4 files changed, 94 insertions, 41 deletions

diff --git a/fs/ocfs2/cluster/tcp_internal.h b/fs/ocfs2/cluster/tcp_internal.h
index ff9e2e2104c2..da42b515cd1d 100644
--- a/fs/ocfs2/cluster/tcp_internal.h
+++ b/fs/ocfs2/cluster/tcp_internal.h
@@ -44,11 +44,14 @@
  * locking semantics of the file system using the protocol.  It should 
  * be somewhere else, I'm sure, but right now it isn't.
  *
+ * New in version 3:
+ *      - Replace dentry votes with a cluster lock
+ *
  * New in version 2:
  *      - full 64 bit i_size in the metadata lock lvbs
  *      - introduction of "rw" lock and pushing meta/data locking down
  */
-#define O2NET_PROTOCOL_VERSION 2ULL
+#define O2NET_PROTOCOL_VERSION 3ULL
 struct o2net_handshake {
         __be64  protocol_version;
         __be64  connector_id;
diff --git a/fs/ocfs2/export.c b/fs/ocfs2/export.c
index ec55ab3c1214..ffcd79749e0d 100644
--- a/fs/ocfs2/export.c
+++ b/fs/ocfs2/export.c
@@ -33,6 +33,7 @@
 
 #include "dir.h"
 #include "dlmglue.h"
+#include "dcache.h"
 #include "export.h"
 #include "inode.h"
 
@@ -77,6 +78,7 @@ static struct dentry *ocfs2_get_dentry(struct super_block *sb, void *vobjp)
                 mlog_errno(-ENOMEM);
                 return ERR_PTR(-ENOMEM);
         }
+        result->d_op = &ocfs2_dentry_ops;
 
         mlog_exit_ptr(result);
         return result;
@@ -127,6 +129,8 @@ static struct dentry *ocfs2_get_parent(struct dentry *child)
                 parent = ERR_PTR(-ENOMEM);
         }
 
+        parent->d_op = &ocfs2_dentry_ops;
+
 bail_unlock:
         ocfs2_meta_unlock(dir, 0);
 
diff --git a/fs/ocfs2/inode.c b/fs/ocfs2/inode.c
index 7bcf69154592..66ca7a82b68a 100644
--- a/fs/ocfs2/inode.c
+++ b/fs/ocfs2/inode.c
@@ -1025,12 +1025,10 @@ void ocfs2_drop_inode(struct inode *inode)
         /* Testing ip_orphaned_slot here wouldn't work because we may
          * not have gotten a delete_inode vote from any other nodes
          * yet. */
-        if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED) {
-                mlog(0, "Inode was orphaned on another node, clearing nlink.\n");
-                inode->i_nlink = 0;
-        }
-
-        generic_drop_inode(inode);
+        if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
+                generic_delete_inode(inode);
+        else
+                generic_drop_inode(inode);
 
         mlog_exit_void();
 }
diff --git a/fs/ocfs2/namei.c b/fs/ocfs2/namei.c
index 0d3e939b1f56..5a942e0123ea 100644
--- a/fs/ocfs2/namei.c
+++ b/fs/ocfs2/namei.c
@@ -199,10 +199,32 @@ static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
         spin_unlock(&oi->ip_lock);
 
 bail_add:
-
         dentry->d_op = &ocfs2_dentry_ops;
         ret = d_splice_alias(inode, dentry);
 
+        if (inode) {
+                /*
+                 * If d_splice_alias() finds a DCACHE_DISCONNECTED
+                 * dentry, it will d_move() it on top of ourse. The
+                 * return value will indicate this however, so in
+                 * those cases, we switch them around for the locking
+                 * code.
+                 *
+                 * NOTE: This dentry already has ->d_op set from
+                 * ocfs2_get_parent() and ocfs2_get_dentry()
+                 */
+                if (ret)
+                        dentry = ret;
+
+                status = ocfs2_dentry_attach_lock(dentry, inode,
+                                                  OCFS2_I(dir)->ip_blkno, 0);
+                if (status) {
+                        mlog_errno(status);
+                        ret = ERR_PTR(status);
+                        goto bail_unlock;
+                }
+        }
+
 bail_unlock:
         /* Don't drop the cluster lock until *after* the d_add --
          * unlink on another node will message us to remove that
@@ -418,6 +440,13 @@ static int ocfs2_mknod(struct inode *dir,
                 goto leave;
         }
 
+        status = ocfs2_dentry_attach_lock(dentry, inode,
+                                          OCFS2_I(dir)->ip_blkno, 1);
+        if (status) {
+                mlog_errno(status);
+                goto leave;
+        }
+
         insert_inode_hash(inode);
         dentry->d_op = &ocfs2_dentry_ops;
         d_instantiate(dentry, inode);
@@ -725,6 +754,13 @@ static int ocfs2_link(struct dentry *old_dentry,
                 goto bail;
         }
 
+        err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno,
+                                       0);
+        if (err) {
+                mlog_errno(err);
+                goto bail;
+        }
+
         atomic_inc(&inode->i_count);
         dentry->d_op = &ocfs2_dentry_ops;
         d_instantiate(dentry, inode);
@@ -743,6 +779,23 @@ bail:
         return err;
 }
 
+/*
+ * Takes and drops an exclusive lock on the given dentry. This will
+ * force other nodes to drop it.
+ */
+static int ocfs2_remote_dentry_delete(struct dentry *dentry)
+{
+        int ret;
+
+        ret = ocfs2_dentry_lock(dentry, 1);
+        if (ret)
+                mlog_errno(ret);
+        else
+                ocfs2_dentry_unlock(dentry, 1);
+
+        return ret;
+}
+
 static int ocfs2_unlink(struct inode *dir,
                         struct dentry *dentry)
 {
@@ -832,8 +885,7 @@ static int ocfs2_unlink(struct inode *dir,
         else
                 inode->i_nlink--;
 
-        status = ocfs2_request_unlink_vote(inode, dentry,
-                                           (unsigned int) inode->i_nlink);
+        status = ocfs2_remote_dentry_delete(dentry);
         if (status < 0) {
                 /* This vote should succeed under all normal
                  * circumstances. */
@@ -1019,7 +1071,6 @@ static int ocfs2_rename(struct inode *old_dir,
         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
                                                     // this is the 1st dirent bh
         nlink_t old_dir_nlink = old_dir->i_nlink, new_dir_nlink = new_dir->i_nlink;
-        unsigned int links_count;
 
         /* At some point it might be nice to break this function up a
          * bit. */
@@ -1093,23 +1144,26 @@ static int ocfs2_rename(struct inode *old_dir,
                 }
         }
 
-        if (S_ISDIR(old_inode->i_mode)) {
-                /* Directories actually require metadata updates to
-                 * the directory info so we can't get away with not
-                 * doing node locking on it. */
-                status = ocfs2_meta_lock(old_inode, handle, NULL, 1);
-                if (status < 0) {
-                        if (status != -ENOENT)
-                                mlog_errno(status);
-                        goto bail;
-                }
-
-                status = ocfs2_request_rename_vote(old_inode, old_dentry);
-                if (status < 0) {
+        /*
+         * Though we don't require an inode meta data update if
+         * old_inode is not a directory, we lock anyway here to ensure
+         * the vote thread on other nodes won't have to concurrently
+         * downconvert the inode and the dentry locks.
+         */
+        status = ocfs2_meta_lock(old_inode, handle, NULL, 1);
+        if (status < 0) {
+                if (status != -ENOENT)
                         mlog_errno(status);
-                        goto bail;
-                }
+                goto bail;
+        }
+
+        status = ocfs2_remote_dentry_delete(old_dentry);
+        if (status < 0) {
+                mlog_errno(status);
+                goto bail;
+        }
 
+        if (S_ISDIR(old_inode->i_mode)) {
                 status = -EIO;
                 old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
                 if (!old_inode_de_bh)
@@ -1123,14 +1177,6 @@ static int ocfs2_rename(struct inode *old_dir,
                 if (!new_inode && new_dir!=old_dir &&
                     new_dir->i_nlink >= OCFS2_LINK_MAX)
                         goto bail;
-        } else {
-                /* Ah, the simple case - we're a file so just send a
-                 * message. */
-                status = ocfs2_request_rename_vote(old_inode, old_dentry);
-                if (status < 0) {
-                        mlog_errno(status);
-                        goto bail;
-                }
         }
 
         status = -ENOENT;
@@ -1202,13 +1248,7 @@ static int ocfs2_rename(struct inode *old_dir,
                         goto bail;
                 }
 
-                if (S_ISDIR(new_inode->i_mode))
-                        links_count = 0;
-                else
-                        links_count = (unsigned int) (new_inode->i_nlink - 1);
-
-                status = ocfs2_request_unlink_vote(new_inode, new_dentry,
-                                                   links_count);
+                status = ocfs2_remote_dentry_delete(new_dentry);
                 if (status < 0) {
                         mlog_errno(status);
                         goto bail;
@@ -1387,6 +1427,7 @@ static int ocfs2_rename(struct inode *old_dir,
                 }
         }
 
+        ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
         status = 0;
 bail:
         if (rename_lock)
@@ -1675,6 +1716,13 @@ static int ocfs2_symlink(struct inode *dir,
                 goto bail;
         }
 
+        status = ocfs2_dentry_attach_lock(dentry, inode,
+                                          OCFS2_I(dir)->ip_blkno, 1);
+        if (status) {
+                mlog_errno(status);
+                goto bail;
+        }
+
         insert_inode_hash(inode);
         dentry->d_op = &ocfs2_dentry_ops;
         d_instantiate(dentry, inode);