Matthew Wilcox [off-list ref] wrote:

That also brings up that there is no set_page_private_2().  I think
that's OK -- you only set PageFsCache() immediately after reading the
page from the server.  But I feel this "unlock_page_private_2" is actually
"clear_page_private_2" -- ie it's equivalent to writeback, not to lock.

How about I do the following:

 (1) Add set_page_private_2() or mark_page_private_2() to set the PG_fscache_2
     bit.  It could take a ref on the page here.

 (2) Rename unlock_page_private_2() to end_page_private_2().  It could drop
     the ref on the page here, but that then means I can't use
     pagevec_release().

 (3) Add wait_on_page_private_2() an analogue of wait_on_page_writeback()
     rather than wait_on_page_locked().

 (4) Provide fscache synonyms of the above.

David