ok this is a first attempt at changing notify_change().....
please look at the following code for any obvious issues. especially the
wait_on_inode functions. I am not sure if a new specific wait_on_attr_inode
should be created or if this change is OK. I am trying to remove the BKL (--
so any "please review all filesystems" comments are unneeded as I know this.)
from the VFS layer and this is just one small portion of that effort.
Craig
Index: linux/fs//attr.c
===================================================================
RCS file: /home/Media/cvs/linux/fs/attr.c,v
retrieving revision 1.1.1.1
diff -u -r1.1.1.1 attr.c
--- linux/fs//attr.c 16 Jan 2002 18:26:33 -0000 1.1.1.1
+++ linux/fs//attr.c 17 Jan 2002 23:30:14 -0000
@@ -111,10 +111,13 @@
return dn_mask;
}
+/* inode_lock is needed to protect i_state */
+
+extern spinlock_t inode_lock;
int notify_change(struct dentry * dentry, struct iattr * attr)
{
struct inode *inode = dentry->d_inode;
- int error;
+ int error,tflag;
time_t now = CURRENT_TIME;
unsigned int ia_valid = attr->ia_valid;
@@ -127,7 +130,25 @@
if (!(ia_valid & ATTR_MTIME_SET))
attr->ia_mtime = now;
- lock_kernel();
+ /*
+ * The special i_state flag I_ATTR_LOCK protects
+ * the following code block this is for serialization
+ * purposes previously managed by lock_kernel();
+ */
+
+ do {
+ wait_on_inode(inode,I_ATTR_LOCK);
+ spin_lock(&inode_lock);
+ if(inode->i_state & I_ATTR_LOCK) {
+ spin_unlock(&inode_lock);
+ }
+ else {
+ inode->i_state =| I_ATTR_LOCK;
+ tflag = 1;
+ spin_unlock(&inode_lock);
+ }
+ } while(!tflag);
+
if (inode->i_op && inode->i_op->setattr)
error = inode->i_op->setattr(dentry, attr);
else {
@@ -140,7 +161,9 @@
error = inode_setattr(inode, attr);
}
}
- unlock_kernel();
+ spin_lock(&inode_lock);
+ inode->i_state =& ~I_ATTR_LOCK;
+ spin_unlock(&inode_lock);
if (!error) {
unsigned long dn_mask = setattr_mask(ia_valid);
if (dn_mask)
Index: linux/fs//inode.c
===================================================================
RCS file: /home/Media/cvs/linux/fs/inode.c,v
retrieving revision 1.1.1.1
diff -u -r1.1.1.1 inode.c
--- linux/fs//inode.c 16 Jan 2002 18:26:32 -0000 1.1.1.1
+++ linux/fs//inode.c 17 Jan 2002 23:01:57 -0000
@@ -163,14 +163,14 @@
spin_unlock(&inode_lock);
}
-static void __wait_on_inode(struct inode * inode)
+static void __wait_on_inode(struct inode * inode,int flag)
{
DECLARE_WAITQUEUE(wait, current);
add_wait_queue(&inode->i_wait, &wait);
repeat:
set_current_state(TASK_UNINTERRUPTIBLE);
- if (inode->i_state & I_LOCK) {
+ if (inode->i_state & flag) {
schedule();
goto repeat;
}
@@ -178,13 +178,6 @@
current->state = TASK_RUNNING;
}
-static inline void wait_on_inode(struct inode *inode)
-{
- if (inode->i_state & I_LOCK)
- __wait_on_inode(inode);
-}
-
-
static inline void write_inode(struct inode *inode, int sync)
{
if (inode->i_sb && inode->i_sb->s_op && inode->i_sb->s_op->write_inode &&
!is_bad_inode(inode))
@@ -250,7 +243,7 @@
if (inode->i_state & I_LOCK) {
__iget(inode);
spin_unlock(&inode_lock);
- __wait_on_inode(inode);
+ __wait_on_inode(inode,I_LOCK);
iput(inode);
spin_lock(&inode_lock);
} else {
@@ -273,7 +266,7 @@
struct inode *inode = list_entry(tmp, struct inode, i_list);
__iget(inode);
spin_unlock(&inode_lock);
- __wait_on_inode(inode);
+ __wait_on_inode(inode,I_LOCK);
iput(inode);
spin_lock(&inode_lock);
}
@@ -435,7 +428,7 @@
sync_one(inode, sync);
spin_unlock(&inode_lock);
if (sync)
- wait_on_inode(inode);
+ wait_on_inode(inode,I_LOCK);
}
else
printk(KERN_ERR "write_inode_now: no super block\n");
@@ -491,7 +484,7 @@
if (need_write_inode_now)
write_inode_now(inode, 1);
else
- wait_on_inode(inode);
+ wait_on_inode(inode,I_LOCK);
return err;
}
@@ -515,7 +508,7 @@
BUG();
if (inode->i_state & I_CLEAR)
BUG();
- wait_on_inode(inode);
+ wait_on_inode(inode,I_LOCK);
DQUOT_DROP(inode);
if (inode->i_sb && inode->i_sb->s_op && inode->i_sb->s_op->clear_inode)
inode->i_sb->s_op->clear_inode(inode);
@@ -904,7 +897,7 @@
spin_unlock(&inode_lock);
destroy_inode(inode);
inode = old;
- wait_on_inode(inode);
+ wait_on_inode(inode,I_LOCK);
}
return inode;
}
@@ -982,7 +975,7 @@
if (inode) {
__iget(inode);
spin_unlock(&inode_lock);
- wait_on_inode(inode);
+ wait_on_inode(inode,I_LOCK);
return inode;
}
spin_unlock(&inode_lock);
Index: linux/include/linux//fs.h
===================================================================
RCS file: /home/Media/cvs/linux/include/linux/fs.h,v
retrieving revision 1.1.1.1
diff -u -r1.1.1.1 fs.h
--- linux/include/linux//fs.h 16 Jan 2002 18:28:27 -0000 1.1.1.1
+++ linux/include/linux//fs.h 17 Jan 2002 22:45:23 -0000
@@ -915,9 +915,15 @@
#define I_LOCK 8
#define I_FREEING 16
#define I_CLEAR 32
+#define I_ADDR_LOCK 64
#define I_DIRTY (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_PAGES)
-
+extern void __wait_on_inode(struct inode *,int);
+static inline void wait_on_inode(struct inode *inode, int flag);
+{
+ if (inode->i_state & flag)
+ __wait_on_inode(inode,flag);
+}
extern void __mark_inode_dirty(struct inode *, int);
static inline void mark_inode_dirty(struct inode *inode)
{
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Wed Jan 23 2002 - 21:00:23 EST