Meanwhile, if you're the sort who bristles about binary-only drivers going into your otherwise trustable kernel, look out for VIA's "support" for the MPEG2 hardware on their EPIA boards, too. -- Heather

up2date SSL error

In Redhat 7.1 thru 9 running up2date fails with this message:

SSL.Error: [('SSL routines', 'SSL3_GET_SERVER_CERTIFICATE', 'certificate verify failed')]

Then it tells you your system's clock may be so wrong it is causing the problem. What is really going on is the certificate and the up2date program is out of date. You need to download the lateest up2date from Redhat.

RHSA-2003:267 for Red Hat Linux: https://rhn.redhat.com/errata/RHSA-2003-267.html

This solved it for me. -- Greg Anderson

[JimD] Might be good to try an rpm --rebuilddb command, too. Just in case rpm segfaults on a corrupted dbm/database.

Snapshot fo current window or desktop

How do I get a snapshot (preferably in gif or jpg format) of the current window in KDE? What I am looking for is the equivalent to MS's Alt-PrtScn then paste to paintbrush and then save to a file.

TIA -- Wes Hegge

[Dan] ksnapshot

Thanks,

I Guess, I am blind as a bat. Right in the "Graphics" submenu is "Screen Capture Program" (aka ksnapshot).

Thanks for the help.

[Thomas] I haven't run KDE for years (I'm an FVWM fan), but I do know that "ImageMagick" offers the "import" utility which does the same thing, as does "xwd".

[Ashwin N] Can be done using Gimp. From the menu choose,

File -> Acquire -> Screenshot

[Heather] For the ImageMagick method I keep a directory named prn and if I like what I captured I rename this image. This happens to also be where I keep documents I only have around to be printed. Here's my bash alias I've been using for awhile:

# capture an X display
# with thanks to the lazy folks at
# http://www.troubleshooters.com/linux/scrshot.htm
# who were kind enough to document how they do 'em.
screenshot ()
{
   import -window root ~/prn/screendump.png
}

Note that the man page for import is actually readable - you can take shots of specific things, not just the whole screen.

There's also an enlightenment epplet that will make screenshots.

Errors while using rpm extension..... Cygwin

Actually i am not using red hat, although it is insalled on my pc. Rather i am using cygwin (http://cygwin.com which is linus-like environment for windows platform - my OS is Windows 2000) for these rpms. It is working fine with .tar and .gz files..but giving problems with all the files of rpm format.

[Ashwin] That is the problem! I have used Cygwin before, but I didn't know that they had even ported over RPM! In any case, if your need is just for a Java SDK, you can download the one available for Windows from the Sun website. Install it and the java, javac and other commandline tools will be available to you under Cygwin. Just remember to update the PATH variable with the directory where the Java binaries are located.

Thanks for the help.........it is working this way.......

[Ashwin] For other Linux utilities that are not available from the Cygwin mirrors, you will have the best chance with the .tar.gz or .tar.bz2 files of those applications.

Cool tool of the week: nntp//rss

Hi,

nntp//rss (http://www.methodize.org/nntprss) is a RSS to NNTP so you can read RSS feeds in your favorite newsreaders. Very nice.

mult headeddsplay

Hello,

I have a NVIDA GForce 2 Go 100 card that supports multiheaded display easily in Windows. Now i want to do the same in Linux. I have RedHat 8.0. What do you suggest that I do?

[Neil] I suggest using Google http://www.google.com/search?q=nvidia+linux+%22multiple+display%22&sourceid=opera&num=0&ie=utf-8&oe=utf-8

It throws up potentially useful stuff like

http://lists.suse.com/archive/suse-linux-e/2001-Jun/att-3012/01-TWINVIEW_README

I suppose you could try the Gentoo Unreal Tournament demo CD, which won't replace anything on your computer. At the very least it's tuned up for NVidia. Find it by typing "unreal" into the search gadget at Freshmeat.Net. -- Heather

Contents:

¶: Greetings From Heather Stern

Efficiency regards running script in a subshell () and a seperate shell

force unmounting of CDROM

IP config files on Red Hat 9

a linux solution for the office

Simple DNS solution with Red Hat 9

Creating RAMDISK

X server crash when starting up RH9 for the first time

Converting from Win2k to Linux

Greetings from Heather Stern

Greetings everyone and welcome again to the world of The Answer Gang. It's been quite hectic for me and not all fun and games... among other things, I was ill last month around submission time and that meant that the TAG column wasn't submitted at all, as I missed the deadline, feeling a little too "dead" at the time. Oh well, I guess we all need a break now and then...

Thomas Adam valiantly threw a hand in to help out, and I have to admit he did more than half the work this time around. He says he's learning an awful lot about perl, too.

The peeve of this month is without a doubt a lack of information and extremely poor descriptions in the compositions of the e-mails sent in to TAG. As a positive note, there have been a record number of hits to:

http://www.linuxgazette.com/tag/ask-the-gang.html

Please, everyone - if you're thinking about asking a question, read that, and ask us what you need as clearly as you can. We understand it is difficult for those who do not speak English very well, but that's rarely been a problem - folks who are so carefully aware of their poor language skills also take a free moment, and only ask what they need to ask, and say what they've tried so far. The point is, if you can't be bothered to ask a clear question, there's far too many messages for us to try to detangle yours.

Regular and attentive readers will note some of the messy messages we have answered. Yes, there's been worse. With a question such as

d00z, cn u h3lp me

... maybe you'll get some chuckles, but you sure won't get an answer. The same goes for you students out there with a take home light quiz. We can spot those a handful of kilometers away, give or take a mile. Maybe you should cc: your professor when you ask us the question, and he can give us the passing marks in your class. The point is to learn a few research skills - so for such questions, search google. Search our KnowedgeBase - it's part of what it's here for. Search TLDP.org and freshmeat if the problem is really about Linux.

And now for what I'd really planned to say last month. I attended Linux World Expo, as I do every year here in the Silicon Valley area, with the prospect of meeting again friends in the open source world from all over the place. But this year, I was also taking a step back and a look around at how the trade show world has changed in its view of Linux. Now I get to write this with the additional perspective of having been invited to PC Expo - a more generic computer trade show.

The View From The Trade Show Floor

I can't comment on the view from the press room since all that's in there is pamphlets, a couple of spare computers, and coffee. Maybe a sandwich tray. The seminars are still seminars and the halls still suck your cellphone dry.

However, out in the exhibit hall, the world has changed a lot.

My first Linux World Expo was in San Jose. IDG had just taken over the project from a local group who wanted to create a Linux conference on the same order as regional conventions run for and by science fiction fans. IDG is a big player who runs a lot of trade shows. They invited big names and posted sponsor banners and the whole nine yards. Jim Dennis (our very own Answer Guy) was invited to speak about security for a half day class, and if the next speaker hadn't shown up, the audience would have kept them from cleaning the room so Jim could continue to talk. The hall was filled with a bunch of booths, many of them small companies, but a few names like Intel spring to mind. I recognized about half the listed speakers by face and about a third of them total would surely recognize me back. Geeks were everywhere, confused managerial and business types were too. T-shirts were plentiful.

They also made quite a splash by having a platinum sponsor pay for a bunch of floor space outright and donate it to be used as a Dot Org pavilion, where projects and linux user groups could have small kiosks and generally have a good time. Dot Org was much better laid out the second time I saw it.

As shows pressed on, toys were on the increase, but shirts amd CDs with products on them were certainly around, too. we saw an increase of booths and as more "generic" presence grew the amount of total IQ in the booths could be seen to be being split evenly, because individually talking to vendors, it was definitely going down. Toys were getting insanely cool - drawings for VW Bugs and motorcycles. I volunteered for the FSF booth. I helped out the Gnome guys. When I worked at Tuxtops I went to both LWE/SF and LWE/NY in that year and had a great time at both places. But Dot Org was becoming a ghetto with hardly any color and the main floor felt increasingly like any other computer show. Something was about to break down.

And it did. A lot of companies had picked up starter capital on the magic of the word Linux. Heck, some of them even were trying to be Linux companies. But VC enthusiasm is no excuse for a poor business plan and when something in the economy turned sour - I'm not sure what, but we'll start at asian money difficulties and work our way up from there - anyone who did not have their hand tightly on the finances watched it all start to head down the drain at tornado speed, and the VCs clamp down their pocketbooks. No more toys. T-shirts only if you sit through the spiel. (Booths that still had IQ points left around would also give them to people who seemed genuinely interested in their products, which if you ask me, is the way it should have always been.) More feet hurt because companies watching their dimes only used a single layer of carpet and not the nice padded layer under it which you can pay extra for. Lots of pamphlets though.

I'm pleased to say that I'm seeing winter's end. There are still insipidly blank faces to be foudn at some booths. And there are what I'll call the "barely Linux" booths - hardware vendors selling server racks, RAID arrays, GPS, scanning printers, and other weird peripherals. But Linux products are getting their own booths and that means they're affording them, I hope. The flavors of products are spreading around, and that's part of why it feels like a normal trade show to me - financial, games, diet plan calculators. Backup programs optimized for Linux, like Storix, are not only around but have competitors. All those things I'd see at a Windows oriented show, I'm seeing.

The Dot Org space still looks plain but it's not hiding so much anymore. There are a lot more projects than there used to be and projects were sharing booths - I recommend a little more platinum get spilled on this next time. It's rather cool to see MBNA America (a large VISA card vendor) hanging out in Dot Org plugging their LinuxFund card. They give away big beach towels if you can sign up for a card.

It's hilarious to see Microsoft booths in "the rookery" as if they're newly born open sourcerors. "Well we do let WinCE developers see some sources." Really? Can they compile Windows from scratch and pour it onto a handheld, then run WinCE office binaries out of the OEM private packages to test their build? "Um, no. They can look at selected drivers and structures." I see. Well, it's better than nothing

Yes, I like what I'm seeing. The frost is melting off the ground and not every seed has sprouted again - but things are getting good.

Over at the more ordinary trade show, PC Expo - for the second year running they had a "Linux Boot Camp" track. For the first year running it's calling itself "TechX>NY" instead and I wonder if they are thinking that the PC is on its way out as the only platform to run desktops or servers. Last year "the Linux track" was one lonely guy presenting nonstop between water cups and snack breaks. This year they invited LPI, Novell (well okay Ximian, and maybe they invited themselves since they're a sponsor for the show), and a few of the Answer Gang. Maybe in a few years they'll have the brass of various "Big Linux" companies knocking down their door trying to be on the speaker's list. I'm advised that our presentations were very well recieved, indeed.

On the show floor itself there was only a little of Linux. Local computing groups knew about it and chatted merrily. Product vendors knew what it was and generally whether they supported it or hadn't tested. I didn't really get "We don't support that! Our customers don't ask for that kind of stuff!" flamage which I'd been seeing a few years ago - in fact quite the opposite, one vendor told me that he'd had requests for support and they were working on it, and he kind of hoped for BSD support too, if it wasn't too hard, though it wasn't on the official roadmap. Outsourcing World (sic) mostly didn't know or care what "A Linux" was though there were a few "outsourced programming" firms I didn't think much of. Okay, the computing sector overall may still be heading downhill in places. But Linux is indeed, looking up.

Efficiency regards running script in a subshell () and a seperate shell

From Nimish kamerkar

Answered By: Thomas Adam

Hi Answer Gang,

Which method is more efficient; Running a command in () or running it in a seperate shell? Can the answer include differences in how the processes are spawned? (i.e. fork exec etc)

[Thomas] OK. This is usually application/situation dependant. If you have something like:

#!/bin/bash

echo $(whereis xterm)

What you are doing there is forcing the command "whereis xterm" to run in a subshell (denoted between ()'s). A "subshell" is just another instance of the command-line interpreter, running your program. Thus, a "subshell" in this instance means that the main shell script is its parent, i.e. assume that we called the script above "parent.sh" then when the subshell executed you'd get:

|-parent.sh
  |-----subshell

This is also known as forking -- i.e. where the process breaks off from the main caller, to form another.

"exec()"'ing a process however, means that the currently running process is replaced by the program that is to be exec'ed. Thus unlike above where-by the subshell ran under a new instance, the following script exec's itself:

See attached exec-anything.bash.txt

What this is doing is it will echo the first two lines, sleep for a second and then re-spawn itself. What you'll see is the same message above. The echo line will NEVER show, because it is after the program ($0 denotes the program's name) that is being (re)exec'ed.

My question is how are invoking a program in a subshell() and invoking it in a seperate shell different, as regards fork & exec. I am asking this as the environment inherited by both is different. I started thinking of this really because I remembered reading somewhere sometime that invoking it in () is more efficient than in a seperate shell. But as far as I can see, both processes should need fork and exec.

Only difference is in (), both the local and global environment variables are initialised, and in a seperate shell, only global environment variables are initialised. By that logic actually, the seperate process should be more efficient than the one in ()!

[Thomas] Those variables that are exported are "global" anyway, so you don't need to describe them in this way.

It depends. A subshell can be efficient if you want to ensure that a task running under another shell script is carried out to completion before the next one is executed (a good example of this would be tarring files over ssh on a pipe).

Of course, invoking a subshell program means that if the parent dies or is kill -9'ed, whatever, then the child process is also killed along with it. That is something you might want to consider.

Normally, when you write a shell script, and you want to end the script by calling another one then you would "exec" the program name, since there is no need for the shell process to fork().

If you need two independant processes to communicate concurrently then using a fork() would be best.

Hello Thomas/Answer Gang,

Thanks for the patient unravelling of the intricacies!

With Warm Regards -Nimish.

force unmounting of CDROM

From Chady Kassouf

Answered By: Thomas Adam, Mike Martin, Karl-Heinz Herrmann

Hello Answer Gang,

I backed up most of my files on CD-Rs that later on appeared to be of very low quality, now none of my CD-ROM of CD-Writer drives manage to read from them, but that's not where the problem is.

[Thomas] As long as you have (hopefully) learnt from the exercise, that's all the matters

The fact that mounting a CDROM in linux locks the drive is making a problem with these CDs.

[Thomas] How so? The whole idea of locking the drive is to do with the way the mount command works.

OK, the CDs are bad, there's no way to read them, but there's the problem of not being able to unmount them, linux will just keep on trying to chew on the bad CD, and killing `cp' will not make it give up. `umount' will either hang forever waiting for system to finish reading from the drive, or it will return but the drive will not be released. same thing will happen with `eject'

[Thomas] If you're trying to umount a /cdrom that is currently being read/write to, then what you should do is:

ps waxf

to see the upper program that is being called below "mount /cdrom", and do:

kill -9 $(pidof <program_name>)

(rude, I know). Then you'll be able to do:

umount /cdrom

I'm using RedHat 7.3, and the guilty drives are a TEAC 40X CDROM and an HP 8200Plus CD-Writer.

[Thomas] What you're describing here is not really a problem with the hardware per se but merely a gripe with the way the kernel and mount mounts a cdrom drive. Locking the drive is there to prevent the FS from screwing up, and allowing a clean change of disks via umount.

It's good to note that while rebooting the machine, init will try to unmount the filesystems but will fail on the drive that's stuck trying to read, `umount2' will kick in and retry for three times before finally giving up and letting the reboot continue.

[Thomas] Umounting drives is done for a reason. I suspect the reason why you cannot unmount /cdrom is due to Zombie processes clogging up your kernel buffer, and the kernel doesn't realised that these have effectively stopped. Usually, I have found in situations like this that a lengthy wait of 30 mins or so, does allow the kernel time to flush itself, and the locked drive is then accessible via an 'eject'.

fuser mount /cdrom

will also help you ascertain this information, as well as the classic:

ps wax

My question is twofold; first, is there a way to tell linux to give up reading and force unmounting a CDROM drive without having to use the safety pin and, hence, lose access to the drive, OR reboot he machine?

[Thomas] Yes, there is. I have done this on some severe occassions. The first thing I would try is to:

umount -f /cdrom

The "-f" flag says to mount to force umounting of it. If that does not work, then edit "/etc/mtab" and remove the entry pertaining to your cdrom drive. In case you are wondering, "/etc/fstab" holds information about drives that can and might be mounted, and "/etc/mtab" is there as a state file for those drives that are currently mounted. Editing the file in this instance is perhaps a good idea.

If you find that this is happening to you on each and every mount, try doing something like:

mount -n /cdrom

which will tell mount NOT to write to /etc/mtab. Typically I have used this on drives whereby "/" has been mounted ro, but I cannot see why it wont work here.

Second, anyone happened to have a similar situation? or might that be a hardware problem?

[Thomas] Reading my answer here will not doubt confirm that I have had experience in this sort of thing. I doubt this is hardware related, but it could be. I would need to know more information about which aspects of your cdrw work/don't work in order to help you further.

[K.-H.] Better use some of the suggestions by Thomas but it seems cdrecord does not care for the "lock" state of a CD. If you issue:

cdrecord -eject

it will simply eject the CD -- regardless if it's locked or mounted. In the best case the kernel recognises the media is gone and the errors given back to "cp" causes all concerned processes to stop. Then a umount should also be possible. Worst case you get a kernel panic for a "damaged" filesystem on the now nonexistant CD (didn't happen here, mostly it's recovering gracefully).

[MikeM] final issue could be a process called fam locking the drive - sometimes if you kill fam, you can then umount the drive.

[Thomas] Indeed, Mike. I had considered this, but then I realised that FAM isn't always loaded on some machines.

FAM (File Alteration Monitor), is also debian package and an absolute cow to compile. It is not distro specific, no. It is used to monitor directory/file changes, so I can see how it might be used in this instance but it is unlikely.

I tried all the options presented by the kind people in this list, the only one that worked though was the cdrecord -eject although it took about 3 minutes to succeed. Thanx for K.H for the solution.

I was able to capture the error that was printed out after a file copy in KDE started to choke:

scsi0: ERROR on channel 0, id 0, lun 0, CDB: Request Sense 00 00 00 40 00
Info fld=0x1a4f7, Current sd0b:00: sense key Medium Error
Additional sense indicates No seek complete
 I/O error: dev 0b:00, sector 431068

[K.-H.] read error from the CD, "Medium error" as you already assumed in your original mail. Probably nothing to save anymore. I too had some CD's which came through the file comparison right after burning. Some time later they gave me plenty of read errors. In my case the read attempts at some point broke off but as I tried to get as much data back as possible I made the reading process to try harder.

IP config files on Red Hat 9

From - E J -

Answered By: Mike Martin, Jay R. Ashworth, Jim Dennis, David Mandala, Kapil Hari Paranjape, Ben Okopnik, Thomas Adam

Please note I am trying to configure IP on a Red Hat 9 system. I would like to see about changing from the default DHCP configuration to a STATIC configuration.

I have tried a couple of things only to trash a test system, in one case I got the system up - but X no longer worked.

I know the hostname is saved in /etc/sysconfig/network.

What are the other files necessary to changing the configuration?

Thanks in Advance.

[MikeM] first of all, for any problem try to do it manually so

once logged in (as root) do

ifdown eth0 ifconfig eth0 address <your ip address> then if you access a router directly do route add gw <name of your default gateway>

If this works add to /etc/hosts a line <your hostname (machie name)> <your machines IP address>

also this may be useful

http://www.faqs.org/docs/linux_network

[jra] But don't worry about that.

Use the RedHat program netconfig, which has a button for DHCP. Turn it off, and fill in the blanks.

[Breen] That'll work, Jay. If EJ wants to know what's going on behind the curtain, the documentation is in

/usr/share/doc/initscripts-<version>/sysconfig.txt

The subject of this thread then changes.... -- Heather

[Breen] /usr/share/doc/initscripts-<version>/sysconfig.txt

[JimD] I felt stupid reading that, too. I've complained for years at the lack of documentation for those. Sometimes I just read through the rc scripts that source the /etc/sysconfig files to figure how which names they want to use. Sometimes I just run the silly GUI config tool (when I can remember which name it goes by in which versions of Red Hat). I had never found this particular file.

I still think it's a bug. Red Hat should include a set of comments in each /etc/sysconfig file that list all of the valid variable names and show examples of the valid values.

Now that the "Red Hat Distribution" is a "community maintained project" rather than a shrink wrapped product ... heck with it; I'll still recommend that advanced Linux users switch to Debian and that businesses pressure Oracle and other ISVs to support Debian.

I think the "Red Hat community project" should set a goal for itself -- to provide a transition to Debian over the next several releases. They can write rpm command wrappers around apt-get/dpkg, add or help Debian add the -V and -K (verification and GnuPG key package signing) features to the .deb package format and the /var/lib/dpkg "database"), etc.

[DavidM] Gack, I hope that never happens, I dislike Debian intensely. I use it on some of my servers and I grate my teeth every time I need to do something with it. Debian needs a lot more then just a rpm wrapper. I'll try not to start a religious war here and simply say you stick with Debian and I'll stick with Red Hat and I hope they get more compatible but I hope that happens by Debian getting closer to Red Hat and not the reverse (except apt-get, that is quite nice).

[Ben] Yeeew. I sure hope that never happens either. I really hate the "papa knows best, just watch the blinkenlights" approach of RedHat, and money is what it takes to get me to play with it these days.

Fortunately, none of this is an issue since we _do have both of these distros plus a whole lot of others, which seems to satisfy a broad range of us picky techie types (say _that 3 times fast with a mouth full of marbles, and your dental problems will be a thing of the past.)

[JimD] In other words, I think a convergence is in order. Let's have both major "community projects" try to consolidate (improving both).

[DavidM] I strongly disagree here. The beauty of Linux and the various projects is there is no "one true way" and those who like it different can have it that way. Otherwise you are just like Microsoft and Apple each having a single "true way".

[JimD] I have similar opinions on gentoo; why create a new distibution when they could have poured that energy into creating a Debian build system that could build your entire distribution from sources, with locally defined optimization and other flags). (BTW: the argument that this results in significant performance gains seems to have been somewhat refuted:

 http://articles.linmagau.org/modules.php?op=modload&name=Sections&file=index&req=viewarticle&artid=227

... though it's possible that this test was flawed.

(That said I've been thinking of installing gentoo on a machine just to see what it's like).

[JimD] I knew my statement was likely to rankle. However, without it degenerating into a religous war, what specific things do you dislike about debian? /etc/sysconfig/network-scripts/ifcfg-* versus /etc/network/interfaces? The installer? Lack of Kickstart? The dpkg commant? dpkg-reconfigure or dpkg-repack? The fact that many packages ask configuration questions during pre-install or post-install? The fact that you can't tell the package manager to exclude the docs during installation?

I want to know because I'd like to see Debian make the necessary changes or options available to make Red Hat users happier.

[DavidM] This is important in the business world where you need 50 machines exactly the same, not close but exactly the same. Also unless you make your own Debian depository the changes in the Debian tree make it all but impossible to do this if you are installing machines over the course of several months.

Another irritant is that Debian package maintainers patch the packages to run in funny places and have odd patches on them. We were running SAMBA on one machine and we were having a major problem that we could not explain. As luck would have it I was able to get tridge to take a look and even he was mystified for several hours. As it turned out the package maintainer had really messed up the package. There is not enough testing done on the edge cases by the Debian developers, so either leave the source alone or truly understand it and test all possible uses of the package which in the case of SAMBA is impossible, the SAMBA developers have spent a huge amount of time getting donated machines available to be able to test the edge cases, there is no way a Debian developer is going to have access to the same resources that the SAMBA developers have.

At tridge's advice and this is advice I've been given many times from others as well. "Use Debian as a base install but for packages you really care about make your own from source as you never quite know what they've done to them." The problem with this of course is if I can't trust what I'm installing then why am I installing it at all?

Followed by the fact that stable is most always too old thus forcing me to use testing which gets broken at times.

[DavidM] I don't think it's possible for Debian developers to make Red Hat users happier, they like to do things their way and the hell with the rest of the folks. An example, change the init levels to match Red Hats, make init 3 the normal run level sans X and level 5 with X. Virtually every other Linux distro is the same as Red Hat very very few are different. But as a business software developer I need to know that if my package it to be installed on both Red Hat and Debian I need to install my startup scripts in different run levels or bad things happen. That is silly.

I agree that one does need to minimize gratuitous differences so tackle the run level differences first, if you can get that changed (which I strongly doubt) perhaps there is a possibility that Debian can become a stronger player in the Corporate market place. What I think you will get is many reasons why Debian is correct but that is irrelevant, it's not about being technically correct it's about minimizing gratuitous differences, and I don't think it's possible for the Debian community to change.

[JimD] Well "funny places" is the main sort of gratuitous change I also object too. FHS was supposed to reduce this (and probably has, somewhat).

We could argue that Red Hat and SuSE put their stuff in "funny" places; then we could both reach for copies of the FHS to bolster our arguments.

I agree about the runlevels and the way the run xdm as an rc script rather than in inittab. Those are gratuitous and Debian should bow to the more widespread convention in both cases.

As for the Samba anecdote: so what. Red Hat applies over a 100 patches to their kernels and they apply patches to many packages, including core packages like Samba. We could exchange package maintainer horror stories for hours. Usually the Debian packaging is better than Red Hat packaging. Debian has a published policy that is derived from the consensus of it's maintainers/developers. There are ongoing discussions on how to change that policy. Maintainers that don't conform to the policy see NMUs (non-maintainer uploads) of their packages and are eventually replaced by new maintainers.

Of course it's possible for them to change. It is easier for us to change Debian than it would be for us to change Red Hat Inc.

[DavidM] Jim, here again we differ, I feel it's easier to change Red Hat all it takes is cash, nothing more nothing less. Changing Debian is akin to herding cats you have to get a large portion of the entire development community into agreement which is damned hard if not impossible to do. You need to join them in overwhelming force in order to out vote them on policy. Any single person is doomed to failure, it was designed that way and it works quite well. Debian is a hackers play land, it was designed to be that and that it will stay, the bylaws they live by are specifically designed to keep it that way and human inertia and apathy will tend to keep it that way.

How long did it take to fix the networking init script that would only start networking and not stop or reset it take? More then 3 years by my count, if you can't fix something like a simple init script how you expect to fix a fundamental difference in philosophy?

[JimD] To effectively change Red Hat Inc. we have to either buy them, or represent a large enough portion of their revenues that they'll listen. To change Debian all we have to do is join them, work on the project and explain our reasons for recommending the change.

My problem with the Red Hat "distribution" is that it's been essentially abandoned by Red Hat Inc. Now, the community will maintain it; RH Inc. will take parts that they like and fold them into their proprietary AS/ES products. However, RH will also make changes that may make AS/ES diverge somewhat from the RH distro.

My comments are mostly intended to discourage people from create yet another distro for broad consumer use.

I'm not saying everyone should just abandon the RH distro -- just try to transition some parts to the Debian base (where that makes sense) and to join Debian in sufficient numbers and quality to effect changes there.

Notice that I'm mostly a proponent of changing Debian in many of the specifics. The one thing that I would push on the other distros is the use of the same package/dependency infrastructure and granularity. I'm not talking about the commands used to manipulate the packages (apt-get, dpkg, aptitude, rpm, etc) --- I personally would prefer to see a command named 'pkg' taking arguments somewhat like the rpm command (except for that idiocy of -U being "upgrade" and install thus necessitating the odd -F --freshen switch to mean just "upgrade"). The difference would be that pkg -i foo would look at your local policy file (/etc/apt/sources.list) and know how to fetch the foo from your preferred location or media.

[DavidM] That is not the only problem, as far as I can tell, when a package changes (is updated) there is no history, the old package is removed and the new inserted replacing it. I could be wrong but the gents I used at my last place were dedicated Debian devotees and they could not find where "outdated" packages were stored. Thus if a package was "updated" you were screwed and could not make an identical from that point forward unless you went to the extreme of maintaining your own Debian depository.

[JimD] If it happens at all I expect it to take several years and releases of each of the two distributions. We've already seen lots of people implementing apt on RPM based distributions. The problem with that superficial approach is that the underlying granularity and dependency cycle problems remain. The real value of Debian in in that acyclic dependency tree.

[Kapil] Joining the distro wars once more ...

Just some random points.

1. While Kickstart is a great idea, systemimager is not a bad replacement. For the more hardcore debian-ers, Kickstart seems too complicated when compared with "dpkg --get-selections" in combination with "dpkg --set-selections".

[Kapil] 2. When comparing "rpm" with "dpkg" (or deb) what is lacking in dpkg is the automated signature/md5sum checking. On the other hand try to unpack an "rpm" on a non-RedHat machine...not all that automatic (you need to install "alien"). To unpack a "deb"...you will (most likely) find "ar" on any machine and so can unpack deb's quite easily.

[Kapil] 3. I don't dislike RedHat as intensely as David seems to dislike Debian! However, the choice was made for me by RedHat when they decided to exclude text-mode tools for installation and management. Even "dselect" is better than nothing except "rpm". I used "purp" on RedHat systems for a while but it was always in "contrib" and often out-of-date.

[Kapil] 4. Re: xdm. I think all daemons were "meant to be" started from rc scripts so in fact the RedHat policy of starting "xdm" from inittab is strange. For example, in the old days it was possible for the console user to somehow "kill" the xdm/gdm/kdm on RedHat (possibly by killing the X server or something) causing all external Xterminal users to be logged out!

[Kapil] 5. Finally, about building things from source. To some extent I agree. It would be nice if we had the following set-up:

A. All the system administrator had to do to was install a base system.

B. Most utility/application packages could be installed by users in (say) /opt/username/packagename/{bin,lib,...}.

[Kapil] C. There would be a user-level package management system that would allow users to mix and match their requirements by creating symlink farms in /home/username/{bin,lib,...} that point to different packages under /opt possibly installed by different users.

[Kapil] This way, those users who need "cutting edge" or even "bleeding edge" tools could install them. The system could remain "stable" and (hopefully) "secure". Such a setup may even be useful on single-user systems as the user could play with upgrades without messing up the running system. Disk space is not really an issue any more unless you use "KDE" or "Gnome" or "Openoffice"

a linux solution for the office

From amitava maity

Answered By: Ben Okopnik, Kapil Hari Paranjape, Heather Stern, Thomas Adam

Hello all!

We have three Windows machines at our office. These are shared by 10 to 12 persons. I am trying to find out if one of these machines could be converted into a Linux box. Hopefully this Linux box can then be configured to meet the demands of 8 terminals connected via serial cables and an appropriate 8 terminal serial adaptor.GUI is necessary at all the terminals. Is this a feasible configuration? Can a Pentium-II, 233MHz machine with an IDE hard disk be used for this purpose?

Linux Gazette, you are doing a great job.

A Maity.

[Ben] Take a look at the Linux Terminal Server Project <http://ltsp.org/>;. It sounds like they'll fit your needs just fine.

Is this a feasible configuration?

[Kapil] There are two possible configurations:

1. A server with a multi-port serial adapter which is connected via
2. A server connected to an ethernet hub to which GUI terminals (such

In modern times, (2) is the way to go unless you already have the full infrastructure of (1). (Where will you find vt100 terminals nowadays?).

If you are going to buy some PC's you can anyway buy them with ethernet cards built-in and a small 16-port ethernet hub. In fact, you can buy "thin clients" if you use a server---in particular, you can save on disk costs.

I would suggest that you do not buy 8 vt100 terminals! This is not cost effective nowadays. Second-hand vt100 terminals are likely to be a hardware headache (since it would appear from your message that you are not a seasoned sysadmin of the unix days) and (more importantly) will not satisfy those of your users who are used to the colorful graphical interface of Windows.

Instead, you can "share" the use of the GNU/Linux machine as you have been sharing the use of the Windows machines---in fact this is a more shareable system!

Advantages over your existing configuration:

1. Users will be able to customize their desktop without interfering
2. Users will be able to run jobs that can run unattended in the
3. You can find and use a lot of useful software without attracting
4. When you get some funds you can add some more Linux machines very

[Heather] Probably, esp. if you use something a bit on the lightweight side for basic window management details (such as fvwm instead of more weighty desktops, abiword with gnumeric instead of staroffice, etc.)

[Heather] Ben's right, LTSP is a good place to look. If these "terminals" are PCs but you'd reather they stay hard-diskless, consider keeping them having CD bays and run knoppix (a live CD distro). Or maybe another liveCD distro (for instance SuSE has a free one) but I'm advised, most of them ask a few sysadminly questions during bootup, and Knoppix doesn't.

Linux Gazette, you are doing a great job. A Maity.

[Heather] Thanks, we love to hear that sort of thing!

Simple DNS solution with Red Hat 9

From - E J -

Answered By: Faber Fedor, Kapil Hari Paranjape, Jim Dennis

[root@localhost sbin]# ./ifconfig eth0 address
192.168.1.103
address: Unknown host
ifconfig: `--help' gives usage information.
[root@localhost sbin]#

Please note I believe I need a DNS server (solution) for my home Red Hat Network. Is there a simple DNS solution I can establish?

I have checked some books - this does not seem to be as simple as editing /etc/host.conf

[Faber] Okay, so? You mis-typed a command. I do it all the time.

[Faber] What are you attempting to do? What is "address"? Is it supposed to be an actual (IP) address? Is it an enviromant variable? What?

[JimD] I'm going out on a limb here to guess that this was supposed to be rendered as:

[root@localhost sbin]# ./ifconfig eth0 address 192.168.1.103
address: Unknown host
ifconfig: `--help' gives usage information.
[root@localhost sbin]#

... cut and pasted from a root shell session.

You wanted to type:

[root@localhost sbin]# ./ifconfig eth0 192.168.1.103

... which would seet your eth0 address to 192.168.1.103 That would also implicitly set the netmask to 255.255.255.0 and the broadcast address to 192.168.1.255. That netmask is the default for the traditional "Class C" network address blocks (all of 192.*.*.* among many others). The broadcast address is then calculated by masking off the high order bit using the netmask then setting that many bits to "on" (1). Then replacing the bits at the end original addresses with the broadcast.

In other words you can often just specify the address without spelling out the other settings. You only have to specify the others when you're network isn't following the "Classical" parameters and defaults.

The problem with your command was that the word "address" was parsed as the name of a host. The ifconfig command then tried to resolve that name into an IP address (presumably via your /etc/hosts file, then DNS --- though that depends on the settings in your nsswitch.conf)

Here's a couple of other examples of ifconfig commands:

# ifconfig eth1 10.0.1.10 netmask 255.255.255.0 broadcast 10.0.1.255 up
# ifconfig eth2 123.45.67.8 netmask 255.255.255.224 broadcast 123.45.67.31

Notice that the address is the one argument that is not prefixed by a literal/keyword or label. It's usually the first argument, though it might work even if you don't follow that convention.

Notice in my last example that we're using a smaller netmask, like the kind you might get from an ISP that was only giving you a block of 30 IP addresses. Long ago I wrote and article on "subnetting and routing" which is still one of the most popular article in LG/TAG history. I've been told it's used for some college TCP/IP fundamentals classes.

Please note I believe I need a DNS server (solution) for my home Red Hat Network. Is there a simple DNS solution I can establish?

[JimD] Perhaps you will need a DNS or other directory service (NIS or LDAP). However, in this case you just needed to look at the error and read the usage section of the --help and/or man page a little more carefully.

It also helps to think about the problem your trying to solve. You're trying to configure a network interface. DNS and other directory services need to use that interface (or some interface) in order to resolve names into IP addresses. That would create a chicken & egg problem if the ifconfig command truly depended on the name resolution. It would be unreasonable to assume that everyone has to run a network directory services daemon on localhost --- and you'd still need it to ifconfig the lo (localhost/loopback) interface.

(That line of reasoning should alert you to the fact that their was something wrong with your premise --- that your conclusion was dubious).

You almost certainly don't want to play with your /etc/host.conf

However, it could be as simple (in this case) as editing /etc/hosts

If you'd put an entry "192.168.1.103 address" as a line in your etc host file then your command would have almost worked. It would complain about extra arguments --- the lookup/resolution of the hostname "address" would have succeeded (assuming you have a normal /etc/nsswitch.conf).

If you put more reasonable address/name pairs in /etc/hosts and you securely distribute those (rsync -e ssh) to your other machines --- you have a working system of host name resolution without DNS NIS or LDAP.

/etc/nsswitch.conf defines the list of services and methods used by glibc (C library) functions to resolve names (hostnames, network names, netgroup, user, group, and service) into numbers (IP addresses, lists of hosts, UIDs, GIDs, and TCP/UDP port numbers). Almost all of the programs on your system are dynamically linked against glibc (a.k.a. just libc). glibc implements resolvers that read /etc/nsswitch.conf and dynamically load /lib/libnss* modules as listed.

Such run-time linkage uses the dlopen() interface. There are two types of dynamic linking in Linux. Link/compile time, such as the way that almost all programs are linked to libc and many programs are linked to libm (the C math functions library). These are listed by the ldd (ld dump --- ld is a non-intuitive mnemonic for "linker"). Run-time linking is done via the dlopen() interface. Any time a program must read a configuration file, command line option or environment setting, or any other run-time source of information to determine which modules to load --- it uses dlopen() Obviously this would be true of the NSS modules since any program that uses any of these name services it has to read /etc/nsswitch.conf to determine which libraries to load (NSS == "name services selection" or "name services subsystem" or something like that).

dlopen() (run-time dynamic linking) is also used by PAM, for PERL and Python binary modules, Apache modules, and XFree86 version 4 and later. You can think of these as being a way to implement some object oriented features in normal C programs. The primary uses of these modules are to extend and/or modularize the functionality of a base binary program.

Thus you can get a custom authentication module (say one of those little electronic credit card PIN tokens) and drop it into your system; add one configuration line and all of the PAM linked programs have been extended to use this module. All without recompiling anything.

As another example you can install XFree86, as compiled by your distribution vendor, and you can install a driver module for your video card from some third party (perhaps even the manufacturer of the card).

I realize I've delved deeply under the hood here --- into details that you won't understand at first reading (and probably don't care about).

My point is that you don't need to run a network name service. Most of the NSS linked programs check local config files /etc/hosts, /etc/passwd, /etc/groups, /etc/services, etc. first. They then check with other services as listed in the /etc/nsswitch.conf. /etc/host.conf is still used, but its usage is somewhat superceded by /etc/nsswitch.conf.

(I'd love to see a good explanation of why we have both nsswitch.conf and host.conf on modern systems --- something at a higher level than their respective man pages).

Creating RAMDISK

From Jose Nathaniel G. Nengasca

Answered By: Ben Okopnik, Kapil Hari Paranjape, Thomas Adam

Hi there,

I just want to create a RAMDISK of 100MB to use as temporary storage for squid cache files, I am using redhat 8.0, using grub bootloader, with 750MB of RAM, the one on linux focus site is rather old (november 1999) howto. Can someone help me with this?

Respectfully yours,

[Kapil] I think you have something confused here. As I understand it:

Squid creates an object cache in memory and periodically saves objects to disk when it runs out of space in memory.

You want to create a virtual (RAM) disk for squid to use as its disk. Instead why don't you increase the amount of memory available to its in-memory object cache?

[Ben] If you have the kernel sources installed, take a look at "/usr/src/kernel-source-<version>/Documentation/filesystems/tmpfs". It's a memory-based file system, and is created (by root) with "mount". Here's an example:

See attached ben-fstab.txt

[Thomas] I think Ben has missed the point. Squid doesn't care for the type of RAM being used. Sure, it needs physical RAM, but this is accessed on-disk by squid's "object cache". It is this that determines how much RAM is being used (see Kapil's answer above).

In /etc/squid.conf you can adjust this by changing:

cache_mem=50MB

to something more appropriate. Don't forget to issue:

squid -k reconfigure

once changes have been made to the file.

X server crash when starting up RH9 for the first time

From Claudiu Spataru

Answered By: Mike Martin, Ben Okopnik, Thomas Adam, Heather Stern

This is a multi-part message in MIME format.

Hello

My X server crashes with the following error messages (see attached log).

My system is an Athlon 2500+, GeForce FX 5200 graphic card, A7N8X deluxe mainboard (not sure how relevant this info is, but added it anyway).

The font server can be stopped and restarted by using '/etc/init.d/xfs start|stop' without any problems. When eliminating the Fontpath line that points to unix/:7100 from the XF86Config file, it complains about a different fontpath that it cannot find and crashes once again. (There is also either no path set to run Xconfigurator or there is no such thing in my installation of RH9.)

Any known solution to the above problem?

[MikeM] dont do this - put it back. If you dont use the xfs server then you need to hardcode the fontpath into /etc/X11/XF86Config

X wont start without xfs running, when X fails to start, is xfs running?

[Ben] That's inaccurate. Either a font server _or a hard-coded font set is sufficient, and there's no advantage that I know of to a server if you're not doing X over a network. My system has run without "xfs" for years now.

[MikeM] I know this is generally true, however the querent is using Red Hat 9

Red Hat does a fair bit with the font server, including dealing with easy adding of fonts, TTF fonts etc.

[MikeM] So to avoid other problems/questions IMHO, it is safe to say that the querent is better off using the installed font server. (ie: dont add unnecesary complications), especially as it sounds like it is actually working.

[MikeM] just did a quick google - its possible that the card is not recognised properly

so try this (as root)

edit the file /etc/X11/XF86Config look for the device section eg:

Section "Device"
	Identifier  "Videocard0"
	Driver      "nv"
	VendorName  "Videocard vendor"
	BoardName   "NVIDIA GeForce 4 (generic)"
	BUSID	    "PCI:1:0:0"
EndSection

and experiment changing the Driver entry to vesa or nv, depending what is there already

Thank you very much for the replies! It was indeed the fact that my graphic card did not get recognized properly and the generic vesa driver did not work in my case. After changing the values for 'BoardName' and 'Driver' as per Mike's suggestion, I was able to start X Windows.

[MikeM] Generic X error problem solving

There are a few very common reasons for X not starting

1. Not enough disk space in / or /var

[MikeM] 2. Font server not starting (can be caused by 1.)