mark_inode_dirty vs mark_inode_dirty_sync

mark_inode_dirty vs mark_inode_dirty_sync

[David Sanchez]

Hi,
Please, could somebody explain me what the mark_inode_dirty* functions
do and what is the difference between mark_inode_dirty and
mark_inode_dirty_sync ?

Thanks


David SANCHEZ

Re: mark_inode_dirty vs mark_inode_dirty_sync

[Dave Kleikamp]

On Fri, 2005-09-02 at 11:55 +0200, David Sanchez wrote:
> Hi,
> Please, could somebody explain me what the mark_inode_dirty* functions
> do and what is the difference between mark_inode_dirty and
> mark_inode_dirty_sync ?

They put the inode on the superblock's dirty list and make the inode as
dirty in the i_state field.  This makes sure that the inode will
eventually be written to disk.

mark_inode_dirty_sync only sets the I_DIRTY_SYNC flag, which does not
imply that any file data was changed.  It is called when a minor change
is made to an inode, such as a timestamp is changed.  Some sync
operations will only write the inode if data was written, so can avoid
writing the an inode that is only dirtied by I_DIRTY_SYNC.

mark_inode_dirty sets I_DIRTY which is I_DIRTY_SYNC | I_DIRTY_DATASYNC |
I_DIRTY_PAGES.  This indicates that the in-memory inode has changes to
the data that have not yet been written to disk.

Shaggy
-- 
David Kleikamp
IBM Linux Technology Center

[PATCH] document mark_inode_dirty & mark_inode_dirty_sync in fs.h

[Dave Kleikamp]

On Fri, 2005-09-02 at 10:46 -0600, Andreas Dilger wrote:
> On Sep 02, 2005  07:42 -0500, Dave Kleikamp wrote:
> > They put the inode on the superblock's dirty list and make the inode as
> > dirty in the i_state field.  This makes sure that the inode will
> > eventually be written to disk.
> > 
> > mark_inode_dirty_sync only sets the I_DIRTY_SYNC flag, which does not
> > imply that any file data was changed.  It is called when a minor change
> > is made to an inode, such as a timestamp is changed.  Some sync
> > operations will only write the inode if data was written, so can avoid
> > writing the an inode that is only dirtied by I_DIRTY_SYNC.
> > 
> > mark_inode_dirty sets I_DIRTY which is I_DIRTY_SYNC | I_DIRTY_DATASYNC |
> > I_DIRTY_PAGES.  This indicates that the in-memory inode has changes to
> > the data that have not yet been written to disk.
> 
> Dave, could you consider submitting a patch to add the above as comments
> to fs.h for future reference?
> 
> Cheers, Andreas

How about this?
=================
Document mark_inode_dirty and mark_inode_dirty_sync in fs.h

Signed-off-by: Dave Kleikamp <shaggy@austin.ibm.com>

diff --git a/include/linux/fs.h b/include/linux/fs.h
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -1048,7 +1048,7 @@ struct super_operations {
 /* Inode state bits.  Protected by inode_lock. */
 #define I_DIRTY_SYNC		1 /* Not dirty enough for O_DATASYNC */
 #define I_DIRTY_DATASYNC	2 /* Data-related inode changes pending */
-#define I_DIRTY_PAGES		4 /* Data-related inode changes pending */
+#define I_DIRTY_PAGES		4 /* Data changes pending */
 #define __I_LOCK		3
 #define I_LOCK			(1 << __I_LOCK)
 #define I_FREEING		16
@@ -1059,11 +1059,19 @@ struct super_operations {
 #define I_DIRTY (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_PAGES)
 
 extern void __mark_inode_dirty(struct inode *, int);
+/*
+ * mark_inode_dirty indicates pending changes to the inode's data.
+ * Puts inode on superblock's dirty list.
+ */
 static inline void mark_inode_dirty(struct inode *inode)
 {
 	__mark_inode_dirty(inode, I_DIRTY);
 }
 
+/*
+ * mark_inode_dirty_sync indicates non-data related changes to the inode,
+ * such as a change to a timestamp.  Puts inode on superblock's dirty list.
+ */
 static inline void mark_inode_dirty_sync(struct inode *inode)
 {
 	__mark_inode_dirty(inode, I_DIRTY_SYNC);

-- 
David Kleikamp
IBM Linux Technology Center