11/udp users
systat 11/tcp users
smtp 25/tcp mail #Simple Mail Transfer
time 37/udp timserver
time 37/tcp timerserver # time
name 42/tcp nameserver
name 42/udp nameserver
whois 43/udp nicname
whois 43/tcp nicname
nameserver 53/tcp domain
nameserver 53/udp domain
The other files in /usr/etc/inet are described in the named section
---- end of part 1/2
Matt Welsh, m...@tc.cornell.edu
"What are you doing, Dave?"
From: m...@TC.Cornell.EDU (Matt Welsh)
Subject: [comp.os.linux] Linux Networking FAQ (Part 2)
Keywords: Linux TCP/IP Networking
Organization: Cornell Theory Center
Date: Thu, 18 Feb 1993 20:33:34 GMT
Approved: linux-annou...@tc.cornell.edu (Matt Welsh)
Last-modified: 17 Feb 1993
This is part 2/2 of the Linux Networking FAQ.
6. Names and name servers, what /etc/hosts is all about.
The internet protocol document defines names, addresses and routes
A name indicates what we seek.
An address indicates where it is.
A route indicates how to get there.
Every network interface attached to a tcp/ip network is identified by a
unique 32-bit IP address. A name (hostname) can be assigned to any
device that has an IP address. Names are assigned to devices because,
compared to numeric Internet addresses, names are easier to remember
and type correctly. In use, most of the tcp/ip software on linux can
interchangeably use name or ip address but whichever is chosen, it is
always the IP address that is used to make connections. Translating
names into addressses isn't simply a "local" issue. The command
'telnet fred.at.linux.net' is expected to work correctly on every host
that is connected to the network. If the machine is connected to the
Internet, hosts all over the world should be able to translate the
name into a valid IP address, therefore, some facility must exist on
the net to translate the name into the numeric IP address.
There are two methods for doing this: one involves using a local lookup
table ('/usr/etc/inet/hosts') and the other uses DNS (Domain Name
System) to remotely interrogate the network for the IP address.
'/usr/etc/inet/hosts' (or /etc/hosts) is a very simple file which
contains a numeric IP address followed by one or more hostname
# /usr/etc/hosts example
# note that the hash is a comment, no text is processed after
# it until the next <cr>
126.96.36.199 csd csdsun csd.uwe.ac.uk csdsun.ac.uk
188.8.131.52 manic manic.uwe.ac.uk # Tom's machine
184.108.40.206 chef chef.uwe.ac.uk # Main waste of money
# other nets
220.127.116.11 hal hal-9000 # local hidden host
18.104.22.168 slave slave.uwe.ac.uk # linux engine 485 25
22.214.171.124 zen zen.uwe.ac.uk # Interactive 2.2.1 386 33
# external nets
Clearly this has a limitation in that on large networks ALL machines
would have to have this information on disk and that could have 1000's
of entries. Just think what that means if an extra 120 machines were
added! 1000's of machines would have to have their /etc/hosts table
updated either by hand or automatic shell scripts calling the list
from a main machine... (see where this is leading?) Enter the DNS
The SLS /etc/inet/hosts file is more involved, it specifies the
router, network, and IP addresses. It's pretty self-explanatory
but you should edit this file (because most everything's set here,
as install.net isn't used with SLS).
6.2 DNS: Domain Name Service
DNS scales well. It doesn't rely on a single large table; it is a
distributed database system that doesn't bog down as the database
grows. DNS currently provides information on approximately 700,000
hosts. DNS also guarentees thst the new host information will be
disseminated to the rest of the network as it is needed.
6.2.1 named: running DNS from your own machine
If you don't have a nameserver (which services DNS requests for you),
and aren't just using loopback, then you can run named on your own
machine. Named will allow you to setup a subsection of the DNS database
for use on your own machine and local network. There are a number of
files to be edited.
/usr/etc/inet/a_hosts_table (can be called anything, usu.
If you have a nameserver, the only file you need is resolv.conf,
where you define your domain name and nameserver's IP address.
These are both set up by install.net. For example:
However, if you're going to run named you need to define the
'nameserver' in resolv.conf to be YOUR OWN IP address. And you need
to provide information in named.boot and a_hosts_table....
primary uwe.ac.uk /usr/etc/inet/a_hosts_table
@ IN SOA slave.uwe.ac.uk. root (
3600 ;refresh every 10 hours
300 ;retry every 6 minutes
36000000;expire after 1000 hours
3600 ) ; default ttl is 100 hours
IN NS slave.uwe.ac.uk.
slave IN A 126.96.36.199
hal IN A 188.8.131.52
zen IN A 184.108.40.206
mother IN A 220.127.116.11
If you're going to run named, resolv.conf, named.boot, and
a_hosts_table will suffice, BUT there are more (for other fun named
6.2.2 More complete list of named setup files
YOU DON'T NEED to run named if you're only using loopback OR if
you have a nameserver. It's a waste of CPU time and memory. But if
you don't have a nameserver or if you just feel like hacking it,
here's a more complete named setup:
resolv.conf: If this file exists, it is read each time a process
using the resolver starts. As a result, the file is
not normally created unless necessary and isn't used
if named is running. You should have it anyway in case
named dies. :)
named.boot: Sets general named parameters and points to the
sources of the domain database information used
by this server. The sources can be local disks or
named.ca: Points to the root domain servers
named.local: Used to locally resolve the loopback address
named.hosts: The zone info file that maps host names to IP addresses
named.rev: the zone file for the reverse domain that maps IP
addresses to host names (you'll prob never touch it
so i'm going to skip it's description unless people
get upset enough to lynch me)
*** STOP PRESS ***
I've just found out from Ross by sheer accident that there is a
program released in comp.sources.unix (volume25) called hostcvt (mutter
mutter) which is supposidly capable of converting /etc/host entries
into the nesessary corrisponding named files.
This program is now available on sunsite.unc.edu for Linux, in
/pub/Linux/system/network. It's also distributed on SLS.
*** RESUME PRESS ***
6.2.3 Where DNS gets its information
The 'named.boot' file points to sources of DNS information.
Some of these sources are local files; others are remote servers. You
only need to create the files referenced in the primary and the cache
DNS commands | functions
directory | Defines a directory for all subsequent file referances
primary | Declares this server as primary for the specified zone
secondary | Declares this server as secondary for the specified zone
cache | Points to the cache file
forwarders | Lists servers to which queries are forwarded
slave | Forces the server to only use the Forwarders
Here are some example setups of the named files.
As mentioned before, if you are going to be using named, this file is
usually disguarded. Otherwise it points to a server that the resolver
is to query for domain information. If no nameserver entries are
contained in the file, the local host is queried for the information.
; cache only server
primary 0.0.127.IN-ADDR.ARPA /usr/etc/inet/named.local
cache . /usr/etc/inet/named.ca
The loopback domain is an in-addr.arpa domain that maps the address
127.0.0.1 to the name localhost. The idea of resolving your own loopback
address makes sense to most people, so most named.boot files contain
; Primary name server boot
primary big.cty.com named.hosts
primary 54.152.IN-ADDR.ARPA named.rev
primary 0.0.127.IN-ADDR-ARPA named.local
cache . named.ca
The directory statement tells named that all subsequent filenames are
relative to the /usr/etc/inet directory. The first primary statement
declares that this is the primary server for the big.cty.com domain and
that the data for that domain is loaded from the file named.hosts. The
second primary statement points to the file that maps IP addresses from
152.54.xxx.xxx to hostnames. This statement says that the local server
is the primary server for the reverse domain 54.152.in-addr.arpa and
that the data for the domain can be loaded from the file named.rev.
6.2.7 DNS Resource Records (RR's)
Resource Records are used in the named files to set attributes of
addresses, networks, and so on. Here's a list of the RR types:
Resource Record Record type function
Start of authority SOA Mark the beginning of a zone's data,
and define parameters that affect the
Name server NS Identifies a domain's name server
Address A Converts a host name to an address
Pointer PT Converts an address to a hostname
Mail Exchange MX Identifies where to deliver mail for a
given domain name
Canonical name CNAME Defines an alias host name
Host information HINFO describes a hosts hardware and OS
Well Known Service WKS Advertises network services
These resourse records are defined in RFC 1033.
The format of DNS resourse records is:
[name] [ttl] IN type data
name: This is the name of the domain object the resource record
references. It can be an individual host or an entire domain.
ttl: time-to-live defines the length of time in seconds that the
information in this resource record should be kept in the
cache. Usually this field is left blank and the default ttl
set in the SOA is used.
IN: Identifies the record as an internet DNS resource record. There
are other classes of records, but they are not used by the DNS
type: Identifies what kind of resourse record this is
data: the information specific to this type of resourse record
6.2.8 The cache Initialization file
The basic 'named.ca' file contains "NS" records that name the root
servers and "A" records tha provide the addresses of the root servers.
A basic 'named.ca' is shown here:
; named.ca - typical setup
; Servers for the root domain
99999999 IN NS tsx-11.mit.edu.
99999999 IN NS nic.funet.fi.
; Root servers by addresses
tsx-11.mit.edu. 99999999 IN A 18.104.22.168
nic.funet.fi. 99999999 IN A 22.214.171.124
Note that the ttl is 99999999 the largest possible size so that the
root servers are never removed from the cache.
6.2.9 The 'named.local' file
The 'named.local' file is used to convert the address 127.0.0.1 (the
loopback address) into the name localhost. It's the zone file for the
reverse domain 0.0.127.in-addr.arpa. Because ALL systems use 127.0.0.1
as the loopback address, this file is virtually identical on every
@ IN SOA slave.uwe.ac.uk. root. (
1 ; serial number
36000 ; refresh every 10 hrs
3600 ; retry after 1 hr
3600000 ; expire after 1000 hrs
36000 ; default ttl is 10 hrs
IN NS slave.uwe.ac.uk.
1 IN PTR localhost.
6.2.10 The 'named.hosts' file
The 'named.hosts' file contains most of the domain information. This
file converts host names to IP addresses, so "A" records predominate,
but it also contains "MX", "CNAME" and other records.
; named.hosts file example
@ IN SOA slave.uwe.ac.uk. probs. (
1 ; serial
36000 ; refresh every X seconds
3600 ; retry every X seconds
3600000 ; expire after X seconds
36000 ; default time to live X seconds
; define nameservers and mailservers
IN NS slave.uwe.ac.uk.
IN MX csd.uwe.ac.uk.
; define localhost
localhost IN A 127.0.0.1
;hosts in this zone
loghost IN A 126.96.36.199
hal IN A 188.8.131.52
zen IN A 184.108.40.206
thing IN A 220.127.116.11
slave IN A 18.104.22.168
IN MX 2 22.214.171.124
servant IN CNAME slave.uwe.ac.uk.
mother IN A 126.96.36.199
; outside domains now follow
csd IN A 188.8.131.52
IN MX 5 184.108.40.206
csdsun IN CNAME csd.uwe.ac.uk.
chef IN A 220.127.116.11
;fictional outside gateway
midway IN A 18.104.22.168
; etc until you have built a reasonable host table
; that you feel will be adaquate for your network
7. NFS: The Network File System
Network filing systems are convenient mechinisms which allow your
machine axcess to more disk space that it actually has by 'borrowing'
disk space from another networked machine for either sharing of common
data or if allowed, the storing of data generated by your machine.
NFS has several benefits:
1) it reduces local disk storage requirements because
a network can store a single copy of a directory, while
the directory continues to be fully accessible to everyone
on the network.
2) NFS simplifies central support tasks, because files can be
updated centrally, yet be available throughout the network.
3) NFS allows users to use familiar UNiX commands to manipulate
files with rather than learning new ones. There is no need
to use rcp/tftp/ftp to copy files, just 'cp' will do.
As of 0.99.2 support has been added into the kernel for running
binaries on both the MSDOS and NFS filesystems (of course the
binaries have to be Linux type binaries to run on your system).
Linus warns that they'll be slower to load and won't be memory
effecient; there are hopes that this will change soon.
Linux now has the following filesystems available for it: minix,
extfs, msdos, proc, isofs, nfs with a view to a compressed
filesystem being worked on (zfs?) all are perfectly transparent to
each other although filename tructation may occur. The reason that I
mention this is that NFS will allow you filename lengths supported by
the type of filesystem you mount eg the HP9000 here supports 15 char
filenames on an NFS mount as does it's MAG-OPT drive whereas the
sun4's offer 255 char filename on their NFS exports.
7.1 The '/etc/exports' file
If you want your machine to be an NFS server for other systems,
you must run nfsd, mountd and edit /etc/exports.
'/etc/exports' allows your machine to decide what local filesystems it
will allow remote clients to NFS mount and decide what access those
clients should have to your filespace.
Example (I just love examples):
/ slave(root_quash) moonbeam(root_quash)
/home slave csdsun
flag | function
ro | read only, this is the default
rw | read and write, used to allow a client to write to that FS
There are other options but these are covered in the README for the
NFS kit and the above are the simplest to get to grips with.
7.2 The /usr/etc/inet/rc.net file
The file 'rc.net' is used to start the named services and nfs the
suggested setup is as follows:
if [ -f /etc/portmap ]
echo "Starting portmapper..."
if [ -f /etc/exports ]
echo "Starting nfsd..."
echo "Starting mountd...."
mount -vt nfs fish:/pub /pub &
mount -vt nfs sparky:/mnt/a /test &
Here if the portmapper isn't running it is started. Once started, it
is now possible to 'hang' the nfsd daemon as well as the mountd daemon
off it. The two mount commands are from the modified mount command
that come with the NFS package and both are run in the background so
that if one of the servers were unreachable the system would continue
to try while going on to finish the system setup and allow root/users
to login. The '-vt nfs' bit isn't nessessary as the mount program
understands the nfs syntax and mounts it as an nfs system but I
include it anyway.
8. '...And on the 6th day she said, "let there be connectivity"...'
All this is well and fine but shows nothing of how to use the various
utilities commonly taken for granted in networking. ie telnet & ftp
Normally people would telnet over a LAN (Large area network)
to a remote site simply to play a mud (multi user dungeon) which runs
on a socket say port number 4000 so the command 'telnet
wopr.magic.mount.mil 4000' would connect to a service offered by that
machine on port 4000. Now then, sockets are most easily perceived as
'openings' in a wall where data may pass through in a uni/bidirectional
fashion, there are any number of ports available for use and quite a
few reserved port sockets can be found in your /etc/services file.
For example by telneting to port 7 of your target machine you should
be able to communicate with the computer by typing in a few charcters
and pressing return. Port 7 is the echo service and any input you type
should be sent back exactly as you sent it. In normal use, however,
telnet connects to port 23 where a login service is provided for
interactive logins to the system.
The canonical usage of telnet is just
where <hostname> is another machine on the net that you want to
Ftp allows the user to transfer files from the host to the target
machine but requires the user to login as (s)he would normally. Once
logged in the user can transfer files both into and out of the machine
with simple commands like 'get text.doc' or 'send report.wps'. Ofter
ftp is used in the 'get' mode and when browsing sites it is usefull to
know that you can peek at the contents of a small README file using
the command 'get README.requirements /dev/tty' which will transfer the
contents of the file to your tty line (in english: the screen)
To start up FTP, just do
where <hostname> is the machine you want to upload/download from.
For public FTP service, login on the remote machine as "anonymous"
and give your e-mail address as the password.
8.3 X11 and networking
After you have networking set up, you can now run X Windows across the
network. For example, you can login to a remote machine in one xterm,
and from that machine run an X program and direct it to display on
your machine. For example, if your Linux machine is called "shoop", on
the remote machine the command
xclock -display shoop:0 &
would display the clock on shoop's display.
Before you can do this, however, you must run the command "xhost" on
shoop to allow the remote machine to display on shoop. If the
remote machine is "loomer", from shoop you must run the command
to give loomer this access.
This is the entire concept underlying X Windows: you can now run huge
programs (such as Mathematica) on remote machines and have them
display on your Linux box.
9. Standalone named Configuration
What follows is an example named configuration for a local (2-machine)
Well after some peer pressure, I see that I'm going to have to include a
standalone configuration in the FAQ as well. According to my
sources/hallucinations, there is an accepted address that is for
'junk' setups so as not to conflict with other machines on the
internet. That address is 192.0.2.xxx where xxx ranges 0..255.
(This address is not routed through the internet so you should be
relatively safe from ip address clashes).
I'm going to assume that your configurations will be held in /etc so
the following files will be referanced there instead of /usr/etc/inet
or /etc/inet. (NOTE: This deviates from the discussion above. /etc
is fine to use instead of /usr/etc/inet as long as you're
A while ago I posted a couple of messages concerning the setup
of the named daemon config. files for a simple isolated
network with a local nameserver. Since nobody responded with
a ready-to-go solution I decided to dig a little deeper into
the subject. As a result I've now got a working nameserver.
This message describes the changes I made. Here goes:
9.1 General Info
My isolated network consists of 2 machines, called whisky
and jenever which are both located in the domain vdm. Whisky
has IP address 192.0.2.1 and jenever has IP address 192.0.2.4.
The nameserver runs on whisky, and jenever accesses whisky to
Starting point is SLS 0.98pl5. This distribution contains
install.net and hostcvt, which are supposed to make network
installation easier, but they where of no help to me. Instead,
I manually changed the files concerned.
9.2 Common changes to files for both machines.
/bin/hostname machine_name added to /etc/rc. Machine_name
stands for either whisky or jenever. This command should
be placed before the /bin/sh rc.local command. Further
hostname commands removed from /etc/rc and /etc/rc.local.
In /etc/inet/rc.net HOSTNAME=softland changed to
HOSTNAME=machine_name. Commented out the IPADDR= line
and inserted IPADDR=192.0.2.1 or IPADDR=192.0.2.4.
ROUTER set to 0.0.0.0 and NET set to 192.0.2.0. In the
third $CONFIG line eth0 changed into eth_if (I use an
Artisoft network card, this isn't necessary for standard
WD network cards).
9.3 Changes for the nameserver (whisky in my case).
For a nameserver the portmap, inetd and named daemons
should be started. This is done in the /etc/rc.net
primary vdm named.hosts
primary 2.0.192.in-addr.arpa named.rev
primary 0.0.127.in-addr.arpa named.local
@ IN SOA whisky.vdm. root.whisky.vdm. (
1 ; Serial
3600 ; Refresh
300 ; Retry
3600000 ; Expire
14400 ) ; Minimum
IN NS whisky.vdm.
localhost IN A 127.0.0.1
whisky IN A 192.0.2.1
jenever IN A 192.0.2.4
@ IN SOA whisky.vdm. root.whisky.vdm. (
IN NS whisky.vdm.
1 IN PTR whisky.vdm.
4 IN PTR jenever.vdm.
@ IN SOA whisky.vdm. root.whisky.vdm. (
IN NS whisky.vdm.
1 IN PTR localhost.
9.4 Changed for a non-nameserver (jenever in my case).
For a non-nameserver only the portmap and inetd daemons
have to be started. The startup of the named daemon in
/etc/inet/rc.net can thus be commented out.
A non-nameserver depends on a nameserver for name
resolution. The non-nameserver is directed to a name-
server by the /etc/resolv.conf file (NOT the
/etc/resolv.conf as mentioned in a lot of doc. files).
So, the /etc/inet/resolv.conf file on jenever contains:
10. Troubleshooting and Common Problems
Here are some of the most common problems with Linux tcp/ip.
One of the most common complaints regards the 'config' command. What
isn't often noted is that this has to be recompiled from the 0.8.1
sources (available currently as tsx-11.mit.edu:
10.2 Library versions
Another problem that crops up is that some binaries that are
distributed require libc.2.2.2 to be present (i.e. the telnet and
ftp in tcpip-0.8. ONLY use the binaries in net-bin-0.2 or a newer
version (which use jump-4.2 or newer) and you're okay.
Other people think that it's their version of libraries that cause the
problem but can't find the source code for the various utils to
recompile. Get the net-src-0.2.tar.Z package from sunsite or tsx-11
and you're set; recompile at will. :)
10.3 kernel errors
You boot the new kernel and suddenly all hell breaks loose... you
have printk's telling you about RPC errors, framepacket errors etc...
it looks a mess but the kernel keeps working... I suggest you grab
HLU's bootdisk and edit your rc files again. Your problem here is most
likely that you have accidentally attemped to use a working IP address
as your own. If it's a sun's, you can expect the sun to lose all
networking capabillity and not recover until lots of drastic commands
are issued (fastboot won't help the guy either). I was asked to do
this so I wasn't too fussed but loads os system admin people out there
will get very ticked off if you do this deliberately.
10.4 named problems
To check that something is working in named when it is run check out
/usr/tmp/named_dumb.db. This is the file that named creates from all
your configuration files. Check it exists, and contains formatted
information similar to your named.hosts file. If it's zero length then
something is wrong with your SOA record heading (A missing '.' perhaps).
10.5 More than one ethernet card in the machine, IRQ conflicts
If you have more than one Ethernet card in your machine OR you have
a device sharing the IRQ of your network card, you may have problems.
Try pulling one of the cards and see what happens, or changing the
IRQ (usually done with jumpers on the card).
In the Linux kernel source, net/tcp/Space.c defines the network
devices to configure. I hear that if you use the 8390/ne2000 driver
on IRQ 5, the entry for the wd8003 card in Space.c will confuse things;
thus just change the #ifdef around the wd entry in Space.c to something
else so it's not compiled in.
The following is provided by Ross Biro.
If you get the message about time outs on the interrupt, you probably
have your irq configured incorrectly. The irq in Space.c (default 5)
MUST match the one on your card.
If nothing happens when you try to use an interface, check the irq
and try to get a new copy of config. Some versions fail to mark the
interface as up (the config.c in tcpip-0.8-fixes should work).
If you get messages about large packets and immpossible sizes to malloc,
you have the memory on your card configured incorrectly, or there is a
conflict with some other piece of hardware. Fix this by checking that
your memory is configued right in Space.c and if it still fails, try
ALL possible locations in memory (people have suggested that higher
seems to work better.)
If you get a message about runt packets, you can safely ignore it
and/or comment the printk (kernel debugging output function) out of
we.c. It indicates either a hardware problem or a initialization
problem in we.c. It only seems to occur on some versions of the SMC
Elite and there is other code to deal with the problem.
Also Note if it works under DOS does NOT mean there is not a hardware
10.6 General ideas
Now then, to give you an idea of what is possible, I'll describe what
I have setup and working. I have X11(Xfree86-1.2) running... In one
xterm I have a dos session going, in another I have a telnet session
connected to a sun (csd), and on that sun, i'm connected to a diku on
the linux machine through 'telnet slave 4000', in yet another xterm I
have an ftp session to yet another sun(chef) pulling CIA 10Megabugger
world map onto an NFS mounted disk on another sun (hal) at a rate of
about 35k/s (+/- 15k). I was going to mount up a swapfile on an NFS
disk but decided against it on the grounds of what might happen if the
external machine fell over while I was using that swapfile.
Some relief can be found on the newsgroup/mailing lists but one thing
that will *REALLY* help is this...
if ((kernel == lastest_on_offer) && (tcpip_broke))
fprintf(std_email,"give blurb about kernel\n");
system("nm ~linux/tools/system | grep <addr_of_err>");
fprintf(std_email,"Conditions of error (recreatable)");
(Sorry about that, we had a compitition to find out who could write
the whackist pseudo C code) more simply stated, the error address that
is reported by the kernel can be used with a kernel system file to
tell us what function broke and how far into it it broke. See below
for more on that:
Several things that can help
1) Upgrade your kernel to the latest one that you can grab
(currently at time of writting 0.99.4). Alternatively
if you are running 0.98.5, all the patches are available
on sunsite.unc.edu:/pub/Linux/system/Network/tcpip, but
as always, think strongly of going to a higher kernel
version as they nearly alwas have all the patches applied
for tcpip and other misc stuff.
2) Join the NET mail channel, you can learn an awful lot
from the guys on this channel (like the various new
copyrighted techniques for tearing out your hair)
3) Try and upgrade your C compiler and libraries to at least
version 2.3.3 if possible.
4) Binary distributions of various network probrams can be
found on sunsite.unc.edu,.. always read the README files
they are there for a reason! (personal show/contacts/etc..)
nic.funet.fi and txs-11.mit.edu also have good variation
in utilities that you can use. Also don't forget that a lot
of network programs will compile reasonably well although,
be prepared for unexpected weeks of fustration.
5) Depending on your type of problem, contacting the author
of the software or the person who ported the software would
be a better choice.
6) If you are experiencing problems with missing files which are
placed where you think they *should* be, it's always worth
trying the following to find out what files are being used
strings <prog> | less
This should show up any hard linked files in the binary.
eg differing versions of telnet will look at /etc/services OR
/usr/etc/inet/services, therefore, it is a good idea to have
a symlink from one to the other eg
ln -s /etc/services /usr/etc/inet/services
7) If the kernel panics, jot down the address next to EIP. Then do
an 'nm /usr/src/linux/tools/system | sort -n' and find out what
function the given EIP address is in. This will help a lot. If
you simply post the panic message to the newsgroup, everyone's
kernel is different so it doesn't tell us much.
7) Complain bitterly to me if I haven't covered your problem
and I'll get it sorted for the next NET-FAQ.
11. Cast of this production
Ross Biro - Without whom all this wouldn't be possible
and who pointed out holes in my documentation.
Also contributed the history of tcp/ip on linux
after he saw my rather perverted view of it.
Mitch DeSouza - Constant alpha tester. Also pointed out mistakes
and made critical and helpfull suggestions (like
getting a spell checker). Also gave me his Tel No.
which I used to annoy him with.
Rick Sladkey - The current author of the NFS client code in the
kernel. He also ported the NFS server and the RPC
Donald Becker - Author of the drop in drivers for the linux kernel
allowing the following cards to be used,
3com503, 3com503/16, NE2000, NE1000 and even a
3com501 (Donald: 'not recommended').
Matt Welsh - Trashed, er... reformatted this document, tried
to clean it up. Wrote the tcp/ip quick start guide and
answers tcp/ip config questions.
The pioneers - These are the fearless people who brazenly marched
their filesystems towards complete oblivion and
watched weeks of work evapourate in milliseconds
without a shred of hate for the OS that they had come
The supporting - You know who you are (probably, depending on how
extras much virtual beer you had last night) for contributing
to the network code with the various bug reports that
inevitably crop up.
Linus Torvalds - The elusive ecentric UNiX kernel coder who probably
burns more CPU time on compiling than anyone else
Here's to a long and healthy kernel development
program and a Nobel equiv award for his efforts.
The critics - For reminding me that it's a thursday... I never
could get the hang of thursday's...
Myself (Phil) - The only sad person to take on the FAQ because I was
getting annoyed at the number of 'petty' tcp/ip code
problems being asked on the net. Besides of which I
wanted to give something useful towards Linux which
I've used since 0.10 (does this make me a veteran?)
Phil (The non spell checking insomniacial/palagerist who never learnt
=--= english grammer)
Matt Welsh, m...@tc.cornell.edu
"What are you doing, Dave?"