Regression seen when HIGHMEM enabled with NFS on 3.1rc4 kernel

Regression seen when HIGHMEM enabled with NFS on 3.1rc4 kernel

[R, Sricharan]

Sorry resending again. My mailer settings thrashed my earlier email.

Hi,
  A kernel crash is observed on 3.1rc4 kernel when HIGHMEM is enabled and
  kernel is booted with a NFS on omap4430sdp. The issue happens in the below
  scenario.

       In file net/sunrpc/xprtsock.c,
         static int xs_send_pagedata( xxx, struct xdr_buf *xdr, ..)
         {
                Struct page **ppage;
                ....
                .....
                ppage = xdr->pages + (base >> PAGE_SHIFT);
                ....
            err = sock->ops->sendpage(sock, *ppage, base, len, flags);

            ...
         }

         1) In the above piece of code, the *ppage value from ops->sendpage
            function is finally passed on to Kmap by the lower level code to
            get the virtual address of the page.
         2) In some corner cases the value of *ppage pointer is NULL.
         3) When highmem is enabled and a NULL pointer is passed to
            Kmap, then kmap finally crashes. But in the case when highmem
            is disabled, then kmap returns a junk value for NULL pointer.

            Highmem Enabled , kmap( NULL )-----> kernel crashes.

           Highmem disabled, kmap( NULL )-----> junk value is returned.
               Subsequently this message is observed on
                                                the console.

                                "RPC call returned error 14"

         4) Now the question is why is the value of *ppage = NULL is passed
            from the above piece of code to lower layers.
           Should that not have handled *ppage = NULL? and kmap should not
            have received a NULL pointer?

Thanks,
  Sricharan
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Regression seen when HIGHMEM enabled with NFS on 3.1rc4 kernel

[Trond Myklebust]

On Fri, 2011-09-09 at 18:40 +0530, R, Sricharan wrote: 
> Sorry resending again. My mailer settings thrashed my earlier email.
> 
> Hi,
>   A kernel crash is observed on 3.1rc4 kernel when HIGHMEM is enabled and
>   kernel is booted with a NFS on omap4430sdp. The issue happens in the below
>   scenario.
> 
>        In file net/sunrpc/xprtsock.c,
>          static int xs_send_pagedata( xxx, struct xdr_buf *xdr, ..)
>          {
>                 Struct page **ppage;
>                 ....
>                 .....
>                 ppage = xdr->pages + (base >> PAGE_SHIFT);
>                 ....
>             err = sock->ops->sendpage(sock, *ppage, base, len, flags);
> 
>             ...
>          }
> 
>          1) In the above piece of code, the *ppage value from ops->sendpage
>             function is finally passed on to Kmap by the lower level code to
>             get the virtual address of the page.
>          2) In some corner cases the value of *ppage pointer is NULL.
>          3) When highmem is enabled and a NULL pointer is passed to
>             Kmap, then kmap finally crashes. But in the case when highmem
>             is disabled, then kmap returns a junk value for NULL pointer.
> 
>             Highmem Enabled , kmap( NULL )-----> kernel crashes.
> 
>            Highmem disabled, kmap( NULL )-----> junk value is returned.
>                Subsequently this message is observed on
>                                                 the console.
> 
>                                 "RPC call returned error 14"
> 
>          4) Now the question is why is the value of *ppage = NULL is passed
>             from the above piece of code to lower layers.
>            Should that not have handled *ppage = NULL? and kmap should not
>             have received a NULL pointer?

I wouldn't expect *ppage to be NULL under any circumstances, so I'm
really curious as to what is happening here.

Could you perhaps add a printk() to that section of code to print out
the values of 'xdr->page_base', 'xdr->page_len', 'len' and 'remainder'
in the case where *ppage == NULL?

Cheers
  Trond

-- 
Trond Myklebust
Linux NFS client maintainer

NetApp
Trond.Myklebust-HgOvQuBEEgTQT0dZR+AlfA@public.gmane.org
www.netapp.com

--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE: Regression seen when HIGHMEM enabled with NFS on 3.1rc4 kernel

[Sricharan R]

Hi Trond,
[....]

>>          1) In the above piece of code, the *ppage value from ops-
>>sendpage
>>             function is finally passed on to Kmap by the lower level
code
>to
>>             get the virtual address of the page.
>>          2) In some corner cases the value of *ppage pointer is NULL.
>>          3) When highmem is enabled and a NULL pointer is passed to
>>             Kmap, then kmap finally crashes. But in the case when
highmem
>>             is disabled, then kmap returns a junk value for NULL
pointer.
>>
>>             Highmem Enabled , kmap( NULL )-----> kernel crashes.
>>
>>            Highmem disabled, kmap( NULL )-----> junk value is returned.
>>                Subsequently this message is observed on
>>                                                 the console.
>>
>>                                 "RPC call returned error 14"
>>
>>          4) Now the question is why is the value of *ppage = NULL is
>passed
>>             from the above piece of code to lower layers.
>>            Should that not have handled *ppage = NULL? and kmap should
>not
>>             have received a NULL pointer?
>
>I wouldn't expect *ppage to be NULL under any circumstances, so I'm
>really curious as to what is happening here.
>
>Could you perhaps add a printk() to that section of code to print out
>the values of 'xdr->page_base', 'xdr->page_len', 'len' and 'remainder'
>in the case where *ppage == NULL?
>


Thanks for the response.
I added a printk just before err = sock->ops->sendpage(sock, *ppage, base,
len, flags);
So here are values when *ppage is NULL.

xdr->page_base= 0xCE9  xdr->page_len=0x400 len=0xE9 remainder=0x0.

Thanks,
 Sricharan
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE: Regression seen when HIGHMEM enabled with NFS on 3.1rc4 kernel

[Trond Myklebust]

On Mon, 2011-09-12 at 11:46 +0530, Sricharan R wrote: 
> Hi Trond,
> [....]
> 
> >>          1) In the above piece of code, the *ppage value from ops-
> >>sendpage
> >>             function is finally passed on to Kmap by the lower level
> code
> >to
> >>             get the virtual address of the page.
> >>          2) In some corner cases the value of *ppage pointer is NULL.
> >>          3) When highmem is enabled and a NULL pointer is passed to
> >>             Kmap, then kmap finally crashes. But in the case when
> highmem
> >>             is disabled, then kmap returns a junk value for NULL
> pointer.
> >>
> >>             Highmem Enabled , kmap( NULL )-----> kernel crashes.
> >>
> >>            Highmem disabled, kmap( NULL )-----> junk value is returned.
> >>                Subsequently this message is observed on
> >>                                                 the console.
> >>
> >>                                 "RPC call returned error 14"
> >>
> >>          4) Now the question is why is the value of *ppage = NULL is
> >passed
> >>             from the above piece of code to lower layers.
> >>            Should that not have handled *ppage = NULL? and kmap should
> >not
> >>             have received a NULL pointer?
> >
> >I wouldn't expect *ppage to be NULL under any circumstances, so I'm
> >really curious as to what is happening here.
> >
> >Could you perhaps add a printk() to that section of code to print out
> >the values of 'xdr->page_base', 'xdr->page_len', 'len' and 'remainder'
> >in the case where *ppage == NULL?
> >
> 
> 
> Thanks for the response.
> I added a printk just before err = sock->ops->sendpage(sock, *ppage, base,
> len, flags);
> So here are values when *ppage is NULL.
> 
> xdr->page_base= 0xCE9  xdr->page_len=0x400 len=0xE9 remainder=0x0.
> 
> Thanks,
>  Sricharan

Can you please tell me what the mount options are for this setup?

Are you running any applications that might be using O_DIRECT writes?

Cheers
  Trond

-- 
Trond Myklebust
Linux NFS client maintainer

NetApp
Trond.Myklebust-HgOvQuBEEgTQT0dZR+AlfA@public.gmane.org
www.netapp.com

--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE: Regression seen when HIGHMEM enabled with NFS on 3.1rc4 kernel

[Trond Myklebust]

On Mon, 2011-09-12 at 10:41 -0400, Trond Myklebust wrote: 
> On Mon, 2011-09-12 at 11:46 +0530, Sricharan R wrote: 
> > Thanks for the response.
> > I added a printk just before err = sock->ops->sendpage(sock, *ppage, base,
> > len, flags);
> > So here are values when *ppage is NULL.
> > 
> > xdr->page_base= 0xCE9  xdr->page_len=0x400 len=0xE9 remainder=0x0.
> > 
> > Thanks,
> >  Sricharan
> 
> Can you please tell me what the mount options are for this setup?

I'm guessing you've got wsize=1024, in which case, can you please try
the following patch?

Cheers
  Trond
8<-------------------------------------------------------------------------- 
>From 7b4a9c76b55dd254431902552528137a2ea5e55d Mon Sep 17 00:00:00 2001
From: Trond Myklebust <Trond.Myklebust@netapp.com>
Date: Mon, 12 Sep 2011 11:47:53 -0400
Subject: [PATCH] NFS: Fix a typo in nfs_flush_multi

Fix a typo which causes an Oops in the RPC layer, when using wsize < 4k.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/write.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/fs/nfs/write.c b/fs/nfs/write.c
index b39b37f..c9bd2a6 100644
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -958,7 +958,7 @@ static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, struct list_head
 		if (!data)
 			goto out_bad;
 		data->pagevec[0] = page;
-		nfs_write_rpcsetup(req, data, wsize, offset, desc->pg_ioflags);
+		nfs_write_rpcsetup(req, data, len, offset, desc->pg_ioflags);
 		list_add(&data->list, res);
 		requests++;
 		nbytes -= len;
-- 
1.7.6



-- 
Trond Myklebust
Linux NFS client maintainer

NetApp
Trond.Myklebust@netapp.com
www.netapp.com

RE: Regression seen when HIGHMEM enabled with NFS on 3.1rc4 kernel

[Sricharan R]

[..]
>>
>> Can you please tell me what the mount options are for this setup?
>
>I'm guessing you've got wsize=1024, in which case, can you please try
>the following patch?
>
The mount options for nfs is rw.

Yes, in my setup wsize=1024 when the issue happened.
I tried your patch and I was not able to see the issue after that,
where as in the other case the issue happened quite frequently.
So I think that the patch fixes the issue.
Thanks a lot for your help.

>Cheers
>  Trond
>8<-----------------------------------------------------------------------
--
>-
>From 7b4a9c76b55dd254431902552528137a2ea5e55d Mon Sep 17 00:00:00 2001
>From: Trond Myklebust <Trond.Myklebust-HgOvQuBEEgTQT0dZR+AlfA@public.gmane.org>
>Date: Mon, 12 Sep 2011 11:47:53 -0400
>Subject: [PATCH] NFS: Fix a typo in nfs_flush_multi
>
>Fix a typo which causes an Oops in the RPC layer, when using wsize < 4k.
>
>Signed-off-by: Trond Myklebust <Trond.Myklebust-HgOvQuBEEgTQT0dZR+AlfA@public.gmane.org>
>---
> fs/nfs/write.c |    2 +-
> 1 files changed, 1 insertions(+), 1 deletions(-)
>
>diff --git a/fs/nfs/write.c b/fs/nfs/write.c
>index b39b37f..c9bd2a6 100644
>--- a/fs/nfs/write.c
>+++ b/fs/nfs/write.c
>@@ -958,7 +958,7 @@ static int nfs_flush_multi(struct
nfs_pageio_descriptor
>*desc, struct list_head
> 		if (!data)
> 			goto out_bad;
> 		data->pagevec[0] = page;
>-		nfs_write_rpcsetup(req, data, wsize, offset,
desc->pg_ioflags);
>+		nfs_write_rpcsetup(req, data, len, offset,
desc->pg_ioflags);
> 		list_add(&data->list, res);
> 		requests++;
> 		nbytes -= len;
>--
>1.7.6
>
>
>
>--
>Trond Myklebust
>Linux NFS client maintainer
>
>NetApp
>Trond.Myklebust-HgOvQuBEEgTQT0dZR+AlfA@public.gmane.org
>www.netapp.com
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Regression seen when HIGHMEM enabled with NFS on 3.1rc4 kernel

[Sricharan R]

Hi,
  A kernel crash is observed on 3.1rc4 kernel when HIGHMEM is enabled and
  kernel is booted with a NFS on omap4430sdp. The issue happens in the
below
  scenario.

       In file net/sunrpc/xprtsock.c,
         static int xs_send_pagedata( xxx, struct xdr_buf *xdr, ..)
         {
		Struct page **ppage;
		....
		.....
		ppage = xdr->pages + (base >> PAGE_SHIFT);
		....
            err = sock->ops->sendpage(sock, *ppage, base, len, flags);

            ...
         }

         1) In the above piece of code, the *ppage value from
ops->sendpage 		function is finally passed on to Kmap by the lower
level code to 		get the virtual address of the page.
	   2) In some corner cases the value of *ppage pointer is NULL.
         3) When highmem is enabled and a NULL pointer is passed to
            Kmap, then kmap finally crashes. But in the case when highmem
            is disabled, then kmap returns a junk value for NULL pointer.

               Highmem Enabled , kmap( NULL )-----> kernel crashes.

               Highmem disabled, kmap( NULL )-----> junk value is
returned.
		   				Subsequently this message
is observed on
						the console.

		                        "RPC call returned error 14"

         4) Now the question is why is the value of *ppage = NULL is
passed
		from the above piece of code to lower layers.
           Should that not have handled *ppage = NULL? and kmap should not
have received a NULL pointer?

Thanks,
  Sricharan