Re: [PATCH] pagecache lock ordering

[Andrew Morton]

Hugh Dickins wrote:
> 
> There's two places, do_buffer_fdatasync

generic_buffer_fdatasync() and hence do_buffer_fdatasync()
are completely unused.  It may be simpler to just trash
them.

> and __find_lock_page_helper,

Yeah.  The code can't deadlock because:

	page_cache_get();
	spin_lock(&pagecache_lock);
	page_cache_release();

we implicitly *know* that page_cache_release won't try
to acquire pagemap_lru_lock, because the page is in the
pagecache and has count=2 or more.  Which is a bit, umm,
subtle.

I get the feeling that a lot of this would be cleaned up
if presence on an LRU contributed to page->count.  It
seems strange, kludgy and probably racy that this is not
the case.

-