Nick Piggin wrote:

> Anyway, clearly a plug friendly bugfix needs to be implemented
> regardless.

I actually think it would be cleaner to introduce a new
mapping_lock_page(mapping, page) or something that does the
unplug on the _mapping_, and have lock_page be callable with
no ref on the inode. Should be a mechanical conversion for
most of the safe callers.

But for 2.6.17, how's this?

-- 
SUSE Labs, Novell Inc.