We found strange blocks layout in our mail server, after careful study, we
got the reason and tried to fix it.

On the very fist attempt to allocate a block to the newly-initialised inode,
if we are trying to add a block at logical file offset "1" then
ext2_find_goal() will incorrectly assume that this was a next_alloc_block
cache hit (because we think the previously-allocated block was at offset
zero).

Net result: why trying to extend a freshly-opened one-block file we end up
deciding to place the second file block at disk block "1", rather than going
off and calling ext2_find_near().

Fix it by checking that we actually do have something valid cached in
next_alloc_goal.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>