Package: xfsprogs Version: 3.1.2-1 Severity: normal
In debian, xfsprogs are built against internal blkid implementation. Upstream recommends using libblkid. One bad side of internal blkid code is that as a result of using it in mkfs.xfs, mkfs loses -s option (it is being silently ignored). -s stands for sector size. One particular case where sector size is important is when doing direct I/O - all operations should be multiple of this sector size and aligned as such. But old blkid always chooses 4096 (4K) as sector size for MD RAID[456] arrays, despite actual size of underlying devices, and it also ignores now exported /sys/block/$device/queue/hw_sector_size attribute of other block devices. As a result, one can't use applications that performs direct-I/O in multiplies of usual 512bytes (such as Oracle10 for example), and it is not possible to use modern large devices with larger sector size properly. XFS amied to be a filesystem for "large" devices, where this is more important than on current consumer disk drives. Recompiling xfsprogs with libblkid from current util-linux-ng immediately fixes this issue. (Actually I had to find this information and recompile xfsprogs locally just to create one XFS filesystem, because mkfs.xfs in Debain didn't work correctly due to this bug). Thanks! -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected]
