Christian,

Never assume that ECKD is CDL again, please. The CMS FORMAT command lays down a 
flawless array of 4k blocks before it stamps a CMS "EDF" filesystem on the 
volume.

You are correct that FBA has no problem with unpartitioned media. But most 
other disk architectures do likewise, including SAN, copper SCSI, IDE, flash, 
CD ROM. In fact, unpartioned is the norm for CD ROM (or DVD) and is required 
for pseudo devices like loop back and logical volumes.

CDL was a terrific addition, but its sole value is to simplify z/OS work. One 
could say perhaps it is CODDLING z/OS, perpetuating the myth that all the 
mainframe world is z/OS.

CDL buys us nothing in a purely Linux environment, or even a mixed environment 
where z/OS does not reference the Linux volumes directly.




-- R;


----- Original Message -----
From: Christian Borntraeger [EMAIL PROTECTED]
Sent: 09/21/2007 12:22 PM
To: LINUX-390@VM.MARIST.EDU
Subject: Re: Suse disk frustration

Am Freitag, 21. September 2007 schrieb Richard Troth:
> Skip the "partitioning" game. Put the filesystem on /dev/dasdd instead
of /dev/dasdd1 and mount that. (be sure all this matches /etc/fstab, of
course)

Rick,

Never suggest that again, please. You really have to know your systems very
well before you can do that.
It can work on FBA disks, but it will eat your file systems on ECKD disk,
formatted with the compatible disk layout. The first track on ECKD disks
usually has a different block size than the rest of the disk. As Linux only
understands block devices with all blocks the same size, the dasd device
drivers simulates a 4kb block on the smaller first blocks. As this block is
only used for the partition table and not for data  this is normally not a
problem.The block is padded with e5.

When you start to use the block as data block, you will lose data:

// fill byte 1000-2999 of dasdd with zeros
# dd if=/dev/zero of=/dev/dasdd bs=1 count=2000 seek=1000
2000+0 records in
2000+0 records out
2000 bytes (2.0 kB) copied, 0.00372012 seconds, 538 kB/s

//it still looks fine
# dd if=/dev/dasdd bs=1 count=2000 skip=1000 | hexdump
2000+0 records in
2000+0 records out
2000 bytes (2.0 kB) copied, 0.00397881 seconds, 503 kB/s
0000000 0000 0000 0000 0000 0000 0000 0000 0000
*
00007d0

//lets drop caches
# chccwdev -d ec13 ; chccwdev -e ec13
Setting device 0.0.ec13 offline
Done
Setting device 0.0.ec13 online
Done

//we now read e5 instead of zeroes. Data is lost!
# dd if=/dev/dasdd bs=1 count=2000 skip=1000 | hexdump
0000000 e5e5 e5e5 e5e5 e5e5 e5e5 e5e5 e5e5 e5e5
*
00007d0
2000+0 records in
2000+0 records out
2000 bytes (2.0 kB) copied, 0.0122469 seconds, 163 kB/s


This behaviour is quite subtle, so please never suggest to create a file
system or an lvm physical volume on a dasd without using a partition. It only
works on a subset of available disk formats.

Christian

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

Reply via email to