Return to BSD News archive
Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!nntp.coast.net!news.sprintlink.net!neonramp.com!cynjut.neonramp.com!cynjut.neonramp.com!not-for-mail
From: burgess@cynjut.neonramp.com (Dave Burgess)
Newsgroups: comp.unix.bsd.netbsd.announce,comp.unix.bsd.freebsd.announce,comp.answers,news.answers
Subject: [comp.unix.bsd] NetBSD, FreeBSD, and 386BSD (0.1) FAQ (Part 6 of 10)
Supersedes: <386bsd-faq-6-822163664@cynjut.neonramp.com>
Followup-To: comp.unix.bsd.netbsd.misc
Date: 20 Jan 1996 16:30:55 -0600
Organization: Dave's House in Omaha
Lines: 788
Approved: news-answers-request@MIT.Edu,cgd@sun-lamp.cs.berkeley.edu
Expires: 02/07/96 16:29:54 CDT
Message-ID: <386bsd-faq-6-822176994@cynjut.neonramp.com>
References: <386bsd-faq-1-822176994@cynjut.neonramp.com>
Reply-To: burgess@cynjut.neonramp.com (386bsd FAQ Maintainer)
NNTP-Posting-Host: cynjut.neonramp.com
Keywords: FAQ 386bsd NetBSD FreeBSD !Linux
X-Posting-Frequency: Posted on/about the 13th and the 27th of every month.
Xref: euryale.cc.adfa.oz.au comp.unix.bsd.netbsd.announce:76 comp.unix.bsd.freebsd.announce:108 comp.answers:14621 news.answers:56251
Posted-By: auto-faq 3.1.1.2
Archive-name: 386bsd-faq/part6
Section 5. (Kernel Replacements)
5.0 Introduction
This section is supposed to document the unusual or optional
kernel add-ons that are available from various places. As
they are included in the mainstream of the various Berkeley
Net Release systems, they will slowly come out of here.
If you know of any replacement parts for the kernel, please
send Dave Burgess (burgess@cynjut.infonet.net) a message
detailing the package (possibly include a README), where it
can be found, and what version of the OS (ie. NetBSD,
386bsd 0.1 + pk 0.2, FreeBSD) it was designed to run under.
5.1 Available Kernel Replacements
5.1.1 keycap/codrv
These server as replacements for the generic pccons driver
that comes (by default) with 386bsd 0.1.
Holger Veit (author of these) writes:
"The same type of driver, but keycap has the version number 0.1.1
and codrv has the version number 0.1.2. The latter is much
improved and downward compatible. Codrv was developed to provide
a universal way of mapping national keyboard layouts during
runtime (ie, not by patching the kernel tables) and providing
better X11 support. Codrv uses a superset of the pc3 terminal
emulation, and a termcap-like database for keymaps (therefore
"keycap"). X11 is supported by two dedicated console raw devices
/dev/kbd and /dev/vga, which avoids all the existing problems
pccons has with X11. The latest version has virtual consoles.
Codrv will become part of patchkit 0.2.4"
5.1.2 pcvt
A superset of pccons, this driver supports virtual consoles,
and some form of database oriented keyboard mappings. It was
also designed to emulate a vt220 terminal as best as possible.
(This section originally identified Joerg Wuensch as the
author. Hellmuth Michealis is the author of pcvt. Joerg was the
author of the original post. This update is from Hellmuth
himself. Apologies from the FAQ staff...)
The last release of pcvt is version 3.00 and was done on March
1st 1994 to the newsgroup comp.sources.misc, Volume 41, Issue
140-152 (part 1-13). Future releases and upgrades will be done
as patches or new releases to that newsgroup.
pcvt was recently put into the kernel sourcetree of
NetBSD-current (pre 1.0) into /sys/arch/i386/isa/pcvt.
pcvt is also available in the FreeBSD contributed tree at
location /usr/ports/util/pcvt.
The pcvt package consists of:
- the driver itself
- complete documentation for installation and operation
- termcap/terminfo, pcvt.el, rc.local, /etc/ttys, xmodmap examples
- cursor, utility to set the cursor size and shape
- fed, a curses-based EGA/VGA character set editor
- fontedit, utility to edit the VT220 downloadable character set
- ispcvt, utility to display the drivers compile time configuration
- kcon, utility to setup national keyboard layouts and remap keys
- keycap, keyboard mapping database library similar to termcap
- loadfont, utility to load up to 4/8 fonts into an EGA/VGA board
- mcon, utility to control/configure a keyboard based mouse emulator
- scon, utility to runtime configure the video part of pcvt
- userkeys, utility to set the VT220 user programmable function keys
- vttest, VT100 compatibility torture test program
- demo, color- characterset- and attribute demos
and more ....
See the README-file for the latest release (3.00) of pcvt for
lots more information and a complete list of the contributors
to this project.
5.1.3 syscons
Another superset of pccons that was designed to emulate SCO as
well as possible. Many of the ioctls from SysV have been
implemented. XFree86 2.0 no longer requires special patches
to be run with kernels using this console driver.
5.1.4 Fast Symbolic Links
The following is taken from the README for the fast sym-links
patch:
"This cruddy but complete hack answers one of the objections to
symlinks: that they are slow, and cost an entire frag. Symlinks
of less than length 60 are stored in the inode itself. Symlinks
longer than this are still in the inode. To make the illusion
of normality complete, dump and fsck also need changing.
Additionally, I made dumpfs verbose to excess."
Fast Symbolic Links are supported natively in FreeBSD and NetBSD.
5.1.5 npx fixes
There are problems with the floating point error handling
routines, and there are fixes available for this problem provided
by Bruce Evans (of Minix-386 fame)
Note that most of the code is applicable to floating point hardware
as opposed to emulation.
The newest version (and now official) fixes to this are in
patchkit-0.2.4.
There are still some nits in the npx emulation code in both FreeBSD
and NetBSD. They are being worked on.
5.1.6 CGD's COM drivers
Chris G. Demetriou (cgd@blah blah blah) has written some COM
drivers for 386bsd. These, among other things, support multi-
port serial packages.
This driver was the basis for the FreeBSD com subsystem. NetBSD
does not use them. There are patch files around that added
some of the missing functionality to NetBSD. Multiport comm
support (the biggest and best feature of the CGD COM driver) is
included in both FreeBSD and NetBSD as standard 'compile in'
features.
5.1.7 The original 386bsd 0.1 wd.c driver doesn't work. [386bsd 0.1
only]
If you are still using 386bsd 0.1 and are having trouble with
the wd driver, Tom Helbekkno took the time to write a
replacement wd driver specifically to fix the problems
identified in the months after 0.1 was released. Much of this
code was pulled into the patch-kit.
The 'Barsoom' driver was used as the basis for the original
NetBSD and FreeBSD drivers, and was the source for much of the
work that has been done to date on this driver.
Unfortunately, the 'barsoom' driver is no longer available, and
E-Mail from Tom indicates that there is very little of the
original 'barsoom' driver left in either FreeBSD or NetBSD. If
you find yourself in this predicament, you REALLY need to
upgrade to either FreeBSD or NetBSD.
5.1.8 Interruptless LPT Driver Kit [386bsd 0.1 only]
This driver was designed with faster performance and lower system
load in mind. See the INSTALL-NOTES that come with the package
for more details and installation information.
This is also included in NetBSD and FreeBSD. Note that with some
printers, it may be prefereable to ignore the status port and rely
on the data port. If you have tried everything else and the
interruptless printer driver still does not work for you, you may
need to play with this.
It has also been determined that the interruptless driver may be
(or already has been) removed from the system. A newer lpt driver
has been developed that removes many of the overhead problems that
the original 386bsd lpt driver had.
5.1.9 A replacement curses program/library.
It is generally accepted that the NetBSD curses can be easily
replaced by the ncurses package. It is more complete and offers
much better support for shared libraries and other advanced
features. The current (early 1995 version) is 1.8.5 and is
available from ftp://netcom.com:/pub/zmbenhal/ncurses/1.8.5.tgz.
5.2 Floppy Disk problems.
One of the most common problems in 386BSD involves working with
new boot sector and/or reformatting a floppy. Dave Silvia
provided this section on using floppy disks.
5.2.1 How do I get a bootable floppy?
Several ways, ranging from brain-dead-but-works to simplest.
Classification into categories is left to the reader (is there
really a difference between 'brain-dead' and 'simple'?:')
1) rawrite (or dd) dist.fs (or fixit.fs) to a disk,
mount it, cd to the mount point, and execute:
rm -rf .
you now have a bootable floppy!;^}
2) Take your existing dist.fs or fixit.fs boot disk and
diskcopy it on a DOS machine. Mount and rm as in 1)
above. Again, you have a bootable floppy!;^}
3) Run disklabel on the floppy, e.g.:
disklabel -w -r fd0a floppy5
where 'floppy5' is a 'name' for an entry in the /etc/disktab
file. You'll get a couple of ioctl errors because writing a
label to a floppy isn't supported (yet?), but the boot blocks
have indeed been written.
4) Write the boot blocks to the floppy:
cat /usr/mdec/fdboot /usr/mdec/bootfd | dd of=/dev/rfd0a
or, more simply:
cat /usr/mdec/fdboot /usr/mdec/bootfd > /dev/rfd0a
Methods 3) and 4) require you to run newfs on the floppy, e.g.:
newfs /dev/rfd0a floppy5
If you have a floppy that was originally bootable, but the boot
blocks were somehow damaged, you can use method 3) or 4) to
restore boot-ability (do _NOT_ run newfs). You _could_, through
the convolutions of copying a floppy whose boot blocks are damaged
to a temporary location and then re-copying to a bootable floppy,
use method 1) or 2) (if you really want to!;^})
5) If the disk is already newfs'ed and is otherwise ready to use,
disklabel will write the boot blocks on the disk. Read the man page
for disklabel.
5.2.2 How do I maximize the space on a mountable floppy disk.
As you all know, when you are working with a floppy, it is usually
more important that the floppy have a lot of room, rather than a
lot of other 'stuff'. Here is the magic incantation that will
maximize the amount of free space on the disk.
newfs -Tfloppy[35] -i[4096 | 8192] -c 80 /dev/fd[0|1]a
This leaves the disk with fewer inodes and only one cylinder group.
5.3 Character Device Driver info
These devices are also often referred to as character devices.
5.3.1 Printers
Configuring a parallel printer for 386bsd requires a working
printer driver to be installed in the kernel. 386bsd 0.1 does
not include a printer driver in the stock distribution kernel.
NetBSD and FreeBSD both include this driver in their stock
manifestations.
It is possible to connect a serial printer to either. This brief
tutorial is provided by Daryl Berryhill
(djberry2@b25info.b25.ingr.com)
The way I got my printer to work.
1) connect a 25 pin to 9 pin null modem cable to printer and
computer.
2) set printer to 9600 baud, 7 data bits, even parity.
3) configure /dev/com1 (DOS COM2) port the same way as the printer
4) add a line to /etc/printcap that says:
lp|local line printer:\
:lp=/dev/com2:wq:sd=/var/spool/lpd:lf=/var/log/lpd-errs:\
:br#9600
5) type "lpr <add filename here>"
6) type "lpd"
and it should start printing.
An obvious point, but make sure that you do NOT start a getty on
on the com port. Check the /etc/ttys file and make sure that
the com port you select is not active.
There have been many reports in the past of people not being able
to get their parallel port printer working. One of the problems
seems to be cables. Another problem may be with the hardware.
A seemingly stupid suggestion is to replace your printer card with
the cheapest parallel port card you can find. I am using a $10
single parallel, two serial port card that I got from Altex.
Works great.
In addition, there are people that want to set up multiple printer
queues using the BSD queueing mechanism. Here is a brief tutorial:
Lpf is mainly intended for processing text and nroffed output...it
isn't anything clean...it will do certain games that will mess up
PCL output.
If you're producing straight PCL or Postscipt (assuming your LJ
takes it), then you want to print directly to the printer w/o any
processing.
What you really want is a "physical" queue that does no processing,
and several logical queues that map back to the physical queue and
do processing...or one "smart" queue that does file content
recognition and then maps to the raw queue.
I do something like this for my DeskJet. There is one raw queue
and several logical queues (some postscript that do different
resoultions and color depth, some text that do various formatting,
etc). When I get the time I'll be trying to set up a "smarter"
queue using aps and maybe some bits from flexfax.
To map logical to physical queues either use a filter that pipes
back into lpr -P<rawqueue> itself, or just point it at the "raw"
queue using something like:
textlp|Text Printing:\
:lp=/dev/null:\
:if=/usr/libexec/lpr/lpf:\
:rm=localhost:\
:rp=lj.raw:
And other entries as needed, you get the idea...to use an output
filter instead of the rm/rp approach (more efficent), you can get
away with something like:
:of=/usr/local/bin/printraw:
where /usr/local/bin/printraw looks like this:
#!/bin/sh
cat | lpr -h -Plj.raw
5.3.2 Terminals/Keyboards
Terminals are relatively simple to add. It involves making sure the
/etc/ttys file identifies the com port (com0, com00, or tty00
depending on your configuration) as an active port and a getty is
running. The man page for ttys and getty help explain this.
Many people report that there are sometimes problems running some
programs on a remote terminal. There are some known bugs in the
terminal handler where the parity and bits per character are
concerned. They are being worked on.
5.3.3 Modems/FAX Modems
5.3.3.1 How do I add a modem to 386BSD:
The first part that confused me was assuming that /dev/com1 is
the same as DOS com1, they're not. /dev/com0 is connected to
COM1 and (I think) /dev/com1 is connected to COM2.
The switch settings for my modem were the same as what I had
under DOS, CTS CD RTS et al were set to follow the actual line
(i.e. my modem can force them high, which I turn off)
Ok that's not too bad.
Now you need to edit the /etc/remote file to include a reference
to the com port. I have only used NetBSD-0.8, so I'm not sure
what the default files are like that come with the other rev's
of 386BSD.
I added the last line (with com0).
--------------------------------------------------------
# @(#)remote 5.2 (Berkeley) 6/30/90
#
...stuff deleted...
# UNIX system definitions
unix1200|1200 Baud dial-out to another UNIX system:\
:el=^U^C^R^O^D^S^Q:ie=%$:oe=^D:tc=dial1200:
unix300|300 Baud dial-out to another UNIX system:\
:el=^U^C^R^O^D^S^Q:ie=%$:oe=^D:tc=dial300:
...stuff deleted...
dial2400|2400 Baud Hayes attributes:\
:dv=/dev/tty19:br#2400:cu=/dev/tty19:at=hayes:du:
dial1200|1200 Baud Hayes attributes:\
:dv=/dev/tty19:br#1200:cu=/dev/tty19:at=hayes:du:
# Hardwired line
com1c|com1:dv=/dev/com1:br#9600:
com1b:dv=/dev/com1:br#2400:
com0:com0:dv=/dev/com0:br#9600:at=hayes:
------------------------------------------------
Ok, now if you are running as root you can use type 'tip com0'
and you should then be talking to your modem. I use kermit to
transfer files, and it wants to create a lock file in (not sure
about the exact path) /var/spool/uucp/lock or something along
those lines. I made the directory world writeable so I could
run kermit with my own uid, rather than root.
Also, you may need to add an entry in /etc/remote for com0.
Thanks also to thombsr@liciren.li.co.uk for information on how
to do this.
New problems have surfaced with the latest releases of NetBSD.
It seems that the paradigm that the com port used to use was
'less than complte' and much of the code has been replaced.
This provides for some interesting new problems. The first is
that the Carrier Detect line is no longer ignored, as it was
before. This means that programs like kermit and tip/cu either
have to be told explicitly to ignore the CD line (in kermit,
for example, you would use the 'set carrier off' in your .kermrc)
or you need to use the 'stty -f /dev/com? clocal' command before
you open the port.
If you have trouble getting the settings to 'stick' it is because
the ports are now initialized to known settings on the last close
of the port. A workaround for this is to use the command:
sleep 1000 < /dev/com?
tip ...
{ or }
kermit ...
This will keep the port open for about 12 minutes while you do
whatever it is you need to do. Once the port is open and your
connection established, the port will not reset until the final
close.
Another symptom of this malady is described below.
When I 'set line' in kermit, it hangs until I hit my escape
character (^\). It *does* set the line (and the rest of my session
is normal), but it's annoying because I can't put 'set line' in
my .kermrc.
The answer is available in the kermit documentation as well as
here (now). The following commands are MY .kermrc file:
set carrier off
set line /dev/com2
set speed 38400
set modem hayes
set dial speed on
set dial timeout 60
The 'set carrier off' tells kermit to ignore the Carrier Detect
line from the modem, and to proceed as if the command were
there. This will (as far as I have ever been able to find
out) fix the specific problem with not being able to set the
line.
5.3.3.2 Adding a modem to NetBSD.
5.3.3.3 Adding a modem to FreeBSD.
5.3.3.4 Adding a Dial-in/Dial-out FAX to NetBSD or FreeBSD.
First, here is the known working configuration for these
instructions:
- HylaFAX 3.0 beta 100.
- Zoom VFX V.32bis Faxmodem;
- Rockwell datapump.
1: Start faxq from rc.local, no options on the command line.
Add a line to your /etc/rc.local which starts up the faxq
program. Do not include any options on the command line.
2: Stary faxgetty from init, i.e. a line in /etc/ttys.
I use the non modem control device; however, it's nonstandard
hardware and I've modified the driver to always return sighup
on lost carrier to solve some sticky problems with non modem
control devices never getting SIGHUP's.
Basically, I just did as the directions said to do. I ran
'faxaddmodem' script to configure the type of modem. I did
have to simplify some lines in the script (the ones executed
in a subshell) since I think my version of bash doesn't handle
subshells correctly.
RTFM and you should be OK unless your modem is braindead and
stupid, not too unlikely tho given the current state of Fax
modems... B^(.
5.3.4 What is the trick for getting Kermit to work with rz and sz?
Add these lines to your .kermrc file. They should do the trick.
define sz !sz \%1 \%2 \%3 \%4 \%5 \%6 \%7 \%8 \%9 < \v(line) > \v(line)
define rz !rz \%1 \%2 \%3 \%4 \%5 \%6 \%7 \%8 \%9 < \v(line) > \v(line)
5.4 Tape Drives
This section should help out for those of you that have either
never used tape drives before, or only have experience with them
as non-Unix devices.
5.4.1 Does the tape need to be formatted?
It depends, but I think usually not. And when it is necessary,
I don't know how it would be done. One thing is for certain,
though, first.... NEVER use the block devices.. erase them and
forget you ever saw them. All operations on tape should be to
the character device (rst0).
5.4.2 If I execute the command 'st -f /dev/st0 status', I get:
Archive/Tandberg? tape drive, residual=0, blocksize=512
Density: high = 16 (0x10), medium = 15 (0xf), low = 5 (0x5)
ds=0
er=0
so to write to tape at high-density (QIC-150), presumably I want
to use a device with minor number +4 (in st.c, density is computed as
minor >> 2 & 0x03, where low density == 3 and high == 1):
You have the idea.. density is controlled by bits 2 and 3
00 = default
01 = hi density
10 = medium density
11 = low density,
Unless the driver knows about you kind of drive the density values
may need to be set by hand before they make any sense.
5.4.3 When is erst0 used?
e stands for 'eject' and is bit 1 of the minor..
e.g. eject on close.. many devices can't actually do this.
There is actually a method to this whole thing:
r = raw (rst0)
e = eject (erst0)
n = No rewind (nrst0 or maybe nerst0)
5.4.4 How is density (bpi) computed? I am using 3M DC 6250 cassettes
which have a 250MB capacity on the Viper 150. But computing the
bits/inch based on 250MB/tape-length (1020 ft.), I get a density
of 171335 bpi, which is nowhere near the 10000 bpi associated
with QIC-150 in the st(1) man page. Why the discrepancy?
These cartridge tapes are written in narrow tracks which
alternately begin at opposite ends of the tape. Track 0 starts
at the beginning of the tape, and Track 1 starts at the other
end, etc.
So, how many times does the tape go backwards and forwards? If
there are 17 tracks, your density is 170000 bpi if it is 10000
bpi per track. The more tracks, the lower the bpi/track.
5.4.5 How is an appropriate block size determined (and in what units
are they specified in the st(1) command)?
QIC 150 and below should stick to 512 byte blocks a write of
1024 bytes from the program will be written as 2 512 byte blocks
with no speed penalty. dd will think it's writing a 1024 byte
block but on tape it's 2 x 512.
Stick to 512 on QIC 150 or less if you ever hope to swap data
with anyone else.
5.4.6 From the 4.3BSD mtio(4) man page, it sounds like data is typically
(traditionally?) stored on tape in eof-terminated sequences of
1K records.
5.4.6.1 Is st's notion of "file" the record sequence between two eof marks?
5.4.6.2 What about a "record"?
5.4.6.3 Is a "record" one "block", as determined by st's "blocksize" command?
If not, what is the connection between them?
5.4.6.4 Can I change the "record" size?
5.4.6.5 When would I want a block size that is different from the default?
1KB is the size of writes used by dd or whatever. QIC specifies
512 byte records (well at least its what people use..) Whatever
you write in will be broken into 512 byte sections. They must be
multiples of 512 though.
If you have written to a tape, a close will automatically append a
filemark (eof mark). You may read the 512 byte blocks back as
512 byte records or as 1024 byte records (in which case you'll
get 2 at once). The bigger the unit, the more efficient.
5.4.7.1 How do I write several archives to a single tape? I tried without
success:
$ st -f /dev/rst4 rewind
$ tar cf /dev/nst4 archive1
$ st -f /dev/nrst4 weof
$ tar cf /dev/nst4 archive2
$ st -f /dev/nrst4 weof
First: throw away the block devices.
'n' stands for 'No-Rewind-on-close' and will leave the tape
positioned ready for another file e.g.
tar -cf /dev/nrst0 archive1
tar -cf /dev/nrst0 archive2
5.4.7.2 Later, I would expect to be able to access, say, archive3 via the fsf
directive to skip over the first two archives. What is the correct
sequence?
st -f /dev/nrst0 rewind
st -f /dev/nrst0 fsf 2
tar -xf /dev/rst0 {files}
5.4.8 Since the Viper 150 writes on QIC-150/120, I guess I don't need
to worry about writing variable-length records? How about reading
a tape written with variable-length records. Is this possible
with the Viper? If so, what's involved?
Who would have written it? :-)
Presently you can't. You`re right. Don't worry about it.
The new 'st' changes will change this somewhat, though.
5.4.9 The very scant documentation that came with my drive mentions
a "selectable buffer disconnect size," whose default is 16K.
This is evidently the "maximum number of bytes that can be
sent over the SCSI bus during a single data transfer phase."
What's that? How is it connected st's "blocksize" command?
Do I want to use 16K blocks, or might I even want to set the
disconnect size to a higher value?
This suggests that 32 512 blocks will be written at a time.
This jives with the tape format for some of the lower density
cartridges (QIC-40 and 80, for example). The tape is written
in blocks of 32 512-byte blocks, with the last three being used
for Error Correction Codes.
Use dd or tar with 16 k blocks and 32 x 512 byte blocks will be
written.
5.4.10 What is "streaming"? When I tar a directory of files to tape,
I notice that the tape often stops. Streaming means it doesn't
stop? How would I get the viper 150 to stream using tar or cpio
or dump?
Use a bigger write size... (more efficient) Try 16k blocks.
5.4.11 Where are all the answers to the above and related questions
written down? Neither on the net nor in the 4.3BSD manuals
nor Administration text which I have could I find this stuff
covered!
They are in the FAQ :-)...
5.4.12 What else should I know? For example, it seems that a new tape
must stretched. How is this done?
Use a blowtorch and a pair of pliers; or you can use the
non-destructive method and run the tape through a complete fast
forward/rewind cycle to get it tight on the spindles.
5.4.13 My tape drive doesn't work.
OK. There are lots of reasons why it may not. The most obvious is
that there are no devices associated with the device in the kernel.
You can check this through the use of the 'dmesg' command. Look
for tape drives.
If your tape drive is connected to your floppy controller, it may
or may not be supported. Several manufacturers of QIC-40/QIC-80
minicartridge drives are supported natively in FreeBSD and
experimentatlly in NetBSD. Some aren't (mine for example, is not).
If your tape drive is a SCSI based drive, your guess is as good as
mine. I don't have one.
5.4.14 I am trying to restore a tape from a FreeBSD machine on a Sun.
What kinds of problems should I expect?
The default blocksize should not be a problem, since they are
both 20K. You may need to use "dd if=/dev/rst0 conv=swab |
<archiver>" instead of extracting directly from tape, just
in case the byte order causes a problem. This is especially
true if you don't use the 'a' and 'c' options in cpio, for
example.
5.5 Network
Network devices for NetBSD and FreeBSD include many types of
Ethernet cards, as well as Serial Line IP and Point to Point
Protocol.
5.5.1 How can I get my system to work as a network router?
The first hurdle to overcome is that the default kernels do not
have the GATEWAY option compiled in. Without this, it is very
nearly impossible to use the kernel as a router.
Once you have the GATEWAY option compiled in, all sorts of things
magically start to work. If you haven't got the GATEWAY option
enabled, you can also use 'sysctl -w net.inet.ip.forwarding=1'
if you are using FreeBSD or NetBSD versions that support that.
Remember, once you build the new kernel, you will need to
install it in the root directory and reboot.
Once you have the forwarding option set, you will need to make
certain that you have not included the '-q' option to routed.
This should be in the routed_flags keyword in /etc/netstart. If
you are using multiple internal LANs, you may also want to
invest in gated instead.
For those folks that are not using routed, you will need to make
certain that you have a static route to your network provider
established.
To test your network capability, try running the following
command:
trceroute -s YOUR_ETHERNET_ADDRESS 129.186.150.150
Check to see where your packets are hanging up. It might be
that someone upstream from you has something broken instead of
simply assuming it is your fault.
5.5.2 I recently has a problem where I got a message that said "panic:
kmem_malloc: mb_map too small". What is the solution to this
problem?
The second hurdle is that sometimes you run out of cluster
allocation space in the kernel. This is probably network-related
and usually shows up when something is being done using the
network (like NFS). The way to get around this would be to
change the value of NMBCLUSTERS in your config file. NMBCLUSTERS
is set at 256 by default, and increased to 512 when the GATEWAY
option is active. To be very safe, you could add
options NMBCLUSTERS=1024
to your config file, and recompile. This is reported to work
with systems that crashed as soon as a large number of people
(75+) were connected to it.
5.6 I want to use my ZIP drive. Are there any weird things I need
to know?
One of the things that "just work" are ZIP drives. The -current
code for both FreeBSD and NetBSD handle ZIP drives very cleanly.
One of the unusual things about ZIP drives is that most people
don't know (and the man page is deliberately vague about) is
once a person has permission to write to the ZIP drive, they can
mount it onto a directory in their space. This is new with the
adoption of BSD 4.4, so it isn't really suprising that it is
new.
--
Dave Burgess (The man of a thousand E-Mail addresses)
386bsd FAQ Maintainer / SysAdmin for the NetBSD system in my spare bedroom
"Just because something is stupid doesn't mean that there isn't someone
that wants to do it...."