SGI admin Frequently Asked Questions (FAQ)
From: sgi-faq@viz.tamu.edu (The SGI FAQ group)
Newsgroups: comp.sys.sgi.misc
Subject: SGI admin Frequently Asked Questions (FAQ)
Date: 6 Jul 2001 05:59:41 GMT
Message-ID: <admin_994399215@viz.tamu.edu>
Reply-To: sgi-faq@viz.tamu.edu (The SGI FAQ group)
Archive-name: sgi/faq/admin
Last-modified: Tue Jun 20 1:00:03 CDT 2000
Posting-Frequency: Twice monthly
URL: http://www-viz.tamu.edu/~sgi-faq/
SGI admin Frequently Asked Questions (FAQ)
This is one of the Silicon Graphics FAQ series, which consists of:
SGI admin FAQ - IRIX system administration
SGI apps FAQ - Applications and miscellaneous programming
SGI audio FAQ - Audio applications and programming
SGI diffs FAQ - Changes to the other FAQs since the last posting
SGI graphics FAQ - Graphics and user environment customization
SGI hardware FAQ - Hardware
SGI impressario FAQ - IRIS Impressario
SGI inventor FAQ - IRIS Inventor
SGI misc FAQ - Introduction & miscellaneous information
SGI movie FAQ - Movies
SGI performer FAQ - IRIS Performer
SGI pointer FAQ - Pointer to the other FAQs
SGI security FAQ - IRIX security
Read the misc FAQ for information about the FAQs themselves. Each FAQ is
posted to comp.sys.sgi.misc and to the news.answers and comp.answers
newsgroups (whose purpose is to store FAQs) twice per month. If you
can't find one of the FAQs with your news program, you can get it from
ftp://viz.tamu.edu/pub/sgi/faq/
ftp://rtfm.mit.edu/pub/usenet/news.answers/sgi/faq/
(rtfm.mit.edu is home to many other FAQs and informational documents,
and is a good place to look if you can't find an answer here.) The FAQs
are on the World Wide Web at
http://www-viz.tamu.edu/~sgi-faq/
If you can't use FTP or WWW, send mail to mail-server@rtfm.mit.edu with
the word 'help' on a line by itself in the text, and it will send you a
document describing how to get files from rtfm.mit.edu by mail. Send the
command 'send usenet/news.answers/sgi/faq/misc' to get the SGI misc FAQ,
and similarly for the other FAQs. Send the command 'send
usenet/news.answers/internet-services/access-via-email' to get the
"Accessing the Internet by E-Mail FAQ".
You may distribute the SGI FAQs freely and we encourage you to do so.
However, you must keep them intact, including headers and this notice,
and you must not charge for or profit from them. Contact us for other
arrangements. We can't be responsible for copies of the SGI FAQs at
sites which we do not control, and copies published on paper or CD-ROM
are certain to be out of date. The contents are accurate as far as we
know, but the usual disclaimers apply. Send additions and changes to
sgi-faq@viz.tamu.edu.
Topics covered in this FAQ:
---------------------------
-1- DIAGNOSTICS
-2- How can I determine which release of IRIX I'm running and which
patches are applied?
-3- How can I determine my SGI's Ethernet (and/or FDDI) address?
-4- My SGI crashed and generated a file, /usr/adm/crash/vmcore.1. How
can I examine this file to see what crashed my system?
-5- How can I find out just about everything about my system at once?
-6- DISKS
-7- How big can files and filesystems be?
-8- My XFS filesystem is corrupt or inconsistent. How do I fix it?
-9- Does IRIX support sparse files?
-10- Why is /debug or /proc full of huge files?
-11- How do I remount the /proc filesystem after accidentally
unmounting it?
-12- How do I extend an existing filesystem onto a new disk?
-13- How do I know if I need more memory and/or swap space?
-14- How much swap space should I have per megabyte of memory?
-15- How can I increase my swap space?
-16- What are virtual and logical swap space? How do they work in IRIX
3.x, 4.0.x and 5.x?
-17- Why is there no way to set up a RAM disk under IRIX?
-18- BOOTING
-19- How can I boot directly into single-user mode?
-20- How can I boot from a non-default disk?
-21- How can I boot my machine using a server on the other side of a
router?
-22- How do I make a bootable tape from an IRIX CD?
-23- Why can't I boot one of the stand-alone programs on a tape or CD?
-24- INSTALLING
-25- Is it possible to remotely install IRIX over a network?
-26- Which IRIX CD is the program 'foo' on?
-27- How can I extract a single file from an 'inst' subsystem?
-28- Why doesn't 'inst' work?
-29- Why doesn't 'inst' work remotely?
-30- I just installed a new version of IRIX and file X is missing.
What should I do?
-31- I reinstalled an IRIX subsystem to restore a missing file or get
rid of a corrupted file, but it didn't help. Why not?
-32- Why do 'inst' and 'showprods' say that a subsystem is an "Unknown
product entry"?
-33- How can I install IRIX onto a second disk which I can then move
to another machine?
-34- How can I copy my system disk onto a second disk which I can then
move to another machine?
-35- How can I share parts of IRIX among multiple machines?
-36- Can I install a "tardist" archive without the "tardist" software?
-37- I'm running 6.5.x, why can't I install the compilers?
-38- NETWORKING
-39- Why isn't my network working?
-40- How can I measure my network's reliability?
-41- How do I add a static route?
-42- How can I make the 'slip' command advertise the Ethernet address
of the SLIP client?
-43- I've just edited inetd.conf, and nothing changed. Why?
-44- Why can't I 'rdist' files between Suns and SGIs?
-45- Why are there "satwrite failure: inetd" messages in my SYSLOG in
IRIX 5.3?
-46- What is causing the "get interface flags" error in my syslog?
-47- Why are network connections between SGIs and Suns much slower
than between SGIs and SGIs or Suns and Suns?
-48- How can I set up IP aliases?
-49- Where is ipfilterd documented?
-50- MAIL
-51- How can I set up 'sendmail' to pass (or not pass) 8-bit
characters?
-52- Why are my mailbox files changing ownership?
-53- Why isn't a valid user getting their mail?
-54- How can SGIs and Suns share a mail spool?
-55- What's an "unknown mailer error"?
-56- What's "mailbox: Error 0"?
-57- Why am I having problems with my NFS-mounted mail spool?
-58- Why are incoming mail addresses case-sensitive in IRIX 5.3?
-59- NFS
-60- How can I tell what hostname to use in /etc/exports?
-61- Why can't I export an NFS-mounted filesystem?
-62- Why can't Ultrix automount SGI filesystems?
-63- Why does 'tar' work strangely on a filesystem mounted from an
SGI?
-64- Why does 'df' report incorrect numbers for a filesystem mounted
from a Sun?
-65- Is 'pcnfsd' available for the SGI?
-66- Can I export a CD-ROM from my SGI to a non-SGI?
-67- How can I read an IRIX (EFS) CD-ROM on a machine which doesn't
use EFS?
-68- How can I get quotas to work on an NFS filesystem?
-69- Why can't some NFS clients NFS-mount IRIX 5.2 filesystems?
-70- What's NFS3?
-71- Why can't Solaris 2.5 clients read directories on NFS3-mounted
IRIX 6.2 filesystems?
-72- Why does my NFS server print "fhtovp_end VFS_GET failed" or
"fhtovp_end vp NULL" to the console/SYSLOG?
-73- PRINTING
-74- Why can't lp(1) read my file?
-75- How can I tell 'lp' to turn banner printing or page reversal off
or on?
-76- How can I use 'lpr' instead of or as well as 'lp'?
-77- How can I print to a printer with a JetDirect card?
-78- Why won't lpr(1) print very large files?
-79- SGI DAEMONS
-80- Why isn't the objectserver working?
-81- What is sending packets to the sgi-dog.mcast.net multicast
address?
-82- Why are the objectserver and directoryserver taking up so much
memory in IRIX 5.3?
-83- Why don't the desktop admin tools recognize root as a privileged
user?
-84- Why doesn't my desktop deal properly with the CD or tape icons?
-85- Why doesn't my mediad start up under IRIX 5.3?
-86- Why do cdman, cdplayer and capture say "permission denied"?
-87- Why can't 'cformat' format floptical disks, and/or why can't
'datman' play audio DATs?
-88- Why is famd hammering my NFS server?
-89- What is causing "/dev/imon: event queue overflow" in my syslog?
-90- BUGS
-91- Why do 'who', 'rusers', etc. show users who aren't really logged
in?
-92- What's wrong with ftpd in IRIX 5.2?
-93- Why isn't /usr/adm/SYSLOG being updated?
-94- I just edited /etc/inittab, and now I can't start up or shut down
my SGI! What's wrong?
-95- Why does timed say "bind: Cannot assign requested address"?
-96- Why does "ALERT: i2cWaitForXferDone: Xfer Done timeout ERROR"
appear in my SYSLOG?
-97- Why has NIS (YP) stopped working under IRIX 6.x?
-98- MISCELLANEOUS
-99- How do I set the number of processes allowed on my machine?
-100- Where can I get a termcap file for 'iris-ansi-net' to install on
my non-SGI system?
-101- How can I make my SGI understand strange terminal types from
other Unix systems?
-102- Can I change my login shell or other password-file info without
being superuser?
-103- How can I administer my Iris without a graphics terminal?
-104- How can I use the visual admin tools on a system with graphics to
administer a system without graphics?
-105- What's /etc/ioctl.syscon?
-106- Can I log out users automatically?
-107- How can I change kernel variables and/or rebuild my kernel?
Subject: -1- DIAGNOSTICS
Date: 15 May 1994 00:00:01 EST
These questions discuss how to find out things about your system.
Subject: -2- How can I determine which release of IRIX I'm running
and which patches are applied?
Date: 28 Apr 2000 11:58:53 PDT
On pre-6.5 systems, a 'uname -a' will give you the Irix version of
the currently running kernel. The 6.5.x release added the '-R' flag
which reports the current revision number (e.g. 6.5.6m). See the
uname(1) manpage for other options.
Of more general use, since the running kernel does not always
reflect installed software, is the versions(1M) command. 'versions'
with no arguments provides a full listing all the installed software
components. 'versions -b | grep patch ' provides a brief listing of
installed patches. Note that the use of patches is depricated in the
6.5 release stream. Software subsystems are now "updated" so in with
a few exceptions (e.g. systems running legato networker), there should
be no patches installed.
IRIX 5.2's System Manager ('chost') has the IRIX version number under
"IRIX Version" and a listing of installed software under "Software"
(the "Show Installed" button).
Subject: -3- How can I determine my SGI's Ethernet (and/or FDDI)
address?
Date: 4 Jun 1997 00:00:01 EST
Many thanks to Miguel Sanchez <miguel@oasis.csd.sgi.com> for
providing the original version of the following discussion, and to
Dave Olson <olson@sgi.com> for comments. Andrew Cherenson
<arc@sgi.com> reminded us that all these methods except the first
apply to FDDI as well, but we'll just say "Ethernet" below.
Every system on an Ethernet network must have a unique Ethernet
address for the network to operate properly. The physical Ethernet
address of your system is the unique number assigned to the Ethernet
hardware on your system. This unique number is assigned to the
manufacturer of your Ethernet hardware by the IEEE (formerly by
Xerox, one of the original developers of Ethernet). This is not to be
confused with the IP address, which can be set arbitrarily.
You may need to determine your system's Ethernet address if your
network manager requires it before connecting your system to a
network. How to do so depends on whether IRIX is running and what
operating system version is loaded. Method 1 only provides the
Ethernet address of the primary interface. If you have multiple
Ethernet interfaces (boards) in a system, use method 2, 3, 4 or 5 to
determine the address(es) of any other interface(s).
METHOD 1: eaddr
If IRIX is not running, and the system is a Personal IRIS (4D20,
25, 30, or 35), Indigo, Crimson, Onyx, Challenge, Indy, O2,
or Indigo2, you can obtain the Ethernet address by typing 'eaddr'
(older machines) or 'printenv eaddr' (newer) at the PROM monitor .
prompt. On some machines (4D30 or later) you can say 'nvram eaddr'
while IRIX is running to get the same result.
METHOD 2: netstat
Under IRIX 4.0.1 or later, you can use the netstat command. For
example,
% /usr/etc/netstat -ia
Name Mtu Network Address Ipkts Ierrs Opkts Oerrs Coll
ec0 1500 siligrph luey7 7765678 21648 384477 0 30338
192.48.200.251
192.0.0.1
08:00:69:06:17:c2
lo0 32880 loopback localhost 41438 0 41438 0 0
192.0.0.1
As seen on the fourth address line, the address of the system
luey7's primary Ethernet interface, "ec0", is 08:00:69:06:17:c2.
METHOD 3: arp
You can obtain the Ethernet address of a Silicon Graphics system
by using another system on your network. 'ping' the system whose
Ethernet address you want, then use 'arp'. For example,
% /usr/etc/ping -c 1 luey6
PING luey6.sgi.com (192.48.200.250): 56 data bytes
64 bytes from 192.48.200.250: icmp_seq=0 ttl=255 time=0 ms
----luey6.sgi.com PING Statistics----
3 packets transmitted, 3 packets received, 0% packet loss
round-trip (ms) min/avg/max = 0/0/0
% /usr/etc/arp luey6
luey6 (192.48.200.250) at 8:0:69:6:c:40
%
METHOD 4: NetVizualyzer/FDDIVizualyzer and the like
SGI's NetVizualyzer/FDDIVizualyzer network monitoring software
and at least one public domain equivalent ('netman', at
ftp://ftp.cs.curtin.edu.au/pub/netman/) allow you to find the
Ethernet address corresponding to any IP address. Read the
manual.
METHOD 5: System Manager
The Network Setup part ('cnet') of the Indigo Magic System Manager
tool ('chost') shows the Ethernet address of each interface.
4DDN: A Special Case
DECnet uses a one-to-one relationship between the DECnet node ID
and the Ethernet address. If the DECnet address is changed the
Ethernet address is changed. DECnet Ethernet addresses always
start with aa:, so you can identify systems running DECnet with
'arp -a'.
4DDN is Silicon Graphics' DECnet interconnection product. The
Ethernet address of an IRIS running 4DDN will change when 4DDN is
started. Method 1 will return the original Ethernet address for
the system. Methods 2-5 will show the Ethernet address currently
in use.
sysinfo
/etc/sysinfo is intended to return a unique identifier, which on
some machines includes part or all of the Ethernet address. This
is best regarded as an amusing coincidence, like HAL's name in
"2001". Don't rely on it.
You can find an Ethernet address from a program most efficiently by
using the SIOCGIFADDR ioctl on a raw socket (SOCK_RAW) using the
RAWPROTO_SNOOP protocol (thanks to David Peter
<davep@isltd.insignia.com>) but the program must run as root. If you
can't run as root, call one of the above programs with system().
Subject: -4- My SGI crashed and generated a file,
/usr/adm/crash/vmcore.1. How can I examine this file to
see what crashed my system?
Date: 28 Apr 2000 11:58:53 PDT
Under Irix 5.3 and above a system crash will generate a number of
files in the /var/adm/crash directory. These files appear in numbered
sets, with each number (n) corresponding to an event. These files
include: unix.n (a copy of the kernel), vmcore.n.comp (a compressed
dump of the memory), and analysis.n (an analysis file generated by the
system using the icrash kernel debugger). Note that these files are
created only is savecore is chkconfig'd on.
Simply viewing the analysis file can often lead to clues regarding the
reason behind a system crash. 'icrash' can also be run manually. See the
icrash(1) manpage for details.
SGI's 'icrash' utility can generate helpful reports and allow
experienced administrators and support people to sift through the core
file. Get patch 813 for IRIX 5.3 or patch 769 for IRIX 6.1. icrash
is part of eoe.sw.unix (and therefore installed by default) on IRIX
6.2 and above.
For machines which lack the icrash utility:
dbx -k /var/adm/crash/{unix,vmcore}.#
t
&putbuf/1000s
Some machines have a special 'dbx' for crash dumps,
/usr/adm/crash/dbx. If it exists, use it instead of /usr/bin/dbx.
The IRIX 5.x Electronic Services package includes a script, 'crpt',
which does this and more automagically. A copy of the IRIX 5.2
version lives at ftp://viz.tamu.edu/pub/sgi/software/admin/crpt.
Subject: -5- How can I find out just about everything about my system
at once?
Date: 06 Apr 1996 00:00:01 EST
IRIX comes with hinv(1M) and gfxinfo(1G). If that's not enough, get
Michael Cooper <mcooper@acamar.usc.edu>'s 'sysinfo', recently ported
to IRIX, from ftp://usc.edu/pub/sysinfo/. Change "PROG" in the
Makefile from "sysinfo" to "Sysinfo" so it doesn't get in the way of
/sbin/sysinfo, which isn't nearly as entertaining but which other
software may require. This is probably also a good place to mention
Dave Olson's 'scsicontrol', at
http://reality.sgi.com/employees/olson/Olson/.
Subject: -6- DISKS
Date: 15 May 1994 00:00:01 EST
These questions deal with disks and swap space.
Subject: -7- How big can files and filesystems be?
Date: 15 Jun 1996 00:00:01 EST
A file on an EFS filesystem can be only 2G in size. The filesystem can
be no more than 8G in size, whether or not it is on a logical volume.
A physical or logical volume can be larger than 8G, but not usefully
so: mkfs still can't make a filesystem larger than 8G on it. (Each
block (512 bytes) has a unique number which must fit into 24 bits; 512
* 2^24 = 8G. See also inode(4).)
Note that 'fx' had a bug in versions of IRIX before 4.0.5H which
prevented it from *exercising* disks larger than 2G, but other 'fx'
functions and other parts of IRIX have no trouble with disks
(filesystems) up to 8G.
Files and filesystems on SGI's new XFS filesystem can be 1 terabyte
(1K gigabytes) in size. On 64-bit machines running IRIX 6.2, a file
will be able to be 9 exabytes (9 gigagigabytes) in size and a
filesystem will be able to be 18 exabytes in size. See the misc FAQ
under "WWW pages" for a pointer to a demonstration.
Subject: -8- My XFS filesystem is corrupt or inconsistent. How do I
fix it?
Date: 28 Apr 2000 11:58:53 PDT
Under IRIX 6.2, get the latest XFS rollup patch and install it.
(At the time of this writing, the latest XFS rollup patch was patch
1768.) This patch comes with xfs_repair, a tool that can often fix
corrupted XFS filesystems. (If the partition you wish to repair is
your boot partition, you will have to boot from the miniroot that
comes with the patch and run xfs_repair from it.) Actual installation
of the patch requires the installation of one of the recent kernel
rollup patches; read the XFS rollup patch release notes before doing
anything else with it.
Under the 6.5.x release stream, 'xfs_repair' and 'xfs_check' are
installed by default as part of eoe.sw.base. Please note that xfs
check and repair operations are performed on unmounted filesystems.
See the xfs_repair(1) manpage for details. Also, it is best to
do an initial scan on a damaged filesystem using 'xfs_repair -n',
(no modify mode) as extremely corrupted filesystems can occasionally
cause an xfs_repair to fail with a core dump.
Subject: -9- Does IRIX support sparse files?
Date: 09 Dec 1995 00:00:01 EST
EFS does not; XFS does.
Subject: -10- Why is /debug or /proc full of huge files?
Date: 10 Dec 1993 00:00:01 EST
Those aren't disk files, they're interfaces to running processes.
Read the debug(4) (IRIX 4.0.x) and/or proc(4) (IRIX 5.x) manpages.
Subject: -11- How do I remount the /proc filesystem after accidentally
unmounting it?
Date: 4 Jun 1997 00:00:01 EST
Run /etc/mntproc as root.
Subject: -12- How do I extend an existing filesystem onto a new disk?
Date: 28 Apr 2000 11:58:53 PDT
Back up the existing filesystem (just in case) then run 'mklv' and
'growfs'. 'mklv' and 'growfs' are nondestructive, so you don't need
to restore the backup unless you screw up. Don't use 'mkfs', which
does destroy existing data.
Under modern IRIX versions (w/ the XFS filesystem) lv (the original
logical volume manager) has been replaced with xlv. The requisite
subsystems are _not_ installed by default. If you wish to use logical
volumes, you must install eoe.sw.xlv, and if you wish to use mirroring
(RAID 1), you must also install eoe.sw.xlvplex. This option requires a
license be purchased from sgi.
To simply grow the filesystem to another disk, you must first create a
concatenated logical volume. This will change the volume headers of the
target disks and let the system know that they should be treated as one
volume. You can then use an 'xfs_growfs' to expand the filesystem on to
the new disk. see the manpages for xlv_make(1), xlv_mgr(1), and
xfs_growfs(1) for more details.
Please not that it is NOT POSSIBLE to have a successfully striped or
concatenated root filesystem. The only available logical volume
configuration for the root filesystem is a mirror (RAID1). Any
attempt to use another logical volume configuration will most
likely work until the next system reboot. Since the Logal volume
management software funtions on the OS level, it is unavailable
during the inital phases of the boot process. Syncing of the
mirrored filesystem will occur after the system comes up.
Subject: -13- How do I know if I need more memory and/or swap space?
Date: 20 Feb 1994 00:00:01 EST
If processes are killed due to lack of memory/swap, you need more
memory and/or swap space. If your CPU is always waiting for swapping
(run 'osview' and look at the "%Swap" entry under "Wait Ratio") you
need more memory.
Subject: -14- How much swap space should I have per megabyte of
memory?
Date: 4 Jun 1997 00:00:01 EST
An oft-recommended ratio is X memory:2.5 X swap, but this may be too
low. Decide how much of your favorite program (plus IRIX) needs to
be resident for good performance and how much doesn't, and make sure
you have enough memory for the former and enough memory plus swap for
the latter. Put "rmem" and "swp" in your ~/.grosview file, run
'gr_osview' and run your favorite program to see what it needs.
Subject: -15- How can I increase my swap space?
Date: 28 Apr 2000 11:58:53 PDT
The Jan/Feb 1993 and May/Jun 94 Pipelines have detailed writeups on how
to do this in IRIX 4.0.x and 5.x respectively. The Jul/Aug Pipeline
has a correction to the latter article. If you like you can call the
TAC and have them fax you the very latest version.
It is also possible to add swapfiles within the filesystem structure.
You can allocate space for swap in the filesystem by doing a 'mkfile'
and adding this file to the system as swap space. See mkfile(1) for more
information. The created file can be added by using a 'swap -a <file>'.
If you intend this to be a permanent resource, make sure that you add
an appropriate line to the /etc/fstab file. See the fstab(4) manpage
for details.
Subject: -16- What are virtual and logical swap space? How do they
work in IRIX 3.x, 4.0.x and 5.x?
Date: 05 Jul 1994 00:00:01 EST
Two terms whose meanings should already be clear: Physical swap space
is an area on disk, either a partition or (in IRIX 5.x) a swap file.
Virtual memory is the sum of physical memory and swap space.
IRIX 3.x accepts a memory request only if enough virtual memory is
free. Even if a process isn't using most of the memory it requested
(which happens often, e.g. when a large process forks and execs a
small process, or with Fortran 77 programs which allocate all storage
statically), its memory is unavailable to other processes until it
exits. IRIX 3.x has no virtual or logical swap space.
In IRIX 4.0.x, IRIX accepts every memory request, and does not
allocate virtual memory until a process actually tries to use it.
This allows programs which request more memory than they use to run
with much less memory than would otherwise be required. If too many
processes actually use their memory requests so that virtual memory
is in danger of filling up, IRIX kills one or more processes. IRIX
usually kills the process which is using the most virtual memory,
which may well not be the process which most recently requested
virtual memory.
IRIX 5.x works like IRIX 4.0.x, but one can set the amount of virtual
memory which IRIX is allowed to overallocate. This amount is called
"virtual swap space". "Logical swap space" is the sum of physical and
virtual swap. There is no virtual swap space by default, so IRIX 5.x
behaves like IRIX 3.x. One can set virtual swap to any amount of
memory; if it is set sufficiently high, memory requests will always
be granted, just like IRIX 4.0.x. Using jargon retroactively, IRIX
4.0.x has an infinite amount of virtual swap space.
Large or infinite amounts of virtual swap space work well for many
people, because most programs don't use all the virtual memory they
request, at least not at once. If your programs do use all their
virtual memory, they'll be killed and you'll see "Process killed due
to insufficient memory/swap" messages in your SYSLOG.
Under IRIX 4.0.x, you can only turn virtual swap off completely by
setting the kernel variable availsmem_accounting to 1. Doing so makes
IRIX 4.0.x behave like IRIX 3.x, allocating memory only if it is
actually available.
Under IRIX 5.x, you can turn virtual swap on or off by doing
'chkconfig vswap off' or 'chkconfig vswap on', or change the size of
virtual swap by editing /etc/config/vswap.options, and rebooting. You
can also use 'swap -v' to do any of these things directly and without
rebooting.
Remember that IRIX 5.x comes with virtual swap turned off and set to
zero. If you were happy with IRIX 4.0.x, you should turn virtual
swap on and set its size to a very large number. If programs are
killed, decrease the size of virtual swap or turn it off.
See the swap(1M) and swapctl(2) manpages for details.
Subject: -17- Why is there no way to set up a RAM disk under IRIX?
Date: 29 Apr 1995 00:00:01 EST
IRIX has no way to set up a RAM disk (a personal computerish term for
a part of memory which can be used like a very fast disk) because it
already has a sophisticated disk caching mechanism, which will
probably do a better job of minimizing disk reads than you would if
you were deciding what to put on a RAM disk.
Subject: -18- BOOTING
Date: 15 May 1994 00:00:01 EST
As the song says, "There must be fifty ways to boot your Iris."
Subject: -19- How can I boot directly into single-user mode?
Date: 8 Mar 1996 00:00:01 CST
Use the PROM monitor's 'single' command.
For machines earlier than 4D35s, whose PROMs don't have that command,
first boot into sash from the PROM monitor with "boot", then type
"boot initstate=s".
Subject: -20- How can I boot from a non-default disk?
Date: 31 May 1995 00:00:01 CST
Says Justin Mason <jmason@iona.ie>: If your disk is SCSI ID 4, do
boot -f dksc(0,4,8)sash dksc(0,4,0)unix root=dks0d4s0
or
setenv bootfile dksc(0,4,8)sash
setenv path dksc(0,4,8)
setenv root dks0d4s0 # This is the tricky part
auto
from the PROM. The first method works once, so that subsequent
reboots use SCSI ID 1, and the second method sets the PROM to boot
from ID 4 every time (until you reset the PROM variables).
Dave Olson <olson@sgi.com> adds: "If you have a separate / and /usr,
you'll probably want to add 'initstate=s' at least the first time, or
the usr partition on your normal root will be mounted. Boot in single
user, then do 'cd /dev; ./MAKEDEV disklinks; multi'. After this, you
won't need 'initstate=s' since the link will now be to the usr
partition on the alternate boot disk.
Subject: -21- How can I boot my machine using a server on the other
side of a router?
Date: 24 Jan 1994 00:00:01 EST
Tell the router to forward BOOTP packets. If it can't, NFS-mount the
remote volumes on another machine on the same subnet and use the
nearby machine for your boot server.
Subject: -22- How do I make a bootable tape from an IRIX CD?
Date: 03 Sep 1995 00:00:01 EST
See the Sep/Oct 1993 Pipeline and/or
ftp://viz.tamu.edu/pub/sgi/software/admin/making-bootable-tape for a
detailed description, or just follow Dave Olson <olson@sgi.com>'s
summary: Take a look at the distcp(1M) manpage, and do something like
tapehost# mount -o ro cdhost:/CDROM /mnt
tapehost# distcp /mnt/dist /dev/nrtape
Note that 'fx', 'ide', and 'sash' for all machines are in the dist/sa
file. 'sa' is an image of the first part of the tape; use 'mkbootape
-f sa -l' to see the contents.
Subject: -23- Why can't I boot one of the stand-alone programs on a
tape or CD?
Date: 03 Apr 1994 00:00:01 EST
One reason is that some CPU names are preceded by periods and some
aren't. Another is that the Indigo R4000 and later CPUs use the
suffix 'ARCS', not 'IP20' or whatever as one might expect from
'hinv'. For example, the correct command to boot fx directly from the
PROM monitor on an Indigo R4000 is 'boot -f dksc(ctlr,unit,8)sashARCS
dksc(ctlr,unit,7)stand/fx.ARCS'. Note the use of 'ARCS' instead of
'IP20' and the missing period in 'sashARCS'.
Subject: -24- INSTALLING
Date: 15 May 1994 00:00:01 EST
These questions discuss software installation.
Subject: -25- Is it possible to remotely install IRIX over a network?
Date: 20 May 1993 00:00:01 CST
Yes. You can install IRIX from a remote machine which has a CD-ROM, a
tape drive, or an IRIX distribution directory. All of these
scenarios (and several others) are described in detail in the "IRIS
Software Installation Guide". Examples are provided.
Subject: -26- Which IRIX CD is the program 'foo' on?
Date: 25 May 1994 00:00:01 EST
Mount the CD and try 'grep foo /CDROM/dist/*.idb'. If you don't get
any output, 'foo' isn't on that CD. If you do, it is, and one of the
fields is the subsystem in which 'foo' lives. Entries in *.idb files
don't have a leading slash so you must leave it out if you grep for a
full path, e.g. 'grep usr/bin/lp /CDROM/dist/*.idb', not 'grep
/usr/bin/lp /CDROM/dist/*.idb'.
Subject: -27- How can I extract a single file from an 'inst'
subsystem?
Date: 05 Feb 1995 00:00:01 EST
Under IRIX 5.2 and earlier, you need to be clever. 'inst' guru Paul
Jackson <pj@sgi.com> reveals all:
- Find the subsystem in which the file lives, as described in the
previous question. For this example we'll extract /sbin/ed, which
lives in eoe1.sw.unix.
- Follow the bouncing prompt:
> su
> cd /usr/tmp
> mkdir -p tmproot/var/inst
> inst -f /CDROM/dist/eoe1 -r /usr/tmp/tmproot
> Inst> keep *
> Inst> install eoe1.sw.unix
> Inst> go
> Inst> q
> ls -l /usr/tmp/tmproot/sbin/ed
-rwxr-xr-x 1 root sys 75480 May 24 13:57 /usr/tmp/tmproot/sbin/ed
- Move your file somwhere else and 'rm -r /usr/tmp/tmproot'.
- That was under IRIX 5.x. Under IRIX 4.0.x or earlier, use
'/usr/tmp/tmproot/usr/lib/inst' for a temporary inst directory
instead of '/usr/tmp/tmproot/var/inst'.
In IRIX 5.3, inst's '-Y' flag allows you to install a single file. See
inst(1M) for more.
Subject: -28- Why doesn't 'inst' work?
Date: 16 Jan 1994 00:00:01 EST
One possibility is that you're using an old 'inst' with new
software. Always use an 'inst' at least as new as what you're
installing.
Subject: -29- Why doesn't 'inst' work remotely?
Date: 05 May 1994 00:00:01 EST
Usually because it can't log in to the machine with the distribution
media. 'inst' uses the guest account to do so, so make sure that
guest on the machine on which you want to install software can rlogin
to guest on the machine with the distribution media without a
password.
Subject: -30- I just installed a new version of IRIX and file X is
missing. What should I do?
Date: 14 Oct 1995 00:00:01 EST
'inst' doesn't deal well with some cases where a file moves from one
subsystem to another between IRIX versions. Upgrading may cause such a
file to disappear entirely. To bring it back, reinstall the subsystem
to which the file belongs in the newer IRIX with "set neweroverride
on" in 'inst' (or, equivalently, remove and reinstall the subsystem).
You might want to install only the missing file; see the entry on
installing a single file above.
Popular examples of this include several headers in /usr/include/sys
in IRIX 4.0.5/IDO 4.1.1, which are in eoe1.sw.unix, the mount(1M)
manpage in IRIX 5.3, which is in eoe1.man.unix, and
/usr/lib/X11/dyDDX/glx.so, also in 5.3, which is in x_eoe.sw.Server.
One way to detect files which have disappeared in this manner in IRIX
5.3 is 'showfiles -B'.
Subject: -31- I reinstalled an IRIX subsystem to restore a missing
file or get rid of a corrupted file, but it didn't help.
Why not?
Date: 13 Apr 1994 00:00:01 EST
'inst' doesn't bother to install a subsystem if the same or a newer
version is already installed. Tell it to install anyway by saying
'set neweroverride' before you say 'go'. Removing the subsystem and
reinstalling it will do more or less the same thing.
Subject: -32- Why do 'inst' and 'showprods' say that a subsystem is an
"Unknown product entry"?
Date: 12 May 1996 00:00:01 EST
Paul "Mr. inst" Jackson of SGI <pj@sgi.com> explains two reasons:
- The "orphan" subsystem is a special subsystem for directories in
which more than one product keeps files. inst will remove these
directories as they become empty. Do NOT remove the subsystem
yourself. In fact, you can only see it due to a cosmetic bug in some
versions of inst.
- The product spec file in /var/inst which describes that subsystem is
damaged, due either to the usual sort of file-damaging problems or
to a bug in IRIX 5.3's inst. This also makes inst think that that
product is version 0, and thus incompatible with every other
subsystem. There is no patch for the bug. The workaround is to
reinstall the affected subsystems with all files restricted, so that
no files are actually installed but the product spec file is
updated, like so:
inst -f whatever -X/
Inst> set exclusions # Should be set to "Value /"
Inst> keep *
Inst> install upgrade
Inst> list i # Should see the unknown entries
Inst> go
Inst> quit
Subject: -33- How can I install IRIX onto a second disk which I can
then move to another machine?
Date: 20 Jan 1994 00:00:01 EST
With difficulty. Many parts of the installation process assume that
you're installing IRIX onto your system disk (SCSI ID 1). Just fiddle
with SCSI ID switches and/or move disks around to make the disk onto
which you want to install IRIX the system disk for the duration of
the installation.
Furthermore, IRIX has many hardware dependencies, so you should only
move system disks between absolutely identical machines. If you want
to make a system disk for a machine without a network connection,
CD-ROM or tape drive, the easiest and safest way is to borrow another
CD-ROM or tape drive.
If you want to try anyway, Justin Mason <jmason@iona.ie> reports that
the following works under IRIX 5.1.1:
Set up the disk, e.g. with SCSI id 4, fx a generic "[bo]otable"
partition setup onto it, and mkfs the partitions. Copy sash, etc.
from your system disk to the new disk with dvhtool. Boot up the
miniroot as usual, go into inst, choose "admin" from the menu and do
the following, replacing SCSI IDs and partition numbers as
appropriate:
umount /root
umount /root/usr
mount /dev/dsk/dks0d4s0 /root
mount /dev/dsk/dks0d4s6 /root/usr
mount # Just to check
return # Go back to main inst menu
Then install as you like.
Subject: -34- How can I copy my system disk onto a second disk which I
can then move to another machine?
Date: 03 Sep 1995 00:00:01 EST
See the article in the Jul/Aug 1992 Pipeline and the addendum in the
Nov/Dec 1992 Pipeline, and note that the warning about hardware
dependencies in the previous question applies here too. Steve
Kotsopoulos <steve@ecf.toronto.edu> has written a script which does
this automatically; you can FTP it from
ftp://viz.tamu.edu/pub/sgi/software/admin/clonedisk. Be sure to read
the comments before running it!
If your machine has FlashPROMs, which are normally updated by 'inst',
you'll need to update them yourself; see flashio(1M) and heed its
warnings.
Subject: -35- How can I share parts of IRIX among multiple machines?
Date: 15 May 1994 00:00:01 EST
Diskless machines, for one. Share trees, for another. Look in the
misc FAQ under "WWW pages" for the "IRIX share trees" WWW page.
Subject: -36- Can I install a "tardist" archive without the "tardist"
software?
Date: 23 Sep 1995 00:00:01 EST
Certainly. A tardist archive is just a tar archive of an 'inst'
distribution. Retrieve the tardist file directly to disk (in Mosaic
or Netscape one does this by shift-clicking the link to the tardist
file), name it something.tar, untar it as you would any tar file, and
install it as you would any inst distribution.
Subject: -37- I'm running 6.5.x, why can't I install the compilers?
Date: 28 Apr 2000 11:58:53 PDT
This is a common problem, having to do with the way that inst deals
with version numbering. Essentially, the distributions on the
Development Foundation and Development Libraries are looking for the
stock 6.5 version of eoe.sw.base. In order to install successfully
from these distributions, you must have the following distributions
open (inst under 6.5 supports having multiple open distributions):
irix-6.5-foundation-1, and overlay disks for your current os level
(e.g irix-6.5.6-installation-tools-and-overlays-1-of-2-11-99, and
irix-6.5.6-overlays-2-of-2-11-99)
Once these distributions are open, you should issue a keep *, so
that nothing will be installed from these distributions. You should
now be able to install the development system with minimal difficulty.
Subject: -38- NETWORKING
Date: 15 May 1994 00:00:01 EST
These questions discuss general networking.
Subject: -39- Why isn't my network working?
Date: 03 Sep 1995 00:00:01 EST
A list of good things to try is at
ftp://viz.tamu.edu/pub/sgi/software/admin/network-checklist.
Subject: -40- How can I measure my network's reliability?
Date: 22 Oct 1995 00:00:01 EST
Don't worry about collisions. They are part of normal operation on a
crowded Ethernet. You *should* worry about late collisions (which are
logged to the console) and lost packets (which you can easily measure
with the command 'ping -fs 3000 -c 1000 someotherhost'), which usually
mean network hardware problems or a misconfigured bridge or router.
See the Ethernet FAQ (posted to comp.dcom.lans.ethernet and FTPable
from ftp://rtfm.mit.edu/pub/usenet/comp.dcom.lans.ethernet/ and other
news.answers archives), Chapter 18 of the "IRIX Advanced Site and
Server Administration Guide" and the Sep/Oct 1993 Pipeline for more.
Subject: -41- How do I add a static route?
Date: 07 Sep 1996 00:00:01 EST
Some sites handle IP routing by designating a routing machine and
having all other hosts define a static route to that machine. The way
to do this on SGIs is in the /etc/init.d/network.local script.
1) Read the paragraph just before the copyright at the top of
/etc/init.d/network and make the links it specifies.
2) Put something like the following in /etc/init.d/network.local,
replacing ROUTER'S.IP.ADDRESS.HERE with the address of your router.
Under IRIX 6.x, omit the "1" at the end of the same line.
#! /bin/sh
IS_ON=/sbin/chkconfig
case "$1" in
'start')
if $IS_ON network; then # network must be chkconfig'ed on
/usr/etc/route add default ROUTER'S.IP.ADDRESS.HERE 1
fi
;;
'stop')
/usr/etc/route delete default ROUTER'S.IP.ADDRESS.HERE ;;
*)
echo "Usage: $0 {start|stop}" ;;
esac
Check the script with 'sh -v /etc/init.d/network.local'.
The above setup will not allow you to NFS-mount (or unmount) disks
from the other side of the static route gracefully, because the route
will be added after the mount attempt during startup and deleted
before the unmount attempt during shutdown. Instead of putting the
routing commands in a separate script, put them in /etc/init.d/network
itself. Put the 'route add' just after the "$ROUTE $RFLUSHFLAG" line
and the 'route delete' after the '/sbin killall mount ...' line.
There is no reason to define a static route and also to run routed. If
you do so, routed will delete the static route and you'll be confused.
'chkconfig routed off'.
Under IRIX 6.2 or later, remove the '1' from the end of the 'route
add' line.
Subject: -42- How can I make the 'slip' command advertise the Ethernet
address of the SLIP client?
Date: 10 Dec 1993 00:00:01 EST
You can't. Just add something like
/usr/etc/arp -s $USER `netstat -ia | grep :` pub
to the shell script in which you start the SLIP process. $USER is the
SLIP client. The 'netstat | grep' part gets the host's Ethernet
address, and 'arp' advertises the host as an ARP server for $USER.
See also the arp(1M) manpage.
Subject: -43- I've just edited inetd.conf, and nothing changed. Why?
Date: 15 Feb 1995 00:00:01 EST
You need to make 'inetd' reread inetd.conf. Do 'killall -HUP inetd'.
If that doesn't work, comment out the entry you modified, 'killall
-HUP inetd', uncomment the entry and 'killall -HUP inetd' again. If
that doesn't work, 'killall -TERM inetd'. If that doesn't work,
reboot.
Subject: -44- Why can't I 'rdist' files between Suns and SGIs?
Date: 30 Jan 1995 00:00:01 EST
Sun's 'rdist' expects SGI's 'rdist' to live in /usr/ucb, but it's
actually in /usr/bsd. Make a symbolic link from /usr/ucb/rdist to
/usr/bsd/rdist and all will be well.
In IRIX 5.3 or later, you may need to use ordist(1).
Subject: -45- Why are there "satwrite failure: inetd" messages in my
SYSLOG in IRIX 5.3?
Date: 4 Jun 1997 00:00:01 EST
Says Ellen Desmond <desmond@orange.engr.sgi.com>:
This is due to a bug in inetd that is manifest only when eoe2.sw.audit
has been installed. Inetd calls satwrite() to generate an audit
record every time it spawns a server. The bug is that the call to
satwrite() is made after the process has done a setuid() to the user
specified for that server in inetd.conf. Because satwrite() is a
superuser-only call, it fails for servers that don't run as root, such
as fingerd.
The bug is fixed by patch 1268. If you don't use auditing, you can just
ignore the messages or remove the eoe2.sw.audit subsystem and rebuild
your kernel.
Subject: -46- What is causing the "get interface flags" error in my
syslog?
Date: 06 Apr 1996 00:00:01 EST
Walter Roberson <roberson@hamer.ibd.nrc.ca> says: As part of its
licensing verification, Framemaker 4.0 attempts to get interface flags
on a fixed ethernet interface instead of using the hardware inventory
functions to figure out what the name of the ethernet is. This results
in an error like
broadcast: ioctl (get interface flags): No such device or address
Speak with Frame Technologies.
Patch 1092 includes fixes for "frame license daemon" and might fix the
above problem; we don't know yet.
Subject: -47- Why are network connections between SGIs and Suns much
slower than between SGIs and SGIs or Suns and Suns?
Date: 13 Aug 1995 00:00:01 EST
Dave Olson <olson@sgi.com> explains: Some older Sun systems (I don't
remember which, but not current systems) have ethernet *chips* (this
isn't an OS issue) that can't handle the recommended 9.6 usec
interpacket gap. This is not a problem when sending packets, but when
receiving packets from faster systems (like SGI Indigos and later) it
results in lost packets, retransmission and major slowdown.
Subject: -48- How can I set up IP aliases?
Date: Wed Sep 22 16:50:53 CDT 1999
An IP alias is an extra IP address which may be in a completely
different subnet and domain than a host's primary IP address. This is
often desired by WWW presence providers. An IP aliasing is a step
beyond a DNS alias, which is an extra hostname for the same IP
address. HTTP requests do not include the host name, so a server has
no way of knowing the host name to which a particular request was
sent. A server *can* determine the IP address to which a particular
request was sent and respond accordingly, so IP aliasing allows one to
have multiple independent servers on a single machine.
Patch 1356 to IRIX 5.3 allows IP aliasing. IRIX 6.2 and later does IP
aliasing without patches. The interim solution once available from
Silicon Surf is no longer available.
See also
http://www.amazing.com/internet/virtual-homer.html
http://www.apache.org/docs/virtual-host.html
http://www.thesphere.com/~dlp/TwoServers/
for general information and descriptions of other implementations.
Subject: -49- Where is ipfilterd documented?
Date: 15 Jun 1996 00:00:01 EST
In IRIX 6.1 and earlier, ipfilterd itself is part of eoe.sw.ipgate but
the ipfilterd(1M) manpage is part of eoe2.sw.netman and the
netsnoop(1M) manpage (ipfilterd uses netsnoop syntax) is part of the
extra-cost NetVisualyzer package. A copy of netsnoop(1M) is at
ftp://viz.tamu.edu/pub/sgi/software/admin/netsnoop.1, and both
manpages are in the uman database (see the misc FAQ under "WWW
pages"). In IRIX 6.2 and later, ipfilterd(1M), netsnoop and
netsnoop(1M) are all part of the base OS.
Note that patch 1249 fixes a number of ipfilterd problems.
Subject: -50- MAIL
Date: 15 May 1994 00:00:01 EST
These questions discuss mail configuration and problems.
Subject: -51- How can I set up 'sendmail' to pass (or not pass) 8-bit
characters?
Date: 30 Jan 1995 00:00:01 EST
Many experts say "don't try". RFC822 requires mail transport agents to
*clear* the eighth bit, and many hosts do. Some which don't may crash
when they get mail with the eighth bit set. Instead, use a
MIME-compatible mail program. MIME, described in RFC1521, is a
standard for enclosing non-RFC822 material in your mail. The apps FAQ
discusses several mail programs which support it.
In IRIXes up to 5.2, one can flout this doctrine by running sendmail
with the '-o8' flag to allow 8-bit characters in message bodies (*not*
headers). Read the description of that option in sendmail(1) for more
reasons not to use it. In IRIX 5.3, sendmail runs in 8-bit mode by
default, but one can use the '-o7' flag to get the old behavior.
Subject: -52- Why are my mailbox files changing ownership?
Date: 15 Oct 1995 00:00:01 EST
If your mail directory is mounted from another machine, your machine
does not have root access to that directory, and the other machine has
BSD-style "restricted chown" (the restricted_chown kernel variable is
turned on or it's not an SGI) /bin/mail will change mail file
ownership when delivering local mail. Without unrestricted chown *or*
root access, /bin/mail is unable to give a mail file back to its owner
after delivering mail. You can fix the problem by turning off re-
stricted chown on the other machine (if it's an SGI) or exporting the
mail directory with root access for your machine. This problem was to
have been fixed in IRIX 5.2, but apparently is still present in 5.3.
Subject: -53- Why isn't a valid user getting their mail?
Date: 24 Jan 1994 00:00:01 EST
IRIX' mail system requires "valid users" to have both valid password
file entries (whether local or via NIS) and home directories. The
latter often trips one up when installing POP servers and whatnot,
where home directories aren't really necessary. Just make a fake
one.
Subject: -54- How can SGIs and Suns share a mail spool?
Date: 05 Feb 1994 00:00:01 EST
Paul Riddle <paulr@umbc.edu> has written up how he did it. Read
ftp://ftp.umbc.edu/pub/sgi/.
Subject: -55- What's an "unknown mailer error"?
Date: 20 Feb 1994 00:00:01 EST
See ftp://viz.tamu.edu/pub/sgi/software/mail/mail-errors.
Subject: -56- What's "mailbox: Error 0"?
Date: 05 Mar 1994 00:00:01 EST
It's a harmless bug; don't worry about it. It is fixed in IRIX
versions 4.0.5H/4.0.5IOP and later.
Subject: -57- Why am I having problems with my NFS-mounted mail spool?
Date: 02 Mar 1996 00:00:01 EST
Perhaps for one of the following reasons:
- IRIX 5.2 NFS has a bug which prevents writing to a zero-length file
if the writing process has group write permission but not group read
(or user write) permission. IRIX 5.2 /bin/mail likes to set the
permissions of /var/mail/<user> to mode 620, exactly what is needed
to exercise the NFS bug. There is no patch for these bugs, but both
are fixed in IRIX 5.3. Meanwhile, you can work around the problem by
chmod'ing your mail files to 660 and setting the appropriate option
in your mail program to make sure that empty mail files are not
removed.
- lockd must be running on the NFS server to allow programs on the
client to lock mail files. If it is not, Zmail/MediaMail will
complain that "file cannot be locked". One could tell MediaMail to
not request a lock with "unset dot_lock", but that would risk mail
file corruption.
- IRIX 5.3 lockd is broken. Typical symptoms include "Warning: could
not lock /var/mail/user after 100 trys." (from /usr/sbin/Mail) and
the presence of user.lock and user.rolock files in /var/mail. Patch
1128 fixes at least some of these problems. If the client and
server are both IRIX 5.3, you can also work around the problem by
changing the mount type to NFS3. (See notes on NFS3 elsewhere in
this FAQ.)
See also "NFS Mounting Mail with IRIX 5.3 and IRIX 6.0.1" in the
Sep/Oct 1995 Pipeline.
Subject: -58- Why are incoming mail addresses case-sensitive in IRIX
5.3?
Date: 31 May 1995 00:00:01 EST
It's a configuration change. To change it back, find each line in your
/etc/sendmail.cf which begins with "M"; these are mailer definitions.
Each has a field beginning with "F="; these are the mailer flags.
Remove the "u" from each mailer flags field.
Subject: -59- NFS
Date: 15 May 1994 00:00:01 EST
These questions discuss NFS.
Subject: -60- How can I tell what hostname to use in /etc/exports?
Date: 07 Feb 1994 00:00:01 EST
NFS servers may need a particular form of a client's name in
/etc/exports to allow the client access. This may not be obvious, for
example if the server is also a router. Log in from the client to the
server and say 'echo $REMOTEHOST' to see what the server thinks the
client is called, and put that in /etc/exports.
The System Manager ('chost') should be able to determine the correct
hostname for you.
Subject: -61- Why can't I export an NFS-mounted filesystem?
Date: 10 Dec 1993 00:00:01 CST
This is known as multi-hop NFS. It is not allowed or supported in
(Sun's) NFS because it is not in general possible to detect errors
such as infinite mount loops, on either the client or the server.
Subject: -62- Why can't Ultrix automount SGI filesystems?
Date: 10 Dec 1993 00:00:01 CST
Ultrix's automount uses an "untrusted" port for mount requests. Add
an '-n' to the mountd lines in /usr/etc/inetd.conf (/etc/inetd.conf
in IRIX 5.x), like so:
mountd/1 stream rpc/tcp wait root /usr/etc/rpc.mountd mountd -n
mountd/1 dgram rpc/udp wait root /usr/etc/rpc.mountd mountd -n
then 'killall mountd' and 'killall -HUP inetd' or reboot.
Subject: -63- Why does 'tar' work strangely on a filesystem mounted
from an SGI?
Date: 15 Oct 1995 00:00:01 EST
When user A extracts a file owned by user B from a tar archive, 'tar'
makes the file owned by user A unless user A is the superuser. Some
systems allow users to give files away (e.g. IRIX); some do not
(e.g. SunOS). On some systems with the restricted behavior (SunOS
among them), 'tar' tries to give the file to user B whether or not
user A is the superuser, assuming that the chown system call will fail
if user A is not. This is not true if user A is using 'tar' on (e.g.)
a Sun to extract files onto a filesystem NFS-mounted from (e.g.) an
SGI. 'tar' may create zero-length files or give away directories and
then be unable to extract files into them.
Work around the problem by doing the 'tar' on the SGI or extracting
onto a Sun filesystem. It is possible that third-party versions of
'tar' (e.g. GNU tar) are smarter; if so, let us know. You could turn
the restricted_chown kernel variable on on the SGI, but be aware that
some programs may depend on unrestricted chown, notably /bin/mail as
discussed elsewhere in this FAQ.
Subject: -64- Why does 'df' report incorrect numbers for a filesystem
mounted from a Sun?
Date: 29 Apr 1995 00:00:01 EST
Sun's filesystem reserves space, usually 10%, and Sun's 'df' reports
only the unreserved space. SGI's filesystem (efs) does not reserve
space, so SGI's 'df' doesn't correct for reserved space.
Subject: -65- Is 'pcnfsd' available for the SGI?
Date: 20 Apr 1996 00:00:01 EST
For IRIX 4.0.x, look in ftp://ftp.sgi.com/support/pcnfsd.sysV/. (Note
that although SGI makes this available, they do not support it.) For
IRIX 5.x, look in ftp://viz.tamu.edu/pub/sgi/software/pcnfsd/ or
http://www.sgi.com/Technology/Connectivity/pcnfsd.html. See also the
pcnfsd bug mentioned in the security FAQ.
Subject: -66- Can I export a CD-ROM from my SGI to a non-SGI?
Date: 10 Dec 1993 00:00:01 EST
Not in IRIX 4.0.x. You can in IRIX 5.x, as you would any other
filesystem.
Subject: -67- How can I read an IRIX (EFS) CD-ROM on a machine which
doesn't use EFS?
Date: 18 Jun 1995 00:00:01 EST
You want 'efslook', at
ftp://viz.tamu.edu/pub/sgi/software/efslook.tar.gz.
Subject: -68- How can I get quotas to work on an NFS filesystem?
Date: 16 Dec 1994 00:00:01 EST
Mount the filesystem with the 'quotas' option, by adding it to
/etc/fstab or the automounter map as appropriate, and make sure the
nfs.sw.nis subsystem, which contains the NFS quota daemon
(/usr/etc/rpc.rquotad) is installed. That's nfs.sw.nis, not
nfs.sw.nfs! See fstab(4), rquotad(1M) and perhaps automount(1M) for
details.
Subject: -69- Why can't some NFS clients NFS-mount IRIX 5.2
filesystems?
Date: 07 Sep 1996 00:00:01 EST
IN IRIX 5.2, SGI's mount daemons (/usr/etc/mount_*) can fool some NFS
clients (TGV's Multinet NFS for VMS, for one) into thinking that the
mount daemons are NFS servers and trying to connect to them as
such. If the client can mount filesystems from your machine when no
mount daemons are running, but not when they are, this is your
problem. Fix it by upgrading to IRIX 5.3 or, if you're daring,
stealing the mount daemons from an IRIX 5.3 machine.
Zsolt Bagoly <bagoly@ludens.elte.hu> reports that DEC OSF1 and Linux
clients can work around this problem with the option "port=2049",
e.g. 'mount -t nfs -o port=2049 server:/path /mount-point'.
Subject: -70- What's NFS3?
Date: 21 Nov 1995 00:00:01 EST
An improved NFS which comes with IRIX 5.3. It is described in detail
in ftp://gatekeeper.dec.com/pub/standards/nfs/ and the
Nov/Dev 1995 Pipeline. Its quirks are still being discovered by users;
one prominent quirk is that it is not supported by the automounter.
Another is that a bug causes ~/.Xauthority files on an NFS3 volume to
be truncated; patch 216 fixes the bug. Another is that it interacts
poorly with XFS; patch 547 fixes that and many other problems.
Subject: -71- Why can't Solaris 2.5 clients read directories on NFS3-
mounted IRIX 6.2 filesystems?
Date: 8 Mar 1997 00:00:01 EST
Solaris 2.5, a 32-bit operating system, can't understand 64-bit data
from IRIX 6.2 NFS3 servers. This will be fixed in Solaris 2.6.
Meanwhile, export the IRIX 6.2 filesystems with the 32bitclients
option. See also 'relnotes nfs 5'.
Subject: -72- Why does my NFS server print "fhtovp_end VFS_GET failed"
or "fhtovp_end vp NULL" to the console/SYSLOG?
Date: 8 Mar 1997 00:00:01 CST
Walter Roberson <roberson@ibd.nrc.ca> explains: You installed an
NFS patch on your server, but you missed installing the NFS client
patch on your 5.3 system. You can either force the mount to use NFS2
or patch your 5.3 NFS client system.
Subject: -73- PRINTING
Date: 15 May 1994 00:00:01 EST
These questions discuss printing.
Subject: -74- Why can't lp(1) read my file?
Date: 10 Dec 1993 00:00:01 EST
lp(1) is setuid, so it can only read world-readable files. You can say
'lp < file' if you don't want to make your file world-readable.
Subject: -75- How can I tell 'lp' to turn banner printing or page
reversal off or on?
Date: 13 Nov 1994 00:00:01 EST
'lp' controls printers via shell scripts, called 'models', which live
in /var/spool/lp/model. When you install a printer, the appropriate
model script is copied to /var/spool/lp/interface/<name-of-printer>.
To temporarily change a printer's behavior, look at the manpage for
its interface script (or, if there is none, the script itself) to see
what options it wants, and pass them to the script with 'lp's '-o'
option. For example, 'lp -o"-nobanner" file' tells a "Generic
Postscript" printer (described in the gpsinterface(1) manpage) to
print 'file' without a banner page.
To permanently change a printer's behavior, edit its interface
script. The following are true for "Generic Postscript" printers,
but the idea is the same for others:
- To turn banner printing off or on, change the line 'BANNER=1' to
'BANNER=0' or vice versa.
- To turn page reversal off or on, change the line
'send=/usr/lib/print/lptops' to 'send="/usr/lib/print/lptops -U"'
(note the quotes) or vice versa.
In IRIX 5.x, you can change these settings in the printpanel. You can
also turn banner printing off on a per-user basis by doing 'echo
nobanner >> /var/spool/lp/settings/<printername>/<yourusername>'.
Subject: -76- How can I use 'lpr' instead of or as well as 'lp'?
Date: 06 Apr 1996 00:00:01 EST
IRIX is based on System V Unix and as such uses the lp(1) printing
protocol. The lp software is sufficient (and preferred) for printing
to local printers, printing to network printers attached to SGIs or
other machines using the lp protocol, and acting as a print server to
machines using the lp protocol. In IRIX 5.3/6.1 or later, it can also
print to remote printers using the lpr protocol and act as an lpr
server. SGI's standard lp interface scripts handle several common
file formats (text, PostScript, RGB images) automatically, and
Impressario (which works through the lp software) handles much more.
The lp software is installed by default.
lpr(1) is the BSD Unix printing protocol. SGI provides and supports
the lpr software only for printing to network printers attached to
computers which use the lpr protocol, e.g. a BSD Unix machine, an HP
printer with an appropriately configured JetDirect card or a
Macintosh running lpDaemon. It can also be used, unofficially, for
controlling a local printer and for a print server. It comes with no
support for handling different file formats at all. The lpr software
is not installed by default. It is in the eoe2.sw.bsdlpr subsystem in
IRIX 4.0.x and print.sw.bsdlpr in IRIX 5.x.
You can use lpr to print to a local printer in one of two ways:
- Set the printer up as an lp printer and write a printcap(4) entry
with an output filter which is just a wrapper around lp. If that
isn't crystal-clear, ask SGI for their writeup on "Integrating The
AT&T Spooler With The BSD LPR Print Spooler". A not-guaranteed-to-
be-up-to-date copy is at
ftp://viz.tamu.edu/pub/sgi/software/print/lpr-to-lp.
- Write a printcap entry and set up the printer as an lpr printer
just as you would on a BSD system. SGI doesn't support this use of
lpr, and comp.sys.sgi.* won't be much help either.
Now that the printer is set up as an lpr printer, you won't be able
to use lp with it directly. You can make lp(1) send files to a
local lpr printer in one of two ways:
- Use an lp interface script that calls lpr. You can use
mkbsdpr(1M), which comes with Impressario 1.1 and later (see the
Impressario FAQ), write one yourself or ask SGI for their writeup
"LPTOLPR, A Model File for LP". A not-guaranteed-to-be-up-to-date
copy of the last is at
ftp://viz.tamu.edu/pub/sgi/software/print/lp-to-lpr.
- Replace lp with a script which calls lpr. One such script is
ftp://viz.tamu.edu/pub/sgi/software/print/lp-wrapper-for-lpr.
You can print to an lpr server in one of two ways:
- Under IRIX 5.3/6.1 or later, use the lp software. Just use
printers(1M) to add the remote printer. The "list printer" button
will not list lpr printers; just enter the name of the remote
printer manually. This may or may not work for a printer with a
JetDirect card; see the next entry.
- Under older IRIXes, or if you're used to it, use the lpr software.
See SGI's "Advanced Site and Server Administration Guide".
You can set up an lpr server in one of two ways:
- Under IRIX 5.3/6.1 or later, you don't need to, because the lp
server is also an lpr server. This is officially documented only in
'relnotes print 3'. There are some caveats:
- Don't define printers to the lp and lpr software (i.e. in
/var/spool/lp/interface and /etc/printcap) with the same name.
- Clients must be in ~lp/.rhosts or /etc/hosts.equiv; lpsched
won't look in /etc/hosts.lpd.
- lpd must not be running when lpsched starts up. If you want to
switch from lpd to lpsched, do '/etc/init.d/bsdlpr stop;
/etc/init.d/lp stop; /etc/init.d/lp start'. /etc/init.d/bsdlpr
can't be chkconfig'ed off, so if you want to leave the lpr
software installed you'll have to move /etc/init.d/bsdlpr to
another name so it won't run on bootup.
- The client must send the control file before the data file. Most
Unixes do; some PCs do not. It might be informative to run lpsched
with the -debug flag, or just try it and see if it works. lpd
accepts either file order; lpsched should, and will in IRIX 6.2.
- Under older IRIXes, or if you can't get the previous solution to
work, or if you're a BSD head, use the lpr software. Set up your
local printer as an lpr printer (see above) and edit /etc/hosts.lpd
to grant access to the clients. Again, SGI won't support this.
Subject: -77- How can I print to a printer with a JetDirect card?
Date: 18 May 1996 00:00:01 EST
Configure the JetDirect card to act as an lpr print server (see the
JetDirect documentation) and set up your SGI to use it as you would
any remote lpr printer (see above). You may also want to set up a
bridge to the 'lp' system; again, see above.
Unless you have a very recent JetDirect card, you will not be able
to use printers(1M) to set up the JetDirect as a remote lpr printer.
You can either use the lpr software on the SGI (see above), or you
can circumvent printers(1M) in one of the following ways:
- Use printers(1M) to add a normal 'lpr' print server with the local
printer name you'll want to use for the JetDirect printer.
- Do '/usr/lib/lputil add /dev/null netface <local-printer-name>'.
- Do '/usr/lib/lpadmin -p<local-printer-name> -mnetface -v /dev/null',
and create by hand all of the directories and little files in
/var/spool/lp.
Each of the above will create an interface script in
/var/spool/lp/interface/<local-printer-name>. Since you subverted the
usual installation process, some of the settings in the script won't
be right yet. Change HOSTNAME to the JetDirect's hostname, NETTYPE to
"bsd" and (thanks to Ken Harris <kenh@netcom.com> for this one) SYNC
to 0. If it's a Postscript printer, change TYPE to "PostScript" and
CONVTYPE to "PostScriptFile".
Bill Poitras <bill@ba.msi.com> adds: To print from a PC using SAMBA
via your SGI, you can either modify the interface script so that
specifying "-oraw" to lp (as suggested in the SAMBA documentation)
will set CONVTYPE to nothing, or set up a second interface to the same
printer and set its CONVTYPE to nothing permanently.
If your JetDirect card is very old, it may not understand the lpr
protocol at all. If it is very, very old, you may have to resort to
sending the file over TCP port 9100 directly. If it is only very old,
you may be able to upgrade the card to a version which does understand
lpr. Upgrading the JetDirect is beyond the scope of this FAQ.
Subject: -78- Why won't lpr(1) print very large files?
Date: 24 Aug 1996 00:00:01 EST
lpd(1M) intentionally places an upper limit on spool file size. To
remove it, add "mx#0" to your printcap(4) entry.
Subject: -79- SGI DAEMONS
Date: 08 Oct 1995 00:00:01 EST
This section deals with the objectserver and its tricky pals, the
directoryserver, famd and mediad.
Subject: -80- Why isn't the objectserver working?
Date: 24 Aug 1996 00:00:01 EST
Install patch 1096. If you still have problems, read on.
First, consider whether you really need the objectserver. Without it,
you'll lose "business cards" and the graphical admin software. They're
probably not worth the headache.
Anne Eagle <annee@sgi.com> posted most of the following:
- Its database may be corrupt. If the objectserver appears to start
OK but crashes later, this is probably the case. Rebuild it like
so:
/etc/init.d/cadmin stop
/etc/init.d/cadmin clean
/etc/init.d/cadmin start
If the preceding doesn't work, try this
/etc/init.d/cadmin stop
mv /var/Cadmin/data /var/Cadmin/data.old
/usr/Cadmin/bin/parseclasses
/etc/init.d/cadmin start
Note that either method destroys "Privileged User" and "Business
Card" information. (This is the ONLY known drawback of rebuilding
your objectserver database, and the ONLY reason why SGI
documentation recommends that you consult with the TAC before doing
so. For most people that means that there's no reason why you
shouldn't rebuild whenever the need arises.)
- One of your system configuration files (including but not limited to
/etc/exports, /etc/fstab, /etc/inittab, /etc/mtab, /etc/passwd and
/etc/printcap) may have minor format problems which don't bother
IRIX proper but do bother the objectserver. Such problems include a
last line which doesn't end with a linefeed, a backspace not
preceded by a space in /etc/exports, or unprintable characters. Gary
Lin <glin@csd.sgi.com> suggests that you ensure that /etc/exports
has explicit -ro or -rw export options and that you remove
continuation lines (\) from /etc/printcap. Ken Gant
<krgant@musetech.com> points out that, as specified in gettydefs(4),
the last line of /etc/gettydefs must be blank. One sign that you
have such a problem is a core file in /var/Cadmin/data. If you find
and fix a problem, rebuild the databases as above.
If you can't find the problem, try the following:
par -s -i -N open -l -SS /usr/Cadmin/bin/objectserver -d
The last file objectserver opens is probably where the problem is.
If you're really desperate, the TAC will give you an objectserver
compiled with -g and help you run dbx on it.
- You may be swamping the objectserver with NIS (YP) users. There are
several ways around this:
- Start a directoryserver on a machine on your local network.
- Use netgroups or the "+user" form in /etc/passwd instead of just
a "+" and rebuild the databases as above.
- Most severely, remove the NIS object definition files so that the
objectserver will not create NIS objects, rebuild the
objectserver database (without the NIS objects) and restart the
objectserver as follows. You will not be able to manipulate NIS
users with Cadmin if you do this.
killall fm
mediad -k
killall objectserver
mv /var/Cadmin/data /var/Cadmin/data.orig
cp -pr /usr/Cadmin/classes /usr/Cadmin/classes.orig
rm /usr/Cadmin/classes/groupObject.op
rm /usr/Cadmin/classes/nisAccountObject.op
rm /usr/Cadmin/classes/peopleNISObject.op
rm /usr/Cadmin/classes/peopleObject.op
/usr/Cadmin/bin/parseclasses
/usr/Cadmin/bin/objectserver
ps -ef | grep obj
Wait until you see 2 objectserver processes running, then do
mediad
fm -lrb &
- Chris Riney <chris.riney@tandy.com> says: "We have just discovered
here at our site that if you do not have a route defined for the
SGI multicast subnet, then objectserver will gobble up memory. I
established a route for 224.0.0.0, and objectserver has been up for
over a week without consuming additional memory." This route is
defined in the stock /etc/init.d/network.
- Andreas Klingler <andreas.klingler@rrze.uni-erlangen.de> fixed his
objectserver by removing /usr/Cadmin/classes/printerObject.op and
then rebuilding /var/Cadmin/data as above.
- David Carrigan <vermeer@panix.com> fixed his objectserver by editing
his /etc/passwd file so userids were in ascending order.
- Tovar ? <tvr@skywebs.com> suggests shutting off your objectserver,
then running 'objecterver -d'.
- Urpo Kotipalo <nightis@raita.oulu.fi> had trouble with shadow
passwords and the objectserver, which he fixed by waiting until
'/etc/init.d/cadmin clean' had finished running pwconv(1M) before
doing '/etc/init.d/cadmin start'.
See also "Indigo Magic Tips and Tricks" in the Sep/Oct 1994 Pipeline
and the entry on the imon queue below.
Subject: -81- What is sending packets to the sgi-dog.mcast.net
multicast address?
Date: 06 Apr 1996 00:00:01 EST
The objectserver. It's using that address intentionally; SGI just
didn't bother to define a new one. Scott Henry <scotth@sgi.com> points
out that if you don't use any directoryservers and want to get rid of
the objectserver multicast packets, you can add '-t 0' to
/etc/config/objectserver.options and they will go away.
Subject: -82- Why are the objectserver and directoryserver taking up
so much memory in IRIX 5.3?
Date: 02 Mar 1996 00:00:01 EST
They really aren't. Doing 'ps -ef' in IRIX 5.3 and looking at the SZ
column shows that these programs have extremely large address spaces,
about 128M each. That's not the same thing as the physical memory they
occupy, which is shown in the RSS column and which should be much
smaller. See ps(1) and the IRIX 5.3 Cadmin release notes for more.
However, there is a bug in the objectserver which makes it grow
continuously when it can't find a directoryserver. SGI claims to have
fixed this in patch 1096; however, some disagree.
Subject: -83- Why don't the desktop admin tools recognize root as a
privileged user?
Date: 28 Jul 1995 00:00:01 EST
The objectserver isn't working. Clean the database as described above.
Subject: -84- Why doesn't my desktop deal properly with the CD or tape
icons?
Date: 02 Mar 1996 00:00:01 EST
Install patch 1096. If you still have problems, read on. Thanks to
Anne Eagle <annee@milamber.csd.sgi.com> for the following:
This is a known bug both in 5.2 and 5.3. Essentially, the desktop
icons stop responding to insertion/removal of mediad, although the
media itself is mounted properly and is accessible either from the
command line or by opening your home directory and then changing to
/CDROM or /floppy or starting cdman by hand or whatever is
appropriate for the media. Double clicking on the icon for the
peripheral results in a dialog saying that the drive is unloaded,
which of course it is not.
Here are a number of workarounds. If the first doesn't work, try the
second, and so on. Below, if the command begins with the "%" symbol,
that command can be executed by any user. Commands beginning with
"#" must be executed by root.
* Restart the File Manager:
% /usr/lib/desktop/telldesktop quit
% fm -lrb
* Restart both the File Manager and mediad:
% /usr/lib/desktop/telldesktop quit
% su -
# /etc/init.d/mediad stop
# /etc/init.d/mediad start
# exit
% fm -lrb
* Restart the File Manager, mediad and the objectserver:
% /usr/lib/desktop/telldesktop quit
% su -
# /etc/init.d/mediad stop
# /etc/init.d/cadmin stop
# /etc/init.d/cadmin start
# /etc/init.d/mediad start
# exit
% fm -lrb
* See if the icons from the Shared Resources in the upper right hand
corner of the System Manager are responsive to insertion/removal of
media. If so, access the icons from the System Manager.
* Reboot the system:
# reboot
* The media will have to be accessed by hand. Open a directory
view either by double-clicking on the home directory icon or
choosing "Desktop/Home Directory" from the Toolchest. Then change
to the appropriate directory for the device. For instance, the
floppy or floptical drives are access by default from /floppy and
the CDROM from /CDROM.
Find the "cdman" icon via "Find/An Icon" and then double click on
it to listen to music CD's over the scsi audio port.
See also the Sep/Oct 1994 Pipeline.
Greg Morlock <gmorlock@osmre.GOV> fixed his device icons by adding
entries to the local copy of /etc/passwd for each NIS user, with no
password and a '+' at the beginning of each userid.
Alternatively, mediad may never have started; see the next question.
Subject: -85- Why doesn't my mediad start up under IRIX 5.3?
Date: 04 Feb 1995 00:00:01 EST
If your mediad is dying with the message "mediad: Initialization of my
address failed. Cannot contact objectserver." to the syslog, your
machine may not be able to get its IP address. Try adding the line
hostresorder local bind
or (if you use NIS)
hostresorder local nis bind
to your /etc/resolv.conf file. Make sure the local machine's hostname
and IP address are in the local /etc/hosts file, and then either
restart the objectserver, mediad, and desktop or reboot.
This problem manifests itself when the objectserver cannot grab the
local machine's IP address from DNS (aka bind). When the resolver is
told to look at the /etc/hosts file first, it should find the local
machine's address without asking DNS about it.
Thanks to Dave Olson and Alexis Cousein (both of SGI) for tracking
this down.
Subject: -86- Why do cdman, cdplayer and capture say "permission
denied"?
Date: 02 Feb 1996 00:00:01 EST
You've installed patch 708 or 1048, which come with a broken mediad.
You really should have patch 1048 or later, which fixes a bad security
hole in the objectserver (see the security FAQ). If you don't care
about automounting CDs, turn off mediad with '/etc/init.d/mediad stop;
/etc/chkconfig mediad off'. If you do, use the mediad from IRIX 5.3,
(which you'll find in /var/inst/patchbase/usr/etc/mediad after
installing the patch) or upgrade to patch 1096 (but see below).
Subject: -87- Why can't 'cformat' format floptical disks, and/or why
can't 'datman' play audio DATs?
Date: 16 Mar 1996 00:00:01 EST
You've installed patch 1096, which breaks cformat(1M) and datman(1).
You really should have patch 1048 or later, which fixes a bad security
hole in the objectserver (see the security FAQ), so you should either
use mkfp(1M) instead of cformat and turn off mediad(1M) when you want
to use datman, or back down to patch 1048 (but see above).
Subject: -88- Why is famd hammering my NFS server?
Date: 24 Oct 1995 00:00:01 EST
It's partly a bug; get patch 165 for IRIX 5.2 and patch 166 for IRIX
5.3. It's partly just famd's nature; you can try to calm it down by
changing its polling interval (6 seconds by default, specified by the
'-t 6' flag) in /etc/inetd.conf. If you do install the patch, don't
forget to add the '-l' flag to famd's line in /etc/inetd.conf.
Subject: -89- What is causing "/dev/imon: event queue overflow" in my
syslog?
Date: 24 Oct 1995 00:00:01 EST
Probably famd. Try the measures in the previous entry. If those don't
work, try increasing the value of QSIZE in /var/sysgen/master.d/imon
and rebuilding your kernel.
Subject: -90- BUGS
Date: 03 Jun 1995 00:00:01 EST
These questions discuss miscellaneous bugs in IRIX. Note also the
entry on bugs and patches in the misc FAQ, and Walter Roberson's patch
lists in the "WWW pages" section of the same FAQ.
Subject: -91- Why do 'who', 'rusers', etc. show users who aren't
really logged in?
Date: 30 Jan 1995 00:00:01 EST
Several bugs in different versions of IRIX cause /etc/utmp to not be
updated properly after a user logs out. Programs incorrectly reporting
logins are correctly reporting the incorrect contents of /etc/utmp.
One such bug is present in an unspecified part of IRIX 4.0.x and fixed
in IRIX 5.x, and another is in ftpd in IRIX 5.2 and fixed in patch 162
and IRIX 5.3.
If one can't fix the bug which causes the incorrect /etc/utmp
entries, one can run a separate program to remove them from /etc/utmp
after the fact. Jeremy DuBois <jer@asdc.com>, David Hinds
<dhinds@allegro.stanford.edu> and Patrick M. Ryan <pat@gsfc.nasa.gov>
have written such programs, which can be found in
ftp://viz.tamu.edu/pub/sgi/software/utmp/. SGI provides such a
program, /usr/sbin/chkutent, in IRIX 5.3; the root crontab runs it
daily.
Subject: -92- What's wrong with ftpd in IRIX 5.2?
Date: 30 Jan 1995 00:00:01 EST
It doesn't maintain utmp properly (see the previous entry) and it dies
during 'mget's. Get patch 162 or upgrade to 5.3.
Subject: -93- Why isn't /usr/adm/SYSLOG being updated?
Date: 20 Dec 1996 00:00:01 CST
Popular causes include:
- running out of disk space. Once syslogd is unable to write to
/usr/adm/SYSLOG, it won't try again until it is `killall -HUP
syslogd`ed.
- installing IRIX 4.0.x and failing to heed the nagging from
the system when it is rebooted to run 'versions changed' and
combine new and old configuration files. In this case, the trouble
is in /usr/spool/cron/crontabs/root.
- Separating fields in /etc/syslog.conf with spaces instead of tabs.
If you use spaces, syslogd will silently segv when it reads that
file.
Subject: -94- I just edited /etc/inittab, and now I can't start up or
shut down my SGI! What's wrong?
Date: 03 Dec 1994 00:00:01 EST
If the last line of /etc/inittab is a comment, init will screw up
horribly. If your machine is still running, remove the comment and
everything will be OK. If not, go to the miniroot, run the shell and
remove the comment from there. The following sequence of commands is
one possible way to do this:
cd /root/etc
cat inittab # Decide how many lines to remove (say three)
wc inittab # See how many lines inittab has (say 120)
head -117 inittab > inittab.new # Keep the first 120 - 3 lines
mv inittab inittab.old
mv inittab.new inittab
cat inittab # Just making sure
and reboot. Don't forget the 'cd'; from the miniroot's point of view,
/etc/inittab is /root/etc/inittab.
The problem should be fixed in IRIX 5.3.
Subject: -95- Why does timed say "bind: Cannot assign requested
address"?
Date: 29 Oct 1994 00:00:01 EST
timed is incompatible with the rld which comes with patchSG0000023,
which is needed for DeltaCC. There are two solutions to this problem
(thanks to Alan Davis <davis@masig.fsu.edu>):
- Get a new timed from SGI.
- Replace the following line in /etc/init.d/network.options (line 664
in an unmodified IRIX 5.2 file)
/usr/etc/timed -M `cat $CONFIG/timed.options 2> /dev/null` &
with
env _RLD_ARGS="-clearstack" /usr/etc/timed `cat $CONFIG/timed.options 2> /dev/null` &
Subject: -96- Why does "ALERT: i2cWaitForXferDone: Xfer Done timeout
ERROR" appear in my SYSLOG?
Date: 22 Oct 1995 00:00:01 EST
These messages appear when the Indycam is unplugged. They are
harmless, but if they bother you you can banish them with patch 486.
Subject: -97- Why has NIS (YP) stopped working under IRIX 6.x?
Date: 04 Jul 1996 00:00:01 EST
Joshua Hart <hart@mcbi-36.med.nyu.edu> writes: Under earlier IRIXes,
NIS domain names could contain uppercase letters. Under IRIX 6.x they
may not; only lowercase letters are permitted.
Subject: -98- MISCELLANEOUS
Date: 15 May 1994 00:00:01 EST
Everything else.
Subject: -99- How do I set the number of processes allowed on my
machine?
Date: 13 Nov 1994 00:00:01 EST
Use systune(1M) to change 'nproc' (in the 'numproc' group of
parameters) and reboot.
Subject: -100- Where can I get a termcap file for 'iris-ansi-net' to
install on my non-SGI system?
Date: 4 Jun 1997 00:00:01 CST
SGIs use terminfo, so you need to translate the terminfo description
to termcap. 'infocmp -Cr iris-ansi-net' will produce an iris-ansi-net
termcap file. See infocmp(1) for more. Note that 'infocmp' is in the
eoe2.sw.terminf subsystem in IRIX 5.x, 6.0.x, and 6.1, and is not
installed by default. (infocmp is part of eoe.sw.unix in IRIX 6.2
and above and is installed by default.)
Subject: -101- How can I make my SGI understand strange terminal types
from other Unix systems?
Date: 4 Jun 1997 00:00:01 EST
If the other system uses terminfo:
Brent L. Bates <blbates@vigyan.com> has been able to copy binary
terminfo files directly between several different types of systems.
(On IRIX 5.x and above, the terminfo files are located under
/usr/share/lib/terminfo.) However, if for some reason this does not
work, use 'infocmp -I whatever > file' to extract the source
erminfo entry for the terminal. Transfer the file to your SGI and
do 'tic file' (as root) to put the entry into the terminfo database.
If the other system uses termcap:
Snip the termcap entry out of /etc/termcap (or wherever) with an
editor, transfer it to your SGI and (as root) do 'captoinfo file
> newfile' and 'tic newfile'.
See also the infocmp(1), captoinfo(1), tic(1) and terminfo(4)
manpages, and make sure you've installed eoe2.sw.terminf (under
IRIX 5.x, 6.0.x, and 6.1), which is not installed by default.
On IRIX 6.2 and above, these utilities are part of eoe.sw.unix.
Many unusual terminal devices can be taken care of in IRIX 6.2 and
above by installing eoe.sw.terminfo, which includes terminfo files
for many devices.
Subject: -102- Can I change my login shell or other password-file info
without being superuser?
Date: 27 Jan 1996 00:00:01 EST
At a graphics terminal, use the User Manager. At a text terminal, if
your account is on NIS (Yellow Pages) use 'ypchpass'. If it isn't
you're stuck, because IRIX has no way to change a local user's
password-file info from the command line. You might ask your
superuser to install one of the many free implementations of 'chsh'
and/or 'chfn' (one is in volume 3 of comp.sources.unix, at
ftp://ftp.uu.net/usenet/comp.sources.unix/volume3/) but they'll want
to make sure that those programs know how to lock /etc/passwd
properly.
-----------------------------
Subject: -103- How can I administer my Iris without a graphics
terminal?
Date: 13 Apr 1994 00:00:01 EST
The visual admin tools in IRIX 4.0.x ('vadmin') need GL, and do not
work on X terminals or workstations without GL. You can use 'sysadm'
on text terminals for some tasks, but beware of bugs and
inadequacies: SGI judged 'sysadm' to be too buggy to be worth
updating for IRIX 5.x.
The visual admin tools in IRIX 5.2 and later should display on any X
display, *except* for the backup/restore tool which is an exact port
from IRIX 4.0.x and requires GL. Some images will be missing when GL
is unavailable, but the tools will function properly. As for text
terminals, you're out of luck: 'sysadm' does not exist in IRIX 5.x.
Of course, you can always use a text editor and write scripts, or see
the next question.
Subject: -104- How can I use the visual admin tools on a system with
graphics to administer a system without graphics?
Date: 12 Feb 1994 00:00:01 EST
rlogin to the graphics-less system and run 'vadmin' (IRIX 4.0.x) or
'chost' (IRIX 5.x). Make sure that the DISPLAY environment variable
is set correctly and that both the vadmin/sysadmdesktop and the
shared library subsystems are installed on the graphics-less system
(which they are in the default installation).
Under IRIX 5.x, look at the READMEs in /var/sysadmdesktop/rsysmanapps
and /var/sysadmdesktop/sysmanapps to find out how to use 'chost' to
run commands on remote systems. Finally, in a future release of IRIX
5.x, the sysadmdesktop tools will be able to manage remote systems
*without* doing an rlogin.
Subject: -105- What's /etc/ioctl.syscon?
Date: 09 Apr 1995 00:00:01 EST
The default 'stty' setings to use on the console in single user mode,
in 'stty -g' format.
Subject: -106- Can I log out users automatically?
Date: 29 Apr 1995 00:00:01 EST
It depends on the sense of "log out" you want. If your users use the
'tcsh' shell you can set the autologout shell variable (and possibly
make it read-only) in a startup script to log out inactive shells.
This won't do anything to an instantiation of tcsh which is waiting
for another program to quit (e.g. a 'more' process in mid-file) or to
programs other than tcsh (e.g. a 'jot' window), and it won't log one
out of the graphics console. One can imagine a program similar to a
screen saver which would log one out after so many minutes of
keyboard/mouse inactivity, but we don't know of one. However, if
you're only worried about security (rather than resource hogging) it
may suffice to use 'xlock' to lock the screen rather than actually
logging one out. For that matter, tcsh can also lock itself rather
than logging one out; this wouldn't be helpful on a graphics console
but might be on a simple terminal.
Subject: -107- How can I change kernel variables and/or rebuild my
kernel?
Date: 14 Oct 1995 00:00:01 EST
To change a kernel variable under IRIX 5.x, use systune(1M). It saves
the tuned kernel in /unix.install; if this file exists, it is moved to
/unix on the next reboot. systune will change the parameters in the
running kernel if it can, and tell you to reboot if it can't.
To actually rebuild the kernel, use 'autoconfig -f'. This leaves the
rebuilt kernel in /unix.install. You must reboot to begin using the
rebuilt kernel; "rebuild your kernel" elsewhere in the FAQs implies
rebooting.
IRIX 4.0.x does not have systune. To change a kernel variable, edit
the file in /usr/sysgen/master.d which contains the variable, rebuild
the kernel and reboot.
------------------------------
End of sgi/faq/admin Digest
******************************
--
The SGI FAQ group <sgi-faq@viz.tamu.edu> http://www-viz.tamu.edu/~sgi-faq/
Finger us for info on the SGI FAQs, or look in ftp://viz.tamu.edu/pub/sgi/.