From Bugzilla Helper: User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.4) Gecko/20030805 Description of problem: Not sure if this is really an sfdisk or Compaq raid driver issue, but this problem doesn't exist on AS 2.1 and does in the latest beta (and probably previous ones). [root@saias11 etc]# sfdisk -s total: 0 blocks [root@saias11 etc]# sfdisk -s /dev/cciss/c0d0 17776560 On a similar machine running AS 2.1: saias10 / 954# sfdisk -s /dev/cciss/c0d0: 17776560 total: 17776560 blocks Version-Release number of selected component (if applicable): util-linux-2.11y-22 How reproducible: Always Steps to Reproduce: 1. run rhel3 beta1 on a Compaq machine w/ raid card that uses cciss driver 2. "sfdisk -s" 3. you get "total: 0 blocks" Expected Results: should see /dev/cciss/* devices in "sfdisk -s" output. Additional info:
this is due to cciss/cpqarray using "c0d0" as the dev node name. sfdisk skips entries that end in a digit, because it thinks it is a partition.
If it is sfdisk skipping the entry for that reason, then it is a new bug in sfdisk (not the HP/Compaq raid driver) and it would have the same broken behavior on machines that use the cpqarray driver (it has names like /dev/ida/c0d0, etc...)
The only workaround I can put in will result in an incorrect count on the 'total:' line. /proc/partitions does not tell me whether a 'partition' is a whole disk, partition, RAID array, LVM group, etc. - implementing this this would need kernel changes. The 2.1AS implementation of sfdisk does really aweful things that shouldn't be forward-ported. Since '-s' works fine if you specify a block device, the consequences are fairly minor. The main problem is coming up with a solution that works with /proc/partitions (perhaps that is to just decide that the 'total:' line should be eliminated - is that acceptable?).
Waiting on feedback from bug reporter
I guess that's acceptable... though I'd think there would be some way to work around this that wasn't as much of a mess as the RH AS 2.1 version.
The fix has been in the devel tree for a long while (pre RHEL3 IIRC) - sorry for not noting that here.
FWIW, the latest util-linux going into the devel tree has a better fix for this issue. The 'total' size count should be back for FC4 and beyond.