Re: [patch] fs: improved handling of page and buffer IO errors

[Jamie Lokier <jamie@shareable.org>]

Miklos Szeredi wrote:
> On Tue, 21 Oct 2008, steve@chygwyn.co wrote:
> > Well I'm not sure why we'd need to distinguish between "page has not
> > been read" and "page has been read but no longer valid". I guess I
> > don't understand why those two cases are not the same from the vfs
> > and filesystem points of view.
> 
> In the first case the page contains random bytes, in the second case
> it contains actual file data, which has become stale, but at some
> point in time it _was_ the contents of the file.
> 
> This is a very important distinction for splice(2) for example.
> Splice does not actually copy data into the pipe buffer, only
> references the pages.  And it can reference pages which are not yet
> up-to-date.  So when the buffers are consumed from the pipe, the
> splice code needs to know if the page contains random junk (never
> brought up-to-date) or data that is, or once was, valid.

So GFS goes to great lengths to ensure that read/write are coherent,
so are mmaps (writable or not), but _splice_ is not coherent in the
sense that it can send invalid but non-random data? :-)

Also, is there still a problem where the data is "valid" but part of
the page may have been zero'd by truncate, which is then transmitted
by splice?

-- Jamie