dd(1)'ing raw disk devices

[Don Lewis]

In article <21427 at ppgbms.UUCP> paul at ppgbms.UUCP (Paul Evan Matz) writes:
>In article <89323.005700BACON at MTUS5.BITNET> BACON at MTUS5.BITNET (Jeffery Bacon) writes:
>>Let's suppose I do the following: (In the hypothetical realm for now:)
>>
>>ctsee8(4)# /bin/dd if=/dev/rsd0c | rsh ctsee5 /bin/dd of=/dev/rsd0c
>>
>>(ctsee8 is a Sun 3/60 running SunOS 4.0.3 with a 327MB scsi; ctsee5 is
>>a diskless 3/50 client of ctsee8, with a formatted 327MB scsi disk
>>attached and configured, but not being used by the machine.)
>>...
>
>When you say "attached and configured", do you mean that:
> a) the diskless 3/50's kernel has been configured with an entry like:
>	disk          sd0 at si0 drive 0 flags 0
>
> b) you have run format on the disk, at which time you defined the 
>    partition layout, which is tranlated into the "disk label".  The
>    partition layout must match the layout of the source disk for the
>    data to be meaningful to the destination disk host.
>
> c) if you intend to use the disk for root and swap, you have also added
>    the appropriate entry to the kernel config file (this is not required
>    to be able to write to the disk, but would be if you eventally what
>    to use it).  The example entry implies that partition "a" is used
>    for the root partition, and partition "b" is used for swap.
>	config        vmunix         root on sd0 swap on sd0
>
>
>>...
>>I know I am copying over the partition table; I can handle that. But what
>>about the defect list? And mapped bad sectors? Or anything else important?
>>
>
>Once the disk really is "configured", using dd or any raw disk device
>access utility will not "copy over the partition table".  The disk
>label is used by the raw device to understand where the boundaries of
>each partition is.  (You may very well corrupt an adjacent file system,
>created via newfs, on a partition with existing data when writing to a
>disk with a differnet partiton/label topology).
>
The disc label is apparently stored in the "a" partition.  Copying the
raw "a" partition from another disc will overwrite the label and partition
information.  I have actually had this happen to me.  Also, this is
why you are not allowed to swap to the "a" partition in SunOS 4.0.x (the
ufs code knows not to overwrite the label but not the swap code).

>On a SCSI disk, each of the eight partitions has a starting SCSI
>logical block address and length (Some may are not used, and are
>set to 0 length).  The way these logical block addresses are specified
>is by starting cylinder and number of cylinders.  You can look at a
>disk's label info by using dkinfo:
>
>dkinfo sd0
>sd0: Emulex MD21 controller at addr 140000, unit # 0
>1018 cylinders 8 heads 34 sectors/track
>a: 16048 sectors (59 cyls)
>   starting cylinder 0
>b: 33456 sectors (123 cyls)
>   starting cylinder 59
>c: 276896 sectors (1018 cyls)
>   starting cylinder 0
>g: 87856 sectors (323 cyls)
>   starting cylinder 182
>h: 139536 sectors (513 cyls)
>   starting cylinder 505
>
>
>By convention, /dev/rsd0c happens to map onto the entire disk.  So if the
>disk you want to "copy" is the same size, and has the same labeling
>setup (exactly), then what you want to do will work, and the data should
>be readable and usable, so long as the machines are the same architecture.
>(You will have to modify some of the host specific files, like /etc/fstab,
>/etc/rc.boot, etc...)
>
>It might be faster to copy only the partitions you really want (no reason 
>to copy sd0b, as it is swap).
I just did it this and it works.  In my case, however, I only wanted
to duplicate the "a" and "g" partitions.  I wanted to change the
allocations between some of the other partitions.  After copying the
"a" partition, I had to re-label the disk to get the partition sizes
back to what I wanted.

>
>Also, most SCSI disks handle the defect list/bad sectors transparently.
>The sd driver just specifies the logical block address and transfer length
>of the data it wants, the disk does the rest.  One really never knows the
>relationship between physical sectoring on the disk and logical block address,
>which is just as well.
The 327 is an ESDI disc hung off a SCSI->ESDI controller.  I don't think
the defect list is handled transparently on this disk.  Also, I think the
defect list is stored at the end of the "c" partition, but is mapped out
of the others by the driver.  I'm not real sure of this info.  I haven't
run across any reliable documentation of this.

>
>The nitty-gritty details of using dd, specifying # of blocks and count,
>still has me mystified, especially when dding between disks and tapes,
>or pipes.  The tar/bar/dump/dd/rsh man pages have lots of interesting stuff
>in them, but there seems to be much they infer (read between the lines, RBTL?)
>If anyone out there has any tips on this, please jump right in!
This is how I did it.
	Boot MUNIX, partition and label the disk
	Boot miniroot
	echo 'address otherhost' >>/etc/hosts
	ifconfig stuff to set my address
	rsh otherhost -n 'dd if=/dev/rsd0a bs=8k' | dd of=/dev/rsd0a obs=8k
	relabel as necessary
	copy other partitions
	mount /dev/sd0a /a
	edit /a/etc/rc.boot and /a/etc/hosts, etc.
	reboot
>
>Good Luck.
>_____________________________________________________________
>|Regards,                           One Campus Drive        |
>|Paul Matz                          PPG Biomedical Systems  |
>|914-741-4685                       Pleasantville, NY. 10570|
>-------------------------------------------------------------
>path ppgbms!moe!paul at philabs.philips.com


--
Don "Truck" Lewis                                     Harris Semiconductor
Internet (if you're lucky): del at semi.harris-atd.com   PO Box 883   MS 62A-028
Internet (if not): del%thrush at trantor.harris-atd.com  Melbourne, FL  32901
UUCP (works): rutgers!soleil!thrush!del               Phone: (407) 729-5205