bug related to do with ioctl(XFS_IOC_RESVSP64)

[Daniel Moore <dxm@xxxxxxxxxxxxxxxxxxxxxxxx>]

on an XFS FS, run "mkfile 1M test" and duck.

so far I've seen:
(running in MIPS mode with pagebuf meta data on)
   a tripped ASSERT in xfs_bmapi

or
(running in native mode (ie it's not an endian issue) 
with pagebuf meta data off)     
   a spew of messages from page_buf saying:
 
May 16 14:59:44 bruce kernel: pbfwa: HOLE ro 0x0 size 0x40000
May 16 14:59:44 bruce kernel: PBFWA: short write written 0 ilen 262144 status 0
May 16 14:59:44 bruce kernel: PBGFWA: short write written 0 ilen 262144 status 
0
May 16 14:59:44 bruce kernel: pbfwa: HOLE ro 0x0 size 0x40000
May 16 14:59:44 bruce kernel: PBFWA: short write written e
May 16 14:59:44 bruce kernel: PBF<4>PBGFWA: short write written 0 ilen 262144 
status 0
May 16 14:59:44 bruce kernel: pbfwa: HOLE ro 0x0 size 0x40000
May 16 14:59:44 bruce kernel: PBFWA: short write written 0 ilen 262144 status 0
May 16 14:59:44 bruce kernel: PBGFWA: short write written 0 ilen 262144 status 
0

mkfile does an ioctl(XFS_IOC_RESVSP64) to reserve space but if the
reserved space is used, things go haywire.

-----------------------------------------------------
 Daniel Moore                  dxm@xxxxxxx
 R&D Software Engineer         Phone: +61-3-98348209
 SGI Performance Tools Group   Fax:   +61-3-98132378
-----------------------------------------------------