NFS infinite loop in filemap_fault()

NFS infinite loop in filemap_fault()

[Miklos Szeredi]

Page fault on NFS apparently goes into an infinite loop if the read on
the server fails.

I don't understand the NFS readpage code, but the filemap_fault() code
looks somewhat suspicious:

	/*
	 * Umm, take care of errors if the page isn't up-to-date.
	 * Try to re-read it _once_. We do this synchronously,
	 * because there really aren't any performance issues here
	 * and we need to check for errors.
	 */
	ClearPageError(page);
	error = mapping->a_ops->readpage(file, page);
	page_cache_release(page);

	if (!error || error == AOP_TRUNCATED_PAGE)
		goto retry_find;

The comment doesn't seem to match what the it actually does: if
->readpage() is asynchronous, then this will just repeat everything,
without any guarantee that it will re-read once.

What am I missing?

Miklos

Re: NFS infinite loop in filemap_fault()

[Miklos Szeredi]

> Page fault on NFS apparently goes into an infinite loop if the read on
> the server fails.
> 
> I don't understand the NFS readpage code, but the filemap_fault() code
> looks somewhat suspicious:
> 
> 	/*
> 	 * Umm, take care of errors if the page isn't up-to-date.
> 	 * Try to re-read it _once_. We do this synchronously,
> 	 * because there really aren't any performance issues here
> 	 * and we need to check for errors.
> 	 */
> 	ClearPageError(page);
> 	error = mapping->a_ops->readpage(file, page);
> 	page_cache_release(page);
> 
> 	if (!error || error == AOP_TRUNCATED_PAGE)
> 		goto retry_find;
> 
> The comment doesn't seem to match what the it actually does: if
> ->readpage() is asynchronous, then this will just repeat everything,
> without any guarantee that it will re-read once.

This patch fixes it.  It's probably wrong in some subtle way though...

Miklos


---
 mm/filemap.c |    6 ++++++
 1 file changed, 6 insertions(+)

Index: linux.git/mm/filemap.c
===================================================================
--- linux.git.orig/mm/filemap.c	2008-05-08 08:17:22.000000000 +0200
+++ linux.git/mm/filemap.c	2008-05-08 08:19:59.000000000 +0200
@@ -1461,6 +1461,12 @@ page_not_uptodate:
 	 */
 	ClearPageError(page);
 	error = mapping->a_ops->readpage(file, page);
+	if (!error && !PageUptodate(page)) {
+		lock_page(page);
+		if (!PageUptodate(page))
+			error = -EIO;
+		unlock_page(page);
+	}
 	page_cache_release(page);
 
 	if (!error || error == AOP_TRUNCATED_PAGE)

Re: NFS infinite loop in filemap_fault()

[Josef Bacik]

On Thu, May 08, 2008 at 08:47:09AM +0200, Miklos Szeredi wrote:
> > Page fault on NFS apparently goes into an infinite loop if the read on
> > the server fails.
> > 
> > I don't understand the NFS readpage code, but the filemap_fault() code
> > looks somewhat suspicious:
> > 
> > 	/*
> > 	 * Umm, take care of errors if the page isn't up-to-date.
> > 	 * Try to re-read it _once_. We do this synchronously,
> > 	 * because there really aren't any performance issues here
> > 	 * and we need to check for errors.
> > 	 */
> > 	ClearPageError(page);
> > 	error = mapping->a_ops->readpage(file, page);
> > 	page_cache_release(page);
> > 
> > 	if (!error || error == AOP_TRUNCATED_PAGE)
> > 		goto retry_find;
> > 
> > The comment doesn't seem to match what the it actually does: if
> > ->readpage() is asynchronous, then this will just repeat everything,
> > without any guarantee that it will re-read once.
> 
> This patch fixes it.  It's probably wrong in some subtle way though...
> 
> Miklos
> 
> 
> ---
>  mm/filemap.c |    6 ++++++
>  1 file changed, 6 insertions(+)
> 
> Index: linux.git/mm/filemap.c
> ===================================================================
> --- linux.git.orig/mm/filemap.c	2008-05-08 08:17:22.000000000 +0200
> +++ linux.git/mm/filemap.c	2008-05-08 08:19:59.000000000 +0200
> @@ -1461,6 +1461,12 @@ page_not_uptodate:
>  	 */
>  	ClearPageError(page);
>  	error = mapping->a_ops->readpage(file, page);
> +	if (!error && !PageUptodate(page)) {

Shouldn't you have (!error || error != AOP_TRUNCATED_PAGE), since the fs can
return AOP_TRUNCATED_PAGE if it needs vfs to try the readpage again?  Things
like OCFS2/GFS2 do this, they send of a lock request and return
AOP_TRUNCATED_PAGE so that when we come back into readpage we are already
holding our lock without blocking while holding the page lock.  Thanks,

Josef

Re: NFS infinite loop in filemap_fault()

[Miklos Szeredi]

> > Index: linux.git/mm/filemap.c
> > ===================================================================
> > --- linux.git.orig/mm/filemap.c	2008-05-08 08:17:22.000000000 +0200
> > +++ linux.git/mm/filemap.c	2008-05-08 08:19:59.000000000 +0200
> > @@ -1461,6 +1461,12 @@ page_not_uptodate:
> >  	 */
> >  	ClearPageError(page);
> >  	error = mapping->a_ops->readpage(file, page);
> > +	if (!error && !PageUptodate(page)) {
> 
> Shouldn't you have (!error || error != AOP_TRUNCATED_PAGE),

That would be a rather useless condition (hint: '!error' means 'error == 0')

> since the fs can
> return AOP_TRUNCATED_PAGE if it needs vfs to try the readpage again?

Yep, I din't touch the 'error != 0' codepath.

Miklos

Re: NFS infinite loop in filemap_fault()

[Josef Bacik]

On Thu, May 08, 2008 at 08:42:03PM +0200, Miklos Szeredi wrote:
> > > Index: linux.git/mm/filemap.c
> > > ===================================================================
> > > --- linux.git.orig/mm/filemap.c	2008-05-08 08:17:22.000000000 +0200
> > > +++ linux.git/mm/filemap.c	2008-05-08 08:19:59.000000000 +0200
> > > @@ -1461,6 +1461,12 @@ page_not_uptodate:
> > >  	 */
> > >  	ClearPageError(page);
> > >  	error = mapping->a_ops->readpage(file, page);
> > > +	if (!error && !PageUptodate(page)) {
> > 
> > Shouldn't you have (!error || error != AOP_TRUNCATED_PAGE),
> 
> That would be a rather useless condition (hint: '!error' means 'error == 0')
> 
> > since the fs can
> > return AOP_TRUNCATED_PAGE if it needs vfs to try the readpage again?
> 
> Yep, I din't touch the 'error != 0' codepath.
>

Doh sorry, I hadn't had my coffee yet.

Josef