Re: [PATCH 00/22] HWPOISON: Intro (v5)

[Nick Piggin]

On Mon, Jun 15, 2009 at 06:09:54PM +0800, Wu Fengguang wrote:
> On Mon, Jun 15, 2009 at 04:14:53PM +0800, Nick Piggin wrote:
> > On Mon, Jun 15, 2009 at 08:44:47AM +0200, Nick Piggin wrote:
> > > Did we verify with filesystem maintainers (eg. btrfs) that the
> > > !ISREG test will be enough to prevent oopses?
> > 
> > BTW. this is quite a significant change I think and not
> > really documented well enough. Previously a filesystem
> > will know exactly when and why pagecache in a mapping
> > under its control will be truncated (as opposed to
> > invalidated).
> > 
> > They even have opportunity to hold locks such as i_mutex.
> > 
> > And depending on what they do, they could do interesting
> > things even with ISREG files.
> > 
> > So, I really think this needs review by filesystem
> > maintainers and it would be far safer to use invalidate
> > until it is known to be safe.
> 
> Nick, we are doing invalidate_complete_page() for !S_ISREG inodes now.
> Do you mean to do invalidate_complete_page() for all inodes for now?

That would make me a lot happier. It is obviously correct because
that is basically what page reclaim and inode reclaim and drop
caches etc does.

Note that I still don't like exporting invalidate_complete_page
fro the same reasons I don't like exporting truncate_complete_page,
so I will ask if you can do an invalidate_inode_page function
along the same lines of the truncate_inode_page one please.


> That's a good suggestion, it shall be able to do the job for most
> pages indeed.

Yes I think it will be far far safer while only introducing
another small class of pages which cannot be recovered (probably
a much smaller set most of the time than the size of the existing
set of pages which cannot be recovered).


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