From 1b4d5b5081d9732e0761abf1b592efc979679b00 Mon Sep 17 00:00:00 2001
From: Herbert Poetzl <herbert@13thfloor.at>
Date: Sat, 20 Oct 2007 01:03:33 +0200
Subject: [PATCH] 

minor redesign for the CoW Link Break locking, to
avoid lower layer problems with unhashed dentries

the locking is now quite similar to the one done
in do_rename() except for the fact that the splice
operation is done outside the lock, but will be
avoided completely if the dentry is invalidated

[ delta-cow-fix17.diff ]
---
 fs/namei.c         |   37 +++++++++++++++++++++++--------------
 fs/super.c         |    1 -
 include/linux/fs.h |    1 -
 3 files changed, 23 insertions(+), 16 deletions(-)

diff --git a/fs/namei.c b/fs/namei.c
index 319af2e..5b7b17f 100644
--- a/fs/namei.c
+++ b/fs/namei.c
@@ -2772,7 +2772,7 @@ struct dentry *cow_break_link(const char *pathname)
 	int ret, mode, pathlen;
 	struct nameidata old_nd, dir_nd;
 	struct dentry *old_dentry, *new_dentry;
-	struct dentry *res;
+	struct dentry *res, *dir;
 	struct vfsmount *old_mnt, *new_mnt;
 	struct file *old_file;
 	struct file *new_file;
@@ -2796,12 +2796,9 @@ struct dentry *cow_break_link(const char *pathname)
 	old_mnt = old_nd.mnt;
 	mode = old_dentry->d_inode->i_mode;
 
-	/* lock filesystem during cow */
-	mutex_lock(&old_dentry->d_inode->i_sb->s_vfs_cow_mutex);
-
 	to = d_path(old_dentry, old_mnt, path, PATH_MAX-2);
 	pathlen = strlen(to);
-	vxdprintk(VXD_CBIT(misc, 2), "old path »%s« [»%*s«:%d]", to,
+	vxdprintk(VXD_CBIT(misc, 2), "old path »%s« [»%.*s«:%d]", to,
 		old_dentry->d_name.len, old_dentry->d_name.name,
 		old_dentry->d_name.len);
 
@@ -2815,7 +2812,7 @@ retry:
 	vxdprintk(VXD_CBIT(misc, 1), "temp copy »%s«", to);
 	/* dir_nd will have refs to dentry and mnt */
 	ret = path_lookup(to,
-		LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE, &dir_nd);
+		LOOKUP_PARENT | LOOKUP_OPEN | LOOKUP_CREATE, &dir_nd);
 	vxdprintk(VXD_CBIT(misc, 2),
 		"path_lookup(new): %d", ret);
 	if (ret < 0)
@@ -2824,19 +2821,20 @@ retry:
 	/* this puppy downs the inode mutex */
 	new_dentry = lookup_create(&dir_nd, 0);
 	vxdprintk(VXD_CBIT(misc, 2),
-		"lookup_create(new): %p [»%*s«:%d]", new_dentry,
+		"lookup_create(new): %p [»%.*s«:%d]", new_dentry,
 		new_dentry->d_name.len, new_dentry->d_name.name,
 		new_dentry->d_name.len);
 	if (!new_dentry || IS_ERR(new_dentry)) {
 		path_release(&dir_nd);
 		goto retry;
 	}
+	dir = dir_nd.dentry;
 
 	ret = vfs_create(dir_nd.dentry->d_inode, new_dentry, mode, &dir_nd);
 	vxdprintk(VXD_CBIT(misc, 2),
 		"vfs_create(new): %d", ret);
 	if (ret == -EEXIST) {
-		mutex_unlock(&dir_nd.dentry->d_inode->i_mutex);
+		mutex_unlock(&dir->d_inode->i_mutex);
 		dput(new_dentry);
 		path_release(&dir_nd);
 		goto retry;
@@ -2846,8 +2844,13 @@ retry:
 		goto out_unlock_new;
 	}
 
-	new_mnt = dir_nd.mnt;
+	/* drop out early */
+	if (d_unhashed(old_dentry)) {
+		res = ERR_PTR(-ENOENT);
+		goto out_unlock_new;
+	}
 
+	new_mnt = dir_nd.mnt;
 	dget(old_dentry);
 	mntget(old_mnt);
 	/* this one cleans up the dentry/mnt in case of failure */
@@ -2891,11 +2894,15 @@ retry:
 		ret = inode_setattr(new_inode, &attr);
 		if (ret) {
 			res = ERR_PTR(ret);
-			goto out_fput_both;	
+			goto out_fput_both;
 		}
 	}
 
 	mutex_lock(&old_dentry->d_inode->i_sb->s_vfs_rename_mutex);
+	ret = -ENOENT;
+	if (d_unhashed(old_dentry))
+		goto out_unlock;
+
 	vxdprintk(VXD_CBIT(misc, 2),
 		"vfs_rename: [»%*s«:%d] -> [»%*s«:%d]",
 		new_dentry->d_name.len, new_dentry->d_name.name,
@@ -2904,8 +2911,11 @@ retry:
 		old_dentry->d_name.len);
 	ret = vfs_rename(dir_nd.dentry->d_inode, new_dentry,
 		old_nd.dentry->d_parent->d_inode, old_dentry);
-	mutex_unlock(&old_dentry->d_inode->i_sb->s_vfs_rename_mutex);
 	vxdprintk(VXD_CBIT(misc, 2), "vfs_rename: %d", ret);
+
+out_unlock:
+	mutex_unlock(&old_dentry->d_inode->i_sb->s_vfs_rename_mutex);
+
 	if (!ret) {
 		res = new_dentry;
 		dget(new_dentry);
@@ -2926,14 +2936,13 @@ out_fput_old:
 	fput(old_file);
 
 out_unlock_new:
-	mutex_unlock(&dir_nd.dentry->d_inode->i_mutex);
+	mutex_unlock(&dir->d_inode->i_mutex);
 	if (IS_ERR(res))
-		vfs_unlink(dir_nd.dentry->d_inode, new_dentry, &dir_nd);
+		vfs_unlink(dir->d_inode, new_dentry, &dir_nd);
 	dput(new_dentry);
 out_rel_both:
 	path_release(&dir_nd);
 out_rel_old:
-	mutex_unlock(&old_dentry->d_inode->i_sb->s_vfs_cow_mutex);
 	path_release(&old_nd);
 out_free_path:
 	kfree(path);
diff --git a/fs/super.c b/fs/super.c
index 2903ecf..8ee97d3 100644
--- a/fs/super.c
+++ b/fs/super.c
@@ -87,7 +87,6 @@ static struct super_block *alloc_super(struct file_system_type *type)
 		s->s_count = S_BIAS;
 		atomic_set(&s->s_active, 1);
 		mutex_init(&s->s_vfs_rename_mutex);
-		mutex_init(&s->s_vfs_cow_mutex);
 		mutex_init(&s->s_dquot.dqio_mutex);
 		mutex_init(&s->s_dquot.dqonoff_mutex);
 		init_rwsem(&s->s_dquot.dqptr_sem);
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 9418ab3..5605b8f 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -979,7 +979,6 @@ struct super_block {
 	 * even looking at it. You had been warned.
 	 */
 	struct mutex s_vfs_rename_mutex;	/* Kludge */
-	struct mutex s_vfs_cow_mutex;
 
 	/* Granularity of c/m/atime in ns.
 	   Cannot be worse than a second */
-- 
1.5.2.5