[XFS] Fix speculative allocation beyond eof
authorLachlan McIlroy <lachlan@redback.melbourne.sgi.com>
Mon, 22 Dec 2008 06:56:49 +0000 (17:56 +1100)
committerLachlan McIlroy <lachlan@redback.melbourne.sgi.com>
Mon, 22 Dec 2008 06:56:49 +0000 (17:56 +1100)
commit9f6c92b9cc2fd41d6c7b493be5637cc5b5659880
tree64ea95c3305d2bba4191f7dec831e623bff8a911
parent4fdc7781799926dca6c3a3bb6e9533a9718c4dea
[XFS] Fix speculative allocation beyond eof

Speculative allocation beyond eof doesn't work properly.  It was
broken some time ago after a code cleanup that moved what is now
xfs_iomap_eof_align_last_fsb() and xfs_iomap_eof_want_preallocate()
out of xfs_iomap_write_delay() into separate functions.  The code
used to use the current file size in various checks but got changed
to be max(file_size, i_new_size).  Since i_new_size is the result
of 'offset + count' then in xfs_iomap_eof_want_preallocate() the
check for '(offset + count) <= isize' will always be true.

ie if 'offset + count' is > ip->i_size then isize will be i_new_size
and equal to 'offset + count'.

This change fixes all the places that used to use the current file
size.

Reviewed-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
fs/xfs/xfs_iomap.c