[patch] dentry_stat fix

[patch] dentry_stat fix

[Maneesh Soni]

Hello Andrew,

This patch the corrects the dentry_stat.nr_unused calculation.

In select_parent() and shrink_dcache_anon() we were not doing 
any adjustments to the nr_unused count after manipulating the dentry_unused 
list. Now the nr_unused count is decremented if the dentry is on dentry_unused
list and is removed from there. 

Further in the same routines, we have to adjust the nr_unused count 
again if the dentry is moved to the end of d_lru list for pruning.


Regards,
Maneesh


diff -urN linux-2.5.67-base/fs/dcache.c linux-2.5.67-dentry_stat/fs/dcache.c
--- linux-2.5.67-base/fs/dcache.c	Mon Apr  7 23:00:42 2003
+++ linux-2.5.67-dentry_stat/fs/dcache.c	Fri Apr 11 15:53:53 2003
@@ -538,13 +538,18 @@
 		struct list_head *tmp = next;
 		struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
 		next = tmp->next;
-		list_del_init(&dentry->d_lru);
 
-		/* don't add non zero d_count dentries 
-		 * back to d_lru list
+		if (!list_empty(&dentry->d_lru)) {
+			dentry_stat.nr_unused--;
+			list_del_init(&dentry->d_lru);
+		}
+		/* 
+		 * move only zero ref count dentries to the end 
+		 * of the unused list for prune_dcache
 		 */
 		if (!atomic_read(&dentry->d_count)) {
 			list_add(&dentry->d_lru, dentry_unused.prev);
+			dentry_stat.nr_unused++;
 			found++;
 		}
 		/*
@@ -609,13 +614,18 @@
 		spin_lock(&dcache_lock);
 		hlist_for_each(lp, head) {
 			struct dentry *this = hlist_entry(lp, struct dentry, d_hash);
-			list_del(&this->d_lru);
+			if (!list_empty(&this->d_lru)) {
+				dentry_stat.nr_unused--;
+				list_del(&this->d_lru);
+			}
 
-			/* don't add non zero d_count dentries 
-			 * back to d_lru list
+			/* 
+			 * move only zero ref count dentries to the end 
+			 * of the unused list for prune_dcache
 			 */
 			if (!atomic_read(&this->d_count)) {
 				list_add_tail(&this->d_lru, &dentry_unused);
+				dentry_stat.nr_unused++;
 				found++;
 			}
 		}

-- 
Maneesh Soni
IBM Linux Technology Center, 
IBM India Software Lab, Bangalore.
Phone: +91-80-5044999 email: maneesh@in.ibm.com
http://lse.sourceforge.net/

-- 
Maneesh Soni
IBM Linux Technology Center, 
IBM India Software Lab, Bangalore.
Phone: +91-80-5044999 email: maneesh@in.ibm.com
http://lse.sourceforge.net/

Re: [patch] dentry_stat fix

[Andrew Morton]

Maneesh Soni <maneesh@in.ibm.com> wrote:
>
> This patch the corrects the dentry_stat.nr_unused calculation.

OK, I didn't even know we had a bug in there...

btw, can you explain to me why shrink_dcache_anon() and select_parent() are
putting dentries at the wrong end of dentry_unused?

Do these dentries have DCACHE_REFERENCED set?

Why shouldn't they get a full two rounds of aging?

It is not clear what's going on in there.  Thanks.

Re: [patch] dentry_stat fix

[Maneesh Soni]

On Mon, Apr 14, 2003 at 02:14:48AM -0700, Andrew Morton wrote:
> Maneesh Soni <maneesh@in.ibm.com> wrote:
> >
> > This patch the corrects the dentry_stat.nr_unused calculation.
> 
> OK, I didn't even know we had a bug in there...
> 
> btw, can you explain to me why shrink_dcache_anon() and select_parent() are
> putting dentries at the wrong end of dentry_unused?
prune_dcache() picks up from this end in first round. It will reset the 
DCACHE_REFERENCED flag and will put it to the front of dentry_unused list.

> 
> Do these dentries have DCACHE_REFERENCED set?
They will have DCACHE_REFERENCED set as they are on the unused list. The flag
get reset in the first round of aging.
 
> Why shouldn't they get a full two rounds of aging?
They do get.  

> It is not clear what's going on in there.  Thanks.

I am posting one more patch which makes the DCACHE_REFERENCED flag uses 
proper following this one.

Maneesh

-- 
Maneesh Soni
IBM Linux Technology Center, 
IBM India Software Lab, Bangalore.
Phone: +91-80-5044999 email: maneesh@in.ibm.com
http://lse.sourceforge.net/

Re: [patch] dentry_stat fix

[Andrew Morton]

Maneesh Soni <maneesh@in.ibm.com> wrote:
>
> On Mon, Apr 14, 2003 at 02:14:48AM -0700, Andrew Morton wrote:
> > Maneesh Soni <maneesh@in.ibm.com> wrote:
> > >
> > > This patch the corrects the dentry_stat.nr_unused calculation.
> > 
> > OK, I didn't even know we had a bug in there...
> > 
> > btw, can you explain to me why shrink_dcache_anon() and select_parent() are
> > putting dentries at the wrong end of dentry_unused?
> prune_dcache() picks up from this end in first round. It will reset the 
> DCACHE_REFERENCED flag and will put it to the front of dentry_unused list.
> 

Sorry, but I still don't understand why they're being put at the "oldest" end
of dentry_unused.

Also, shrink_dcache_anon() does:

	prune_dcache(found);

I hope we're not assuming that all the dentries which were just added will be
freed?  They hae the referenced bit set, and new dentries can be added...