Merge branch 'alloc_path' of git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh...
[linux-2.6.git] / fs / btrfs / tree-log.c
index 3a72a1b..babee65 100644 (file)
@@ -17,6 +17,7 @@
  */
 
 #include <linux/sched.h>
+#include <linux/slab.h>
 #include "ctree.h"
 #include "transaction.h"
 #include "disk-io.h"
 #define LOG_INODE_EXISTS 1
 
 /*
+ * directory trouble cases
+ *
+ * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
+ * log, we must force a full commit before doing an fsync of the directory
+ * where the unlink was done.
+ * ---> record transid of last unlink/rename per directory
+ *
+ * mkdir foo/some_dir
+ * normal commit
+ * rename foo/some_dir foo2/some_dir
+ * mkdir foo/some_dir
+ * fsync foo/some_dir/some_file
+ *
+ * The fsync above will unlink the original some_dir without recording
+ * it in its new location (foo2).  After a crash, some_dir will be gone
+ * unless the fsync of some_file forces a full commit
+ *
+ * 2) we must log any new names for any file or dir that is in the fsync
+ * log. ---> check inode while renaming/linking.
+ *
+ * 2a) we must log any new names for any file or dir during rename
+ * when the directory they are being removed from was logged.
+ * ---> check inode and old parent dir during rename
+ *
+ *  2a is actually the more important variant.  With the extra logging
+ *  a crash might unlink the old name without recreating the new one
+ *
+ * 3) after a crash, we must go through any directories with a link count
+ * of zero and redo the rm -rf
+ *
+ * mkdir f1/foo
+ * normal commit
+ * rm -rf f1/foo
+ * fsync(f1)
+ *
+ * The directory f1 was fully removed from the FS, but fsync was never
+ * called on f1, only its parent dir.  After a crash the rm -rf must
+ * be replayed.  This must be able to recurse down the entire
+ * directory tree.  The inode link count fixup code takes care of the
+ * ugly details.
+ */
+
+/*
  * stages for the tree walking.  The first
  * stage (0) is to only pin down the blocks we find
  * the second stage (1) is to make sure that all the inodes
 #define LOG_WALK_REPLAY_INODES 1
 #define LOG_WALK_REPLAY_ALL 2
 
-static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
+static int btrfs_log_inode(struct btrfs_trans_handle *trans,
                             struct btrfs_root *root, struct inode *inode,
                             int inode_only);
 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
                             struct btrfs_root *root,
                             struct btrfs_path *path, u64 objectid);
+static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
+                                      struct btrfs_root *root,
+                                      struct btrfs_root *log,
+                                      struct btrfs_path *path,
+                                      u64 dirid, int del_all);
 
 /*
  * tree logging is a special write ahead log used to make sure that
@@ -78,91 +127,6 @@ static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
  */
 
 /*
- * btrfs_add_log_tree adds a new per-subvolume log tree into the
- * tree of log tree roots.  This must be called with a tree log transaction
- * running (see start_log_trans).
- */
-static int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
-                     struct btrfs_root *root)
-{
-       struct btrfs_key key;
-       struct btrfs_root_item root_item;
-       struct btrfs_inode_item *inode_item;
-       struct extent_buffer *leaf;
-       struct btrfs_root *new_root = root;
-       int ret;
-       u64 objectid = root->root_key.objectid;
-
-       leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
-                                     BTRFS_TREE_LOG_OBJECTID,
-                                     trans->transid, 0, 0, 0);
-       if (IS_ERR(leaf)) {
-               ret = PTR_ERR(leaf);
-               return ret;
-       }
-
-       btrfs_set_header_nritems(leaf, 0);
-       btrfs_set_header_level(leaf, 0);
-       btrfs_set_header_bytenr(leaf, leaf->start);
-       btrfs_set_header_generation(leaf, trans->transid);
-       btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
-
-       write_extent_buffer(leaf, root->fs_info->fsid,
-                           (unsigned long)btrfs_header_fsid(leaf),
-                           BTRFS_FSID_SIZE);
-       btrfs_mark_buffer_dirty(leaf);
-
-       inode_item = &root_item.inode;
-       memset(inode_item, 0, sizeof(*inode_item));
-       inode_item->generation = cpu_to_le64(1);
-       inode_item->size = cpu_to_le64(3);
-       inode_item->nlink = cpu_to_le32(1);
-       inode_item->nbytes = cpu_to_le64(root->leafsize);
-       inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
-
-       btrfs_set_root_bytenr(&root_item, leaf->start);
-       btrfs_set_root_generation(&root_item, trans->transid);
-       btrfs_set_root_level(&root_item, 0);
-       btrfs_set_root_refs(&root_item, 0);
-       btrfs_set_root_used(&root_item, 0);
-
-       memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
-       root_item.drop_level = 0;
-
-       btrfs_tree_unlock(leaf);
-       free_extent_buffer(leaf);
-       leaf = NULL;
-
-       btrfs_set_root_dirid(&root_item, 0);
-
-       key.objectid = BTRFS_TREE_LOG_OBJECTID;
-       key.offset = objectid;
-       btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
-       ret = btrfs_insert_root(trans, root->fs_info->log_root_tree, &key,
-                               &root_item);
-       if (ret)
-               goto fail;
-
-       new_root = btrfs_read_fs_root_no_radix(root->fs_info->log_root_tree,
-                                              &key);
-       BUG_ON(!new_root);
-
-       WARN_ON(root->log_root);
-       root->log_root = new_root;
-
-       /*
-        * log trees do not get reference counted because they go away
-        * before a real commit is actually done.  They do store pointers
-        * to file data extents, and those reference counts still get
-        * updated (along with back refs to the log tree).
-        */
-       new_root->ref_cows = 0;
-       new_root->last_trans = trans->transid;
-fail:
-       return ret;
-}
-
-/*
  * start a sub transaction and setup the log tree
  * this increments the log tree writer count to make the people
  * syncing the tree wait for us to finish
@@ -171,19 +135,40 @@ static int start_log_trans(struct btrfs_trans_handle *trans,
                           struct btrfs_root *root)
 {
        int ret;
+       int err = 0;
+
+       mutex_lock(&root->log_mutex);
+       if (root->log_root) {
+               if (!root->log_start_pid) {
+                       root->log_start_pid = current->pid;
+                       root->log_multiple_pids = false;
+               } else if (root->log_start_pid != current->pid) {
+                       root->log_multiple_pids = true;
+               }
+
+               root->log_batch++;
+               atomic_inc(&root->log_writers);
+               mutex_unlock(&root->log_mutex);
+               return 0;
+       }
+       root->log_multiple_pids = false;
+       root->log_start_pid = current->pid;
        mutex_lock(&root->fs_info->tree_log_mutex);
        if (!root->fs_info->log_root_tree) {
                ret = btrfs_init_log_root_tree(trans, root->fs_info);
-               BUG_ON(ret);
+               if (ret)
+                       err = ret;
        }
-       if (!root->log_root) {
+       if (err == 0 && !root->log_root) {
                ret = btrfs_add_log_tree(trans, root);
-               BUG_ON(ret);
+               if (ret)
+                       err = ret;
        }
-       atomic_inc(&root->fs_info->tree_log_writers);
-       root->fs_info->tree_log_batch++;
        mutex_unlock(&root->fs_info->tree_log_mutex);
-       return 0;
+       root->log_batch++;
+       atomic_inc(&root->log_writers);
+       mutex_unlock(&root->log_mutex);
+       return err;
 }
 
 /*
@@ -199,13 +184,27 @@ static int join_running_log_trans(struct btrfs_root *root)
        if (!root->log_root)
                return -ENOENT;
 
-       mutex_lock(&root->fs_info->tree_log_mutex);
+       mutex_lock(&root->log_mutex);
        if (root->log_root) {
                ret = 0;
-               atomic_inc(&root->fs_info->tree_log_writers);
-               root->fs_info->tree_log_batch++;
+               atomic_inc(&root->log_writers);
        }
-       mutex_unlock(&root->fs_info->tree_log_mutex);
+       mutex_unlock(&root->log_mutex);
+       return ret;
+}
+
+/*
+ * This either makes the current running log transaction wait
+ * until you call btrfs_end_log_trans() or it makes any future
+ * log transactions wait until you call btrfs_end_log_trans()
+ */
+int btrfs_pin_log_trans(struct btrfs_root *root)
+{
+       int ret = -ENOENT;
+
+       mutex_lock(&root->log_mutex);
+       atomic_inc(&root->log_writers);
+       mutex_unlock(&root->log_mutex);
        return ret;
 }
 
@@ -213,12 +212,13 @@ static int join_running_log_trans(struct btrfs_root *root)
  * indicate we're done making changes to the log tree
  * and wake up anyone waiting to do a sync
  */
-static int end_log_trans(struct btrfs_root *root)
+int btrfs_end_log_trans(struct btrfs_root *root)
 {
-       atomic_dec(&root->fs_info->tree_log_writers);
-       smp_mb();
-       if (waitqueue_active(&root->fs_info->tree_log_wait))
-               wake_up(&root->fs_info->tree_log_wait);
+       if (atomic_dec_and_test(&root->log_writers)) {
+               smp_mb();
+               if (waitqueue_active(&root->log_writer_wait))
+                       wake_up(&root->log_writer_wait);
+       }
        return 0;
 }
 
@@ -275,12 +275,9 @@ static int process_one_buffer(struct btrfs_root *log,
                              struct extent_buffer *eb,
                              struct walk_control *wc, u64 gen)
 {
-       if (wc->pin) {
-               mutex_lock(&log->fs_info->pinned_mutex);
-               btrfs_update_pinned_extents(log->fs_info->extent_root,
-                                           eb->start, eb->len, 1);
-               mutex_unlock(&log->fs_info->pinned_mutex);
-       }
+       if (wc->pin)
+               btrfs_pin_extent(log->fs_info->extent_root,
+                                eb->start, eb->len, 0);
 
        if (btrfs_buffer_uptodate(eb, gen)) {
                if (wc->write)
@@ -336,11 +333,17 @@ static noinline int overwrite_item(struct btrfs_trans_handle *trans,
                        goto insert;
 
                if (item_size == 0) {
-                       btrfs_release_path(root, path);
+                       btrfs_release_path(path);
                        return 0;
                }
                dst_copy = kmalloc(item_size, GFP_NOFS);
                src_copy = kmalloc(item_size, GFP_NOFS);
+               if (!dst_copy || !src_copy) {
+                       btrfs_release_path(path);
+                       kfree(dst_copy);
+                       kfree(src_copy);
+                       return -ENOMEM;
+               }
 
                read_extent_buffer(eb, src_copy, src_ptr, item_size);
 
@@ -358,13 +361,13 @@ static noinline int overwrite_item(struct btrfs_trans_handle *trans,
                 * sync
                 */
                if (ret == 0) {
-                       btrfs_release_path(root, path);
+                       btrfs_release_path(path);
                        return 0;
                }
 
        }
 insert:
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
        /* try to insert the key into the destination tree */
        ret = btrfs_insert_empty_item(trans, root, path,
                                      key, item_size);
@@ -379,10 +382,9 @@ insert:
                } else if (found_size < item_size) {
                        ret = btrfs_extend_item(trans, root, path,
                                                item_size - found_size);
-                       BUG_ON(ret);
                }
        } else if (ret) {
-               BUG();
+               return ret;
        }
        dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
                                        path->slots[0]);
@@ -433,52 +435,9 @@ insert:
                                                   trans->transid);
                }
        }
-
-       if (overwrite_root &&
-           key->type == BTRFS_EXTENT_DATA_KEY) {
-               int extent_type;
-               struct btrfs_file_extent_item *fi;
-
-               fi = (struct btrfs_file_extent_item *)dst_ptr;
-               extent_type = btrfs_file_extent_type(path->nodes[0], fi);
-               if (extent_type == BTRFS_FILE_EXTENT_REG ||
-                   extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
-                       struct btrfs_key ins;
-                       ins.objectid = btrfs_file_extent_disk_bytenr(
-                                                       path->nodes[0], fi);
-                       ins.offset = btrfs_file_extent_disk_num_bytes(
-                                                       path->nodes[0], fi);
-                       ins.type = BTRFS_EXTENT_ITEM_KEY;
-
-                       /*
-                        * is this extent already allocated in the extent
-                        * allocation tree?  If so, just add a reference
-                        */
-                       ret = btrfs_lookup_extent(root, ins.objectid,
-                                                 ins.offset);
-                       if (ret == 0) {
-                               ret = btrfs_inc_extent_ref(trans, root,
-                                               ins.objectid, ins.offset,
-                                               path->nodes[0]->start,
-                                               root->root_key.objectid,
-                                               trans->transid, key->objectid);
-                       } else {
-                               /*
-                                * insert the extent pointer in the extent
-                                * allocation tree
-                                */
-                               ret = btrfs_alloc_logged_extent(trans, root,
-                                               path->nodes[0]->start,
-                                               root->root_key.objectid,
-                                               trans->transid, key->objectid,
-                                               &ins);
-                               BUG_ON(ret);
-                       }
-               }
-       }
 no_copy:
        btrfs_mark_buffer_dirty(path->nodes[0]);
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
        return 0;
 }
 
@@ -489,18 +448,16 @@ no_copy:
 static noinline struct inode *read_one_inode(struct btrfs_root *root,
                                             u64 objectid)
 {
+       struct btrfs_key key;
        struct inode *inode;
-       inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
-       if (inode->i_state & I_NEW) {
-               BTRFS_I(inode)->root = root;
-               BTRFS_I(inode)->location.objectid = objectid;
-               BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
-               BTRFS_I(inode)->location.offset = 0;
-               btrfs_read_locked_inode(inode);
-               unlock_new_inode(inode);
 
-       }
-       if (is_bad_inode(inode)) {
+       key.objectid = objectid;
+       key.type = BTRFS_INODE_ITEM_KEY;
+       key.offset = 0;
+       inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
+       if (IS_ERR(inode)) {
+               inode = NULL;
+       } else if (is_bad_inode(inode)) {
                iput(inode);
                inode = NULL;
        }
@@ -530,6 +487,7 @@ static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
        u64 extent_end;
        u64 alloc_hint;
        u64 start = key->offset;
+       u64 saved_nbytes;
        struct btrfs_file_extent_item *item;
        struct inode *inode = NULL;
        unsigned long size;
@@ -560,7 +518,7 @@ static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
         * file.  This must be done before the btrfs_drop_extents run
         * so we don't try to drop this extent.
         */
-       ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
+       ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
                                       start, 0);
 
        if (ret == 0 &&
@@ -585,23 +543,101 @@ static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
                 * we don't have to do anything
                 */
                if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
-                       btrfs_release_path(root, path);
+                       btrfs_release_path(path);
                        goto out;
                }
        }
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
 
+       saved_nbytes = inode_get_bytes(inode);
        /* drop any overlapping extents */
-       ret = btrfs_drop_extents(trans, root, inode,
-                        start, extent_end, start, &alloc_hint);
+       ret = btrfs_drop_extents(trans, inode, start, extent_end,
+                                &alloc_hint, 1);
        BUG_ON(ret);
 
-       /* insert the extent */
-       ret = overwrite_item(trans, root, path, eb, slot, key);
-       BUG_ON(ret);
+       if (found_type == BTRFS_FILE_EXTENT_REG ||
+           found_type == BTRFS_FILE_EXTENT_PREALLOC) {
+               u64 offset;
+               unsigned long dest_offset;
+               struct btrfs_key ins;
+
+               ret = btrfs_insert_empty_item(trans, root, path, key,
+                                             sizeof(*item));
+               BUG_ON(ret);
+               dest_offset = btrfs_item_ptr_offset(path->nodes[0],
+                                                   path->slots[0]);
+               copy_extent_buffer(path->nodes[0], eb, dest_offset,
+                               (unsigned long)item,  sizeof(*item));
+
+               ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
+               ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
+               ins.type = BTRFS_EXTENT_ITEM_KEY;
+               offset = key->offset - btrfs_file_extent_offset(eb, item);
+
+               if (ins.objectid > 0) {
+                       u64 csum_start;
+                       u64 csum_end;
+                       LIST_HEAD(ordered_sums);
+                       /*
+                        * is this extent already allocated in the extent
+                        * allocation tree?  If so, just add a reference
+                        */
+                       ret = btrfs_lookup_extent(root, ins.objectid,
+                                               ins.offset);
+                       if (ret == 0) {
+                               ret = btrfs_inc_extent_ref(trans, root,
+                                               ins.objectid, ins.offset,
+                                               0, root->root_key.objectid,
+                                               key->objectid, offset);
+                               BUG_ON(ret);
+                       } else {
+                               /*
+                                * insert the extent pointer in the extent
+                                * allocation tree
+                                */
+                               ret = btrfs_alloc_logged_file_extent(trans,
+                                               root, root->root_key.objectid,
+                                               key->objectid, offset, &ins);
+                               BUG_ON(ret);
+                       }
+                       btrfs_release_path(path);
+
+                       if (btrfs_file_extent_compression(eb, item)) {
+                               csum_start = ins.objectid;
+                               csum_end = csum_start + ins.offset;
+                       } else {
+                               csum_start = ins.objectid +
+                                       btrfs_file_extent_offset(eb, item);
+                               csum_end = csum_start +
+                                       btrfs_file_extent_num_bytes(eb, item);
+                       }
+
+                       ret = btrfs_lookup_csums_range(root->log_root,
+                                               csum_start, csum_end - 1,
+                                               &ordered_sums, 0);
+                       BUG_ON(ret);
+                       while (!list_empty(&ordered_sums)) {
+                               struct btrfs_ordered_sum *sums;
+                               sums = list_entry(ordered_sums.next,
+                                               struct btrfs_ordered_sum,
+                                               list);
+                               ret = btrfs_csum_file_blocks(trans,
+                                               root->fs_info->csum_root,
+                                               sums);
+                               BUG_ON(ret);
+                               list_del(&sums->list);
+                               kfree(sums);
+                       }
+               } else {
+                       btrfs_release_path(path);
+               }
+       } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
+               /* inline extents are easy, we just overwrite them */
+               ret = overwrite_item(trans, root, path, eb, slot, key);
+               BUG_ON(ret);
+       }
 
-       /* btrfs_drop_extents changes i_bytes & i_blocks, update it here */
-       inode_add_bytes(inode, extent_end - start);
+       inode_set_bytes(inode, saved_nbytes);
        btrfs_update_inode(trans, root, inode);
 out:
        if (inode)
@@ -635,14 +671,21 @@ static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
        btrfs_dir_item_key_to_cpu(leaf, di, &location);
        name_len = btrfs_dir_name_len(leaf, di);
        name = kmalloc(name_len, GFP_NOFS);
+       if (!name)
+               return -ENOMEM;
+
        read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
 
        inode = read_one_inode(root, location.objectid);
-       BUG_ON(!inode);
+       if (!inode) {
+               kfree(name);
+               return -EIO;
+       }
 
        ret = link_to_fixup_dir(trans, root, path, location.objectid);
        BUG_ON(ret);
+
        ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
        BUG_ON(ret);
        kfree(name);
@@ -673,7 +716,7 @@ static noinline int inode_in_dir(struct btrfs_root *root,
                        goto out;
        } else
                goto out;
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
 
        di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
        if (di && !IS_ERR(di)) {
@@ -684,7 +727,7 @@ static noinline int inode_in_dir(struct btrfs_root *root,
                goto out;
        match = 1;
 out:
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
        return match;
 }
 
@@ -713,6 +756,9 @@ static noinline int backref_in_log(struct btrfs_root *log,
        int match = 0;
 
        path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
+
        ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
        if (ret != 0)
                goto out;
@@ -755,18 +801,13 @@ static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
 {
        struct inode *dir;
        int ret;
-       struct btrfs_key location;
        struct btrfs_inode_ref *ref;
-       struct btrfs_dir_item *di;
        struct inode *inode;
        char *name;
        int namelen;
        unsigned long ref_ptr;
        unsigned long ref_end;
-
-       location.objectid = key->objectid;
-       location.type = BTRFS_INODE_ITEM_KEY;
-       location.offset = 0;
+       int search_done = 0;
 
        /*
         * it is possible that we didn't log all the parent directories
@@ -779,7 +820,10 @@ static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
                return -ENOENT;
 
        inode = read_one_inode(root, key->objectid);
-       BUG_ON(!dir);
+       if (!inode) {
+               iput(dir);
+               return -EIO;
+       }
 
        ref_ptr = btrfs_item_ptr_offset(eb, slot);
        ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
@@ -794,7 +838,7 @@ again:
        read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
 
        /* if we already have a perfect match, we're done */
-       if (inode_in_dir(root, path, dir->i_ino, inode->i_ino,
+       if (inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
                         btrfs_inode_ref_index(eb, ref),
                         name, namelen)) {
                goto out;
@@ -807,7 +851,10 @@ again:
         * existing back reference, and we don't want to create
         * dangling pointers in the directory.
         */
-conflict_again:
+
+       if (search_done)
+               goto insert;
+
        ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
        if (ret == 0) {
                char *victim_name;
@@ -843,41 +890,26 @@ conflict_again:
                        if (!backref_in_log(log, key, victim_name,
                                            victim_name_len)) {
                                btrfs_inc_nlink(inode);
-                               btrfs_release_path(root, path);
+                               btrfs_release_path(path);
+
                                ret = btrfs_unlink_inode(trans, root, dir,
                                                         inode, victim_name,
                                                         victim_name_len);
-                               kfree(victim_name);
-                               btrfs_release_path(root, path);
-                               goto conflict_again;
                        }
                        kfree(victim_name);
                        ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
                }
                BUG_ON(ret);
-       }
-       btrfs_release_path(root, path);
 
-       /* look for a conflicting sequence number */
-       di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
-                                        btrfs_inode_ref_index(eb, ref),
-                                        name, namelen, 0);
-       if (di && !IS_ERR(di)) {
-               ret = drop_one_dir_item(trans, root, path, dir, di);
-               BUG_ON(ret);
-       }
-       btrfs_release_path(root, path);
-
-
-       /* look for a conflicting name */
-       di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
-                                  name, namelen, 0);
-       if (di && !IS_ERR(di)) {
-               ret = drop_one_dir_item(trans, root, path, dir, di);
-               BUG_ON(ret);
+               /*
+                * NOTE: we have searched root tree and checked the
+                * coresponding ref, it does not need to check again.
+                */
+               search_done = 1;
        }
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
 
+insert:
        /* insert our name */
        ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
                             btrfs_inode_ref_index(eb, ref));
@@ -896,76 +928,23 @@ out:
        BUG_ON(ret);
 
 out_nowrite:
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
        iput(dir);
        iput(inode);
        return 0;
 }
 
-/*
- * replay one csum item from the log tree into the subvolume 'root'
- * eb, slot and key all refer to the log tree
- * path is for temp use by this function and should be released on return
- *
- * This copies the checksums out of the log tree and inserts them into
- * the subvolume.  Any existing checksums for this range in the file
- * are overwritten, and new items are added where required.
- *
- * We keep this simple by reusing the btrfs_ordered_sum code from
- * the data=ordered mode.  This basically means making a copy
- * of all the checksums in ram, which we have to do anyway for kmap
- * rules.
- *
- * The copy is then sent down to btrfs_csum_file_blocks, which
- * does all the hard work of finding existing items in the file
- * or adding new ones.
- */
-static noinline int replay_one_csum(struct btrfs_trans_handle *trans,
-                                     struct btrfs_root *root,
-                                     struct btrfs_path *path,
-                                     struct extent_buffer *eb, int slot,
-                                     struct btrfs_key *key)
+static int insert_orphan_item(struct btrfs_trans_handle *trans,
+                             struct btrfs_root *root, u64 offset)
 {
        int ret;
-       u32 item_size = btrfs_item_size_nr(eb, slot);
-       u64 cur_offset;
-       u16 csum_size =
-               btrfs_super_csum_size(&root->fs_info->super_copy);
-       unsigned long file_bytes;
-       struct btrfs_ordered_sum *sums;
-       struct btrfs_sector_sum *sector_sum;
-       unsigned long ptr;
-
-       file_bytes = (item_size / csum_size) * root->sectorsize;
-       sums = kzalloc(btrfs_ordered_sum_size(root, file_bytes), GFP_NOFS);
-       if (!sums)
-               return -ENOMEM;
-
-       INIT_LIST_HEAD(&sums->list);
-       sums->len = file_bytes;
-       sums->bytenr = key->offset;
+       ret = btrfs_find_orphan_item(root, offset);
+       if (ret > 0)
+               ret = btrfs_insert_orphan_item(trans, root, offset);
+       return ret;
+}
 
-       /*
-        * copy all the sums into the ordered sum struct
-        */
-       sector_sum = sums->sums;
-       cur_offset = key->offset;
-       ptr = btrfs_item_ptr_offset(eb, slot);
-       while (item_size > 0) {
-               sector_sum->bytenr = cur_offset;
-               read_extent_buffer(eb, &sector_sum->sum, ptr, csum_size);
-               sector_sum++;
-               item_size -= csum_size;
-               ptr += csum_size;
-               cur_offset += root->sectorsize;
-       }
 
-       /* let btrfs_csum_file_blocks add them into the file */
-       ret = btrfs_csum_file_blocks(trans, root->fs_info->csum_root, sums);
-       BUG_ON(ret);
-       kfree(sums);
-       return 0;
-}
 /*
  * There are a few corners where the link count of the file can't
  * be properly maintained during replay.  So, instead of adding
@@ -987,12 +966,15 @@ static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
        unsigned long ptr;
        unsigned long ptr_end;
        int name_len;
+       u64 ino = btrfs_ino(inode);
 
-       key.objectid = inode->i_ino;
+       key.objectid = ino;
        key.type = BTRFS_INODE_REF_KEY;
        key.offset = (u64)-1;
 
        path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
 
        while (1) {
                ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
@@ -1005,7 +987,7 @@ static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
                }
                btrfs_item_key_to_cpu(path->nodes[0], &key,
                                      path->slots[0]);
-               if (key.objectid != inode->i_ino ||
+               if (key.objectid != ino ||
                    key.type != BTRFS_INODE_REF_KEY)
                        break;
                ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
@@ -1024,15 +1006,26 @@ static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
                if (key.offset == 0)
                        break;
                key.offset--;
-               btrfs_release_path(root, path);
+               btrfs_release_path(path);
        }
-       btrfs_free_path(path);
+       btrfs_release_path(path);
        if (nlink != inode->i_nlink) {
                inode->i_nlink = nlink;
                btrfs_update_inode(trans, root, inode);
        }
        BTRFS_I(inode)->index_cnt = (u64)-1;
 
+       if (inode->i_nlink == 0) {
+               if (S_ISDIR(inode->i_mode)) {
+                       ret = replay_dir_deletes(trans, root, NULL, path,
+                                                ino, 1);
+                       BUG_ON(ret);
+               }
+               ret = insert_orphan_item(trans, root, ino);
+               BUG_ON(ret);
+       }
+       btrfs_free_path(path);
+
        return 0;
 }
 
@@ -1064,23 +1057,30 @@ static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
                        break;
 
                ret = btrfs_del_item(trans, root, path);
-               BUG_ON(ret);
+               if (ret)
+                       goto out;
 
-               btrfs_release_path(root, path);
+               btrfs_release_path(path);
                inode = read_one_inode(root, key.offset);
-               BUG_ON(!inode);
+               if (!inode)
+                       return -EIO;
 
                ret = fixup_inode_link_count(trans, root, inode);
                BUG_ON(ret);
 
                iput(inode);
 
-               if (key.offset == 0)
-                       break;
-               key.offset--;
+               /*
+                * fixup on a directory may create new entries,
+                * make sure we always look for the highset possible
+                * offset
+                */
+               key.offset = (u64)-1;
        }
-       btrfs_release_path(root, path);
-       return 0;
+       ret = 0;
+out:
+       btrfs_release_path(path);
+       return ret;
 }
 
 
@@ -1099,7 +1099,8 @@ static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
        struct inode *inode;
 
        inode = read_one_inode(root, objectid);
-       BUG_ON(!inode);
+       if (!inode)
+               return -EIO;
 
        key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
        btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
@@ -1107,7 +1108,7 @@ static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
 
        ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
 
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
        if (ret == 0) {
                btrfs_inc_nlink(inode);
                btrfs_update_inode(trans, root, inode);
@@ -1186,10 +1187,14 @@ static noinline int replay_one_name(struct btrfs_trans_handle *trans,
        int ret;
 
        dir = read_one_inode(root, key->objectid);
-       BUG_ON(!dir);
+       if (!dir)
+               return -EIO;
 
        name_len = btrfs_dir_name_len(eb, di);
        name = kmalloc(name_len, GFP_NOFS);
+       if (!name)
+               return -ENOMEM;
+
        log_type = btrfs_dir_type(eb, di);
        read_extent_buffer(eb, name, (unsigned long)(di + 1),
                   name_len);
@@ -1200,7 +1205,7 @@ static noinline int replay_one_name(struct btrfs_trans_handle *trans,
                exists = 1;
        else
                exists = 0;
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
 
        if (key->type == BTRFS_DIR_ITEM_KEY) {
                dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
@@ -1213,7 +1218,7 @@ static noinline int replay_one_name(struct btrfs_trans_handle *trans,
        } else {
                BUG();
        }
-       if (!dst_di || IS_ERR(dst_di)) {
+       if (IS_ERR_OR_NULL(dst_di)) {
                /* we need a sequence number to insert, so we only
                 * do inserts for the BTRFS_DIR_INDEX_KEY types
                 */
@@ -1244,18 +1249,17 @@ static noinline int replay_one_name(struct btrfs_trans_handle *trans,
        if (key->type == BTRFS_DIR_INDEX_KEY)
                goto insert;
 out:
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
        kfree(name);
        iput(dir);
        return 0;
 
 insert:
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
        ret = insert_one_name(trans, root, path, key->objectid, key->offset,
                              name, name_len, log_type, &log_key);
 
-       if (ret && ret != -ENOENT)
-               BUG();
+       BUG_ON(ret && ret != -ENOENT);
        goto out;
 }
 
@@ -1282,6 +1286,8 @@ static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
        ptr_end = ptr + item_size;
        while (ptr < ptr_end) {
                di = (struct btrfs_dir_item *)ptr;
+               if (verify_dir_item(root, eb, di))
+                       return -EIO;
                name_len = btrfs_dir_name_len(eb, di);
                ret = replay_one_name(trans, root, path, eb, di, key);
                BUG_ON(ret);
@@ -1370,7 +1376,7 @@ next:
        *end_ret = found_end;
        ret = 0;
 out:
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
        return ret;
 }
 
@@ -1408,6 +1414,11 @@ again:
        ptr_end = ptr + item_size;
        while (ptr < ptr_end) {
                di = (struct btrfs_dir_item *)ptr;
+               if (verify_dir_item(root, eb, di)) {
+                       ret = -EIO;
+                       goto out;
+               }
+
                name_len = btrfs_dir_name_len(eb, di);
                name = kmalloc(name_len, GFP_NOFS);
                if (!name) {
@@ -1417,23 +1428,26 @@ again:
                read_extent_buffer(eb, name, (unsigned long)(di + 1),
                                  name_len);
                log_di = NULL;
-               if (dir_key->type == BTRFS_DIR_ITEM_KEY) {
+               if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
                        log_di = btrfs_lookup_dir_item(trans, log, log_path,
                                                       dir_key->objectid,
                                                       name, name_len, 0);
-               } else if (dir_key->type == BTRFS_DIR_INDEX_KEY) {
+               } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
                        log_di = btrfs_lookup_dir_index_item(trans, log,
                                                     log_path,
                                                     dir_key->objectid,
                                                     dir_key->offset,
                                                     name, name_len, 0);
                }
-               if (!log_di || IS_ERR(log_di)) {
+               if (IS_ERR_OR_NULL(log_di)) {
                        btrfs_dir_item_key_to_cpu(eb, di, &location);
-                       btrfs_release_path(root, path);
-                       btrfs_release_path(log, log_path);
+                       btrfs_release_path(path);
+                       btrfs_release_path(log_path);
                        inode = read_one_inode(root, location.objectid);
-                       BUG_ON(!inode);
+                       if (!inode) {
+                               kfree(name);
+                               return -EIO;
+                       }
 
                        ret = link_to_fixup_dir(trans, root,
                                                path, location.objectid);
@@ -1455,7 +1469,7 @@ again:
                        ret = 0;
                        goto out;
                }
-               btrfs_release_path(log, log_path);
+               btrfs_release_path(log_path);
                kfree(name);
 
                ptr = (unsigned long)(di + 1);
@@ -1463,8 +1477,8 @@ again:
        }
        ret = 0;
 out:
-       btrfs_release_path(root, path);
-       btrfs_release_path(log, log_path);
+       btrfs_release_path(path);
+       btrfs_release_path(log_path);
        return ret;
 }
 
@@ -1482,7 +1496,7 @@ static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
                                       struct btrfs_root *root,
                                       struct btrfs_root *log,
                                       struct btrfs_path *path,
-                                      u64 dirid)
+                                      u64 dirid, int del_all)
 {
        u64 range_start;
        u64 range_end;
@@ -1512,10 +1526,14 @@ again:
        range_start = 0;
        range_end = 0;
        while (1) {
-               ret = find_dir_range(log, path, dirid, key_type,
-                                    &range_start, &range_end);
-               if (ret != 0)
-                       break;
+               if (del_all)
+                       range_end = (u64)-1;
+               else {
+                       ret = find_dir_range(log, path, dirid, key_type,
+                                            &range_start, &range_end);
+                       if (ret != 0)
+                               break;
+               }
 
                dir_key.offset = range_start;
                while (1) {
@@ -1541,13 +1559,14 @@ again:
                                break;
 
                        ret = check_item_in_log(trans, root, log, path,
-                                               log_path, dir, &found_key);
+                                               log_path, dir,
+                                               &found_key);
                        BUG_ON(ret);
                        if (found_key.offset == (u64)-1)
                                break;
                        dir_key.offset = found_key.offset + 1;
                }
-               btrfs_release_path(root, path);
+               btrfs_release_path(path);
                if (range_end == (u64)-1)
                        break;
                range_start = range_end + 1;
@@ -1558,11 +1577,11 @@ next_type:
        if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
                key_type = BTRFS_DIR_LOG_INDEX_KEY;
                dir_key.type = BTRFS_DIR_INDEX_KEY;
-               btrfs_release_path(root, path);
+               btrfs_release_path(path);
                goto again;
        }
 out:
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
        btrfs_free_path(log_path);
        iput(dir);
        return ret;
@@ -1586,7 +1605,6 @@ static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
        struct btrfs_path *path;
        struct btrfs_root *root = wc->replay_dest;
        struct btrfs_key key;
-       u32 item_size;
        int level;
        int i;
        int ret;
@@ -1599,17 +1617,16 @@ static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
                return 0;
 
        path = btrfs_alloc_path();
-       BUG_ON(!path);
+       if (!path)
+               return -ENOMEM;
 
        nritems = btrfs_header_nritems(eb);
        for (i = 0; i < nritems; i++) {
                btrfs_item_key_to_cpu(eb, &key, i);
-               item_size = btrfs_item_size_nr(eb, i);
 
                /* inode keys are done during the first stage */
                if (key.type == BTRFS_INODE_ITEM_KEY &&
                    wc->stage == LOG_WALK_REPLAY_INODES) {
-                       struct inode *inode;
                        struct btrfs_inode_item *inode_item;
                        u32 mode;
 
@@ -1618,27 +1635,23 @@ static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
                        mode = btrfs_inode_mode(eb, inode_item);
                        if (S_ISDIR(mode)) {
                                ret = replay_dir_deletes(wc->trans,
-                                        root, log, path, key.objectid);
+                                        root, log, path, key.objectid, 0);
                                BUG_ON(ret);
                        }
                        ret = overwrite_item(wc->trans, root, path,
                                             eb, i, &key);
                        BUG_ON(ret);
 
-                       /* for regular files, truncate away
-                        * extents past the new EOF
+                       /* for regular files, make sure corresponding
+                        * orhpan item exist. extents past the new EOF
+                        * will be truncated later by orphan cleanup.
                         */
                        if (S_ISREG(mode)) {
-                               inode = read_one_inode(root,
-                                                      key.objectid);
-                               BUG_ON(!inode);
-
-                               ret = btrfs_truncate_inode_items(wc->trans,
-                                       root, inode, inode->i_size,
-                                       BTRFS_EXTENT_DATA_KEY);
+                               ret = insert_orphan_item(wc->trans, root,
+                                                        key.objectid);
                                BUG_ON(ret);
-                               iput(inode);
                        }
+
                        ret = link_to_fixup_dir(wc->trans, root,
                                                path, key.objectid);
                        BUG_ON(ret);
@@ -1659,10 +1672,6 @@ static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
                        ret = replay_one_extent(wc->trans, root, path,
                                                eb, i, &key);
                        BUG_ON(ret);
-               } else if (key.type == BTRFS_EXTENT_CSUM_KEY) {
-                       ret = replay_one_csum(wc->trans, root, path,
-                                             eb, i, &key);
-                       BUG_ON(ret);
                } else if (key.type == BTRFS_DIR_ITEM_KEY ||
                           key.type == BTRFS_DIR_INDEX_KEY) {
                        ret = replay_one_dir_item(wc->trans, root, path,
@@ -1680,7 +1689,6 @@ static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
                                   struct walk_control *wc)
 {
        u64 root_owner;
-       u64 root_gen;
        u64 bytenr;
        u64 ptr_gen;
        struct extent_buffer *next;
@@ -1710,25 +1718,26 @@ static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
 
                parent = path->nodes[*level];
                root_owner = btrfs_header_owner(parent);
-               root_gen = btrfs_header_generation(parent);
 
                next = btrfs_find_create_tree_block(root, bytenr, blocksize);
-
-               wc->process_func(root, next, wc, ptr_gen);
+               if (!next)
+                       return -ENOMEM;
 
                if (*level == 1) {
+                       ret = wc->process_func(root, next, wc, ptr_gen);
+                       if (ret)
+                               return ret;
+
                        path->slots[*level]++;
                        if (wc->free) {
                                btrfs_read_buffer(next, ptr_gen);
 
                                btrfs_tree_lock(next);
+                               btrfs_set_lock_blocking(next);
                                clean_tree_block(trans, root, next);
                                btrfs_wait_tree_block_writeback(next);
                                btrfs_tree_unlock(next);
 
-                               ret = btrfs_drop_leaf_ref(trans, root, next);
-                               BUG_ON(ret);
-
                                WARN_ON(root_owner !=
                                        BTRFS_TREE_LOG_OBJECTID);
                                ret = btrfs_free_reserved_extent(root,
@@ -1751,38 +1760,7 @@ static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
        WARN_ON(*level < 0);
        WARN_ON(*level >= BTRFS_MAX_LEVEL);
 
-       if (path->nodes[*level] == root->node)
-               parent = path->nodes[*level];
-       else
-               parent = path->nodes[*level + 1];
-
-       bytenr = path->nodes[*level]->start;
-
-       blocksize = btrfs_level_size(root, *level);
-       root_owner = btrfs_header_owner(parent);
-       root_gen = btrfs_header_generation(parent);
-
-       wc->process_func(root, path->nodes[*level], wc,
-                        btrfs_header_generation(path->nodes[*level]));
-
-       if (wc->free) {
-               next = path->nodes[*level];
-               btrfs_tree_lock(next);
-               clean_tree_block(trans, root, next);
-               btrfs_wait_tree_block_writeback(next);
-               btrfs_tree_unlock(next);
-
-               if (*level == 0) {
-                       ret = btrfs_drop_leaf_ref(trans, root, next);
-                       BUG_ON(ret);
-               }
-               WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
-               ret = btrfs_free_reserved_extent(root, bytenr, blocksize);
-               BUG_ON(ret);
-       }
-       free_extent_buffer(path->nodes[*level]);
-       path->nodes[*level] = NULL;
-       *level += 1;
+       path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
 
        cond_resched();
        return 0;
@@ -1794,16 +1772,13 @@ static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
                                 struct walk_control *wc)
 {
        u64 root_owner;
-       u64 root_gen;
        int i;
        int slot;
        int ret;
 
        for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
                slot = path->slots[i];
-               if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
-                       struct extent_buffer *node;
-                       node = path->nodes[i];
+               if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
                        path->slots[i]++;
                        *level = i;
                        WARN_ON(*level == 0);
@@ -1816,25 +1791,22 @@ static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
                                parent = path->nodes[*level + 1];
 
                        root_owner = btrfs_header_owner(parent);
-                       root_gen = btrfs_header_generation(parent);
-                       wc->process_func(root, path->nodes[*level], wc,
+                       ret = wc->process_func(root, path->nodes[*level], wc,
                                 btrfs_header_generation(path->nodes[*level]));
+                       if (ret)
+                               return ret;
+
                        if (wc->free) {
                                struct extent_buffer *next;
 
                                next = path->nodes[*level];
 
                                btrfs_tree_lock(next);
+                               btrfs_set_lock_blocking(next);
                                clean_tree_block(trans, root, next);
                                btrfs_wait_tree_block_writeback(next);
                                btrfs_tree_unlock(next);
 
-                               if (*level == 0) {
-                                       ret = btrfs_drop_leaf_ref(trans, root,
-                                                                 next);
-                                       BUG_ON(ret);
-                               }
-
                                WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
                                ret = btrfs_free_reserved_extent(root,
                                                path->nodes[*level]->start,
@@ -1865,7 +1837,8 @@ static int walk_log_tree(struct btrfs_trans_handle *trans,
        int orig_level;
 
        path = btrfs_alloc_path();
-       BUG_ON(!path);
+       if (!path)
+               return -ENOMEM;
 
        level = btrfs_header_level(log->node);
        orig_level = level;
@@ -1897,15 +1870,11 @@ static int walk_log_tree(struct btrfs_trans_handle *trans,
                        next = path->nodes[orig_level];
 
                        btrfs_tree_lock(next);
+                       btrfs_set_lock_blocking(next);
                        clean_tree_block(trans, log, next);
                        btrfs_wait_tree_block_writeback(next);
                        btrfs_tree_unlock(next);
 
-                       if (orig_level == 0) {
-                               ret = btrfs_drop_leaf_ref(trans, log,
-                                                         next);
-                               BUG_ON(ret);
-                       }
                        WARN_ON(log->root_key.objectid !=
                                BTRFS_TREE_LOG_OBJECTID);
                        ret = btrfs_free_reserved_extent(log, next->start,
@@ -1921,99 +1890,258 @@ static int walk_log_tree(struct btrfs_trans_handle *trans,
                }
        }
        btrfs_free_path(path);
-       if (wc->free)
-               free_extent_buffer(log->node);
        return ret;
 }
 
-static int wait_log_commit(struct btrfs_root *log)
+/*
+ * helper function to update the item for a given subvolumes log root
+ * in the tree of log roots
+ */
+static int update_log_root(struct btrfs_trans_handle *trans,
+                          struct btrfs_root *log)
+{
+       int ret;
+
+       if (log->log_transid == 1) {
+               /* insert root item on the first sync */
+               ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
+                               &log->root_key, &log->root_item);
+       } else {
+               ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
+                               &log->root_key, &log->root_item);
+       }
+       return ret;
+}
+
+static int wait_log_commit(struct btrfs_trans_handle *trans,
+                          struct btrfs_root *root, unsigned long transid)
 {
        DEFINE_WAIT(wait);
-       u64 transid = log->fs_info->tree_log_transid;
+       int index = transid % 2;
 
+       /*
+        * we only allow two pending log transactions at a time,
+        * so we know that if ours is more than 2 older than the
+        * current transaction, we're done
+        */
        do {
-               prepare_to_wait(&log->fs_info->tree_log_wait, &wait,
-                               TASK_UNINTERRUPTIBLE);
-               mutex_unlock(&log->fs_info->tree_log_mutex);
-               if (atomic_read(&log->fs_info->tree_log_commit))
+               prepare_to_wait(&root->log_commit_wait[index],
+                               &wait, TASK_UNINTERRUPTIBLE);
+               mutex_unlock(&root->log_mutex);
+
+               if (root->fs_info->last_trans_log_full_commit !=
+                   trans->transid && root->log_transid < transid + 2 &&
+                   atomic_read(&root->log_commit[index]))
+                       schedule();
+
+               finish_wait(&root->log_commit_wait[index], &wait);
+               mutex_lock(&root->log_mutex);
+       } while (root->log_transid < transid + 2 &&
+                atomic_read(&root->log_commit[index]));
+       return 0;
+}
+
+static int wait_for_writer(struct btrfs_trans_handle *trans,
+                          struct btrfs_root *root)
+{
+       DEFINE_WAIT(wait);
+       while (atomic_read(&root->log_writers)) {
+               prepare_to_wait(&root->log_writer_wait,
+                               &wait, TASK_UNINTERRUPTIBLE);
+               mutex_unlock(&root->log_mutex);
+               if (root->fs_info->last_trans_log_full_commit !=
+                   trans->transid && atomic_read(&root->log_writers))
                        schedule();
-               finish_wait(&log->fs_info->tree_log_wait, &wait);
-               mutex_lock(&log->fs_info->tree_log_mutex);
-       } while (transid == log->fs_info->tree_log_transid &&
-               atomic_read(&log->fs_info->tree_log_commit));
+               mutex_lock(&root->log_mutex);
+               finish_wait(&root->log_writer_wait, &wait);
+       }
        return 0;
 }
 
 /*
  * btrfs_sync_log does sends a given tree log down to the disk and
  * updates the super blocks to record it.  When this call is done,
- * you know that any inodes previously logged are safely on disk
+ * you know that any inodes previously logged are safely on disk only
+ * if it returns 0.
+ *
+ * Any other return value means you need to call btrfs_commit_transaction.
+ * Some of the edge cases for fsyncing directories that have had unlinks
+ * or renames done in the past mean that sometimes the only safe
+ * fsync is to commit the whole FS.  When btrfs_sync_log returns -EAGAIN,
+ * that has happened.
  */
 int btrfs_sync_log(struct btrfs_trans_handle *trans,
                   struct btrfs_root *root)
 {
+       int index1;
+       int index2;
+       int mark;
        int ret;
-       unsigned long batch;
        struct btrfs_root *log = root->log_root;
-
-       mutex_lock(&log->fs_info->tree_log_mutex);
-       if (atomic_read(&log->fs_info->tree_log_commit)) {
-               wait_log_commit(log);
-               goto out;
+       struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
+       unsigned long log_transid = 0;
+
+       mutex_lock(&root->log_mutex);
+       index1 = root->log_transid % 2;
+       if (atomic_read(&root->log_commit[index1])) {
+               wait_log_commit(trans, root, root->log_transid);
+               mutex_unlock(&root->log_mutex);
+               return 0;
        }
-       atomic_set(&log->fs_info->tree_log_commit, 1);
+       atomic_set(&root->log_commit[index1], 1);
+
+       /* wait for previous tree log sync to complete */
+       if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
+               wait_log_commit(trans, root, root->log_transid - 1);
 
        while (1) {
-               batch = log->fs_info->tree_log_batch;
-               mutex_unlock(&log->fs_info->tree_log_mutex);
-               schedule_timeout_uninterruptible(1);
-               mutex_lock(&log->fs_info->tree_log_mutex);
-
-               while (atomic_read(&log->fs_info->tree_log_writers)) {
-                       DEFINE_WAIT(wait);
-                       prepare_to_wait(&log->fs_info->tree_log_wait, &wait,
-                                       TASK_UNINTERRUPTIBLE);
-                       mutex_unlock(&log->fs_info->tree_log_mutex);
-                       if (atomic_read(&log->fs_info->tree_log_writers))
-                               schedule();
-                       mutex_lock(&log->fs_info->tree_log_mutex);
-                       finish_wait(&log->fs_info->tree_log_wait, &wait);
+               unsigned long batch = root->log_batch;
+               if (root->log_multiple_pids) {
+                       mutex_unlock(&root->log_mutex);
+                       schedule_timeout_uninterruptible(1);
+                       mutex_lock(&root->log_mutex);
                }
-               if (batch == log->fs_info->tree_log_batch)
+               wait_for_writer(trans, root);
+               if (batch == root->log_batch)
                        break;
        }
 
-       ret = btrfs_write_and_wait_marked_extents(log, &log->dirty_log_pages);
+       /* bail out if we need to do a full commit */
+       if (root->fs_info->last_trans_log_full_commit == trans->transid) {
+               ret = -EAGAIN;
+               mutex_unlock(&root->log_mutex);
+               goto out;
+       }
+
+       log_transid = root->log_transid;
+       if (log_transid % 2 == 0)
+               mark = EXTENT_DIRTY;
+       else
+               mark = EXTENT_NEW;
+
+       /* we start IO on  all the marked extents here, but we don't actually
+        * wait for them until later.
+        */
+       ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
        BUG_ON(ret);
-       ret = btrfs_write_and_wait_marked_extents(root->fs_info->log_root_tree,
-                              &root->fs_info->log_root_tree->dirty_log_pages);
+
+       btrfs_set_root_node(&log->root_item, log->node);
+
+       root->log_batch = 0;
+       root->log_transid++;
+       log->log_transid = root->log_transid;
+       root->log_start_pid = 0;
+       smp_mb();
+       /*
+        * IO has been started, blocks of the log tree have WRITTEN flag set
+        * in their headers. new modifications of the log will be written to
+        * new positions. so it's safe to allow log writers to go in.
+        */
+       mutex_unlock(&root->log_mutex);
+
+       mutex_lock(&log_root_tree->log_mutex);
+       log_root_tree->log_batch++;
+       atomic_inc(&log_root_tree->log_writers);
+       mutex_unlock(&log_root_tree->log_mutex);
+
+       ret = update_log_root(trans, log);
+
+       mutex_lock(&log_root_tree->log_mutex);
+       if (atomic_dec_and_test(&log_root_tree->log_writers)) {
+               smp_mb();
+               if (waitqueue_active(&log_root_tree->log_writer_wait))
+                       wake_up(&log_root_tree->log_writer_wait);
+       }
+
+       if (ret) {
+               BUG_ON(ret != -ENOSPC);
+               root->fs_info->last_trans_log_full_commit = trans->transid;
+               btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
+               mutex_unlock(&log_root_tree->log_mutex);
+               ret = -EAGAIN;
+               goto out;
+       }
+
+       index2 = log_root_tree->log_transid % 2;
+       if (atomic_read(&log_root_tree->log_commit[index2])) {
+               btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
+               wait_log_commit(trans, log_root_tree,
+                               log_root_tree->log_transid);
+               mutex_unlock(&log_root_tree->log_mutex);
+               ret = 0;
+               goto out;
+       }
+       atomic_set(&log_root_tree->log_commit[index2], 1);
+
+       if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
+               wait_log_commit(trans, log_root_tree,
+                               log_root_tree->log_transid - 1);
+       }
+
+       wait_for_writer(trans, log_root_tree);
+
+       /*
+        * now that we've moved on to the tree of log tree roots,
+        * check the full commit flag again
+        */
+       if (root->fs_info->last_trans_log_full_commit == trans->transid) {
+               btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
+               mutex_unlock(&log_root_tree->log_mutex);
+               ret = -EAGAIN;
+               goto out_wake_log_root;
+       }
+
+       ret = btrfs_write_and_wait_marked_extents(log_root_tree,
+                               &log_root_tree->dirty_log_pages,
+                               EXTENT_DIRTY | EXTENT_NEW);
        BUG_ON(ret);
+       btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
 
        btrfs_set_super_log_root(&root->fs_info->super_for_commit,
-                                log->fs_info->log_root_tree->node->start);
+                               log_root_tree->node->start);
        btrfs_set_super_log_root_level(&root->fs_info->super_for_commit,
-                      btrfs_header_level(log->fs_info->log_root_tree->node));
+                               btrfs_header_level(log_root_tree->node));
+
+       log_root_tree->log_batch = 0;
+       log_root_tree->log_transid++;
+       smp_mb();
+
+       mutex_unlock(&log_root_tree->log_mutex);
+
+       /*
+        * nobody else is going to jump in and write the the ctree
+        * super here because the log_commit atomic below is protecting
+        * us.  We must be called with a transaction handle pinning
+        * the running transaction open, so a full commit can't hop
+        * in and cause problems either.
+        */
+       btrfs_scrub_pause_super(root);
+       write_ctree_super(trans, root->fs_info->tree_root, 1);
+       btrfs_scrub_continue_super(root);
+       ret = 0;
 
-       write_ctree_super(trans, log->fs_info->tree_root, 2);
-       log->fs_info->tree_log_transid++;
-       log->fs_info->tree_log_batch = 0;
-       atomic_set(&log->fs_info->tree_log_commit, 0);
+       mutex_lock(&root->log_mutex);
+       if (root->last_log_commit < log_transid)
+               root->last_log_commit = log_transid;
+       mutex_unlock(&root->log_mutex);
+
+out_wake_log_root:
+       atomic_set(&log_root_tree->log_commit[index2], 0);
        smp_mb();
-       if (waitqueue_active(&log->fs_info->tree_log_wait))
-               wake_up(&log->fs_info->tree_log_wait);
+       if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
+               wake_up(&log_root_tree->log_commit_wait[index2]);
 out:
-       mutex_unlock(&log->fs_info->tree_log_mutex);
-       return 0;
+       atomic_set(&root->log_commit[index1], 0);
+       smp_mb();
+       if (waitqueue_active(&root->log_commit_wait[index1]))
+               wake_up(&root->log_commit_wait[index1]);
+       return ret;
 }
 
-/* * free all the extents used by the tree log.  This should be called
- * at commit time of the full transaction
- */
-int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
+static void free_log_tree(struct btrfs_trans_handle *trans,
+                         struct btrfs_root *log)
 {
        int ret;
-       struct btrfs_root *log;
-       struct key;
        u64 start;
        u64 end;
        struct walk_control wc = {
@@ -2021,52 +2149,44 @@ int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
                .process_func = process_one_buffer
        };
 
-       if (!root->log_root)
-               return 0;
-
-       log = root->log_root;
        ret = walk_log_tree(trans, log, &wc);
        BUG_ON(ret);
 
        while (1) {
                ret = find_first_extent_bit(&log->dirty_log_pages,
-                                   0, &start, &end, EXTENT_DIRTY);
+                               0, &start, &end, EXTENT_DIRTY | EXTENT_NEW);
                if (ret)
                        break;
 
-               clear_extent_dirty(&log->dirty_log_pages,
-                                  start, end, GFP_NOFS);
+               clear_extent_bits(&log->dirty_log_pages, start, end,
+                                 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
        }
 
-       log = root->log_root;
-       ret = btrfs_del_root(trans, root->fs_info->log_root_tree,
-                            &log->root_key);
-       BUG_ON(ret);
-       root->log_root = NULL;
-       kfree(root->log_root);
-       return 0;
+       free_extent_buffer(log->node);
+       kfree(log);
 }
 
 /*
- * helper function to update the item for a given subvolumes log root
- * in the tree of log roots
+ * free all the extents used by the tree log.  This should be called
+ * at commit time of the full transaction
  */
-static int update_log_root(struct btrfs_trans_handle *trans,
-                          struct btrfs_root *log)
+int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
 {
-       u64 bytenr = btrfs_root_bytenr(&log->root_item);
-       int ret;
-
-       if (log->node->start == bytenr)
-               return 0;
+       if (root->log_root) {
+               free_log_tree(trans, root->log_root);
+               root->log_root = NULL;
+       }
+       return 0;
+}
 
-       btrfs_set_root_bytenr(&log->root_item, log->node->start);
-       btrfs_set_root_generation(&log->root_item, trans->transid);
-       btrfs_set_root_level(&log->root_item, btrfs_header_level(log->node));
-       ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
-                               &log->root_key, &log->root_item);
-       BUG_ON(ret);
-       return ret;
+int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
+                            struct btrfs_fs_info *fs_info)
+{
+       if (fs_info->log_root_tree) {
+               free_log_tree(trans, fs_info->log_root_tree);
+               fs_info->log_root_tree = NULL;
+       }
+       return 0;
 }
 
 /*
@@ -2099,7 +2219,9 @@ int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
        struct btrfs_dir_item *di;
        struct btrfs_path *path;
        int ret;
+       int err = 0;
        int bytes_del = 0;
+       u64 dir_ino = btrfs_ino(dir);
 
        if (BTRFS_I(dir)->logged_trans < trans->transid)
                return 0;
@@ -2112,17 +2234,30 @@ int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
 
        log = root->log_root;
        path = btrfs_alloc_path();
-       di = btrfs_lookup_dir_item(trans, log, path, dir->i_ino,
+       if (!path) {
+               err = -ENOMEM;
+               goto out_unlock;
+       }
+
+       di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
                                   name, name_len, -1);
-       if (di && !IS_ERR(di)) {
+       if (IS_ERR(di)) {
+               err = PTR_ERR(di);
+               goto fail;
+       }
+       if (di) {
                ret = btrfs_delete_one_dir_name(trans, log, path, di);
                bytes_del += name_len;
                BUG_ON(ret);
        }
-       btrfs_release_path(log, path);
-       di = btrfs_lookup_dir_index_item(trans, log, path, dir->i_ino,
+       btrfs_release_path(path);
+       di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
                                         index, name, name_len, -1);
-       if (di && !IS_ERR(di)) {
+       if (IS_ERR(di)) {
+               err = PTR_ERR(di);
+               goto fail;
+       }
+       if (di) {
                ret = btrfs_delete_one_dir_name(trans, log, path, di);
                bytes_del += name_len;
                BUG_ON(ret);
@@ -2134,12 +2269,16 @@ int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
        if (bytes_del) {
                struct btrfs_key key;
 
-               key.objectid = dir->i_ino;
+               key.objectid = dir_ino;
                key.offset = 0;
                key.type = BTRFS_INODE_ITEM_KEY;
-               btrfs_release_path(log, path);
+               btrfs_release_path(path);
 
                ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
+               if (ret < 0) {
+                       err = ret;
+                       goto fail;
+               }
                if (ret == 0) {
                        struct btrfs_inode_item *item;
                        u64 i_size;
@@ -2155,14 +2294,19 @@ int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
                        btrfs_mark_buffer_dirty(path->nodes[0]);
                } else
                        ret = 0;
-               btrfs_release_path(log, path);
+               btrfs_release_path(path);
        }
-
+fail:
        btrfs_free_path(path);
+out_unlock:
        mutex_unlock(&BTRFS_I(dir)->log_mutex);
-       end_log_trans(root);
+       if (ret == -ENOSPC) {
+               root->fs_info->last_trans_log_full_commit = trans->transid;
+               ret = 0;
+       }
+       btrfs_end_log_trans(root);
 
-       return 0;
+       return err;
 }
 
 /* see comments for btrfs_del_dir_entries_in_log */
@@ -2184,10 +2328,14 @@ int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
        log = root->log_root;
        mutex_lock(&BTRFS_I(inode)->log_mutex);
 
-       ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino,
+       ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
                                  dirid, &index);
        mutex_unlock(&BTRFS_I(inode)->log_mutex);
-       end_log_trans(root);
+       if (ret == -ENOSPC) {
+               root->fs_info->last_trans_log_full_commit = trans->transid;
+               ret = 0;
+       }
+       btrfs_end_log_trans(root);
 
        return ret;
 }
@@ -2214,13 +2362,14 @@ static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
        else
                key.type = BTRFS_DIR_LOG_INDEX_KEY;
        ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
-       BUG_ON(ret);
+       if (ret)
+               return ret;
 
        item = btrfs_item_ptr(path->nodes[0], path->slots[0],
                              struct btrfs_dir_log_item);
        btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
        btrfs_mark_buffer_dirty(path->nodes[0]);
-       btrfs_release_path(log, path);
+       btrfs_release_path(path);
        return 0;
 }
 
@@ -2239,18 +2388,20 @@ static noinline int log_dir_items(struct btrfs_trans_handle *trans,
        struct btrfs_key max_key;
        struct btrfs_root *log = root->log_root;
        struct extent_buffer *src;
+       int err = 0;
        int ret;
        int i;
        int nritems;
        u64 first_offset = min_offset;
        u64 last_offset = (u64)-1;
+       u64 ino = btrfs_ino(inode);
 
        log = root->log_root;
-       max_key.objectid = inode->i_ino;
+       max_key.objectid = ino;
        max_key.offset = (u64)-1;
        max_key.type = key_type;
 
-       min_key.objectid = inode->i_ino;
+       min_key.objectid = ino;
        min_key.type = key_type;
        min_key.offset = min_offset;
 
@@ -2263,18 +2414,17 @@ static noinline int log_dir_items(struct btrfs_trans_handle *trans,
         * we didn't find anything from this transaction, see if there
         * is anything at all
         */
-       if (ret != 0 || min_key.objectid != inode->i_ino ||
-           min_key.type != key_type) {
-               min_key.objectid = inode->i_ino;
+       if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
+               min_key.objectid = ino;
                min_key.type = key_type;
                min_key.offset = (u64)-1;
-               btrfs_release_path(root, path);
+               btrfs_release_path(path);
                ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
                if (ret < 0) {
-                       btrfs_release_path(root, path);
+                       btrfs_release_path(path);
                        return ret;
                }
-               ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
+               ret = btrfs_previous_item(root, path, ino, key_type);
 
                /* if ret == 0 there are items for this type,
                 * create a range to tell us the last key of this type.
@@ -2292,7 +2442,7 @@ static noinline int log_dir_items(struct btrfs_trans_handle *trans,
        }
 
        /* go backward to find any previous key */
-       ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
+       ret = btrfs_previous_item(root, path, ino, key_type);
        if (ret == 0) {
                struct btrfs_key tmp;
                btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
@@ -2301,9 +2451,13 @@ static noinline int log_dir_items(struct btrfs_trans_handle *trans,
                        ret = overwrite_item(trans, log, dst_path,
                                             path->nodes[0], path->slots[0],
                                             &tmp);
+                       if (ret) {
+                               err = ret;
+                               goto done;
+                       }
                }
        }
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
 
        /* find the first key from this transaction again */
        ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
@@ -2323,12 +2477,14 @@ static noinline int log_dir_items(struct btrfs_trans_handle *trans,
                for (i = path->slots[0]; i < nritems; i++) {
                        btrfs_item_key_to_cpu(src, &min_key, i);
 
-                       if (min_key.objectid != inode->i_ino ||
-                           min_key.type != key_type)
+                       if (min_key.objectid != ino || min_key.type != key_type)
                                goto done;
                        ret = overwrite_item(trans, log, dst_path, src, i,
                                             &min_key);
-                       BUG_ON(ret);
+                       if (ret) {
+                               err = ret;
+                               goto done;
+                       }
                }
                path->slots[0] = nritems;
 
@@ -2342,7 +2498,7 @@ static noinline int log_dir_items(struct btrfs_trans_handle *trans,
                        goto done;
                }
                btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
-               if (tmp.objectid != inode->i_ino || tmp.type != key_type) {
+               if (tmp.objectid != ino || tmp.type != key_type) {
                        last_offset = (u64)-1;
                        goto done;
                }
@@ -2350,22 +2506,29 @@ static noinline int log_dir_items(struct btrfs_trans_handle *trans,
                        ret = overwrite_item(trans, log, dst_path,
                                             path->nodes[0], path->slots[0],
                                             &tmp);
-
-                       BUG_ON(ret);
-                       last_offset = tmp.offset;
+                       if (ret)
+                               err = ret;
+                       else
+                               last_offset = tmp.offset;
                        goto done;
                }
        }
 done:
-       *last_offset_ret = last_offset;
-       btrfs_release_path(root, path);
-       btrfs_release_path(log, dst_path);
+       btrfs_release_path(path);
+       btrfs_release_path(dst_path);
 
-       /* insert the log range keys to indicate where the log is valid */
-       ret = insert_dir_log_key(trans, log, path, key_type, inode->i_ino,
-                                first_offset, last_offset);
-       BUG_ON(ret);
-       return 0;
+       if (err == 0) {
+               *last_offset_ret = last_offset;
+               /*
+                * insert the log range keys to indicate where the log
+                * is valid
+                */
+               ret = insert_dir_log_key(trans, log, path, key_type,
+                                        ino, first_offset, last_offset);
+               if (ret)
+                       err = ret;
+       }
+       return err;
 }
 
 /*
@@ -2397,7 +2560,8 @@ again:
                ret = log_dir_items(trans, root, inode, path,
                                    dst_path, key_type, min_key,
                                    &max_key);
-               BUG_ON(ret);
+               if (ret)
+                       return ret;
                if (max_key == (u64)-1)
                        break;
                min_key = max_key + 1;
@@ -2431,8 +2595,8 @@ static int drop_objectid_items(struct btrfs_trans_handle *trans,
 
        while (1) {
                ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
-
-               if (ret != 1)
+               BUG_ON(ret == 0);
+               if (ret < 0)
                        break;
 
                if (path->slots[0] == 0)
@@ -2446,91 +2610,12 @@ static int drop_objectid_items(struct btrfs_trans_handle *trans,
                        break;
 
                ret = btrfs_del_item(trans, log, path);
-               BUG_ON(ret);
-               btrfs_release_path(log, path);
-       }
-       btrfs_release_path(log, path);
-       return 0;
-}
-
-static noinline int copy_extent_csums(struct btrfs_trans_handle *trans,
-                                     struct list_head *list,
-                                     struct btrfs_root *root,
-                                     u64 disk_bytenr, u64 len)
-{
-       struct btrfs_ordered_sum *sums;
-       struct btrfs_sector_sum *sector_sum;
-       int ret;
-       struct btrfs_path *path;
-       struct btrfs_csum_item *item = NULL;
-       u64 end = disk_bytenr + len;
-       u64 item_start_offset = 0;
-       u64 item_last_offset = 0;
-       u32 diff;
-       u32 sum;
-       u16 csum_size = btrfs_super_csum_size(&root->fs_info->super_copy);
-
-       sums = kzalloc(btrfs_ordered_sum_size(root, len), GFP_NOFS);
-
-       sector_sum = sums->sums;
-       sums->bytenr = disk_bytenr;
-       sums->len = len;
-       list_add_tail(&sums->list, list);
-
-       path = btrfs_alloc_path();
-       while (disk_bytenr < end) {
-               if (!item || disk_bytenr < item_start_offset ||
-                   disk_bytenr >= item_last_offset) {
-                       struct btrfs_key found_key;
-                       u32 item_size;
-
-                       if (item)
-                               btrfs_release_path(root, path);
-                       item = btrfs_lookup_csum(NULL, root, path,
-                                                disk_bytenr, 0);
-                       if (IS_ERR(item)) {
-                               ret = PTR_ERR(item);
-                               if (ret == -ENOENT || ret == -EFBIG)
-                                       ret = 0;
-                               sum = 0;
-                               printk(KERN_INFO "log no csum found for "
-                                      "byte %llu\n",
-                                      (unsigned long long)disk_bytenr);
-                               item = NULL;
-                               btrfs_release_path(root, path);
-                               goto found;
-                       }
-                       btrfs_item_key_to_cpu(path->nodes[0], &found_key,
-                                             path->slots[0]);
-
-                       item_start_offset = found_key.offset;
-                       item_size = btrfs_item_size_nr(path->nodes[0],
-                                                      path->slots[0]);
-                       item_last_offset = item_start_offset +
-                               (item_size / csum_size) *
-                               root->sectorsize;
-                       item = btrfs_item_ptr(path->nodes[0], path->slots[0],
-                                             struct btrfs_csum_item);
-               }
-               /*
-                * this byte range must be able to fit inside
-                * a single leaf so it will also fit inside a u32
-                */
-               diff = disk_bytenr - item_start_offset;
-               diff = diff / root->sectorsize;
-               diff = diff * csum_size;
-
-               read_extent_buffer(path->nodes[0], &sum,
-                                  ((unsigned long)item) + diff,
-                                  csum_size);
-found:
-               sector_sum->bytenr = disk_bytenr;
-               sector_sum->sum = sum;
-               disk_bytenr += root->sectorsize;
-               sector_sum++;
+               if (ret)
+                       break;
+               btrfs_release_path(path);
        }
-       btrfs_free_path(path);
-       return 0;
+       btrfs_release_path(path);
+       return ret;
 }
 
 static noinline int copy_items(struct btrfs_trans_handle *trans,
@@ -2554,6 +2639,9 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
 
        ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
                           nr * sizeof(u32), GFP_NOFS);
+       if (!ins_data)
+               return -ENOMEM;
+
        ins_sizes = (u32 *)ins_data;
        ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
 
@@ -2563,9 +2651,12 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
        }
        ret = btrfs_insert_empty_items(trans, log, dst_path,
                                       ins_keys, ins_sizes, nr);
-       BUG_ON(ret);
+       if (ret) {
+               kfree(ins_data);
+               return ret;
+       }
 
-       for (i = 0; i < nr; i++) {
+       for (i = 0; i < nr; i++, dst_path->slots[0]++) {
                dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
                                                   dst_path->slots[0]);
 
@@ -2598,59 +2689,58 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
                        extent = btrfs_item_ptr(src, start_slot + i,
                                                struct btrfs_file_extent_item);
 
+                       if (btrfs_file_extent_generation(src, extent) < trans->transid)
+                               continue;
+
                        found_type = btrfs_file_extent_type(src, extent);
                        if (found_type == BTRFS_FILE_EXTENT_REG ||
                            found_type == BTRFS_FILE_EXTENT_PREALLOC) {
-                               u64 ds = btrfs_file_extent_disk_bytenr(src,
-                                                                  extent);
-                               u64 dl = btrfs_file_extent_disk_num_bytes(src,
-                                                                     extent);
-                               u64 cs = btrfs_file_extent_offset(src, extent);
-                               u64 cl = btrfs_file_extent_num_bytes(src,
-                                                                    extent);;
+                               u64 ds, dl, cs, cl;
+                               ds = btrfs_file_extent_disk_bytenr(src,
+                                                               extent);
+                               /* ds == 0 is a hole */
+                               if (ds == 0)
+                                       continue;
+
+                               dl = btrfs_file_extent_disk_num_bytes(src,
+                                                               extent);
+                               cs = btrfs_file_extent_offset(src, extent);
+                               cl = btrfs_file_extent_num_bytes(src,
+                                                               extent);
                                if (btrfs_file_extent_compression(src,
                                                                  extent)) {
                                        cs = 0;
                                        cl = dl;
                                }
-                               /* ds == 0 is a hole */
-                               if (ds != 0) {
-                                       ret = btrfs_inc_extent_ref(trans, log,
-                                                  ds, dl,
-                                                  dst_path->nodes[0]->start,
-                                                  BTRFS_TREE_LOG_OBJECTID,
-                                                  trans->transid,
-                                                  ins_keys[i].objectid);
-                                       BUG_ON(ret);
-                                       ret = copy_extent_csums(trans,
-                                               &ordered_sums,
+
+                               ret = btrfs_lookup_csums_range(
                                                log->fs_info->csum_root,
-                                               ds + cs, cl);
-                                       BUG_ON(ret);
-                               }
+                                               ds + cs, ds + cs + cl - 1,
+                                               &ordered_sums, 0);
+                               BUG_ON(ret);
                        }
                }
-               dst_path->slots[0]++;
        }
 
        btrfs_mark_buffer_dirty(dst_path->nodes[0]);
-       btrfs_release_path(log, dst_path);
+       btrfs_release_path(dst_path);
        kfree(ins_data);
 
        /*
         * we have to do this after the loop above to avoid changing the
         * log tree while trying to change the log tree.
         */
+       ret = 0;
        while (!list_empty(&ordered_sums)) {
                struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
                                                   struct btrfs_ordered_sum,
                                                   list);
-               ret = btrfs_csum_file_blocks(trans, log, sums);
-               BUG_ON(ret);
+               if (!ret)
+                       ret = btrfs_csum_file_blocks(trans, log, sums);
                list_del(&sums->list);
                kfree(sums);
        }
-       return 0;
+       return ret;
 }
 
 /* log a single inode in the tree log.
@@ -2667,7 +2757,7 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
  *
  * This handles both files and directories.
  */
-static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
+static int btrfs_log_inode(struct btrfs_trans_handle *trans,
                             struct btrfs_root *root, struct inode *inode,
                             int inode_only)
 {
@@ -2677,44 +2767,47 @@ static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
        struct btrfs_key max_key;
        struct btrfs_root *log = root->log_root;
        struct extent_buffer *src = NULL;
-       u32 size;
+       int err = 0;
        int ret;
        int nritems;
        int ins_start_slot = 0;
        int ins_nr;
+       u64 ino = btrfs_ino(inode);
 
        log = root->log_root;
 
        path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
        dst_path = btrfs_alloc_path();
+       if (!dst_path) {
+               btrfs_free_path(path);
+               return -ENOMEM;
+       }
 
-       min_key.objectid = inode->i_ino;
+       min_key.objectid = ino;
        min_key.type = BTRFS_INODE_ITEM_KEY;
        min_key.offset = 0;
 
-       max_key.objectid = inode->i_ino;
+       max_key.objectid = ino;
+
+       /* today the code can only do partial logging of directories */
+       if (!S_ISDIR(inode->i_mode))
+           inode_only = LOG_INODE_ALL;
+
        if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
                max_key.type = BTRFS_XATTR_ITEM_KEY;
        else
                max_key.type = (u8)-1;
        max_key.offset = (u64)-1;
 
-       /*
-        * if this inode has already been logged and we're in inode_only
-        * mode, we don't want to delete the things that have already
-        * been written to the log.
-        *
-        * But, if the inode has been through an inode_only log,
-        * the logged_trans field is not set.  This allows us to catch
-        * any new names for this inode in the backrefs by logging it
-        * again
-        */
-       if (inode_only == LOG_INODE_EXISTS &&
-           BTRFS_I(inode)->logged_trans == trans->transid) {
+       ret = btrfs_commit_inode_delayed_items(trans, inode);
+       if (ret) {
                btrfs_free_path(path);
                btrfs_free_path(dst_path);
-               goto out;
+               return ret;
        }
+
        mutex_lock(&BTRFS_I(inode)->log_mutex);
 
        /*
@@ -2726,12 +2819,14 @@ static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
 
                if (inode_only == LOG_INODE_EXISTS)
                        max_key_type = BTRFS_XATTR_ITEM_KEY;
-               ret = drop_objectid_items(trans, log, path,
-                                         inode->i_ino, max_key_type);
+               ret = drop_objectid_items(trans, log, path, ino, max_key_type);
        } else {
                ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
        }
-       BUG_ON(ret);
+       if (ret) {
+               err = ret;
+               goto out_unlock;
+       }
        path->keep_locks = 1;
 
        while (1) {
@@ -2742,13 +2837,12 @@ static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
                        break;
 again:
                /* note, ins_nr might be > 0 here, cleanup outside the loop */
-               if (min_key.objectid != inode->i_ino)
+               if (min_key.objectid != ino)
                        break;
                if (min_key.type > max_key.type)
                        break;
 
                src = path->nodes[0];
-               size = btrfs_item_size_nr(src, path->slots[0]);
                if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
                        ins_nr++;
                        goto next_slot;
@@ -2760,7 +2854,10 @@ again:
 
                ret = copy_items(trans, log, dst_path, src, ins_start_slot,
                                 ins_nr, inode_only);
-               BUG_ON(ret);
+               if (ret) {
+                       err = ret;
+                       goto out_unlock;
+               }
                ins_nr = 1;
                ins_start_slot = path->slots[0];
 next_slot:
@@ -2776,10 +2873,13 @@ next_slot:
                        ret = copy_items(trans, log, dst_path, src,
                                         ins_start_slot,
                                         ins_nr, inode_only);
-                       BUG_ON(ret);
+                       if (ret) {
+                               err = ret;
+                               goto out_unlock;
+                       }
                        ins_nr = 0;
                }
-               btrfs_release_path(root, path);
+               btrfs_release_path(path);
 
                if (min_key.offset < (u64)-1)
                        min_key.offset++;
@@ -2794,74 +2894,212 @@ next_slot:
                ret = copy_items(trans, log, dst_path, src,
                                 ins_start_slot,
                                 ins_nr, inode_only);
-               BUG_ON(ret);
+               if (ret) {
+                       err = ret;
+                       goto out_unlock;
+               }
                ins_nr = 0;
        }
        WARN_ON(ins_nr);
        if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
-               btrfs_release_path(root, path);
-               btrfs_release_path(log, dst_path);
-               BTRFS_I(inode)->log_dirty_trans = 0;
+               btrfs_release_path(path);
+               btrfs_release_path(dst_path);
                ret = log_directory_changes(trans, root, inode, path, dst_path);
-               BUG_ON(ret);
+               if (ret) {
+                       err = ret;
+                       goto out_unlock;
+               }
        }
        BTRFS_I(inode)->logged_trans = trans->transid;
+out_unlock:
        mutex_unlock(&BTRFS_I(inode)->log_mutex);
 
        btrfs_free_path(path);
        btrfs_free_path(dst_path);
+       return err;
+}
 
-       mutex_lock(&root->fs_info->tree_log_mutex);
-       ret = update_log_root(trans, log);
-       BUG_ON(ret);
-       mutex_unlock(&root->fs_info->tree_log_mutex);
+/*
+ * follow the dentry parent pointers up the chain and see if any
+ * of the directories in it require a full commit before they can
+ * be logged.  Returns zero if nothing special needs to be done or 1 if
+ * a full commit is required.
+ */
+static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
+                                              struct inode *inode,
+                                              struct dentry *parent,
+                                              struct super_block *sb,
+                                              u64 last_committed)
+{
+       int ret = 0;
+       struct btrfs_root *root;
+       struct dentry *old_parent = NULL;
+
+       /*
+        * for regular files, if its inode is already on disk, we don't
+        * have to worry about the parents at all.  This is because
+        * we can use the last_unlink_trans field to record renames
+        * and other fun in this file.
+        */
+       if (S_ISREG(inode->i_mode) &&
+           BTRFS_I(inode)->generation <= last_committed &&
+           BTRFS_I(inode)->last_unlink_trans <= last_committed)
+                       goto out;
+
+       if (!S_ISDIR(inode->i_mode)) {
+               if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
+                       goto out;
+               inode = parent->d_inode;
+       }
+
+       while (1) {
+               BTRFS_I(inode)->logged_trans = trans->transid;
+               smp_mb();
+
+               if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
+                       root = BTRFS_I(inode)->root;
+
+                       /*
+                        * make sure any commits to the log are forced
+                        * to be full commits
+                        */
+                       root->fs_info->last_trans_log_full_commit =
+                               trans->transid;
+                       ret = 1;
+                       break;
+               }
+
+               if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
+                       break;
+
+               if (IS_ROOT(parent))
+                       break;
+
+               parent = dget_parent(parent);
+               dput(old_parent);
+               old_parent = parent;
+               inode = parent->d_inode;
+
+       }
+       dput(old_parent);
 out:
-       return 0;
+       return ret;
 }
 
-int btrfs_log_inode(struct btrfs_trans_handle *trans,
-                   struct btrfs_root *root, struct inode *inode,
-                   int inode_only)
+static int inode_in_log(struct btrfs_trans_handle *trans,
+                struct inode *inode)
 {
-       int ret;
+       struct btrfs_root *root = BTRFS_I(inode)->root;
+       int ret = 0;
 
-       start_log_trans(trans, root);
-       ret = __btrfs_log_inode(trans, root, inode, inode_only);
-       end_log_trans(root);
+       mutex_lock(&root->log_mutex);
+       if (BTRFS_I(inode)->logged_trans == trans->transid &&
+           BTRFS_I(inode)->last_sub_trans <= root->last_log_commit)
+               ret = 1;
+       mutex_unlock(&root->log_mutex);
        return ret;
 }
 
+
 /*
  * helper function around btrfs_log_inode to make sure newly created
  * parent directories also end up in the log.  A minimal inode and backref
  * only logging is done of any parent directories that are older than
  * the last committed transaction
  */
-int btrfs_log_dentry(struct btrfs_trans_handle *trans,
-                   struct btrfs_root *root, struct dentry *dentry)
+int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
+                   struct btrfs_root *root, struct inode *inode,
+                   struct dentry *parent, int exists_only)
 {
-       int inode_only = LOG_INODE_ALL;
+       int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
        struct super_block *sb;
-       int ret;
+       struct dentry *old_parent = NULL;
+       int ret = 0;
+       u64 last_committed = root->fs_info->last_trans_committed;
+
+       sb = inode->i_sb;
+
+       if (btrfs_test_opt(root, NOTREELOG)) {
+               ret = 1;
+               goto end_no_trans;
+       }
+
+       if (root->fs_info->last_trans_log_full_commit >
+           root->fs_info->last_trans_committed) {
+               ret = 1;
+               goto end_no_trans;
+       }
+
+       if (root != BTRFS_I(inode)->root ||
+           btrfs_root_refs(&root->root_item) == 0) {
+               ret = 1;
+               goto end_no_trans;
+       }
+
+       ret = check_parent_dirs_for_sync(trans, inode, parent,
+                                        sb, last_committed);
+       if (ret)
+               goto end_no_trans;
+
+       if (inode_in_log(trans, inode)) {
+               ret = BTRFS_NO_LOG_SYNC;
+               goto end_no_trans;
+       }
+
+       ret = start_log_trans(trans, root);
+       if (ret)
+               goto end_trans;
+
+       ret = btrfs_log_inode(trans, root, inode, inode_only);
+       if (ret)
+               goto end_trans;
+
+       /*
+        * for regular files, if its inode is already on disk, we don't
+        * have to worry about the parents at all.  This is because
+        * we can use the last_unlink_trans field to record renames
+        * and other fun in this file.
+        */
+       if (S_ISREG(inode->i_mode) &&
+           BTRFS_I(inode)->generation <= last_committed &&
+           BTRFS_I(inode)->last_unlink_trans <= last_committed) {
+               ret = 0;
+               goto end_trans;
+       }
 
-       start_log_trans(trans, root);
-       sb = dentry->d_inode->i_sb;
+       inode_only = LOG_INODE_EXISTS;
        while (1) {
-               ret = __btrfs_log_inode(trans, root, dentry->d_inode,
-                                       inode_only);
-               BUG_ON(ret);
-               inode_only = LOG_INODE_EXISTS;
+               if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
+                       break;
 
-               dentry = dentry->d_parent;
-               if (!dentry || !dentry->d_inode || sb != dentry->d_inode->i_sb)
+               inode = parent->d_inode;
+               if (root != BTRFS_I(inode)->root)
                        break;
 
-               if (BTRFS_I(dentry->d_inode)->generation <=
-                   root->fs_info->last_trans_committed)
+               if (BTRFS_I(inode)->generation >
+                   root->fs_info->last_trans_committed) {
+                       ret = btrfs_log_inode(trans, root, inode, inode_only);
+                       if (ret)
+                               goto end_trans;
+               }
+               if (IS_ROOT(parent))
                        break;
+
+               parent = dget_parent(parent);
+               dput(old_parent);
+               old_parent = parent;
        }
-       end_log_trans(root);
-       return 0;
+       ret = 0;
+end_trans:
+       dput(old_parent);
+       if (ret < 0) {
+               BUG_ON(ret != -ENOSPC);
+               root->fs_info->last_trans_log_full_commit = trans->transid;
+               ret = 1;
+       }
+       btrfs_end_log_trans(root);
+end_no_trans:
+       return ret;
 }
 
 /*
@@ -2873,12 +3111,13 @@ int btrfs_log_dentry(struct btrfs_trans_handle *trans,
 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
                          struct btrfs_root *root, struct dentry *dentry)
 {
-       u64 gen;
-       gen = root->fs_info->last_trans_new_blockgroup;
-       if (gen > root->fs_info->last_trans_committed)
-               return 1;
-       else
-               return btrfs_log_dentry(trans, root, dentry);
+       struct dentry *parent = dget_parent(dentry);
+       int ret;
+
+       ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent, 0);
+       dput(parent);
+
+       return ret;
 }
 
 /*
@@ -2895,22 +3134,25 @@ int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
        struct btrfs_key tmp_key;
        struct btrfs_root *log;
        struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
-       u64 highest_inode;
        struct walk_control wc = {
                .process_func = process_one_buffer,
                .stage = 0,
        };
 
-       fs_info->log_root_recovering = 1;
        path = btrfs_alloc_path();
-       BUG_ON(!path);
+       if (!path)
+               return -ENOMEM;
 
-       trans = btrfs_start_transaction(fs_info->tree_root, 1);
+       fs_info->log_root_recovering = 1;
+
+       trans = btrfs_start_transaction(fs_info->tree_root, 0);
+       BUG_ON(IS_ERR(trans));
 
        wc.trans = trans;
        wc.pin = 1;
 
-       walk_log_tree(trans, log_root_tree, &wc);
+       ret = walk_log_tree(trans, log_root_tree, &wc);
+       BUG_ON(ret);
 
 again:
        key.objectid = BTRFS_TREE_LOG_OBJECTID;
@@ -2928,24 +3170,23 @@ again:
                }
                btrfs_item_key_to_cpu(path->nodes[0], &found_key,
                                      path->slots[0]);
-               btrfs_release_path(log_root_tree, path);
+               btrfs_release_path(path);
                if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
                        break;
 
                log = btrfs_read_fs_root_no_radix(log_root_tree,
                                                  &found_key);
-               BUG_ON(!log);
-
+               BUG_ON(IS_ERR(log));
 
                tmp_key.objectid = found_key.offset;
                tmp_key.type = BTRFS_ROOT_ITEM_KEY;
                tmp_key.offset = (u64)-1;
 
                wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
+               BUG_ON(IS_ERR_OR_NULL(wc.replay_dest));
 
-               BUG_ON(!wc.replay_dest);
-
-               btrfs_record_root_in_trans(wc.replay_dest);
+               wc.replay_dest->log_root = log;
+               btrfs_record_root_in_trans(trans, wc.replay_dest);
                ret = walk_log_tree(trans, log, &wc);
                BUG_ON(ret);
 
@@ -2954,20 +3195,17 @@ again:
                                                      path);
                        BUG_ON(ret);
                }
-               ret = btrfs_find_highest_inode(wc.replay_dest, &highest_inode);
-               if (ret == 0) {
-                       wc.replay_dest->highest_inode = highest_inode;
-                       wc.replay_dest->last_inode_alloc = highest_inode;
-               }
 
                key.offset = found_key.offset - 1;
+               wc.replay_dest->log_root = NULL;
                free_extent_buffer(log->node);
+               free_extent_buffer(log->commit_root);
                kfree(log);
 
                if (found_key.offset == 0)
                        break;
        }
-       btrfs_release_path(log_root_tree, path);
+       btrfs_release_path(path);
 
        /* step one is to pin it all, step two is to replay just inodes */
        if (wc.pin) {
@@ -2994,3 +3232,94 @@ again:
        kfree(log_root_tree);
        return 0;
 }
+
+/*
+ * there are some corner cases where we want to force a full
+ * commit instead of allowing a directory to be logged.
+ *
+ * They revolve around files there were unlinked from the directory, and
+ * this function updates the parent directory so that a full commit is
+ * properly done if it is fsync'd later after the unlinks are done.
+ */
+void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
+                            struct inode *dir, struct inode *inode,
+                            int for_rename)
+{
+       /*
+        * when we're logging a file, if it hasn't been renamed
+        * or unlinked, and its inode is fully committed on disk,
+        * we don't have to worry about walking up the directory chain
+        * to log its parents.
+        *
+        * So, we use the last_unlink_trans field to put this transid
+        * into the file.  When the file is logged we check it and
+        * don't log the parents if the file is fully on disk.
+        */
+       if (S_ISREG(inode->i_mode))
+               BTRFS_I(inode)->last_unlink_trans = trans->transid;
+
+       /*
+        * if this directory was already logged any new
+        * names for this file/dir will get recorded
+        */
+       smp_mb();
+       if (BTRFS_I(dir)->logged_trans == trans->transid)
+               return;
+
+       /*
+        * if the inode we're about to unlink was logged,
+        * the log will be properly updated for any new names
+        */
+       if (BTRFS_I(inode)->logged_trans == trans->transid)
+               return;
+
+       /*
+        * when renaming files across directories, if the directory
+        * there we're unlinking from gets fsync'd later on, there's
+        * no way to find the destination directory later and fsync it
+        * properly.  So, we have to be conservative and force commits
+        * so the new name gets discovered.
+        */
+       if (for_rename)
+               goto record;
+
+       /* we can safely do the unlink without any special recording */
+       return;
+
+record:
+       BTRFS_I(dir)->last_unlink_trans = trans->transid;
+}
+
+/*
+ * Call this after adding a new name for a file and it will properly
+ * update the log to reflect the new name.
+ *
+ * It will return zero if all goes well, and it will return 1 if a
+ * full transaction commit is required.
+ */
+int btrfs_log_new_name(struct btrfs_trans_handle *trans,
+                       struct inode *inode, struct inode *old_dir,
+                       struct dentry *parent)
+{
+       struct btrfs_root * root = BTRFS_I(inode)->root;
+
+       /*
+        * this will force the logging code to walk the dentry chain
+        * up for the file
+        */
+       if (S_ISREG(inode->i_mode))
+               BTRFS_I(inode)->last_unlink_trans = trans->transid;
+
+       /*
+        * if this inode hasn't been logged and directory we're renaming it
+        * from hasn't been logged, we don't need to log it
+        */
+       if (BTRFS_I(inode)->logged_trans <=
+           root->fs_info->last_trans_committed &&
+           (!old_dir || BTRFS_I(old_dir)->logged_trans <=
+                   root->fs_info->last_trans_committed))
+               return 0;
+
+       return btrfs_log_inode_parent(trans, root, inode, parent, 1);
+}
+