[PATCH] mm: add replace_page_cache_page() function

[PATCH] mm: add replace_page_cache_page() function

[Miklos Szeredi]

From: Miklos Szeredi <mszeredi@suse.cz>

This function basically does:

     remove_from_page_cache(old);
     page_cache_release(old);
     add_to_page_cache_locked(new);

Except it does this atomically, so there's no possibility for the
"add" to fail because of a race.

This is used by fuse to move pages into the page cache.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
---
 fs/fuse/dev.c           |   10 ++++------
 include/linux/pagemap.h |    1 +
 mm/filemap.c            |   41 +++++++++++++++++++++++++++++++++++++++++
 3 files changed, 46 insertions(+), 6 deletions(-)

Index: linux-2.6/mm/filemap.c
===================================================================
--- linux-2.6.orig/mm/filemap.c	2010-12-15 16:39:55.000000000 +0100
+++ linux-2.6/mm/filemap.c	2010-12-15 16:41:24.000000000 +0100
@@ -389,6 +389,47 @@ int filemap_write_and_wait_range(struct
 }
 EXPORT_SYMBOL(filemap_write_and_wait_range);
 
+int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
+{
+	int error;
+
+	VM_BUG_ON(!PageLocked(old));
+	VM_BUG_ON(!PageLocked(new));
+	VM_BUG_ON(new->mapping);
+
+	error = mem_cgroup_cache_charge(new, current->mm,
+					gfp_mask & GFP_RECLAIM_MASK);
+	if (error)
+		goto out;
+
+	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
+	if (error == 0) {
+		struct address_space *mapping = old->mapping;
+		pgoff_t offset = old->index;
+
+		page_cache_get(new);
+		new->mapping = mapping;
+		new->index = offset;
+
+		spin_lock_irq(&mapping->tree_lock);
+		__remove_from_page_cache(old);
+		error = radix_tree_insert(&mapping->page_tree, offset, new);
+		BUG_ON(error);
+		mapping->nrpages++;
+		__inc_zone_page_state(new, NR_FILE_PAGES);
+		if (PageSwapBacked(new))
+			__inc_zone_page_state(new, NR_SHMEM);
+		spin_unlock_irq(&mapping->tree_lock);
+		radix_tree_preload_end();
+		mem_cgroup_uncharge_cache_page(old);
+		page_cache_release(old);
+	} else
+		mem_cgroup_uncharge_cache_page(new);
+out:
+	return error;
+}
+EXPORT_SYMBOL_GPL(replace_page_cache_page);
+
 /**
  * add_to_page_cache_locked - add a locked page to the pagecache
  * @page:	page to add
Index: linux-2.6/include/linux/pagemap.h
===================================================================
--- linux-2.6.orig/include/linux/pagemap.h	2010-12-15 16:39:39.000000000 +0100
+++ linux-2.6/include/linux/pagemap.h	2010-12-15 16:41:24.000000000 +0100
@@ -457,6 +457,7 @@ int add_to_page_cache_lru(struct page *p
 				pgoff_t index, gfp_t gfp_mask);
 extern void remove_from_page_cache(struct page *page);
 extern void __remove_from_page_cache(struct page *page);
+int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
 
 /*
  * Like add_to_page_cache_locked, but used to add newly allocated pages:
Index: linux-2.6/fs/fuse/dev.c
===================================================================
--- linux-2.6.orig/fs/fuse/dev.c	2010-12-15 16:39:39.000000000 +0100
+++ linux-2.6/fs/fuse/dev.c	2010-12-15 16:41:24.000000000 +0100
@@ -729,14 +729,12 @@ static int fuse_try_move_page(struct fus
 	if (WARN_ON(PageMlocked(oldpage)))
 		goto out_fallback_unlock;
 
-	remove_from_page_cache(oldpage);
-	page_cache_release(oldpage);
-
-	err = add_to_page_cache_locked(newpage, mapping, index, GFP_KERNEL);
+	err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
 	if (err) {
-		printk(KERN_WARNING "fuse_try_move_page: failed to add page");
-		goto out_fallback_unlock;
+		unlock_page(newpage);
+		return err;
 	}
+
 	page_cache_get(newpage);
 
 	if (!(buf->flags & PIPE_BUF_FLAG_LRU))

[PATCH] mm: add replace_page_cache_page() function

[Miklos Szeredi]

From: Miklos Szeredi <mszeredi@suse.cz>

This function basically does:

     remove_from_page_cache(old);
     page_cache_release(old);
     add_to_page_cache_locked(new);

Except it does this atomically, so there's no possibility for the
"add" to fail because of a race.

This is used by fuse to move pages into the page cache.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
---
 fs/fuse/dev.c           |   10 ++++------
 include/linux/pagemap.h |    1 +
 mm/filemap.c            |   41 +++++++++++++++++++++++++++++++++++++++++
 3 files changed, 46 insertions(+), 6 deletions(-)

Index: linux-2.6/mm/filemap.c
===================================================================
--- linux-2.6.orig/mm/filemap.c	2010-12-15 16:39:55.000000000 +0100
+++ linux-2.6/mm/filemap.c	2010-12-15 16:41:24.000000000 +0100
@@ -389,6 +389,47 @@ int filemap_write_and_wait_range(struct
 }
 EXPORT_SYMBOL(filemap_write_and_wait_range);
 
+int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
+{
+	int error;
+
+	VM_BUG_ON(!PageLocked(old));
+	VM_BUG_ON(!PageLocked(new));
+	VM_BUG_ON(new->mapping);
+
+	error = mem_cgroup_cache_charge(new, current->mm,
+					gfp_mask & GFP_RECLAIM_MASK);
+	if (error)
+		goto out;
+
+	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
+	if (error == 0) {
+		struct address_space *mapping = old->mapping;
+		pgoff_t offset = old->index;
+
+		page_cache_get(new);
+		new->mapping = mapping;
+		new->index = offset;
+
+		spin_lock_irq(&mapping->tree_lock);
+		__remove_from_page_cache(old);
+		error = radix_tree_insert(&mapping->page_tree, offset, new);
+		BUG_ON(error);
+		mapping->nrpages++;
+		__inc_zone_page_state(new, NR_FILE_PAGES);
+		if (PageSwapBacked(new))
+			__inc_zone_page_state(new, NR_SHMEM);
+		spin_unlock_irq(&mapping->tree_lock);
+		radix_tree_preload_end();
+		mem_cgroup_uncharge_cache_page(old);
+		page_cache_release(old);
+	} else
+		mem_cgroup_uncharge_cache_page(new);
+out:
+	return error;
+}
+EXPORT_SYMBOL_GPL(replace_page_cache_page);
+
 /**
  * add_to_page_cache_locked - add a locked page to the pagecache
  * @page:	page to add
Index: linux-2.6/include/linux/pagemap.h
===================================================================
--- linux-2.6.orig/include/linux/pagemap.h	2010-12-15 16:39:39.000000000 +0100
+++ linux-2.6/include/linux/pagemap.h	2010-12-15 16:41:24.000000000 +0100
@@ -457,6 +457,7 @@ int add_to_page_cache_lru(struct page *p
 				pgoff_t index, gfp_t gfp_mask);
 extern void remove_from_page_cache(struct page *page);
 extern void __remove_from_page_cache(struct page *page);
+int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
 
 /*
  * Like add_to_page_cache_locked, but used to add newly allocated pages:
Index: linux-2.6/fs/fuse/dev.c
===================================================================
--- linux-2.6.orig/fs/fuse/dev.c	2010-12-15 16:39:39.000000000 +0100
+++ linux-2.6/fs/fuse/dev.c	2010-12-15 16:41:24.000000000 +0100
@@ -729,14 +729,12 @@ static int fuse_try_move_page(struct fus
 	if (WARN_ON(PageMlocked(oldpage)))
 		goto out_fallback_unlock;
 
-	remove_from_page_cache(oldpage);
-	page_cache_release(oldpage);
-
-	err = add_to_page_cache_locked(newpage, mapping, index, GFP_KERNEL);
+	err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
 	if (err) {
-		printk(KERN_WARNING "fuse_try_move_page: failed to add page");
-		goto out_fallback_unlock;
+		unlock_page(newpage);
+		return err;
 	}
+
 	page_cache_get(newpage);
 
 	if (!(buf->flags & PIPE_BUF_FLAG_LRU))

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[Rik van Riel]

On 12/15/2010 10:49 AM, Miklos Szeredi wrote:
> From: Miklos Szeredi<mszeredi@suse.cz>
>
> This function basically does:
>
>       remove_from_page_cache(old);
>       page_cache_release(old);
>       add_to_page_cache_locked(new);
>
> Except it does this atomically, so there's no possibility for the
> "add" to fail because of a race.
>
> This is used by fuse to move pages into the page cache.
>
> Signed-off-by: Miklos Szeredi<mszeredi@suse.cz>

Acked-by: Rik van Riel <riel@redhat.com>

-- 
All rights reversed

Re: [PATCH] mm: add replace_page_cache_page() function

[Rik van Riel]

On 12/15/2010 10:49 AM, Miklos Szeredi wrote:
> From: Miklos Szeredi<mszeredi@suse.cz>
>
> This function basically does:
>
>       remove_from_page_cache(old);
>       page_cache_release(old);
>       add_to_page_cache_locked(new);
>
> Except it does this atomically, so there's no possibility for the
> "add" to fail because of a race.
>
> This is used by fuse to move pages into the page cache.
>
> Signed-off-by: Miklos Szeredi<mszeredi@suse.cz>

Acked-by: Rik van Riel <riel@redhat.com>

-- 
All rights reversed

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[Minchan Kim]

On Thu, Dec 16, 2010 at 12:49 AM, Miklos Szeredi <miklos@szeredi.hu> wrote:
> From: Miklos Szeredi <mszeredi@suse.cz>
>
> This function basically does:
>
>     remove_from_page_cache(old);
>     page_cache_release(old);
>     add_to_page_cache_locked(new);
>
> Except it does this atomically, so there's no possibility for the
> "add" to fail because of a race.
>
> This is used by fuse to move pages into the page cache.

Please write down why fuse need this new atomic function in description.

>
> Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
> ---
>  fs/fuse/dev.c           |   10 ++++------
>  include/linux/pagemap.h |    1 +
>  mm/filemap.c            |   41 +++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 46 insertions(+), 6 deletions(-)
>
> Index: linux-2.6/mm/filemap.c
> ===================================================================
> --- linux-2.6.orig/mm/filemap.c 2010-12-15 16:39:55.000000000 +0100
> +++ linux-2.6/mm/filemap.c      2010-12-15 16:41:24.000000000 +0100
> @@ -389,6 +389,47 @@ int filemap_write_and_wait_range(struct
>  }
>  EXPORT_SYMBOL(filemap_write_and_wait_range);
>

This function is exported.
Please, add function description

> +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> +{
> +       int error;
> +
> +       VM_BUG_ON(!PageLocked(old));
> +       VM_BUG_ON(!PageLocked(new));
> +       VM_BUG_ON(new->mapping);
> +
> +       error = mem_cgroup_cache_charge(new, current->mm,
> +                                       gfp_mask & GFP_RECLAIM_MASK);
> +       if (error)
> +               goto out;
> +
> +       error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> +       if (error == 0) {
> +               struct address_space *mapping = old->mapping;
> +               pgoff_t offset = old->index;
> +
> +               page_cache_get(new);
> +               new->mapping = mapping;
> +               new->index = offset;
> +
> +               spin_lock_irq(&mapping->tree_lock);
> +               __remove_from_page_cache(old);
> +               error = radix_tree_insert(&mapping->page_tree, offset, new);
> +               BUG_ON(error);
> +               mapping->nrpages++;
> +               __inc_zone_page_state(new, NR_FILE_PAGES);
> +               if (PageSwapBacked(new))
> +                       __inc_zone_page_state(new, NR_SHMEM);
> +               spin_unlock_irq(&mapping->tree_lock);
> +               radix_tree_preload_end();
> +               mem_cgroup_uncharge_cache_page(old);
> +               page_cache_release(old);

Why do you release reference of old?

> +       } else
> +               mem_cgroup_uncharge_cache_page(new);
> +out:
> +       return error;
> +}
> +EXPORT_SYMBOL_GPL(replace_page_cache_page);
> +
>  /**
>  * add_to_page_cache_locked - add a locked page to the pagecache
>  * @page:      page to add
> Index: linux-2.6/include/linux/pagemap.h
> ===================================================================
> --- linux-2.6.orig/include/linux/pagemap.h      2010-12-15 16:39:39.000000000 +0100
> +++ linux-2.6/include/linux/pagemap.h   2010-12-15 16:41:24.000000000 +0100
> @@ -457,6 +457,7 @@ int add_to_page_cache_lru(struct page *p
>                                pgoff_t index, gfp_t gfp_mask);
>  extern void remove_from_page_cache(struct page *page);
>  extern void __remove_from_page_cache(struct page *page);
> +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
>
>  /*
>  * Like add_to_page_cache_locked, but used to add newly allocated pages:
> Index: linux-2.6/fs/fuse/dev.c
> ===================================================================
> --- linux-2.6.orig/fs/fuse/dev.c        2010-12-15 16:39:39.000000000 +0100
> +++ linux-2.6/fs/fuse/dev.c     2010-12-15 16:41:24.000000000 +0100
> @@ -729,14 +729,12 @@ static int fuse_try_move_page(struct fus
>        if (WARN_ON(PageMlocked(oldpage)))
>                goto out_fallback_unlock;
>
> -       remove_from_page_cache(oldpage);
> -       page_cache_release(oldpage);
> -
> -       err = add_to_page_cache_locked(newpage, mapping, index, GFP_KERNEL);
> +       err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
>        if (err) {
> -               printk(KERN_WARNING "fuse_try_move_page: failed to add page");
> -               goto out_fallback_unlock;
> +               unlock_page(newpage);
> +               return err;
>        }
> +
>        page_cache_get(newpage);
>
>        if (!(buf->flags & PIPE_BUF_FLAG_LRU))
>
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@kvack.org.  For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
> Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
>



-- 
Kind regards,
Minchan Kim

Re: [PATCH] mm: add replace_page_cache_page() function

[Minchan Kim]

On Thu, Dec 16, 2010 at 12:49 AM, Miklos Szeredi <miklos@szeredi.hu> wrote:
> From: Miklos Szeredi <mszeredi@suse.cz>
>
> This function basically does:
>
>     remove_from_page_cache(old);
>     page_cache_release(old);
>     add_to_page_cache_locked(new);
>
> Except it does this atomically, so there's no possibility for the
> "add" to fail because of a race.
>
> This is used by fuse to move pages into the page cache.

Please write down why fuse need this new atomic function in description.

>
> Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
> ---
>  fs/fuse/dev.c           |   10 ++++------
>  include/linux/pagemap.h |    1 +
>  mm/filemap.c            |   41 +++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 46 insertions(+), 6 deletions(-)
>
> Index: linux-2.6/mm/filemap.c
> ===================================================================
> --- linux-2.6.orig/mm/filemap.c 2010-12-15 16:39:55.000000000 +0100
> +++ linux-2.6/mm/filemap.c      2010-12-15 16:41:24.000000000 +0100
> @@ -389,6 +389,47 @@ int filemap_write_and_wait_range(struct
>  }
>  EXPORT_SYMBOL(filemap_write_and_wait_range);
>

This function is exported.
Please, add function description

> +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> +{
> +       int error;
> +
> +       VM_BUG_ON(!PageLocked(old));
> +       VM_BUG_ON(!PageLocked(new));
> +       VM_BUG_ON(new->mapping);
> +
> +       error = mem_cgroup_cache_charge(new, current->mm,
> +                                       gfp_mask & GFP_RECLAIM_MASK);
> +       if (error)
> +               goto out;
> +
> +       error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> +       if (error == 0) {
> +               struct address_space *mapping = old->mapping;
> +               pgoff_t offset = old->index;
> +
> +               page_cache_get(new);
> +               new->mapping = mapping;
> +               new->index = offset;
> +
> +               spin_lock_irq(&mapping->tree_lock);
> +               __remove_from_page_cache(old);
> +               error = radix_tree_insert(&mapping->page_tree, offset, new);
> +               BUG_ON(error);
> +               mapping->nrpages++;
> +               __inc_zone_page_state(new, NR_FILE_PAGES);
> +               if (PageSwapBacked(new))
> +                       __inc_zone_page_state(new, NR_SHMEM);
> +               spin_unlock_irq(&mapping->tree_lock);
> +               radix_tree_preload_end();
> +               mem_cgroup_uncharge_cache_page(old);
> +               page_cache_release(old);

Why do you release reference of old?

> +       } else
> +               mem_cgroup_uncharge_cache_page(new);
> +out:
> +       return error;
> +}
> +EXPORT_SYMBOL_GPL(replace_page_cache_page);
> +
>  /**
>  * add_to_page_cache_locked - add a locked page to the pagecache
>  * @page:      page to add
> Index: linux-2.6/include/linux/pagemap.h
> ===================================================================
> --- linux-2.6.orig/include/linux/pagemap.h      2010-12-15 16:39:39.000000000 +0100
> +++ linux-2.6/include/linux/pagemap.h   2010-12-15 16:41:24.000000000 +0100
> @@ -457,6 +457,7 @@ int add_to_page_cache_lru(struct page *p
>                                pgoff_t index, gfp_t gfp_mask);
>  extern void remove_from_page_cache(struct page *page);
>  extern void __remove_from_page_cache(struct page *page);
> +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
>
>  /*
>  * Like add_to_page_cache_locked, but used to add newly allocated pages:
> Index: linux-2.6/fs/fuse/dev.c
> ===================================================================
> --- linux-2.6.orig/fs/fuse/dev.c        2010-12-15 16:39:39.000000000 +0100
> +++ linux-2.6/fs/fuse/dev.c     2010-12-15 16:41:24.000000000 +0100
> @@ -729,14 +729,12 @@ static int fuse_try_move_page(struct fus
>        if (WARN_ON(PageMlocked(oldpage)))
>                goto out_fallback_unlock;
>
> -       remove_from_page_cache(oldpage);
> -       page_cache_release(oldpage);
> -
> -       err = add_to_page_cache_locked(newpage, mapping, index, GFP_KERNEL);
> +       err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
>        if (err) {
> -               printk(KERN_WARNING "fuse_try_move_page: failed to add page");
> -               goto out_fallback_unlock;
> +               unlock_page(newpage);
> +               return err;
>        }
> +
>        page_cache_get(newpage);
>
>        if (!(buf->flags & PIPE_BUF_FLAG_LRU))
>
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@kvack.org.  For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
> Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
>



-- 
Kind regards,
Minchan Kim
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] mm: add replace_page_cache_page() function

[Minchan Kim]

On Thu, Dec 16, 2010 at 12:49 AM, Miklos Szeredi <miklos@szeredi.hu> wrote:
> From: Miklos Szeredi <mszeredi@suse.cz>
>
> This function basically does:
>
>     remove_from_page_cache(old);
>     page_cache_release(old);
>     add_to_page_cache_locked(new);
>
> Except it does this atomically, so there's no possibility for the
> "add" to fail because of a race.
>
> This is used by fuse to move pages into the page cache.

Please write down why fuse need this new atomic function in description.

>
> Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
> ---
>  fs/fuse/dev.c           |   10 ++++------
>  include/linux/pagemap.h |    1 +
>  mm/filemap.c            |   41 +++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 46 insertions(+), 6 deletions(-)
>
> Index: linux-2.6/mm/filemap.c
> ===================================================================
> --- linux-2.6.orig/mm/filemap.c 2010-12-15 16:39:55.000000000 +0100
> +++ linux-2.6/mm/filemap.c      2010-12-15 16:41:24.000000000 +0100
> @@ -389,6 +389,47 @@ int filemap_write_and_wait_range(struct
>  }
>  EXPORT_SYMBOL(filemap_write_and_wait_range);
>

This function is exported.
Please, add function description

> +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> +{
> +       int error;
> +
> +       VM_BUG_ON(!PageLocked(old));
> +       VM_BUG_ON(!PageLocked(new));
> +       VM_BUG_ON(new->mapping);
> +
> +       error = mem_cgroup_cache_charge(new, current->mm,
> +                                       gfp_mask & GFP_RECLAIM_MASK);
> +       if (error)
> +               goto out;
> +
> +       error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> +       if (error == 0) {
> +               struct address_space *mapping = old->mapping;
> +               pgoff_t offset = old->index;
> +
> +               page_cache_get(new);
> +               new->mapping = mapping;
> +               new->index = offset;
> +
> +               spin_lock_irq(&mapping->tree_lock);
> +               __remove_from_page_cache(old);
> +               error = radix_tree_insert(&mapping->page_tree, offset, new);
> +               BUG_ON(error);
> +               mapping->nrpages++;
> +               __inc_zone_page_state(new, NR_FILE_PAGES);
> +               if (PageSwapBacked(new))
> +                       __inc_zone_page_state(new, NR_SHMEM);
> +               spin_unlock_irq(&mapping->tree_lock);
> +               radix_tree_preload_end();
> +               mem_cgroup_uncharge_cache_page(old);
> +               page_cache_release(old);

Why do you release reference of old?

> +       } else
> +               mem_cgroup_uncharge_cache_page(new);
> +out:
> +       return error;
> +}
> +EXPORT_SYMBOL_GPL(replace_page_cache_page);
> +
>  /**
>  * add_to_page_cache_locked - add a locked page to the pagecache
>  * @page:      page to add
> Index: linux-2.6/include/linux/pagemap.h
> ===================================================================
> --- linux-2.6.orig/include/linux/pagemap.h      2010-12-15 16:39:39.000000000 +0100
> +++ linux-2.6/include/linux/pagemap.h   2010-12-15 16:41:24.000000000 +0100
> @@ -457,6 +457,7 @@ int add_to_page_cache_lru(struct page *p
>                                pgoff_t index, gfp_t gfp_mask);
>  extern void remove_from_page_cache(struct page *page);
>  extern void __remove_from_page_cache(struct page *page);
> +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
>
>  /*
>  * Like add_to_page_cache_locked, but used to add newly allocated pages:
> Index: linux-2.6/fs/fuse/dev.c
> ===================================================================
> --- linux-2.6.orig/fs/fuse/dev.c        2010-12-15 16:39:39.000000000 +0100
> +++ linux-2.6/fs/fuse/dev.c     2010-12-15 16:41:24.000000000 +0100
> @@ -729,14 +729,12 @@ static int fuse_try_move_page(struct fus
>        if (WARN_ON(PageMlocked(oldpage)))
>                goto out_fallback_unlock;
>
> -       remove_from_page_cache(oldpage);
> -       page_cache_release(oldpage);
> -
> -       err = add_to_page_cache_locked(newpage, mapping, index, GFP_KERNEL);
> +       err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
>        if (err) {
> -               printk(KERN_WARNING "fuse_try_move_page: failed to add page");
> -               goto out_fallback_unlock;
> +               unlock_page(newpage);
> +               return err;
>        }
> +
>        page_cache_get(newpage);
>
>        if (!(buf->flags & PIPE_BUF_FLAG_LRU))
>
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@kvack.org.  For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
> Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
>



-- 
Kind regards,
Minchan Kim

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[KAMEZAWA Hiroyuki]

On Wed, 15 Dec 2010 16:49:58 +0100
Miklos Szeredi <miklos@szeredi.hu> wrote:

> From: Miklos Szeredi <mszeredi@suse.cz>
> 
> This function basically does:
> 
>      remove_from_page_cache(old);
>      page_cache_release(old);
>      add_to_page_cache_locked(new);
> 
> Except it does this atomically, so there's no possibility for the
> "add" to fail because of a race.
> 
> This is used by fuse to move pages into the page cache.
> 
> Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
> ---
>  fs/fuse/dev.c           |   10 ++++------
>  include/linux/pagemap.h |    1 +
>  mm/filemap.c            |   41 +++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 46 insertions(+), 6 deletions(-)
> 
> Index: linux-2.6/mm/filemap.c
> ===================================================================
> --- linux-2.6.orig/mm/filemap.c	2010-12-15 16:39:55.000000000 +0100
> +++ linux-2.6/mm/filemap.c	2010-12-15 16:41:24.000000000 +0100
> @@ -389,6 +389,47 @@ int filemap_write_and_wait_range(struct
>  }
>  EXPORT_SYMBOL(filemap_write_and_wait_range);
>  
> +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> +{
> +	int error;
> +
> +	VM_BUG_ON(!PageLocked(old));
> +	VM_BUG_ON(!PageLocked(new));
> +	VM_BUG_ON(new->mapping);
> +
> +	error = mem_cgroup_cache_charge(new, current->mm,
> +					gfp_mask & GFP_RECLAIM_MASK);

Hmm, then, the page will be recharged to "current" instead of the memcg
where "old" was under control. Is this design ? If so, why ?

In mm/migrate.c, following is called.

	 charge = mem_cgroup_prepare_migration(page, newpage, &mem);
	....do migration....
        if (!charge)
                mem_cgroup_end_migration(mem, page, newpage);

BTW, off topic, in fuse/dev.c

add_to_page_cache_locked(page)

is called and this page is "charged" to memory cgroup. But, IIUC, this page
will be never be on LRU and cannot be reclaimed by memory cgroup.
I think this looks like a memory leak at rmdir() of memory cgroup and
rmdir will fail wish -EBUSY always.

So, I'd like to change this call something like as

add_to_page_cache_locked_and_no_memory_cgroup_control().

So, I think just dropping this memory cgroup related code is okay for us
because this is a replacement for add_to_page_cache_locked() which seems
problematic.
This will put pages on fuse's private radix-tree out of control.

Or, is it possible to drain these radix-tree pages at rmdir() of memory
cgroup by some call ?

Thanks,
-Kame


> +	if (error)
> +		goto out;
> +
> +	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> +	if (error == 0) {
> +		struct address_space *mapping = old->mapping;
> +		pgoff_t offset = old->index;
> +
> +		page_cache_get(new);
> +		new->mapping = mapping;
> +		new->index = offset;
> +
> +		spin_lock_irq(&mapping->tree_lock);
> +		__remove_from_page_cache(old);
> +		error = radix_tree_insert(&mapping->page_tree, offset, new);
> +		BUG_ON(error);
> +		mapping->nrpages++;
> +		__inc_zone_page_state(new, NR_FILE_PAGES);
> +		if (PageSwapBacked(new))
> +			__inc_zone_page_state(new, NR_SHMEM);
> +		spin_unlock_irq(&mapping->tree_lock);
> +		radix_tree_preload_end();
> +		mem_cgroup_uncharge_cache_page(old);
> +		page_cache_release(old);
> +	} else
> +		mem_cgroup_uncharge_cache_page(new);
> +out:
> +	return error;
> +}
> +EXPORT_SYMBOL_GPL(replace_page_cache_page);
> +
>  /**
>   * add_to_page_cache_locked - add a locked page to the pagecache
>   * @page:	page to add
> Index: linux-2.6/include/linux/pagemap.h
> ===================================================================
> --- linux-2.6.orig/include/linux/pagemap.h	2010-12-15 16:39:39.000000000 +0100
> +++ linux-2.6/include/linux/pagemap.h	2010-12-15 16:41:24.000000000 +0100
> @@ -457,6 +457,7 @@ int add_to_page_cache_lru(struct page *p
>  				pgoff_t index, gfp_t gfp_mask);
>  extern void remove_from_page_cache(struct page *page);
>  extern void __remove_from_page_cache(struct page *page);
> +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
>  
>  /*
>   * Like add_to_page_cache_locked, but used to add newly allocated pages:
> Index: linux-2.6/fs/fuse/dev.c
> ===================================================================
> --- linux-2.6.orig/fs/fuse/dev.c	2010-12-15 16:39:39.000000000 +0100
> +++ linux-2.6/fs/fuse/dev.c	2010-12-15 16:41:24.000000000 +0100
> @@ -729,14 +729,12 @@ static int fuse_try_move_page(struct fus
>  	if (WARN_ON(PageMlocked(oldpage)))
>  		goto out_fallback_unlock;
>  
> -	remove_from_page_cache(oldpage);
> -	page_cache_release(oldpage);
> -
> -	err = add_to_page_cache_locked(newpage, mapping, index, GFP_KERNEL);
> +	err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
>  	if (err) {
> -		printk(KERN_WARNING "fuse_try_move_page: failed to add page");
> -		goto out_fallback_unlock;
> +		unlock_page(newpage);
> +		return err;
>  	}
> +
>  	page_cache_get(newpage);
>  
>  	if (!(buf->flags & PIPE_BUF_FLAG_LRU))
> 
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@kvack.org.  For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
> Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
> 

Re: [PATCH] mm: add replace_page_cache_page() function

[KAMEZAWA Hiroyuki]

On Wed, 15 Dec 2010 16:49:58 +0100
Miklos Szeredi <miklos@szeredi.hu> wrote:

> From: Miklos Szeredi <mszeredi@suse.cz>
> 
> This function basically does:
> 
>      remove_from_page_cache(old);
>      page_cache_release(old);
>      add_to_page_cache_locked(new);
> 
> Except it does this atomically, so there's no possibility for the
> "add" to fail because of a race.
> 
> This is used by fuse to move pages into the page cache.
> 
> Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
> ---
>  fs/fuse/dev.c           |   10 ++++------
>  include/linux/pagemap.h |    1 +
>  mm/filemap.c            |   41 +++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 46 insertions(+), 6 deletions(-)
> 
> Index: linux-2.6/mm/filemap.c
> ===================================================================
> --- linux-2.6.orig/mm/filemap.c	2010-12-15 16:39:55.000000000 +0100
> +++ linux-2.6/mm/filemap.c	2010-12-15 16:41:24.000000000 +0100
> @@ -389,6 +389,47 @@ int filemap_write_and_wait_range(struct
>  }
>  EXPORT_SYMBOL(filemap_write_and_wait_range);
>  
> +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> +{
> +	int error;
> +
> +	VM_BUG_ON(!PageLocked(old));
> +	VM_BUG_ON(!PageLocked(new));
> +	VM_BUG_ON(new->mapping);
> +
> +	error = mem_cgroup_cache_charge(new, current->mm,
> +					gfp_mask & GFP_RECLAIM_MASK);

Hmm, then, the page will be recharged to "current" instead of the memcg
where "old" was under control. Is this design ? If so, why ?

In mm/migrate.c, following is called.

	 charge = mem_cgroup_prepare_migration(page, newpage, &mem);
	....do migration....
        if (!charge)
                mem_cgroup_end_migration(mem, page, newpage);

BTW, off topic, in fuse/dev.c

add_to_page_cache_locked(page)

is called and this page is "charged" to memory cgroup. But, IIUC, this page
will be never be on LRU and cannot be reclaimed by memory cgroup.
I think this looks like a memory leak at rmdir() of memory cgroup and
rmdir will fail wish -EBUSY always.

So, I'd like to change this call something like as

add_to_page_cache_locked_and_no_memory_cgroup_control().

So, I think just dropping this memory cgroup related code is okay for us
because this is a replacement for add_to_page_cache_locked() which seems
problematic.
This will put pages on fuse's private radix-tree out of control.

Or, is it possible to drain these radix-tree pages at rmdir() of memory
cgroup by some call ?

Thanks,
-Kame


> +	if (error)
> +		goto out;
> +
> +	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> +	if (error == 0) {
> +		struct address_space *mapping = old->mapping;
> +		pgoff_t offset = old->index;
> +
> +		page_cache_get(new);
> +		new->mapping = mapping;
> +		new->index = offset;
> +
> +		spin_lock_irq(&mapping->tree_lock);
> +		__remove_from_page_cache(old);
> +		error = radix_tree_insert(&mapping->page_tree, offset, new);
> +		BUG_ON(error);
> +		mapping->nrpages++;
> +		__inc_zone_page_state(new, NR_FILE_PAGES);
> +		if (PageSwapBacked(new))
> +			__inc_zone_page_state(new, NR_SHMEM);
> +		spin_unlock_irq(&mapping->tree_lock);
> +		radix_tree_preload_end();
> +		mem_cgroup_uncharge_cache_page(old);
> +		page_cache_release(old);
> +	} else
> +		mem_cgroup_uncharge_cache_page(new);
> +out:
> +	return error;
> +}
> +EXPORT_SYMBOL_GPL(replace_page_cache_page);
> +
>  /**
>   * add_to_page_cache_locked - add a locked page to the pagecache
>   * @page:	page to add
> Index: linux-2.6/include/linux/pagemap.h
> ===================================================================
> --- linux-2.6.orig/include/linux/pagemap.h	2010-12-15 16:39:39.000000000 +0100
> +++ linux-2.6/include/linux/pagemap.h	2010-12-15 16:41:24.000000000 +0100
> @@ -457,6 +457,7 @@ int add_to_page_cache_lru(struct page *p
>  				pgoff_t index, gfp_t gfp_mask);
>  extern void remove_from_page_cache(struct page *page);
>  extern void __remove_from_page_cache(struct page *page);
> +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
>  
>  /*
>   * Like add_to_page_cache_locked, but used to add newly allocated pages:
> Index: linux-2.6/fs/fuse/dev.c
> ===================================================================
> --- linux-2.6.orig/fs/fuse/dev.c	2010-12-15 16:39:39.000000000 +0100
> +++ linux-2.6/fs/fuse/dev.c	2010-12-15 16:41:24.000000000 +0100
> @@ -729,14 +729,12 @@ static int fuse_try_move_page(struct fus
>  	if (WARN_ON(PageMlocked(oldpage)))
>  		goto out_fallback_unlock;
>  
> -	remove_from_page_cache(oldpage);
> -	page_cache_release(oldpage);
> -
> -	err = add_to_page_cache_locked(newpage, mapping, index, GFP_KERNEL);
> +	err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
>  	if (err) {
> -		printk(KERN_WARNING "fuse_try_move_page: failed to add page");
> -		goto out_fallback_unlock;
> +		unlock_page(newpage);
> +		return err;
>  	}
> +
>  	page_cache_get(newpage);
>  
>  	if (!(buf->flags & PIPE_BUF_FLAG_LRU))
> 
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@kvack.org.  For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
> Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
> 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[Miklos Szeredi]

On Thu, 16 Dec 2010, Minchan Kim wrote:
> On Thu, Dec 16, 2010 at 12:49 AM, Miklos Szeredi <miklos@szeredi.hu> wrote:
> > From: Miklos Szeredi <mszeredi@suse.cz>
> >
> > This function basically does:
> >
> >     remove_from_page_cache(old);
> >     page_cache_release(old);
> >     add_to_page_cache_locked(new);
> >
> > Except it does this atomically, so there's no possibility for the
> > "add" to fail because of a race.
> >
> > This is used by fuse to move pages into the page cache.
> 
> Please write down why fuse need this new atomic function in description.

Okay.

> >
> > Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
> > ---
> >  fs/fuse/dev.c           |   10 ++++------
> >  include/linux/pagemap.h |    1 +
> >  mm/filemap.c            |   41 +++++++++++++++++++++++++++++++++++++++++
> >  3 files changed, 46 insertions(+), 6 deletions(-)
> >
> > Index: linux-2.6/mm/filemap.c
> > ===================================================================
> > --- linux-2.6.orig/mm/filemap.c 2010-12-15 16:39:55.000000000 +0100
> > +++ linux-2.6/mm/filemap.c      2010-12-15 16:41:24.000000000 +0100
> > @@ -389,6 +389,47 @@ int filemap_write_and_wait_range(struct
> >  }
> >  EXPORT_SYMBOL(filemap_write_and_wait_range);
> >
> 
> This function is exported.
> Please, add function description

Right, will do.

> > +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> > +{
> > +       int error;
> > +
> > +       VM_BUG_ON(!PageLocked(old));
> > +       VM_BUG_ON(!PageLocked(new));
> > +       VM_BUG_ON(new->mapping);
> > +
> > +       error = mem_cgroup_cache_charge(new, current->mm,
> > +                                       gfp_mask & GFP_RECLAIM_MASK);
> > +       if (error)
> > +               goto out;
> > +
> > +       error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> > +       if (error == 0) {
> > +               struct address_space *mapping = old->mapping;
> > +               pgoff_t offset = old->index;
> > +
> > +               page_cache_get(new);
> > +               new->mapping = mapping;
> > +               new->index = offset;
> > +
> > +               spin_lock_irq(&mapping->tree_lock);
> > +               __remove_from_page_cache(old);
> > +               error = radix_tree_insert(&mapping->page_tree, offset, new);
> > +               BUG_ON(error);
> > +               mapping->nrpages++;
> > +               __inc_zone_page_state(new, NR_FILE_PAGES);
> > +               if (PageSwapBacked(new))
> > +                       __inc_zone_page_state(new, NR_SHMEM);
> > +               spin_unlock_irq(&mapping->tree_lock);
> > +               radix_tree_preload_end();
> > +               mem_cgroup_uncharge_cache_page(old);
> > +               page_cache_release(old);
> 
> Why do you release reference of old?

That's the page cache reference we release.  Just like we acquire the
page cache reference for "new" above.

I suspect it's historic that page_cache_release() doesn't drop the
page cache ref.

Thanks for the review.

Miklos

> > +       } else
> > +               mem_cgroup_uncharge_cache_page(new);
> > +out:
> > +       return error;
> > +}
> > +EXPORT_SYMBOL_GPL(replace_page_cache_page);
> > +
> >  /**
> >  * add_to_page_cache_locked - add a locked page to the pagecache
> >  * @page:      page to add
> > Index: linux-2.6/include/linux/pagemap.h
> > ===================================================================
> > --- linux-2.6.orig/include/linux/pagemap.h      2010-12-15 16:39:39.000000000 +0100
> > +++ linux-2.6/include/linux/pagemap.h   2010-12-15 16:41:24.000000000 +0100
> > @@ -457,6 +457,7 @@ int add_to_page_cache_lru(struct page *p
> >                                pgoff_t index, gfp_t gfp_mask);
> >  extern void remove_from_page_cache(struct page *page);
> >  extern void __remove_from_page_cache(struct page *page);
> > +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
> >
> >  /*
> >  * Like add_to_page_cache_locked, but used to add newly allocated pages:
> > Index: linux-2.6/fs/fuse/dev.c
> > ===================================================================
> > --- linux-2.6.orig/fs/fuse/dev.c        2010-12-15 16:39:39.000000000 +0100
> > +++ linux-2.6/fs/fuse/dev.c     2010-12-15 16:41:24.000000000 +0100
> > @@ -729,14 +729,12 @@ static int fuse_try_move_page(struct fus
> >        if (WARN_ON(PageMlocked(oldpage)))
> >                goto out_fallback_unlock;
> >
> > -       remove_from_page_cache(oldpage);
> > -       page_cache_release(oldpage);
> > -
> > -       err = add_to_page_cache_locked(newpage, mapping, index, GFP_KERNEL);
> > +       err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
> >        if (err) {
> > -               printk(KERN_WARNING "fuse_try_move_page: failed to add page");
> > -               goto out_fallback_unlock;
> > +               unlock_page(newpage);
> > +               return err;
> >        }
> > +
> >        page_cache_get(newpage);
> >
> >        if (!(buf->flags & PIPE_BUF_FLAG_LRU))
> >
> > --
> > To unsubscribe, send a message with 'unsubscribe linux-mm' in
> > the body to majordomo@kvack.org.  For more info on Linux MM,
> > see: http://www.linux-mm.org/ .
> > Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
> > Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
> >
> 
> 
> 
> -- 
> Kind regards,
> Minchan Kim
> 

Re: [PATCH] mm: add replace_page_cache_page() function

[Miklos Szeredi]

On Thu, 16 Dec 2010, Minchan Kim wrote:
> On Thu, Dec 16, 2010 at 12:49 AM, Miklos Szeredi <miklos@szeredi.hu> wrote:
> > From: Miklos Szeredi <mszeredi@suse.cz>
> >
> > This function basically does:
> >
> > A  A  remove_from_page_cache(old);
> > A  A  page_cache_release(old);
> > A  A  add_to_page_cache_locked(new);
> >
> > Except it does this atomically, so there's no possibility for the
> > "add" to fail because of a race.
> >
> > This is used by fuse to move pages into the page cache.
> 
> Please write down why fuse need this new atomic function in description.

Okay.

> >
> > Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
> > ---
> > A fs/fuse/dev.c A  A  A  A  A  | A  10 ++++------
> > A include/linux/pagemap.h | A  A 1 +
> > A mm/filemap.c A  A  A  A  A  A | A  41 +++++++++++++++++++++++++++++++++++++++++
> > A 3 files changed, 46 insertions(+), 6 deletions(-)
> >
> > Index: linux-2.6/mm/filemap.c
> > ===================================================================
> > --- linux-2.6.orig/mm/filemap.c 2010-12-15 16:39:55.000000000 +0100
> > +++ linux-2.6/mm/filemap.c A  A  A 2010-12-15 16:41:24.000000000 +0100
> > @@ -389,6 +389,47 @@ int filemap_write_and_wait_range(struct
> > A }
> > A EXPORT_SYMBOL(filemap_write_and_wait_range);
> >
> 
> This function is exported.
> Please, add function description

Right, will do.

> > +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> > +{
> > + A  A  A  int error;
> > +
> > + A  A  A  VM_BUG_ON(!PageLocked(old));
> > + A  A  A  VM_BUG_ON(!PageLocked(new));
> > + A  A  A  VM_BUG_ON(new->mapping);
> > +
> > + A  A  A  error = mem_cgroup_cache_charge(new, current->mm,
> > + A  A  A  A  A  A  A  A  A  A  A  A  A  A  A  A  A  A  A  gfp_mask & GFP_RECLAIM_MASK);
> > + A  A  A  if (error)
> > + A  A  A  A  A  A  A  goto out;
> > +
> > + A  A  A  error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> > + A  A  A  if (error == 0) {
> > + A  A  A  A  A  A  A  struct address_space *mapping = old->mapping;
> > + A  A  A  A  A  A  A  pgoff_t offset = old->index;
> > +
> > + A  A  A  A  A  A  A  page_cache_get(new);
> > + A  A  A  A  A  A  A  new->mapping = mapping;
> > + A  A  A  A  A  A  A  new->index = offset;
> > +
> > + A  A  A  A  A  A  A  spin_lock_irq(&mapping->tree_lock);
> > + A  A  A  A  A  A  A  __remove_from_page_cache(old);
> > + A  A  A  A  A  A  A  error = radix_tree_insert(&mapping->page_tree, offset, new);
> > + A  A  A  A  A  A  A  BUG_ON(error);
> > + A  A  A  A  A  A  A  mapping->nrpages++;
> > + A  A  A  A  A  A  A  __inc_zone_page_state(new, NR_FILE_PAGES);
> > + A  A  A  A  A  A  A  if (PageSwapBacked(new))
> > + A  A  A  A  A  A  A  A  A  A  A  __inc_zone_page_state(new, NR_SHMEM);
> > + A  A  A  A  A  A  A  spin_unlock_irq(&mapping->tree_lock);
> > + A  A  A  A  A  A  A  radix_tree_preload_end();
> > + A  A  A  A  A  A  A  mem_cgroup_uncharge_cache_page(old);
> > + A  A  A  A  A  A  A  page_cache_release(old);
> 
> Why do you release reference of old?

That's the page cache reference we release.  Just like we acquire the
page cache reference for "new" above.

I suspect it's historic that page_cache_release() doesn't drop the
page cache ref.

Thanks for the review.

Miklos

> > + A  A  A  } else
> > + A  A  A  A  A  A  A  mem_cgroup_uncharge_cache_page(new);
> > +out:
> > + A  A  A  return error;
> > +}
> > +EXPORT_SYMBOL_GPL(replace_page_cache_page);
> > +
> > A /**
> > A * add_to_page_cache_locked - add a locked page to the pagecache
> > A * @page: A  A  A page to add
> > Index: linux-2.6/include/linux/pagemap.h
> > ===================================================================
> > --- linux-2.6.orig/include/linux/pagemap.h A  A  A 2010-12-15 16:39:39.000000000 +0100
> > +++ linux-2.6/include/linux/pagemap.h A  2010-12-15 16:41:24.000000000 +0100
> > @@ -457,6 +457,7 @@ int add_to_page_cache_lru(struct page *p
> > A  A  A  A  A  A  A  A  A  A  A  A  A  A  A  A pgoff_t index, gfp_t gfp_mask);
> > A extern void remove_from_page_cache(struct page *page);
> > A extern void __remove_from_page_cache(struct page *page);
> > +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
> >
> > A /*
> > A * Like add_to_page_cache_locked, but used to add newly allocated pages:
> > Index: linux-2.6/fs/fuse/dev.c
> > ===================================================================
> > --- linux-2.6.orig/fs/fuse/dev.c A  A  A  A 2010-12-15 16:39:39.000000000 +0100
> > +++ linux-2.6/fs/fuse/dev.c A  A  2010-12-15 16:41:24.000000000 +0100
> > @@ -729,14 +729,12 @@ static int fuse_try_move_page(struct fus
> > A  A  A  A if (WARN_ON(PageMlocked(oldpage)))
> > A  A  A  A  A  A  A  A goto out_fallback_unlock;
> >
> > - A  A  A  remove_from_page_cache(oldpage);
> > - A  A  A  page_cache_release(oldpage);
> > -
> > - A  A  A  err = add_to_page_cache_locked(newpage, mapping, index, GFP_KERNEL);
> > + A  A  A  err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
> > A  A  A  A if (err) {
> > - A  A  A  A  A  A  A  printk(KERN_WARNING "fuse_try_move_page: failed to add page");
> > - A  A  A  A  A  A  A  goto out_fallback_unlock;
> > + A  A  A  A  A  A  A  unlock_page(newpage);
> > + A  A  A  A  A  A  A  return err;
> > A  A  A  A }
> > +
> > A  A  A  A page_cache_get(newpage);
> >
> > A  A  A  A if (!(buf->flags & PIPE_BUF_FLAG_LRU))
> >
> > --
> > To unsubscribe, send a message with 'unsubscribe linux-mm' in
> > the body to majordomo@kvack.org. A For more info on Linux MM,
> > see: http://www.linux-mm.org/ .
> > Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
> > Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
> >
> 
> 
> 
> -- 
> Kind regards,
> Minchan Kim
> 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[Miklos Szeredi]

On Thu, 16 Dec 2010, KAMEZAWA Hiroyuki wrote:
> On Wed, 15 Dec 2010 16:49:58 +0100
> Miklos Szeredi <miklos@szeredi.hu> wrote:
> 
> > From: Miklos Szeredi <mszeredi@suse.cz>
> > 
> > This function basically does:
> > 
> >      remove_from_page_cache(old);
> >      page_cache_release(old);
> >      add_to_page_cache_locked(new);
> > 
> > Except it does this atomically, so there's no possibility for the
> > "add" to fail because of a race.
> > 
> > This is used by fuse to move pages into the page cache.
> > 
> > Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
> > ---
> >  fs/fuse/dev.c           |   10 ++++------
> >  include/linux/pagemap.h |    1 +
> >  mm/filemap.c            |   41 +++++++++++++++++++++++++++++++++++++++++
> >  3 files changed, 46 insertions(+), 6 deletions(-)
> > 
> > Index: linux-2.6/mm/filemap.c
> > ===================================================================
> > --- linux-2.6.orig/mm/filemap.c	2010-12-15 16:39:55.000000000 +0100
> > +++ linux-2.6/mm/filemap.c	2010-12-15 16:41:24.000000000 +0100
> > @@ -389,6 +389,47 @@ int filemap_write_and_wait_range(struct
> >  }
> >  EXPORT_SYMBOL(filemap_write_and_wait_range);
> >  
> > +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> > +{
> > +	int error;
> > +
> > +	VM_BUG_ON(!PageLocked(old));
> > +	VM_BUG_ON(!PageLocked(new));
> > +	VM_BUG_ON(new->mapping);
> > +
> > +	error = mem_cgroup_cache_charge(new, current->mm,
> > +					gfp_mask & GFP_RECLAIM_MASK);
> 
> Hmm, then, the page will be recharged to "current" instead of the memcg
> where "old" was under control. Is this design ? If so, why ?

No, I just haven't thought about it.

Porbably charging "new" to where "old" was charged is the logical
thing to do here.

> 
> In mm/migrate.c, following is called.
> 
> 	 charge = mem_cgroup_prepare_migration(page, newpage, &mem);
> 	....do migration....
>         if (!charge)
>                 mem_cgroup_end_migration(mem, page, newpage);
> 
> BTW, off topic, in fuse/dev.c
> 
> add_to_page_cache_locked(page)

This is the call which the above patch replaces with
replace_page_cache_page().  So if I fix replace_page_cache_page() to
charge "newpage" to the correct memory cgroup, that should solve all
problems, no?

Thanks for the review.

Miklos

Re: [PATCH] mm: add replace_page_cache_page() function

[Miklos Szeredi]

On Thu, 16 Dec 2010, KAMEZAWA Hiroyuki wrote:
> On Wed, 15 Dec 2010 16:49:58 +0100
> Miklos Szeredi <miklos@szeredi.hu> wrote:
> 
> > From: Miklos Szeredi <mszeredi@suse.cz>
> > 
> > This function basically does:
> > 
> >      remove_from_page_cache(old);
> >      page_cache_release(old);
> >      add_to_page_cache_locked(new);
> > 
> > Except it does this atomically, so there's no possibility for the
> > "add" to fail because of a race.
> > 
> > This is used by fuse to move pages into the page cache.
> > 
> > Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
> > ---
> >  fs/fuse/dev.c           |   10 ++++------
> >  include/linux/pagemap.h |    1 +
> >  mm/filemap.c            |   41 +++++++++++++++++++++++++++++++++++++++++
> >  3 files changed, 46 insertions(+), 6 deletions(-)
> > 
> > Index: linux-2.6/mm/filemap.c
> > ===================================================================
> > --- linux-2.6.orig/mm/filemap.c	2010-12-15 16:39:55.000000000 +0100
> > +++ linux-2.6/mm/filemap.c	2010-12-15 16:41:24.000000000 +0100
> > @@ -389,6 +389,47 @@ int filemap_write_and_wait_range(struct
> >  }
> >  EXPORT_SYMBOL(filemap_write_and_wait_range);
> >  
> > +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> > +{
> > +	int error;
> > +
> > +	VM_BUG_ON(!PageLocked(old));
> > +	VM_BUG_ON(!PageLocked(new));
> > +	VM_BUG_ON(new->mapping);
> > +
> > +	error = mem_cgroup_cache_charge(new, current->mm,
> > +					gfp_mask & GFP_RECLAIM_MASK);
> 
> Hmm, then, the page will be recharged to "current" instead of the memcg
> where "old" was under control. Is this design ? If so, why ?

No, I just haven't thought about it.

Porbably charging "new" to where "old" was charged is the logical
thing to do here.

> 
> In mm/migrate.c, following is called.
> 
> 	 charge = mem_cgroup_prepare_migration(page, newpage, &mem);
> 	....do migration....
>         if (!charge)
>                 mem_cgroup_end_migration(mem, page, newpage);
> 
> BTW, off topic, in fuse/dev.c
> 
> add_to_page_cache_locked(page)

This is the call which the above patch replaces with
replace_page_cache_page().  So if I fix replace_page_cache_page() to
charge "newpage" to the correct memory cgroup, that should solve all
problems, no?

Thanks for the review.

Miklos

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[Minchan Kim]

On Thu, Dec 16, 2010 at 12:59:58PM +0100, Miklos Szeredi wrote:
> On Thu, 16 Dec 2010, Minchan Kim wrote:
> > > +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> > > +{
> > > + ?? ?? ?? int error;
> > > +
> > > + ?? ?? ?? VM_BUG_ON(!PageLocked(old));
> > > + ?? ?? ?? VM_BUG_ON(!PageLocked(new));
> > > + ?? ?? ?? VM_BUG_ON(new->mapping);
> > > +
> > > + ?? ?? ?? error = mem_cgroup_cache_charge(new, current->mm,
> > > + ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? gfp_mask & GFP_RECLAIM_MASK);
> > > + ?? ?? ?? if (error)
> > > + ?? ?? ?? ?? ?? ?? ?? goto out;
> > > +
> > > + ?? ?? ?? error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> > > + ?? ?? ?? if (error == 0) {
> > > + ?? ?? ?? ?? ?? ?? ?? struct address_space *mapping = old->mapping;
> > > + ?? ?? ?? ?? ?? ?? ?? pgoff_t offset = old->index;
> > > +
> > > + ?? ?? ?? ?? ?? ?? ?? page_cache_get(new);
> > > + ?? ?? ?? ?? ?? ?? ?? new->mapping = mapping;
> > > + ?? ?? ?? ?? ?? ?? ?? new->index = offset;
> > > +
> > > + ?? ?? ?? ?? ?? ?? ?? spin_lock_irq(&mapping->tree_lock);
> > > + ?? ?? ?? ?? ?? ?? ?? __remove_from_page_cache(old);
> > > + ?? ?? ?? ?? ?? ?? ?? error = radix_tree_insert(&mapping->page_tree, offset, new);
> > > + ?? ?? ?? ?? ?? ?? ?? BUG_ON(error);
> > > + ?? ?? ?? ?? ?? ?? ?? mapping->nrpages++;
> > > + ?? ?? ?? ?? ?? ?? ?? __inc_zone_page_state(new, NR_FILE_PAGES);
> > > + ?? ?? ?? ?? ?? ?? ?? if (PageSwapBacked(new))
> > > + ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? __inc_zone_page_state(new, NR_SHMEM);
> > > + ?? ?? ?? ?? ?? ?? ?? spin_unlock_irq(&mapping->tree_lock);
> > > + ?? ?? ?? ?? ?? ?? ?? radix_tree_preload_end();
> > > + ?? ?? ?? ?? ?? ?? ?? mem_cgroup_uncharge_cache_page(old);
> > > + ?? ?? ?? ?? ?? ?? ?? page_cache_release(old);
> > 
> > Why do you release reference of old?
> 
> That's the page cache reference we release.  Just like we acquire the
> page cache reference for "new" above.

I mean current page cache handling semantic and page reference counting semantic
is separeated. For example, remove_from_page_cache doesn't drop the reference of page.
That's because we need more works after drop the page from page cache.
Look at shmem_writepage, truncate_complete_page.

You makes the general API and caller might need works before the old page 
is free. So how about this?

err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
if (err) {
        ...
}

page_cache_release(oldpage); /* drop ref of page cache */


> 
> I suspect it's historic that page_cache_release() doesn't drop the
> page cache ref.

Sorry I can't understand your words.

> 
> Thanks for the review.
> 
> Miklos
-- 
Kind regards,
Minchan Kim

Re: [PATCH] mm: add replace_page_cache_page() function

[Minchan Kim]

On Thu, Dec 16, 2010 at 12:59:58PM +0100, Miklos Szeredi wrote:
> On Thu, 16 Dec 2010, Minchan Kim wrote:
> > > +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> > > +{
> > > + ?? ?? ?? int error;
> > > +
> > > + ?? ?? ?? VM_BUG_ON(!PageLocked(old));
> > > + ?? ?? ?? VM_BUG_ON(!PageLocked(new));
> > > + ?? ?? ?? VM_BUG_ON(new->mapping);
> > > +
> > > + ?? ?? ?? error = mem_cgroup_cache_charge(new, current->mm,
> > > + ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? gfp_mask & GFP_RECLAIM_MASK);
> > > + ?? ?? ?? if (error)
> > > + ?? ?? ?? ?? ?? ?? ?? goto out;
> > > +
> > > + ?? ?? ?? error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> > > + ?? ?? ?? if (error == 0) {
> > > + ?? ?? ?? ?? ?? ?? ?? struct address_space *mapping = old->mapping;
> > > + ?? ?? ?? ?? ?? ?? ?? pgoff_t offset = old->index;
> > > +
> > > + ?? ?? ?? ?? ?? ?? ?? page_cache_get(new);
> > > + ?? ?? ?? ?? ?? ?? ?? new->mapping = mapping;
> > > + ?? ?? ?? ?? ?? ?? ?? new->index = offset;
> > > +
> > > + ?? ?? ?? ?? ?? ?? ?? spin_lock_irq(&mapping->tree_lock);
> > > + ?? ?? ?? ?? ?? ?? ?? __remove_from_page_cache(old);
> > > + ?? ?? ?? ?? ?? ?? ?? error = radix_tree_insert(&mapping->page_tree, offset, new);
> > > + ?? ?? ?? ?? ?? ?? ?? BUG_ON(error);
> > > + ?? ?? ?? ?? ?? ?? ?? mapping->nrpages++;
> > > + ?? ?? ?? ?? ?? ?? ?? __inc_zone_page_state(new, NR_FILE_PAGES);
> > > + ?? ?? ?? ?? ?? ?? ?? if (PageSwapBacked(new))
> > > + ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? __inc_zone_page_state(new, NR_SHMEM);
> > > + ?? ?? ?? ?? ?? ?? ?? spin_unlock_irq(&mapping->tree_lock);
> > > + ?? ?? ?? ?? ?? ?? ?? radix_tree_preload_end();
> > > + ?? ?? ?? ?? ?? ?? ?? mem_cgroup_uncharge_cache_page(old);
> > > + ?? ?? ?? ?? ?? ?? ?? page_cache_release(old);
> > 
> > Why do you release reference of old?
> 
> That's the page cache reference we release.  Just like we acquire the
> page cache reference for "new" above.

I mean current page cache handling semantic and page reference counting semantic
is separeated. For example, remove_from_page_cache doesn't drop the reference of page.
That's because we need more works after drop the page from page cache.
Look at shmem_writepage, truncate_complete_page.

You makes the general API and caller might need works before the old page 
is free. So how about this?

err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
if (err) {
        ...
}

page_cache_release(oldpage); /* drop ref of page cache */


> 
> I suspect it's historic that page_cache_release() doesn't drop the
> page cache ref.

Sorry I can't understand your words.

> 
> Thanks for the review.
> 
> Miklos
-- 
Kind regards,
Minchan Kim

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[KAMEZAWA Hiroyuki]

On Thu, 16 Dec 2010 13:05:44 +0100
Miklos Szeredi <miklos@szeredi.hu> wrote:

> On Thu, 16 Dec 2010, KAMEZAWA Hiroyuki wrote:

> > Hmm, then, the page will be recharged to "current" instead of the memcg
> > where "old" was under control. Is this design ? If so, why ?
> 
> No, I just haven't thought about it.
> 
> Porbably charging "new" to where "old" was charged is the logical
> thing to do here.
> 
> > 
> > In mm/migrate.c, following is called.
> > 
> > 	 charge = mem_cgroup_prepare_migration(page, newpage, &mem);
> > 	....do migration....
> >         if (!charge)
> >                 mem_cgroup_end_migration(mem, page, newpage);
> > 
> > BTW, off topic, in fuse/dev.c
> > 
> > add_to_page_cache_locked(page)
> 
> This is the call which the above patch replaces with
> replace_page_cache_page().  So if I fix replace_page_cache_page() to
> charge "newpage" to the correct memory cgroup, that should solve all
> problems, no?
> 
No. memory cgroup expects all pages should be found on LRU. But, IIUC,
pages on this radix-tree will not be on LRU. So, memory cgroup can't find
it at destroying cgroup and can't reduce "usage" of resource to be 0.
This makes rmdir() returns -EBUSY.

I'm sorry if this page will be on LRU, somewhere.

Thanks,
-Kame

Re: [PATCH] mm: add replace_page_cache_page() function

[KAMEZAWA Hiroyuki]

On Thu, 16 Dec 2010 13:05:44 +0100
Miklos Szeredi <miklos@szeredi.hu> wrote:

> On Thu, 16 Dec 2010, KAMEZAWA Hiroyuki wrote:

> > Hmm, then, the page will be recharged to "current" instead of the memcg
> > where "old" was under control. Is this design ? If so, why ?
> 
> No, I just haven't thought about it.
> 
> Porbably charging "new" to where "old" was charged is the logical
> thing to do here.
> 
> > 
> > In mm/migrate.c, following is called.
> > 
> > 	 charge = mem_cgroup_prepare_migration(page, newpage, &mem);
> > 	....do migration....
> >         if (!charge)
> >                 mem_cgroup_end_migration(mem, page, newpage);
> > 
> > BTW, off topic, in fuse/dev.c
> > 
> > add_to_page_cache_locked(page)
> 
> This is the call which the above patch replaces with
> replace_page_cache_page().  So if I fix replace_page_cache_page() to
> charge "newpage" to the correct memory cgroup, that should solve all
> problems, no?
> 
No. memory cgroup expects all pages should be found on LRU. But, IIUC,
pages on this radix-tree will not be on LRU. So, memory cgroup can't find
it at destroying cgroup and can't reduce "usage" of resource to be 0.
This makes rmdir() returns -EBUSY.

I'm sorry if this page will be on LRU, somewhere.

Thanks,
-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[Hugh Dickins]

On Fri, 17 Dec 2010, Minchan Kim wrote:
> On Thu, Dec 16, 2010 at 12:59:58PM +0100, Miklos Szeredi wrote:
> > On Thu, 16 Dec 2010, Minchan Kim wrote:
> > > 
> > > Why do you release reference of old?
> > 
> > That's the page cache reference we release.  Just like we acquire the
> > page cache reference for "new" above.
> 
> I mean current page cache handling semantic and page reference counting semantic
> is separeated. For example, remove_from_page_cache doesn't drop the reference of page.
> That's because we need more works after drop the page from page cache.
> Look at shmem_writepage, truncate_complete_page.

I disagree with you there: I like the way Miklos made it symmetric,
I like the way delete_from_swap_cache drops the swap cache reference,
I dislike the way remove_from_page_cache does not - I did once try to
change that, but did a bad job, messed up reiserfs or reiser4 I forget
which, retreated in shame.

In both the examples you give, shmem_writepage and truncate_complete_page,
the caller has to be holding their own reference, in part because they
locked the page, and will need to unlock it before releasing their ref.
I think that would be true of any replace_page_cache_page caller.

> 
> You makes the general API and caller might need works before the old page 
> is free. So how about this?
> 
> err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
> if (err) {
>         ...
> }
> 
> page_cache_release(oldpage); /* drop ref of page cache */
> 
> 
> > 
> > I suspect it's historic that page_cache_release() doesn't drop the
> > page cache ref.
> 
> Sorry I can't understand your words.

Me neither: I believe Miklos meant __remove_from_page_cache() rather
than page_cache_release() in that instance.

Hugh

Re: [PATCH] mm: add replace_page_cache_page() function

[Hugh Dickins]

On Fri, 17 Dec 2010, Minchan Kim wrote:
> On Thu, Dec 16, 2010 at 12:59:58PM +0100, Miklos Szeredi wrote:
> > On Thu, 16 Dec 2010, Minchan Kim wrote:
> > > 
> > > Why do you release reference of old?
> > 
> > That's the page cache reference we release.  Just like we acquire the
> > page cache reference for "new" above.
> 
> I mean current page cache handling semantic and page reference counting semantic
> is separeated. For example, remove_from_page_cache doesn't drop the reference of page.
> That's because we need more works after drop the page from page cache.
> Look at shmem_writepage, truncate_complete_page.

I disagree with you there: I like the way Miklos made it symmetric,
I like the way delete_from_swap_cache drops the swap cache reference,
I dislike the way remove_from_page_cache does not - I did once try to
change that, but did a bad job, messed up reiserfs or reiser4 I forget
which, retreated in shame.

In both the examples you give, shmem_writepage and truncate_complete_page,
the caller has to be holding their own reference, in part because they
locked the page, and will need to unlock it before releasing their ref.
I think that would be true of any replace_page_cache_page caller.

> 
> You makes the general API and caller might need works before the old page 
> is free. So how about this?
> 
> err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
> if (err) {
>         ...
> }
> 
> page_cache_release(oldpage); /* drop ref of page cache */
> 
> 
> > 
> > I suspect it's historic that page_cache_release() doesn't drop the
> > page cache ref.
> 
> Sorry I can't understand your words.

Me neither: I believe Miklos meant __remove_from_page_cache() rather
than page_cache_release() in that instance.

Hugh

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[Minchan Kim]

Hi Hugh,

On Fri, Dec 17, 2010 at 10:21 AM, Hugh Dickins <hughd@google.com> wrote:
> On Fri, 17 Dec 2010, Minchan Kim wrote:
>> On Thu, Dec 16, 2010 at 12:59:58PM +0100, Miklos Szeredi wrote:
>> > On Thu, 16 Dec 2010, Minchan Kim wrote:
>> > >
>> > > Why do you release reference of old?
>> >
>> > That's the page cache reference we release.  Just like we acquire the
>> > page cache reference for "new" above.
>>
>> I mean current page cache handling semantic and page reference counting semantic
>> is separeated. For example, remove_from_page_cache doesn't drop the reference of page.
>> That's because we need more works after drop the page from page cache.
>> Look at shmem_writepage, truncate_complete_page.
>
> I disagree with you there: I like the way Miklos made it symmetric,
> I like the way delete_from_swap_cache drops the swap cache reference,
> I dislike the way remove_from_page_cache does not - I did once try to
> change that, but did a bad job, messed up reiserfs or reiser4 I forget
> which, retreated in shame.

I agree symmetric is good. I just said current fact which is that
remove_from_page_cache doesn't release ref.
So I thought we have to match current semantic to protect confusing.
Okay. I will not oppose current semantics.
Instead of it, please add it (ex, caller should hold the page
reference) in function description.

I am happy to hear that you tried it.
Although it is hard, I think it's very valuable thing.
Could you give me hint to googling your effort and why it is failed?

>
> In both the examples you give, shmem_writepage and truncate_complete_page,
> the caller has to be holding their own reference, in part because they
> locked the page, and will need to unlock it before releasing their ref.
> I think that would be true of any replace_page_cache_page caller.

Agree.

>
>>
>> You makes the general API and caller might need works before the old page
>> is free. So how about this?
>>
>> err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
>> if (err) {
>>         ...
>> }
>>
>> page_cache_release(oldpage); /* drop ref of page cache */
>>
>>
>> >
>> > I suspect it's historic that page_cache_release() doesn't drop the
>> > page cache ref.
>>
>> Sorry I can't understand your words.
>
> Me neither: I believe Miklos meant __remove_from_page_cache() rather
> than page_cache_release() in that instance.

Maybe. :)

Thanks for the comment, Hugh.
>
> Hugh
>



-- 
Kind regards,
Minchan Kim

Re: [PATCH] mm: add replace_page_cache_page() function

[Minchan Kim]

Hi Hugh,

On Fri, Dec 17, 2010 at 10:21 AM, Hugh Dickins <hughd@google.com> wrote:
> On Fri, 17 Dec 2010, Minchan Kim wrote:
>> On Thu, Dec 16, 2010 at 12:59:58PM +0100, Miklos Szeredi wrote:
>> > On Thu, 16 Dec 2010, Minchan Kim wrote:
>> > >
>> > > Why do you release reference of old?
>> >
>> > That's the page cache reference we release.  Just like we acquire the
>> > page cache reference for "new" above.
>>
>> I mean current page cache handling semantic and page reference counting semantic
>> is separeated. For example, remove_from_page_cache doesn't drop the reference of page.
>> That's because we need more works after drop the page from page cache.
>> Look at shmem_writepage, truncate_complete_page.
>
> I disagree with you there: I like the way Miklos made it symmetric,
> I like the way delete_from_swap_cache drops the swap cache reference,
> I dislike the way remove_from_page_cache does not - I did once try to
> change that, but did a bad job, messed up reiserfs or reiser4 I forget
> which, retreated in shame.

I agree symmetric is good. I just said current fact which is that
remove_from_page_cache doesn't release ref.
So I thought we have to match current semantic to protect confusing.
Okay. I will not oppose current semantics.
Instead of it, please add it (ex, caller should hold the page
reference) in function description.

I am happy to hear that you tried it.
Although it is hard, I think it's very valuable thing.
Could you give me hint to googling your effort and why it is failed?

>
> In both the examples you give, shmem_writepage and truncate_complete_page,
> the caller has to be holding their own reference, in part because they
> locked the page, and will need to unlock it before releasing their ref.
> I think that would be true of any replace_page_cache_page caller.

Agree.

>
>>
>> You makes the general API and caller might need works before the old page
>> is free. So how about this?
>>
>> err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
>> if (err) {
>>         ...
>> }
>>
>> page_cache_release(oldpage); /* drop ref of page cache */
>>
>>
>> >
>> > I suspect it's historic that page_cache_release() doesn't drop the
>> > page cache ref.
>>
>> Sorry I can't understand your words.
>
> Me neither: I believe Miklos meant __remove_from_page_cache() rather
> than page_cache_release() in that instance.

Maybe. :)

Thanks for the comment, Hugh.
>
> Hugh
>



-- 
Kind regards,
Minchan Kim

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[Hugh Dickins]

On Fri, 17 Dec 2010, Minchan Kim wrote:
> On Fri, Dec 17, 2010 at 10:21 AM, Hugh Dickins <hughd@google.com> wrote:
> >
> > I disagree with you there: I like the way Miklos made it symmetric,
> > I like the way delete_from_swap_cache drops the swap cache reference,
> > I dislike the way remove_from_page_cache does not - I did once try to
> > change that, but did a bad job, messed up reiserfs or reiser4 I forget
> > which, retreated in shame.
> 
> I agree symmetric is good. I just said current fact which is that
> remove_from_page_cache doesn't release ref.
> So I thought we have to match current semantic to protect confusing.
> Okay. I will not oppose current semantics.
> Instead of it, please add it (ex, caller should hold the page
> reference) in function description.
> 
> I am happy to hear that you tried it.
> Although it is hard, I think it's very valuable thing.
> Could you give me hint to googling your effort and why it is failed?

http://lkml.org/lkml/2004/10/24/140

Hugh

Re: [PATCH] mm: add replace_page_cache_page() function

[Hugh Dickins]

On Fri, 17 Dec 2010, Minchan Kim wrote:
> On Fri, Dec 17, 2010 at 10:21 AM, Hugh Dickins <hughd@google.com> wrote:
> >
> > I disagree with you there: I like the way Miklos made it symmetric,
> > I like the way delete_from_swap_cache drops the swap cache reference,
> > I dislike the way remove_from_page_cache does not - I did once try to
> > change that, but did a bad job, messed up reiserfs or reiser4 I forget
> > which, retreated in shame.
> 
> I agree symmetric is good. I just said current fact which is that
> remove_from_page_cache doesn't release ref.
> So I thought we have to match current semantic to protect confusing.
> Okay. I will not oppose current semantics.
> Instead of it, please add it (ex, caller should hold the page
> reference) in function description.
> 
> I am happy to hear that you tried it.
> Although it is hard, I think it's very valuable thing.
> Could you give me hint to googling your effort and why it is failed?

http://lkml.org/lkml/2004/10/24/140

Hugh

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[Minchan Kim]

On Fri, Dec 17, 2010 at 11:10 AM, Hugh Dickins <hughd@google.com> wrote:
> On Fri, 17 Dec 2010, Minchan Kim wrote:
>> On Fri, Dec 17, 2010 at 10:21 AM, Hugh Dickins <hughd@google.com> wrote:
>> >
>> > I disagree with you there: I like the way Miklos made it symmetric,
>> > I like the way delete_from_swap_cache drops the swap cache reference,
>> > I dislike the way remove_from_page_cache does not - I did once try to
>> > change that, but did a bad job, messed up reiserfs or reiser4 I forget
>> > which, retreated in shame.
>>
>> I agree symmetric is good. I just said current fact which is that
>> remove_from_page_cache doesn't release ref.
>> So I thought we have to match current semantic to protect confusing.
>> Okay. I will not oppose current semantics.
>> Instead of it, please add it (ex, caller should hold the page
>> reference) in function description.
>>
>> I am happy to hear that you tried it.
>> Although it is hard, I think it's very valuable thing.
>> Could you give me hint to googling your effort and why it is failed?
>
> http://lkml.org/lkml/2004/10/24/140

Thanks.

Now we have only 3 callers of remove_from_page_cache in mmtom.

1. truncate_huge_page
2. shmem_writepage
3. truncate_complete_page
4. fuse_try_move_page

It seems all of caller hold the page reference so It's ok to change
the semantic of remove_from_page_cache.
Okay. I will do that.

>
> Hugh
>



-- 
Kind regards,
Minchan Kim

Re: [PATCH] mm: add replace_page_cache_page() function

[Minchan Kim]

On Fri, Dec 17, 2010 at 11:10 AM, Hugh Dickins <hughd@google.com> wrote:
> On Fri, 17 Dec 2010, Minchan Kim wrote:
>> On Fri, Dec 17, 2010 at 10:21 AM, Hugh Dickins <hughd@google.com> wrote:
>> >
>> > I disagree with you there: I like the way Miklos made it symmetric,
>> > I like the way delete_from_swap_cache drops the swap cache reference,
>> > I dislike the way remove_from_page_cache does not - I did once try to
>> > change that, but did a bad job, messed up reiserfs or reiser4 I forget
>> > which, retreated in shame.
>>
>> I agree symmetric is good. I just said current fact which is that
>> remove_from_page_cache doesn't release ref.
>> So I thought we have to match current semantic to protect confusing.
>> Okay. I will not oppose current semantics.
>> Instead of it, please add it (ex, caller should hold the page
>> reference) in function description.
>>
>> I am happy to hear that you tried it.
>> Although it is hard, I think it's very valuable thing.
>> Could you give me hint to googling your effort and why it is failed?
>
> http://lkml.org/lkml/2004/10/24/140

Thanks.

Now we have only 3 callers of remove_from_page_cache in mmtom.

1. truncate_huge_page
2. shmem_writepage
3. truncate_complete_page
4. fuse_try_move_page

It seems all of caller hold the page reference so It's ok to change
the semantic of remove_from_page_cache.
Okay. I will do that.

>
> Hugh
>



-- 
Kind regards,
Minchan Kim

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[Miklos Szeredi]

On Fri, 17 Dec 2010, KAMEZAWA Hiroyuki wrote:
> No. memory cgroup expects all pages should be found on LRU. But, IIUC,
> pages on this radix-tree will not be on LRU. So, memory cgroup can't find
> it at destroying cgroup and can't reduce "usage" of resource to be 0.
> This makes rmdir() returns -EBUSY.

Oh, right.  Yes, the page will be on the LRU (it needs to be,
otherwise the VM coulnd't reclaim it).  After the
add_to_page_cache_locked is this:

	if (!(buf->flags & PIPE_BUF_FLAG_LRU))
		lru_cache_add_file(newpage);

It will add the page to the LRU, unless it's already on it.

Thanks,
Miklos

Re: [PATCH] mm: add replace_page_cache_page() function

[Miklos Szeredi]

On Fri, 17 Dec 2010, KAMEZAWA Hiroyuki wrote:
> No. memory cgroup expects all pages should be found on LRU. But, IIUC,
> pages on this radix-tree will not be on LRU. So, memory cgroup can't find
> it at destroying cgroup and can't reduce "usage" of resource to be 0.
> This makes rmdir() returns -EBUSY.

Oh, right.  Yes, the page will be on the LRU (it needs to be,
otherwise the VM coulnd't reclaim it).  After the
add_to_page_cache_locked is this:

	if (!(buf->flags & PIPE_BUF_FLAG_LRU))
		lru_cache_add_file(newpage);

It will add the page to the LRU, unless it's already on it.

Thanks,
Miklos

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[Miklos Szeredi]

On Fri, 17 Dec 2010, Minchan Kim wrote:
> >> >
> >> > I suspect it's historic that page_cache_release() doesn't drop the
> >> > page cache ref.
> >>
> >> Sorry I can't understand your words.
> >
> > Me neither: I believe Miklos meant __remove_from_page_cache() rather
> > than page_cache_release() in that instance.
> 
> Maybe. :)

Yeah, I did mean remove_from_page_cache :)

Thanks,
Miklos

Re: [PATCH] mm: add replace_page_cache_page() function

[Miklos Szeredi]

On Fri, 17 Dec 2010, Minchan Kim wrote:
> >> >
> >> > I suspect it's historic that page_cache_release() doesn't drop the
> >> > page cache ref.
> >>
> >> Sorry I can't understand your words.
> >
> > Me neither: I believe Miklos meant __remove_from_page_cache() rather
> > than page_cache_release() in that instance.
> 
> Maybe. :)

Yeah, I did mean remove_from_page_cache :)

Thanks,
Miklos

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[KAMEZAWA Hiroyuki]

On Fri, 17 Dec 2010 16:51:44 +0100
Miklos Szeredi <miklos@szeredi.hu> wrote:

> On Fri, 17 Dec 2010, KAMEZAWA Hiroyuki wrote:
> > No. memory cgroup expects all pages should be found on LRU. But, IIUC,
> > pages on this radix-tree will not be on LRU. So, memory cgroup can't find
> > it at destroying cgroup and can't reduce "usage" of resource to be 0.
> > This makes rmdir() returns -EBUSY.
> 
> Oh, right.  Yes, the page will be on the LRU (it needs to be,
> otherwise the VM coulnd't reclaim it).  After the
> add_to_page_cache_locked is this:
> 
> 	if (!(buf->flags & PIPE_BUF_FLAG_LRU))
> 		lru_cache_add_file(newpage);
> 
> It will add the page to the LRU, unless it's already on it.
> 

Thank you for clarification. 

Thanks,
-Kame

Re: [PATCH] mm: add replace_page_cache_page() function

[KAMEZAWA Hiroyuki]

On Fri, 17 Dec 2010 16:51:44 +0100
Miklos Szeredi <miklos@szeredi.hu> wrote:

> On Fri, 17 Dec 2010, KAMEZAWA Hiroyuki wrote:
> > No. memory cgroup expects all pages should be found on LRU. But, IIUC,
> > pages on this radix-tree will not be on LRU. So, memory cgroup can't find
> > it at destroying cgroup and can't reduce "usage" of resource to be 0.
> > This makes rmdir() returns -EBUSY.
> 
> Oh, right.  Yes, the page will be on the LRU (it needs to be,
> otherwise the VM coulnd't reclaim it).  After the
> add_to_page_cache_locked is this:
> 
> 	if (!(buf->flags & PIPE_BUF_FLAG_LRU))
> 		lru_cache_add_file(newpage);
> 
> It will add the page to the LRU, unless it's already on it.
> 

Thank you for clarification. 

Thanks,
-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[Miklos Szeredi]

On Tue, 11 Jan 2011, Daisuke Nishimura wrote:
> > What I recommend is below. (Please see the newest -mm because of a bug fix for
> > mem cgroup) Considering page management on radix-tree, it can be considerd as
> > a kind of page-migration, which replaces pages on radix-tree.
> > 
> > ==
> > 
> > > +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> > > +{
> > > +	int error;
> > > +
> > > +	VM_BUG_ON(!PageLocked(old));
> > > +	VM_BUG_ON(!PageLocked(new));
> > > +	VM_BUG_ON(new->mapping);
> > > +
> > 	struct mem_cgroup *memcg;
> > 
> I think it should be initialized to NULL.
> 
> > 	error = mem_cgroup_prepare_migration(old, new, &memcg);
> 
> I want some comments like:
> 
> 	/*
> 	 * This is not page migration, but prepare_migration and end_migration
> 	 * does enough work for charge replacement.
> 	 */
> 
> > 	#
> > 	# This function will charge against "newpage". But this expects
> > 	# the caller allows GFP_KERNEL gfp_mask. 
> > 	# After this, the newpage is in "charged" state.
> > 	if (error)
> > 		return -ENOMEM;
> > 
> > > +	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> > > +	if (!error) {
> > > +		struct address_space *mapping = old->mapping;
> > > +		pgoff_t offset = old->index;
> > > +
> > > +		page_cache_get(new);
> > > +		new->mapping = mapping;
> > > +		new->index = offset;
> > > +
> > > +		spin_lock_irq(&mapping->tree_lock);
> > > +		__remove_from_page_cache(old);
> > > +		error = radix_tree_insert(&mapping->page_tree, offset, new);
> > > +		BUG_ON(error);
> > > +		mapping->nrpages++;
> > > +		__inc_zone_page_state(new, NR_FILE_PAGES);
> > > +		if (PageSwapBacked(new))
> > > +			__inc_zone_page_state(new, NR_SHMEM);
> > > +		spin_unlock_irq(&mapping->tree_lock);
> > > +		radix_tree_preload_end();
> > 
> > > +		mem_cgroup_replace_cache_page(old, new); <== remove this.
> > 
> > 		mem_cgroup_end_migraton(memcg, old, new, true);
> > 
> > > +		page_cache_release(old);
> > > +	} 
> > 	else 
> > 		mem_cgroup_end_migration(memcg, old, new, false);
> > 
> > 	# Here, if the 4th argument is true, old page is uncharged.
> > 	# if the 4th argument is false, the new page is uncharged.
> > 	# Then, "charge" of the old page will be migrated onto the new page
> > 	# if replacement is done.
> > 
> > 
> > 
> > > +
> > > +	return error;
> > > +}
> > > +EXPORT_SYMBOL_GPL(replace_page_cache_page);
> > > +
> > 
> > ==
> > 
> > I think this is enough simple and this covers all memory cgroup's racy
> > problems.
> > 
> I agree.

Thanks for the comments.

Yeah, using existing infrastructure is undoubtedly simpler and less
prone to bugs.  So going with this for a first implementation might
do.

However, replace_page_cache_page() is meant to be very efficient,
otherwise any performance won by not copying the page contents are
lost to the cost of page replacement.  My guess is,
mem_cgroup_prepare_migration()/end_migration() are to heavyweight for
this.

Thanks,
Miklos

Re: [PATCH] mm: add replace_page_cache_page() function

[Miklos Szeredi]

On Tue, 11 Jan 2011, Daisuke Nishimura wrote:
> > What I recommend is below. (Please see the newest -mm because of a bug fix for
> > mem cgroup) Considering page management on radix-tree, it can be considerd as
> > a kind of page-migration, which replaces pages on radix-tree.
> > 
> > ==
> > 
> > > +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> > > +{
> > > +	int error;
> > > +
> > > +	VM_BUG_ON(!PageLocked(old));
> > > +	VM_BUG_ON(!PageLocked(new));
> > > +	VM_BUG_ON(new->mapping);
> > > +
> > 	struct mem_cgroup *memcg;
> > 
> I think it should be initialized to NULL.
> 
> > 	error = mem_cgroup_prepare_migration(old, new, &memcg);
> 
> I want some comments like:
> 
> 	/*
> 	 * This is not page migration, but prepare_migration and end_migration
> 	 * does enough work for charge replacement.
> 	 */
> 
> > 	#
> > 	# This function will charge against "newpage". But this expects
> > 	# the caller allows GFP_KERNEL gfp_mask. 
> > 	# After this, the newpage is in "charged" state.
> > 	if (error)
> > 		return -ENOMEM;
> > 
> > > +	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> > > +	if (!error) {
> > > +		struct address_space *mapping = old->mapping;
> > > +		pgoff_t offset = old->index;
> > > +
> > > +		page_cache_get(new);
> > > +		new->mapping = mapping;
> > > +		new->index = offset;
> > > +
> > > +		spin_lock_irq(&mapping->tree_lock);
> > > +		__remove_from_page_cache(old);
> > > +		error = radix_tree_insert(&mapping->page_tree, offset, new);
> > > +		BUG_ON(error);
> > > +		mapping->nrpages++;
> > > +		__inc_zone_page_state(new, NR_FILE_PAGES);
> > > +		if (PageSwapBacked(new))
> > > +			__inc_zone_page_state(new, NR_SHMEM);
> > > +		spin_unlock_irq(&mapping->tree_lock);
> > > +		radix_tree_preload_end();
> > 
> > > +		mem_cgroup_replace_cache_page(old, new); <== remove this.
> > 
> > 		mem_cgroup_end_migraton(memcg, old, new, true);
> > 
> > > +		page_cache_release(old);
> > > +	} 
> > 	else 
> > 		mem_cgroup_end_migration(memcg, old, new, false);
> > 
> > 	# Here, if the 4th argument is true, old page is uncharged.
> > 	# if the 4th argument is false, the new page is uncharged.
> > 	# Then, "charge" of the old page will be migrated onto the new page
> > 	# if replacement is done.
> > 
> > 
> > 
> > > +
> > > +	return error;
> > > +}
> > > +EXPORT_SYMBOL_GPL(replace_page_cache_page);
> > > +
> > 
> > ==
> > 
> > I think this is enough simple and this covers all memory cgroup's racy
> > problems.
> > 
> I agree.

Thanks for the comments.

Yeah, using existing infrastructure is undoubtedly simpler and less
prone to bugs.  So going with this for a first implementation might
do.

However, replace_page_cache_page() is meant to be very efficient,
otherwise any performance won by not copying the page contents are
lost to the cost of page replacement.  My guess is,
mem_cgroup_prepare_migration()/end_migration() are to heavyweight for
this.

Thanks,
Miklos

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[Miklos Szeredi]

On Tue, 11 Jan 2011, Daisuke Nishimura wrote:
> > What I recommend is below. (Please see the newest -mm because of a bug fix for
> > mem cgroup) Considering page management on radix-tree, it can be considerd as
> > a kind of page-migration, which replaces pages on radix-tree.
> > 
> > ==
> > 
> > > +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> > > +{
> > > +	int error;
> > > +
> > > +	VM_BUG_ON(!PageLocked(old));
> > > +	VM_BUG_ON(!PageLocked(new));
> > > +	VM_BUG_ON(new->mapping);
> > > +
> > 	struct mem_cgroup *memcg;
> > 
> I think it should be initialized to NULL.
> 
> > 	error = mem_cgroup_prepare_migration(old, new, &memcg);
> 
> I want some comments like:
> 
> 	/*
> 	 * This is not page migration, but prepare_migration and end_migration
> 	 * does enough work for charge replacement.
> 	 */
> 
> > 	#
> > 	# This function will charge against "newpage". But this expects
> > 	# the caller allows GFP_KERNEL gfp_mask. 
> > 	# After this, the newpage is in "charged" state.
> > 	if (error)
> > 		return -ENOMEM;
> > 
> > > +	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> > > +	if (!error) {
> > > +		struct address_space *mapping = old->mapping;
> > > +		pgoff_t offset = old->index;
> > > +
> > > +		page_cache_get(new);
> > > +		new->mapping = mapping;
> > > +		new->index = offset;
> > > +
> > > +		spin_lock_irq(&mapping->tree_lock);
> > > +		__remove_from_page_cache(old);
> > > +		error = radix_tree_insert(&mapping->page_tree, offset, new);
> > > +		BUG_ON(error);
> > > +		mapping->nrpages++;
> > > +		__inc_zone_page_state(new, NR_FILE_PAGES);
> > > +		if (PageSwapBacked(new))
> > > +			__inc_zone_page_state(new, NR_SHMEM);
> > > +		spin_unlock_irq(&mapping->tree_lock);
> > > +		radix_tree_preload_end();
> > 
> > > +		mem_cgroup_replace_cache_page(old, new); <== remove this.
> > 
> > 		mem_cgroup_end_migraton(memcg, old, new, true);
> > 
> > > +		page_cache_release(old);
> > > +	} 
> > 	else 
> > 		mem_cgroup_end_migration(memcg, old, new, false);
> > 
> > 	# Here, if the 4th argument is true, old page is uncharged.
> > 	# if the 4th argument is false, the new page is uncharged.
> > 	# Then, "charge" of the old page will be migrated onto the new page
> > 	# if replacement is done.
> > 
> > 
> > 
> > > +
> > > +	return error;
> > > +}
> > > +EXPORT_SYMBOL_GPL(replace_page_cache_page);
> > > +
> > 
> > ==
> > 
> > I think this is enough simple and this covers all memory cgroup's racy
> > problems.
> > 
> I agree.

Thanks for the comments.

Yeah, using existing infrastructure is undoubtedly simpler and less
prone to bugs.  So going with this for a first implementation might
do.

However, replace_page_cache_page() is meant to be very efficient,
otherwise any performance won by not copying the page contents are
lost to the cost of page replacement.  My guess is,
mem_cgroup_prepare_migration()/end_migration() are to heavyweight for
this.

Thanks,
Miklos

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[Minchan Kim]

Hi Miklos,

On Fri, Jan 07, 2011 at 07:22:41PM +0100, Miklos Szeredi wrote:
> Here's an updated patch, addressing the review comments.
> 
> Hiroyuki-san, can you please review the newly introduced
> mem_cgroup_replace_cache_page(), as I'm not fully familiar with the
> memory cgroup code.
> 
> Thanks,
> Miklos
> ---
> 
> From: Miklos Szeredi <mszeredi@suse.cz>
> Subject: mm: add replace_page_cache_page() function
> 
> This function basically does:
> 
>      remove_from_page_cache(old);
>      page_cache_release(old);
>      add_to_page_cache_locked(new);
> 
> Except it does this atomically, so there's no possibility for the
> "add" to fail because of a race.
> 
> This is used by fuse to move pages into the page cache.
> 
> Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
> ---
>  fs/fuse/dev.c              |   10 +++------
>  include/linux/memcontrol.h |    8 +++++++
>  include/linux/pagemap.h    |    1 
>  mm/filemap.c               |   50 +++++++++++++++++++++++++++++++++++++++++++++
>  mm/memcontrol.c            |   38 ++++++++++++++++++++++++++++++++++
>  5 files changed, 101 insertions(+), 6 deletions(-)
> 
> Index: linux-2.6/mm/filemap.c
> ===================================================================
> --- linux-2.6.orig/mm/filemap.c	2011-01-07 17:53:39.000000000 +0100
> +++ linux-2.6/mm/filemap.c	2011-01-07 19:14:45.000000000 +0100
> @@ -390,6 +390,56 @@ int filemap_write_and_wait_range(struct
>  EXPORT_SYMBOL(filemap_write_and_wait_range);
>  
>  /**
> + * replace_page_cache_page - replace a pagecache page with a new one
> + * @old:	page to be replaced
> + * @new:	page to replace with
> + * @gfp_mask:	page allocation mode
> + *
> + * This function replaces a page in the pagecache with a new one.  On
> + * success it acquires the pagecache reference for the new page and
> + * drop it for the old page.  Both the old and new pages must be
> + * locked.  This function does not add the new page to the LRU, the
> + * caller must do that.
> + *
> + * The remove + add is atomic.  The only way this function can fail is
> + * memory allocation failure.
> + */
> +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> +{
> +	int error;
> +
> +	VM_BUG_ON(!PageLocked(old));
> +	VM_BUG_ON(!PageLocked(new));
> +	VM_BUG_ON(new->mapping);
> +
> +	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> +	if (!error) {
> +		struct address_space *mapping = old->mapping;
> +		pgoff_t offset = old->index;
> +
> +		page_cache_get(new);
> +		new->mapping = mapping;
> +		new->index = offset;
> +
> +		spin_lock_irq(&mapping->tree_lock);
> +		__remove_from_page_cache(old);

Since v1, I tried to change page reference accouting semantics of
remove_from_page_cache. At last, it changes API name from __remove_from_page_cache
to __delete_from_page_cache.
So this conflicts my series.
If you are not urgent, could you resend based on my series after Andrew merges it?
Or if Andrew has a concern about my series and he merged your patch, I will resend
my series include fixing this part of your patch.

Please wait Andrew's next step.

-- 
Kind regards,
Minchan Kim

Re: [PATCH] mm: add replace_page_cache_page() function

[Minchan Kim]

Hi Miklos,

On Fri, Jan 07, 2011 at 07:22:41PM +0100, Miklos Szeredi wrote:
> Here's an updated patch, addressing the review comments.
> 
> Hiroyuki-san, can you please review the newly introduced
> mem_cgroup_replace_cache_page(), as I'm not fully familiar with the
> memory cgroup code.
> 
> Thanks,
> Miklos
> ---
> 
> From: Miklos Szeredi <mszeredi@suse.cz>
> Subject: mm: add replace_page_cache_page() function
> 
> This function basically does:
> 
>      remove_from_page_cache(old);
>      page_cache_release(old);
>      add_to_page_cache_locked(new);
> 
> Except it does this atomically, so there's no possibility for the
> "add" to fail because of a race.
> 
> This is used by fuse to move pages into the page cache.
> 
> Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
> ---
>  fs/fuse/dev.c              |   10 +++------
>  include/linux/memcontrol.h |    8 +++++++
>  include/linux/pagemap.h    |    1 
>  mm/filemap.c               |   50 +++++++++++++++++++++++++++++++++++++++++++++
>  mm/memcontrol.c            |   38 ++++++++++++++++++++++++++++++++++
>  5 files changed, 101 insertions(+), 6 deletions(-)
> 
> Index: linux-2.6/mm/filemap.c
> ===================================================================
> --- linux-2.6.orig/mm/filemap.c	2011-01-07 17:53:39.000000000 +0100
> +++ linux-2.6/mm/filemap.c	2011-01-07 19:14:45.000000000 +0100
> @@ -390,6 +390,56 @@ int filemap_write_and_wait_range(struct
>  EXPORT_SYMBOL(filemap_write_and_wait_range);
>  
>  /**
> + * replace_page_cache_page - replace a pagecache page with a new one
> + * @old:	page to be replaced
> + * @new:	page to replace with
> + * @gfp_mask:	page allocation mode
> + *
> + * This function replaces a page in the pagecache with a new one.  On
> + * success it acquires the pagecache reference for the new page and
> + * drop it for the old page.  Both the old and new pages must be
> + * locked.  This function does not add the new page to the LRU, the
> + * caller must do that.
> + *
> + * The remove + add is atomic.  The only way this function can fail is
> + * memory allocation failure.
> + */
> +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> +{
> +	int error;
> +
> +	VM_BUG_ON(!PageLocked(old));
> +	VM_BUG_ON(!PageLocked(new));
> +	VM_BUG_ON(new->mapping);
> +
> +	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> +	if (!error) {
> +		struct address_space *mapping = old->mapping;
> +		pgoff_t offset = old->index;
> +
> +		page_cache_get(new);
> +		new->mapping = mapping;
> +		new->index = offset;
> +
> +		spin_lock_irq(&mapping->tree_lock);
> +		__remove_from_page_cache(old);

Since v1, I tried to change page reference accouting semantics of
remove_from_page_cache. At last, it changes API name from __remove_from_page_cache
to __delete_from_page_cache.
So this conflicts my series.
If you are not urgent, could you resend based on my series after Andrew merges it?
Or if Andrew has a concern about my series and he merged your patch, I will resend
my series include fixing this part of your patch.

Please wait Andrew's next step.

-- 
Kind regards,
Minchan Kim

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[Daisuke Nishimura]

> What I recommend is below. (Please see the newest -mm because of a bug fix for
> mem cgroup) Considering page management on radix-tree, it can be considerd as
> a kind of page-migration, which replaces pages on radix-tree.
> 
> ==
> 
> > +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> > +{
> > +	int error;
> > +
> > +	VM_BUG_ON(!PageLocked(old));
> > +	VM_BUG_ON(!PageLocked(new));
> > +	VM_BUG_ON(new->mapping);
> > +
> 	struct mem_cgroup *memcg;
> 
I think it should be initialized to NULL.

> 	error = mem_cgroup_prepare_migration(old, new, &memcg);

I want some comments like:

	/*
	 * This is not page migration, but prepare_migration and end_migration
	 * does enough work for charge replacement.
	 */

> 	#
> 	# This function will charge against "newpage". But this expects
> 	# the caller allows GFP_KERNEL gfp_mask. 
> 	# After this, the newpage is in "charged" state.
> 	if (error)
> 		return -ENOMEM;
> 
> > +	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> > +	if (!error) {
> > +		struct address_space *mapping = old->mapping;
> > +		pgoff_t offset = old->index;
> > +
> > +		page_cache_get(new);
> > +		new->mapping = mapping;
> > +		new->index = offset;
> > +
> > +		spin_lock_irq(&mapping->tree_lock);
> > +		__remove_from_page_cache(old);
> > +		error = radix_tree_insert(&mapping->page_tree, offset, new);
> > +		BUG_ON(error);
> > +		mapping->nrpages++;
> > +		__inc_zone_page_state(new, NR_FILE_PAGES);
> > +		if (PageSwapBacked(new))
> > +			__inc_zone_page_state(new, NR_SHMEM);
> > +		spin_unlock_irq(&mapping->tree_lock);
> > +		radix_tree_preload_end();
> 
> > +		mem_cgroup_replace_cache_page(old, new); <== remove this.
> 
> 		mem_cgroup_end_migraton(memcg, old, new, true);
> 
> > +		page_cache_release(old);
> > +	} 
> 	else 
> 		mem_cgroup_end_migration(memcg, old, new, false);
> 
> 	# Here, if the 4th argument is true, old page is uncharged.
> 	# if the 4th argument is false, the new page is uncharged.
> 	# Then, "charge" of the old page will be migrated onto the new page
> 	# if replacement is done.
> 
> 
> 
> > +
> > +	return error;
> > +}
> > +EXPORT_SYMBOL_GPL(replace_page_cache_page);
> > +
> 
> ==
> 
> I think this is enough simple and this covers all memory cgroup's racy
> problems.
> 
I agree.

Thanks,
Daisuke Nishimura.

Re: [PATCH] mm: add replace_page_cache_page() function

[Daisuke Nishimura]

> What I recommend is below. (Please see the newest -mm because of a bug fix for
> mem cgroup) Considering page management on radix-tree, it can be considerd as
> a kind of page-migration, which replaces pages on radix-tree.
> 
> ==
> 
> > +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> > +{
> > +	int error;
> > +
> > +	VM_BUG_ON(!PageLocked(old));
> > +	VM_BUG_ON(!PageLocked(new));
> > +	VM_BUG_ON(new->mapping);
> > +
> 	struct mem_cgroup *memcg;
> 
I think it should be initialized to NULL.

> 	error = mem_cgroup_prepare_migration(old, new, &memcg);

I want some comments like:

	/*
	 * This is not page migration, but prepare_migration and end_migration
	 * does enough work for charge replacement.
	 */

> 	#
> 	# This function will charge against "newpage". But this expects
> 	# the caller allows GFP_KERNEL gfp_mask. 
> 	# After this, the newpage is in "charged" state.
> 	if (error)
> 		return -ENOMEM;
> 
> > +	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> > +	if (!error) {
> > +		struct address_space *mapping = old->mapping;
> > +		pgoff_t offset = old->index;
> > +
> > +		page_cache_get(new);
> > +		new->mapping = mapping;
> > +		new->index = offset;
> > +
> > +		spin_lock_irq(&mapping->tree_lock);
> > +		__remove_from_page_cache(old);
> > +		error = radix_tree_insert(&mapping->page_tree, offset, new);
> > +		BUG_ON(error);
> > +		mapping->nrpages++;
> > +		__inc_zone_page_state(new, NR_FILE_PAGES);
> > +		if (PageSwapBacked(new))
> > +			__inc_zone_page_state(new, NR_SHMEM);
> > +		spin_unlock_irq(&mapping->tree_lock);
> > +		radix_tree_preload_end();
> 
> > +		mem_cgroup_replace_cache_page(old, new); <== remove this.
> 
> 		mem_cgroup_end_migraton(memcg, old, new, true);
> 
> > +		page_cache_release(old);
> > +	} 
> 	else 
> 		mem_cgroup_end_migration(memcg, old, new, false);
> 
> 	# Here, if the 4th argument is true, old page is uncharged.
> 	# if the 4th argument is false, the new page is uncharged.
> 	# Then, "charge" of the old page will be migrated onto the new page
> 	# if replacement is done.
> 
> 
> 
> > +
> > +	return error;
> > +}
> > +EXPORT_SYMBOL_GPL(replace_page_cache_page);
> > +
> 
> ==
> 
> I think this is enough simple and this covers all memory cgroup's racy
> problems.
> 
I agree.

Thanks,
Daisuke Nishimura.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: add replace_page_cache_page() function

[KAMEZAWA Hiroyuki]

On Fri, 07 Jan 2011 19:22:41 +0100
Miklos Szeredi <miklos@szeredi.hu> wrote:

> Here's an updated patch, addressing the review comments.
> 
> Hiroyuki-san, can you please review the newly introduced
> mem_cgroup_replace_cache_page(), as I'm not fully familiar with the
> memory cgroup code.
> 

Ok. Ccing Nishimura. see below.

> Thanks,
> Miklos
> ---
> 
> From: Miklos Szeredi <mszeredi@suse.cz>
> Subject: mm: add replace_page_cache_page() function
> 
> This function basically does:
> 
>      remove_from_page_cache(old);
>      page_cache_release(old);
>      add_to_page_cache_locked(new);
> 
> Except it does this atomically, so there's no possibility for the
> "add" to fail because of a race.
> 
> This is used by fuse to move pages into the page cache.
> 
> Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
> ---
>  fs/fuse/dev.c              |   10 +++------
>  include/linux/memcontrol.h |    8 +++++++
>  include/linux/pagemap.h    |    1 
>  mm/filemap.c               |   50 +++++++++++++++++++++++++++++++++++++++++++++
>  mm/memcontrol.c            |   38 ++++++++++++++++++++++++++++++++++
>  5 files changed, 101 insertions(+), 6 deletions(-)
> 
> Index: linux-2.6/mm/filemap.c
> ===================================================================
> --- linux-2.6.orig/mm/filemap.c	2011-01-07 17:53:39.000000000 +0100
> +++ linux-2.6/mm/filemap.c	2011-01-07 19:14:45.000000000 +0100
> @@ -390,6 +390,56 @@ int filemap_write_and_wait_range(struct
>  EXPORT_SYMBOL(filemap_write_and_wait_range);
>  
>  /**
> + * replace_page_cache_page - replace a pagecache page with a new one
> + * @old:	page to be replaced
> + * @new:	page to replace with
> + * @gfp_mask:	page allocation mode
> + *
> + * This function replaces a page in the pagecache with a new one.  On
> + * success it acquires the pagecache reference for the new page and
> + * drop it for the old page.  Both the old and new pages must be
> + * locked.  This function does not add the new page to the LRU, the
> + * caller must do that.
> + *
> + * The remove + add is atomic.  The only way this function can fail is
> + * memory allocation failure.
> + */
> +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> +{
> +	int error;
> +
> +	VM_BUG_ON(!PageLocked(old));
> +	VM_BUG_ON(!PageLocked(new));
> +	VM_BUG_ON(new->mapping);
> +
> +	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> +	if (!error) {
> +		struct address_space *mapping = old->mapping;
> +		pgoff_t offset = old->index;
> +
> +		page_cache_get(new);
> +		new->mapping = mapping;
> +		new->index = offset;
> +
> +		spin_lock_irq(&mapping->tree_lock);
> +		__remove_from_page_cache(old);
> +		error = radix_tree_insert(&mapping->page_tree, offset, new);
> +		BUG_ON(error);
> +		mapping->nrpages++;
> +		__inc_zone_page_state(new, NR_FILE_PAGES);
> +		if (PageSwapBacked(new))
> +			__inc_zone_page_state(new, NR_SHMEM);
> +		spin_unlock_irq(&mapping->tree_lock);
> +		radix_tree_preload_end();
> +		mem_cgroup_replace_cache_page(old, new);
> +		page_cache_release(old);
> +	}
> +
> +	return error;
> +}
> +EXPORT_SYMBOL_GPL(replace_page_cache_page);
> +
> +/**
>   * add_to_page_cache_locked - add a locked page to the pagecache
>   * @page:	page to add
>   * @mapping:	the page's address_space
> Index: linux-2.6/include/linux/pagemap.h
> ===================================================================
> --- linux-2.6.orig/include/linux/pagemap.h	2011-01-07 17:53:39.000000000 +0100
> +++ linux-2.6/include/linux/pagemap.h	2011-01-07 19:14:45.000000000 +0100
> @@ -457,6 +457,7 @@ int add_to_page_cache_lru(struct page *p
>  				pgoff_t index, gfp_t gfp_mask);
>  extern void remove_from_page_cache(struct page *page);
>  extern void __remove_from_page_cache(struct page *page);
> +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
>  
>  /*
>   * Like add_to_page_cache_locked, but used to add newly allocated pages:
> Index: linux-2.6/fs/fuse/dev.c
> ===================================================================
> --- linux-2.6.orig/fs/fuse/dev.c	2011-01-07 17:53:39.000000000 +0100
> +++ linux-2.6/fs/fuse/dev.c	2011-01-07 19:14:45.000000000 +0100
> @@ -737,14 +737,12 @@ static int fuse_try_move_page(struct fus
>  	if (WARN_ON(PageMlocked(oldpage)))
>  		goto out_fallback_unlock;
>  
> -	remove_from_page_cache(oldpage);
> -	page_cache_release(oldpage);
> -
> -	err = add_to_page_cache_locked(newpage, mapping, index, GFP_KERNEL);
> +	err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
>  	if (err) {
> -		printk(KERN_WARNING "fuse_try_move_page: failed to add page");
> -		goto out_fallback_unlock;
> +		unlock_page(newpage);
> +		return err;
>  	}
> +
>  	page_cache_get(newpage);
>  
>  	if (!(buf->flags & PIPE_BUF_FLAG_LRU))
> Index: linux-2.6/include/linux/memcontrol.h
> ===================================================================
> --- linux-2.6.orig/include/linux/memcontrol.h	2011-01-07 17:53:39.000000000 +0100
> +++ linux-2.6/include/linux/memcontrol.h	2011-01-07 19:14:45.000000000 +0100
> @@ -95,6 +95,9 @@ mem_cgroup_prepare_migration(struct page
>  extern void mem_cgroup_end_migration(struct mem_cgroup *mem,
>  	struct page *oldpage, struct page *newpage);
>  
> +extern void mem_cgroup_replace_cache_page(struct page *oldpage,
> +					  struct page *newpage);
> +
>  /*
>   * For memory reclaim.
>   */
> @@ -236,6 +239,11 @@ static inline void mem_cgroup_end_migrat
>  {
>  }
>  
> +static inline void mem_cgroup_replace_cache_page(struct page *oldpage,
> +					 	struct page *newpage)
> +{
> +}
> +
>  static inline int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
>  {
>  	return 0;
> Index: linux-2.6/mm/memcontrol.c
> ===================================================================
> --- linux-2.6.orig/mm/memcontrol.c	2011-01-07 17:53:39.000000000 +0100
> +++ linux-2.6/mm/memcontrol.c	2011-01-07 19:20:41.000000000 +0100
> @@ -2905,6 +2905,44 @@ void mem_cgroup_end_migration(struct mem
>  }
>  
>  /*
> + * This function moves the charge from oldpage to newpage.  The new
> + * page must not be already charged.
> + */
> +void mem_cgroup_replace_cache_page(struct page *oldpage, struct page *newpage)
> +{
> +	struct page_cgroup *old_pc;
> +	struct page_cgroup *new_pc;
> +	struct mem_cgroup *mem;
> +
> +	if (mem_cgroup_disabled())
> +		return;
> +
> +	old_pc = lookup_page_cgroup(oldpage);
> +	lock_page_cgroup(old_pc);
> +	if (!PageCgroupUsed(old_pc)) {
> +		unlock_page_cgroup(old_pc);
> +		return;
> +	}
> +
> +	mem = old_pc->mem_cgroup;
> +	css_get(&mem->css);
> +	ClearPageCgroupUsed(old_pc);
> +	unlock_page_cgroup(old_pc);
> +
> +	new_pc = lookup_page_cgroup(newpage);
> +	lock_page_cgroup(new_pc);
> +	BUG_ON(PageCgroupUsed(new_pc));
> +
> +	new_pc->mem_cgroup = mem;
> +	smp_wmb();
> +	SetPageCgroupCache(new_pc);
> +	SetPageCgroupUsed(new_pc);
> +	unlock_page_cgroup(new_pc);
> +	css_put(&mem->css);
> +}

I think some of moving flags are lacked and this new function is not necessary.


What I recommend is below. (Please see the newest -mm because of a bug fix for
mem cgroup) Considering page management on radix-tree, it can be considerd as
a kind of page-migration, which replaces pages on radix-tree.

==

> +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> +{
> +	int error;
> +
> +	VM_BUG_ON(!PageLocked(old));
> +	VM_BUG_ON(!PageLocked(new));
> +	VM_BUG_ON(new->mapping);
> +
	struct mem_cgroup *memcg;

	error = mem_cgroup_prepare_migration(old, new, &memcg);
	#
	# This function will charge against "newpage". But this expects
	# the caller allows GFP_KERNEL gfp_mask. 
	# After this, the newpage is in "charged" state.
	if (error)
		return -ENOMEM;

> +	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> +	if (!error) {
> +		struct address_space *mapping = old->mapping;
> +		pgoff_t offset = old->index;
> +
> +		page_cache_get(new);
> +		new->mapping = mapping;
> +		new->index = offset;
> +
> +		spin_lock_irq(&mapping->tree_lock);
> +		__remove_from_page_cache(old);
> +		error = radix_tree_insert(&mapping->page_tree, offset, new);
> +		BUG_ON(error);
> +		mapping->nrpages++;
> +		__inc_zone_page_state(new, NR_FILE_PAGES);
> +		if (PageSwapBacked(new))
> +			__inc_zone_page_state(new, NR_SHMEM);
> +		spin_unlock_irq(&mapping->tree_lock);
> +		radix_tree_preload_end();

> +		mem_cgroup_replace_cache_page(old, new); <== remove this.

		mem_cgroup_end_migraton(memcg, old, new, true);

> +		page_cache_release(old);
> +	} 
	else 
		mem_cgroup_end_migration(memcg, old, new, false);

	# Here, if the 4th argument is true, old page is uncharged.
	# if the 4th argument is false, the new page is uncharged.
	# Then, "charge" of the old page will be migrated onto the new page
	# if replacement is done.



> +
> +	return error;
> +}
> +EXPORT_SYMBOL_GPL(replace_page_cache_page);
> +

==

I think this is enough simple and this covers all memory cgroup's racy
problems.

Thanks,
-Kame

Re: [PATCH] mm: add replace_page_cache_page() function

[KAMEZAWA Hiroyuki]

On Fri, 07 Jan 2011 19:22:41 +0100
Miklos Szeredi <miklos@szeredi.hu> wrote:

> Here's an updated patch, addressing the review comments.
> 
> Hiroyuki-san, can you please review the newly introduced
> mem_cgroup_replace_cache_page(), as I'm not fully familiar with the
> memory cgroup code.
> 

Ok. Ccing Nishimura. see below.

> Thanks,
> Miklos
> ---
> 
> From: Miklos Szeredi <mszeredi@suse.cz>
> Subject: mm: add replace_page_cache_page() function
> 
> This function basically does:
> 
>      remove_from_page_cache(old);
>      page_cache_release(old);
>      add_to_page_cache_locked(new);
> 
> Except it does this atomically, so there's no possibility for the
> "add" to fail because of a race.
> 
> This is used by fuse to move pages into the page cache.
> 
> Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
> ---
>  fs/fuse/dev.c              |   10 +++------
>  include/linux/memcontrol.h |    8 +++++++
>  include/linux/pagemap.h    |    1 
>  mm/filemap.c               |   50 +++++++++++++++++++++++++++++++++++++++++++++
>  mm/memcontrol.c            |   38 ++++++++++++++++++++++++++++++++++
>  5 files changed, 101 insertions(+), 6 deletions(-)
> 
> Index: linux-2.6/mm/filemap.c
> ===================================================================
> --- linux-2.6.orig/mm/filemap.c	2011-01-07 17:53:39.000000000 +0100
> +++ linux-2.6/mm/filemap.c	2011-01-07 19:14:45.000000000 +0100
> @@ -390,6 +390,56 @@ int filemap_write_and_wait_range(struct
>  EXPORT_SYMBOL(filemap_write_and_wait_range);
>  
>  /**
> + * replace_page_cache_page - replace a pagecache page with a new one
> + * @old:	page to be replaced
> + * @new:	page to replace with
> + * @gfp_mask:	page allocation mode
> + *
> + * This function replaces a page in the pagecache with a new one.  On
> + * success it acquires the pagecache reference for the new page and
> + * drop it for the old page.  Both the old and new pages must be
> + * locked.  This function does not add the new page to the LRU, the
> + * caller must do that.
> + *
> + * The remove + add is atomic.  The only way this function can fail is
> + * memory allocation failure.
> + */
> +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> +{
> +	int error;
> +
> +	VM_BUG_ON(!PageLocked(old));
> +	VM_BUG_ON(!PageLocked(new));
> +	VM_BUG_ON(new->mapping);
> +
> +	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> +	if (!error) {
> +		struct address_space *mapping = old->mapping;
> +		pgoff_t offset = old->index;
> +
> +		page_cache_get(new);
> +		new->mapping = mapping;
> +		new->index = offset;
> +
> +		spin_lock_irq(&mapping->tree_lock);
> +		__remove_from_page_cache(old);
> +		error = radix_tree_insert(&mapping->page_tree, offset, new);
> +		BUG_ON(error);
> +		mapping->nrpages++;
> +		__inc_zone_page_state(new, NR_FILE_PAGES);
> +		if (PageSwapBacked(new))
> +			__inc_zone_page_state(new, NR_SHMEM);
> +		spin_unlock_irq(&mapping->tree_lock);
> +		radix_tree_preload_end();
> +		mem_cgroup_replace_cache_page(old, new);
> +		page_cache_release(old);
> +	}
> +
> +	return error;
> +}
> +EXPORT_SYMBOL_GPL(replace_page_cache_page);
> +
> +/**
>   * add_to_page_cache_locked - add a locked page to the pagecache
>   * @page:	page to add
>   * @mapping:	the page's address_space
> Index: linux-2.6/include/linux/pagemap.h
> ===================================================================
> --- linux-2.6.orig/include/linux/pagemap.h	2011-01-07 17:53:39.000000000 +0100
> +++ linux-2.6/include/linux/pagemap.h	2011-01-07 19:14:45.000000000 +0100
> @@ -457,6 +457,7 @@ int add_to_page_cache_lru(struct page *p
>  				pgoff_t index, gfp_t gfp_mask);
>  extern void remove_from_page_cache(struct page *page);
>  extern void __remove_from_page_cache(struct page *page);
> +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
>  
>  /*
>   * Like add_to_page_cache_locked, but used to add newly allocated pages:
> Index: linux-2.6/fs/fuse/dev.c
> ===================================================================
> --- linux-2.6.orig/fs/fuse/dev.c	2011-01-07 17:53:39.000000000 +0100
> +++ linux-2.6/fs/fuse/dev.c	2011-01-07 19:14:45.000000000 +0100
> @@ -737,14 +737,12 @@ static int fuse_try_move_page(struct fus
>  	if (WARN_ON(PageMlocked(oldpage)))
>  		goto out_fallback_unlock;
>  
> -	remove_from_page_cache(oldpage);
> -	page_cache_release(oldpage);
> -
> -	err = add_to_page_cache_locked(newpage, mapping, index, GFP_KERNEL);
> +	err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
>  	if (err) {
> -		printk(KERN_WARNING "fuse_try_move_page: failed to add page");
> -		goto out_fallback_unlock;
> +		unlock_page(newpage);
> +		return err;
>  	}
> +
>  	page_cache_get(newpage);
>  
>  	if (!(buf->flags & PIPE_BUF_FLAG_LRU))
> Index: linux-2.6/include/linux/memcontrol.h
> ===================================================================
> --- linux-2.6.orig/include/linux/memcontrol.h	2011-01-07 17:53:39.000000000 +0100
> +++ linux-2.6/include/linux/memcontrol.h	2011-01-07 19:14:45.000000000 +0100
> @@ -95,6 +95,9 @@ mem_cgroup_prepare_migration(struct page
>  extern void mem_cgroup_end_migration(struct mem_cgroup *mem,
>  	struct page *oldpage, struct page *newpage);
>  
> +extern void mem_cgroup_replace_cache_page(struct page *oldpage,
> +					  struct page *newpage);
> +
>  /*
>   * For memory reclaim.
>   */
> @@ -236,6 +239,11 @@ static inline void mem_cgroup_end_migrat
>  {
>  }
>  
> +static inline void mem_cgroup_replace_cache_page(struct page *oldpage,
> +					 	struct page *newpage)
> +{
> +}
> +
>  static inline int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
>  {
>  	return 0;
> Index: linux-2.6/mm/memcontrol.c
> ===================================================================
> --- linux-2.6.orig/mm/memcontrol.c	2011-01-07 17:53:39.000000000 +0100
> +++ linux-2.6/mm/memcontrol.c	2011-01-07 19:20:41.000000000 +0100
> @@ -2905,6 +2905,44 @@ void mem_cgroup_end_migration(struct mem
>  }
>  
>  /*
> + * This function moves the charge from oldpage to newpage.  The new
> + * page must not be already charged.
> + */
> +void mem_cgroup_replace_cache_page(struct page *oldpage, struct page *newpage)
> +{
> +	struct page_cgroup *old_pc;
> +	struct page_cgroup *new_pc;
> +	struct mem_cgroup *mem;
> +
> +	if (mem_cgroup_disabled())
> +		return;
> +
> +	old_pc = lookup_page_cgroup(oldpage);
> +	lock_page_cgroup(old_pc);
> +	if (!PageCgroupUsed(old_pc)) {
> +		unlock_page_cgroup(old_pc);
> +		return;
> +	}
> +
> +	mem = old_pc->mem_cgroup;
> +	css_get(&mem->css);
> +	ClearPageCgroupUsed(old_pc);
> +	unlock_page_cgroup(old_pc);
> +
> +	new_pc = lookup_page_cgroup(newpage);
> +	lock_page_cgroup(new_pc);
> +	BUG_ON(PageCgroupUsed(new_pc));
> +
> +	new_pc->mem_cgroup = mem;
> +	smp_wmb();
> +	SetPageCgroupCache(new_pc);
> +	SetPageCgroupUsed(new_pc);
> +	unlock_page_cgroup(new_pc);
> +	css_put(&mem->css);
> +}

I think some of moving flags are lacked and this new function is not necessary.


What I recommend is below. (Please see the newest -mm because of a bug fix for
mem cgroup) Considering page management on radix-tree, it can be considerd as
a kind of page-migration, which replaces pages on radix-tree.

==

> +int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> +{
> +	int error;
> +
> +	VM_BUG_ON(!PageLocked(old));
> +	VM_BUG_ON(!PageLocked(new));
> +	VM_BUG_ON(new->mapping);
> +
	struct mem_cgroup *memcg;

	error = mem_cgroup_prepare_migration(old, new, &memcg);
	#
	# This function will charge against "newpage". But this expects
	# the caller allows GFP_KERNEL gfp_mask. 
	# After this, the newpage is in "charged" state.
	if (error)
		return -ENOMEM;

> +	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
> +	if (!error) {
> +		struct address_space *mapping = old->mapping;
> +		pgoff_t offset = old->index;
> +
> +		page_cache_get(new);
> +		new->mapping = mapping;
> +		new->index = offset;
> +
> +		spin_lock_irq(&mapping->tree_lock);
> +		__remove_from_page_cache(old);
> +		error = radix_tree_insert(&mapping->page_tree, offset, new);
> +		BUG_ON(error);
> +		mapping->nrpages++;
> +		__inc_zone_page_state(new, NR_FILE_PAGES);
> +		if (PageSwapBacked(new))
> +			__inc_zone_page_state(new, NR_SHMEM);
> +		spin_unlock_irq(&mapping->tree_lock);
> +		radix_tree_preload_end();

> +		mem_cgroup_replace_cache_page(old, new); <== remove this.

		mem_cgroup_end_migraton(memcg, old, new, true);

> +		page_cache_release(old);
> +	} 
	else 
		mem_cgroup_end_migration(memcg, old, new, false);

	# Here, if the 4th argument is true, old page is uncharged.
	# if the 4th argument is false, the new page is uncharged.
	# Then, "charge" of the old page will be migrated onto the new page
	# if replacement is done.



> +
> +	return error;
> +}
> +EXPORT_SYMBOL_GPL(replace_page_cache_page);
> +

==

I think this is enough simple and this covers all memory cgroup's racy
problems.

Thanks,
-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[PATCH] mm: add replace_page_cache_page() function

[Miklos Szeredi]

Here's an updated patch, addressing the review comments.

Hiroyuki-san, can you please review the newly introduced
mem_cgroup_replace_cache_page(), as I'm not fully familiar with the
memory cgroup code.

Thanks,
Miklos
---

From: Miklos Szeredi <mszeredi@suse.cz>
Subject: mm: add replace_page_cache_page() function

This function basically does:

     remove_from_page_cache(old);
     page_cache_release(old);
     add_to_page_cache_locked(new);

Except it does this atomically, so there's no possibility for the
"add" to fail because of a race.

This is used by fuse to move pages into the page cache.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
---
 fs/fuse/dev.c              |   10 +++------
 include/linux/memcontrol.h |    8 +++++++
 include/linux/pagemap.h    |    1 
 mm/filemap.c               |   50 +++++++++++++++++++++++++++++++++++++++++++++
 mm/memcontrol.c            |   38 ++++++++++++++++++++++++++++++++++
 5 files changed, 101 insertions(+), 6 deletions(-)

Index: linux-2.6/mm/filemap.c
===================================================================
--- linux-2.6.orig/mm/filemap.c	2011-01-07 17:53:39.000000000 +0100
+++ linux-2.6/mm/filemap.c	2011-01-07 19:14:45.000000000 +0100
@@ -390,6 +390,56 @@ int filemap_write_and_wait_range(struct
 EXPORT_SYMBOL(filemap_write_and_wait_range);
 
 /**
+ * replace_page_cache_page - replace a pagecache page with a new one
+ * @old:	page to be replaced
+ * @new:	page to replace with
+ * @gfp_mask:	page allocation mode
+ *
+ * This function replaces a page in the pagecache with a new one.  On
+ * success it acquires the pagecache reference for the new page and
+ * drop it for the old page.  Both the old and new pages must be
+ * locked.  This function does not add the new page to the LRU, the
+ * caller must do that.
+ *
+ * The remove + add is atomic.  The only way this function can fail is
+ * memory allocation failure.
+ */
+int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
+{
+	int error;
+
+	VM_BUG_ON(!PageLocked(old));
+	VM_BUG_ON(!PageLocked(new));
+	VM_BUG_ON(new->mapping);
+
+	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
+	if (!error) {
+		struct address_space *mapping = old->mapping;
+		pgoff_t offset = old->index;
+
+		page_cache_get(new);
+		new->mapping = mapping;
+		new->index = offset;
+
+		spin_lock_irq(&mapping->tree_lock);
+		__remove_from_page_cache(old);
+		error = radix_tree_insert(&mapping->page_tree, offset, new);
+		BUG_ON(error);
+		mapping->nrpages++;
+		__inc_zone_page_state(new, NR_FILE_PAGES);
+		if (PageSwapBacked(new))
+			__inc_zone_page_state(new, NR_SHMEM);
+		spin_unlock_irq(&mapping->tree_lock);
+		radix_tree_preload_end();
+		mem_cgroup_replace_cache_page(old, new);
+		page_cache_release(old);
+	}
+
+	return error;
+}
+EXPORT_SYMBOL_GPL(replace_page_cache_page);
+
+/**
  * add_to_page_cache_locked - add a locked page to the pagecache
  * @page:	page to add
  * @mapping:	the page's address_space
Index: linux-2.6/include/linux/pagemap.h
===================================================================
--- linux-2.6.orig/include/linux/pagemap.h	2011-01-07 17:53:39.000000000 +0100
+++ linux-2.6/include/linux/pagemap.h	2011-01-07 19:14:45.000000000 +0100
@@ -457,6 +457,7 @@ int add_to_page_cache_lru(struct page *p
 				pgoff_t index, gfp_t gfp_mask);
 extern void remove_from_page_cache(struct page *page);
 extern void __remove_from_page_cache(struct page *page);
+int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
 
 /*
  * Like add_to_page_cache_locked, but used to add newly allocated pages:
Index: linux-2.6/fs/fuse/dev.c
===================================================================
--- linux-2.6.orig/fs/fuse/dev.c	2011-01-07 17:53:39.000000000 +0100
+++ linux-2.6/fs/fuse/dev.c	2011-01-07 19:14:45.000000000 +0100
@@ -737,14 +737,12 @@ static int fuse_try_move_page(struct fus
 	if (WARN_ON(PageMlocked(oldpage)))
 		goto out_fallback_unlock;
 
-	remove_from_page_cache(oldpage);
-	page_cache_release(oldpage);
-
-	err = add_to_page_cache_locked(newpage, mapping, index, GFP_KERNEL);
+	err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
 	if (err) {
-		printk(KERN_WARNING "fuse_try_move_page: failed to add page");
-		goto out_fallback_unlock;
+		unlock_page(newpage);
+		return err;
 	}
+
 	page_cache_get(newpage);
 
 	if (!(buf->flags & PIPE_BUF_FLAG_LRU))
Index: linux-2.6/include/linux/memcontrol.h
===================================================================
--- linux-2.6.orig/include/linux/memcontrol.h	2011-01-07 17:53:39.000000000 +0100
+++ linux-2.6/include/linux/memcontrol.h	2011-01-07 19:14:45.000000000 +0100
@@ -95,6 +95,9 @@ mem_cgroup_prepare_migration(struct page
 extern void mem_cgroup_end_migration(struct mem_cgroup *mem,
 	struct page *oldpage, struct page *newpage);
 
+extern void mem_cgroup_replace_cache_page(struct page *oldpage,
+					  struct page *newpage);
+
 /*
  * For memory reclaim.
  */
@@ -236,6 +239,11 @@ static inline void mem_cgroup_end_migrat
 {
 }
 
+static inline void mem_cgroup_replace_cache_page(struct page *oldpage,
+					 	struct page *newpage)
+{
+}
+
 static inline int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
 {
 	return 0;
Index: linux-2.6/mm/memcontrol.c
===================================================================
--- linux-2.6.orig/mm/memcontrol.c	2011-01-07 17:53:39.000000000 +0100
+++ linux-2.6/mm/memcontrol.c	2011-01-07 19:20:41.000000000 +0100
@@ -2905,6 +2905,44 @@ void mem_cgroup_end_migration(struct mem
 }
 
 /*
+ * This function moves the charge from oldpage to newpage.  The new
+ * page must not be already charged.
+ */
+void mem_cgroup_replace_cache_page(struct page *oldpage, struct page *newpage)
+{
+	struct page_cgroup *old_pc;
+	struct page_cgroup *new_pc;
+	struct mem_cgroup *mem;
+
+	if (mem_cgroup_disabled())
+		return;
+
+	old_pc = lookup_page_cgroup(oldpage);
+	lock_page_cgroup(old_pc);
+	if (!PageCgroupUsed(old_pc)) {
+		unlock_page_cgroup(old_pc);
+		return;
+	}
+
+	mem = old_pc->mem_cgroup;
+	css_get(&mem->css);
+	ClearPageCgroupUsed(old_pc);
+	unlock_page_cgroup(old_pc);
+
+	new_pc = lookup_page_cgroup(newpage);
+	lock_page_cgroup(new_pc);
+	BUG_ON(PageCgroupUsed(new_pc));
+
+	new_pc->mem_cgroup = mem;
+	smp_wmb();
+	SetPageCgroupCache(new_pc);
+	SetPageCgroupUsed(new_pc);
+	unlock_page_cgroup(new_pc);
+	css_put(&mem->css);
+}
+
+
+/*
  * A call to try to shrink memory usage on charge failure at shmem's swapin.
  * Calling hierarchical_reclaim is not enough because we should update
  * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.

[PATCH] mm: add replace_page_cache_page() function

[Miklos Szeredi]

Here's an updated patch, addressing the review comments.

Hiroyuki-san, can you please review the newly introduced
mem_cgroup_replace_cache_page(), as I'm not fully familiar with the
memory cgroup code.

Thanks,
Miklos
---

From: Miklos Szeredi <mszeredi@suse.cz>
Subject: mm: add replace_page_cache_page() function

This function basically does:

     remove_from_page_cache(old);
     page_cache_release(old);
     add_to_page_cache_locked(new);

Except it does this atomically, so there's no possibility for the
"add" to fail because of a race.

This is used by fuse to move pages into the page cache.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
---
 fs/fuse/dev.c              |   10 +++------
 include/linux/memcontrol.h |    8 +++++++
 include/linux/pagemap.h    |    1 
 mm/filemap.c               |   50 +++++++++++++++++++++++++++++++++++++++++++++
 mm/memcontrol.c            |   38 ++++++++++++++++++++++++++++++++++
 5 files changed, 101 insertions(+), 6 deletions(-)

Index: linux-2.6/mm/filemap.c
===================================================================
--- linux-2.6.orig/mm/filemap.c	2011-01-07 17:53:39.000000000 +0100
+++ linux-2.6/mm/filemap.c	2011-01-07 19:14:45.000000000 +0100
@@ -390,6 +390,56 @@ int filemap_write_and_wait_range(struct
 EXPORT_SYMBOL(filemap_write_and_wait_range);
 
 /**
+ * replace_page_cache_page - replace a pagecache page with a new one
+ * @old:	page to be replaced
+ * @new:	page to replace with
+ * @gfp_mask:	page allocation mode
+ *
+ * This function replaces a page in the pagecache with a new one.  On
+ * success it acquires the pagecache reference for the new page and
+ * drop it for the old page.  Both the old and new pages must be
+ * locked.  This function does not add the new page to the LRU, the
+ * caller must do that.
+ *
+ * The remove + add is atomic.  The only way this function can fail is
+ * memory allocation failure.
+ */
+int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
+{
+	int error;
+
+	VM_BUG_ON(!PageLocked(old));
+	VM_BUG_ON(!PageLocked(new));
+	VM_BUG_ON(new->mapping);
+
+	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
+	if (!error) {
+		struct address_space *mapping = old->mapping;
+		pgoff_t offset = old->index;
+
+		page_cache_get(new);
+		new->mapping = mapping;
+		new->index = offset;
+
+		spin_lock_irq(&mapping->tree_lock);
+		__remove_from_page_cache(old);
+		error = radix_tree_insert(&mapping->page_tree, offset, new);
+		BUG_ON(error);
+		mapping->nrpages++;
+		__inc_zone_page_state(new, NR_FILE_PAGES);
+		if (PageSwapBacked(new))
+			__inc_zone_page_state(new, NR_SHMEM);
+		spin_unlock_irq(&mapping->tree_lock);
+		radix_tree_preload_end();
+		mem_cgroup_replace_cache_page(old, new);
+		page_cache_release(old);
+	}
+
+	return error;
+}
+EXPORT_SYMBOL_GPL(replace_page_cache_page);
+
+/**
  * add_to_page_cache_locked - add a locked page to the pagecache
  * @page:	page to add
  * @mapping:	the page's address_space
Index: linux-2.6/include/linux/pagemap.h
===================================================================
--- linux-2.6.orig/include/linux/pagemap.h	2011-01-07 17:53:39.000000000 +0100
+++ linux-2.6/include/linux/pagemap.h	2011-01-07 19:14:45.000000000 +0100
@@ -457,6 +457,7 @@ int add_to_page_cache_lru(struct page *p
 				pgoff_t index, gfp_t gfp_mask);
 extern void remove_from_page_cache(struct page *page);
 extern void __remove_from_page_cache(struct page *page);
+int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
 
 /*
  * Like add_to_page_cache_locked, but used to add newly allocated pages:
Index: linux-2.6/fs/fuse/dev.c
===================================================================
--- linux-2.6.orig/fs/fuse/dev.c	2011-01-07 17:53:39.000000000 +0100
+++ linux-2.6/fs/fuse/dev.c	2011-01-07 19:14:45.000000000 +0100
@@ -737,14 +737,12 @@ static int fuse_try_move_page(struct fus
 	if (WARN_ON(PageMlocked(oldpage)))
 		goto out_fallback_unlock;
 
-	remove_from_page_cache(oldpage);
-	page_cache_release(oldpage);
-
-	err = add_to_page_cache_locked(newpage, mapping, index, GFP_KERNEL);
+	err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
 	if (err) {
-		printk(KERN_WARNING "fuse_try_move_page: failed to add page");
-		goto out_fallback_unlock;
+		unlock_page(newpage);
+		return err;
 	}
+
 	page_cache_get(newpage);
 
 	if (!(buf->flags & PIPE_BUF_FLAG_LRU))
Index: linux-2.6/include/linux/memcontrol.h
===================================================================
--- linux-2.6.orig/include/linux/memcontrol.h	2011-01-07 17:53:39.000000000 +0100
+++ linux-2.6/include/linux/memcontrol.h	2011-01-07 19:14:45.000000000 +0100
@@ -95,6 +95,9 @@ mem_cgroup_prepare_migration(struct page
 extern void mem_cgroup_end_migration(struct mem_cgroup *mem,
 	struct page *oldpage, struct page *newpage);
 
+extern void mem_cgroup_replace_cache_page(struct page *oldpage,
+					  struct page *newpage);
+
 /*
  * For memory reclaim.
  */
@@ -236,6 +239,11 @@ static inline void mem_cgroup_end_migrat
 {
 }
 
+static inline void mem_cgroup_replace_cache_page(struct page *oldpage,
+					 	struct page *newpage)
+{
+}
+
 static inline int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
 {
 	return 0;
Index: linux-2.6/mm/memcontrol.c
===================================================================
--- linux-2.6.orig/mm/memcontrol.c	2011-01-07 17:53:39.000000000 +0100
+++ linux-2.6/mm/memcontrol.c	2011-01-07 19:20:41.000000000 +0100
@@ -2905,6 +2905,44 @@ void mem_cgroup_end_migration(struct mem
 }
 
 /*
+ * This function moves the charge from oldpage to newpage.  The new
+ * page must not be already charged.
+ */
+void mem_cgroup_replace_cache_page(struct page *oldpage, struct page *newpage)
+{
+	struct page_cgroup *old_pc;
+	struct page_cgroup *new_pc;
+	struct mem_cgroup *mem;
+
+	if (mem_cgroup_disabled())
+		return;
+
+	old_pc = lookup_page_cgroup(oldpage);
+	lock_page_cgroup(old_pc);
+	if (!PageCgroupUsed(old_pc)) {
+		unlock_page_cgroup(old_pc);
+		return;
+	}
+
+	mem = old_pc->mem_cgroup;
+	css_get(&mem->css);
+	ClearPageCgroupUsed(old_pc);
+	unlock_page_cgroup(old_pc);
+
+	new_pc = lookup_page_cgroup(newpage);
+	lock_page_cgroup(new_pc);
+	BUG_ON(PageCgroupUsed(new_pc));
+
+	new_pc->mem_cgroup = mem;
+	smp_wmb();
+	SetPageCgroupCache(new_pc);
+	SetPageCgroupUsed(new_pc);
+	unlock_page_cgroup(new_pc);
+	css_put(&mem->css);
+}
+
+
+/*
  * A call to try to shrink memory usage on charge failure at shmem's swapin.
  * Calling hierarchical_reclaim is not enough because we should update
  * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>