Tape backup for UFS and ZFS filesystems under Solaris 10
If you want UFS filesystems to be snapshot, then you must create a backing store to house them.
With ZFS available, just create a filesystem at /snap.
# zfs create pool/snap
# zfs set mountpoint=/snap pool/snap
# chmod 700 /snap
Lacking ZFS, a symlink from /snap to /var/tmp will suffice (but see the note below):
# ln -s /var/tmp /snap
To disable UFS snapshots altogether, you could simply set the SNAPSHOT_UFS
environment variable to 0:
# env SNAPSHOT_UFS=0 ...
If no /snap is found, then the script will print a warning but continue without UFS snapshots.
You can configure exclusions and filesystems which you want backed up "live"
(without being snapshot).
The fileformat for both these configuration files is one filesystem mointpoint
per-line.
If you create a /snap zfs, then it should be excluded; if you backed /snap in
/var, then /var should not be snapshot.
# vi etc/exclude
/snap
# vi etc/nosnap
/var
The backup scripts supports configuration of arbitrary pre- and post-backup
scripts via etc/scripts.
If ntpd runs on the system, then it must be stopped in order to snapshot the
root filesystem. Examples for Solaris 9/10 are included.
Finally, add an appropriate entry to the crontab:
# crontab -e
30 2 * * 2-6 /sysadmin/backup/bin/oxbackup backup
Were you backing up to a remote tape, pass the appropriate arguments:
30 2 * * 2-6 /sysadmin/backup/bin/oxbackup -f csdns:/dev/rmt/1cn backup
In order to expose standard checkback functionality, add an alias to ~root/.cshrc:
# vi ~root/.cshrc
alias checkback '/sysadmin/backup/bin/oxbackup check'
Sometimes, the script will falsely report a failure on a zfs mountpoint when it encounters a filename containing UTF-8 characters, for example:
==> BEGIN: dump of /crm1 via pax: 2007-07-19 01:57:51
--> running pax...
pax: invalid character in UTF-8 conversion of 'oracle_uat/tstcrmdb/9.2.0/network/admin/dg_prim_diag_lcrm_&×tamp.out'
pax: file (oracle_uat/tstcrmdb/9.2.0/network/admin/dg_prim_diag_lcrm_&×tamp.out): UTF-8 conversion failed.
==> ERROR: dump of /crm1 failed: 2007-07-19 03:32:57
234612447+0 records in
1862003+1 records out
These can generally be safely ignored, though it may be wise to notify the owner of the offending file that it's not being backed up. If unsure, you can verify the amount of data written to tape by multiplying the records out by output blocksize, obs=63k. For example 1,862,003 * 63KB = 117,306,189 KB = 111.9 GB. Check the size of the filesystem by multiplying the zfs used by compressratio, in this case giving us ~108.1 GB:
# zfs get used,compressratio st1/crm1
NAME PROPERTY VALUE SOURCE
st1/crm1 used 41.4G -
st1/crm1 compressratio 2.61x -
The oxbackup script contains a wrapper around basic restore functionality, accessed via bin/oxbackup restore [mountpoint].
If mountpoint is a UFS partition this will position the tape correctly before launching ufsrestore.
For a zfs mountpoint it will position the tape then prompt for a space-separated list of fnmatch(3C) patterns to restore (relative to mountpoint, no leading ./).
A typical restore from a zfs might look something like this:
# ./oxbackup toc
--> rewinding tape
Backup of tcisdb1: 2007-07-19 01:10:02
1 ufs /dev/dsk/c0t0d0s0 /
2 ufs /dev/dsk/c0t0d0s3 /var
3 zfs pool/app /app
4 zfs st1/auth1 /auth1
5 zfs pool/brookes /brookes
6 zfs st1/cis /cis
7 zfs st1/cmis /cmis
8 zfs st1/crm1 /crm1
9 zfs pool/data /data
10 zfs st1/ecsis1 /ecsis1
11 zfs st1/edm1 /edm1
12 zfs st1/eis1 /eis1
13 zfs st1/fin1 /fin1
14 zfs pool/logs /logs
15 zfs st1/oxshare1 /oxshare1
16 zfs pool/sysadmin /sysadmin
17 zfs pool/sysadmin/home /sysadmin/home
18 zfs pool/users /users
1+1 records in
1+1 records out
# cd /tmp/foo
# ./oxbackup restore /crm1
==> initiating restore from /crm1
--> WARNING: restore will be to the current directory (/tmp/foo)!
--> rewinding tape
1+1 records in
1+1 records out
--> matched entry in table of contents: 8 zfs st1/crm1 /crm1
--> pax does not support interaction
Enter file patterns to restore: oracle/lcrmdb
USTAR format archive extended
oracle/lcrmdb/9.2.0/inventory/Components21/oracle.swd.oui/2.2.0.19.0/installlog.xml
oracle/lcrmdb/9.2.0/inventory/ContentsXML/comps.xml
oracle/lcrmdb/9.2.0/inventory/ContentsXML/libs.xml
1862003+1 records in
234612447+0 records out
483.47u 2004.12s 1:25:54.58 48.2%
When prompted with 'Enter file patterns to restore:', it is requesting the subfolders of the folder (so in the example it's crm1/oracle/lcrmdb that's being restored).
[csdns:~]> mkdir trestore
[csdns:~]> cd trestore
[csdns:~/trestore]> /sysadmin/backup/bin/oxbackup toc
--> [13:34:43] rewinding tape
/dev/rmt/0cn: write protected or reserved.
++> [13:34:43] ERROR: accessing tape
Exit 1
[csdns:~/trestore]> pfexec tcsh
info: TCSHRC: sourced from CSDNS
[csdns:~/trestore]# /sysadmin/backup/bin/oxbackup toc
--> [13:35:17] rewinding tape
Backup of dcisapp1 [2014-10-23 03:10:02]
1 ufs /dev/md/dsk/d10 /
2 ufs /dev/md/dsk/d30 /var
3 ufs /dev/md/dsk/d70 /cis
4 ufs /dev/md/dsk/d40 /brookes
5 ufs /dev/md/dsk/d50 /app
6 ufs /dev/md/dsk/d60 /users
0+1 records in
0+1 records out
[csdns:~/trestore]# /sysadmin/backup/bin/oxbackup position 6
[csdns:~/trestore]# ufsrestore iv
Verify volume and initialize maps
Media block size is 126
Dump date: 23 October 2014 05:10:03 BST
Dumped from: the epoch
Level 0 dump of an unlisted file system on dcisapp1:/dev/fssnap/0
Label: none
Extract directories from tape
Initialize symbol table.
ufsrestore > pwd
/
ufsrestore > ls
.:
2 *./ 10 oraform/ 19043 p0074784/
2 *../ 13 oraoemag/ 150714 p0074807/
175029 brookes/ 72916 orassom/ 30326 p0074855/
116209 dcrmapp/ 19002 p0023009/ 169003 p0075330/
3 lost+found/ 25600 p0054111/ 43064 p0075347/
86778 misbuild/ 25601 p0054131/ 80344 p0075634/
44817 miscron/ 42185 p0054270/ 338874 p0075715/
86779 misdploy/ 152536 p0071665/ 85246 p0076594/
128086 ora10clt/ 19044 p0072338/ 175980 p0076801/
14 ora92clt/ 6 p0073279/ 175037 p0076923/
15 oracrma3/ 25605 p0073604/ 72606 p0077178/
150402 oracron/ 7 p0073611/ 121305 p0077268/
8 oraecsm/ 25602 p0073923/ 755 quotas
102409 oraecsws/ 25603 p0074093/ 11 webecs1/
9 oraedmm/ 152560 p0074397/ 12 webedm1/
ufsrestore > cd p0074093/
ufsrestore > pwd
/p0074093
ufsrestore > ls
./p0074093:
27155 #afiedt.buf#
25603 ./
2 *../
34380 .Xauthority
...
42747 zztemp
ufsrestore > add *
Make node ./p0074093
...
Make node ./p0074093/xfer
ufsrestore > marked
./p0074093:
27155 *#afiedt.buf#
...
42747 *zztemp
ufsrestore > extract
Extract requested files
You have not read any volumes yet.
Unless you know which volume your file(s) are on you should start
with the last volume and work towards the first.
Specify next volume #:
Specify next volume #: 1
extract file ./p0074093/news/xx.
...
extract file ./p0074093/srec.logx
write error extracting inode 35142, name ./p0074093/srec.logx
write: No space left on device
5.81u 6.42s 14:02.19 1.4%
Exit 1
[csdns:~/trestore]#
The following outlines a manual restore, for example from a Solaris boot CD.
- Read the table of contents:
# mt rewind
# dd if=/dev/rmt/0c
Backup of tcisdb1: 2007-07-19 01:10:02
1 ufs /dev/dsk/c0t0d0s0 /
2 ufs /dev/dsk/c0t0d0s3 /var
3 zfs pool/app /app
4 zfs st1/auth1 /auth1
5 zfs pool/brookes /brookes
6 zfs st1/cis /cis
7 zfs st1/cmis /cmis
8 zfs st1/crm1 /crm1
9 zfs pool/data /data
10 zfs st1/ecsis1 /ecsis1
11 zfs st1/edm1 /edm1
12 zfs st1/eis1 /eis1
13 zfs st1/fin1 /fin1
14 zfs pool/logs /logs
15 zfs st1/oxshare1 /oxshare1
16 zfs pool/sysadmin /sysadmin
17 zfs pool/sysadmin/home /sysadmin/home
18 zfs pool/users /users
- Restore: Below, the mt command is used to "Rewind, fast forward to beginning of file X". This operation has been seen to fail silently on csvleadm1's second tape drive. After running
# mt asf 1
It is worth running the following to check that the drive has actually positioned the tape to the file number you have specified:
# mt status
HP DAT-72 tape drive:
sense key(0x12)= EOF residual= 0 retries= 0
file no= 1 block no= 0
If you still get a zero ( 0 ) on the file no= entry then try a different tape drive
2a. UFS restore:
The first thing you will need to know is which file number on the tape holds the directories and files you want to restore. This is available from the output of the table of contents acquired using the command:
#dd if=/dev/rmt/0c
Restores can be started interactively, wich drops you to a ufsrestore> shell. The function letter 'i' achieves this (see commands below). Note the 'mt asf 1' command moves the tape to file number 1 on the tape, having first rewound it. If you find you are in the wrong location on the tape, you can use 'mt fsf 1' to move forwards 1 EOF marker, without rewinding to the start first.
# mt asf 1
# cd /restore_path
# ufsrestore ivf /dev/rmt/0cn
Note that user directories can be in a variety of locations. eg. For systems Administration:
/brookes/sysadmin
/home/sysadmin
For standard users:
/users
Make sure you know where the files should be located on the server before trying to locate them on the tape, it will save you time.
Files to be restored need to be marked using the 'add' command. They can be located on the tape using the 'ls' command. (Top level directories on the tape do not require a leading '/').
ufsrestore > ls
.:
2 *./ 13 oraoemag/ 19043 p0074784/
2 *../ 72916 orassom/ 150714 p0074807/
116209 dcrmapp/ 19002 p0023009/ 30326 p0074855/
3 lost+found/ 25600 p0054111/ 144765 p0074930/
86778 misbuild/ 25601 p0054131/ 169003 p0075330/
44817 miscron/ 25604 p0054270/ 43064 p0075347/
Mark a directory and all of its files for restore using 'add'. The output will display all the files marked for restore.
ufsrestore> add p0054270
Make node ./p0054270/.subversion
Make node ./p0054270/.subversion/auth
Make node ./p0054270/.subversion/auth/svn.simple
To initiate the restore of files use the 'extract' command. You will be prompted for the volume to read. The volume refers to the tape used to backup the filesystem, if the particular filesystem backup ran over more than one tape you may need to specify something other than 1 here.
ufsrestore> extract
Extract requested files
You have not read any volumes yet.
Unless you know which volume your file(s) are on you should start
with the last volume and work towards the first.
Specify next volume #: 1
At the end of the restore you will be prompted wether or not to reset permissions on the restored directory. Generally you will not want to do this, so can respond 'N'.
2b. ZFS restore, files matching fnmatch PATTERN (nothing if you want to restore everything):
# mt asf 3
# cd /restore_path
# dd bs=1024k if=/dev/rmt/0cn | pax -r -pe PATTERN
As an example, csvleadm1 is attached to two tape drives. One is for backups of itself, the other for csvledr1 which a physical server in Wheatly H127.
- Identify the correct drive, by sticking a tape in, then ejecting it by specifying the path to the device
# mt -f /dev/rmt/0cn offline
Take a look to see if the drive you expected to eject has done so. If it has, record the device path used. If not, try:
# mt -f /dev/rmt/1cn offline
Keep going up the numbers until the the tape is eject from the drive you want to use.
- Read the table of contents:
# mt -f /dev/rmt/1cn rewind
# dd if=/dev/rmt/1c
Backup of csvledr1 [2012-08-23 01:10:06]
1 zfs rpool/ROOT/S10u6-20090629 /
2 zfs rpool/ROOT/S10u6-20090629/var /var
3 zfs rpool/ROOT/S10u6-20090629/opt /opt
4 zfs rpool/COMMON/brookes /brookes
5 zfs rpool/COMMON/logs /logs
6 zfs rpool /rpool
7 zfs rpool/ROOT /rpool/ROOT
8 zfs st1/sync-csvledb1 /sync-csvledb1
9 zfs st1/sync-tvle-db1 /sync-tvle-db1
10 zfs rpool/COMMON/sysadmin /sysadmin
11 zfs rpool/COMMON/sysadmin/home /sysadmin/home
12 zfs rpool/COMMON/users /users
1+1 records in
1+1 records out
- Restore Below, the mt command is used to "Rewind, fast forward to beginning of file X". This operation has been seen to fail silently on csvleadm1's second tape drive. After running
# mt -f /dev/rmt/1cn asf 1
It is worth running the following to check that the drive has actually positioned the tape to the file number you have specified:
# mt -f /dev/rmt/1cn status
HP DAT-72 tape drive:
sense key(0x12)= EOF residual= 0 retries= 0
file no= 1 block no= 0
If you still get a zero ( 0 ) on the file no= entry then try a different tape drive
3a. UFS restore:
# mt -f /dev/rmt/1c asf 1
# cd /restore_path
# ufsrestore ivf /dev/rmt/1cn
3b. ZFS restore
# mt -f /dev/rmt/1c asf 5
# cd /restore_path
# dd bs=1024k if=/dev/rmt/1cn | pax -r -pe PATTERN