Re: invalidate_inode_pages in 2.5.32/3

[Andrew Morton]

Trond Myklebust wrote:
> 
> >>>>> " " == Andrew Morton <[e-mail]> writes:
> 
>      > It's a bit worrisome if NFS is dependent upon successful
>      > pagecache takedown in invalidate_inode_pages.
> 
> Why? We don't use all that buffer crap, so in principle
> invalidate_inode_page() is only supposed to fail for us if
> 
>   - page is locked (i.e. new read in progress or something like that)
>   - page is refcounted (by something like mmap()).

    - someone took a temporary ref on the page.

    Possibly it is the deferred LRU addition code.  Try running
    lru_add_drain() before invalidate_inode_pages().
 
> neither of which should be the case here.

Probably, it worked OK with the global locking because nobody was taking
a temp ref against those pages.

Please tell me exactly what semantics NFS needs in there.  Does
truncate_inode_pages() do the wrong thing?