Messages in this thread | | | Date | Fri, 22 Oct 2004 19:04:03 +0200 | From | Andrea Arcangeli <> | Subject | Re: [PATCH] zap_pte_range should not mark non-uptodate pages dirty |
| |
On Fri, Oct 22, 2004 at 06:17:44PM +0200, Andrea Arcangeli wrote: > So let's forget the mmapped case and let's fix the bh screwup in 2.6.
I propose a solution to fix the coherency problem that afflicts 2.6 if invalidate_complete_page fails. If the page is still PagePrivate despite we called try_to_release_page (which means clearing the uptodate bitflag wouldn't help), we should simply return an error to the O_DIRECT writes. That way the error would not be overlooked by the database writing. This is a minium guarantee we have to provide: if we fail invalidate cause the O_DIRECT write to fail.
Of course we should return write failures as well in the mmapped case, but let's ignore the mmapped case for now.
The real showstopper bug is try_to_release_page failing and preventing the coherency protocol to work even without mmaps. This could never happen in 2.4, in 2.4 as worse ext3 would get an hearth attack, which is much better than silent corruption in the backup. - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |