This document was originally written by Daniel Moore , on Jan 12, 1998. Substantially rewritten by Alexandre Oliva . Additional corrections and additions from Murf and Ralf Fassel . It describes how to restore files backed up with amanda either with or without amanda tools. All these cases assume you're trying to restore a complete disk, that is, you've replaced the lost disk with a new one, or created a new filesystem on it. Tweaking with the arguments to restore (not amrestore), you will be able to restore individual files. Also, this text does not cover amrecover, a program that provides a text user interface similar to interactive restore (restore -i), but it allows you to select individual files to recover and automatically determines the tapes where they were stored. The backups must be performed with the `index' option enabled for this to work. I considered 6 cases. 1) Client machine fails, non-system critical. 2) Client machine fails, system critical disk. 3) Server machine fails, non-system critical, non-amanda disk. 4) Server machine fails, system critical, non-amanda disk. 5) Server machine fails, non-system critical, amanda disk, with db. 6) Server machine fails, non-system critical, amanda disk, with binaries. The server machine (machine Aaron) is solaris, the client machine (machine Barney) is sunos. Cases: 1) Client machine fails, non-system critical. Example: /home fails on Barney. First, use amadmin to find the tapes most recently used to backup the partition. amadmin info Barney '/home$' Current info for Barney /home: Stats: dump rates (kps), Full: 41.1, 33.1, 38.8 Incremental: 9.5, 2.1, 24.7 compressed size, Full: 63.1%, 54.0%, 52.9% Incremental: 43.7%, 15.5%, 47.8% Dumps: lev datestmp tape file origK compK secs 0 19971223 Barney01 16 329947 208032 5060 1 19980108 Barney16 8 1977 864 91 2 19971222 Barney06 7 1874 672 83 3 19970926 Barney03 11 12273 3040 211 This tells us that we will need two tapes to do a full restore (Barney01, Barney16). Note that, even if Barney06 and Barney03 are listed, they are actually older than the full backup, so they should not be used to restore any data. Log into Barney. Cd to the /home directory. Insert the tape with the level 0 dump on it into the tape drive of Aaron. Become super-user in the client host and run (replace with the username under which amanda runs): rsh -n -l Aaron amrestore -p /dev/rmt/0cn Barney '/home\$' | restore -ivbf 2 - This requires client root to have login access to @Aaron, with a .rhosts entry (.amandahosts won't do). If you use ssh, you may be able to type a password in order to be authenticated. Another alternative is to start the operation in the server, and rsh to the client. You should be the amanda user or root in the tape server and run: amrestore -p /dev/rmt/0cn Barney '/home$' | rsh Barney -l root /usr/etc/restore -ivbf 2 - If you don't want to use rsh at all, you may run: amrestore /dev/rmt/0cn Barney '/home$' This should create a file whose name contains the hostname, directory name, dump level and dump date of the backup. Now you have to move this file to the client somehow: you may use NFS, rcp, ftp, floppy disks :-), whatever. Suppose you rename that file to `home.0'. Then, on the client, you should become root and run: restore -ivbf 2 home.0 Repeat one of these steps, incrementing the level of the dump, until there are no more available backups. 2) Client machine fails, system critical disk. Example: / fails on Barney. First of all, boot off the CD, and reinstall the system critical partition, restoring it to vendor supplied state. Then, go through all of Scenario 1. 3) Server machine fails, non-system critical, non-amanda disk. Proceed just as described in Scenario 1. However, you won't have to go through the rsh process, because you can just use amrestore to replace the lost data directly. 4) Server machine fails, system critical, non-amanda disk. Example: / on Aaron First of all, boot off the CD, and reinstall the system critical partition, restoring it to vendor supplied state. Then, follow steps in Scenario 3. 5) Server machine fails, non-system critical, amanda disk, with db. An example: /opt on Aaron If the disk that the amanda database is toast, then you need to rebuild the database. The easiest way to do it is to take the text file that you had mailed to you via the 'amadmin export' command, and import via the 'amadmin import' command. Then you should be able to follow the steps outlined in Scenario 4. Note that amanda does not mail the exported database automatically; you may add this to the crontab entry that runs amanda. Maybe it's a good idea to print out the text files as well and store the last 2 dumpcycles worth of paper (the disc text files might have got toasted as well). From the paper you still are able to reconstruct where your discs are. 6) Server machine fails, non-system critical, amanda disk, with binaries. Example: /usr/local on Aaron This is where things get hairy. If the disk with the amanda binaries on it is toast, you have three options. i) reinstall the amanda binaries from another tape, on which you have conveniently backed up the binaries within the last couple of weeks (not using amanda). ii) recompile amanda, making sure not to overwrite your db files. iii) use dd to read amanda formatted tapes. This is the option I am going to explore most fully, because this seems the most likely to occur. a) Find out the device name used by amanda, by looking in amanda.conf. Turns out to be /dev/rmt/0cn for this system. If amanda.conf isn't at hand: this must be a non-rewinding tape device specifier (which I believe the trailing `n' stands for). b) Look over the copy of the output of 'amadmin export', and find out which tapes /usr/local was backed up on. c) Grab the tapes that /opt was backed up on, and stick the level 0 into the drive. cd to /usr/local. d) Skip the first record, which is just the tape header, by using the appropriate tape command. mt -f /dev/rmt/0cn fsf 1 e) Now you want to start looking for /usr/local on this tape. dd if=/dev/rmt/0cn bs=32k skip=1 | gzip -d | /usr/sbin/ufsrestore -ivf - This command gives us an interactive restore of this record, including telling us what partition, what host, and what level the backup was. The gzip -d portion of the pipe can be omitted if there was no compression. f) If you don't find /usr/local on the first try, quit ufsrestore, and move forward one record. mt -f /dev/rmt/0cn fsf 1 and try the dd/restore command shown above. Do this until you find /opt on the disk. Another possibility: quick and dirty tape index in case you don't know which partition /usr/local was on: (from ralf@atg.venture.de) #!/bin/sh TAPEDEV= while mt -f $TAPEDEV fsf 1 ; do dd if=$TAPEDEV bs=32k count=1 | head -1 sleep 1 done Example output: AMANDA: FILE 19971220 uri /root-sun4 lev 1 comp .gz program DUMP AMANDA: FILE 19971220 uri /misc lev 1 comp .gz program DUMP AMANDA: FILE 19971220 uri / lev 1 comp .gz program DUMP g) Restore the amanda binaries (what else do you need??), and then bail out of ufsrestore. You can use amrestore, as in Scenario 3.