Re: Linux 2.5.71

[Trond Myklebust <trond.myklebust@fys.uio.no>]

>>>>> " " == Linus Torvalds <torvalds@transmeta.com> writes:

     > Trond, any ideas?

Here's the patch I posted to Bugzilla earlier today. The problem is
that whereas using the dentry->d_hash list for private purposes was
previously harmless, it is not allowed with the RCU lists, because we
depend on the unhashed node continuing to point back to the original
list.

Cheers,
   Trond

diff -u --recursive --new-file linux-2.5.71/net/sunrpc/rpc_pipe.c linux-2.5.71-fix_rpcpipe/net/sunrpc/rpc_pipe.c
--- linux-2.5.71/net/sunrpc/rpc_pipe.c	2003-06-11 19:24:29.000000000 -0700
+++ linux-2.5.71-fix_rpcpipe/net/sunrpc/rpc_pipe.c	2003-06-14 16:58:21.000000000 -0700
@@ -472,30 +472,37 @@
 rpc_depopulate(struct dentry *parent)
 {
 	struct inode *dir = parent->d_inode;
-	HLIST_HEAD(head);
 	struct list_head *pos, *next;
-	struct dentry *dentry;
+	struct dentry *dentry, *dvec[10];
+	int n = 0;
 
 	down(&dir->i_sem);
+repeat:
 	spin_lock(&dcache_lock);
 	list_for_each_safe(pos, next, &parent->d_subdirs) {
 		dentry = list_entry(pos, struct dentry, d_child);
+		spin_lock(&dentry->d_lock);
 		if (!d_unhashed(dentry)) {
 			dget_locked(dentry);
 			__d_drop(dentry);
-			hlist_add_head(&dentry->d_hash, &head);
-		}
+			spin_unlock(&dentry->d_lock);
+			dvec[n++] = dentry;
+			if (n == ARRAY_SIZE(dvec))
+				break;
+		} else
+			spin_unlock(&dentry->d_lock);
 	}
 	spin_unlock(&dcache_lock);
-	while (!hlist_empty(&head)) {
-		dentry = list_entry(head.first, struct dentry, d_hash);
-		/* Private list, so no dcache_lock needed and use __d_drop */
-		__d_drop(dentry);
-		if (dentry->d_inode) {
-			rpc_inode_setowner(dentry->d_inode, NULL);
-			simple_unlink(dir, dentry);
-		}
-		dput(dentry);
+	if (n) {
+		do {
+			dentry = dvec[--n];
+			if (dentry->d_inode) {
+				rpc_inode_setowner(dentry->d_inode, NULL);
+				simple_unlink(dir, dentry);
+			}
+			dput(dentry);
+		} while (n);
+		goto repeat;
 	}
 	up(&dir->i_sem);
 }