Re: writethrough of mmapped pages

[Andrew Morton <akpm@osdl.org>]

Steve French <smfltc@us.ibm.com> wrote:
>
> Is there an easy way to locally force the page cache to do in effect
> writethrough of mmapped pages (when I lose my write caching token in
> cifs vfs e.g.)?  Standard writes can be hooked easily enough by wrapping
> the inode operation for write and only calling generic_write if they are
> (still) cacheable locally which is what I do in current code, but
> mmapped writes still go through the page cache presumably and thus would
> have to be synced if a non-mmmapped read followed an mmapped write of an
> overlapping range.

You mean that you want the filesytem to know whenever an application
attempts to modify a mmapped page?  After the modification, rather than
before?

Not sure how to do that - even if you modify the MM to take a trap on every
write the page still hasn't been modified on entry to the fault handler.