TAKE 965636 - xfs_setfilesize panics on null xfs_inode

TAKE 965636 - xfs_setfilesize panics on null xfs_inode

[David Chinner]

Handle null returned from xfs_vtoi() in xfs_setfilesize().


Date:  Mon Jun  4 18:27:21 AEST 2007
Workarea:  chook.melbourne.sgi.com:/build/dgc/isms/2.6.x-xfs
Inspected by:  tes@sgi.com, olaf@sgi.com

The following file(s) were checked into:
  longdrop.melbourne.sgi.com:/isms/linux/2.6.x-xfs-melb


Modid:  xfs-linux-melb:xfs-kern:28777a
fs/xfs/linux-2.6/xfs_aops.c - 1.145 - changed
http://oss.sgi.com/cgi-bin/cvsweb.cgi/xfs-linux/linux-2.6/xfs_aops.c.diff?r1=text&tr1=1.145&r2=text&tr2=1.144&f=h
	- If we get a null xfs_inode in xfs_setfilesize(), we can't update
	  the file size. Don't even bother trying.

Re: TAKE 965636 - xfs_setfilesize panics on null xfs_inode

[Christoph Hellwig]

On Mon, Jun 04, 2007 at 06:28:12PM +1000, David Chinner wrote:
> Handle null returned from xfs_vtoi() in xfs_setfilesize().

This doesn't make any sense at all.  xfs_inodes always live longer than
vnodes.  What's the backtrace of the problem you're seeing?

Re: TAKE 965636 - xfs_setfilesize panics on null xfs_inode

[David Chinner]

On Mon, Jun 04, 2007 at 09:32:22AM +0100, Christoph Hellwig wrote:
> On Mon, Jun 04, 2007 at 06:28:12PM +1000, David Chinner wrote:
> > Handle null returned from xfs_vtoi() in xfs_setfilesize().
> 
> This doesn't make any sense at all.  xfs_inodes always live longer than
> vnodes.  What's the backtrace of the problem you're seeing?

The I/O completion handlers are used by CXFS clients as well.
Hence we can't assume that the vnode is backed by a xfs_inode....

Cheers,

Dave.
-- 
Dave Chinner
Principal Engineer
SGI Australian Software Group

Re: TAKE 965636 - xfs_setfilesize panics on null xfs_inode

[Christoph Hellwig]

On Mon, Jun 04, 2007 at 06:46:08PM +1000, David Chinner wrote:
> On Mon, Jun 04, 2007 at 09:32:22AM +0100, Christoph Hellwig wrote:
> > On Mon, Jun 04, 2007 at 06:28:12PM +1000, David Chinner wrote:
> > > Handle null returned from xfs_vtoi() in xfs_setfilesize().
> > 
> > This doesn't make any sense at all.  xfs_inodes always live longer than
> > vnodes.  What's the backtrace of the problem you're seeing?
> 
> The I/O completion handlers are used by CXFS clients as well.
> Hence we can't assume that the vnode is backed by a xfs_inode....

Please switch them over to something different instead of crapping up
the code like this