[PATCH] mm: silently skip readahead for DAX inodes

[PATCH] mm: silently skip readahead for DAX inodes

[Ross Zwisler]

For DAX inodes we need to be careful to never have page cache pages in the
mapping->page_tree.  This radix tree should be composed only of DAX
exceptional entries and zero pages.

ltp's readahead02 test was triggering a warning because we were trying to
insert a DAX exceptional entry but found that a page cache page had already
been inserted into the tree.  This page was being inserted into the radix
tree in response to a readahead(2) call.

Readahead doesn't make sense for DAX inodes, but we don't want it to report
a failure either.  Instead, we just return success and don't do any work.

Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reported-by: Jeff Moyer <jmoyer@redhat.com>
Cc: stable@vger.kernel.org
---
 mm/readahead.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/mm/readahead.c b/mm/readahead.c
index 65ec288..a9ba1be 100644
--- a/mm/readahead.c
+++ b/mm/readahead.c
@@ -8,6 +8,7 @@
  */
 
 #include <linux/kernel.h>
+#include <linux/dax.h>
 #include <linux/gfp.h>
 #include <linux/export.h>
 #include <linux/blkdev.h>
@@ -544,6 +545,9 @@ do_readahead(struct address_space *mapping, struct file *filp,
 	if (!mapping || !mapping->a_ops)
 		return -EINVAL;
 
+	if (dax_mapping(mapping))
+		return 0;
+
 	return force_page_cache_readahead(mapping, filp, index, nr);
 }
 
-- 
2.9.0

Re: [PATCH] mm: silently skip readahead for DAX inodes

[Andrew Morton]

On Wed, 24 Aug 2016 14:37:12 -0600 Ross Zwisler <ross.zwisler@linux.intel.com> wrote:

> For DAX inodes we need to be careful to never have page cache pages in the
> mapping->page_tree.  This radix tree should be composed only of DAX
> exceptional entries and zero pages.
> 
> ltp's readahead02 test was triggering a warning because we were trying to
> insert a DAX exceptional entry but found that a page cache page had already
> been inserted into the tree.  This page was being inserted into the radix
> tree in response to a readahead(2) call.
> 
> Readahead doesn't make sense for DAX inodes, but we don't want it to report
> a failure either.  Instead, we just return success and don't do any work.
> 
> --- a/mm/readahead.c
> +++ b/mm/readahead.c
> @@ -8,6 +8,7 @@
>   */
>  
>  #include <linux/kernel.h>
> +#include <linux/dax.h>
>  #include <linux/gfp.h>
>  #include <linux/export.h>
>  #include <linux/blkdev.h>
> @@ -544,6 +545,9 @@ do_readahead(struct address_space *mapping, struct file *filp,
>  	if (!mapping || !mapping->a_ops)
>  		return -EINVAL;
>  
> +	if (dax_mapping(mapping))
> +		return 0;
> +

Please don't force readers to go spend minutes putzing around in the
git tree trying to understand your code.
/* these things considered useful! */

[PATCH v2] mm: silently skip readahead for DAX inodes

[Ross Zwisler]

For DAX inodes we need to be careful to never have page cache pages in the
mapping->page_tree.  This radix tree should be composed only of DAX
exceptional entries and zero pages.

ltp's readahead02 test was triggering a warning because we were trying to
insert a DAX exceptional entry but found that a page cache page had
already been inserted into the tree.  This page was being inserted into the
radix tree in response to a readahead(2) call.

Readahead doesn't make sense for DAX inodes, but we don't want it to report
a failure either.  Instead, we just return success and don't do any work.

Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reported-by: Jeff Moyer <jmoyer@redhat.com>
---

Changes from v1:
 - Added a comment so readers don't have to go putzing around in the git
   tree to understand why we're doing what we're doing. :)  (akpm)

---
 mm/readahead.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/mm/readahead.c b/mm/readahead.c
index 65ec288..c8a955b 100644
--- a/mm/readahead.c
+++ b/mm/readahead.c
@@ -8,6 +8,7 @@
  */
 
 #include <linux/kernel.h>
+#include <linux/dax.h>
 #include <linux/gfp.h>
 #include <linux/export.h>
 #include <linux/blkdev.h>
@@ -544,6 +545,14 @@ do_readahead(struct address_space *mapping, struct file *filp,
 	if (!mapping || !mapping->a_ops)
 		return -EINVAL;
 
+	/*
+	 * Readahead doesn't make sense for DAX inodes, but we don't want it
+	 * to report a failure either.  Instead, we just return success and
+	 * don't do any work.
+	 */
+	if (dax_mapping(mapping))
+		return 0;
+
 	return force_page_cache_readahead(mapping, filp, index, nr);
 }
 
-- 
2.9.0

Re: [PATCH v2] mm: silently skip readahead for DAX inodes

[Andrew Morton]

On Wed, 24 Aug 2016 16:14:29 -0600 Ross Zwisler <ross.zwisler@linux.intel.com> wrote:

> For DAX inodes we need to be careful to never have page cache pages in the
> mapping->page_tree.  This radix tree should be composed only of DAX
> exceptional entries and zero pages.
> 
> ltp's readahead02 test was triggering a warning because we were trying to
> insert a DAX exceptional entry but found that a page cache page had
> already been inserted into the tree.  This page was being inserted into the
> radix tree in response to a readahead(2) call.
> 
> Readahead doesn't make sense for DAX inodes, but we don't want it to report
> a failure either.  Instead, we just return success and don't do any work.
> 
> Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
> Reported-by: Jeff Moyer <jmoyer@redhat.com>

cc:stable [4.5+] ?

[PATCH v2 RESEND] mm: silently skip readahead for DAX inodes

[Ross Zwisler]

For DAX inodes we need to be careful to never have page cache pages in the
mapping->page_tree.  This radix tree should be composed only of DAX
exceptional entries and zero pages.

ltp's readahead02 test was triggering a warning because we were trying to
insert a DAX exceptional entry but found that a page cache page had
already been inserted into the tree.  This page was being inserted into the
radix tree in response to a readahead(2) call.

Readahead doesn't make sense for DAX inodes, but we don't want it to report
a failure either.  Instead, we just return success and don't do any work.

Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reported-by: Jeff Moyer <jmoyer@redhat.com>
Cc: <stable@vger.kernel.org>    [4.5+]
---

Changes from v1:
 - Added a comment so readers don't have to go putzing around in the git
   tree to understand why we're doing what we're doing. :)  (akpm)
 - Resending, adding stable@vger.kernel.org.  Thank you, akpm, for the
   catch.

---
 mm/readahead.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/mm/readahead.c b/mm/readahead.c
index 65ec288..c8a955b 100644
--- a/mm/readahead.c
+++ b/mm/readahead.c
@@ -8,6 +8,7 @@
  */
 
 #include <linux/kernel.h>
+#include <linux/dax.h>
 #include <linux/gfp.h>
 #include <linux/export.h>
 #include <linux/blkdev.h>
@@ -544,6 +545,14 @@ do_readahead(struct address_space *mapping, struct file *filp,
 	if (!mapping || !mapping->a_ops)
 		return -EINVAL;
 
+	/*
+	 * Readahead doesn't make sense for DAX inodes, but we don't want it
+	 * to report a failure either.  Instead, we just return success and
+	 * don't do any work.
+	 */
+	if (dax_mapping(mapping))
+		return 0;
+
 	return force_page_cache_readahead(mapping, filp, index, nr);
 }
 
-- 
2.9.0