At 06:33 PM 1/15/97 +0100, Michael Fink wrote:
>We seem to be able to avoid most of (all of?) the mentioned problems
>by proceeding for recovery of UNIX clients as follows (not tested very
>thoroughly yet, but seems to work fine for SGI IRIX clients):
>
>- Partition new disk (using external documentation of partition layout)
> and/or make logical volumes (whatever is needed)
>- Boot miniroot (i.e. minimal root file system on swap partition)
>- Install minimal ADSM client software on miniroot (tar image)
>- mkfs "true" root and usr partitions
>- mkdir /root, mount respective partitions over /root and /root/usr
>- Restore / to /root and /usr to /root/usr using adsmc
>- mkdir other mount mpoints
>- cd /dev ; MAKEDEV (create devices)
>- Recreate and restore whatever other filesystems were damaged in
> analogous manner.
>- Reboot
The other day I had an opportunity to use ADSM to simplify recovering an SGI
running IRIX 6.2 from a failing disk. The disk was reporting recovered read
errors, but since we're running AFS and the cache was also on that disk, I
figured it was time to do a rebuild.
Success was achieved. Here are my notes; this is a pretty standard SGI IRIX
6.2 setup on an R4600 Indy except that we're also running AFS.
1. Install ADSM client software, run incremental backup with
"domain /" in dsm.opt. Write down/memorize client password.
[Note: this machine has two partitions: / and /usr/vice/cache]
2. Boot fx from IRIX 6.2 CD-ROM, reformat disk, run surface checks.
Partitioning sizes were changed but not the partition numbers.
3. Boot installation tools from CD-ROM.
4. Follow prompts to format / as an XFS filesystem. Installation
scripts automagically mount this as /root/.
5. Once at inst>, type "admin sh" to get a miniroot shell.
6. mkfs -t efs /dev/dsk/dks0d1s7
7. mkdir -p /root/usr/vice/cache
8. mount /dev/dsk/dks0d1s7 /root/usr/vice/cache
9. hostname <old machine hostname>
10. ifconfig ec0 A.B.C.D netmask 255.255.255.0 up
[Note: the ADSM server is on the same subnet. If this had not been
the case I would have done a "route add default E.F.G.H" command to
get a route to the ADSM server into the kernel.]
11. tftp'd sgi.v2r1m6.IP20995.tar from a tftp server to /usr/tmp.
[Note: IRIX miniroot doesn't have ftp, just tftp]
12. mkdir -p /usr/adsm ; cd /usr/adsm ; tar xvf /usr/tmp/sgi*tar \
adsm-sgi.client.tar.Z ; zcat adsm-sgi.client.tar.Z | tar xvf -
13. cp dsm.opt.smp dsm.opt
14. cat > dsm.sys
server ralph
commethod tcpip
tcpport 1500
tcpserveraddress A.B.C.X
<control-d>
[Note: I suppose I could have tftp'd this rather than typing it
in by hand, but I didn't bother to put the file on tftp server.]
15. Run /usr/adsm/dsmc and test with "query filespace" to verify
connectivity, correct node name, and correct password.
16. Then, at the dsmc prompt: dsmc> restore / /root -subdir=yes
[Note: this stopped from time to time asking to overwrite files;
if I did it again I would probably use the -replace=all option.]
17. After about an hour, all the files were restored from the ADSM
2.1.5.12 server on an RS/6000-59H reading from an IBM 3590 tape.
18. rm -rf /usr/tmp/sgi*tar /usr/adsm
19. cd /root/dev ; ./MAKEDEV
20. cd /CDROM/stand ; /root/sbin/dvhtool -v create sashARCS sash
21. mkdir /root/afs
22. exit miniroot shell, quit inst, allow to reboot
It didn't go as smoothly as it looks above; steps 18-21 took a few reboot
attempts to get worked out. Unless you do step #20, you must boot sash from
the CD-ROM and then run /unix from the appropriate partition.
There are some patches from SGI that update sash; after performing this
"bare metal" restore from CD-ROM and ADSM, I forced a reinstall of patch
#1397 which is the current (as of the last time I checked) XFS rollup patch.
This wrote the latest and greatest sash to the boot partition on the disk.
Note that this process restored /usr/adsm/*; however, you may want to test
dsmc again to make sure that if you're using passwordaccess=generate you
still have agreement between the ADSM client and server.
--
Bill Nickless nickless AT mcs.anl DOT gov +1 630 252
7390
Bill Nickless nickless AT mcs.anl DOT gov +1 630 252
7390
PGP 2.6.2 Key fingerprint = 0E 0F 16 80 C5 B1 69 52 E1 44 1A A5 0E 1B 74 F7
http://www.mcs.anl.gov/people/nickless
|