Table of Contents
Please read the notes that correspond to the architectures you are installing for. If you find additional gotchas, or anything incorrect in these notes, please send your updates to mailto://[email protected] after checking that they are not known/fixed problems in the Amanda patches page: http://www.amanda.org/patches/.
You may have compilation errors on statfs.c if you're running, on a Solaris 2.6 host, a gcc that was not build on a Solaris 2.6 host. This happens because gcc stores fixed copies of some Solaris header files on an internal directory. You must rebuild gcc if you get this kind of trouble. Note, however, that gcc 2.7.2.3 does not support Solaris 2.6, you should upgrade to 2.8.0 or higher, or egcs.
You may get errors running make:
Assure that you use the BSD-version of make, usually /usr/ccs/bin/make
.
Add /usr/ccs/bin
to the path before running configure.
According to Julian Stevens <[email protected]>
, the format
of inetd on Trusted Solaris 1.2 is a bit different. Just before
the user name, you should insert the word `untrusted':
amanda dgram udp wait untrusted amuser /usr/local/libexec/amandad amandad
The patch-system script is *NOT* aware of this detail; you must fix it yourself.
A bug in GNU-tar 1.12 causes it to miscalculate (in fact, to misreport) the size of filesystems. A patch for GNU-tar is available in the patches directory.
The Ultrix dump program contains an explicit check that it is being run by root. This defeats the usual practice of a non-root "operator" userid for dumps. For this reason, the rundump program (a setuid-root wrapper for dump) is enabled by default. If you find rundump is not necessary for you, just run
configure --without-rundump
The Ultrix restore program will fail if it is not connected to a tty.
Since the restore program is invoked in the clients in order to create
index files, and it the client is not connected to a tty, index
creation will fail. Using GNU-tar instead of DUMP is an option.
Thanks to Edwin Wiles <[email protected]>
for the
investigation. Another alternative proposed by Martyn Johnson
<[email protected]>
is to use a modified restore program:
use a binary program editor and replace `/dev/tty' with `/dev/nul',
for instance, and link /dev/nul to /dev/null. Note that the chosen
file name must be exactly 8 bytes long, otherwise you'll break the
restore program. A nice one-liner perl script by Martyn Johnson will
do the trick (make sure you preserve a copy of the original restore
program, it will be rewritten by running this script!):
perl -pi -e 'BEGIN { $/ = "/dev/tty" } s-$/-/dev/nul-' restore
The Ultrix C compiler seems to be too broken to compile Amanda. You are likely to need gcc or egcs.
You may run into an internal /bin/sh limit when running the configure script. The error message is:
./configure: sh internal 2K buffer overflow
Using /bin/posix/sh usually works around the problem. Another solution is to install GNU bash and use it instead of /bin/sh.
If `configure' complains about not finding `lex', you'll have to get `flex' installed. Look for its URL in Amanda Installation Notes.
If you use logical volumes, you may refer to mountpoints or full device pathnames instead of device names in the disk list file.
According to Stan Brown <[email protected]>
, amverify won't work with
HP/UX's stock mt. The work-around is to install GNU cpio, that
contains an implementation of mt, and edit amverify so that MT
points to GNU mt and MTOPT is `-f', or reconfigure and rebuild amanda
from scratch, making sure it finds GNU mt before the stock mt in the
PATH.
If you have vxfs filesystems to back up, Amanda will pick vxdump automatically.
GNU-tar 1.12 will incorrectly report the size of backups. There's a patch in the patches directory that fixes this problem.
The use of `amhpfixdevs' is deprecated, since you can list
mount-points or full device names in the disklist
. The script may be
removed in future releases of Amanda.
Sometimes you may get the error message `Link severed' from an HP/UX
client. This is just a cute name the developers of HP/UX found for
`Network error'. Reported by Michael Brehl <[email protected]>
Linux hosts intended to back up efs partitions with dump should install the dump package, as it is not installed by default on most Linux distributions. It is possible to find compiled versions of dump on most Linux sites and CD-ROMs.
Note, however, that DUMP for Linux has been quite unreliable,
sometimes being unable to perform backups, producing weird error
messages such as `master/slave protocol botched', and sometimes
creating unrestorable dump images, especially of active filesystems.
The first problem can sometimes be fixed by cleaning up outdated
entries in /etc/dumpdates
, but your best bet is probably GNU-tar.
Make sure the user that runs configure has permission to run the dump program, otherwise configure may misdetect an ability of dump to accept a -E (for estimates) switch.
GNU-tar 1.11.8, distributed with some Linux versions, will cause index failures (Index returned -1). Upgrading to GNU-tar 1.12 fixes this problem. This is not a Linux-specific problem, but it is quite common in this platform.
Amanda now supports the ftape driver version 3.04d. It adjusts the blocksize automatically to 32k and supports QIC volume tables. More details can be found in the file ZFTAPE in this directory.
Some releases of dump for Linux, such as the one distributed with
Debian 2.0, have modified dump so that it stores dumpdates in
/var/lib. If this is your case, you should create a link to it in
/etc. Suggested by David Wolfskill <[email protected]>
According to Michael Galloway <[email protected]>
, the stock
DUX4 dump is broken. There is a patch available at
ftp://ftp.service.digital.com/public/dunix/v4.0b/duv40bas00005-19970926.README
When both dump and vdump are available, Amanda will use vdump for backing up advfs filesystems only, and dump will be used for the rest. If you'd rather back up all filesystems with vdump, #undef DUMP in config/config.h after running configure.
Unfortunately, the output of `dump -E' incorrectly matches a line of
output from SAMBA, which gets Amanda's estimate process confused.
client-src/sendsize.c
will refuse to compile if both
HAVE_DUMP_ESTIMATE and SAMBA_CLIENT are defined in config/config.h
.
Amanda will work correctly if you undefine HAVE_DUMP_ESTIMATE in
config/config.h
; if you prefer to have incorrect estimates for SAMBA
backups, follow the instructions in client-src/sendsize.c
on removing the
compile-time error.
According to Oren Laadan <[email protected]>
, DEC compiler version
"DEC C V5.2-033 on Digital UNIX V4.0 (Rev. 564)" (obtained with "cc -V")
does not build Amanda properly, in particular, taper.c. Using gcc is OK.
The use of `amsinixfixdevs' is deprecated, since you can list mount-points or full device names in the disklist. The script may be removed in future releases of Amanda.
Sinix port originally by Michael Schmitz <[email protected]>
.
When setting the tape device name in either amanda.conf
or one of the
changer configuration files, make sure you specify the "variable" device
name, which has a 'v' on the end. If not, IRIX will write 4KByte blocks
instead of the 32KByte blocks Amanda tells it to. This apparantly works
OK unless you take the tape to a non-IRIX system, where amrestore will
complain about a short (4096) read.
If you do end up in this situation, the dd command to use to pipe into your system restore program is:
dd if=/dev/whatever bs=4k skip=8 | ...
Jean-Francois Malouin <[email protected]>
reports that, if you are going to use an IRIX host as the tape server,
you *must* patch your system with *all* kernel and scsi rollup
patches, otherwise you may end up with useless tapes, due to tape
rewinding after short periods of inactivity. See
http://www-viz.tamu.edu/~sgi-faq/ for more details.
Some unpatched version of xfsdump are reported as not printing
estimates. This causes estimates to take much longer than they had
to, because backups are always performed twice. According to Mike
Acar <[email protected]>
, patch 2333 for IRIX 5.3 and 6.1 will
fix this problem.
Seems like SGI make program is a bit broken, in a way that causes it to rebuild some files if doesn't have to if you happen to run make again. Using GNU make fixes this problem.
If you have xfs filesystems to back up, Amanda will pick xfsdump automatically.
Luc Lalonde <[email protected]>
contributed the following notes:
If you use a jukebox, you must set the ownership of the robot-changer
device to the Amanda operator:group in /etc/ioperms
. Here's my
configuration:
/etc/ioperms: /dev/scsi/sc8d6l0 0600 amanda backup
Otherwise the ownership:group is changed to "root:sys" after each reboot.
When you do upgrades, check the file /var/sysgen/master.d/scsi
to see if
it has changed. Otherwise your jukebox could be rendered unuseable. In
particular, check if it has been replaced by a new version and renamed
to "./scsi.O/.".
If you use the Amanda package provided by freeware.sgi.com, you are not affected by the first question since at compile time the Amanda operator is "root:sys".
Jens Krause <[email protected]>
has reported some problems with
GNU-tar 1.12
on SCO Release 5. Although the `sparse' files patch was applied,
GNU-tar would consistently crash. GNUtar had to be built linked with
malloc-libraries, and the `--sparse' switches had to be removed from
client-src/sendbackup-gnutar.c
and client-src/sendsize.c
.
chg-scsi was not updated to support the new camlib.h-dependent chio.h, so chg-scsi will be automatically disabled if camlib.h is found. You may use chg-chio instead.
Sep. 28th, 2004: Jason Miller <[email protected]>
reported problems with setting up the
Amanda-client on FreeBSD 4.9. He wrote:
Due to the need for read permissions for Amanda-client the default user and group for this on FreeBSD 4.9 is "operator:operator" which I found a write up on that as well. Just a note the port wanted to install it with these user permissions by default and I initially changed them to match my Redhat 9.0 install. So just doing a
make distclean uninstall install Amanda_SERVER=servernamefixed that for me. Then I just followed the below instructions and everything was good to go.
Refer to this link for more details: http://www.freebsd.org/cgi/query-pr.cgi?pr=59302.
Nicolas Ecarnot <[email protected]>
discovered that for
FreeBSD 5.1 (maybe earlier, and surely further), you have to set the
net.inet.udp.maxdgram TCP/IP variable to 65535. The default is 9216,
and this is a problem when trying to backup a large number of clients
as indicated by errors in during amcheck or the estimate phase.
You can just run the command:
sysctl net.inet.udp.maxdgram=63535but this won't last until the next reboot.
To make it permanent, just add this line:
net.inet.udp.maxdgram=65535
in the file /etc/sysctl.conf
.
sendsize is reported to coredump on AIX 4.3.3, this is a linking problem, try configuring Amanda with the option "--disable-shared".
Although Amanda won't run standalone on MS-Windows hosts, it is possible to use it to back up their disks, by using SAMBA. Please read Backup PC hosts using Samba for more information.
SAMBA may be unable to back up some files due to file locking
restrictions. Particularly, paging and registry files usually present
problems. Backing up page files is pointless, but registry files are
quite important to back up. It is possible to create regular files
that contain registry information by using the Regback utility, from
the Windows NT Resource Kit. Unfortunately, it is not part of the
Windows NT standard distribution, you have to purchase it separately.
Thanks to Ernie Oporto <[email protected]>
for the tip.
For notes on how to setup Amanda under Apple's OS X, please refer to
http://web.brandeis.edu/pages/view/Bio/AmandaMacOSXCompileNotes,
written by Steven Karel<[email protected]>
.
Thanks to Jose L.Hales-Garcia <[email protected]>
for the tip.