Linux-Development-Sys Digest #295, Volume #6 Sun, 17 Jan 99 17:14:23 EST
Contents:
Re: file access in device driver? (Malte Starostik)
INT13 disk driver (Allan Jensen)
Re: The new interface in Java - IS VERY IMPORTANT! (Christopher B. Browne)
Re: 2 stacks? (Richard Jones)
starting a debugger from within the application that you would like to (Wim Delvaux)
Re: mmap problem (Daniel R. Grayson)
Re: Update from 2.0.29 to 2.0.36 (Fotis D. Zagoras)
Re: How to find lost ext2 partitions (Tero Pelander)
Re: Flush UDP on close()? (Andi Kleen)
The new interface in Java - IS VERY IMPORTANT! ("Marinho Brandão")
Re: - deprecated - why? (David Steuber)
Re: 2.2.0pre7 problem with can't find map file (Chris Vine)
Re: disheartened gnome developer ([EMAIL PROTECTED])
2.2.0pre7 dowsnt load modules (Staffan Nilsson)
KERNEL 2.0.36 - Unable to open an initial console (Dick Repasky)
----------------------------------------------------------------------------
From: Malte Starostik <[EMAIL PROTECTED]>
Subject: Re: file access in device driver?
Date: Tue, 12 Jan 1999 19:19:44 +0100
Frank Sweetser wrote:
> Malte Starostik <[EMAIL PROTECTED]> writes:
>
> > Hi!
> > While just plaing around writing a character device driver, I came
> > accross the problem that I need to read something like a config file.
> > When insmoding the module, I always get unresolved symbol errors for
> > open and close...
> > Can I use these calls at all? Are there any other ways to access a file
> > from within a kernel module?
>
> no. there's no API for doing such things from within the kernel. the
> reccomended way seems to be to configure the device via ioctl()'s passed
> from a userspace program to the device.
I agree that ioctl is the better way. Though, if anyone´s interested, I figured
out how to do it:unsigned long old_fs;
int fd, cb;
char buf[256];
old_fs = get_fs();
set_fs(get_ds());
if ((fd = sys_open("filename", 0, 0)) != -1) {
cb = sys_read(fd, buf, sizeof(buf));
sys_close(fd);
}
set_fs(old_fs);
Will work, while switching fs is required to not get EFAULT on sys_open. buf
must be in kernel space or you´ll get a wonderful Oops.
-Malte
------------------------------
From: Allan Jensen <[EMAIL PROTECTED]>
Subject: INT13 disk driver
Date: Sun, 17 Jan 1999 18:45:49 +0100
Hi there,
Would any of you know of a disk driver for Linux that utilizes INT13 (the BIOS
interface)
as the Disk I/O driver?
I am in quite a need for such one, for a project I'm involved in.
The problem is that I got a disk controller which is by no means standard IDE.
On DOS level it does behave as a regular hard drive, but as soon as you try to go
below the
BIOS interface, nothing (and I do mean _nothing_) is standard IDE.
Therefore, I'd like to know if a INT13 driver would be available for Linux. Yes, I
know it's
a ridiculous way to go, that it costs huge amounts of performance, that I ought to
try to get hold of the driver specs etc.etc, but trust me - I've considered and tried
all
of this, and I've had no luck.
Therefore, I need to find an INT13 driver if such one exists. If it's not available, I
may
look into writing a driver on my own (even though I'm no experienced kernel hacker).
So.. If someone knows of such a driver, or may want to help me on this one, please
drop me
a mail ASAP, as I need this pretty quick.
Thank you so much in advance.
Best regards,
-Allan Jensen
The way to love anything is to realize that it might be lost.
------------------------------
From: [EMAIL PROTECTED] (Christopher B. Browne)
Subject: Re: The new interface in Java - IS VERY IMPORTANT!
Reply-To: [EMAIL PROTECTED]
Date: Sun, 17 Jan 1999 18:39:15 GMT
On Sun, 17 Jan 1999 13:36:59 -0200, Marinho Brandão <[EMAIL PROTECTED]>
posted:
>************* READ THIS BECAUSE MOST IMPORTANT ********************
>Are you tired from that horrible interface of X-Windows? Iam also. And
>have a big idea as solution!
>
>Before, I likely that know if is possible to work in graphic mode at
>Java. I not, send for me a message informing this. This is only a
>project.
There is nothing in particular about using Java that will make such an
effort either better or worse; Java is only a language.
If the use of Java overpowers the quality of the API, you should look at XTC
<http://www.cs.umb.edu/~eugene/XTC> which is a GUI library written in "pure
Java" that should solve your X problems.
Alternatively, if you'd rather look at good GUI abstractions, you should
consider libraries such as Qt, FLTK, GTK, Tk, and OpenStep, all of which may
be used to provide APIs that many claim are nice that run atop Xlib and
other graphical substrates.
--
Those who do not understand Unix are condemned to reinvent it, poorly.
-- Henry Spencer <http://www.hex.net/~cbbrowne/lsf.html>
[EMAIL PROTECTED] - "What have you contributed to Linux today?..."
------------------------------
From: Richard Jones <[EMAIL PROTECTED]>
Subject: Re: 2 stacks?
Date: Sat, 16 Jan 1999 19:49:19 +0000
David Wragg <[EMAIL PROTECTED]> wrote:
: Richard Jones <[EMAIL PROTECTED]> writes:
:> In fact, with gcc and pure C, alloca isn't necessary
:> either because gcc extends C to provide useful (albeit
:> non-portable) ways to allocate variable-sized arrays
:> on the stack.
: C9X adds variable length arrays too, but the details are significantly
: different from the gcc facility.
That's interesting. Can you tell us more or
point to an online resource about this? I
heard (pure rumour and speculation :-) that
a lot of C++ features like default arguments
and overloaded functions might be added ...
Rich.
--
- Richard Jones. Linux contractor London and SE areas. -
- Very boring homepage at: http://www.annexia.demon.co.uk/ -
- You are currently the 1,991,243,100th visitor to this signature. -
- Original message content Copyright (C) 1998 Richard Jones. -
------------------------------
From: Wim Delvaux <[EMAIL PROTECTED]>
Subject: starting a debugger from within the application that you would like to
Date: Sun, 17 Jan 1999 15:08:12 +0100
Hi all,
I would like to be able to start a debugger (e.g. ddd) from within the
application that needs to be debugged.
This is because the application I write detects problems (e.g. memory
overruns) and it would be great to
the user to have some kind of dialog box pop up to ask whether he would
like to start debugging.
I know that the way to start debugging is to call the strace system call
but I do not know what happens
if I call strace to 'myself' just before I forked/execed the debugger
with my PID. Will the strace call issued
from the debugger Work ? or Not ? Does anybody have any other ideas how
to do this ?
Thanx
WIm Delvaux
PS. In windows you do this by triggering IRQ. THis IRQ is caught by the
kernel and will prompt for the debugger.
------------------------------
From: [EMAIL PROTECTED] (Daniel R. Grayson)
Subject: Re: mmap problem
Date: 17 Jan 1999 13:15:53 -0600
Marcin Szyllo <[EMAIL PROTECTED]> writes:
> Hi there,
>
> I'm battling to understand why mmap (as in the attached example) does not
> allow me to mmap a block device - a partition... Why NOT?
Probably only because no one has written such code for the kernel yet.
In linux/drivers/block/hd.c we see a NULL entry for the mmap routine.
static struct file_operations hd_fops = {
NULL, /* lseek - default */
block_read, /* read - general block-dev read */
block_write, /* write - general block-dev write */
NULL, /* readdir - bad */
NULL, /* poll */
hd_ioctl, /* ioctl */
NULL, /* mmap */
hd_open, /* open */
hd_release, /* release */
block_fsync /* fsync */
};
------------------------------
From: [EMAIL PROTECTED] (Fotis D. Zagoras)
Subject: Re: Update from 2.0.29 to 2.0.36
Date: Sun, 17 Jan 1999 15:05:22 GMT
[EMAIL PROTECTED] (Fotis D. Zagoras) wrote:
>Hello,
>I have updated my 2.0.29 kernel to 2.0.36.
>The new kernel works fine but i get always
>the following 3 first lines in my /var/adm/syslog file. :
>Jan 11 19:48:55 kassiopea kernel: Error in symbol table input.
>Jan 11 19:48:55 kassiopea kernel: Cannot find map file.
>Jan 11 19:48:55 kassiopea kernel: Memory: sized by int13 0e801h
>etc ....
>The problem is with the first 2 lines. This messages are produced
>by klogd deamon (I suppose) and maybe have some relation with
>the System.map file. Any suggestion to resolve this problem.
>Thank You, Fotis.
My /usr/src/linux/Makefile puts the 2 exit files: vmlinuz and
System.map in directory /boot; So i have changed the
/etc/rc.d/rc.inet2 file in the line that fire the klogd daemon.
This line is made: klogd -k /boot/System.map
With this change in have resolved only the 2nd error line:
(Cannot find map file) but stiill remains the 1st error line:
(Error in symbol table input).
I think that this partial solution is not radical and that the problem
is deeper - maybe a conflict between the newer kernel (2.0.36)
and the version of klogd (1.3-0).
I have sent e-mail to the creator of klogd ([EMAIL PROTECTED]). You can
find this address in the man page of klogd.
Thank you, Fotis
------------------------------
From: [EMAIL PROTECTED] (Tero Pelander)
Subject: Re: How to find lost ext2 partitions
Date: 17 Jan 1999 18:28:24 GMT
[EMAIL PROTECTED] (Yann Dupont) writes:
>Hello. I have a problem : I have two disks where the partition tables have
>been erased by error (human error :)
>The ext2 partitions haven't be damaged - or the damage is minimal.
...
You could try the gpart program that tries to guess the partition tables.
If you can't find on an ftp server try...
http://www.stud.uni-hannover.de/user/76201/gpart/
------------------------------
From: Andi Kleen <[EMAIL PROTECTED]>
Subject: Re: Flush UDP on close()?
Date: 12 Jan 1999 18:04:02 +0100
In article <77eth2$[EMAIL PROTECTED]>,
[EMAIL PROTECTED] writes:
> Is there no way of knowing _when_ the kernel has sent all my UDP data, and
> it's safe to close() my UDP socket? Is there no way of forcing data to be
> flushed, or have my write() calls block until data has been really sent?
UDP data is always flushed (=put into the device send queue) as soon as
possible, there is no other buffering in the kernel. The UDP sendmsg
function "forgets" about the kernel packet as soon as it is queued (the
only information left is in the write buffer allocation counter that you
can see with netstat). You can always get the size of the write buffer
with ioctl(sk, TIOCQOUTQ, &amount), when amount == 0 all packets have been
sent. There is no way to sleep for this going to zero, UDP doesn't support
linger close. If you care about your data you need some acking/retransmit
scheme anyways.
-Andi
------------------------------
From: "Marinho Brandão" <[EMAIL PROTECTED]>
Subject: The new interface in Java - IS VERY IMPORTANT!
Date: Sun, 17 Jan 1999 13:36:59 -0200
************* READ THIS BECAUSE MOST IMPORTANT ********************
Are you tired from that horrible interface of X-Windows? Iam also. And
have a big idea as solution!
Before, I likely that know if is possible to work in graphic mode at
Java. I not, send for me a message informing this. This is only a
project.
The idea is: a graphic interface for Linux rapid, simple, small, and
most beauty that the Windows.
This interface will be call as "JavaMyc" in that text.
Is anything as: insted of windows, have boxes in 3D; and insted of the
desktop, have the infinite! So we will can to go round a box and see the
side ou the behind, this is wonderful!!! And continue: the language used
will the Java. Because Java? You saw the samples of JDK? Have a from it
that have a spheres, you click and push, pull, drag and it round one for
behind of other, etc. Imagine make this with programs! Imagine thet
icons floating in the black background! All will be FREE in internet, as
the Linux Kernel.
Applications, drives, classes, API objects and libraries will be class
in Java. The difference is a public int variable: public int
__TYPECLASS__. The JavaMyc read it and define the type. Simple no?
public int __TYPECLASS__ inexistent -> classe
public int __TYPECLASS__ = 0 -> executable
public int __TYPECLASS__ = 1 -> library
public int __TYPECLASS__ = 2 -> API object (only a package)
public int __TYPECLASS__ = 3 -> drive
This signify that will be 99% immunity from virus (Java classes are
secure)
=================================================================
That is the directory relationchip:
/bin -> classes for call of JavaMyc
/src -> open source-code
/lib -> the add-in libraries from JavaMyc
/api -> the classes from API: is equals the java.api package, every
class is an object. So, is simple adding a new object
/drv -> the drives of devices
/cfg -> the files with parameters. One of it is the cfg.ini, that have
the system parameters. A parameter is: FILELANG=<path of file with
values of variables with text to display>, with this, is simple
translate the JavaMyc for germany, spain, portuguese or any
languages.
/lang -> default directory from language files. Examp.: ptg.lng,
eng.lng, spa.lng, etc.
/init -> links or classes to automatic initialize with the JavaMyc
/menu -> is the menu directory, equal the "Menu Iniciar" directory of
Windows95 SO.
=================================================================
Extended because this is a wonderful project?
Increment and modify this ideas for a perfect graphic interface, ok!
I am waiting the reply.
Ah, and sorry by bed english, I am brazilian.
===================================
by Marinho Brandão... hehe...
[EMAIL PROTECTED]
www.lagoserradamesa.tur.br/marinho
viva Linux!!! morte ao rWindows!!!
Programmer in C, C++, Java, Pascal and Assembly.
------------------------------
From: David Steuber <[EMAIL PROTECTED]>
Crossposted-To: comp.os.linux.advocacy
Subject: Re: - deprecated - why?
Date: 17 Jan 1999 13:32:30 -0500
[EMAIL PROTECTED] (Christopher B. Browne) writes:
-> I see the smileys; I get the joke; anyone that is actually serious about
-> the notion that free software might dominate against proprietary software
-> and ignores the issue that "we" have to start setting the standards,
-> *and paying for the process,* is pretty naive.
Why should it cost anything to create a standard? Does ANSI or the
ISO have a monopoly on standards? What if we had an Open Standards
Organization that served as a standards body that operated in
essentially the same mannor as the FSF?
The reason for ISO and ANSI was to provide _industry_ standards. The
industry was made up of commercial entities. They become obsolete in
an environment where anybody can float a 'standard' and if it is any
good it will be adopted. Like Perl for instance.
We will set standards. The process will be a democratic one. The
voters will be anyone who can write code that implements the
standards. No corporate sponsered consortium needs to be paid for the
process.
I just did a quick check. oso.org and openstandards.org are taken. I
like openstandards.org better as a domain name. The record was
created on 25, Dec, 1998. Missed it by that much. Neither of the two
domains appear to have a website.
--
David Steuber
http://www.david-steuber.com
s/trashcan/david/ to reply by mail
SYSTEM ALERT: /dev/null is full. Please delete any unnecessary files.
------------------------------
From: [EMAIL PROTECTED] (Chris Vine)
Subject: Re: 2.2.0pre7 problem with can't find map file
Date: Sun, 17 Jan 1999 15:48:41 GMT
On 17 Jan 1999 00:17:12 GMT, Frank Hale <[EMAIL PROTECTED]>
wrote:
>Yes that was the problem I had the wrong version on here 1.3.27 which I
>got from rawhide.redhat.com I think they need to sync there directories
>since rawhide is supposed to be the latest and greatest, I owe you a
>virtual beer. Thanx alot. I downloaded the binary and it works only
>catch is you still need to supply the symlink or it will still complain
>it can't find map file.
>
>ln -s System.map-2.2.0pre7 System.map
>
That's odd. Mine definitely does not need symlinking, so I can run
multiple kernels without having to re-symlink. I have
System.map-2.0.36, System.map-2.2.0-pre5 and System.map-2.2.0-pre7 all
in /boot, and all are correctly identified, depending which kernel I
am booting via lilo.
You may have used the wrong format - that is, System.map-2.2.0pre7 is
wrong. It should be System.map-2.2.0-pre7
Chris.
--
If replying by e-mail, remove the --nospam--
------------------------------
From: [EMAIL PROTECTED]
Crossposted-To: comp.os.linux.advocacy,comp.os.linux.development.apps,comp.os.linux.x
Subject: Re: disheartened gnome developer
Date: Sun, 17 Jan 1999 21:23:06 GMT
In article <77qk82$[EMAIL PROTECTED]>,
Navindra Umanee <[EMAIL PROTECTED]> wrote:
> Okay, Qt handles/captures destroy and delete events by default so I'll
> quote code that shows you how to write your own slots.
>
FYI, GTK, captures destroy and delete events by default as well. The example
was showing how to capture it yourself so that you might do something with
it, i.e you might want to take some action before the appication exits like
pop up a dialog that says "do you want to save your changes" This was
explained in the notes that were in the tutorial. I'm sure there's a way to
do it in Qt as well. But, my point is you shouldn't compare code when you
don't fully understand what the code is doing and why.
[slot example snipped]
FWIW, your slot example is not completle. The line "public slots:" is not C++
and needs to be processed by a special pre-processor. You must configure your
make file to call that preprocessor. The pre-processor will then generate an
additional cpp and header file that also need to be handled by the make file.
I don't know how well GNU automake can handle this stuff, perhaps a couple of
extra m4 macros all that's needed.
But were still comparing apples to oranges. In Gtk, I have no need to create
slots. I can connect an object's signal to a callback that can operate on any
other object.
> So in your judgement most people developing with GTK and GNOME tend to
> use C and the other bindings aren't really mature yet? (IOW at the
> moment, bindings or not, C is simply the language of choice for GTK or
> GNOME developers.)
Gnome is a work in progress. It's not even at 1.0 yet. While the Gnome
developers chose the GTK toolkit because GTK was designed to be easily
extensible, for the most part they are not the ones working on the bindings.
In addition, most of the people working on the binding are not so much
GTK/Gnome fans as they are fans of their respective languages. That is to say
many of the binding developers are more interested in enhancing their
respective languages than they are in enhancing Gnome in particular. Gnome
just happens to be the easiest path to a toolkit binding for their language.
And that's exactly what the Gnome developers wanted. But since Gnome is not
complete, it follows that the bindings are even less complete.
I have seen some GTK apps in Perl, Python, and C++. But even those binding
are incomplete at this time, so it's only natural the majority of GTK apps at
this time are in C. And while in the future we will see more apps with the
bindings, But I don't think we will see less apps in C, instead we will see
more apps total.
But as has been stated repeatedly by myself and others, programming GTK in C
is rather easy. If you think it's ugly though, your entitled to feel that
way.
> Agreed, I'm hoping someone will do in depth comparisons some day. :)
>
Considering both KDE and Gnome are likely to flourish, there are likely to be
many. Of course, some will be made by idiots who don't know WTF they are
talking about, others will be done by people who are obviously biased, and a
few will be fair comparisons. And there will always be religious rantings in
regards to all of them.
Linux is about freedom, Navindra. Do what you want and let others do what they
want.
Perry
============= Posted via Deja News, The Discussion Network ============
http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own
------------------------------
From: Staffan Nilsson <[EMAIL PROTECTED]>
Subject: 2.2.0pre7 dowsnt load modules
Date: Mon, 18 Jan 1999 16:51:44 +0000
Hello
I know this problem, or similar ones, have been discussed already, but I
still have no luck running 2.2.0pre7 on a standard redhat 5.2
installation. The problem is that the kernel doesnt load any modules,
and therefore is rather useless. I tried installing modutils 2.1.121
from rahide, but they didnt would (see previous discussion). Someone
suggested getting the source rpm and recompiling,so I did. It compiled
fine, but after installation, my standard 2.0.36 kernel didnt find its
modules. I didnt even bother trying to run 2.2.0.
So now I am back to the old standard 2.0.36.
Can anybody help?
/Staffan
------------------------------
From: [EMAIL PROTECTED] (Dick Repasky)
Subject: KERNEL 2.0.36 - Unable to open an initial console
Date: 17 Jan 1999 21:38:31 GMT
Reply-To: [EMAIL PROTECTED]
I receive the message from kernel 2.0.36:
Unable to open an initial console
just after the kernel mounts the root file system and unmount old
root. Here are the exact messages:
UMSDOS Beta 0.6 (comptatibility level 0.4, fast msdos)
VFS: Mounted root (UMSDOS filesystem)
Trying to unmount old root ... ok
Unable to open an initial console
The kernel then stops dead.
You should know that this is a root file system that I have populated.
I have considered the following two possibilities:
1) A device file is missing from the root file system.
Here is a list of /dev devices that seem relevant listed
when the the root file system is mounted as an UMSDOS
file system by a linux system running on a different disk.
cr--r--r-- 1 root root 4, 0 May 5 1998 console
pr--r--r-- 1 root root 0 Jan 16 02:05 initctl|
cr--r--r-- 1 root root 1, 3 May 5 1998 null
cr--r--r-- 1 root root 4, 0 May 5 1998 systty
cr--r--r-- 1 root root 5, 0 May 5 1998 tty
cr--r--r-- 1 root root 4, 0 May 5 1998 tty0
cr--r--r-- 1 root root 4, 1 Jan 16 01:27 tty1
cr--r--r-- 1 root root 4, 2 Jan 16 01:57 tty2
cr--r--r-- 1 root root 4, 3 May 5 1998 tty3
cr--r--r-- 1 root root 4, 4 May 5 1998 tty4
cr--r--r-- 1 root root 4, 5 May 5 1998 tty5
cr--r--r-- 1 root root 4, 6 May 5 1998 tty6
cr--r--r-- 1 root root 4, 64 May 5 1998 ttyS0
cr--r--r-- 1 root root 4, 65 May 5 1998 ttyS1
cr--r--r-- 1 root root 3, 0 May 5 1998 ttyp0
cr--r--r-- 1 root root 3, 1 May 5 1998 ttyp1
cr--r--r-- 1 root root 3, 2 May 5 1998 ttyp2
cr--r--r-- 1 root root 3, 3 May 5 1998 ttyp3
cr--r--r-- 1 root root 3, 4 May 5 1998 ttyp4
cr--r--r-- 1 root root 3, 5 May 5 1998 ttyp5
cr--r--r-- 1 root root 3, 6 May 5 1998 ttyp6
cr--r--r-- 1 root root 3, 7 May 5 1998 ttyp7
cr--r--r-- 1 root root 3, 8 May 5 1998 ttyp8
cr--r--r-- 1 root root 3, 9 May 5 1998 ttyp9
cr--r--r-- 1 root root 3, 10 May 5 1998 ttypa
cr--r--r-- 1 root root 3, 11 May 5 1998 ttypb
cr--r--r-- 1 root root 3, 12 May 5 1998 ttypc
cr--r--r-- 1 root root 3, 13 May 5 1998 ttypd
cr--r--r-- 1 root root 3, 14 May 5 1998 ttype
cr--r--r-- 1 root root 3, 15 May 5 1998 ttypf
cr--r--r-- 1 root root 1, 5 May 5 1998 zero
2) The kernel hasn't actually mounted the root file system
properly, and it cannot therefore find /dev.
I have no way of judging this, but I suspect it because the
last act of an initrd linuxrc script is to pass the id of
the root file system to the kernel. The root file system is
/dev/sda1, and the command line within linuxrc that sets the
root file system is:
echo 2049 > /proc/sys/kernel/real-root-dev
Subsequent "cat /proc/sys/kernel/real-root-dev" yields 2049.
Note that there is no other dos partition on any disk on the
system that could be mounted as dos/umsdos.
I also suspect this hypothesis because the umssync operation
that I used to set up the file system was run on the linux directory
rather than the actual mount point (i.e., /mnt/bld/linux
rather than /mnt/bld). (Note that all symbolic links, named pipes
device files and file permissions with in the /mnt/bld/linux
directory tree are fine.)
Another possibly relevant fact: The error occurs after an initrd linuxrc
script exits. The linuxrc script is very active. It actually mounts
the root-partition-to-be, a floppy as well as a CD-ROM and then builds
the root file system beginning with the "C:\linux" directory. The new
root file system actually checks out very well. During test runs
in which linuxrc is a symbolic link to ash, I can from a shell prompt
manually run the script, and once it is finished but before control
is returned to the kernel, I can chroot to the root-to-be (chroot
/mnt/bld/linux /bin/bash) and run scripts and binaries from the root
file system. Linuxrc unmounts the cd, floppy, root-file-system-to-be
and the proc file system before returning control to the kernel.
I assume that the error occurs before the kernel reaches for
/sbin/init. True?
So, what could be wrong that would produce the error that I have?
Thanks,
Dick
--
Remove the underscore from my e-mail address to reply by mail.
------------------------------
** FOR YOUR REFERENCE **
The service address, to which questions about the list itself and requests
to be added to or deleted from it should be directed, is:
Internet: [EMAIL PROTECTED]
You can send mail to the entire list (and comp.os.linux.development.system) via:
Internet: [EMAIL PROTECTED]
Linux may be obtained via one of these FTP sites:
ftp.funet.fi pub/Linux
tsx-11.mit.edu pub/linux
sunsite.unc.edu pub/Linux
End of Linux-Development-System Digest
******************************
