[PATCH] ceph: switch back to testing for NULL folio->private in ceph_dirty_folio

[PATCH] ceph: switch back to testing for NULL folio->private in ceph_dirty_folio

[Jeff Layton]

Willy requested that we change this back to warning on folio->private
being non-NULl. He's trying to kill off the PG_private flag, and so we'd
like to catch where it's non-NULL.

Add a VM_WARN_ON_FOLIO (since it doesn't exist yet) and change over to
using that instead of VM_BUG_ON_FOLIO along with testing the ->private
pointer.

Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 fs/ceph/addr.c          | 2 +-
 include/linux/mmdebug.h | 9 +++++++++
 2 files changed, 10 insertions(+), 1 deletion(-)

diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
index b43cc01a61db..b24d6bdb91db 100644
--- a/fs/ceph/addr.c
+++ b/fs/ceph/addr.c
@@ -122,7 +122,7 @@ static bool ceph_dirty_folio(struct address_space *mapping, struct folio *folio)
 	 * Reference snap context in folio->private.  Also set
 	 * PagePrivate so that we get invalidate_folio callback.
 	 */
-	VM_BUG_ON_FOLIO(folio_test_private(folio), folio);
+	VM_WARN_ON_FOLIO(folio->private, folio);
 	folio_attach_private(folio, snapc);
 
 	return ceph_fscache_dirty_folio(mapping, folio);
diff --git a/include/linux/mmdebug.h b/include/linux/mmdebug.h
index d7285f8148a3..5107bade2ab2 100644
--- a/include/linux/mmdebug.h
+++ b/include/linux/mmdebug.h
@@ -54,6 +54,15 @@ void dump_mm(const struct mm_struct *mm);
 	}								\
 	unlikely(__ret_warn_once);					\
 })
+#define VM_WARN_ON_FOLIO(cond, folio)		({			\
+	int __ret_warn = !!(cond);					\
+									\
+	if (unlikely(__ret_warn)) {					\
+		dump_page(&folio->page, "VM_WARN_ON_FOLIO(" __stringify(cond)")");\
+		WARN_ON(1);						\
+	}								\
+	unlikely(__ret_warn);						\
+})
 #define VM_WARN_ON_ONCE_FOLIO(cond, folio)	({			\
 	static bool __section(".data.once") __warned;			\
 	int __ret_warn_once = !!(cond);					\
-- 
2.36.1

Re: [PATCH] ceph: switch back to testing for NULL folio->private in ceph_dirty_folio

[Xiubo Li]

On 6/10/22 11:40 PM, Jeff Layton wrote:
> Willy requested that we change this back to warning on folio->private
> being non-NULl. He's trying to kill off the PG_private flag, and so we'd
> like to catch where it's non-NULL.
>
> Add a VM_WARN_ON_FOLIO (since it doesn't exist yet) and change over to
> using that instead of VM_BUG_ON_FOLIO along with testing the ->private
> pointer.
>
> Cc: Matthew Wilcox <willy@infradead.org>
> Signed-off-by: Jeff Layton <jlayton@kernel.org>
> ---
>   fs/ceph/addr.c          | 2 +-
>   include/linux/mmdebug.h | 9 +++++++++
>   2 files changed, 10 insertions(+), 1 deletion(-)
>
> diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
> index b43cc01a61db..b24d6bdb91db 100644
> --- a/fs/ceph/addr.c
> +++ b/fs/ceph/addr.c
> @@ -122,7 +122,7 @@ static bool ceph_dirty_folio(struct address_space *mapping, struct folio *folio)
>   	 * Reference snap context in folio->private.  Also set
>   	 * PagePrivate so that we get invalidate_folio callback.
>   	 */
> -	VM_BUG_ON_FOLIO(folio_test_private(folio), folio);
> +	VM_WARN_ON_FOLIO(folio->private, folio);
>   	folio_attach_private(folio, snapc);
>   
>   	return ceph_fscache_dirty_folio(mapping, folio);
> diff --git a/include/linux/mmdebug.h b/include/linux/mmdebug.h
> index d7285f8148a3..5107bade2ab2 100644
> --- a/include/linux/mmdebug.h
> +++ b/include/linux/mmdebug.h
> @@ -54,6 +54,15 @@ void dump_mm(const struct mm_struct *mm);
>   	}								\
>   	unlikely(__ret_warn_once);					\
>   })
> +#define VM_WARN_ON_FOLIO(cond, folio)		({			\
> +	int __ret_warn = !!(cond);					\
> +									\
> +	if (unlikely(__ret_warn)) {					\
> +		dump_page(&folio->page, "VM_WARN_ON_FOLIO(" __stringify(cond)")");\
> +		WARN_ON(1);						\
> +	}								\
> +	unlikely(__ret_warn);						\
> +})
>   #define VM_WARN_ON_ONCE_FOLIO(cond, folio)	({			\
>   	static bool __section(".data.once") __warned;			\
>   	int __ret_warn_once = !!(cond);					\

All tests passed except the known issues.

Merged into testing branch, thanks Jeff!

-- Xiubo

Re: [PATCH] ceph: switch back to testing for NULL folio->private in ceph_dirty_folio

[Xiubo Li]

On 6/10/22 11:40 PM, Jeff Layton wrote:
> Willy requested that we change this back to warning on folio->private
> being non-NULl. He's trying to kill off the PG_private flag, and so we'd
> like to catch where it's non-NULL.
>
> Add a VM_WARN_ON_FOLIO (since it doesn't exist yet) and change over to
> using that instead of VM_BUG_ON_FOLIO along with testing the ->private
> pointer.
>
> Cc: Matthew Wilcox <willy@infradead.org>
> Signed-off-by: Jeff Layton <jlayton@kernel.org>
> ---
>   fs/ceph/addr.c          | 2 +-
>   include/linux/mmdebug.h | 9 +++++++++
>   2 files changed, 10 insertions(+), 1 deletion(-)
>
> diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
> index b43cc01a61db..b24d6bdb91db 100644
> --- a/fs/ceph/addr.c
> +++ b/fs/ceph/addr.c
> @@ -122,7 +122,7 @@ static bool ceph_dirty_folio(struct address_space *mapping, struct folio *folio)
>   	 * Reference snap context in folio->private.  Also set
>   	 * PagePrivate so that we get invalidate_folio callback.
>   	 */
> -	VM_BUG_ON_FOLIO(folio_test_private(folio), folio);
> +	VM_WARN_ON_FOLIO(folio->private, folio);
>   	folio_attach_private(folio, snapc);
>   
>   	return ceph_fscache_dirty_folio(mapping, folio);
> diff --git a/include/linux/mmdebug.h b/include/linux/mmdebug.h
> index d7285f8148a3..5107bade2ab2 100644
> --- a/include/linux/mmdebug.h
> +++ b/include/linux/mmdebug.h
> @@ -54,6 +54,15 @@ void dump_mm(const struct mm_struct *mm);
>   	}								\
>   	unlikely(__ret_warn_once);					\
>   })
> +#define VM_WARN_ON_FOLIO(cond, folio)		({			\
> +	int __ret_warn = !!(cond);					\
> +									\
> +	if (unlikely(__ret_warn)) {					\
> +		dump_page(&folio->page, "VM_WARN_ON_FOLIO(" __stringify(cond)")");\
> +		WARN_ON(1);						\
> +	}								\
> +	unlikely(__ret_warn);						\
> +})

I have fixed the compile warning reported by kernel test robot by 
defining it in case the DEBUG_VM is disabled in testing branch.

-- Xiubo


>   #define VM_WARN_ON_ONCE_FOLIO(cond, folio)	({			\
>   	static bool __section(".data.once") __warned;			\
>   	int __ret_warn_once = !!(cond);					\

Re: [PATCH] ceph: switch back to testing for NULL folio->private in ceph_dirty_folio

[Matthew Wilcox]

On Mon, Jun 13, 2022 at 08:48:40AM +0800, Xiubo Li wrote:
> 
> On 6/10/22 11:40 PM, Jeff Layton wrote:
> > Willy requested that we change this back to warning on folio->private
> > being non-NULl. He's trying to kill off the PG_private flag, and so we'd
> > like to catch where it's non-NULL.
> > 
> > Add a VM_WARN_ON_FOLIO (since it doesn't exist yet) and change over to
> > using that instead of VM_BUG_ON_FOLIO along with testing the ->private
> > pointer.
> > 
> > Cc: Matthew Wilcox <willy@infradead.org>
> > Signed-off-by: Jeff Layton <jlayton@kernel.org>
> > ---
> >   fs/ceph/addr.c          | 2 +-
> >   include/linux/mmdebug.h | 9 +++++++++
> >   2 files changed, 10 insertions(+), 1 deletion(-)
> > 
> > diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
> > index b43cc01a61db..b24d6bdb91db 100644
> > --- a/fs/ceph/addr.c
> > +++ b/fs/ceph/addr.c
> > @@ -122,7 +122,7 @@ static bool ceph_dirty_folio(struct address_space *mapping, struct folio *folio)
> >   	 * Reference snap context in folio->private.  Also set
> >   	 * PagePrivate so that we get invalidate_folio callback.
> >   	 */
> > -	VM_BUG_ON_FOLIO(folio_test_private(folio), folio);
> > +	VM_WARN_ON_FOLIO(folio->private, folio);
> >   	folio_attach_private(folio, snapc);
> >   	return ceph_fscache_dirty_folio(mapping, folio);

I found a couple of places where page->private needs to be NULLed out.
Neither of them are Ceph's fault.  I decided that testing whether
folio->private and PG_private are in agreement was better done in
folio_unlock() than in any of the other potential places we could
check for it.

diff --git a/mm/filemap.c b/mm/filemap.c
index 8ef861297ffb..acef71f75e78 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -1535,6 +1535,9 @@ void folio_unlock(struct folio *folio)
 	BUILD_BUG_ON(PG_waiters != 7);
 	BUILD_BUG_ON(PG_locked > 7);
 	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
+	VM_BUG_ON_FOLIO(!folio_test_private(folio) &&
+			!folio_test_swapbacked(folio) &&
+			folio_get_private(folio), folio);
 	if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0)))
 		folio_wake_bit(folio, PG_locked);
 }
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 2e2a8b5bc567..af0751a79c19 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -2438,6 +2438,7 @@ static void __split_huge_page_tail(struct page *head, int tail,
 			page_tail);
 	page_tail->mapping = head->mapping;
 	page_tail->index = head->index + tail;
+	page_tail->private = 0;
 
 	/* Page flags must be visible before we make the page non-compound. */
 	smp_wmb();
diff --git a/mm/migrate.c b/mm/migrate.c
index eb62e026c501..fa8e36e74f0d 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -1157,6 +1157,8 @@ static int unmap_and_move(new_page_t get_new_page,
 	newpage = get_new_page(page, private);
 	if (!newpage)
 		return -ENOMEM;
+	BUG_ON(compound_order(newpage) != compound_order(page));
+	newpage->private = 0;
 
 	rc = __unmap_and_move(page, newpage, force, mode);
 	if (rc == MIGRATEPAGE_SUCCESS)

Re: [PATCH] ceph: switch back to testing for NULL folio->private in ceph_dirty_folio

[Xiubo Li]

On 6/19/22 11:49 AM, Matthew Wilcox wrote:
> On Mon, Jun 13, 2022 at 08:48:40AM +0800, Xiubo Li wrote:
>> On 6/10/22 11:40 PM, Jeff Layton wrote:
>>> Willy requested that we change this back to warning on folio->private
>>> being non-NULl. He's trying to kill off the PG_private flag, and so we'd
>>> like to catch where it's non-NULL.
>>>
>>> Add a VM_WARN_ON_FOLIO (since it doesn't exist yet) and change over to
>>> using that instead of VM_BUG_ON_FOLIO along with testing the ->private
>>> pointer.
>>>
>>> Cc: Matthew Wilcox <willy@infradead.org>
>>> Signed-off-by: Jeff Layton <jlayton@kernel.org>
>>> ---
>>>    fs/ceph/addr.c          | 2 +-
>>>    include/linux/mmdebug.h | 9 +++++++++
>>>    2 files changed, 10 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
>>> index b43cc01a61db..b24d6bdb91db 100644
>>> --- a/fs/ceph/addr.c
>>> +++ b/fs/ceph/addr.c
>>> @@ -122,7 +122,7 @@ static bool ceph_dirty_folio(struct address_space *mapping, struct folio *folio)
>>>    	 * Reference snap context in folio->private.  Also set
>>>    	 * PagePrivate so that we get invalidate_folio callback.
>>>    	 */
>>> -	VM_BUG_ON_FOLIO(folio_test_private(folio), folio);
>>> +	VM_WARN_ON_FOLIO(folio->private, folio);
>>>    	folio_attach_private(folio, snapc);
>>>    	return ceph_fscache_dirty_folio(mapping, folio);
> I found a couple of places where page->private needs to be NULLed out.
> Neither of them are Ceph's fault.  I decided that testing whether
> folio->private and PG_private are in agreement was better done in
> folio_unlock() than in any of the other potential places we could
> check for it.

Hi Willy,

Cool. I will test this patch today. Thanks!

-- Xiubo

> diff --git a/mm/filemap.c b/mm/filemap.c
> index 8ef861297ffb..acef71f75e78 100644
> --- a/mm/filemap.c
> +++ b/mm/filemap.c
> @@ -1535,6 +1535,9 @@ void folio_unlock(struct folio *folio)
>   	BUILD_BUG_ON(PG_waiters != 7);
>   	BUILD_BUG_ON(PG_locked > 7);
>   	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
> +	VM_BUG_ON_FOLIO(!folio_test_private(folio) &&
> +			!folio_test_swapbacked(folio) &&
> +			folio_get_private(folio), folio);
>   	if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0)))
>   		folio_wake_bit(folio, PG_locked);
>   }
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 2e2a8b5bc567..af0751a79c19 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -2438,6 +2438,7 @@ static void __split_huge_page_tail(struct page *head, int tail,
>   			page_tail);
>   	page_tail->mapping = head->mapping;
>   	page_tail->index = head->index + tail;
> +	page_tail->private = 0;
>   
>   	/* Page flags must be visible before we make the page non-compound. */
>   	smp_wmb();
> diff --git a/mm/migrate.c b/mm/migrate.c
> index eb62e026c501..fa8e36e74f0d 100644
> --- a/mm/migrate.c
> +++ b/mm/migrate.c
> @@ -1157,6 +1157,8 @@ static int unmap_and_move(new_page_t get_new_page,
>   	newpage = get_new_page(page, private);
>   	if (!newpage)
>   		return -ENOMEM;
> +	BUG_ON(compound_order(newpage) != compound_order(page));
> +	newpage->private = 0;
>   
>   	rc = __unmap_and_move(page, newpage, force, mode);
>   	if (rc == MIGRATEPAGE_SUCCESS)
>