Re: [PATCH] Scalable page cache

[Linus Torvalds]

On Mon, 26 Nov 2001, Andrew Morton wrote:
>
> We've called block_prepare_write(), which has done the kmap.
> But even though block_prepare_write() returned success, this
> call to the filesystem's ->prepare_write() is about to fail.

That's _way_ too intimate knowledge of how block_prepare_write() works (or
doesn't work).

How about sending me a patch that removes all the kmap/kunmap crap from
_both_ ext3 and block_prepare/commit_write.

> There have been a number of mistakes made over this particular kmap()
> operation.  NFS client had it wrong for a while. I think sct had
> some proposal for making it more robust.

It _is_ more robust. You are only battling remnants from an older age when
it wasn't.

		Linus