dump restore problem

[Matthew Dillon]

:
:David O'Brien writes:
: > On Mon, Apr 28, 2003 at 09:13:20AM -0700, Matthew Dillon wrote:
: > >     Use restore's 'r' option instead of 'x' if you want an exact restore.
: > >     'x' is used when restoring part of a backup, 'r' is used when restoring
: > >     the entire backup.  For example, if you just wanted to restore the 'etc
: > 
: > IIRC, 'r' doesn't restore the exact owner, modification time, and mode.
: > _______________________________________________
: >
:BTW my original point was to make somekind of best practices how to
:migrate ufs1 to ufs2.  Dump says maximum block size is 1000 and it
:doesn't work.  Why this doesn't work?  I understand that maximum block
:size is 64k when writing through device layer (physio?) but what's
:happening on this case?
:
:  Tomppa

    I don't recall the differences David suggest.  that doesn't sound right.
    Both 'x' and 'r' will restore all aspects of the file if possible, 
    including the owner, modes, and so forth, regardless of your umask.
    The -r option must be used if you are restoring through several
    incremental backups because only -r tracks auxillary information between
    the incrementals.  The -x option must be used if you only want to pull
    out a portion of a backup rather then the whole thing.

    -x will query you in regards to setting the owner and mode of '.' when
    it is done.  -r will not.  Using -x to do a full restore when you have
    a single level 0 dump file (or pipe) will probably work, but I don't
    know how it will treat hard links.  At BEST I always used -r to restore
    an entire partition, even if I only wanted to get a piece of it, because
    -x did weird things sometimes.

    -

    In regards to the block size, the physical device layer is irrelevant
    if you are dumping/restoring with files or pipes.  However, dump and
    restore may still do odd things if you specify numbers above 127.  I
    don't know why.  There is no real need to use a larger block size, it
    isn't going to make the operation any more efficient and, in fact, may
    make things LESS efficient because it will blow out your machine's L1
    cache on every block.

    That's my take.  If I am wrong about the options, Dave, I would 
    appreciate an example.

					-Matt
					Matthew Dillon 
					<dillon at backplane.com>