NetBSD Documentation: Other FAQs and HOWTOs

General questions about NetBSD

What changed between NetBSD releases? (top)

A relatively up-to-date list of all changes (including those in NetBSD-current) is available.

There is also a CHANGES file in the top level directory of each release, and maintained online for previous and the next release.

Is NetBSD a 64 bit operating system? (top)

NetBSD runs on both 32 bit and 64 bit CPUs. On the alpha CPU the kernel and userland run in a mode known as LP64, where ints are 32 bit, and longs and pointers (and the address space) are 64 bit. The amd64 and sparc64 ports runs in both 32-bit and 64-bit configurations. A 64 bit kernel (LP64) supports 32-bit applications in 64-bit mode. A 32 bit kernel, which uses 32 bits for all of integers, pointers and long integers (IPL32) can not run 64 bit programs.

On other 64 bit capable CPUs, such as R4000 and newer MIPS (used by a number of ports) NetBSD runs in 32 bit mode though is in the process of being modified to 64 bit.

All NetBSD ports support 64 bit arithmetic, both in the kernel and in userland programs. In the kernel this is used for various counters and for FFS, which is the 64 bit 'Fast File System'.

Is NetBSD exportable from the US? (top)

Up to (and including) NetBSD 1.4.x, everything except the security distribution (i.e. secr.tgz) was exportable from the US. The secr.tgz distribution file contained the libraries for crypt. Starting with NetBSD 1.5, crypto code is included in the NetBSD base distribution. Please refer to the NetBSD document on exporting restricted code or binaries.

For what is NetBSD used? (top)

In the words of Chris G. Demetriou (cgd@NetBSD.org):

“ NetBSD, in general, is meant as a "stable research platform" -- that is, a system that can be used for commercial, home, and research work... what _you_ do with it is up to you. In general, those of us working on NetBSD are trying to improve the system in whatever way we can -- support for more hardware, stability, performance, documentation... ”

See our features page for an idea of what opportunities NetBSD provides you with!

User configuration

Adding a user (top)

You can use the useradd(8) command to add users on recent NetBSD release, older releases can install the sysutils/user package to get the same functionality. Editing /etc/passwd will not work. If you want to edit the password database, use vipw(8). See also pwd_mkdb(8) for more information.

Changing the information displayed via `finger`? (top)

Use the chfn(1) and chsh(1) commands, or use vipw(8) to edit /etc/master.passwd

Problems logging in as, or 'su'ing to, root (top)

This covers a number of possible problems:

Unable to login directly as root

NetBSD will only allow direct root login on terminals marked as 'secure' in /etc/ttys (see ttys(5)). To be able to login directly as root across the network (not recommended) you will need to add 'secure' to the end of every network terminal line. For those who prefer not to have to edit each line you can (as root) type:
cd /etc cp ttys ttys.orig cat ttys.orig | sed 's/network$/network secure/' > ttys
Unable to 'su' to root

NetBSD checks the contents of the 'wheel' group in /etc/group for any attempt to su(1) to root. If it is empty (wheel:*:0:) then anyone with the correct password can su to root. Otherwise it should be a comma separated (no spaces!) list of the permitted accounts.
Unable to 'su' to anyone
Type "ls -l /usr/bin/su". The first 'word' on the output should be '-r-sr-xr-x'. If it is anything else (such as '-r-xr-xr-x'). Type (as root) "chmod 4555 /usr/bin/su". The 's' in that position indicates it is a 'setuid' binary, which in this case allows 'su' to run with root privileges.

This problem can be caused by not giving the '-p' (preserve permissions) flag to tar when extracting the distribution. If this was the case then re-extract all the tarfiles (except etc.tar.gz) as various other utilities will also be broken.

Adding something to your path (top)

For example, to add /usr/X11R6/bin to your path, edit the file .login in your home directory and add the following:

set path = ( /usr/X11R6/bin $path )

This only works for the C-shell csh(1), and derivatives such as shells/tcsh. For sh(1) and derivatives (like shells/bash2), the correct syntax is:

PATH=/usr/X11R6/bin:$PATH
export PATH

Setting something for all users (top)

You can set something for all users running csh(1) or shells/tcsh by adding it to one of the following files:

/etc/csh.cshrc

This is run every time a new shell is opened, before any .cshrc in the user's home directory.
/etc/csh.login

This is run when the user logs in, after anything in /etc/csh.cshrc, but before any .login in the user's home directory.
/etc/csh.logout

This is run when the user logs out, before any .logout in the user's home directory.

Cursor key editing or filename completion in a shell (top)

This depends on the type of shell you are using:

sh(1) (Bourne shell)

This shell is used in single-user mode, and can enable command line editing with the command 'set -E'. This can be added to the /.profile file so you always have command line editing in single-user mode.
csh(1) ('C' shell)

This is the standard user shell, it supports filename completion ('set filec' and use the ESCAPE key) but does not support command line editing. However you can install shells/tcsh, or shells/bash2 from the NetBSD packages collection, both of which have filename completion and command line editing enabled by default.
ksh(1) (Korn Shell)

This is a public domain version of the Korn Shell, aka pdksh. It supports command line editing in both Emacs or Vi mode ('set -o {emacs,vi}'). In Emacs mode, use two escapes for file completion (use the command 'bind "^I"=complete' to change it to TAB instead). Use 'set -o vi-tabcomplete' or 'set -o vi-esccomplete' to get Vi mode with completion by TAB or ESC, respectively.

The public domain korn shell is part of the NetBSD base system, but is also available from the NetBSD packages collection, under shells/pdksh for other operating systems. The KSH-93 is also available in shells/ast-ksh.
bash (GNU Bourne Again Shell)

'bash' is another extended shell with command line editing, this time based on 'sh'. You can install it from the NetBSD packages collection, under shells/bash2.
tcsh (an extended 'C' shell)

'tcsh' is a near perfect superset of csh, with many useful extra features, including filename completion (use TAB) and command line editing. You can install it from the NetBSD packages collection, under shells/tcsh.
Other shells

The NetBSD packages collection also contains various other command line shells under the shells category.

Specifying a language for messages (top)

NetBSD uses the 'LANG' environment variable determine which language to use for error messages and suchlike. This can be set in csh or tcsh with the command 'setenv LANG XX' where XX is the two letter country code. For a list of supported codes type 'ls /usr/share/nls'.

Please note: Only certain messages have been translated. To test, set the LANG variable and type

  cd /a_directory_that_does_not_exist

Printing and Scanning

Printing man pages as PostScript (top)

Man pages are held in nroff(1) source in /usr/share/man/manX/manpage.X, where X is the section number (eg: 1 in the case of nroff). To convert them into PostScript one can use:

groff -Tps -mandoc /usr/share/man/manX/manpage.X

For example to convert the ls(1) man page into PostScript and send it directly to the default printer via lpr(1) you can use.

groff -Tps -mandoc /usr/share/man/man1/ls.1 |lpr

If you wish to print on a different type of printer you may want to use the print/ghostscript package from pkgsrc.

Printing to a remote SMB printer from NetBSD (top)

Please see this page for details.

Configuring a HP JetDirect Card via DHCP (top)

If you have a JetDirect print server such as the JetDirect EX Plus, you can use dhcpd(8) under NetBSD to configure the print server. The steps are: Create an entry in /etc/dhcpd.conf for the JetDirect server:

options jd-tftp-cfg		code 144 = string;
options jd-tftp-server		code 150 = string;

allow				bootp;

# [...]

host hp690c {
	fixed-address		hp690c;			# set to the DNS name or address of the JetDirect
	hardware ethernet	00:60:b0:00:11:22;	# set to the MAC address of the JetDirect
	option	log-servers	servername;		# set to the DNS name or address of log server
	option	jd-tftp-server	servername;		# set to the DNS name or address of tftp server
	option	jd-tftp-cfg	"hpnp/hp690c.cfg";	# name of the JetDirect config file
}

See dhcpd.conf(5) for more information on this file.

Make sure that tftpd is enabled in /etc/inetd.conf, Note the argument to "-s" on the tftp line of /etc/inetd.conf. This is the tftpd(8) root directory. For the rest of this example it is assumed that the tftpd(8) root directory is set to /tftpboot.

Make sure that dhcpd is enabled in /etc/rc.conf.

Next create a Network Printer Interface (NPI) configuration file. In this example, the file is /tftpboot/hpnp/hp690c.cfg. This file should look something like:

name: printer name
location: Location of the Printer 
contact: Contact Person 
idle-timeout: 1800
banner: 0

If you made changed to /etc/inetd.conf, make sure inetd(8) re-reads inetd.conf(5) by doing kill -HUP `cat /var/run/inetd.pid`. When you reboot the JetDirect print server (or printer with HP JetDirect card), the server should get configured by DHCP. On some printers, you may have to explicitly enable BOOTP/DHCP configuration from the front panel.

Why doesn't sh suppress the "burst" page on an HP JetDirect? (top)

The "burst" page is a function of the host with the locally attached printer, or in this case, the dedicated hardware emulating lpd(8). When NetBSD is used as a remote spooler, with rm, it never adds an additional burst page. The JetDirect's "burst" page is an unusual feature in a dedicated device. HP (TM) actually calls it a "trailer". You can disable it as above, by setting banner: 0, or simply telnet to port 23 of the device, and issue the command banner 0, then quit.

Some JetDirect units may need their firmware upgraded to add the banner command. You will need to use the appropriate JetDirect network management tools available from HP to do this. You may need to set the SNMP read and write community strings to known values in order to perform this upgrade. To set these, use the following lines in the NPI configuration file.

get-community-name: read-password
set-community-name: write-password

(See the above section on configuring the JetDirect cards with DHCP.) Just be aware of the security ramifications of leaving these entries in a publicly available (via TFTP) configuration file...

Scanning under NetBSD (top)

NetBSD supports various different SCSI scanners using the graphics/sane-backends scanning package. Frontends for accessing the API provided by that package are in the graphics/sane-frontends package, or it can be called directly by graphics tools such as the graphics/gimp.

Note: in versions of NetBSD before 1.4.2 the in kernel ss(4) scanner driver may cause problems with graphics/sane-backends and some scanners - the solution is either to recompile a kernel without ss(4) and use the uk(4) device, or to upgrade to 1.4.2 or later.

Common (and less common) problems

/usr/bin/nroff: not found when running 'man' (top)

You need to install the 'text' set, text.tgz. See your base install notes.

'Unable to lock mailbox: Permission denied' when reading mail (top)

Check the permissions on the /var/mail directory. They should be 'drwxrwxrwt':

% ls -ld /var/mail
drwxrwxrwt  2 root  wheel  512 Nov  6 08:21 /var/mail

If they are not, as root run 'chmod 1777 /var/mail'.

netstat: kvm_read kvm_read: Bad address when running '`netstat -r`' (top)

This is nothing to worry about. Basically, the file /netbsd is not your current kernel. A number of programs (such as ps, who, systat etc.) and libkvm kvm(3), access /netbsd to learn what's going on in the kernel. So you can simply rename your current kernel to /netbsd to make this kind of error go away. Make sure that you are not overwriting a working kernel when you do this, though, unless that is exactly what you intend to do.

When I try '`netstat -r`' I get an endless stream of question marks. What's wrong? (top)

That's a mis-match between libkvm/netstat and /netbsd. Or it's that you're running a kernel that's not named /netbsd. Other symptoms of this problem are likely to be that who, ps, ifconfig, and systat will not work either. If you update your binaries and your kernel at the same time, then you should be OK.

`w`, `ps`, and `netstat` do not work after upgrading to a new kernel (top)

One of two things. Either your currently booted kernel isn't named /netbsd or else you have a mismatch between your kernel and the binaries you are using. In the first case, simply making a link from your currently booted kernel to /netbsd will solve the problem.

In the second case, dynamically linked binaries can usually be fixed by upgrading libkvm to match your new kernel. Statically linked binaries need to be replaced with more recent versions. Since they are statically linked, if you are going to recompile them yourself, you need to rebuild libkvm.a before you rebuild the program in question.

Thanks to John Wittkowski (jpw@netscape.com), here is a list of most of the programs (besides /bin/ps) that depend on libkvm (all of these are in /usr/bin):

fstat
gdb
ipcs
netstat
nfsstat
systat
uptime
vmstat
w

After changing kernels, 'ps' reports "proc size mismatch" (top)

Like the previous three questions on this subject, the answer is most likely that your libkvm is out of sync with your kernel or binaries. To solve this problem, you can either get a binary distribution which matches your kernel, or you can build you own by following the instructions below:

If you get the "proc size mismatch" error and you determine
that you need to update your libs, here's what to do:

1. Get all the source code. If you're not willing to do this
   and recompile things than you'll have to find someone who
   will do it for you and you can try installing everything
   by hand.

2. Make sure that your include files are up to date. Do 
   this by:
      cd /usr/src
      make includes
   This will take a while. I had some trouble with this
   because some of the Makefiles didn't have the INSTALL
   variable defined. Whenever the "make includes" failed,
   I went to the last directory listed and added the
   following line to the Makefile:
      INSTALL=/usr/bin/install
   I had to do this several times before the make finished
   without any errors.

   (If you make sure that /usr/bin/make and all the files in
   /usr/share/mk are up to date first, the above difficulties
   can probably be avoided)

3. Rebuild the libkvm and install it:
      cd /usr/src/lib/libkvm
      make
      make install
   Note that in order to get the libkvm to compile on
   my system I had to add the following link:
      cd /usr/include/machine
      ln -s ../m68k/kcore.h kcore.h
   This may have been a quirk of my system so try 
   compiling without it first.

4. Then rebuild the binaries that are STATICALLY linked to
   libkvm. The only statically linked program that I am 
   aware of is "/bin/ps". To rebuild ps, simply:
      cd /usr/src/bin/ps
      make
      make install

5. You may or may not need to rebuild the binaries that
   are dynamically linked to libkvm. This is because (I
   think) if the major version number of the lib changes
   then the old binaries will expect the old version
   number and not work with the newer version of the lib.
   For example, my old libkvm was libkvm.so.4.0. The new
   one was libkvm.so.5.0. Without recompiling the 
   dynamically linked binaries, it would still complain
   about "proc size mismatch" (if the 4.0 lib was still
   there) or some lib missing error (if the 4.0 lib
   had been removed from /usr/lib). If the version minor
   number changes (4.0 to 4.1, for example), I think it
   will run with a warning and so you may not need to
   recompile all of these things.

   The dynamically linked binaries that I am aware of
   will give the  "proc size mismatch" error (if the 4.0
   lib was still there) or some lib missing error (if the
   4.0 lib had been removed from /usr/lib). If the version
   minor number changes (4.0 to 4.1, for example), I think
   it will run with a warning, and you may not need to
   recompile all of these things.

   The dynamically linked binaries that I am aware of
   are:
      /usr/bin/fstat
      /usr/bin/gdb
      /usr/bin/ipcs
      /usr/bin/netstat
      /usr/bin/nfsstat
      /usr/bin/systat
      /usr/bin/uptime (linked to /usr/bin/w)
      /usr/bin/vmstat
      /usr/bin/w
   Note that /usr/bin/uptime is a link to /usr/bin/w and
   will be set up properly when you do the "make install"
   for w.

   To recompile these, do the following:
      cd /usr/src/usr.bin/<cmd>
      make 
      make install
   For example, to recompile /usr/bin/vmstat:
      cd /usr/src/usr.bin/vmstat
      make
      make install

Much thanks to John Wittkowski (jpw@netscape.com) for providing such a detailed answer for this one.

A little background: the reason that these system utilities need to grovel through kernel memory to get information, including going through all the trouble mentioned above, is that this method also works on kernel crash dumps, which can be quite a useful feature when debugging kernels.

'hash map "Alias0":unsafe map file `/etc/aliases`: No such file or directory' is displayed (top)

You can fix this by typing 'newaliases' as root.

Sendmail reports this when the aliases(5) file is not present. Note that sendmail(8) actually uses /etc/aliases.db (which is a fast lookup hash(3) map built from /etc/aliases), so the error message can be a little misleading. This should only really be noticed in NetBSD 1.3 to 1.4, other versions have a different /etc/sendmail.cf which instructs sendmail to automatically rebuild /etc/aliases.db as needed. If you are not running at least NetBSD 1.4.1 you should probably upgrade to a newer formal release.

How do I prevent `portmap` messages from showing up in shells, when logged in as root (top)

Use security/sudo and never log in as root.

If you must log in as root, then you can edit your /etc/syslog.conf file and remove the necessary facilities to prevent the annoying messages. In most cases, it's sufficient to disable auth.debug in the line reading:

*.notice;auth.debug                                     root

This will eliminate those annoying portmap [pid]: connect from some.other.host ... messages. Be sure to kill -HUP the pid of your syslogd process.

You may also wish to look at the FAQ entry for using an xconsole.

How do I get umlauts and other localised characters displayed? (top)

Per default, NetBSD 1.6 has already numerous locales installed. You can find them in /usr/share/locales. Choose the locale most suitable for your location and set the LC_ALL environment variable accordingly. Note that it is not sufficient to set LC_ALL to, say, ``de_DE'', but that you should use the full locale, such as ``de_DE.ISO8859-15''.

For older versions of NetBSD, here's a tip from the mail archives:

Grab locale.tgz
Unpack it in /: tar plzvxCf / locale.tgz
In your shell, set LC_ALL=iso_8859_1 (make it permanent in .cshrc, .profile, ...)
Restart your shell or re-login
See your shell, vi, etc. accept all these umlauts from your keyboard: öäüÄÖÜ...

Disks & Filesystems

Adding a disk to an existing system (top)

Halt the system
Connect the new drive
If your machine has a PROM or BIOS that can list drives, check that the drive is recognised:
- recent sparc: Press [STOP][A] before the boot starts, then at the "ok" prompt, probe-scsi.
- i386 with adaptec BIOS: Press [CTRL][A] before the boot starts to enter the BIOS.
- other: more details welcomed :)
boot
Check dmesg(8) output to confirm that the drive is recognized

Write a disklabel(5) including the partitions using disklabel(8)

(if the disk doesn't have a disklabel yet, you'll probably do something like:

 disklabel -i -I sd1 # ignore any "sd1: no disk label"

Partition 'c' is reserved for the entire NetBSD portion of a disk, and on some ports (notably i386), 'd' is similarly reserved for the entire disk. The following is a sample session setting the entire disk to be one filesystem 'e':

	partition> e
	Filesystem type [?] [unused]: 4.2BSD
	Start offset [0c, 0s, 0M]: 
	Partition size ('$' for all remaining) [0c, 0s, 0M]: $
 	e: 234375000         0     4.2BSD      0     0     0   # (Cyl.    0 - 232514*)
	partition> W
	Label disk [n]? y
	partition> Q

Inside disklabel -i you can use '?' for additional help.

For each filesystem added, newfs <filesystem> (where <filesystem> is the partition's raw device, like /dev/rsd1e, /dev/rwd1f, ...)
Put the filesystems into /etc/fstab
fsck(8) them:
fsck -f <filesystem>
mount(8) them:
mount -va

How to use soft updates (softdeps)? (top)

Frank van der Linden (frank@wins.uva.nl) has brought Kirk McKusick's trickle sync + FFS soft update code into the main tree.

To use it, do the following:

Compile a new kernel with "options SOFTDEP" in your kernel config file. This option is enabled in GENERIC kernels.
Run the new kernel.
In /etc/fstab, add the "softdep" mount option to all filesystems that you want to use softdeps with. For example change the line reading
/dev/wd0e /usr ffs rw 1 2
to:
/dev/wd0e /usr ffs rw,softdep 1 2
Reboot
Have fun!

Are there any measured improvements of softdep? (top)

The fact that meta-data isn't immediately written to disk makes the system feel a lot faster. No numbers are available to show this at this moment.

Paul Vixie has run a benchmark on storing and retrieving (in random order) 75.000 files into a directory, on both a system with and without softdep. The results show that file access time for both creating and locating files in large directories are generally better on systems with soft dependencies. (The softdep system in this case was FreeBSD 4.2, the non-softdep system was BSD/OS 3.1. Similar behaviour will be visible on a NetBSD system with and without softdep).

How resilient is the Fast File System (FFS), FFS with softdeps and NetBSD's ext2fs implementation with respect to unclean shutdowns? (top)

The FFS takes care to correctly order all metadata operations, as well as to ensure that all metadata operations precede operations on the data to which they refer, so that the file system may be guaranteed to be recoverable after a crash. The last N seconds of file data may not be recoverable, where N is the syncer interval, but the file system metadata will be. N is usually 30.

With softdeps running, you've got almost the same guarantee. With softdeps, you have the guarantee that you will get a consistent snapshot of the file system as it was at some particular point in time before the crash. So you don't know, as you did without softdeps, that, for example, if you did an atomic operation such as a rename of a lock file, the lock file will actually be there; but you do know that the directory it was in won't be trashed and you do know that ordering dependencies between that atomic operation and future atomic operations will have been preserved, so if you are depending on atomic operations to control, say, some database-like process (e.g. writing mail spool files in batches, gathering data from a transaction system, etc.) you can safely start back up where you appear to have left off.

NetBSD's ext2fs implementation gives you the traditional FFS guarantee about metadata (unlike the Linux implementation), so you can actually use it with more confidence than you can use the native ext2fs in Linux. The downside is that it's a bit slower, but that's because it actually does the right thing if the system crashes, instead of potentially eating your file system itself.

Moving /usr from root to a different partition (top)

Make sure that the desired new location (e.g. /dev/sd0e) does not overlap any existing partitions. Check the output from disklabel(8).
If it does not already contain a filesystem, run
# newfs /dev/sd0e
Mount it and copy across data:
# mount /dev/sd0e /mnt
# cd /usr
# pax -rw -pe . /mnt
# umount /mnt
Edit /etc/fstab and add '/dev/sd0e /usr ffs rw 1 2' (change /dev/sd0e for your situation).
Shutdown to single user mode and make new mountpoint:
# shutdown now
# mv /usr /usr.old
# mkdir /usr
Bring system up, test, and when all OK, 'rm -rf /usr.old'.

How to mount a 'read-only' filesystem read/write? (top)

Use mount(8)'s -u (update) switch: 'mount -u /'. More than likely, you booted single-user which mounts the root filesystem read-only by default. This gives you a chance to run fsck by hand before bringing the system up into multi-user. One other way to mount all the filesystems in /etc/fstab as read/write is to use 'mount -a'.

However, the best way to do this is to go into multi-user mode by exiting from your single-user shell. This will mount all of the filesystems in /etc/fstab in whatever mode they are listed.

When is it safe to use fsck? (top)

Only use fsck(8) on an unmounted or read-only filesystem. Running fsck(8) on a r/w filesystem is dangerous and could corrupt the filesystem. If you bring the system up into single-user mode, the system will be mounted read-only. From there you can do an 'fsck -p' to check all filesystems before bringing the system up into multi-user mode. If the filesystem has previously been marked "clean," and you still want to check it, you can use the -f flag to fsck(8).

If you do make any changes to the filesystem using fsck, it is probably best to type 'reboot -n' to reboot the machine immediately without syncing the disks.

For what is `/kern` used? (top)

It is used for mounting the kernfs filesystem. See the mount_kernfs(8) manual page.

For what is `/proc` used? (top)

It is used for mounting the procfs filesystem. Consult The Design and Implementation of the 4.3BSD UNIX Operating System by Leffler, McKusick, et al., p. 104-5, 436. Also see the mount_procfs(8) manual page.

This is normally used by creating a '/proc' directory as root, then adding the following to /etc/fstab (see fstab(5)):

/proc /proc procfs rw 0 0

As /proc in NetBSD is intended for process-related data only, other entries known from linux are not present. This can be changed by adding the "linux" mount option to the file system flags, though:

/proc /proc procfs rw,-olinux 0 0

Encrypted swap with cgd(4) (top)

Since the topic of swap encryption pops up from time to time, this information could be useful for some people who are interested in encrypting their swap space but are not sure how to set it up.

Currently there is one limitation, cgd devices are created with disklabel containing only one slice marked as 4.2BSD which is not suitable for swapping into. This is the reason why using cgd drive as a swap device is not straightforward. Roland Dowdeswell is looking into this and it should be resolved in the near future.

A real world example: Our swap device wd0b will be configured as cgd1, using 256 bit aes-cbc key. Step by step instructions follows:

  1) generate a paramsfile for the cgd device.  the default file would be
     /etc/cgd/wd0b but i will use /etc/cgd/swap for better illustration:

     # cgdconfig -g -V none -k randomkey aes-cbc > /etc/cgd/swap

  2) configure the cgd device over wd0b, (make sure wd0b is not used as a
     swap device at this moment):

     # cgdconfig cgd1 /dev/wd0b /etc/cgd/swap

  3) edit disklabel for cgd1, we need to create a valid swap slice.  save
     the disklabel to a file:

     # disklabel -e -I cgd1

     # /dev/rcgd1d:
     type: cgd
     disk: cgd
     label: default label
     flags:
     bytes/sector: 512
     sectors/track: 2048
     tracks/cylinder: 1
     sectors/cylinder: 2048
     cylinders: 128
     total sectors: 263655
     rpm: 3600
     interleave: 1
     trackskew: 0
     cylinderskew: 0
     headswitch: 0           # microseconds
     track-to-track seek: 0  # microseconds
     drivedata: 0

     4 partitions:
     #        size    offset     fstype  [fsize bsize cpg/sgs]
      d:    263655         0     4.2BSD      0     0     0   # (Cyl.    0 - 128*)

     create a swap slice and change label (that's not really necessary, just for
     our information):

     # /dev/rcgd1d:
     type: cgd
     disk: cgd
     label: swap
     flags:
     bytes/sector: 512
     sectors/track: 2048
     tracks/cylinder: 1
     sectors/cylinder: 2048
     cylinders: 128
     total sectors: 263655
     rpm: 3600
     interleave: 1
     trackskew: 0
     cylinderskew: 0
     headswitch: 0           # microseconds
     track-to-track seek: 0  # microseconds
     drivedata: 0

     4 partitions:
     #        size    offset     fstype  [fsize bsize cpg/sgs]
      a:    263655         0       swap                      # (Cyl.    0 - 128*)

     and save it to a file:

     # disklabel cgd1 > /etc/cgd/swap.disklabel

  4) set up the cgd device to be configured automatically at boot.  add the
     following line into your /etc/cgd/cgd.conf:

     cgd1 /dev/wd0b /etc/cgd/swap

  5) now we need to restore our disklabel to the newly created cgd device at
     boot time.  create /etc/rc.conf.d/cgd with this simple function:

     swap_device="cgd1"
     swap_disklabel="/etc/cgd/swap.disklabel"
     start_postcmd="cgd_swap"

     cgd_swap()
     {
             if [ -f $swap_disklabel ]; then
                     disklabel -R -r $swap_device $swap_disklabel
             fi
     }

  6) add the cgd device into /etc/fstab as a swap device:

     /dev/cgd1a none swap sw 0 0

That's all. Now you can activate your swap partition with 'swapctl -a /dev/cgd1a' or just reboot.

What you see above is a setup from my laptop.

   Device      1K-blocks     Used    Avail Capacity  Priority
   /dev/cgd1a     131827     8828   122999     7%    0

Boot-Related Questions

What is single-user mode? Why and how should you use it? (top)

'single-user' mode is when the kernel has booted, but only runs a single shell on the console. If you boot single-user only the root filesystem will be mounted, and it will be mounted read-only. Among the many uses for single-user mode are:

fsck your filesystem due to suspected corruption.
edit corrupted system configuration files in /etc.
install a new set of binaries which you downloaded or compiled yourself.
make a backup of your filesystem without worrying about interference from user processes.

You can enter single-user mode by running 'shutdown now' as root on a running system, or by booting the system with the '-s' flag (the method for doing this varies from port to port).

Once booted single-user you may find some of the following commands useful:

fsck -p ; mount -va

Check the filesystems with fsck(8), then make them available with mount(8).
set -E

Enable command line editing.
TERM=vt220 ; export TERM

Set and export the TERM variable, to permit editors such as vi(1) to work. Note: You may need to use a different terminal type from 'vt220', depending on the console type.
sh /etc/rc.d/network start

Bring up the network. If you have any nfs mounted filesystems you may want to do this before the fsck and mount. If you have a system prior to NetBSD 1.5, the command to use is sh /etc/netstart.

How to switch from single-user to multi-user (top)

Exit the single-user shell with "exit" or ctrl-D.

The system will boot up to multi-user mode, starting all kind of services, etc. Note that disks are not checked when going from single-user to multi-user mode!

How to read the boot messages? (top)

The system message buffer can be displayed with dmesg(8). This is usually (in 1.4 and later) automatically stored in /var/run/dmesg.boot immediately after each boot. If kernfs (see mount_kernfs(8)) is enabled, as in install kernels, you can also more /kern/msgbuf.

How to shutdown the machine? (top)

To drop back into single-user mode, use 'shutdown now'. To halt the machine, use 'shutdown -h now'. On some ports, you may need to use 'shutdown -p now' to actually power off the machine. Please see the shutdown(8) manual page for complete details.

How to reboot the machine? (top)

Use 'shutdown -r now' or 'reboot'. See the shutdown(8) or reboot(8) manual pages for further details.

Starting a program at boot time (top)

To start programs at system boot time, any commands can be added to /etc/rc.local, which will be ran at the end of the boot process.

As an alternative mostly for non-root users, the Vixie cron(8) (used in NetBSD)'s special @reboot time specification can be used to start a given program at cron(8) startup time, which usually coincides with the system booting. This also allows users without root privileges to run commands automatically at startup.

Memory/Swap Configuration

How much swap space (top)

On older unix systems it was recommended to have twice as much swap space as RAM. This was because active pages in RAM had to have pages allocated in swap, effectively limiting the total virtual memory to the size of the swap space. On NetBSD this is no longer true, so total virtual memory is approximately swap + RAM.

Under NetBSD there are three primary uses for swap space:

Additional virtual memory

Less active pages containing modified data can be migrated to swap space allowing virtual memory to be larger than physical memory.
An area for kernel core dumps

This is then read by savecore(8), on boot. For this to work the primary swap partition must be slightly larger than the total RAM.
For mfs(8) filesystems

Classically mounted on /tmp in order to speed up any programs using /tmp, by adding the following to the fstab(5) file:
swap /tmp mfs rw,-s=SIZE 0 0
where SIZE is in 512byte blocks. The space required is taken from the swap space as the filesystem is used.

The 'correct' value for swap is dependent on the usage of the system, but for some general rules:

A good base size is the larger of RAM or 32MB, plus the total size of any mfs(8) filesystems.
If you intend to run X(7), add the larger of 1/2 of RAM or 16MB.
Always try to split swap across as many disks as possible to increase the available swap bandwidth.
Guess high, rather than low - running out of swap can cause the system to have to kill running processes.

Adding more swap to a running system (top)

Systems should be configured with sufficient swap space, as per the How much swap entry.

There are three ways to add swap to a configured system:

Rebuild the system disk, using a larger swap partition.

While arguably the cleanest solution, this obviously involves significant disruption.
Add an additional swap partition.

When you add a disk it is usually a good idea to include a small 'b' partition for additional swap, and add an entry of the form:
/dev/DISKb none swap sw 0 0
to your /etc/fstab file to automatically enable on boot. (Where DISK is the disk name such as 'sd1' or 'wd2').
Create a swap file.

The quickest and simplest option is to create a file in an existing filesystem, and use it as additional swap space. While accessing this will be slightly slower than dedicated swap partitions (due to fragmentation and filesystem overhead), it is ideal for solving a temporary swap shortage.

To add 10 MBytes of swap space is as easy as selecting a filesystem with sufficient free space and:
# dd if=/dev/zero bs=1m count=10 of=/somefilesystem/swap # chmod 600 /somefilesystem/swap # swapctl -a -p 1 /somefilesystem/swap
The dd(1) command creates a 10 MByte file `/somefilesystem/swap'. This swap file needs to be chmod(1) 600 so that unprivileged users may not read its contents (swapctl(8) will reject world readable files). Then swapctl(8) command adds /somefilesystem/swap to the system swap space at priority 1. Priority 0 is the (default) highest priority, and since swapping to files is slightly slower we only want the system to use the file when conventional swap has all been used.

To make swapping to that file permanent and enable it on every reboot, put something like the following into /etc/fstab:
/somefilesystem/swap none swap sw,priority=1 0 0

Tuning Virtual Memory to improve performance on low-memory systems (top)

According to a post by Simon Burge on the tech-kern list, it is possible to greatly improve the performance of small-memory machines by tweaking the virtual memory settings. While it is in your best interests to read his entire post and decide for yourself what your settings should be, the gist is that you'll want to run sysctl -w vm.anonmax=95. According to the sysctl(3) page, this changes "The percentage of physical memory which will be reclaimed from other types of memory usage to store anonymous application data."