Did a "yum update" and noticed a filesystem created with cryptsetup using a stock FC4 system was no longer mounting. Rolling back to 1.0-1 fixed the problem. The following shell session adequately illustrates the problem: [root@viper ~]# dd if=/dev/zero of=foobar bs=1M count=10 10+0 records in 10+0 records out [root@viper ~]# rpm -q cryptsetup-luks cryptsetup-luks-1.0-1 [root@viper ~]# losetup /dev/loop0 foobar [root@viper ~]# echo blahblah | cryptsetup -c serpent-cbc-essiv:sha256 -s 256 create cp0 /dev/loop0 [root@viper ~]# dd if=/dev/zero of=/dev/mapper/cp0 dd: writing to `/dev/mapper/cp0': No space left on device 20481+0 records in 20480+0 records out [root@viper ~]# md5sum /dev/loop0 fc001241299983f3d59c1ec234c56f6c /dev/loop0 [root@viper ~]# yum -y update cryptsetup-luks &>/dev/null [root@viper ~]# rpm -q cryptsetup-luks cryptsetup-luks-1.0.1-0.fc4 [root@viper ~]# dmsetup remove cp0 [root@viper ~]# echo blahblah | cryptsetup -c serpent-cbc-essiv:sha256 -s 256 create cp0 /dev/loop0 [root@viper ~]# dd if=/dev/zero of=/dev/mapper/cp0 dd: writing to `/dev/mapper/cp0': No space left on device 20481+0 records in 20480+0 records out [root@viper ~]# md5sum /dev/loop0 44afd074d4b9f1b0d73b01198901fe4a /dev/loop0 Notice the md5sum of the loop-backed file is different between the two versions.
From: http://luks.endorphin.org/dm-crypt - Fixed hash inconsistancy bug between cryptsetup-luks and original for piped passwords. So, it's either be compatibile with earlier releases of cryptsetup (i.e., FC3), or compatible with what was shipped in FC4. :/
Interesting dilemma. Seeing as the regression is actually in 1.0-1 and not 1.0.1-0, and backward compatibility with FC3 is probably more important, I suppose I'd be inclined to mark this NOTABUG, with apologies to annoyed users like myself who have to rebuild their volumes. :)
Yeah, fixing the regression from FC3 was sort of the reason for the upgrade. It's unfortunate that that wasn't noticed before FC4.