lkml.org 
[lkml]   [2000]   [Dec]   [21]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [PATCH] Re: fs corruption with invalidate_buffers()
On Thu, Dec 21, 2000 at 05:01:00PM -0800, Linus Torvalds wrote:
>
>
> On Fri, 22 Dec 2000, Jan Niehusmann wrote:
> >
> > The test I did initially was the following:
> >
> > if(!atomic_read(&bh->b_count) &&
> > (destroy_dirty_buffers || !buffer_dirty(bh))
> > && ! (bh->b_page && bh->b_page->mapping)
> > )
> >
> > That is, I was explicitely checking for a mapped page. It worked well, too.
> > Is this more reasonable?
>
> I'd suggest just doing this instead (warning: cut-and-paste in xterm, so
> white-space damage):

> which just ignores mapped buffers entirely (and doesn't test for
> bh->b_page being non-NULL, because that shouldn't be allowed anyway).

Yes, looks good to me, and passes some tests. Here is a patch that has not
been cut and pasted:

--- linux/fs/buffer.c.orig Thu Dec 21 20:30:03 2000
+++ linux/fs/buffer.c Fri Dec 22 02:11:29 2000
@@ -643,7 +643,12 @@
continue;
for (i = nr_buffers_type[nlist]; i > 0 ; bh = bh_next, i--) {
bh_next = bh->b_next_free;
+
+ /* Another device? */
if (bh->b_dev != dev)
+ continue;
+ /* Part of a mapping? */
+ if (bh->b_page->mapping)
continue;
if (buffer_locked(bh)) {
atomic_inc(&bh->b_count);

I have one additional question: invalidate_buffers normally gets called if
someone wants to make sure that, after the call, read accesses to a device
really go to the device and don't get served by a cache. Is there some
mechanismn that does the same thing to mapped pages?

Jan


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:28    [W:0.084 / U:3.552 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site