The Answer Gang

Dav jeg syntes at det er en gode side du har med en masse gode brugbare r�d . men det er ikke det jeg vil , je har et problem som du m�ske kan hj�lpe mig med . Jeg har en 450 mhz p3 cpu som jeg gerne vil have overclocket jeg har et asus bundkort model :p2b/f1440bx agp atx. Jeg ved ikke om at jeg skal have noget extra k�ling p� n�r det kun er til 500 mhz da mit bundkort ikke kan tage mere.enanden ting er at jeg ikke ved hvordan jeg g�r s� jeg h�ber at du vil hj�lpe mig.JEg h�ber at du vil hl�lpe mig med mine sp�rgsm�l.

hilsen kasper

Andswer Dude,

Your response to Tony Grant on the Plug and Play board problems (#36) can

setserial /dev/ttyS3 uart 16550A port 0x2e8 irq 10

to the /etc/rc.d/rc.S file it'll be run on every boot (duh) and correct the problem. Of course the IRQ and IO need to be changed. The chipset of 16550A is pretty much standard and most likely won't need changed...but if it does, you can always grab it easily. All that command line does is force's the box to accept the comport and recognize that it can in fact be used. Dammed defaults tend to only recognize Com1-Com3...Hopefully the next RH, Caldera OL, or Debian should have that fix (since Slackware is just..well....lacking....nobody has hopes for that to ever get itself in gear.)

- -=Razor=-
- -=Buzz=-

[Jim] The code fragment you've included doesn't specify what scripting language you're using. It isn't a valid fragment of bash, PERL, or even csh. For one thing, the common UNIX scripting languages are case sensitive. Thus your capitalization of "If" and "Then" are enough to cause this fragment to fail under most interpreters.

Of than that there isn't enough context or code here to guess what scripting language you're trying to use. However, the 'set' command isn't used in most Linux scripting languages (at least not for "setting values to variables"). csh, TCL (and 'expect', a TCL derivative) and the MS-DOS batch language, use the "set" command for variable assignments.

This leads me to suspect that your code sample is in "MS-DOS batch" or some sort of psuedo-syntax.

To do this with bash (or Korn shell or any similar interpreter) you'd use something like:

    #!/bin/sh
    a=$(cksum $1)
    b=$(cksum $2)
    if [ "$a" = "$b" ] ; then
       ...
       else
       ...
       fi

...assuming that you were calling the script with two parameters, the names of the two files. Note: the $( ... ) expressions are the key here. They "capture" the output from the enclosed command(s) and substitute those result into the expression in which they $(....) expressions have appeared. This is called "command substition" (traditionally rendered as `...` using backticks). This "command substitution" feature is one of the shell's most powerful and useful scripting mechanisms and it allows us to seamless assign the output from any normal command (internal, or external) to shell variables.

(Note: Some very old Bourne shells might not recognized the

$(...) form and thus may require the backtick form. However, all UNIX shells should be able to do command substitution. I've never heard of one that didn't. csh/tcsh also requires the backticks, and can't use the more legible $(...) form).

Actually this is an oversimplification. The GNU 'cksum' command prints output of the form:

2839321845 1516 /path/file.name

Obviously if I take the output of two of these commands, with DIFFERENT FILENAMES the full text of each output will be different even if the checksums are the same. I need to extract just the checksums, or at least filter out the differences in the filenames.

My first thought was that the cksum command might have some switches or options to suppress the extraneous output. It seems like the need to get just the numeric checksum value would be pretty common. However, it appears that the FSF maintainer for this utility doesn't agree with me. So we have to isolate it ourselves. That's only a minor nuisance (taking far less time for me to do than to explain).

There are a couple of ways I can do that. Here's the first that comes to mind. Just insert the following at the top of the script.

   function cksum () {
      command cksum $1 | {
	  read a b x
	  echo $a $b
	  }
  }

This creates a local shell function which over-rides the output of the external cksum command. The "command" command forces the shell to execute the command (bypassing the shell functions and aliases --- and prevent a recursion loop).

All I do here is pipe the output into a command that reads the first and second fields (the part I want to keep). I read the rest of the output into a "throwaway" variable (which I expediently call "x"). Then I just echo out the two pits of info I cared about (the checksum and the size) leaving off the "rest." This trick of using the read command to filter out fields that I want from lines of input is pretty handy. It's a reasonable advantage over using the external 'cut' command because read and echo are internal commands. Also 'cut' defaults to using tabs as delimiters while I usually want to "cut" on any whitespace (any number of tabs or spaces).

The advantage of writing this little shell function into our script is that I can leave the rest of the script alone. I don't have to re-write it. Of course it's better to avoid the name collision. I could name my function "checksum" (and avoid having to use the "command" command). Even if I do rename the shell function I can leave my "command" command as is. It doesn't hurt anything.

Naturally I could have also just piped the output of each of these cksum command through cut like so:

a=$(cksum $1 | cut -d" " -f 1-2 )

... which works fine. Of course it is a little less maintainable. Even though I'm only calling this expression twice --- it's still better to consolidate it into a shell function so it really works the say way in both invocations. Otherwise a slight difference to one of the invocations could silently cause the later comparison to always and erroneously fail.

Note that we don't have to use "if... then ... else .... fi" in most shell scripts. We can shorten this script to:

[ "$(checksum $1)" = "$(checksum $2)" ] && .... || ....

(assuming I made my checksum shell function as before).

... where the command after the && is the same as you'd put after the "then" token in the earlier script. The command after the || operator is similar to the "else" block, but it would be execute if the checksums didn't match or if the if the command in the && clause returned a non-zero value (an error). This is frequently what you actually want in shell programming; though the differences can be subtle and important.

Note: the && and || operators take a single command. If you want to perform a block of commands under those conditionals you'll want to use command grouping or possibly a subshell --- using the {...} (braces/grouping) or (...) (subshell) syntax.

One "gotchya" that crops up in bash 2.x when using "grouping" is this:

{ foo; bar }

... was accepted under bash 1.x and is an error under bash 2.x --- it's because the closing brace is being taken as an argument to the bar command. This is technically correct for the parser (it was a bug in bash 1.x that allowed the command to work).

So, good shell scripting requires that we us this syntax:

{ foo; bar; }

(or simply put the braces, particularly the closing brace, after a line end, perhaps on its own line).

That's basic shell scripting.

[Jim] Actually the [ .... ] doesn't "imply" the test command. [ is really a built-in alias for 'test' (and it generally also exists as a symbolic link to the /usr/bin/test command, for those shells which don't implement it as a built-in).

When the command 'test' is called under the name '[' then it requires the ']' as a delimiter. That's actually a bit silly, since the shell is still doing it's own parsing, and the shell "knows" when the command ends quite independently of this "]" marker (which the shell ignores as it's just another argument to the '[' command.

However, these are just syntactic anomalies. It's really better for beginning shell scripters to use the 'test' command (so that the really internalize that it is really just a command like any other Unix command. It is not a "feature of the language" --- it's just a command that processes a list of command line arguments and returns and exit value. (This is as true of '[' but it's less obvious to people who've been exposed to any other programming languages.

[Jim] Although bash allows this, the external 'test' command requires that we use the = and will give an error if we try to use ==

So, depending on bash' permissiveness is less portable.

Incidentally, another approach we could have used (given the original problem) is to do something like:

pushd $(dirname $1)
a=$(cksum $(basename $1 ))
cd $(dirname $2)
b=$(cksum $(basename $2 ))
popd
....

... this relies on the fact that the files being compare have the same names but reside in different directories. However, it seems really bad to impose that constraint on our shell script even though this particular application/situation allows it. It would make the resulting script useless for most other situations. However, the approach I recommended (filtering out the filename with and read/echo pair or a 'cut' command) gives us a more general script that we can re-use for similar purposes.

This example does show the use of the very handy 'basename' and 'dirname' commands. It also shows that the $(...) form of command substitution can be nested (which overcomes a limitation of the older `...` backtick form).

[Jim] Once you have local copies of the file, why not just use the 'cmp' command. The cksum command is already going to read the whole file. You've already burned up the bandwidth (transfer the whole files to the local machine).

So what's wrong with:

 if cmp -s /old/path/file1 /new/path/file1
    then
      ...
    else
      ...
    fi

That seems quite a bit simpler.

Also, let's assume that you have two directories. A script to compare corresponding files in them would look something like:

 for i in $1/*; do
   cmp -s $i $2/$(basename $i) && # they're O.K ...
	|| # Ooops: corrupt file
 done

(This assumes that you're call it with just two parameters, the names of the old and new directories).

Alternatively you can have a script take a directory name (the "new" directory for argument's sake) and a list of files as probably provided by a "wildcard" (globbing) pattern.

That would look something like:

 d=$1
 [ -d "$d" ] || exit 1
 shift
 for i; do
    if cmp $i  $d/$( basename $i )
       then
          ....
       else
          ....
       fi
 done

... Here again I'm using the basename command. I could also use the "parameter substitution" feature of the shell instead of basename: ${i##*/} However, I find that form to be almost unreadable. If performance where an issue I might hide the ${1##*/} in a shell function that I'd name "basename" (and I'd toss in ${1%/*} as "dirname"). That would be a bit quicker for large directories since basename and dirname are external commands. So using them entails quite a bit of fork()'ing and exec()'ing. Naturally the ${...} parameter substitution features are always internal if they are supported at all.

[Jim] Of course the "folder" (directory name) is part of what's being compared. The "$(.....)" are expressions that evaluate to text strings. The contents of those strings are set to the output of the commands that are included in the parentheses. The [ (test) command takes a list of arguments and operators. In this case the arguments are two strings (substitutes by the $(...) expressions) and the = operator. Note that the "=" sign here is just an argument to the test command --- which is also know as the '[' command. The closing ']' is just an argument that the 'test' command requires when it is called under the '[' name.

Now, if you think about it you'll see that the '[' command has no reasonable way of "knowing" that you only care about the checksum values of the two strings. It was give a couple of strings and an argument (the "=" sign). So it (the test command) will return a value (exit code, errorlevel) based on whether the two strings are identical.

[Jim] I don't recommand that. The 'test' command will probably emit an error about the format of the operands to the -eq option/operator.

[Jim] See my long response of a few minutes ago. The answer is simple, we include a bit in the $(....) expressions that filters out the irrelevant text. I do this by over-riding the cksum (external) command with my own shell function, but the concept is the same.

Note: I dove into that message and my earlier response before seeing that others had tried to help you with your question.

Henry C. White

HELP

AnswerGang: Michael Williams, Heather Stern

From WwSHADOWMASTERwW on Thu, 29 Jun 2000

Listen. I just installed RedHat Linux 6.2 and I cannot get my modem to work. I did the test and modem test on the set up manu and is does detect it but stays at the initializing Modem prompt.. What do I Do I can t find anyone who can answer this for me HELP.....I am using the KDE work station setup..please tell em Step by Step on how to do this I would appreciate it very much
PS I am not using Gnome!

[Michael] Is your modem internal? If it is, then there's a fair chance it's a 'WinModem'. These are modems designed to work within MSWindows. Since they use drivers written for MSWindows to work, it is very difficult [currently impossible] to get them working under Linux. If this is the case, then your best bet is to buy a new external modem. They're reasonably priced, and will work with all OS's.

[Michael] What distribution are you using? I'm guessing it's Caldera, since that attempts to set up the modem at installation.

[ No, he said RH 6.2, but that's an interesting factoid, so it stays. --Heather ]

You don't actually have to 'install' the modem as you would have to do in Win98. To use a modem, firstly find out its comm port. It'll probably be in Comm1 or 2. Under Linux, these appear as /dev/cua0 and /dev/cua1. You'll also need to know the modems speed. If it's a new modem it should be 57000 k's a second. Now, to use this goto kppp under the internet selection of the KDE 'start' menu.

It's pretty self explanatary from here onwards. Enter your comm port - try from 1 - 4 ( cua0 - cua3 ), until you find which port your modem uses. Enter your modem's speed, and then your ISP's details. Unless you have other problems, that should allow you to use the internet.

Binfmt/Exec Format Errors in /linuxrc on initrd

AnswerGang: Jim Dennis

[ Folks, while our Answer Gang does read technese as well as English, it helps if you use some connective grammar... little things like "when I used 'cat whateverfile' it said " or "with kerneloptthingy=nnn I can see syscalls blah() blah() blabla()". This one had to be translated, and my wildest guess is Fuchangdong uses some sort of kernel debugging that he didn't describe to us. --Heather ]

From fuchangdong on Mon, 17 Jul 2000

please give me some help,i didn't know how to explain at my implementing embeded os. fuchangdong

You're trying to use Linux for an embedded system?

http://www.sohu.com/sas/temp/twoyear/2year.html http://www.sohu.com

hi :

i now have a question,please give me help, i use initrd and ramdisk to complete embedded linux on my hardware. first ,i create a initrd.img from command mkinitrd.and a bigger root fs:ram.img.gz ,to lilo it,and reboot it

You're using the Linux initrd (initial RAM disk) feature. You use the mkinitrd command to create your RAM disk image install that and your kernel onto the target hardware (which I presume is x86 because...) you then run /sbin/lilo on that and try to boot it.

at init process,do_basic_setup,this line :

kernel_thread(do_linuxrc,"/linuxrc",0);

at this function: do_linuxrc()

execve(shell,argv,envp_init); it return -1 ,and errno is 8,this tell that it is "exec format error"

so i can't to exec linuxrc script file.

According to the kernel sources it is calling the kernel_thread(do_linuxrc,...) function and the do_linuxrc function returns a failure on the execve(), with the errno global set to 8, which translates to "exec format error" according to the strerror()/perror() function.

linuxrc's content is :

#!/bin/sh
ls -l
and chmod 0777 linuxrc

The /linuxrc is a trivial (test) shell script. You've tried marking that as executable with the chmod 0777 command.

so i can't know what wrong with me? why initrd.img cant't be load right? but i find :

ret = open("/linuxrc",O_RDONLY,0); ret = success.

If you (patch the kernel?) to simply open the file you don't see any error.

and infomation have : mount root filesystem (ext2);

You think you have an ext2 filesystem mounted on root at this point? (It's not clear how you are getting this info).

so i can't get reason ,please give me help? linux is redhat 6.2 linux kernel is 2.2.12-20

The development environment is a Red Hat 6.2 system and you're using a 2.2.12-20 kernel.

after, i test this ,give me these information: i add modprobe/insmod command in initrd.img, reboot it, this system give me information: " kmod:failed to load /sbin/modprobe -s -k binfmt-0000"

When you try to run a modprobe command in the initrd.img you get a kmod binfmt error.

execve() call do_execve(),do_execve() call request_mode() ,request_mod() call exec_modprobe(),so it's path is right. but i can see this inforamtion ,at boot ,system load script ,aout,elf binfmt. so i can't know greater!!! please give me help !!!

This last bit of typing is utter gibberish. Actually your whole message is basically incomprehensible. However, I've echoed a guess after each fragment of what you've said to see if I could understand the question.

It sounds to me like you are somehow missing some of the necessary binfmt loaders from your kernel. Now there are a couple of options in the 'make config' scripts that allow you to enable or disable a couple of different types of executable (binfmt) loaders. You generally need at least one of them compiled directly into the kernel (so that it can execute a linuxrc and/or an init(8) process).

I don't think it's possible to build a kernel without statically linking one of a.out (COFF) or ELF. If 'make menuconfig' somehow let you pull that off, it's a bug in the Makefiles and dependencies.

You need one of those.

In addition I've never seen an option to leave out the text/script binfmt loader. That is the loader that handles text files and uses the #!/.../ line to execute most scripts.

However, it would seem that you have somehow managed to do this. I could see it if you had been applying your own patches to the kernel code, or if you were hand editing or bypassing the Makefiles with some of your own.

I suppose English is not your native language (given the distressing incompetance of your message). I supposed you should look for a (Chinese?) users group, newsgroup, mailing list or other forum where you can have someone translate your question into English.

Other than that try recompiling your kernel and ensuring that the ELF executable support (under "General Setup") is set to "Y" (NOT "M" and definitely NOT "N").

To quote the help text that is associated with that menu config option:

	Saying M or N here is dangerous because some
	programs on your system might be in ELF format.

It is highly unlikely that you are somehow managing to compile your core shell and other software in a.out format. That actually might be quite useful for embedded systems work --- but the older format and the tools to generate them haven't been used by any general purpuse distribution in a few years. The only remaining a.out distribution that I know of is David Parsons' Mastodon (http://www.pell.portland.or.us/~orc/Mastodon).

So, I think you can safely leave out the other binfmt loaders.

BTW: You also MUST have one of the filesystem types statically linked into the kernel. You can just go through and blindly mark EVERYTHING as modular. It won't work. The initial RAMdisk will have to be in some filesystem format (minix, ext2, something). Of course it would be possible to use the ROMfs. This is much different than initrd --- it's readonly and you have to make the filesystem using a genromfs utility AND you'd have to link your ROMFS into your kernel. I don't know of anyone that actually uses ROMFS.

Anyway, I suspect that the reason your shell script isn't working is that the kernel can't load the shell interpreter. The reason it can't load the shell interpreter is because your shell is probably in ELF (executable linking format) and you left the ELF loader out or put it in as a module. Of course the insmod/modprobe programs are also in ELF format --- and the kmod (kernel loader module) requires access to those in order to actually load any modules. (kmod doesn't load modules, it spawns a kernel thread, which runs modprobe to do the actual work. You can read /usr/src/linux/kernel/kmod.c to see that.

I hope that helps.

Mandrake and the Missing Modem

AnswerGang: Jim Dennis

From Michael Hudson on Tue, 04 Jul 2000

Hi yall,

First off let me tell you that I am completely new to the Linux world! I have been <Stuck> with Windoze most of my computing life.. I have only recently discoverd this whole new world! So please make you answers as simple as possible to understand.. Thanx in advance!

I have recently installed Linux Mandrake on my K6 Machine. I am running it Dual Boot with Windoze.. I am having some reall problems setting up my modem to actually work..

I think this is solely down to my lack of knowledge towards Linux... Could NE1 give me some advice?

Yours, Michael Hudson.

You're also having "some reall" [sic] problems describing your problem. Read back through your message. Try to pretend that you were getting this from some stranger. Do you really think there is enough detail provided for any mere mortal to devine what you problem could be?

I understand that you're a novice a Linux. However, you could put a little energy into the questions you're going to ask.

What did you try to do? Did you run some program to try to "set up" you modem? What do you mean by "set up"? What kind of modem is it? If you ran some program or command to try ot "set up" your modem; WHAT DID IT DO? Did you get a error message? What were you expecting the modem to do? What did it do?

Did you read any manuals or do searches through any Internet web search engines?

Anyway, the problem is probably that you probably have a "winmodem" or a "softmodem" or some other useless piece of junk that isn't really a modem. If you go back to the Linux Gazette (which you should have read in order to get this e-mail address) and you peruse the FAQ and maybe search on the word "modem" you'll find about 100 other messages where I've talked about modems, Linux, using modems under Linux, testing to see if your modem is supported by Linux, and especially about why "winmodems" are such losers.

Making the Laptop's Fan Run

AnswerGang: Jim Dennis, Heather Stern

From Allen Tate on Thu, 27 Jul 2000

Anyone out there know anything about making the cooling fan run on a laptop running Linux? Seems I read something somewhere about running a module that made the fan run. Any advice is appreciated.

[Jim] What makes you think you need a special module or driver to control your system's fan?

On any reasonable equipment the fan should run when it is needed without any software support required. The hardware should include its own thermostat which should operate completely indendently of the OS.

(Actually there's a good argument that we should be producing better hardware that runs cooler, with lower power consumption. So that fans would be unecessary for most laptops and general purpose computing devices. That's what Transmeta --- the company for which Linus works --- has recently introduced to the PC market).

Anyway, I don't know of any module that "makes the fan run." or anything like that. The closest I can think of would be the ACPI kernel features (ACPI is an advanced and somewhat complicated alternative to APM --- advanced power management). That would require that you get a daemon to call those kernel functions from user space. Under Debian you'd just use the command 'apt-get install acpid' to fetch and install that daemon, under other Linux distributions you'd have to hunt for it on your CDs, and/or look for it on their FTP contrib sites, etc).

There is also a package called "LM_Sensors" which allows one to monitor some values such as CPU temperature, fan speed, power supply voltage, etc. There are a number of motherboards which use an LM78 or similar chip and sensor set to allow software access to these sorts of metrics. Under Debian you could get the sources to this package using 'apt-get source lm-sensors' which will fetch the original package sources and the Debian maintainer's patches and unpack them under your current directory. I usually do that sort of thing from my /usr/src/debian directory.

LM_Sensors consists of a kernel patch (you must recompile your kernel to add these features) and some user space utilities for querying the kernel driver.

I highly recommend LM_Sensors to sysadmins who are maintaining servers at co-located facilities and in server closets. Those are places where having this information available via software can save a great deal of downtime and damage. (The audible alarms that might be in your case to warn of fan failures and overheating aren't very useful when there's no one there to hear them. Also the typical machine room has to much fan and air conditioning noise for anyone to hear the failure of one system).

However, I don't know if any laptops have any of the support LM78 or similar sensor features. So that's probably not useful to you.

... he replied ...

Thanks for the advice. I look into it.

MX Records and Precedence Values

AnswerGang: Jim Dennis, Mike Orr

From Todd Tredeau on Sat, 01 Jul 2000

I am trying to understand mx records, and the role the play in relationship to a backup queue server. I have two mail servers mx1.wisernet.com and mx2.wisernet.com, I also have a third emergency back server, to be manually added if I need it.

If the primary mail store is on mx1 then should the priority be higher or lower?

like mx1.wisernet.com 10 (primary) mx2.wisernet.com 20 (backup)....

your help would be greatly appreciated, I have all sorts of mail problems....Actually my antispam software was working so well at one point, I couldn't send messages from mx1 to mx2 and so on... got that straightened out though. Nice web site.....

[ Thanks! -- Heather ]

Todd

The MX record with the lowest value will have the highest priority. Think of it as the "distance to user's mailboxes" and consider that the various MTAs (mail transport agents) which are relaying a piece of mail are each seeking to get the mail closer to its final destination.

Of course the host with the lowest MX value will either have to accept the mail or there will have to be an accessible route to an A record of the host. (Note: CNAMES are never supposed to be used for mail exchanges). Normally we have MX and A (address) records for any host that is supposed to receive mail.

In general there is nothing special to setting up backup MX relationships. It used to be that you could simply add the appropriate MX records to your domain zones. These days there is one extra step.

In recent years it has become almost mandatory for sites to limit their mail relaying. Before the advent of widespread spamming it was common to allow "promiscous relaying." That basically meant that my mail servers would attempt to forward/relay/deliver any piece of e-mail that landed on them, regardless of where it was from and regardless of who it was to. That was basically a fault tolerance feature. If a bit of e-mail got mis-routed and landed on my server --- the server would just try to get it delivered anyway. That was common courtesy in a co-operative Internet.

However, the spammers ruined all of that forever. They would dump one item of e-mail, generally with a couple thousand recipient addresses, onto any open relay. This allows the spammer to use a small bit of their own bandwidth (as provided by a 14.4 or 28.8 modem) while leeching much more bandwidth (a few thousand times their "investment") off of the rest of the Internet and the host of the open relay in particular.

So now we have to also configure the MTA on our backup MX hosts to access mail to our domain. (Obviously that's no problem if we're talking about additional hosts within our domain --- they presumably are already configured to accept/relay mail for us. It is also true of cases where we want to set up mutual backup MX services for and with other domains. (Thus if the connection(s) into our domain is/are down, or if some regional outages prevent some customers from reaching us directly, but still allow connections to one of our MX partners, then the mail works its way towards us. The correspondents feed their mail up to any available MX server, so the mail doesn't languish on thier systems.

That's the idea, anyway. I've had some people question whether configuring backup MX services is still appropriate in the modern Internet. Personally I think it is. However, there are valid arguments on both sides of this issue.

[The way I heard it, if the primary mail server is down, a secondary server's job is to accept the message and keep trying to forward it to the primary server, with a longer-than-usual retry timeout. This prevents the mail from bouncing needlessly if the primary server is down for a while. Note that the secondary server cannot deliver the message itself, since the recipient is not a local user on that machine. --Mike]

Re: unable to open a initial console

Also: A Short Guide on How to do Backups and Recovery:

AnswerGang: Jim Dennis

From Asghar Nafarieh on Tue, 25 Jul 2000

Hi,

I hope you can help me on this problem. After booting my linux server (RedHat6.0) It goos through booting and comes back with the above prompts and hangs there. I have hat this machine running for 6 months and this is the first time this is happenning. I have a lot of data in there. I tried to use the resuce disk but I don't know how to get to the hard disk to check the problems. I appreciate your help.

Thanks, -Asghar

This error message basically means that the kernel was unable to find a console on which it could run init.

That suggests that it can't find your /dev directory (on the root filesystem) or that it can't find the appropriate /dev/tty* and /dev/console device nodes thereunder.

This is most commonly caused by one of two problems:

1. Perhaps you removed or damaged the /dev/* nodes that the kernel needs.
2. Perhaps the kernel is mounting the wrong filesystem on the root directory (a filesystem which doesn't HAVE a /dev directory).

So, here's how you use a rescue diskette the troubleshoot this sort of problem:

1. Boot from the rescue diskette.
2. Mount your root filesystem. Use a command like:

   mount /dev/hda3 /mnt

3. Look for a .../dev/console device thereunder. Use a command like:

   ls -l /mnt/dev/console

   It should look something like:

   crw-r--r--    1 root     root       5,   1 Jul 21 14:50 /dev/console
   

If it's there then you want to try booting from your hard drive again. This time, at the LILO prompt you'd interrupt the boot process and pass the kernel some options.

When you see LILO press the [CapsLock] or the [ScrollLock] key. Then hit the [Tab] key. That should give you a list of available boot labels ("linux" and "dos" for example). You'd type something like 'linux root=/dev/hda3 init=/bin/sh' (Be sure to refer to the same device, hda3, or whatever, as you did when mounting your root fs under the rescue diskette).

In this case I've specified the kernel option "init=/bin/sh" just for further troubleshooting. If that comes up O.K. you can then type 'exec /sbin/init 6' to force the system to shutdown and reboot under the normal init.

I realize, from the tone of your question, that this may all be a bit confusing to you. You don't mention what you've done to the system between the time that it was working and the time that this error started occurring. I can guess at a few possibilities, but I'd only be guessing.

For example: if you are someone else with administrative access to that system had built a new kernel it might be that you built it with a faulty "rootfs" flag. A Linux kernel as a point to the default root filesystem device and partition compiled into it. If it isn't passed a root= parameter, than the this compiled in pointer specified which device the kernel will try to find and which partition it will try to mount as root. Normally the LILO boot loader has a root= directive in it. That is usually in the "global" section and is used for any "stanza" which doesn't over-ride it. When we are typing in root= directives at the LILO prompt we are over-riding both the kernel's default and LILO's stored option.

As you can infer from the foregoing the Linux kernel mounts a root filesystem and then it opens a console device. That done it prints alot of messages to the screen, and runs the init program. It looks in several places like /sbin, /etc, and /bin, for a program named 'init' then it looks for /bin/sh as a failsafe. Failling all those the kernel will print an error message like: "No init found. Try passing init= option to kernel."

(You can read the kernel source code for these actions in /usr/src/linux/init/main.c).

Note that I haven't addressed the issue of whether there is a Linux filesystem, recognized by your kernel, available. If you had no Linux filesystem there, you'd be getting a error more like: "VFS Kernel Panic: Unable to mount root" or "VFS: Cannot open root device" (depending on whether the filesystem/partition was nonexistent or corrupt, or whether the device couldn't even be found).

I've also left out any discussion of the initrd (initial RAM disk). Red Hat does tend to use these, though they are not necessary for most systems. Here's a little bit about how those work:

If you are using an initrd, then the loader (LILO) must load the kernel, and the initrd into memory. It then passes the kernel an option. The kernel (with initrd support enabled) will then allocate memory for a RAM disk, and decompress the initrd image into that memory. Normally the initrd will contain a compressed filesystem image. (It's actually possible for it to contain other sorts of data, but that's not a feature that I've ever heard of anyone using).

Once the initrd (RAMdisk) has been initialized and populated, the kernel temporarily mounts that as the root filesystem and attempts to execute a command called /linuxrc. After that command exits, then the regular root filesytem is mounted, and the normal init process is run.

Note that this is basically a hook between the kernel's initialization and the normal root fileystem mount and init process. Often the initrd will have no effect on the regular boot process. However the most common case is for the initrd to contain some modular device drivers, and for the /linuxrc to load them. This is intended to allow the kernel to access devices for which it only has modular (rather than compiled in) drivers.

(Usually I suggest that users learn how to compile their own kernel, statically including their main disk interface and network adapter drivers. That obviates the need for an initrd, making the whole system a tiny bit easier to maintain and troubleshoot).

I mention all of this in your case because it's possible that you kernel is fine, your root filesystem is fine but that your initrd has been corrupted and is setting the rootfs flag to some

For more details about this initrd subsystem you can read /usr/src/linux/Documentation/initrd.txt

Of course I should also take this opportunity to give the standard parental lecture about the need to make and test backups. However, I don't have a really good resource to which I can refer you. I don't know of a well-written "System Recovery HOWTO" and I should take it upon myself to write one. (The third chapter of my book on system administration is a start --- but it doesn't go down to step-by-step details).

Let's just say this for now:

If you end up re-installing here are some tips to make recovery from these sorts of disasters much easier:

First, during installation, create at least three or four partitions. I like using lots of partitions. You want to have partitions for root (/), system (/usr), and data (/home) at least.

I like to have an alternative root filesystem (/mnt/altroot) (which is normally not mounted) and a /var partition. Then I may add other partitions based on the needs of a specific machine. I usually create /tmp and /usr/local partitions, and sometimes I add /var/spool and/or /var/spool/news partitions for some mail and news servers.

One of the reasons for this partitioning is to facilitate system and data recovery. Most problems will only affect one of your filesystems. For example, if your root filesystem is damaged (as it appears has happened in your case) then you can just reformat and restore that without worrying about your data (which should mostly be stored on /home and/or /usr/local).

If you have a separate /boot partition it can be mounted read-only most of the time (just remounted in read-write mode when you are installing a new kernel). That can also work around limitations of older BIOS' and versions of LILO with regards to the infamous 1024 cylinder limit. If you keep an extra "alternative root" filesystem you can maintain a "mirror" (replication of) the root filesystem on that, with copies of all the system configuration data (from under /etc). Then when your root fs is damaged you can simply boot from the altroot using the root= kernel/LILO option while booting. (You could also use the root= directive when booting from a floppy disk or bootable rescue CD).

You can copy all of your root fs to the alternative root with a sequence of commands something like:

     mount /dev/hdc8 /mnt/altroot
     cp -ax / /mnt/altroot
     umount /mnt/altroot

... assuming that you already have created a /mnt/altroot mountpoint (using mkdir) and that you have a partition like /dev/hdc8, the fourth extended partition on the primary IDE drive of the secondary controller, with a valid filesystem thereon. Once your create an altroot partition

I suggest keeping /usr as a separate filesystem for two reasons. You can keep it mounted read-only most of the time (remounting it in read-write mode during major system upgrades and while installing new packages). That makes it more difficult for it to get damaged and might even protect your system from some of the sloppier "script kiddy" exploits (it's not a real security feature, a better exploit will remount filesystems read-only before installing a rootkit).

Of course keeping /home as a separate partition should be fairly obvious. If you're using your system in a sane fashion, most of your data should be under /home. That means that you can focus on backing that system up. The other filesystems should change somewhat less often, and you can be assured that the programs, libraries and other files are store on them are recoverable (from your installation CDs, and the Internet at large) or are expendable (temporary files, caches, logs, etc).

Under Linux there are many different ways to perform a backup. In general you can use 'tar', 'cpio' and/or the 'dump' commands for individual systems, or you can use the free AMANDA package for setting up a networked client/server backup infrastructure.

Each has its advantages and disadvantages. You could also get BRU (the backup and recovery utility) which is probably the most popular among several commercial Linux backup packages.

Of course you need more than software to do backups. You need to have places to store these backups (media) and a device to handle the media. Some of your choices are tape drives, CD-R or CDRW, magneto optical or any of various types of removable storage ranging from floppies through LS120, Zip, Jaz, etc.

Most systems sold these days don't include any backup devices. With common disk drive capacities of several gigabytes, we can't count 1.44Mb floppies as a reasonable backup device. (Even in the days of 100 and 200 Mb hard drives, no one was using floppies to do full system backups). Managing a thousand or more floppies per hard drive is absurd.

Even the systems that sell with LS120 or Zip(tm) drives aren't really meeting the backup/recovery needs of an average user. It wasn't too bad for one and two gigabyte systems (10 to 20 disks) but it's not reasonable for the 6 to 18 gigabyte hard drives we're seeing now (60 to 200 disks). Even CD-R or CDRW are barely adequate for backing up individual systems (at 650Mb each you need about a dozen discs for a typical drive, and I'd need almost 30 of them to backup my laptop).

So the only reasonable way to do full system backups on most moderns PCs is to use tape drives. A 4mm DAT3 tape can store 12 Gb uncompressed. DLT tape drive capacities range from 20 to 70 Gb. There are other drives ranging from 250Mb (FT) through over 100 Gb and most are supported by Linux drivers.

The biggest problems with tape drives is that they are expensive. A good tape drive costs as much as a cheap PC.

Let's say you bought a 4mm DAT drive (and a SCSI controller to go with it). You could to a backup of your whole system with a command like:

     tar cSlvf /dev/st0 / /usr /home ...

... Note: here I'm not using compression, and I am using the "S" (--sparse: note that's a capital "S") and "l" (--one-file-system a lower case "ell") options to 'tar'. I'm assuming the first (usually the only) tape drive which is called /dev/st0 (or /dev/nst0 if you want to prevent the system from rewind the tape after the access). I'm listing the top level directory of each locally mounted filesystem (the mount points). Using this technique avoids inadvertantly backing up /proc (a virtual filesystem) and any network mounted or other unusual filesystems. Obviously you'd only list those filesystems that made sense for your system (read your /etc/fstab for a list).

I could add a "z" flag to force 'tar' to compress the data, however that usually causes latency issues (the data doesn't "stream" or flow smoothly to the tape drive). Since the tape must be moving under the read-write head at a constant velocity, if the data doesn't stream you'll get "shoeshining." The most common causes of this are compression and networking. So, in those cases you'd use a command more like:

     tar cSlvf - / /usr /home ...  | buffer -o /dev/st0

(Here, I've changed 'tar' to write it's output into the pipe --- to stdout technically --- and added the buffer command which using a bunch of shared memory and a pair of read/write processes to "smooth out" the data flow).

Hint: You should write down the exact command you used to write your data on any tapes that you've created. This allows any good sysadmin to figure out what command is required to restore the data.

To restore a system using such a tape you'd follow the following procedure:

1. Boot from a rescue diskette or CD (or onto your altroot)
2. Mount up a temporary filesystem using a command like: mount /dev/hda5 /tmp (or make sure your RAM disk has a few meg of free space).
3. Restore a table of contents (index) of your tar file to /tmp/files using a command like: tar tf /dev/st0 > /tmp/files
4. Restore your /etc/passwd and /etc/group files from the tape. Overwrite those in your rescue system's (RAM disk based) /etc directory.

   NOTE: This must be done in order to ensure that all the OTHER files that you restore will have their proper ownership and permissions. Otherwise you are quite likely to end up with all the files on the system owned by the root user (depends on the version of 'tar'). Trust me, you need to do this. This may be a bit time consuming, since the tar command will go throug the entire tape to find those two files. (It does make more sense in practice to do do different backups to your tapes, one of just the root filesystem, or even just the /etc directory, and the other containing the rest. However, it is more complicated to understand and explain, as you're dealing with "multi-member" tapes and have to know how to use the 'mt' command with the nst0 device node to skip tape "members" (files). This method will work, albeit slowly).

   To do this selective restore use a command like:

   		   tar xf /dev/st0 ./etc/passwd ./etc/group
   

   Note: when you did the backup as I described above the GNU tar command will have prepended each filename with "./"; if you weren't using GNU tar you should modify the command I listed to create the backup by inserting a cd / command before it, and changing each directory/mountpoint reference to ./ ./usr, etc. Of course, if you weren't using GNU tar then the S and l options might not work anyway. Those are GNU extensions.

5. For each corrupted/damaged filesystem:
   1. backup/copy any accessible files that are newer than your last backup.
   2. reformat using the 'mkfs' command. Use the -c option to check for bad blocks.
   3. mount that filesystem under /mnt in the same (relative) place where it would go under normal operations. For example a filesystem that would normally be located under / would be under mnt, and one that was usually under /usr would go under /mnt/usr, and one that was under /usr/local would now be mounted under /mnt/usr/local/ (see your old /etc/fstab for details, restore that to /tmp if necessary).

      Note: It may make sense to mount any undamaged filesystems read-only as part of this process ... so that the whole directory tree will appear more like you expect as you're working, but helping you avoid accidentally over-writing or damaging your (previously) undamaged filesystems. Obviously this is simpler if you're restoring to a whole new disk or system --- and are thus restoring EVERYTHING.

   4. restore the files that were on that filesystem.

If you are restoring a whole system (there were no undamaged filesystems) then you can simply use a command sequence like:

		 cd /mnt && tar xpvf /dev/st0

(after you've mounted up all the filesystems under /mnt in the correct relationship).

If you need to restore individual filesystems you'd still cd to /mnt, then you'd issue a command like:

		 tar xpvf /dev/st0 ./home ./var ...

where ./home ./var ... are the list of top level

directories below which you want to restore your files.

If you just want to restore a small list of files (you can't use "*.txt" or other wildcard patterns on the 'tar' command line) then the best method is to use a "take list." Take the "index" (table of contents file) that you generated back in step 3 and either edit or "grep" it for the list files that you want. Filter out or delete the names of all the files that you don't want. Then use a command like:

		 tar xpvTf /tmp/takelist /dev/st0 ./home ./var ...

... assuming that you stored the list of files you want in /tmp/takelist.

If you know of a regular expression that uniquely describes the files you want to restore you can use a command like:

		 grep "^\./home/docs/.*\.txt" /tmp/filelist |
		    tar xpvTf - /dev/st0 ./home ./var ...

... to get them without having to create a "takelist" file. Here we are forcing 'tar' to "take" its list of files from "stdin" (the command pipeline in this case).

I realize that all of this seems complicated. However, that's about as easy as I can make it for people using the stock Linux tools. If that's too complicated, then you might want to consider trying something like BRU (which has menu and GUI screens in addition to its command line utilities). Personally I think those are really as complicated, but some of that complication is hidden from the common cases and only comes out to bite you during moments of extreme stress --- like when your system is unusable while you're trying to restore your root filesystem).

BTW: you don't have to buy a tape drive for every computer on your network. Linux and other UNIX systems can easily share tape drives using their standard tools. For example you can use, 'ssh' (or 'rsh' if you have NO security requirements) and the 'buffer' program to redirect any 'tar', 'cpio' or 'dump' backup (or restore) to a tape drive on a remote system.

Then you can use commands like:

     tar cSlvf - / /usr /home ...  | ssh -l bakoper tapehost buffer -o /dev/st0

... to do your backups. (In this case I'm using ssh to access a "backup operator" account (bakoper) on the host named "tapehost", and I'm directing my tar output to a 'buffer' process on that remote system). Obviously there's more do it than that. You have to co-ordinate all the access to those tapes --- since it wouldn't do to have each machine over-writing one tape. But that's what professional sysadmins are for. They can write the scripts and handle all the scheduling, tape changing etc.

unable to open a initial console

... he replied ...

Jim,

The file /dev/console was missing as well as /var/log/*. I think my server was compromised by a DNS attack. I was running old version of bind. I noticed there is a directory ADMROCKS in /var/named which implies bind overflow. I upgrated my OS and things back to normal.

Thanks for the tips,
-Asghar

RE: uninstall

AnswerGang: Ben Okopnik, Jim Dennis

From erwin on Fri, 30 Jun 2000

If i want to install a package from binary source, i put the command tar -XXX foo.tar.gz, "make", and then "make install" .... What I have to do if i want uninstall that package?

[Ben] First, be a bit careful about syntax when using "tar"; for historic reasons, the '-' is not just a syntax "preceder" but a part of the syntax itself, signifying piped input. "tar xzf foo.tar.gz" would be the correct way to "untar and defeather" the package; "tar xvzf foo.tar.gz" would print some useful info while doing so.

As to uninstalling the package - this is where one of the disadvantages of *.tar.gz packages shows up: since most of them do not follow any kind of a filesystem standard or a set of install/uninstall rules (unless you're talking about packages from a standard Linux distrib), the process can range from "simple" to "I'd rather have a root canal".

Since you didn't say that you're using a package from, e.g., Slackware, which I believe has a specific uninstall procedure, I'm going to assume the worst case - that you're talking about a random tarball pulled off the Net somewhere, meaning that it could be anything at all. So, here we go...

Easy version: type "make uninstall". Some software authors have enough mercy in their hearts on people like me and you to include an uninstall routine in their makefile. If it works, burn a Windows CD as an offering and be happy.

More complex version: If the above process comes back with an error ("No rule to make target `uninstall'. Stop."), the next step is to examine the makefile itself. This can be an ugly, confusing, frustrating process if you're not used to reading makefiles - but since we're only looking for 'targets' (things like "all:", "install:", "clean:", and "uninstall:"), here's a shortcut -

grep : makefile

This will print all the target names contained in the makefile, possibly along with a bit of unrelated junk. The line you're looking for may be named something like "remove:", "purge:", "expunge:", or a number of other things - but what that target should have, as the listed action (run "make -n <target_name>" to see what commands would be executed by that option), is the deletion of everything done by the "install:" target. If you find one that fits, rerun "make" with that switch.

"Crawling on broken glass" version: if you can't find anything like that, then you have to remove everything manually. In a number of cases, I've found that the least painful way to do it is by 1) running "make -n install > uninstall" and examining the created file to see exactly what is done by that target, 2) deleting all the compilation statements ("gcc [...]" or "g++ [...]" and the like) and reversing the action of all the "mkdir", "cp", and "install" statements (i.e., "rm -rf" the created directories and "rm" the individual files that fall outside that hierarchy), and 3) running what remains as a shell script to execute those actions (". uninstall").

Of course, if the "install" target is simple enough - say, copying one or two files into /usr/bin - just delete those.

On a more general note, you should _always_ examine any makefile that you're about to run (with at least a cursory glance to see if an "uninstall" target exists): since some programs require installation by the root user, a stray "rm -rf" could cause you a lot of grief. This requires learning to read makefiles - but, in my opinion, this is a rather useful skill anyway. Using Midnight Commander to view the makefiles can be very helpful in this, since it highlights the syntax, which visually breaks up the file into more easily readable units.

... he replied ...

Thank you for information and correction I miss to interpretate syntax "tar" with preceder ("-") and without preceder. Could you explain what is the main different between command "tar -zxvf" and "tar zxvf". In many linux (linux howto..) and other unix clone articles I found "tar" command with preceder and sometime without preceder, which one a correct?

[Ben] If you examine the stated syntax carefully, you will find that both are correct: as is usual with Linux, There's More Than One Way To Do It. The dash ('-') in "tar" syntax (as in a number of other utilities) indicates "piped" input. Here are two versions of a command line that performs the same operation:

tar xvzf foo.tgz
gzip -dc foo.tgz | tar xv -

The differences are the following:

1. In the first case, "gzip" is invoked by "tar", via the "z" switch; in the second case, it is used explicitly. As I understand it, "tar" did not originally have this capability - this may explain why some folks would use the second version (i.e., a habit from previous usage). As well, I believe that a number of users are unaware of this "built-in decompression" in "tar" - and a name like "foo.tar.gz" seems to just beg for two tools to process it...<grin>
2. The 'f' switch precedes the name of the file that "tar" should process. In the second case, since the input to "tar" is piped from the output of "gzip", '-' is substituted for 'f' to indicate this. The 'z' switch is also eliminated, since the decompression is done explicitly by "gzip".

For LOTS of further info (prepare to spend an entire evening or so), read the "tar" man page.

Basica Fascist SysAdmin's Laundry List

AnswerGang: Jim Dennis

From Edwin Ferguson on Tue, 04 Jul 2000

Hello I am hoping that you can help me , even with your busy schedule, can you tell me how I can stop my network user from running chat room programs and instant messaging programs like ICQ , Yahoo and MSN. I use a linux box as a firewall and proxy server. I am running Red Hat 6.1, is there a way to also prevent them from running Real Player and other such applications that take up plenty bandwidth. Then finally how can I actually see what sites they are visiting and in turn block out porn sites etc. Your assitance is very much appreciated.

Edwin Ferguson Technical Support

What you've presented here is the basic laundry list of the "fascist sysadmin?" You're trying to enforce an acceptable use policy based on the assumption that your users are trying to waste your bandwidth and your company's time and other resources.

You could spend a considerable amount of time tightening your packet filters, eliminating routing and IP masquerading in favor of application layer proxies, monitoring your proxy logs, installing and/or writing filtering software etc.

If you're users are motivated to break the rules and violate these policies then you'll probably find yourself in an escalating "cybercombat" with some of the more "hacker" oriented among them.

Ultimately this is a recipe for disaster.

Now, back to your questions:

Instead of making a list of all the things that you "don't want them doing" try turning it around to ask: "What services should my users be able to access?"

If all they need is e-mail, then you can block all IP routing masquerading and proxying for all the client systems. You then run a local mail server that is allowed to relay mail from the Internet. That's that! If they need access to a selected dozen or hundred external web sites, consider installling Squid (http://www.squid-cache.org) (an Internet caching deamon) and SquidGuard (http://www.nbs.at/linux/Squidguard/installation.html) (a filtering module for Squid) and define your acceptable list accordingly.

If you remain more vague about what you policies are then you'll just enough up with an ever growing laundry list. It's obviously that the list you gave here isn't comprehensive; you tossed in "and block porn sites etc" as an afterthought. That approach will grow to consume all of your time and creative energy. Be sure to explain this to your management, assuming that they are pushing on you to pursue this tack.

The bottom line is that the there are some policies that are best enforced by human means (specifically by the HR department). Otherwise it may well be that your best recommendation will read something like:

"For each user we hire one full-time armed guard. Each guard is assigned a user, stands over his or her shoulder with weapon locked, loaded and aimed at the victim's temple...."

(Of course you management might try doing some MANAGEMENT. If the users are busy with their work, and if the management has reasonable productivity metrics and sane methods for monitoring behaviour --- then abuses of your precious bandwidth should be relatively limited ... unless management is spending all ITS time in IRC on the porno channels!).

More on TCP Wrappers and telnet Connection Delays

AnswerGang: Jim Dennis

From Hari P Kolasani on Wed, 26 Jul 2000

Hi,

I was looking at this issue:- http://tech.buffalostate.edu/LDP/LDP/LG/issue38/tag/32.html, and I did not understand your solution correctly.

Can you please let me know what I need to do in order for telnet to work without any pause?

I happen to see similar problem for FTP also.

Thanks Hari Koalsani

If you look at some of the other back issues (search on the string "tcpd" you can see that I've tried to explain the issue a few times and at great length.

Basically there are three ways to approach this:

1. Abandon telnet; use ssh instead.
2. Fix your reverse DNS zones. Make the PTR records consistent with the A (address/host) records.
3. Remove TCP Wrappers protection from the telnet service on this host. Change the line in the /etc/inetd.conf file that reads something like:

telnet	stream	tcp	nowait	telnetd.telnetd	/usr/sbin/tcpd	/usr/sbin/in.telnetd

to look more like:

telnet	stream	tcp	nowait	telnetd.telnetd	/usr/sbin/in.telnetd in.telnetd

Personally I suggest that you use both methods 1 and 2. Use ssh, which USUALLY doesn't use tcpd or libwrap, the library which implements tcpd access controls, AND fix your DNS zones so that your hosts have proper PTR records.

As I said, I've written many pages on this topic. I'm not going to re-hash it again. Hopefully this summary will get you on the right track. If you still can't understand what is going on and how to do this you should consider calling a tech support service (Linuxcare does offer single-incident tech support calls, though they are a bit expensive; there may be other companies still doing this), or hire a Linux consultant in your area (look in the Linux Consultants HOWTO http://www.linuxdoc.org/HOWTO/Consultants-HOWTO.html for one list of them).

They can provide hand holding services. A good consultant can and will show you how to handle these sorts of things for yourself, and will ask some questions regarding your needs, and recommend comprehensive solutions.

I would ask about why you are using telnet, who needs access to the system, what level and form of access they need, etc. I can simply answer questions, but a good consultant will ask more questions than he or she answers --- to make sure that you're getting the right answers. Given my constraints here, I don't have the luxury of doing in-depth requirements analysis for this column. (Also note that I'm not currently available for consulting contracts, Starshine Technical Services is currently in hiatus).

Linux in a Windows NT Domain (under a PDC)

AnswerGang: Jim Dennis

From Maenard Martinez on Tue, 25 Jul 2000

Is it possible to connect the Linux Red Hat 6.0 (costum installed) to the network wherein the PDC is a Windows NT 4.0 Server? Do I need additional tools to connect it? Is it similar to UNIX X-windows?

Thanks, Maenard

Basically all interoperation between Linux (and other forms of UNIX) and the Microsoft Windows family of network protocols (SMB used by OS/2 LANManager and LANServer, WfW, Win '9x, NT, and W2K) is done through the free Samba package.

Normally Samba allows a Linux or other UNIX system to act as an SMB file and print server. There are various ways of getting Linux to act as an SMB client (including the smbclient program, which is basically like using "FTP" to an SMB server, and the smbfs kernel option that allows one to mount SMB shares basically as though they were NFS exports).

Now, when it comes to having Linux act as a client in an MS Windows "domain" (under a PDC, or primary domain controller) it takes a bit of extra work. Recently the Andrew Tridgell and his Samba team have been working on a package called "winbind." Tridge demonstrated it to me last time he was in San Francisco.

Basically you configure and run the winbind daemon, point it at your PDC (and BDCs?) and it can do host and user lookups, (and user authentication?) for you. I guess there is also a libnss (name services selector) module that is also included, so you could edit your Linux system's /etc/nsswitch.conf to add this, just as you might to force glibc linked programs to query NIS, NIS+, LDAP or other directory services.

Now I should point out two things about what Tridge showed me. First, it was under development at the time. It probably still is. You'd want to look at the Samba web pages and read about the current state of the code --- but it may not be ready for use on production systems. (I hear that some sites are already using it in production, but basically that's because it's their only choice). The other thing I should mention is that I got the basic "salesman's" demo. That's not any fault of Tridge's (he wasn't trying to "sell" it to me and he certainly can get into the technical nitty gritty to any level that I could understand). It's just that we didn't have much time to spend together. As usual we were both pressed for time.

(I'm writing this on a train, which is why I can't look for more details at the Samba site for you. So, point your browser at: http://www.samba.org for more details.

automating windows telnet to linux

AnswerGang: Ben Okopnik

From Mike Miller on Sun, 23 Jul 2000

Hi, I'm having a bit of trouble trying to figure out a way to automate my dial up process. Say I'm sitting here at my hewlett packard and I want to get on the internet.... I have to open a telnet window, logon as root on my linuxbox, and type ppp-go. I already have a script for my isp login name and password. Is there program out there that would possibly open a telnet window, type root and password, and enter ppp-go, sort of a dial on demand? Also, is there a way to disconnect from my isp from my hewlett packard without opening telnet and using ppp-off?

Short answer: IP-Masq and "diald". The man page and the HOWTO were on the NY Times Bestseller list for 16 weeks straight. <grin>

OK, let's see - you don't say what your setup is; for that matter, we have no information whatsoever, other than reasonable guesses. From these clues, I gather the following: you have a Windows box (3.1? 95/98? NT?) connected to a Linux machine on a local network. The Linux box is the one with the connection (ISDN? Dial-up? Telepathic?) to your ISP. If this is correct, then the explanation to follow may be of use; my main reason for answering this is that it's a relatively common setup, and other people may find it useful as well.

The first thing that's needed is IP-Masq and SLIP compiled into your kernel; depending on your distro and version, it may already be done. IP-Masq is a NAT (Network Address Translation) program; what it does, in effect, is make your LAN look like a single IP address to the "outside world", i.e., no matter which machine you use to surf, telnet, etc., all requests will come from (and all replies will be sent to) your IP-Masq router, which will then route the traffic inside the LAN.

"Diald" is a 'dial-on-demand' daemon (that requires SLIP) that will establish a connection to your ISP whenever you request an "outside" IP - i.e., if you fire up Netscape and ask for www.slashdot.com, "diald" will see that the address is non-local and establish a connection by dialing up. It will also, if you want it to, disconnect automatically after a period of inactivity.

What does this mean in practical terms? You never have to think about dialing from either of your machines again - just open your browser and start surfing, or telnet to anywhere, or ping at will. The first response will take 30 seconds or so (the period required for the dial-up connection), but that's it. As automatic as it gets.

The IP-Masquerading HOWTO (sorry, no URL - I'm writing this at sea, and don't have access to the Net) takes you step-by-step through the process of setting up IP-Masq, and the "diald" man page and documentation are very detailed, with lots of examples for various situations.

Telnet Clients for Windows and Linux

AnswerGang: Heather Stern

Hello Answer Guy, or Gang perhaps,

I would like to ask your help on something that's been bugging for some time.

I work in a company where Windows and Microsoft in general are the standard for the desktop and I more or less manage to survive the daily routine (Windows 98 only crashes a few times a week, which is a big improvement over Windows 95). However, for my technical support activity I use two Linux boxes, old 486s recycled because no Windows 9x would run on them, at least not without reducing productivity to 10%. I'm very happy with them and I just couldn't not do without them. One runs Slackware 3.4, the other Debian 2.2.

The only problem I have is to telnet into them from my Windows machines (as this is an internal network I don't need to use SSH and similar). That is, any telnet client works fine but whenever I need to use applications like Midnight Commander (wonderful tool) or even VI, some keys, namely function and navigation keys, do not work. The test I normally do when I try a new client is to run MC and try all the function keys. I have tried the standard Windows client, Netmanage, and several others. The only client that somehow achieves about some success if the new CRT 3.1, from Van Dyke, www.vandyke.com. It has a Linux terminal and keyboard type and with it I can use F6 to F10 with no problems but F1 to F5 seem not be working at all. I have tried all the different combinations, like VT100 terminal and Linux keyboard, and so on (for some obscure reason F5 does not work at all, with any client).

Teraterm is trainable. (As a side note it also has an ssh add-in available.) You might also try whatever Hummingbird offers for telnet services, they have been doing terminal emulators for a l...o...n...g time, and of all the possible results you should be able to pick one on your side, and a matching TERM variable on the Linux side.

But it's worth noting that there are big stacks of vt-something terminal types. When I was playing with a Solaris box at one point (!) the "standard" Windows telnet behaved best if I set the term variable to "vt100-nav" (no advanced video, has some sort of effect on the way it handles the last screen column). You probably want to try a bunch of the TERM variables anyway, because lame little telnet announces itself as "ansi" but isn't close enough to that spec either. For that matter, the telnet that comes with it also offers vt52 emulation, and you can try that ...

The Keyboard HowTo does not say anything on this issue, so I wonder whether you had any information you may be willing to share.

There's no reason something about remapping the Linux console driver's idea of keys would have any effect whatsoever on a remote connection (whether ssh or telnet)

I know you do not use Windows since several years but maybe you have come across this problem in the past.

Best of luck with it; if you need to keep looking for a configurable enough client, try winfiles.com or Tucows.

Anyway, thanks a lot for your help and should you need any additional information, please feel free to contact me at any time.

Best regards.

ROBERTO URBAN

... he replied ...

Heather,

Thanks for your quick response. I'll act on your information right away. Thanks again.

Best regards.
ROBERTO URBAN

Port 80 Telnet

AnswerGang: Srinivasa Shikaripura, Mike Orr

From Nick Adams on Tue, 11 Jul 2000

Hello, Quick question. I want to change my port to accept telnet connections to port 80. This enables me to connect from behind my proxy at work. How do I do this? Thanks,

Nick Adams

[Sas] hi,

If I understand your problem, "you want to telnet to your personal machine which is behind a http proxy, from outside the proxy network".

My quick answer would be it is not possible.

If you are behind a http proxy, then you can't connet to your machine using telnet from outside. Since proxy talks only in HTTP protocol, your telnet clint from outside wouldn't be able to talk to your machine through it.

Coming to other part of the question on how to make the telnetd accept telnet connections on port 80, you may need to modify your '/etc/services' and /etc/inded.conf'.

Hope that helps.

Cheers, -Sas

[Mike Orr] There exist telnet-via-web applications, but they have to be installed on the host (i.e., proxy) machine. I've never used them, so I don't know anything more about them.

[Sas] Thanks for the info.

I agree with you that with custorm programs to handle Telnet proxy we could telnet over proxy. But with a standard apache/Netscape/IIS proxy web server it is not possible. Also, the proxy admin needs to install and enable corresponding telnet port to outside world, which may be risky.

Here is one server which does telnet proxy:

http://www.nabe-intl.co.jp/faqs/telfaqs.html#tel001

Just FYI.

-Sas

Connection Refused

From Yu-Kang Tsao on Wed, 26 Jul 2000

AnswerGang: Jim Dennis

Hi James:

Now I am setting up a linux red hat 6.2 server box in our NT LAN and I am trying to telnet connect to that box from one of the NT workstation in our NT LAN. But it gives me connectiong refuse message. Would you help me telnet connect to linux box ? Thank you very much.

Sincerely
Nathan

You probably don't have DNS, specifically your reverse DNS zones (PTR records) properly configured.

Linux includes a package called TCP Wrappers (tcpd) which allows you to control which systems can connect to which services. This control is based on the contents of two configuration files (/etc/hosts.allow and /etc/hosts.deny) which can contain host/domain name and IP address patterns that "allow" or "deny" access to specific services.

You could disable this feature by editing your /etc/inetd.conf file and changing a line that reads something like:

telnet stream tcp nowait telnetd.telnetd /usr/sbin/tcpd /usr/sbin/in.telnetd

to something that looks more like:

telnet stream tcp nowait telnetd.telnetd /usr/sbin/in.telnetd /usr/sbin/in.telnetd

(Note: THESE ARE EACY JUST ON ONE LINE! THE TRAILING BACKSLASH is for e-mail/browser legibility) some of the details might differ abit. This example is from my Debian laptop and Red Hat has slightly different paths and permissions in some cases).

You should search the back issues of LG for hosts.allow and tcpd for other (more detailed) discussions of this issue. It is an FAQ. Of course you can also read the man pages for hosts_access(5), hosts_options(5) and tcpd(8) for more details on how to use this package.

Note: You should also consider banning telnet from your networks. I highly recommend that you search the LG back issues for references to 'ssh' for discussions that relate to that. Basically, the telnet protocol leaves your systems susceptible to sniffing (and session hijacking, among other problems) and therefore greatly increases your chances of getting cracked, and greatly increases the amount of damage that an intruder or disgruntled local user can do to your systems. 'ssh' and its alternatives are MUCH safer.

... he replied ...

Hi Jim:

I also want to thank you for advising me ban telnet from my networks. I will ban telnet from my networks. Thanks a lot.

Nathan

Loadlin trouble

From sarnold on Fri, 07 Jul 2000 on the L.U.S.T List

AnswerGang: Jim Dennis

On 3 Jul 00, at 18:07, Number 4 wrote:

I've just installed Loadlin on my Win98 partition and can't get it to boot my kernel (bzImage type). When I try to load the kernel, with all of the proper parameters set, it gives an "invalid compressed format" error message and the system is halted. I think the problem is that when I copy the kernel onto the windoze

partition, it is automatically converted from Linux binary format (two-digit hex numbers in brackets) to DOS binary format (many weird ASCII characters). Does anyone know how to remedy this? Thanks.

I have no idea what you think you're saying in this last statement. Binary is binary. If a file is copied as a stream of binary octets (bytes) than it will the same file on any platform that supports 8-bit bytes. There is no "Linux binary" vs. "DOS binary" (in terms of file formats). Of course the executable binaries have much different formats (in fact Linux supports a.out, ELF and some others, while MS-DOS support .COM and .EXE).

However, the Linux kernel is not "executed" by DOS. It is loaded LOADLIN.EXE (which obviously an MS-DOS .EXE binary executable file). However, the kernel image is generally a compressed kernel image in ELF format with a small executable stub/header. It is formatted so that it could be dropped unto a floppy and directly booted (so the first sector of a Linux kernel image is basically just like a floppy boot sector). Other loaders (like LILO, SYSLINUX and LOADLIN.EXE copy the kernel into memory and jump into a different entry point (past the "boot record" and unto the part that allocates extended memory and decompresses the kernel into it)

I hope you can see that your characterization of "hex digits" vs. "weird ASCII characters" is hopeless confused. Those are both different ways of viewing or representing the same binary data. The fact that they appeared to be different is probably an artifact of the tool you were using to view them. To actually tell if the file was modified as it was copied, use the the cmp (or at least the diff) command and check its return value.

If the files are different then look to see if you have your FAT/MSDOS filesystem mounted with the "convert" options enabled. This was a feature in earlier Linux kernels that applied to some of the FAT, VFAT, and UMSDOS filesystems. I think it as been dropped from more recent kernels (or is at least depracated). It was intended to automatically convert TEXT files as they were copied to or from MS-DOS compatible filesystems. However, it is known to have caused many problems and the consensus in the Linux kernel community seems to be that kernel filesystem drivers should NOT modify the contents of files as they are stored or retrieved. (I'm inclined to agree --- let the applications be modified to handle the format differences gracefully).

Of course the TEXT file formats differ among UNIX/Linux, MS-DOS, and MacOS systems. It all depends on the line termination conventions. Linux/UNIX use just "newlines" (LF, linefeed, a single character: ASCII hex 0x0A, '\n' in C strings) while MacOS uses just the carriage return (CR, ASCII hex 0x0D, '\r' in C) and MS-DOS uses the highly irritating CRLF (2 characters: carriage return, line feed, ASCII hex 0D0A sequences, or "\r\n" in C). I've seen some MS-DOS editors freak out when presented with text files that had LFCR line boundaries (reversed CR and LF sequences). However, most of them could handle that and some/most could handle UNIX and Mac style text files.

(Of course most GNU and free text editors and tools can handle any of these formats and there are many little scripts and tools to convert a text file into any of the appropriate formats. Some day, someone ought to write a really top notch "text file" library that automatically detects the line feed convention on open and defaults to preserving that throughout the rest of the operation --- with options to coerce a conversion as necessary).

(The reason I say the MS-DOS form is so irritating is that it messages with the sizes of the file. Having two character line boundaries then breaks quite a few other assumptions about the text of the file.

I don't use loadlin, but I think the 2.2.x kernels need to be bzipped. 2.0.x kernels use the older compression; you could try an older kernel, or maybe boot your kernel off a floppy disk. Is there some reason why you can't install to an e2fs partition and use lilo?

Sorry, that's about all I can think of (on the last morning of a holiday weekend

Steve

Older versions of any loader (LILO, LOADLIN.EXE, SYSLINUX) my not be able to handle bzipped kernels. However recent versions (as in the last two or three YEARS) should be able to cope with them.

I suspect that it is more likely that he's corrupting the kernel image as he's copying it.

Linux vs. MS Exchange for Mail Server

From sas on Sun, 16 Jul 2000

AnswerGang: Jim Dennis

Christine Rancapero was published in the Mailbag:

Do you have an issue regarding the advantages and disadvantages of migrating linux mail server to an MS exchange? Your help is gratefully appreciated....thank you very much

One of our more active readers this month replied -

Advantages of moving from Linux mail server to MS Exchange:

1. Improves MS revenue, there by improving its financial status (very crucial after the DOJ battle)
2. When ever there is a "Mellissa" or "I LOVE YOU" virus, MS Exchange get clogged for 3 days and you could enjoy vacations, long weekends, frequently. (Anyway there will be MS to show finger at!)
3. You could have the pleasure of raising invoices for Pentium IV (V, VI, which ever is latest), 1 GB main memory, Windows 2000 systems and I tell you it is a good administrative experience... ))

Dis-advantages of not moving to MS Exchange:

1. I have been on Netscape, IMAP, *nix mail for 2 years in my company and have accessed it from all sorts of environments and locations (dial-up, international) and had no problem with it. Bad luck, I couldn't tell my manager why I couldn't complete my assignments (only if it were to be MS Exchange!)

[Disclaimer: No hard feeling please. It is not a flaim bait.

Just my experience with *nix mail and my colleguages experience with MS Exchange]

cheers -Sas

All humor aside this would not be so much of a "special issue" (of LG) as a white paper. Here are some thoughts:

Linux (and free software) vs. MS NT + MS Exchange for E-mail

The first observation to make is that we are comparing apples to fish heads. Linux is an operating system kernel. There are many packages that can supply standard mail services under Linux. Basically the UNIX/Linux e-mail model involves MTA (mail transport agents), MSA (mail storage/access agents) and MUAs (mail user agents). There are also a variety of utilities that don't really quite fit in any of these categories.

[ Our LG Editor, Mike, thought Jim's next part describing an overview of Linux mail services was so good, he split it into a seperate article: http://linuxgazette.net/issue56/dennis.html

Summarized: there are a several MTAs, a number of ways to apply administrative policy -- more complicated policy takes much more planning. You can also get the LDA (local delivery agent) involved, and apply rules or filters at the email client level. This certainly includes responders such as the common 'vacation/out of office' note. With shell scripts invoking small utilities, certain kinds of recovery are easier on the sysadmin; small utlilities for the user (like 'biff' to spot new mail) exist too. Goodness knows what mail client the user may have - he has so many choices.

-- Heather. ]

This is all in contrast to Microsoft's approach. With Microsoft you are almost forced to use the MS Outlook client, and the MS Exchange server. They referred to that as "integrated." They also basically require that you use their "Back Office" for and "SMS" products for some management features, and their WINS (or the newer ActiveDirectory?) for directory services.

One of the costs of all this integration is CONTROL. You must set up your network, your routers, and your servers in one of the approved Microsoft ways in order for any of to work. You can't have one "farm" (cluster) of servers (say outside your firewall, possibly with some geographic dispersion) recieving and relaying mail with another cluster of servers (say inside your firewall, at specific regional and departmental offices. You can't make your e-mail address names follow one convention (abstraction) such as "[email protected]" while the actual underlying routing and storage archictecture follows a different model (such as [email protected]).

The UNIX/Linux model is scalable. That's proven by the fact that it's used by well over 80% of the Internet (obviously the largest interconnecting set of computer e-mail networks in history).

As usual if the Microsoft package doesn't do what you want you'll have to do without. There is very little option for administrators and users to customize the operations. Even if you do try to customize your Microsoft installation their internal complexity, tight coupling (integration) and overall fragility result in steep learning curves, and high risks (the packages you add in are more likely to conflict with other, seemingly unrelated, parts of the system or with other subsystems).

Obviously with the Linux tools there are no arbitrary limits placed on number of users, number of accounts, number of sent or received messages, sizes of messages, etc. While some specific tools may bump into limits, more often the default configuration, or the wise administrator, will impose constraints based on their own capacity planning needs and their own policies. (Like when I modified my sendmail.cf to set limits after the incident I described above).

With the Microsoft approach you're required to pay for every user; and those costs will probably become ANNUAL expenses (as Microsoft foists thier ASP software "subscription" model on their customers).

In addition, of course, the Microsoft approach emphasizes the convenience for their programmers and the needs of their marketing people over the security of your users. That's why we are regularly treated to the perrennial debacle of the e-mail macro virus epidemics (Melissa, ILOVEU, LoveBug, etc). These macro viruses are basically caused by the very same programming flaws that gave us the WinWord and Excel Macro viruses (and they are written in basically the same language). Similar bugs seem to have been found in Explorer.

Microsoft thrives of shallow whizzy "features" and one of the easiest way to implement those is through poorly designed obscure "dynamic content" hooks which treat "special" data as programs. Those are precisely the kinds of "features" that are most attractive to cybervandals and most easily exploited. Once they've been put into a system and used by other components on that system then they can't be removed or disabled (all in the name of backwards compatibility).

Of course that hallowed "backward compatibility" will only be honored to the degree that suits Microsoft's whims. They will deliberately or neglectfully break their APIs in order to force users and ISVs (independent software developers) to upgrade existing products as a requirement to upgrading other (seemingly unrelated) subsystems.

Thus an upgrade to the latest Powerpoint may entail an upgrade to the rest of MS Office, which may require upgrades to the OS and thus to the mail client (Outlook or Express) and thence possibly right up to the mail server (Exchange) and the server's OS (NT to W2K). Microsoft generally benefits from such domino effects; though they do have to exhibit some restraint. That's particularly true since they have enough trouble getting any single product to ship on schedule and they can't try to sync them all for really massive coups.

This is another cost of integration. The "integrated" systems become rigid and hard to maintain, harder to upgrade or enhance, impossible to troubleshoot or repair.

Open systems are characterized by modularity --- separate components interacting through common APIs (sometimes via shared libraries), and communicating via published protocols. Open systems generally have multiple combinations of clients and servers. Of course that has its cost. Some of these components will fail to implement their protocols in interoperable ways some of the time. Sometimes this will require revisions to the protocols, more often to the components. Some combinations of components will not work, or will be a bad idea for other reasons. Often the same functions will be implemented at multiple different points (duplication of feature sets).

Overall these systems will be more robust, more resilient, and more flexible. It will be possible for an organization to tailor their system to meet their needs.

Such systems do require skilled, professional administrators (or least consultants for the initial deployments, and for follow-up support). However, the "easy to use" MS Windows based systems, and even the famed "intuitive" MacOS networks required trained professionals for most non-trivial networks.

Ultimately you should consider the availability of expertise in your IT decisions. Hire people with broad experience and a willingness to learn. Then ask them what systems they prefer to manage.

Copyright © 2000, the respective authors

Collection Copyright © 2000, Linux Gazette
Published in Linux Gazette Issue 56 August 2000

HTML transformation by Heather Stern of Tuxtops, Inc., http://www.tuxtops.com/