Re: page_launder() bug

[Marcelo Tosatti]

On Mon, 7 May 2001, Linus Torvalds wrote:

> In fact, it might even clean stuff up. Who knows? At least
> page_launder() would not need to know about magic dead swap pages, because
> the decision would be entirely in writepage().
> 
> And there aren't that many writepage() implementations in the kernel to
> fix up (and the fixup tends to be two simple added lines of code for most
> of them - just the "if (!priority) return").
> 
> Also note how some filesystems might use the writepage() callback even
> with a zero priority as a hint that things are approaching the point where
> we need to start flushing, which might make a difference for deciding when
> to try to write a log entry, for example.

Moreover, the filesystem may want to return "-1" even if "priority" is
non-zero --- think about delayed allocations. (the XFS guys were just
complaining about page_launder() not checking the return value of
writepage() so they could not do this kind of thing with delayed
allocations sometime ago).

> Now, I'm not saying this is _the_ solution to it, but I don't see any
> really clean alternatives.

I like it --- it pushes down control to the pagers so they can be smarter.

Will send a patch later.