swalloc/allocsize question

[Dave Chinner]

On Mon, Aug 03, 2015 at 01:27:03PM -0400, Joe Landman wrote:
> Hi folks:
> 
>   I've noticed some "odd" behavior while experimenting with xfs
> mounts on relatively recent kernels (3.18 series).  Basically, the
> swalloc,allocsize=x options are being modified to something
> different.  That is:
> 
> root at usn-04:~/burn-in/fio# uname -r
> 3.18.12.scalable
> 
> root at usn-04:~/burn-in/fio# mount -o
> inode64,noatime,nodiratime,swalloc,allocsize=1280k /dev/sda /data/1
> 
> root at usn-04:~/burn-in/fio# mount | grep sda
> /dev/sda on /data/1 type xfs
> (rw,noatime,nodiratime,swalloc,attr2,inode64,allocsize=256k,noquota)
> 
> This is a hardware RAID6 with a 128k chunk size and 12 elements (so
> 10 data drives, thus the 1280k allocsize for stripe width
> allocation).  Is this something specific to xfs itself, or is this
> an issue in the mount tools ?

Please provide:

	- xfs_info output from the filesystem in question
	- dmesg output of all the mount messages

Dave.
-- 
Dave Chinner
david at fromorbit.com