Btrfs: tree logging unlink/rename fixes

The tree logging code allows individual files or directories to be logged
without including operations on other files and directories in the FS.
It tries to commit the minimal set of changes to disk in order to
fsync the single file or directory that was sent to fsync or O_SYNC.

The tree logging code was allowing files and directories to be unlinked
if they were part of a rename operation where only one directory
in the rename was in the fsync log.  This patch adds a few new rules
to the tree logging.

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.  The commit isn't done during the unlink,
but it is forced the next time we try to log the parent directory.

Solution: record transid of last unlink/rename per directory when the
directory wasn't already logged.  For renames this is only done when
renaming to a different 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.  This way we make sure not to lose files that are unlinked during
the same transaction.

2a) we must log any new names for any file or dir during rename
when the directory they are being removed from was logged.

2a is actually the more important variant.  Without 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.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

7 files changed, 372 insertions, 98 deletions

diff --git a/fs/btrfs/btrfs_inode.h b/fs/btrfs/btrfs_inode.h
index 72677ce2b74f..3af4cfb5654c 100644
--- a/fs/btrfs/btrfs_inode.h
+++ b/fs/btrfs/btrfs_inode.h
@@ -86,12 +86,6 @@ struct btrfs_inode {
 	 */
 	u64 logged_trans;
 
-	/*
-	 * trans that last made a change that should be fully fsync'd.  This
-	 * gets reset to zero each time the inode is logged
-	 */
-	u64 log_dirty_trans;
-
 	/* total number of bytes pending delalloc, used by stat to calc the
 	 * real block usage of the file
 	 */
@@ -121,6 +115,13 @@ struct btrfs_inode {
 	/* the start of block group preferred for allocations. */
 	u64 block_group;
 
+	/* the fsync log has some corner cases that mean we have to check
+	 * directories to see if any unlinks have been done before
+	 * the directory was logged.  See tree-log.c for all the
+	 * details
+	 */
+	u64 last_unlink_trans;
+
 	struct inode vfs_inode;
 };
 
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 4ddce91cf3f9..2737facbd341 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -695,7 +695,12 @@ struct btrfs_fs_info {
 
 	u64 generation;
 	u64 last_trans_committed;
-	u64 last_trans_new_blockgroup;
+
+	/*
+	 * this is updated to the current trans every time a full commit
+	 * is required instead of the faster short fsync log commits
+	 */
+	u64 last_trans_log_full_commit;
 	u64 open_ioctl_trans;
 	unsigned long mount_opt;
 	u64 max_extent;
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 8933d15a240f..0c482e0d7c43 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -5897,7 +5897,7 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
 
 	extent_root = root->fs_info->extent_root;
 
-	root->fs_info->last_trans_new_blockgroup = trans->transid;
+	root->fs_info->last_trans_log_full_commit = trans->transid;
 
 	cache = kzalloc(sizeof(*cache), GFP_NOFS);
 	if (!cache)
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index f06c275644b7..32d10a617613 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -1173,8 +1173,11 @@ out_nolock:
 			ret = btrfs_log_dentry_safe(trans, root,
 						    file->f_dentry);
 			if (ret == 0) {
-				btrfs_sync_log(trans, root);
-				btrfs_end_transaction(trans, root);
+				ret = btrfs_sync_log(trans, root);
+				if (ret == 0)
+					btrfs_end_transaction(trans, root);
+				else
+					btrfs_commit_transaction(trans, root);
 			} else {
 				btrfs_commit_transaction(trans, root);
 			}
@@ -1266,8 +1269,11 @@ int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
 	if (ret > 0) {
 		ret = btrfs_commit_transaction(trans, root);
 	} else {
-		btrfs_sync_log(trans, root);
-		ret = btrfs_end_transaction(trans, root);
+		ret = btrfs_sync_log(trans, root);
+		if (ret == 0)
+			ret = btrfs_end_transaction(trans, root);
+		else
+			ret = btrfs_commit_transaction(trans, root);
 	}
 	mutex_lock(&dentry->d_inode->i_mutex);
 out:
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 9b4faac50c18..bffd79faffb5 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -2246,8 +2246,6 @@ int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
 	ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
 					 inode, dir->i_ino);
 	BUG_ON(ret != 0 && ret != -ENOENT);
-	if (ret != -ENOENT)
-		BTRFS_I(dir)->log_dirty_trans = trans->transid;
 
 	ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
 					   dir, index);
@@ -2280,6 +2278,9 @@ static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 	trans = btrfs_start_transaction(root, 1);
 
 	btrfs_set_trans_block_group(trans, dir);
+
+	btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
+
 	ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
 				 dentry->d_name.name, dentry->d_name.len);
 
@@ -3042,7 +3043,7 @@ static noinline void init_btrfs_i(struct inode *inode)
 	bi->disk_i_size = 0;
 	bi->flags = 0;
 	bi->index_cnt = (u64)-1;
-	bi->log_dirty_trans = 0;
+	bi->last_unlink_trans = 0;
 	extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
 	extent_io_tree_init(&BTRFS_I(inode)->io_tree,
 			     inode->i_mapping, GFP_NOFS);
@@ -3786,6 +3787,8 @@ static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
 		drop_inode = 1;
 
 	nr = trans->blocks_used;
+
+	btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
 	btrfs_end_transaction_throttle(trans, root);
 fail:
 	if (drop_inode) {
@@ -4666,6 +4669,15 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
 
 	trans = btrfs_start_transaction(root, 1);
 
+	/*
+	 * this is an ugly little race, but the rename is required to make
+	 * sure that if we crash, the inode is either at the old name
+	 * or the new one.  pinning the log transaction lets us make sure
+	 * we don't allow a log commit to come in after we unlink the
+	 * name but before we add the new name back in.
+	 */
+	btrfs_pin_log_trans(root);
+
 	btrfs_set_trans_block_group(trans, new_dir);
 
 	btrfs_inc_nlink(old_dentry->d_inode);
@@ -4673,6 +4685,9 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
 	new_dir->i_ctime = new_dir->i_mtime = ctime;
 	old_inode->i_ctime = ctime;
 
+	if (old_dentry->d_parent != new_dentry->d_parent)
+		btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
+
 	ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
 				 old_dentry->d_name.name,
 				 old_dentry->d_name.len);
@@ -4704,7 +4719,14 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
 	if (ret)
 		goto out_fail;
 
+	btrfs_log_new_name(trans, old_inode, old_dir,
+				       new_dentry->d_parent);
 out_fail:
+
+	/* this btrfs_end_log_trans just allows the current
+	 * log-sub transaction to complete
+	 */
+	btrfs_end_log_trans(root);
 	btrfs_end_transaction_throttle(trans, root);
 out_unlock:
 	return ret;
diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c
index 405439ca4c45..1b7f04a8f168 100644
--- a/fs/btrfs/tree-log.c
+++ b/fs/btrfs/tree-log.c
@@ -35,6 +35,49 @@
 #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
@@ -47,12 +90,17 @@
 #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
@@ -133,10 +181,25 @@ static int join_running_log_trans(struct btrfs_root *root)
 }
 
 /*
+ * 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;
+}
+
+/*
  * 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)
 {
 	if (atomic_dec_and_test(&root->log_writers)) {
 		smp_mb();
@@ -602,6 +665,7 @@ static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
 
 	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);
@@ -803,6 +867,7 @@ conflict_again:
 					    victim_name_len)) {
 				btrfs_inc_nlink(inode);
 				btrfs_release_path(root, path);
+
 				ret = btrfs_unlink_inode(trans, root, dir,
 							 inode, victim_name,
 							 victim_name_len);
@@ -921,13 +986,20 @@ static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
 		key.offset--;
 		btrfs_release_path(root, path);
 	}
-	btrfs_free_path(path);
+	btrfs_release_path(root, 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 && S_ISDIR(inode->i_mode)) {
+		ret = replay_dir_deletes(trans, root, NULL, path,
+					 inode->i_ino, 1);
+		BUG_ON(ret);
+	}
+	btrfs_free_path(path);
+
 	return 0;
 }
 
@@ -970,9 +1042,12 @@ static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
 
 		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;
@@ -1312,11 +1387,11 @@ 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,
@@ -1377,7 +1452,7 @@ 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;
@@ -1407,10 +1482,14 @@ 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) {
@@ -1436,7 +1515,8 @@ 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;
@@ -1513,7 +1593,7 @@ 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,
@@ -1850,7 +1930,8 @@ static int update_log_root(struct btrfs_trans_handle *trans,
 	return ret;
 }
 
-static int wait_log_commit(struct btrfs_root *root, unsigned long transid)
+static int wait_log_commit(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root, unsigned long transid)
 {
 	DEFINE_WAIT(wait);
 	int index = transid % 2;
@@ -1864,9 +1945,12 @@ static int wait_log_commit(struct btrfs_root *root, unsigned long transid)
 		prepare_to_wait(&root->log_commit_wait[index],
 				&wait, TASK_UNINTERRUPTIBLE);
 		mutex_unlock(&root->log_mutex);
-		if (root->log_transid < transid + 2 &&
+
+		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 &&
@@ -1874,14 +1958,16 @@ static int wait_log_commit(struct btrfs_root *root, unsigned long transid)
 	return 0;
 }
 
-static int wait_for_writer(struct btrfs_root *root)
+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 (atomic_read(&root->log_writers))
+		if (root->fs_info->last_trans_log_full_commit !=
+		    trans->transid && atomic_read(&root->log_writers))
 			schedule();
 		mutex_lock(&root->log_mutex);
 		finish_wait(&root->log_writer_wait, &wait);
@@ -1892,7 +1978,14 @@ static int wait_for_writer(struct btrfs_root *root)
 /*
  * 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)
@@ -1906,7 +1999,7 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
 	mutex_lock(&root->log_mutex);
 	index1 = root->log_transid % 2;
 	if (atomic_read(&root->log_commit[index1])) {
-		wait_log_commit(root, root->log_transid);
+		wait_log_commit(trans, root, root->log_transid);
 		mutex_unlock(&root->log_mutex);
 		return 0;
 	}
@@ -1914,18 +2007,26 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
 
 	/* wait for previous tree log sync to complete */
 	if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
-		wait_log_commit(root, root->log_transid - 1);
+		wait_log_commit(trans, root, root->log_transid - 1);
 
 	while (1) {
 		unsigned long batch = root->log_batch;
 		mutex_unlock(&root->log_mutex);
 		schedule_timeout_uninterruptible(1);
 		mutex_lock(&root->log_mutex);
-		wait_for_writer(root);
+
+		wait_for_writer(trans, root);
 		if (batch == root->log_batch)
 			break;
 	}
 
+	/* 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;
+	}
+
 	ret = btrfs_write_and_wait_marked_extents(log, &log->dirty_log_pages);
 	BUG_ON(ret);
 
@@ -1961,16 +2062,29 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
 
 	index2 = log_root_tree->log_transid % 2;
 	if (atomic_read(&log_root_tree->log_commit[index2])) {
-		wait_log_commit(log_root_tree, log_root_tree->log_transid);
+		wait_log_commit(trans, log_root_tree,
+				log_root_tree->log_transid);
 		mutex_unlock(&log_root_tree->log_mutex);
 		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(log_root_tree, log_root_tree->log_transid - 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);
 
-	wait_for_writer(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) {
+		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);
@@ -1995,7 +2109,9 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
 	 * in and cause problems either.
 	 */
 	write_ctree_super(trans, root->fs_info->tree_root, 2);
+	ret = 0;
 
+out_wake_log_root:
 	atomic_set(&log_root_tree->log_commit[index2], 0);
 	smp_mb();
 	if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
@@ -2008,7 +2124,8 @@ out:
 	return 0;
 }
 
-/* * free all the extents used by the tree log.  This should be called
+/*
+ * 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)
@@ -2142,7 +2259,7 @@ int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
 
 	btrfs_free_path(path);
 	mutex_unlock(&BTRFS_I(dir)->log_mutex);
-	end_log_trans(root);
+	btrfs_end_log_trans(root);
 
 	return 0;
 }
@@ -2169,7 +2286,7 @@ int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
 	ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino,
 				  dirid, &index);
 	mutex_unlock(&BTRFS_I(inode)->log_mutex);
-	end_log_trans(root);
+	btrfs_end_log_trans(root);
 
 	return ret;
 }
@@ -2569,7 +2686,7 @@ 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)
 {
@@ -2595,28 +2712,17 @@ static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
 	min_key.offset = 0;
 
 	max_key.objectid = inode->i_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) {
-		btrfs_free_path(path);
-		btrfs_free_path(dst_path);
-		goto out;
-	}
 	mutex_lock(&BTRFS_I(inode)->log_mutex);
 
 	/*
@@ -2703,7 +2809,6 @@ next_slot:
 	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;
 		ret = log_directory_changes(trans, root, inode, path, dst_path);
 		BUG_ON(ret);
 	}
@@ -2712,19 +2817,58 @@ next_slot:
 
 	btrfs_free_path(path);
 	btrfs_free_path(dst_path);
-out:
 	return 0;
 }
 
-int btrfs_log_inode(struct btrfs_trans_handle *trans,
-		    struct btrfs_root *root, struct inode *inode,
-		    int inode_only)
+/*
+ * 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;
+	int ret = 0;
+	struct btrfs_root *root;
 
-	start_log_trans(trans, root);
-	ret = __btrfs_log_inode(trans, root, inode, inode_only);
-	end_log_trans(root);
+	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 (parent == sb->s_root)
+			break;
+
+		parent = parent->d_parent;
+		inode = parent->d_inode;
+
+	}
+out:
 	return ret;
 }
 
@@ -2734,31 +2878,53 @@ int btrfs_log_inode(struct btrfs_trans_handle *trans,
  * 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;
+	int ret = 0;
+	u64 last_committed = root->fs_info->last_trans_committed;
+
+	sb = inode->i_sb;
+
+	if (root->fs_info->last_trans_log_full_commit >
+	    root->fs_info->last_trans_committed) {
+		ret = 1;
+		goto end_no_trans;
+	}
+
+	ret = check_parent_dirs_for_sync(trans, inode, parent,
+					 sb, last_committed);
+	if (ret)
+		goto end_no_trans;
 
 	start_log_trans(trans, root);
-	sb = dentry->d_inode->i_sb;
-	while (1) {
-		ret = __btrfs_log_inode(trans, root, dentry->d_inode,
-					inode_only);
-		BUG_ON(ret);
-		inode_only = LOG_INODE_EXISTS;
 
-		dentry = dentry->d_parent;
-		if (!dentry || !dentry->d_inode || sb != dentry->d_inode->i_sb)
+	ret = btrfs_log_inode(trans, root, inode, inode_only);
+	BUG_ON(ret);
+	inode_only = LOG_INODE_EXISTS;
+
+	while (1) {
+		if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
 			break;
 
-		if (BTRFS_I(dentry->d_inode)->generation <=
-		    root->fs_info->last_trans_committed)
+		inode = parent->d_inode;
+		if (BTRFS_I(inode)->generation >
+		    root->fs_info->last_trans_committed) {
+			ret = btrfs_log_inode(trans, root, inode, inode_only);
+			BUG_ON(ret);
+		}
+		if (parent == sb->s_root)
 			break;
+
+		parent = parent->d_parent;
 	}
-	end_log_trans(root);
-	return 0;
+	ret = 0;
+	btrfs_end_log_trans(root);
+end_no_trans:
+	return ret;
 }
 
 /*
@@ -2770,12 +2936,8 @@ 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);
+	return btrfs_log_inode_parent(trans, root, dentry->d_inode,
+				      dentry->d_parent, 0);
 }
 
 /*
@@ -2894,3 +3056,74 @@ 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)
+{
+	/*
+	 * 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;
+
+	/*
+	 * 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);
+}
+
diff --git a/fs/btrfs/tree-log.h b/fs/btrfs/tree-log.h
index b9409b32ed02..d09c7609e16b 100644
--- a/fs/btrfs/tree-log.h
+++ b/fs/btrfs/tree-log.h
@@ -22,14 +22,9 @@
 int btrfs_sync_log(struct btrfs_trans_handle *trans,
 		   struct btrfs_root *root);
 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root);
-int btrfs_log_dentry(struct btrfs_trans_handle *trans,
-		    struct btrfs_root *root, struct dentry *dentry);
 int btrfs_recover_log_trees(struct btrfs_root *tree_root);
 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
 			  struct btrfs_root *root, struct dentry *dentry);
-int btrfs_log_inode(struct btrfs_trans_handle *trans,
-		    struct btrfs_root *root, struct inode *inode,
-		    int inode_only);
 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
 				 struct btrfs_root *root,
 				 const char *name, int name_len,
@@ -38,4 +33,16 @@ int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
 			       struct btrfs_root *root,
 			       const char *name, int name_len,
 			       struct inode *inode, u64 dirid);
+int btrfs_join_running_log_trans(struct btrfs_root *root);
+int btrfs_end_log_trans(struct btrfs_root *root);
+int btrfs_pin_log_trans(struct btrfs_root *root);
+int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
+		    struct btrfs_root *root, struct inode *inode,
+		    struct dentry *parent, int exists_only);
+void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
+			     struct inode *dir, struct inode *inode,
+			     int for_rename);
+int btrfs_log_new_name(struct btrfs_trans_handle *trans,
+			struct inode *inode, struct inode *old_dir,
+			struct dentry *parent);
 #endif