Re: [PATCH] [13/16] HWPOISON: The high level memory error handler in the VM v5

Re: [PATCH] [13/16] HWPOISON: The high level memory error handler in the VM v5

[Nick Piggin]

On Wed, Jun 03, 2009 at 08:46:47PM +0200, Andi Kleen wrote:
> +static int me_pagecache_clean(struct page *p, unsigned long pfn)
> +{
> +	struct address_space *mapping;
> +
> +	if (!isolate_lru_page(p))
> +		page_cache_release(p);
> +
> +	/*
> +	 * Now truncate the page in the page cache. This is really
> +	 * more like a "temporary hole punch"
> +	 * Don't do this for block devices when someone else
> +	 * has a reference, because it could be file system metadata
> +	 * and that's not safe to truncate.
> +	 */
> +	mapping = page_mapping(p);
> +	if (mapping && S_ISBLK(mapping->host->i_mode) && page_count(p) > 1) {
> +		printk(KERN_ERR
> +			"MCE %#lx: page looks like a unsupported file system metadata page\n",
> +			pfn);
> +		return FAILED;
> +	}

page_count check is racy. Hmm, S_ISBLK should handle xfs's private mapping.
AFAIK btrfs has a similar private mapping but a quick grep does not show
up S_IFBLK anywhere, so I don't know what the situation is there.

Unfortunately though, the linear mapping is not the only metadata mapping
a filesystem might have. Many work on directories in seperate mappings
(ext2, for example, which is where I first looked and will still oops with
your check).

Also, others may have other interesting inodes they use for metadata. Do
any of them go through the pagecache? I dont know. The ext3 journal,
for example? How does that work?

Unfortunately I don't know a good way to detect regular data mappings
easily. Ccing linux-fsdevel. Until that is worked out, you'd need to
use the safe pagecache invalidate rather than unsafe truncate.


> +	if (mapping) {
> +		truncate_inode_page(mapping, p);
> +		if (page_has_private(p) && !try_to_release_page(p, GFP_NOIO)) {
> +			pr_debug(KERN_ERR "MCE %#lx: failed to release buffers\n",
> +				pfn);
> +			return FAILED;
> +		}
> +	}
> +	return RECOVERED;
> +}

Re: [PATCH] [13/16] HWPOISON: The high level memory error handler in the VM v5

[Nick Piggin]

On Tue, Jun 09, 2009 at 11:51:55AM +0200, Nick Piggin wrote:
> On Wed, Jun 03, 2009 at 08:46:47PM +0200, Andi Kleen wrote:
> > +static int me_pagecache_clean(struct page *p, unsigned long pfn)
> > +{
> > +	struct address_space *mapping;
> > +
> > +	if (!isolate_lru_page(p))
> > +		page_cache_release(p);
> > +
> > +	/*
> > +	 * Now truncate the page in the page cache. This is really
> > +	 * more like a "temporary hole punch"
> > +	 * Don't do this for block devices when someone else
> > +	 * has a reference, because it could be file system metadata
> > +	 * and that's not safe to truncate.
> > +	 */
> > +	mapping = page_mapping(p);
> > +	if (mapping && S_ISBLK(mapping->host->i_mode) && page_count(p) > 1) {
> > +		printk(KERN_ERR
> > +			"MCE %#lx: page looks like a unsupported file system metadata page\n",
> > +			pfn);
> > +		return FAILED;
> > +	}
> 
> page_count check is racy. Hmm, S_ISBLK should handle xfs's private mapping.
> AFAIK btrfs has a similar private mapping but a quick grep does not show
> up S_IFBLK anywhere, so I don't know what the situation is there.
> 
> Unfortunately though, the linear mapping is not the only metadata mapping
> a filesystem might have. Many work on directories in seperate mappings
> (ext2, for example, which is where I first looked and will still oops with
> your check).
> 
> Also, others may have other interesting inodes they use for metadata. Do
> any of them go through the pagecache? I dont know. The ext3 journal,
> for example? How does that work?
> 
> Unfortunately I don't know a good way to detect regular data mappings
> easily. Ccing linux-fsdevel. Until that is worked out, you'd need to
> use the safe pagecache invalidate rather than unsafe truncate.

Maybe just testing S_ISREG would be better. Definitely safer than
ISBLK.

Note that for !ISREG files, then you can still attempt the
non-destructive invalidate (after extracting a suitable function
similarly to the truncate one). Most likely the fs is not using
the page right now, so it should give bit more coverage.

I still don't exactly know about, say, ext3 journal. Probably
it doesn't use pagecache anyway. Do any other filesystems do
crazy things with S_ISREG files? They probably deserve to oops
if they do ;)