Network File System Utilities Release Notice

-== ---~
Network File System Utilities
Release Notice
Document No. 710-003730-009
December 1991
CONVEX Computer Corporation
Richardson, Texas USA
Network File System Utilities Vl 0.0
Release Notice
Document No. 710-003730-009
Copyright 1991 CONVEX Computer Corporation.
All rights reserved.
This document is copyrighted. This document may not, in whole or part, be copied, duplicated, reproduced, translated, electronically stored, or reduced to machine readable form without prior written consent from CONVEX Computer
Although the material contained herein has been carefully reviewed, CONVEX
Computer Corporation (CONVEX) does not warrant it to be free of errors or omissions. CONVEX reserves the right to make corrections, updates, revisions or
changes to the information contained herein. CONVEX does not warrant the material described herein to be free of patent infringement.
Unless provided otherwise in writing with CONVEX Computer Corporation
(CONVEX), the program described herein is provided as is without warranty of
any kind, either expressed or implied, including, but not limited to the implied
warranties of merchantability and fitness for a particular purpose. Some states do
not allow the exclusion of implied warranties. The above exclusion may not be applicable to all purchasers because warranty rights can vary from state to state. In
no event will CONVEX be liable to anyone for special, collateral, incidental or consequential damages, including any lost profits or lost savings, arising out of the use
or inability to use this program. CONVEX will not be liable even if it has been notified of the possibility of such damage by the purchaser or any third party.
CONVEX and the CONVEX logo ("C") are registered trademarks of CONVEX
Computer Corporation.
ConvexOS is a trademark of CONVEX Computer Corporation.
NFS and SunOS are trademarks of Sun Microsystems, Inc.
Ethernet is a trademark of Xerox Corporation.
Printed in the United States of America
This document describes the VtO.O Release of the CONVEX Network File System
(NFS) software. Always refer to this release notice before reporting problems: your
questions may be answered herein. Fixes and workarounds listed in this document
may help you avoid rediscovering known problems.
The CONVEX NFS release includes the following functionality:
• NFS (Network File System) - provides transparent access to remote file systems in
a heterogeneous network.
• NIS (Yellow Pages) - provides a distributed network lookup service.
• NETdisk - allows diskless workstations to boot from a CONVEX system.
• RPC/XDR (Remote Procedure Call/External Data Representation)- a set of libraries which implement network transactions. The NFS products are built on the
RPC/XDR libraries.
• RPC Protocol Compiler (rpcgen)- a tool which generates C code to implement the
RPC protocol.
• REX (Remote EXecution)- a program which executes a command remotely using
• Automount - a program that allows on-demand mounting of NFS file systems.
• Secure NFS - a version of NFS whose transaction requests are protected by an encryption protocol.
.. Note
The Secure NFS funcHonallty Is not available to lntemaHonal customers oulslde of the
U.S. and canada at this Hme due to legal restrlcHons on the export of encrypHon algorithms by the U.S. Department of State's Offtce of MunlHon Control.
The remaining sections in this document include the following information:
the contents of this distribution.
notes and cautions about use of the software.
known restrictions in the software.
bugs fixed in this release.
known software problems.
the release documentation.
To install the software described in this document, consult the Installation Procedures,
ConvexOS and Utilities V10.0.
Contents of This
The distribution package for this release of CONVEX Network File System Utilities
consists of
• This release notice
• Distribution media for the software
• Documentation
The specific contents of the software and documentation distribution are described
Software tor USA Distribution
Qty Type
Mag Tape
Part Number
ConvexOS Utilities, USA VlO.O
Software tor lnternatlonal Distribution
Qty Type
Mag Tape
Part Number
ConvexOS Utilities, INTL V10.0
CONVEX NFS Release Notice
Notes and
This section contains useful information and words of caution about the product.
• The VtO.O release of the CONVEX Network File System is bundled on the same
tape as the VtO.O release of ConvexOS and Utilities. To extract this product, you
must know the activation key for your system. If you have not received an activation key with the product, contact the CONVEX Technical Assistance Center for
instructions. In the continental United States, call 1 (800) 952-0379.
Prerequisites for this release of the CONVEX Network File System Utilities are as
-your system must be running the VtO.O release of the ConvexOS operating system
-your system must be running the VtO.O release of the ConvexOS Utilities
-your system must be running the Vl 0.0 release of the CONVEX Internet Services
• If you received the USA distribution tape, Secure NFS will automatically be installed when you select the CONVEX Network File System Utilities for installation.
TouseNFSonyoursystem,startupsystemdaemonsin the /etc/re. local system
start-up file, copying lines from I etc/ re. local. vlO. O. See the CONVEX Network File System System Manager's Guide for complete instructions on how to set up
NFS, the automounter, Secure NFS and NIS for your site.
CONVEX NFS Release Notice
New Features
The lock manager has been upgraded to the NFS 4.1 version. With this version of the
lock manager, all local file locking is done within the Operating system. The lock
manager daemon (rpcJockd) is now only involved in locking files accessed via nfs
and serving lock requests from remote systems.
with non-Convex
NIS (yp) masters
If you are using a non-CONVEX machine as the master NIS server (such as a machine
from Sun Microsystems), change the NIS configuration on that machine. Convex NIS
expects the NIS services map and rpc map to be built so that it may be searched quickly with a single yp_match access. Sun and some other vendors do not build these
maps, and the master NIS Makefile must change to build the map so as to allow the
Convex to find the information it needs. This change will not interfere with NIS accesses by machines from other vendors, as it only supplies additional infonnation for
To make this change, copy lines from the file /usrI etc/yp/Makefile on the CONVEX.
Find the string services. time, copy all lines down to the string
protocols. time to a temporary file and move the temporary file onto the nonCONVEX machine. Add the contents of the temporary file to the master YP Makefile
on the non-CONVEX machine.
On a Sun running SunOS 4.0 or greater, look for the Makefile in /var /yp, while on
older Suns look for it in /etc/yp. Go to that directory and replace the appropriate
lines in the Makefile with those from the temporary file. Once the Makefile is ready,
remove the files services.time and rpc.time. Build the new maps with make. The
CONVEX can now access the services and rpc maps correctly.
Do not set the system domainname if you do not plan on running the NIS system. In
general, if ypbind is running, the domainname must be set; otherwise the domain-
name should not be set.
between NIS (yp)
and sendmail
The default send.mail configuration does not support the use of the YP /NIS alias
map. The result of this is that sending mail to a YP alias will fail unless the sendmail
configuration file is altered.
The steps required to enable YP aliases within sendmail are:
1. Search in the file /usr/lib/sendmail.d for a line that begins with the characters
"OA". An example would be:
OA/usr /lib/aliases
2. Add the line "Op" immediately after the OA line. An example would look like:
OA/usr /lib/aliases
3. Refreeze the send.mail configuration file by executing a "I usr /lib/ sendmail -bz".
After perfonning step three, YP mail aliases should be enabled.
CONVEX NFS Rel~e Notice
Altering NIS (yp)
netgroup maps
H the NIS(yp) netgroup map on a running system is altered, the mount daemon
(mountd) will not see the changes until the mount daemon is killed and restarted or
the system is rebooted.
The procedure to kill and restart the mountd is as follows:
1. find the pid of the mountd ''ps -aux I grep mountd"
2. kill-1 <the pid found in step one>.
3. find the pid of the Internet services daemon "ps -aux I grep inetd"
4. kill -1 <the pid found in step one>.
All mounts made after this point will have knowledge of the changes to the netgroup
CONVEX NFS Release Notice
Current documentation for CONVEX NFS Utilities VlO.O includes the following
manuals and addenda:
CONVEX NFS Concepts, First Edition
CONVEX NFS System Manager's Guide, Third Edition
CONVEX NFS Reference Set, Third Edition.
CONVEX NFS Programmer's Reference, Second Edition.
CONVEX NFS Release Notice
Altering NIS (yp)
netgroup maps
If the NIS(yp) netgroup map on a running system is altered, the mount daemon
(mountd) will not see the changes until the mount daemon is killed and restarted or
the system is rebooted.
The procedure to kill and restart the mountd is as follows:
1. find the pid of the mountd "ps -aux I grep mountd"
2. kill-1 <the pid found in step one>.
3. find the pid of the Internet services daemon "ps -aux I grep inetd"
4. kill -1 <the pid found in step one>.
All mounts made after this point will have knowledge of the changes to the netgroup
CONVEX NFS Release Notice
between NIS and
the Name Server
The NIS server normally uses the YP maps to answer all queries, which can be a problem in networks using the Berkeley Internet Name Domain Server (BIND). If desired,
NIS can be configured to use BIND to resolve host information. This allows simpler
configuration of NIS client machines, since these machines can get accurate hosts information without being configured to use BIND themselves. This procedure works
on both ConvexOS V9.1 and V10.0. See Chapter 9 of the Internet Services System Manager's Guide for more information on BIND.
To configure NIS to use BIND, these changes are necessary: all NIS server daemons
must be restarted with a new command line option; the file /usr I etc/yp/Makefile must be changed to build theNIS hosts .byname and hosts .byaddr maps
differently; and the hosts .byname and hosts .byaddr maps must be rebuilt and
pushed out to all NIS servers.
First, edit I etc/re.local on all NIS server machines and add "-i" to the invocation of
/usr /etc/ypserv as follows:
if [ -f /usr/etc/ypserv -a -d /etc/yp/'/bin/domainname' ]; then
/usr/etc/ypserv -i & echo -n ' ypserv'
The -i flag ensures that ypserv will attempt to use BIND without regard to the
/etc/use_nameserver file. Next, edit /usr/etc/yp/Makefile to add "-b" to
the invocation of makedbm in the lines shown below:
hosts.time: $(DIR)/hosts
-@if [ -f $(DIR)/hosts ]; then\
umask 022; \
sed -e "/At/d" -e s/f.*$$// $(DIR)/hosts I $(STDHOSTS) \
I awk '{for (i = 2; i <=NF; i++) print $$i, $$0}' \
I $(MAKEDBM) -b - $(DOMDIR)/hosts.byname; \
$(STDHOSTS) $(DIR)/hosts I \
awk 'BEGIN {OFS="\t";} $$1 !- /A#/ {print $$1,
$$0}' \
I $(MAKEDBM) -b - $(DOMDIR)/hosts.byaddr; \
touch hosts.time; \
echo "updated hosts"; \
This change adds a special YP_ INTERDOMAIN key to the NIS hosts maps to tell the
YP server that it should make host database lookups via BIND. Finally, use the following steps to update and push out the new NIS maps:
t cd /usr/etc/yp
t rm hosts.time
# make
Do not distribute these maps to CONVEX NIS 181Y81S which a.. not set up to use BIND
by default (I.e••, which do not have the tlle /etc/u-_n...•8Z'V9r) until they ant running ypaerv with the ··I" flag. Otherwise, the VP ....,., wlll enter an endless loop.
CONVEX NFS Release Notice
NETdisk changes
NETdisk has undergone some major improvements to permit it to work with recent
SunOS release tapes. The use of the NETdisk tools has not changed, though users
should be aware of the following:
• The support for sharing of non-executable files such as 'man' pages is no longer
supported by these scripts since Sun Mircosystems has dropped their support for
this feature. If you were using this feature in the past, you may encounter difficulties if you try to add new clients.
• Though Sun native installs now support multiple versions and multiple architectures, configurations such as these are not supported by Convex. Such installations fundamentally work, but the root prototype tree installed in
/usr/etc/install/proto is shared for all architectures, which may cause problems
with multiple versions in particular.
• The SunOS 4.1.1 release is the first Sun release known to contain compressed 'tar'
archives. Convex does not supply 'uncompress', the tool needed to work with
these archives, at this time. The 'compress' tools are in the public domain and
available from many sources, including the Convex Users Group tape. If you run
INSTALL(8) with 'uncompress' in your search path, you will be able to install
SunOS 4.1.1 successfully.
Convex has tested these scripts to work correctly with the following SunOS release
CONVEX NFS Release Notice
SunOS 4.0.3
Current documentation for CONVEX NFS Utilities VlO.O includes the following
manuals and addenda:
CONVEX NFS Concepts, First Edition
CONVEX NFS System Manllger's Guide, Third Edition
CONVEX NFS Reference Set, Third Edition.
CONVEX NFS Programmer's Reference, Second Edition.
CONVEX NFS Release Notice
NFS Module
This section describes the module names used in the following sections describing restrictions, bugs and bug fixes in CONVEX NFS.
Module Name
Short Description
User and kernel level rpc and xdr
User and kernel level secure RPC
Automounter, automount(8)
Network based lock manager
Yellow Pages
RPC protocol compiler, rpcgen(l)
Remote file system quotas, rquotad(8c)
Network File System
Remote Execution, rex(lc) and rexd(8c)
Remote write to all users
rusers(tc), rusersd(lc)
Remote system status
Configuration and Management
CONVEX NFS man pages
CONVEX NFS Concepts manual
CONVEX NFS Reference set
CONVEX NFS System Manager's Guide
CONVEX NFS Master Index
CONVEX NFS Release Notice
Fixed Software
This section describes the bug fixes that have been made since the last release of CONVEX NFS.
Running automount without NIS configured results in the following error messages
being logged:
<date hostname> automount[10346]: YP bind failed: args to yp function are bad
<date hostname> automount[10346]: auto.master: args to yp function are bad
Resolution: The automounter has been changed to check if NIS (yp) is running before
calling any NIS functions.
PR-17313 PR-20108
The automounter can hang with the following deadlock. The automounter is in clntkudp_callit_addr() doing a getattr on a remote file system. The function clntkudp_callit_addr() is waiting to get access to the output buffer for the client handle being
used for the request. However, the output buffer is referenced by an mbuf which is
on the automounter's socket receive buffer.
Resolution: The problem has been fixed in clntkudp_callit_addr() by waiting on the
client handle output buffer to become free before returning from
clntkudp_callit_addr(). Note that this is the method used by Sun NFS.
PR-20166 PR-20434
When attempting to mount exported filesystems from the NOS/VE system via the
automounter the mounts fail. Normal mounts work fine.
Resolution: The automounter has been modified to use a different method of pinging
the remote mountd.
If the mount point specified in an automounter direct map contains a symlink, the
mount point will not be resolved to the real /tmp_mnt... mount point when it is accessed. The mount point appears to be a link to nowhere and users can't access the
files in the server directory via the mount point path.
Resolution: The automounter has been modified to allow the map entry paths
(mount points) to contain symlinks.
The man page for mount states that the grpid option can be used for setting the group
ID for a file the same as the group ID of the directory (default for the CONVEX). This
does not work for a mounted file system on a Sun (files reside on the Sun). Setting the
SETGID bit on the Sun directory gives the correct result.
Resolution: The automounter has been modified to make the 'grpid' option operate
CONVEX NFS Release Notice
The automounter leaves a link to nowhere: lrwxrwxrwx xxxxxx xxxxxx-> when
the directory it is trying to mount is not exported to the machine on which the automounter is running.
Resolution: Error logging was added to the automounter so that the case of trying to
mount a remote directory which is not exported or which does not contain the client
in its access list, will be easier to diagnose. The empty symbolic link for the mountpoint is the correct behavior and is the way the SunOS 4.1 automounter handles the
same situation. The error logging is displayed if the verbose command line parameter is supplied when starting the automounter.
The automounter will core dump if an indirect map of 100+ entries is used.
Resolution: The automounter has been fixed so that it does not corrupt its process
stack when processing a large indirect map.
When NFS is selected for optional product installation from an International Utilities
tape, the install script checks for disk space needed for Secure NFS on the SPU, and
(if applicable) complains: - Insufficient space on the spu for Secure NFS Please
free 2Mb on /mnt/os. Note that Secure NFS is not even on the international tape,
contrary to what the install notes suggest. At least there is no way to select Secure
Resolution: The install scripts will no longer check available SPU space unless a secure NF'S sysgen is going to be performed during the product installation.
The lock daemon's total SZ and RSS seem to grow proportionately to the total time
used by the process. This seems indicative of a memory leak due to an improperly
free()ed mallocO.
Resolution: The lock manager has been greatly changed for ConvexOS VlO.O and the
known memory leaks are now fixed.
CONVEX NFS Release Notice
The fcntl(2) system call can be interrupted when stopped and started while blocked,
waiting for a file lock resulting in the failure of the lock request.
Resolution: Blocking fcntl() file lock requests can now be stopped and started with.out
interrupting the fcntl() system call.
PR-19385 PR-20219 PR-20967
The rpc.lockd daemon aborted and dumped core.
Resolution: The lock manager has been altered and restructured for ConvexOS VlO.O,
resulting in a much more stable lock manager.
If a file resides on machine A and a process that runs on machine B (which NFSmounts the filesystem containing the file) locks the file with lockf (fd, F_TLOCK, 180),
and machine B crashes, it is not possible to lock the file from either system after B is
rebooted because of permission problems.
Resolution: This problem has been fixed.
The "SEE ALSO" section of the publickey.3r man page has a line truncated to: "RPC
Programmer's Manual in"
Resolution: The man page for publickey(3r) has been corrected to refer to the RPC
Programming Guide in the Convex NFS Reference Set.
The rpc.statd program supports the -dlevel option, but the man page makes no mention of any options. The -dlevel option needs to be explained on the man page.
Resolution: The statd.8c man page has been updated to include descriptions of the d and -D command line parameters.
The man pages for domainname(8), getdomainname(2) and setdomainname(2) do
not mention that they are for NIS(YP) domains as opposed to the Internet Domain
Name. The NAME and DESCRIPTION sections should have NIS added before the
word "domain".
Resolution: The man pagesdomainname.1 and getdomainname.2 have been updated
to clarify that the domainname referred to is the yellow pages domain.
CONVEX NFS Release Notice
The intro.3r man page refers to non-existent man pages. The following are referred
to but have no man pages: klm_prot (3r)nlm_prot (3r) sm_inter.3r yp (3r)
Resolution: Added the man pages for klm_prot3r, nlm_prot.3r, sm_inter.3r, yp.3r
and edited the descriptions of the standard RPC services in intro3r.
The exports(5) and netgroup(5) man pages should state that to use netgroups you
must also be running and using NIS(YP).
Resolution: A note has been added to the man pages exports.5 and netgroups.S indicating that netgroup relies on NIS(YP).
The netdisk install script "setup_client" does not work unless the system is running
NIS (Yellow Pages).
Resolution: setup_client has been modified to get the domain name by executing
/bin/ domainname instead of parsing the re.local file. If the YP service is to be used
during the client installation, the domain name must be set and yp must be running
when setup_client is executed. If the domain name is not set, YP cannot be running
and will not be used by setup_clicnt.
If there is a domain name configured, setup_client will use 'ypmatch' and thus, the
domain name _must_ be set and yp _must_ be running when setup_dient is executed.
The permissions on /etc/rmtab are not secure enough. It seems the permissions
should be rw-r-r- but not rw-rw-rw.
Resolution: The permissions of /etc/rmtab have been changed to be rw-r-r-.
The I re.• files are currently setup to perform NFS mounts before they have started their own NFS server daemons(mountd). The portmap (mountd) daemon should
be started before any NFS client requests are performed. This would alleviate the
problem with systems that are both NFS servers and clients, that must timeout the
NFS mounts before they can start the NFS server daemons.
Resolution: The start-up of inetd has been moved to re.local from re.std. Now, inetd
will be running and able to start mountd before NFS mounts are attempted.
CONVEX NFS Release Notice
When a filesystem is exported with a-anon=-1, the authentication fails during the
Resolution: The authentication now functions properly with-anon=-1.
Inconsistencies in the treatment of uids in the Unix kernel result in problems with
NFS access when a negative uid is used. In particular, a directory can be created but
the permissions will not allow use of that directory. This is a problem with root access
when the filesystem is exported with "anon=-2" and with access by PC clients, which
often use uid -2.
Resolution: File access with a negative uid is now consistent in its behavior.
An NFS operation is leaving files named .nfs?????? around which can take up a large
amount of disk space. It appears to be caused by programs that unlink temporary
Resolution: NFS operations on open and unlinked files now properly clean up upon
If the udpcksum tunable parameter is set to 1 in the bootcmd.local file, NFS will no
longer work.
. Resolution: There were two problems: 1. setting wrong IP source add before UDP
checkless calculation, 2. overwriting IP header length when calculating UDP checkless. Both problems are fixed.
Unlinking open files over NFS on a machine running bids can in some cases result in
a system hang.
Resolution: The problem was fixed by not inserting unlinked/open files into the directory name lookup cache.
When a "mv" is performed of a file as large or larger than the amount of physical
memory in the system to an NFS mounted fleshiest, all system memory can potentially be allocated to buffering data for the "me" command and the system can deadlock.
CONVEX NFS Release Notice
Resolution: NFS operations are no longer capable of buffering an unlimited amount
of memory during transfer operations.
The NFS biod daemons can potentially deadlock during interaction with the virtual
memory system.
Resolution: The deadlock has been eliminated.
Killing an nfs process can result in a system hang if it is killed while the NFS server is
executing client requests.
Resolution: The nfs process now shuts down properly when it is terminated.
An nfsd process can hang during a read directory operation if the directory entry on
the filesystem is corrupted.
Resolution: The nfs code that reads directories now does sanity checks on some aspects of the directory information passed to it from the filesystem.
The owner of a file which does not have read or write permission for the owner cannot read or write the file over NFS if the owner is -2 (for anon=-2 mounts).
Resolution: The problem has been fixed.
The NFS server code defines a global linked list (rfsfreesp) which is not protected by
a semaphore. This problem could potentially cause corruption of the rfsfreesp list
and a system panic.
Resolution: The list rfsfreesp is now protected by a semaphore.
A table involved in nfs/rpc authentication became corrupted. The corruption caused
a system panic.
Resolution: The source of the authentication table corruption has been fixed.
There is currently a limit on how long an NFS access list can be in the I etc/ exports
file. If a list is specified that exceeds the limit, no error is returned and the behavior
is the same as if there were no access list.
Resolution: The exportfs command has been modified to display a warning if an entry in the /etc/exports file is too long.
CONVEX NFS Release Notice
The NFS server code does not support read-dir request sizes of less than 512 bytes.
Resolution: Support for read-dir requests less than 512 bytes has been added to the
NFS server code.
If 'root' runs passwd to change the root password, the passwd application wants to
use the new password to encrypt the secret key for root in the yp map publickey.byname. The string variable 'newpass' is passed to the reencrypt_secret() function.
But in the main() function, the newpass string variable is not initialized if the user id
is 0. Thus, the secret key for root will be encrypted using a non-valid password and
is then stored in the yp map.
Resolution: /bin/passwd was changed to properly initialize the string used to re-encrypt root's secret DES key in the publickey.byname YP map.
The rpcgen utility should check for read/write failures and call perror() if an error occurs.
Resolution: Error checking and diagnostic messages have been added to rpcgen in the
case of failed reads and writes.
A User process can hang on the Oient side when accessing a file over NFS over the
Ultranet uip interface.
Resolution: The problem has been fixed.
CONVEX NFS Release Notice
Files including /usr/include/des_crypt.h will not compile correctly.
Resolution: The file /usr/include/des_crypt.h has been fixed so that files including
it will compile correctly.
PR-10828 PR-11788
ypserv and ypbind should use syslog to report error conditions.
Resolution: The ypbind and ypserv daemons now report all errors via syslog messages.
When an ordinary user changes his or her password with yppasswd, yppasswd
wants to encrypt the users secret key in the publickey.byname yp map using the new
password. yppasswd does a getnetname() to get the name of the user doing the pass-
word change. The library function getnetname() uses the process' effective user id to
determine which user name to return. The yppasswd application is running with the
effective user ID of root since the file mode is suid as root.
Resolution: The correct effective user ID is now set by yppasswd before getnetname()
is called to get the users network name.
The yppasswdd daemon does not execute a mkpasswd after updating the I etc/passwd file, thus local logins and applications that use the password dbm database files
Resolution: The yp password daemon (rpc.yppasswdd) has been modified to add the
'mkpasswd' command to the command string which does the ypmake if the-m flag
is supplied.
The system accounting utility I etc/sa will du mp core when executed on an NIS(yp)
Resolution: The problem has been fixed.
CONVEX NFS Release Notice
Known Software
This section lists the known problems with the CONVEX Network File System Uti\1ties software as of December, 1991. Problems reported after this date may not be reflected in the document. Please refer to this list prior to reporting a problem to ensure
that it has not been previously reported. Serious problems include workarounds if
they are known.
PR-18535 PR-18535
If the SunOS installation is done in a fresh, cleaned-out file system and /tftpboot
points to /filesystem/tftpboot but the directory isn't there, the installation of the client fails.
Workaround: Do mkdir /filesystem/tftpboot and retry the installation.
The daemon rpc.bootparamd can dump core if it fails in an attempt to read kernel
If the search path is sufficiently strange, programs called from INSTALL can fail to
find binaries they need. The reason is that INSTALL is not exporting its PATH variable.
The command 'rpcinfo -u host nlockmgr' produces the following output: program
100021 version 1 ready and waiting rpcinfo: RPC: Procedure unavailable program
100021 version 2 is not available
Mountd should be started automatically by inetd if it has aborted.
CONVEX NFS Release Notice
Reading and writing to a named pipe can hang forever if the named pipe is on different file systems but is the same underlying object. Somehow, because of NFS, the system does not realiz.e it is the same object. However, NFS is suppose to be transparent
to the user.
If the "noac" flag is used to do an NFS mount from a X9.0 system to a V8.1 system,
then a "trace/BPT trap" message will appear when an executable is run.
When a filesystem is exported to a specific list of machines, many of which are HP
workstations, it is exported -rw, and the HPs mount it -rw. The first 16 machines in
the host list work correctly, but all HP workstations after the 16th machine do not
have write access to the filesystem. Other CONVEX machines get write access to the
filesystem regardless of its position in the host list.
PR-15585 PR-19430
The library version of chkpnt is having problems with fdpath failing under certain
circumstances. This appears to be related to NFS mounting the filesystems.
PR-19297 PR-19337 PR-19688
If a file system is mounted from a COC4680 to a CONVEX machine with trailers enabled on either system, file access problems occur on files containing approximately
923 bytes. File accessing programs, such as "cat", fail with ''RPC timed out" messages.
Files larger than 925 bytes and smaller than 922 are handled without any problems.
The NFS client code can become completely stuck if one NFS server goes down. This
is due to the NFS client code queuing requests for biods even when no biods are waiting for work. The NFS client code will never execute a read or write request in the
application process context When close()s are executed, they can become stuck waiting on buffers to be flushed which are on the biod queue.
While running a user application, a file that is on an NFS file system is sometimes
reported to be smaller than it actually is on the local (server) disk.
Due to the long latency times for local file access introduced by file migration systems, it is possible for NFS client and server code to hang during interaction with a
file migration system.
The DES authentication table for nfs/rpc in the kernel can become corrupted resulting in a system panic.
CONVEX NFS Release Notice
On page 3-3 of the NFS User's Guide, the paragraph which starts: ''Note the convexs
and eclipse subdirectories" there is no eclipse subdirectory in the example listing.
There is a "tinman" subdirectory, however.
Hone kills a stopped rex process, the rexd on the remote host doesn't know that this
has happened and the rexd will not properly clean up the executing processes.
Rex will hang in select once the user exits from a remote interactive rex session. The
remote shell and rexd exit normally.
If get_myaddress() is unable to open a socket file descriptor, it syslogs an error message and exits. It should not exit.
After a power failure, there can be a problem with statd on machines all starting at
the same time. The statd process will try to contact all known machines and not listen
for any incoming calls. The problem is that statd will not listen to requests while
they are trying to contact other statds. When many of them start up at one time, the
result is many processes talking and few listening.
Workaround: This can be worked around by removing the contents of the /etc/sm
and the I etc/ sm.bak directories. This, however, breaks any outstanding file locks.
CONVEX NFS Release Notice
There is no discussion of the publickey, netid, and rpc maps in the NIS(YP) Chapter.
All three are in the /etc/yp/Makefile.
There is no explanation in the System Manager's Guide on how the number of nfsd
and biod daemons can change the performance of NFS.
PR-20580 PR-20675
The NIS chapter should mention that for send mail to access aliases in the NIS mail.aliases maps, the sendmail 0 option must be set to p.
On page 3-19 of the Convex NFS System Manager's Guide, Using the Automounter
(Order No. DSW-113), the manual should include the following error messages: ''Domain name not set. Cannot use YP." "YP bind failed: can't communicate with ypbind".
The following error message is not documented in the NFS System Manager's Guide.
mount: serverhost:/exported/fs server not responding: RPC: Authentication error;
why= Invalid client credential mount: giving up on: /exports/fs
The NFS Manager's Guide should mention somewhere the necessity for the nobody
entry (nobody:•:65534:65534::/:) in the /etc/passwd file.
A new log message has been added to the automounter which should be documented in the NFS System Manager's Guide. The message is:
path: <pathname> does not exist on host: <hostname>
Where <pathname> is the pathname that was mounted from the remote
(including any subdirectories from indirect maps using':') and <hostname> is the
name of the NFS server. To correct the problem,
create the specified directory on
the NFS server system.
The NFS System Manager's Guide should document the procedure for getting mail
to work with the YP aliases map.
CONVEX NFS Release Notice
The procedure for changing the netgroup map, then getting mount to recognize the
new netgroup entry should be added to the NFS smg.
If a system configured to use YP is booted at a time when no YP server is available,
the boot does not complete and nothing can be done to the system short of a reboot.
We have two convex machines. One (A) is the YP master. It is running ypserv, ypbind
and yppasswdd. The second machine (B) is a YP slave. it is running ypbind and yppasswdd. Reading maps on both machines works fine. However, problems arise
when a user wants to change his password. on (A) password checking is done against
/etc/passwd, not against the YP map. on (B) password checking is done against the
YP map. Changing a password on (A) with /bin/passwd does not change the YP
Attempting to build yp maps for a non-existent domain will result in changes to the
maps in the default domain.
CONVEX NFS Release Notice
CONVEX Network File System Utilities Release Notice
Document No. 710-003730-009
Download PDF
Similar pages