From: Linus Nielsen Feltzing <linus_at_haxx.se>
Date: Fri, 08 Nov 2002 16:08:06 +0100

Greg Haerr wrote:
> Bjorn/Linus,
> I've got a buffer-cached version of the fat32 disk working,
> caching for the time being only data sectors (currently it's just
> modified file.c, not fat.c for testing) I hadn't quite realized what a great
> job Linus did optimizing the routines for extremely fast
> sequential reading and writing, and my original design didn't
> handle that...

It was Björn, actually.

> Anyways, another question: currently, the creat() call
> does an open(...,O_WRONLY), lseeks are disallowed,
> and the fat chain is truncated at close time past the
> last cluster written. Do we want to make a distinction for
> the new write code so that creat() and open(...,O_WRONLY)
> act differently, that is, open(...,O_WRONLY) allows
> seeking and appending to end-of-file? This would allow logfiles,
> etc to be written, which can't be handled now.

We (Björn) are working to get that functionality, with the O_RDWR flag.

/Linus
Received on 2002-11-08