Provide an iget variant without unlocking the inode and ->read_inode
call. This is needed for XFS and IIRC the reiserfs folks wanted it,
too.
Tested in 2.5 for more than half a year and in 2.4-ac/-aa, and
the varoius vendor trees for a long time.
--- 1.37/fs/inode.c Thu Jul 10 11:51:08 2003
+++ edited/fs/inode.c Tue Aug 5 01:42:38 2003
@@ -834,6 +839,20 @@
return inode;
}
+void unlock_new_inode(struct inode *inode)
+{
+ /*
+ * This is special! We do not need the spinlock
+ * when clearing I_LOCK, because we're guaranteed
+ * that nobody else tries to do anything about the
+ * state of the inode when it is locked, as we
+ * just created it (so there can be no old holders
+ * that haven't tested I_LOCK).
+ */
+ inode->i_state &= ~(I_LOCK|I_NEW);
+ wake_up(&inode->i_wait);
+}
+
/*
* This is called without the inode lock held.. Be careful.
*
@@ -856,31 +875,13 @@
list_add(&inode->i_list, &inode_in_use);
list_add(&inode->i_hash, head);
inode->i_ino = ino;
- inode->i_state = I_LOCK;
+ inode->i_state = I_LOCK|I_NEW;
spin_unlock(&inode_lock);
- /* reiserfs specific hack right here. We don't
- ** want this to last, and are looking for VFS changes
- ** that will allow us to get rid of it.
- ** -- [email protected]
- */
- if (sb->s_op->read_inode2) {
- sb->s_op->read_inode2(inode, opaque) ;
- } else {
- sb->s_op->read_inode(inode);
- }
-
/*
- * This is special! We do not need the spinlock
- * when clearing I_LOCK, because we're guaranteed
- * that nobody else tries to do anything about the
- * state of the inode when it is locked, as we
- * just created it (so there can be no old holders
- * that haven't tested I_LOCK).
+ * Return the locked inode with I_NEW set, the
+ * caller is responsible for filling in the contents
*/
- inode->i_state &= ~I_LOCK;
- wake_up(&inode->i_wait);
-
return inode;
}
@@ -960,8 +961,7 @@
return inode;
}
-
-struct inode *iget4(struct super_block *sb, unsigned long ino, find_inode_t find_actor, void *opaque)
+struct inode *iget4_locked(struct super_block *sb, unsigned long ino, find_inode_t find_actor, void *opaque)
{
struct list_head * head = inode_hashtable + hash(sb,ino);
struct inode * inode;
--- 1.73/include/linux/fs.h Sun Aug 3 16:50:01 2003
+++ edited/include/linux/fs.h Thu Aug 7 12:44:44 2003
@@ -966,6 +969,7 @@
#define I_LOCK 8
#define I_FREEING 16
#define I_CLEAR 32
+#define I_NEW 64
#define I_DIRTY (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_PAGES)
@@ -1391,12 +1396,47 @@
extern void force_delete(struct inode *);
extern struct inode * igrab(struct inode *);
extern ino_t iunique(struct super_block *, ino_t);
+extern void unlock_new_inode(struct inode *);
typedef int (*find_inode_t)(struct inode *, unsigned long, void *);
-extern struct inode * iget4(struct super_block *, unsigned long, find_inode_t, void *);
+
+extern struct inode * iget4_locked(struct super_block *, unsigned long,
+ find_inode_t, void *);
+
+static inline struct inode *iget4(struct super_block *sb, unsigned long ino,
+ find_inode_t find_actor, void *opaque)
+{
+ struct inode *inode = iget4_locked(sb, ino, find_actor, opaque);
+
+ if (inode && (inode->i_state & I_NEW)) {
+ /*
+ * reiserfs-specific kludge that is expected to go away ASAP.
+ */
+ if (sb->s_op->read_inode2)
+ sb->s_op->read_inode2(inode, opaque);
+ else
+ sb->s_op->read_inode(inode);
+ unlock_new_inode(inode);
+ }
+
+ return inode;
+}
+
static inline struct inode *iget(struct super_block *sb, unsigned long ino)
{
- return iget4(sb, ino, NULL, NULL);
+ struct inode *inode = iget4_locked(sb, ino, NULL, NULL);
+
+ if (inode && (inode->i_state & I_NEW)) {
+ sb->s_op->read_inode(inode);
+ unlock_new_inode(inode);
+ }
+
+ return inode;
+}
+
+static inline struct inode *iget_locked(struct super_block *sb, unsigned long ino)
+{
+ return iget4_locked(sb, ino, NULL, NULL);
}
extern void clear_inode(struct inode *);
--- 1.67/kernel/ksyms.c Sun Aug 3 16:50:01 2003
+++ edited/kernel/ksyms.c Tue Aug 5 01:44:42 2003
@@ -143,7 +143,8 @@
EXPORT_SYMBOL(fget);
EXPORT_SYMBOL(igrab);
EXPORT_SYMBOL(iunique);
-EXPORT_SYMBOL(iget4);
+EXPORT_SYMBOL(iget4_locked);
+EXPORT_SYMBOL(unlock_new_inode);
EXPORT_SYMBOL(iput);
EXPORT_SYMBOL(inode_init_once);
EXPORT_SYMBOL(force_delete);
Helllo!
On Mon, Aug 25, 2003 at 04:07:14PM +0200, Christoph Hellwig wrote:
> Provide an iget variant without unlocking the inode and ->read_inode
> call. This is needed for XFS and IIRC the reiserfs folks wanted it,
> too.
No. This patch is useless for our purposes. (and coda/nfs ).
We wanted to get rid of a race where inode allocation and
filling fs specific parts are done non-atomically.
The patch below does not achieve this. We still fill inode private part
outside of inode_lock locked region.
Bye,
Oleg
On Tue, Aug 26, 2003 at 03:27:16PM +0400, Oleg Drokin wrote:
> The patch below does not achieve this. We still fill inode private part
> outside of inode_lock locked region.
-ENOPATCH :)
Hello!
> Mail-Followup-To: Christoph Hellwig <[email protected]>,
Hm, very interesting header, I'd say. No wonder I'm getting errors replying to
your emails.
On Tue, Aug 26, 2003 at 03:48:09PM +0200, Christoph Hellwig wrote:
> > The patch below does not achieve this. We still fill inode private part
> > outside of inode_lock locked region.
> -ENOPATCH :)
I meant the patch in the email you sent and to which I answered originally ;)
Bye,
Oleg
Hello!
On Tue, Aug 26, 2003 at 04:30:24PM +0200, Christoph Hellwig wrote:
> > > Mail-Followup-To: Christoph Hellwig <[email protected]>,
> > Hm, very interesting header, I'd say. No wonder I'm getting errors replying to
> > your emails.
> Well, I got the same from you although I though only in the Cc line
> which I removed.
That's because I hit "reply" first time and have not looked at the CC list.
> > > > The patch below does not achieve this. We still fill inode private part
> > > > outside of inode_lock locked region.
> > > -ENOPATCH :)
> > I meant the patch in the email you sent and to which I answered originally ;)
> Sorry, I missed the 'not' when reading and though you had an alternate
> patch
Actually, I have altenate patch (but no, I have not tried to attach it) ;)
I have full iget5_locked backport (that I was sending around some time ago). Only it
will break every iget4 user not in the tree (by changing VFS API), so I guess it is not an option.
But if people think it is good idea to adapt this change, I can easily resurrect that patch, of course.
Bye,
Oleg
On Tue, Aug 26, 2003 at 05:54:42PM +0400, Oleg Drokin wrote:
> Hello!
>
> > Mail-Followup-To: Christoph Hellwig <[email protected]>,
>
> Hm, very interesting header, I'd say. No wonder I'm getting errors replying to
> your emails.
Well, I got the same from you although I though only in the Cc line
which I removed.
>
> > > The patch below does not achieve this. We still fill inode private part
> > > outside of inode_lock locked region.
> > -ENOPATCH :)
>
> I meant the patch in the email you sent and to which I answered originally ;)
Sorry, I missed the 'not' when reading and though you had an alternate
patch