SUMMARY of a SUMMARY: dump multiple filesystems on one tape?

From: Jeff Greer (jeff@datapage.com)
Date: Thu Jun 15 1995 - 06:49:08 CDT


Well, here we are again.

First let me apologize for not trying all of the examples that I included in my first summary. Of the three methods I summarized I am hearing from many folks that only one of them is correct.

The following is my summary to my summary.
Michael Shon (mshon@sunrock.east.sun.com) prompted this summary of a summary and included below is some of what he sent as a reply to my original summary.

Michael was not the only one who pointed out that two of the origianl three methods did not work and could be dangerous. Thanks to everyone who did.

The one and only:

  Using the non-rewinding device /dev/nrst? (4.x) do the following.
        # dump [options] /dump/file /file/system1
        # dump [options] /dump/file /file/system2

The elipses myth:

|>SYNOPSIS
| /usr/etc/dump [ options [ arguments ] ] filesystem
| /usr/etc/dump [ options [ arguments ] ] filename ...
|>
|SNIP
|>
|>You can see the single "filesystem" mentioned above, but there are
|>ellipsis after the "filename" signifying multiple filenames.
| ^^^^^^^^ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|

Filesystem refers to a device name. The device does not need to be mounted.
Filename refers to the pathname of a non-device file
 -- within a *mounted filesystem*. --

If the first filename is a device name, then all other filenames are ignored,
otherwise the filenames must ALL be within the same filesystem.
This restriction is not mentioned in the 4.x manpage,
but is in the 5.x manpage.
4.x enforces it anyway:

  4.x: dump 0f /tmp/dumpfile /etc /usr/etc
        DUMP: Dumping /dev/rid000a (/) to /tmp/dumpfile
        DUMP: /usr/etc is not on device /dev/rid000a
        DUMP: The ENTIRE dump is aborted.

 5.x:files_to_dump is required and must be the last argument on
     the command line. It specifies the files to dump. Usually
     it identifies a whole file system by its raw device name
     (for example, /dev/rdsk/c0t3d0s6). ....
     ... In all cases, the files must be contained in the same
     file system, and the file system must be
     local to the system where ufsdump is being run.

>The form of this command is:
> dump [options] /dev/nrst? /dev/sd0a /dev/sd0g ect.

The example above is NOT a valid form
         [the devices after the first are ignored].

The following IS a valid form.
        dump [options] device /export/home/user1 /export/home/user2

BUT data from only ONE filesystem is dumped.

The third way dosn't work either.

> What you can do for example is to mount the partition number 2
>(for SunOS : /dev/sd0c or for Solaris : /dev/dsk/c0t3d0s2) and there you will
>be able to make a full dump of your disk, because, by default, the partition
>number 2 is contains all the data of your different file system.

Partition "c" [s2] may allow access to all blocks of the disk,
but the only *filesystem* [a structured, mountable collection of blocks]
accessible on that partition is the one which begins at sector 0,
which is usually the one defined for partition "a" [s0] .
If parition "c" [s2] is mounted at the same time that partition "a" [s0]
is mounted, filesystem corruption may result, depending on what is
done on the filesystem. If it is mounted read-only, it is much less risky,
but potential for other problems exists.
 
So, not only is it a Bad Idea, but it will NOT allow you to dump all of the
filesystems on the device [unless there is only one filesystem there].
In addition, it does not have to be mounted in order to dump it -
just use the device name - [but that serves no real purpose anyway].

And of course check your backups before they are needed.



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:10:26 CDT