Btrfs: Add a write ahead tree log to optimize synchronous operations
Chris Mason [Fri, 5 Sep 2008 20:13:11 +0000 (16:13 -0400)]
File syncs and directory syncs are optimized by copying their
items into a special (copy-on-write) log tree.  There is one log tree per
subvolume and the btrfs super block points to a tree of log tree roots.

After a crash, items are copied out of the log tree and back into the
subvolume.  See tree-log.c for all the details.

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

17 files changed:
fs/btrfs/Makefile
fs/btrfs/btrfs_inode.h
fs/btrfs/compat.h
fs/btrfs/ctree.c
fs/btrfs/ctree.h
fs/btrfs/dir-item.c
fs/btrfs/disk-io.c
fs/btrfs/disk-io.h
fs/btrfs/extent-tree.c
fs/btrfs/file.c
fs/btrfs/inode.c
fs/btrfs/root-tree.c
fs/btrfs/transaction.c
fs/btrfs/transaction.h
fs/btrfs/tree-defrag.c
fs/btrfs/tree-log.c [new file with mode: 0644]
fs/btrfs/tree-log.h [new file with mode: 0644]

index 3a01065..b7addbf 100644 (file)
@@ -7,8 +7,7 @@ btrfs-y := super.o ctree.o extent-tree.o print-tree.o root-tree.o dir-item.o \
           transaction.o bit-radix.o inode.o file.o tree-defrag.o \
           extent_map.o sysfs.o struct-funcs.o xattr.o ordered-data.o \
           extent_io.o volumes.o async-thread.o ioctl.o locking.o orphan.o \
-          ref-cache.o acl.o export.o
-
+          ref-cache.o export.o tree-log.o acl.o
 else
 
 # Normal Makefile
index 111f905..fcc8cf2 100644 (file)
@@ -33,6 +33,7 @@ struct btrfs_inode {
        struct extent_io_tree io_failure_tree;
        struct mutex csum_mutex;
        struct mutex extent_mutex;
+       struct mutex log_mutex;
        struct inode vfs_inode;
        struct btrfs_ordered_inode_tree ordered_tree;
 
@@ -44,10 +45,17 @@ struct btrfs_inode {
 
        struct list_head delalloc_inodes;
 
+       /* full 64 bit generation number */
+       u64 generation;
+
        /*
         * transid of the trans_handle that last modified this inode
         */
        u64 last_trans;
+       /*
+        * transid that last logged this inode
+        */
+       u64 logged_trans;
        u64 delalloc_bytes;
        u64 disk_i_size;
        u32 flags;
index d45fb37..b0ed188 100644 (file)
@@ -22,6 +22,21 @@ static inline struct dentry *d_obtain_alias(struct inode *inode)
 }
 #endif
 
+#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
+static inline void btrfs_drop_nlink(struct inode *inode)
+{
+       inode->i_nlink--;
+}
+
+static inline void btrfs_inc_nlink(struct inode *inode)
+{
+       inode->i_nlink++;
+}
+#else
+# define btrfs_drop_nlink(inode) drop_nlink(inode)
+# define btrfs_inc_nlink(inode)        inc_nlink(inode)
+#endif
+
 /*
  * Even if AppArmor isn't enabled, it still has different prototypes.
  * Add more distro/version pairs here to declare which has AppArmor applied.
index 7114faa..5791240 100644 (file)
@@ -60,7 +60,7 @@ void btrfs_free_path(struct btrfs_path *p)
        kmem_cache_free(btrfs_path_cachep, p);
 }
 
-void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
+void noinline btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
 {
        int i;
 
@@ -176,7 +176,7 @@ int btrfs_copy_root(struct btrfs_trans_handle *trans,
        return 0;
 }
 
-int __btrfs_cow_block(struct btrfs_trans_handle *trans,
+int noinline __btrfs_cow_block(struct btrfs_trans_handle *trans,
                             struct btrfs_root *root,
                             struct extent_buffer *buf,
                             struct extent_buffer *parent, int parent_slot,
@@ -294,7 +294,7 @@ int __btrfs_cow_block(struct btrfs_trans_handle *trans,
        return 0;
 }
 
-int btrfs_cow_block(struct btrfs_trans_handle *trans,
+int noinline btrfs_cow_block(struct btrfs_trans_handle *trans,
                    struct btrfs_root *root, struct extent_buffer *buf,
                    struct extent_buffer *parent, int parent_slot,
                    struct extent_buffer **cow_ret, u64 prealloc_dest)
@@ -677,9 +677,10 @@ static int noinline check_block(struct btrfs_root *root,
  *
  * slot may point to max if the key is bigger than all of the keys
  */
-static int generic_bin_search(struct extent_buffer *eb, unsigned long p,
-                             int item_size, struct btrfs_key *key,
-                             int max, int *slot)
+static noinline int generic_bin_search(struct extent_buffer *eb,
+                                      unsigned long p,
+                                      int item_size, struct btrfs_key *key,
+                                      int max, int *slot)
 {
        int low = 0;
        int high = max;
@@ -765,7 +766,7 @@ static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
        return -1;
 }
 
-static struct extent_buffer *read_node_slot(struct btrfs_root *root,
+static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
                                   struct extent_buffer *parent, int slot)
 {
        int level = btrfs_header_level(parent);
@@ -781,7 +782,7 @@ static struct extent_buffer *read_node_slot(struct btrfs_root *root,
                       btrfs_node_ptr_generation(parent, slot));
 }
 
-static int balance_level(struct btrfs_trans_handle *trans,
+static noinline int balance_level(struct btrfs_trans_handle *trans,
                         struct btrfs_root *root,
                         struct btrfs_path *path, int level)
 {
@@ -1128,8 +1129,9 @@ static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
 /*
  * readahead one full node of leaves
  */
-static void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
-                            int level, int slot, u64 objectid)
+static noinline void reada_for_search(struct btrfs_root *root,
+                                     struct btrfs_path *path,
+                                     int level, int slot, u64 objectid)
 {
        struct extent_buffer *node;
        struct btrfs_disk_key disk_key;
@@ -1201,7 +1203,8 @@ static void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
        }
 }
 
-static void unlock_up(struct btrfs_path *path, int level, int lowest_unlock)
+static noinline void unlock_up(struct btrfs_path *path, int level,
+                              int lowest_unlock)
 {
        int i;
        int skip_level = level;
@@ -1759,8 +1762,9 @@ static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
  *
  * returns 0 on success and < 0 on failure
  */
-static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
-                     *root, struct btrfs_path *path, int level)
+static noinline int split_node(struct btrfs_trans_handle *trans,
+                              struct btrfs_root *root,
+                              struct btrfs_path *path, int level)
 {
        u64 root_gen;
        struct extent_buffer *c;
@@ -1874,7 +1878,8 @@ static int leaf_space_used(struct extent_buffer *l, int start, int nr)
  * the start of the leaf data.  IOW, how much room
  * the leaf has left for both items and data
  */
-int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf)
+int noinline btrfs_leaf_free_space(struct btrfs_root *root,
+                                  struct extent_buffer *leaf)
 {
        int nritems = btrfs_header_nritems(leaf);
        int ret;
@@ -2283,9 +2288,11 @@ out:
  *
  * returns 0 if all went well and < 0 on failure.
  */
-static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
-                     *root, struct btrfs_key *ins_key,
-                     struct btrfs_path *path, int data_size, int extend)
+static noinline int split_leaf(struct btrfs_trans_handle *trans,
+                              struct btrfs_root *root,
+                              struct btrfs_key *ins_key,
+                              struct btrfs_path *path, int data_size,
+                              int extend)
 {
        u64 root_gen;
        struct extent_buffer *l;
@@ -3079,6 +3086,7 @@ int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
  * was nothing in the tree that matched the search criteria.
  */
 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
+                        struct btrfs_key *max_key,
                         struct btrfs_path *path, int cache_only,
                         u64 min_trans)
 {
@@ -3093,6 +3101,7 @@ int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
 again:
        cur = btrfs_lock_root_node(root);
        level = btrfs_header_level(cur);
+       WARN_ON(path->nodes[level]);
        path->nodes[level] = cur;
        path->locks[level] = 1;
 
@@ -3107,6 +3116,8 @@ again:
 
                /* at level = 0, we're done, setup the path and exit */
                if (level == 0) {
+                       if (slot >= nritems)
+                               goto find_next_key;
                        ret = 0;
                        path->slots[level] = slot;
                        btrfs_item_key_to_cpu(cur, &found_key, slot);
@@ -3123,6 +3134,8 @@ again:
                        u64 blockptr;
                        u64 gen;
                        struct extent_buffer *tmp;
+                       struct btrfs_disk_key disk_key;
+
                        blockptr = btrfs_node_blockptr(cur, slot);
                        gen = btrfs_node_ptr_generation(cur, slot);
                        if (gen < min_trans) {
@@ -3132,6 +3145,14 @@ again:
                        if (!cache_only)
                                break;
 
+                       if (max_key) {
+                               btrfs_node_key(cur, &disk_key, slot);
+                               if (comp_keys(&disk_key, max_key) >= 0) {
+                                       ret = 1;
+                                       goto out;
+                               }
+                       }
+
                        tmp = btrfs_find_tree_block(root, blockptr,
                                            btrfs_level_size(root, level - 1));
 
@@ -3143,14 +3164,16 @@ again:
                                free_extent_buffer(tmp);
                        slot++;
                }
+find_next_key:
                /*
                 * we didn't find a candidate key in this node, walk forward
                 * and find another one
                 */
                if (slot >= nritems) {
-                       ret = btrfs_find_next_key(root, path, min_key, level,
+                       path->slots[level] = slot;
+                       sret = btrfs_find_next_key(root, path, min_key, level,
                                                  cache_only, min_trans);
-                       if (ret == 0) {
+                       if (sret == 0) {
                                btrfs_release_path(root, path);
                                goto again;
                        } else {
@@ -3351,6 +3374,7 @@ int btrfs_previous_item(struct btrfs_root *root,
 {
        struct btrfs_key found_key;
        struct extent_buffer *leaf;
+       u32 nritems;
        int ret;
 
        while(1) {
@@ -3362,9 +3386,20 @@ int btrfs_previous_item(struct btrfs_root *root,
                        path->slots[0]--;
                }
                leaf = path->nodes[0];
+               nritems = btrfs_header_nritems(leaf);
+               if (nritems == 0)
+                       return 1;
+               if (path->slots[0] == nritems)
+                       path->slots[0]--;
+
                btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
                if (found_key.type == type)
                        return 0;
+               if (found_key.objectid < min_objectid)
+                       break;
+               if (found_key.objectid == min_objectid &&
+                   found_key.type < type)
+                       break;
        }
        return 1;
 }
index b305ae7..6532b60 100644 (file)
@@ -77,6 +77,10 @@ struct btrfs_ordered_sum;
 /* orhpan objectid for tracking unlinked/truncated files */
 #define BTRFS_ORPHAN_OBJECTID -5ULL
 
+/* does write ahead logging to speed up fsyncs */
+#define BTRFS_TREE_LOG_OBJECTID -6ULL
+#define BTRFS_TREE_LOG_FIXUP_OBJECTID -7ULL
+
 /*
  * All files have objectids higher than this.
  */
@@ -276,6 +280,7 @@ struct btrfs_super_block {
        __le64 generation;
        __le64 root;
        __le64 chunk_root;
+       __le64 log_root;
        __le64 total_bytes;
        __le64 bytes_used;
        __le64 root_dir_objectid;
@@ -287,6 +292,7 @@ struct btrfs_super_block {
        __le32 sys_chunk_array_size;
        u8 root_level;
        u8 chunk_root_level;
+       u8 log_root_level;
        struct btrfs_dev_item dev_item;
        char label[BTRFS_LABEL_SIZE];
        u8 sys_chunk_array[BTRFS_SYSTEM_CHUNK_ARRAY_SIZE];
@@ -392,7 +398,10 @@ struct btrfs_timespec {
  * make a new item type
  */
 struct btrfs_inode_item {
+       /* nfs style generation number */
        __le64 generation;
+       /* transid that last touched this inode */
+       __le64 transid;
        __le64 size;
        __le64 nblocks;
        __le64 block_group;
@@ -409,8 +418,13 @@ struct btrfs_inode_item {
        struct btrfs_timespec otime;
 } __attribute__ ((__packed__));
 
+struct btrfs_dir_log_item {
+       __le64 end;
+} __attribute__ ((__packed__));
+
 struct btrfs_dir_item {
        struct btrfs_disk_key location;
+       __le64 transid;
        __le16 data_len;
        __le16 name_len;
        u8 type;
@@ -505,6 +519,9 @@ struct btrfs_fs_info {
        struct btrfs_root *tree_root;
        struct btrfs_root *chunk_root;
        struct btrfs_root *dev_root;
+
+       /* the log root tree is a directory of all the other log roots */
+       struct btrfs_root *log_root_tree;
        struct radix_tree_root fs_roots_radix;
 
        struct extent_io_tree free_space_cache;
@@ -518,6 +535,7 @@ struct btrfs_fs_info {
 
        u64 generation;
        u64 last_trans_committed;
+       u64 last_trans_new_blockgroup;
        u64 open_ioctl_trans;
        unsigned long mount_opt;
        u64 max_extent;
@@ -527,6 +545,9 @@ struct btrfs_fs_info {
        wait_queue_head_t transaction_throttle;
        wait_queue_head_t transaction_wait;
        wait_queue_head_t async_submit_wait;
+
+       wait_queue_head_t tree_log_wait;
+
        struct btrfs_super_block super_copy;
        struct btrfs_super_block super_for_commit;
        struct block_device *__bdev;
@@ -535,6 +556,7 @@ struct btrfs_fs_info {
        struct backing_dev_info bdi;
        spinlock_t hash_lock;
        struct mutex trans_mutex;
+       struct mutex tree_log_mutex;
        struct mutex transaction_kthread_mutex;
        struct mutex cleaner_mutex;
        struct mutex alloc_mutex;
@@ -544,8 +566,13 @@ struct btrfs_fs_info {
        struct list_head trans_list;
        struct list_head hashers;
        struct list_head dead_roots;
+
        atomic_t nr_async_submits;
        atomic_t nr_async_bios;
+       atomic_t tree_log_writers;
+       atomic_t tree_log_commit;
+       unsigned long tree_log_batch;
+       u64 tree_log_transid;
 
        /*
         * this is used by the balancing code to wait for all the pending
@@ -583,6 +610,7 @@ struct btrfs_fs_info {
        struct completion kobj_unregister;
        int do_barriers;
        int closing;
+       int log_root_recovering;
        atomic_t throttles;
        atomic_t throttle_gen;
 
@@ -596,6 +624,7 @@ struct btrfs_fs_info {
        u64 delalloc_bytes;
        u64 last_alloc;
        u64 last_data_alloc;
+       u64 last_log_alloc;
 
        spinlock_t ref_cache_lock;
        u64 total_ref_cache_size;
@@ -632,6 +661,7 @@ struct btrfs_root {
        struct btrfs_leaf_ref_tree *ref_tree;
        struct btrfs_leaf_ref_tree ref_tree_struct;
        struct btrfs_dirty_root *dirty_root;
+       struct btrfs_root *log_root;
 
        struct btrfs_root_item root_item;
        struct btrfs_key root_key;
@@ -640,6 +670,7 @@ struct btrfs_root {
        struct kobject root_kobj;
        struct completion kobj_unregister;
        struct mutex objectid_mutex;
+       struct mutex log_mutex;
 
        u64 objectid;
        u64 last_trans;
@@ -692,6 +723,8 @@ struct btrfs_root {
  * dir items are the name -> inode pointers in a directory.  There is one
  * for every name in a directory.
  */
+#define BTRFS_DIR_LOG_ITEM_KEY  14
+#define BTRFS_DIR_LOG_INDEX_KEY 15
 #define BTRFS_DIR_ITEM_KEY     16
 #define BTRFS_DIR_INDEX_KEY    17
 /*
@@ -703,7 +736,8 @@ struct btrfs_root {
  */
 #define BTRFS_CSUM_ITEM_KEY    19
 
-/* reserve 20-31 for other file stuff */
+
+/* reserve 21-31 for other file/dir stuff */
 
 /*
  * root items point to tree roots.  There are typically in the root
@@ -938,6 +972,7 @@ BTRFS_SETGET_FUNCS(inode_ref_index, struct btrfs_inode_ref, index, 64);
 
 /* struct btrfs_inode_item */
 BTRFS_SETGET_FUNCS(inode_generation, struct btrfs_inode_item, generation, 64);
+BTRFS_SETGET_FUNCS(inode_transid, struct btrfs_inode_item, transid, 64);
 BTRFS_SETGET_FUNCS(inode_size, struct btrfs_inode_item, size, 64);
 BTRFS_SETGET_FUNCS(inode_nblocks, struct btrfs_inode_item, nblocks, 64);
 BTRFS_SETGET_FUNCS(inode_block_group, struct btrfs_inode_item, block_group, 64);
@@ -1126,10 +1161,13 @@ static inline void btrfs_set_item_key(struct extent_buffer *eb,
        write_eb_member(eb, item, struct btrfs_item, key, disk_key);
 }
 
+BTRFS_SETGET_FUNCS(dir_log_end, struct btrfs_dir_log_item, end, 64);
+
 /* struct btrfs_dir_item */
 BTRFS_SETGET_FUNCS(dir_data_len, struct btrfs_dir_item, data_len, 16);
 BTRFS_SETGET_FUNCS(dir_type, struct btrfs_dir_item, type, 8);
 BTRFS_SETGET_FUNCS(dir_name_len, struct btrfs_dir_item, name_len, 16);
+BTRFS_SETGET_FUNCS(dir_transid, struct btrfs_dir_item, transid, 64);
 
 static inline void btrfs_dir_item_key(struct extent_buffer *eb,
                                      struct btrfs_dir_item *item,
@@ -1301,7 +1339,11 @@ BTRFS_SETGET_STACK_FUNCS(super_root_level, struct btrfs_super_block,
 BTRFS_SETGET_STACK_FUNCS(super_chunk_root, struct btrfs_super_block,
                         chunk_root, 64);
 BTRFS_SETGET_STACK_FUNCS(super_chunk_root_level, struct btrfs_super_block,
-                        chunk_root_level, 64);
+                        chunk_root_level, 8);
+BTRFS_SETGET_STACK_FUNCS(super_log_root, struct btrfs_super_block,
+                        log_root, 64);
+BTRFS_SETGET_STACK_FUNCS(super_log_root_level, struct btrfs_super_block,
+                        log_root_level, 8);
 BTRFS_SETGET_STACK_FUNCS(super_total_bytes, struct btrfs_super_block,
                         total_bytes, 64);
 BTRFS_SETGET_STACK_FUNCS(super_bytes_used, struct btrfs_super_block,
@@ -1405,6 +1447,12 @@ static inline struct dentry *fdentry(struct file *file) {
 }
 
 /* extent-tree.c */
+int btrfs_lookup_extent(struct btrfs_root *root, struct btrfs_path *path,
+                       u64 start, u64 len);
+int btrfs_update_pinned_extents(struct btrfs_root *root,
+                               u64 bytenr, u64 num, int pin);
+int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
+                       struct btrfs_root *root, struct extent_buffer *leaf);
 int btrfs_cross_ref_exists(struct btrfs_trans_handle *trans,
                           struct btrfs_root *root,
                           struct btrfs_key *key, u64 bytenr);
@@ -1448,6 +1496,11 @@ int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
                                u64 root_objectid, u64 ref_generation,
                                u64 owner, u64 owner_offset,
                                struct btrfs_key *ins);
+int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
+                               struct btrfs_root *root,
+                               u64 root_objectid, u64 ref_generation,
+                               u64 owner, u64 owner_offset,
+                               struct btrfs_key *ins);
 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
                                  struct btrfs_root *root,
                                  u64 num_bytes, u64 min_alloc_size,
@@ -1488,9 +1541,9 @@ int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
                        struct btrfs_key *key, int lowest_level,
                        int cache_only, u64 min_trans);
 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
+                        struct btrfs_key *max_key,
                         struct btrfs_path *path, int cache_only,
                         u64 min_trans);
-
 int btrfs_cow_block(struct btrfs_trans_handle *trans,
                    struct btrfs_root *root, struct extent_buffer *buf,
                    struct extent_buffer *parent, int parent_slot,
@@ -1656,6 +1709,18 @@ int btrfs_csum_truncate(struct btrfs_trans_handle *trans,
 #define PageChecked PageFsMisc
 #endif
 
+int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
+                      struct btrfs_root *root,
+                      struct inode *dir, struct inode *inode,
+                      const char *name, int name_len);
+int btrfs_add_link(struct btrfs_trans_handle *trans,
+                  struct inode *parent_inode, struct inode *inode,
+                  const char *name, int name_len, int add_backref, u64 index);
+int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
+                              struct btrfs_root *root,
+                              struct inode *inode, u64 new_size,
+                              u32 min_type);
+
 int btrfs_start_delalloc_inodes(struct btrfs_root *root);
 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end);
 int btrfs_writepages(struct address_space *mapping,
@@ -1715,6 +1780,7 @@ int btrfs_update_inode(struct btrfs_trans_handle *trans,
 long btrfs_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
 
 /* file.c */
+int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync);
 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end);
 int btrfs_check_file(struct btrfs_root *root, struct inode *inode);
 extern struct file_operations btrfs_file_operations;
index 1250946..e4f3009 100644 (file)
@@ -96,6 +96,7 @@ int btrfs_insert_xattr_item(struct btrfs_trans_handle *trans,
        btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
        btrfs_set_dir_type(leaf, dir_item, BTRFS_FT_XATTR);
        btrfs_set_dir_name_len(leaf, dir_item, name_len);
+       btrfs_set_dir_transid(leaf, dir_item, trans->transid);
        btrfs_set_dir_data_len(leaf, dir_item, data_len);
        name_ptr = (unsigned long)(dir_item + 1);
        data_ptr = (unsigned long)((char *)name_ptr + name_len);
@@ -142,6 +143,7 @@ int btrfs_insert_dir_item(struct btrfs_trans_handle *trans, struct btrfs_root
        btrfs_set_dir_type(leaf, dir_item, type);
        btrfs_set_dir_data_len(leaf, dir_item, 0);
        btrfs_set_dir_name_len(leaf, dir_item, name_len);
+       btrfs_set_dir_transid(leaf, dir_item, trans->transid);
        name_ptr = (unsigned long)(dir_item + 1);
 
        write_extent_buffer(leaf, name, name_ptr, name_len);
@@ -169,6 +171,7 @@ second_insert:
        btrfs_set_dir_type(leaf, dir_item, type);
        btrfs_set_dir_data_len(leaf, dir_item, 0);
        btrfs_set_dir_name_len(leaf, dir_item, name_len);
+       btrfs_set_dir_transid(leaf, dir_item, trans->transid);
        name_ptr = (unsigned long)(dir_item + 1);
        write_extent_buffer(leaf, name, name_ptr, name_len);
        btrfs_mark_buffer_dirty(leaf);
index 8e7a938..a4373db 100644 (file)
@@ -41,6 +41,7 @@
 #include "async-thread.h"
 #include "locking.h"
 #include "ref-cache.h"
+#include "tree-log.h"
 
 #if 0
 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
@@ -694,6 +695,18 @@ struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 }
 
 
+int btrfs_write_tree_block(struct extent_buffer *buf)
+{
+       return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
+                                     buf->start + buf->len - 1, WB_SYNC_NONE);
+}
+
+int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
+{
+       return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
+                                 buf->start, buf->start + buf->len -1);
+}
+
 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
                                      u32 blocksize, u64 parent_transid)
 {
@@ -732,15 +745,6 @@ int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
        return 0;
 }
 
-int wait_on_tree_block_writeback(struct btrfs_root *root,
-                                struct extent_buffer *buf)
-{
-       struct inode *btree_inode = root->fs_info->btree_inode;
-       wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
-                                       buf);
-       return 0;
-}
-
 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
                        u32 stripesize, struct btrfs_root *root,
                        struct btrfs_fs_info *fs_info,
@@ -771,6 +775,7 @@ static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
        spin_lock_init(&root->node_lock);
        spin_lock_init(&root->list_lock);
        mutex_init(&root->objectid_mutex);
+       mutex_init(&root->log_mutex);
 
        btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
        root->ref_tree = &root->ref_tree_struct;
@@ -809,11 +814,74 @@ static int find_and_setup_root(struct btrfs_root *tree_root,
        return 0;
 }
 
-struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
-                                              struct btrfs_key *location)
+int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
+                            struct btrfs_fs_info *fs_info)
+{
+       struct extent_buffer *eb;
+       int ret;
+
+       if (!fs_info->log_root_tree)
+               return 0;
+
+       eb = fs_info->log_root_tree->node;
+
+       WARN_ON(btrfs_header_level(eb) != 0);
+       WARN_ON(btrfs_header_nritems(eb) != 0);
+
+       ret = btrfs_free_extent(trans, fs_info->tree_root,
+                               eb->start, eb->len,
+                               BTRFS_TREE_LOG_OBJECTID, 0, 0, 0, 1);
+       BUG_ON(ret);
+
+       free_extent_buffer(eb);
+       kfree(fs_info->log_root_tree);
+       fs_info->log_root_tree = NULL;
+       return 0;
+}
+
+int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
+                            struct btrfs_fs_info *fs_info)
 {
        struct btrfs_root *root;
        struct btrfs_root *tree_root = fs_info->tree_root;
+
+       root = kzalloc(sizeof(*root), GFP_NOFS);
+       if (!root)
+               return -ENOMEM;
+
+       __setup_root(tree_root->nodesize, tree_root->leafsize,
+                    tree_root->sectorsize, tree_root->stripesize,
+                    root, fs_info, BTRFS_TREE_LOG_OBJECTID);
+
+       root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
+       root->root_key.type = BTRFS_ROOT_ITEM_KEY;
+       root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
+       root->ref_cows = 0;
+
+       root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
+                                           BTRFS_TREE_LOG_OBJECTID,
+                                           0, 0, 0, 0, 0);
+
+       btrfs_set_header_nritems(root->node, 0);
+       btrfs_set_header_level(root->node, 0);
+       btrfs_set_header_bytenr(root->node, root->node->start);
+       btrfs_set_header_generation(root->node, trans->transid);
+       btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
+
+       write_extent_buffer(root->node, root->fs_info->fsid,
+                           (unsigned long)btrfs_header_fsid(root->node),
+                           BTRFS_FSID_SIZE);
+       btrfs_mark_buffer_dirty(root->node);
+       btrfs_tree_unlock(root->node);
+       fs_info->log_root_tree = root;
+       return 0;
+}
+
+struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
+                                              struct btrfs_key *location)
+{
+       struct btrfs_root *root;
+       struct btrfs_fs_info *fs_info = tree_root->fs_info;
        struct btrfs_path *path;
        struct extent_buffer *l;
        u64 highest_inode;
@@ -863,11 +931,13 @@ out:
                                     blocksize, 0);
        BUG_ON(!root->node);
 insert:
-       root->ref_cows = 1;
-       ret = btrfs_find_highest_inode(root, &highest_inode);
-       if (ret == 0) {
-               root->highest_inode = highest_inode;
-               root->last_inode_alloc = highest_inode;
+       if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
+               root->ref_cows = 1;
+               ret = btrfs_find_highest_inode(root, &highest_inode);
+               if (ret == 0) {
+                       root->highest_inode = highest_inode;
+                       root->last_inode_alloc = highest_inode;
+               }
        }
        return root;
 }
@@ -907,7 +977,7 @@ struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
        if (root)
                return root;
 
-       root = btrfs_read_fs_root_no_radix(fs_info, location);
+       root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
        if (IS_ERR(root))
                return root;
        ret = radix_tree_insert(&fs_info->fs_roots_radix,
@@ -1250,16 +1320,18 @@ struct btrfs_root *open_ctree(struct super_block *sb,
        u32 blocksize;
        u32 stripesize;
        struct buffer_head *bh;
-       struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
+       struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
                                                 GFP_NOFS);
-       struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
+       struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
                                               GFP_NOFS);
        struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
                                                GFP_NOFS);
-       struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
+       struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
                                                GFP_NOFS);
-       struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
+       struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
                                              GFP_NOFS);
+       struct btrfs_root *log_tree_root;
+
        int ret;
        int err = -EINVAL;
 
@@ -1343,6 +1415,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
        mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
 
        mutex_init(&fs_info->trans_mutex);
+       mutex_init(&fs_info->tree_log_mutex);
        mutex_init(&fs_info->drop_mutex);
        mutex_init(&fs_info->alloc_mutex);
        mutex_init(&fs_info->chunk_mutex);
@@ -1352,6 +1425,10 @@ struct btrfs_root *open_ctree(struct super_block *sb,
        init_waitqueue_head(&fs_info->transaction_throttle);
        init_waitqueue_head(&fs_info->transaction_wait);
        init_waitqueue_head(&fs_info->async_submit_wait);
+       init_waitqueue_head(&fs_info->tree_log_wait);
+       atomic_set(&fs_info->tree_log_commit, 0);
+       atomic_set(&fs_info->tree_log_writers, 0);
+       fs_info->tree_log_transid = 0;
 
 #if 0
        ret = add_hasher(fs_info, "crc32c");
@@ -1532,7 +1609,26 @@ struct btrfs_root *open_ctree(struct super_block *sb,
        if (!fs_info->transaction_kthread)
                goto fail_cleaner;
 
+       if (btrfs_super_log_root(disk_super) != 0) {
+               u32 blocksize;
+               u64 bytenr = btrfs_super_log_root(disk_super);
+
+               blocksize =
+                    btrfs_level_size(tree_root,
+                                     btrfs_super_log_root_level(disk_super));
 
+               log_tree_root = kzalloc(sizeof(struct btrfs_root),
+                                                     GFP_NOFS);
+
+               __setup_root(nodesize, leafsize, sectorsize, stripesize,
+                            log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
+
+               log_tree_root->node = read_tree_block(tree_root, bytenr,
+                                                     blocksize, 0);
+               ret = btrfs_recover_log_trees(log_tree_root);
+               BUG_ON(ret);
+       }
+       fs_info->last_trans_committed = btrfs_super_generation(disk_super);
        return tree_root;
 
 fail_cleaner:
index 2562a27..6b6fdc6 100644 (file)
@@ -45,7 +45,7 @@ struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
                                      struct btrfs_key *location,
                                      const char *name, int namelen);
-struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
+struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
                                               struct btrfs_key *location);
 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
                                              struct btrfs_key *location);
@@ -74,4 +74,10 @@ int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
                        extent_submit_bio_hook_t *submit_bio_hook);
 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone);
 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info);
+int btrfs_write_tree_block(struct extent_buffer *buf);
+int btrfs_wait_tree_block_writeback(struct extent_buffer *buf);
+int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
+                            struct btrfs_fs_info *fs_info);
+int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
+                            struct btrfs_fs_info *fs_info);
 #endif
index e63b3b4..646b914 100644 (file)
@@ -496,6 +496,23 @@ static int match_extent_ref(struct extent_buffer *leaf,
        return ret == 0;
 }
 
+/* simple helper to search for an existing extent at a given offset */
+int btrfs_lookup_extent(struct btrfs_root *root, struct btrfs_path *path,
+                       u64 start, u64 len)
+{
+       int ret;
+       struct btrfs_key key;
+
+       maybe_lock_mutex(root);
+       key.objectid = start;
+       key.offset = len;
+       btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
+       ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
+                               0, 0);
+       maybe_unlock_mutex(root);
+       return ret;
+}
+
 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
                                          struct btrfs_root *root,
                                          struct btrfs_path *path, u64 bytenr,
@@ -1409,7 +1426,7 @@ static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
 }
 
 
-static int update_pinned_extents(struct btrfs_root *root,
+int btrfs_update_pinned_extents(struct btrfs_root *root,
                                u64 bytenr, u64 num, int pin)
 {
        u64 len;
@@ -1492,7 +1509,7 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
                                            EXTENT_DIRTY);
                if (ret)
                        break;
-               update_pinned_extents(root, start, end + 1 - start, 0);
+               btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
                clear_extent_dirty(unpin, start, end, GFP_NOFS);
                set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
                if (need_resched()) {
@@ -1538,14 +1555,11 @@ static int finish_current_insert(struct btrfs_trans_handle *trans,
                clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
                                  GFP_NOFS);
 
-               eb = btrfs_find_tree_block(extent_root, ins.objectid,
+               eb = btrfs_find_create_tree_block(extent_root, ins.objectid,
                                           ins.offset);
 
-               if (!btrfs_buffer_uptodate(eb, trans->transid)) {
-                       mutex_unlock(&extent_root->fs_info->alloc_mutex);
+               if (!btrfs_buffer_uptodate(eb, trans->transid))
                        btrfs_read_buffer(eb, trans->transid);
-                       mutex_lock(&extent_root->fs_info->alloc_mutex);
-               }
 
                btrfs_tree_lock(eb);
                level = btrfs_header_level(eb);
@@ -1585,13 +1599,20 @@ static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
                struct extent_buffer *buf;
                buf = btrfs_find_tree_block(root, bytenr, num_bytes);
                if (buf) {
+                       /* we can reuse a block if it hasn't been written
+                        * and it is from this transaction.  We can't
+                        * reuse anything from the tree log root because
+                        * it has tiny sub-transactions.
+                        */
                        if (btrfs_buffer_uptodate(buf, 0) &&
                            btrfs_try_tree_lock(buf)) {
                                u64 transid =
                                    root->fs_info->running_transaction->transid;
                                u64 header_transid =
                                        btrfs_header_generation(buf);
-                               if (header_transid == transid &&
+                               if (btrfs_header_owner(buf) !=
+                                   BTRFS_TREE_LOG_OBJECTID &&
+                                   header_transid == transid &&
                                    !btrfs_header_flag(buf,
                                               BTRFS_HEADER_FLAG_WRITTEN)) {
                                        clean_tree_block(NULL, root, buf);
@@ -1603,7 +1624,7 @@ static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
                        }
                        free_extent_buffer(buf);
                }
-               update_pinned_extents(root, bytenr, num_bytes, 1);
+               btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
        } else {
                set_extent_bits(&root->fs_info->pending_del,
                                bytenr, bytenr + num_bytes - 1,
@@ -1801,7 +1822,7 @@ static int del_pending_extents(struct btrfs_trans_handle *trans, struct
                                  GFP_NOFS);
                if (!test_range_bit(&extent_root->fs_info->extent_ins,
                                    start, end, EXTENT_LOCKED, 0)) {
-                       update_pinned_extents(extent_root, start,
+                       btrfs_update_pinned_extents(extent_root, start,
                                              end + 1 - start, 1);
                        ret = __free_extent(trans, extent_root,
                                             start, end + 1 - start,
@@ -1919,6 +1940,12 @@ static int noinline find_free_extent(struct btrfs_trans_handle *trans,
        if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
                last_ptr = &root->fs_info->last_data_alloc;
        }
+       if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
+               last_ptr = &root->fs_info->last_log_alloc;
+               if (!last_ptr == 0 && root->fs_info->last_alloc) {
+                       *last_ptr = root->fs_info->last_alloc + empty_cluster;
+               }
+       }
 
        if (last_ptr) {
                if (*last_ptr)
@@ -2268,6 +2295,35 @@ int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
        maybe_unlock_mutex(root);
        return ret;
 }
+
+/*
+ * this is used by the tree logging recovery code.  It records that
+ * an extent has been allocated and makes sure to clear the free
+ * space cache bits as well
+ */
+int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
+                               struct btrfs_root *root,
+                               u64 root_objectid, u64 ref_generation,
+                               u64 owner, u64 owner_offset,
+                               struct btrfs_key *ins)
+{
+       int ret;
+       struct btrfs_block_group_cache *block_group;
+
+       maybe_lock_mutex(root);
+       block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
+       cache_block_group(root, block_group);
+
+       clear_extent_dirty(&root->fs_info->free_space_cache,
+                          ins->objectid, ins->objectid + ins->offset - 1,
+                          GFP_NOFS);
+       ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid,
+                                           ref_generation, owner,
+                                           owner_offset, ins);
+       maybe_unlock_mutex(root);
+       return ret;
+}
+
 /*
  * finds a free extent and does all the dirty work required for allocation
  * returns the key for the extent through ins, and a tree buffer for
@@ -2350,9 +2406,8 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
        return buf;
 }
 
-static int noinline drop_leaf_ref_no_cache(struct btrfs_trans_handle *trans,
-                                          struct btrfs_root *root,
-                                          struct extent_buffer *leaf)
+int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
+                       struct btrfs_root *root, struct extent_buffer *leaf)
 {
        u64 leaf_owner;
        u64 leaf_generation;
@@ -2402,9 +2457,9 @@ static int noinline drop_leaf_ref_no_cache(struct btrfs_trans_handle *trans,
        return 0;
 }
 
-static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
-                                        struct btrfs_root *root,
-                                        struct btrfs_leaf_ref *ref)
+static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
+                                       struct btrfs_root *root,
+                                       struct btrfs_leaf_ref *ref)
 {
        int i;
        int ret;
@@ -2512,7 +2567,7 @@ static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
                    btrfs_header_nritems(cur))
                        break;
                if (*level == 0) {
-                       ret = drop_leaf_ref_no_cache(trans, root, cur);
+                       ret = btrfs_drop_leaf_ref(trans, root, cur);
                        BUG_ON(ret);
                        break;
                }
@@ -2552,7 +2607,7 @@ static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
                        btrfs_node_key_to_cpu(cur, &key, path->slots[*level]);
                        ref = btrfs_lookup_leaf_ref(root, bytenr);
                        if (ref) {
-                               ret = drop_leaf_ref(trans, root, ref);
+                               ret = cache_drop_leaf_ref(trans, root, ref);
                                BUG_ON(ret);
                                btrfs_remove_leaf_ref(root, ref);
                                btrfs_free_leaf_ref(root, ref);
@@ -3628,6 +3683,8 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
        extent_root = root->fs_info->extent_root;
        block_group_cache = &root->fs_info->block_group_cache;
 
+       root->fs_info->last_trans_new_blockgroup = trans->transid;
+
        cache = kzalloc(sizeof(*cache), GFP_NOFS);
        BUG_ON(!cache);
        cache->key.objectid = chunk_offset;
index e9e86fb..84ecf3a 100644 (file)
@@ -36,6 +36,8 @@
 #include "btrfs_inode.h"
 #include "ioctl.h"
 #include "print-tree.h"
+#include "tree-log.h"
+#include "locking.h"
 #include "compat.h"
 
 
@@ -988,10 +990,27 @@ out_nolock:
        *ppos = pos;
 
        if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
-               err = sync_page_range(inode, inode->i_mapping,
-                                     start_pos, num_written);
+               struct btrfs_trans_handle *trans;
+
+               err = btrfs_fdatawrite_range(inode->i_mapping, start_pos,
+                                            start_pos + num_written -1,
+                                            WB_SYNC_NONE);
+               if (err < 0)
+                       num_written = err;
+
+               err = btrfs_wait_on_page_writeback_range(inode->i_mapping,
+                                start_pos, start_pos + num_written - 1);
                if (err < 0)
                        num_written = err;
+
+               trans = btrfs_start_transaction(root, 1);
+               ret = btrfs_log_dentry_safe(trans, root, file->f_dentry);
+               if (ret == 0) {
+                       btrfs_sync_log(trans, root);
+                       btrfs_end_transaction(trans, root);
+               } else {
+                       btrfs_commit_transaction(trans, root);
+               }
        } else if (num_written > 0 && (file->f_flags & O_DIRECT)) {
 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
                do_sync_file_range(file, start_pos,
@@ -1019,8 +1038,7 @@ int btrfs_release_file(struct inode * inode, struct file * filp)
        return 0;
 }
 
-static int btrfs_sync_file(struct file *file,
-                          struct dentry *dentry, int datasync)
+int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
 {
        struct inode *inode = dentry->d_inode;
        struct btrfs_root *root = BTRFS_I(inode)->root;
@@ -1043,6 +1061,8 @@ static int btrfs_sync_file(struct file *file,
        }
        mutex_unlock(&root->fs_info->trans_mutex);
 
+       filemap_fdatawait(inode->i_mapping);
+
        /*
         * ok we haven't committed the transaction yet, lets do a commit
         */
@@ -1054,7 +1074,16 @@ static int btrfs_sync_file(struct file *file,
                ret = -ENOMEM;
                goto out;
        }
-       ret = btrfs_commit_transaction(trans, root);
+
+       ret = btrfs_log_dentry_safe(trans, root, file->f_dentry);
+       if (ret < 0)
+               goto out;
+       if (ret > 0) {
+               ret = btrfs_commit_transaction(trans, root);
+       } else {
+               btrfs_sync_log(trans, root);
+               ret = btrfs_end_transaction(trans, root);
+       }
 out:
        return ret > 0 ? EIO : ret;
 }
index 43d3f26..65df9d8 100644 (file)
@@ -46,6 +46,8 @@
 #include "volumes.h"
 #include "ordered-data.h"
 #include "xattr.h"
+#include "compat.h"
+#include "tree-log.h"
 
 struct btrfs_iget_args {
        u64 ino;
@@ -586,6 +588,7 @@ nocow:
                          &ordered_extent->list);
 
        btrfs_ordered_update_i_size(inode, ordered_extent);
+       btrfs_update_inode(trans, root, inode);
        btrfs_remove_ordered_extent(inode, ordered_extent);
 
        /* once for us */
@@ -593,7 +596,6 @@ nocow:
        /* once for the tree */
        btrfs_put_ordered_extent(ordered_extent);
 
-       btrfs_update_inode(trans, root, inode);
        btrfs_end_transaction(trans, root);
        return 0;
 }
@@ -1007,7 +1009,8 @@ void btrfs_read_locked_inode(struct inode *inode)
        inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 
        inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
-       inode->i_generation = btrfs_inode_generation(leaf, inode_item);
+       BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
+       inode->i_generation = BTRFS_I(inode)->generation;
        inode->i_rdev = 0;
        rdev = btrfs_inode_rdev(leaf, inode_item);
 
@@ -1056,7 +1059,8 @@ make_bad:
        make_bad_inode(inode);
 }
 
-static void fill_inode_item(struct extent_buffer *leaf,
+static void fill_inode_item(struct btrfs_trans_handle *trans,
+                           struct extent_buffer *leaf,
                            struct btrfs_inode_item *item,
                            struct inode *inode)
 {
@@ -1082,7 +1086,8 @@ static void fill_inode_item(struct extent_buffer *leaf,
                                inode->i_ctime.tv_nsec);
 
        btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
-       btrfs_set_inode_generation(leaf, item, inode->i_generation);
+       btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
+       btrfs_set_inode_transid(leaf, item, trans->transid);
        btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
        btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
        btrfs_set_inode_block_group(leaf, item,
@@ -1112,7 +1117,7 @@ int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
        inode_item = btrfs_item_ptr(leaf, path->slots[0],
                                  struct btrfs_inode_item);
 
-       fill_inode_item(leaf, inode_item, inode);
+       fill_inode_item(trans, leaf, inode_item, inode);
        btrfs_mark_buffer_dirty(leaf);
        btrfs_set_inode_last_trans(trans, inode);
        ret = 0;
@@ -1122,14 +1127,12 @@ failed:
 }
 
 
-static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
-                             struct btrfs_root *root,
-                             struct inode *dir,
-                             struct dentry *dentry)
+int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
+                      struct btrfs_root *root,
+                      struct inode *dir, struct inode *inode,
+                      const char *name, int name_len)
 {
        struct btrfs_path *path;
-       const char *name = dentry->d_name.name;
-       int name_len = dentry->d_name.len;
        int ret = 0;
        struct extent_buffer *leaf;
        struct btrfs_dir_item *di;
@@ -1160,13 +1163,12 @@ static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
        btrfs_release_path(root, path);
 
        ret = btrfs_del_inode_ref(trans, root, name, name_len,
-                                 dentry->d_inode->i_ino,
-                                 dentry->d_parent->d_inode->i_ino, &index);
+                                 inode->i_ino,
+                                 dir->i_ino, &index);
        if (ret) {
                printk("failed to delete reference to %.*s, "
                       "inode %lu parent %lu\n", name_len, name,
-                      dentry->d_inode->i_ino,
-                      dentry->d_parent->d_inode->i_ino);
+                      inode->i_ino, dir->i_ino);
                goto err;
        }
 
@@ -1183,21 +1185,25 @@ static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
        ret = btrfs_delete_one_dir_name(trans, root, path, di);
        btrfs_release_path(root, path);
 
-       dentry->d_inode->i_ctime = dir->i_ctime;
+       ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
+                                        inode, dir->i_ino);
+       BUG_ON(ret);
+
+       ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
+                                          dir, index);
+       BUG_ON(ret);
 err:
        btrfs_free_path(path);
-       if (!ret) {
-               btrfs_i_size_write(dir, dir->i_size - name_len * 2);
-               dir->i_mtime = dir->i_ctime = CURRENT_TIME;
-               btrfs_update_inode(trans, root, dir);
-#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
-               dentry->d_inode->i_nlink--;
-#else
-               drop_nlink(dentry->d_inode);
-#endif
-               ret = btrfs_update_inode(trans, root, dentry->d_inode);
-               dir->i_sb->s_dirt = 1;
-       }
+       if (ret)
+               goto out;
+
+       btrfs_i_size_write(dir, dir->i_size - name_len * 2);
+       inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
+       btrfs_update_inode(trans, root, dir);
+       btrfs_drop_nlink(inode);
+       ret = btrfs_update_inode(trans, root, inode);
+       dir->i_sb->s_dirt = 1;
+out:
        return ret;
 }
 
@@ -1218,7 +1224,8 @@ static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
        trans = btrfs_start_transaction(root, 1);
 
        btrfs_set_trans_block_group(trans, dir);
-       ret = btrfs_unlink_trans(trans, root, dir, dentry);
+       ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
+                                dentry->d_name.name, dentry->d_name.len);
 
        if (inode->i_nlink == 0)
                ret = btrfs_orphan_add(trans, inode);
@@ -1256,7 +1263,8 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
                goto fail_trans;
 
        /* now the directory is empty */
-       err = btrfs_unlink_trans(trans, root, dir, dentry);
+       err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
+                                dentry->d_name.name, dentry->d_name.len);
        if (!err) {
                btrfs_i_size_write(inode, 0);
        }
@@ -1283,10 +1291,10 @@ fail:
  * min_type is the minimum key type to truncate down to.  If set to 0, this
  * will kill all the items on this inode, including the INODE_ITEM_KEY.
  */
-static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
-                                  struct btrfs_root *root,
-                                  struct inode *inode,
-                                  u32 min_type)
+noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
+                                       struct btrfs_root *root,
+                                       struct inode *inode,
+                                       u64 new_size, u32 min_type)
 {
        int ret;
        struct btrfs_path *path;
@@ -1307,7 +1315,9 @@ static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
        int extent_type = -1;
        u64 mask = root->sectorsize - 1;
 
-       btrfs_drop_extent_cache(inode, inode->i_size & (~mask), (u64)-1);
+       if (root->ref_cows)
+               btrfs_drop_extent_cache(inode,
+                                       new_size & (~mask), (u64)-1);
        path = btrfs_alloc_path();
        path->reada = -1;
        BUG_ON(!path);
@@ -1324,7 +1334,13 @@ search_again:
                goto error;
        }
        if (ret > 0) {
-               BUG_ON(path->slots[0] == 0);
+               /* there are no items in the tree for us to truncate, we're
+                * done
+                */
+               if (path->slots[0] == 0) {
+                       ret = 0;
+                       goto error;
+               }
                path->slots[0]--;
        }
 
@@ -1358,10 +1374,10 @@ search_again:
                }
                if (found_type == BTRFS_CSUM_ITEM_KEY) {
                        ret = btrfs_csum_truncate(trans, root, path,
-                                                 inode->i_size);
+                                                 new_size);
                        BUG_ON(ret);
                }
-               if (item_end < inode->i_size) {
+               if (item_end < new_size) {
                        if (found_type == BTRFS_DIR_ITEM_KEY) {
                                found_type = BTRFS_INODE_ITEM_KEY;
                        } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
@@ -1378,7 +1394,7 @@ search_again:
                        btrfs_set_key_type(&key, found_type);
                        goto next;
                }
-               if (found_key.offset >= inode->i_size)
+               if (found_key.offset >= new_size)
                        del_item = 1;
                else
                        del_item = 0;
@@ -1394,7 +1410,7 @@ search_again:
                        if (!del_item) {
                                u64 orig_num_bytes =
                                        btrfs_file_extent_num_bytes(leaf, fi);
-                               extent_num_bytes = inode->i_size -
+                               extent_num_bytes = new_size -
                                        found_key.offset + root->sectorsize - 1;
                                extent_num_bytes = extent_num_bytes &
                                        ~((u64)root->sectorsize - 1);
@@ -1402,7 +1418,7 @@ search_again:
                                                         extent_num_bytes);
                                num_dec = (orig_num_bytes -
                                           extent_num_bytes);
-                               if (extent_start != 0)
+                               if (root->ref_cows && extent_start != 0)
                                        dec_i_blocks(inode, num_dec);
                                btrfs_mark_buffer_dirty(leaf);
                        } else {
@@ -1413,22 +1429,29 @@ search_again:
                                num_dec = btrfs_file_extent_num_bytes(leaf, fi);
                                if (extent_start != 0) {
                                        found_extent = 1;
-                                       dec_i_blocks(inode, num_dec);
+                                       if (root->ref_cows)
+                                               dec_i_blocks(inode, num_dec);
+                               }
+                               if (root->ref_cows) {
+                                       root_gen =
+                                               btrfs_header_generation(leaf);
                                }
-                               root_gen = btrfs_header_generation(leaf);
                                root_owner = btrfs_header_owner(leaf);
                        }
                } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
                        if (!del_item) {
-                               u32 newsize = inode->i_size - found_key.offset;
-                               dec_i_blocks(inode, item_end + 1 -
-                                           found_key.offset - newsize);
-                               newsize =
-                                   btrfs_file_extent_calc_inline_size(newsize);
+                               u32 size = new_size - found_key.offset;
+
+                               if (root->ref_cows) {
+                                       dec_i_blocks(inode, item_end + 1 -
+                                                   found_key.offset - size);
+                               }
+                               size =
+                                   btrfs_file_extent_calc_inline_size(size);
                                ret = btrfs_truncate_item(trans, root, path,
-                                                         newsize, 1);
+                                                         size, 1);
                                BUG_ON(ret);
-                       } else {
+                       } else if (root->ref_cows) {
                                dec_i_blocks(inode, item_end + 1 -
                                             found_key.offset);
                        }
@@ -1666,7 +1689,7 @@ void btrfs_delete_inode(struct inode *inode)
        trans = btrfs_start_transaction(root, 1);
 
        btrfs_set_trans_block_group(trans, inode);
-       ret = btrfs_truncate_in_trans(trans, root, inode, 0);
+       ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
        if (ret) {
                btrfs_orphan_del(NULL, inode);
                goto no_delete_lock;
@@ -1753,15 +1776,20 @@ static int fixup_tree_root_location(struct btrfs_root *root,
        return 0;
 }
 
-static int btrfs_init_locked_inode(struct inode *inode, void *p)
+static noinline void init_btrfs_i(struct inode *inode)
 {
-       struct btrfs_iget_args *args = p;
-       inode->i_ino = args->ino;
-       BTRFS_I(inode)->root = args->root;
-       BTRFS_I(inode)->delalloc_bytes = 0;
-       inode->i_mapping->writeback_index = 0;
-       BTRFS_I(inode)->disk_i_size = 0;
-       BTRFS_I(inode)->index_cnt = (u64)-1;
+       struct btrfs_inode *bi = BTRFS_I(inode);
+
+       bi->i_acl = NULL;
+       bi->i_default_acl = NULL;
+
+       bi->generation = 0;
+       bi->last_trans = 0;
+       bi->logged_trans = 0;
+       bi->delalloc_bytes = 0;
+       bi->disk_i_size = 0;
+       bi->flags = 0;
+       bi->index_cnt = (u64)-1;
        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);
@@ -1771,6 +1799,15 @@ static int btrfs_init_locked_inode(struct inode *inode, void *p)
        btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
        mutex_init(&BTRFS_I(inode)->csum_mutex);
        mutex_init(&BTRFS_I(inode)->extent_mutex);
+       mutex_init(&BTRFS_I(inode)->log_mutex);
+}
+
+static int btrfs_init_locked_inode(struct inode *inode, void *p)
+{
+       struct btrfs_iget_args *args = p;
+       inode->i_ino = args->ino;
+       init_btrfs_i(inode);
+       BTRFS_I(inode)->root = args->root;
        return 0;
 }
 
@@ -2263,21 +2300,10 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
         * btrfs_get_inode_index_count has an explanation for the magic
         * number
         */
+       init_btrfs_i(inode);
        BTRFS_I(inode)->index_cnt = 2;
-
-       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);
-       extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
-                            inode->i_mapping, GFP_NOFS);
-       btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
-       INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
-       mutex_init(&BTRFS_I(inode)->csum_mutex);
-       mutex_init(&BTRFS_I(inode)->extent_mutex);
-       BTRFS_I(inode)->delalloc_bytes = 0;
-       inode->i_mapping->writeback_index = 0;
-       BTRFS_I(inode)->disk_i_size = 0;
        BTRFS_I(inode)->root = root;
+       BTRFS_I(inode)->generation = trans->transid;
 
        if (mode & S_IFDIR)
                owner = 0;
@@ -2290,7 +2316,6 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
                new_inode_group = group;
        }
        BTRFS_I(inode)->block_group = new_inode_group;
-       BTRFS_I(inode)->flags = 0;
 
        key[0].objectid = objectid;
        btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
@@ -2318,7 +2343,7 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
        inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
        inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
                                  struct btrfs_inode_item);
-       fill_inode_item(path->nodes[0], inode_item, inode);
+       fill_inode_item(trans, path->nodes[0], inode_item, inode);
 
        ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
                             struct btrfs_inode_ref);
@@ -2349,38 +2374,34 @@ static inline u8 btrfs_inode_type(struct inode *inode)
        return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
 }
 
-static int btrfs_add_link(struct btrfs_trans_handle *trans,
-                           struct dentry *dentry, struct inode *inode,
-                           int add_backref, u64 index)
+int btrfs_add_link(struct btrfs_trans_handle *trans,
+                  struct inode *parent_inode, struct inode *inode,
+                  const char *name, int name_len, int add_backref, u64 index)
 {
        int ret;
        struct btrfs_key key;
-       struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
-       struct inode *parent_inode = dentry->d_parent->d_inode;
+       struct btrfs_root *root = BTRFS_I(parent_inode)->root;
 
        key.objectid = inode->i_ino;
        btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
        key.offset = 0;
 
-       ret = btrfs_insert_dir_item(trans, root,
-                                   dentry->d_name.name, dentry->d_name.len,
-                                   dentry->d_parent->d_inode->i_ino,
+       ret = btrfs_insert_dir_item(trans, root, name, name_len,
+                                   parent_inode->i_ino,
                                    &key, btrfs_inode_type(inode),
                                    index);
        if (ret == 0) {
                if (add_backref) {
                        ret = btrfs_insert_inode_ref(trans, root,
-                                            dentry->d_name.name,
-                                            dentry->d_name.len,
-                                            inode->i_ino,
-                                            parent_inode->i_ino,
-                                            index);
+                                                    name, name_len,
+                                                    inode->i_ino,
+                                                    parent_inode->i_ino,
+                                                    index);
                }
                btrfs_i_size_write(parent_inode, parent_inode->i_size +
-                                  dentry->d_name.len * 2);
+                                  name_len * 2);
                parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
-               ret = btrfs_update_inode(trans, root,
-                                        dentry->d_parent->d_inode);
+               ret = btrfs_update_inode(trans, root, parent_inode);
        }
        return ret;
 }
@@ -2389,7 +2410,9 @@ static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
                            struct dentry *dentry, struct inode *inode,
                            int backref, u64 index)
 {
-       int err = btrfs_add_link(trans, dentry, inode, backref, index);
+       int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
+                                inode, dentry->d_name.name,
+                                dentry->d_name.len, backref, index);
        if (!err) {
                d_instantiate(dentry, inode);
                return 0;
@@ -2513,19 +2536,7 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
                inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
                inode->i_fop = &btrfs_file_operations;
                inode->i_op = &btrfs_file_inode_operations;
-               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);
-               extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
-                                    inode->i_mapping, GFP_NOFS);
-               INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
-               mutex_init(&BTRFS_I(inode)->csum_mutex);
-               mutex_init(&BTRFS_I(inode)->extent_mutex);
-               BTRFS_I(inode)->delalloc_bytes = 0;
-               BTRFS_I(inode)->disk_i_size = 0;
-               inode->i_mapping->writeback_index = 0;
                BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
-               btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
        }
        dir->i_sb->s_dirt = 1;
        btrfs_update_inode_block_group(trans, inode);
@@ -2556,11 +2567,7 @@ static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
        if (inode->i_nlink == 0)
                return -ENOENT;
 
-#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
-       inode->i_nlink++;
-#else
-       inc_nlink(inode);
-#endif
+       btrfs_inc_nlink(inode);
        err = btrfs_check_free_space(root, 1, 0);
        if (err)
                goto fail;
@@ -2650,7 +2657,9 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
        if (err)
                goto out_fail;
 
-       err = btrfs_add_link(trans, dentry, inode, 0, index);
+       err = btrfs_add_link(trans, dentry->d_parent->d_inode,
+                                inode, dentry->d_name.name,
+                                dentry->d_name.len, 0, index);
        if (err)
                goto out_fail;
 
@@ -3221,7 +3230,7 @@ static void btrfs_truncate(struct inode *inode)
        if (ret)
                goto out;
        /* FIXME, add redo link to tree so we don't leak on crash */
-       ret = btrfs_truncate_in_trans(trans, root, inode,
+       ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
                                      BTRFS_EXTENT_DATA_KEY);
        btrfs_update_inode(trans, root, inode);
 
@@ -3304,6 +3313,7 @@ struct inode *btrfs_alloc_inode(struct super_block *sb)
        if (!ei)
                return NULL;
        ei->last_trans = 0;
+       ei->logged_trans = 0;
        btrfs_ordered_inode_tree_init(&ei->ordered_tree);
        ei->i_acl = BTRFS_ACL_NOT_CACHED;
        ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
@@ -3463,31 +3473,39 @@ static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
 
        btrfs_set_trans_block_group(trans, new_dir);
 
-       old_dentry->d_inode->i_nlink++;
+       btrfs_inc_nlink(old_dentry->d_inode);
        old_dir->i_ctime = old_dir->i_mtime = ctime;
        new_dir->i_ctime = new_dir->i_mtime = ctime;
        old_inode->i_ctime = ctime;
 
-       ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
+       ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
+                                old_dentry->d_name.name,
+                                old_dentry->d_name.len);
        if (ret)
                goto out_fail;
 
        if (new_inode) {
                new_inode->i_ctime = CURRENT_TIME;
-               ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
+               ret = btrfs_unlink_inode(trans, root, new_dir,
+                                        new_dentry->d_inode,
+                                        new_dentry->d_name.name,
+                                        new_dentry->d_name.len);
                if (ret)
                        goto out_fail;
                if (new_inode->i_nlink == 0) {
-                       ret = btrfs_orphan_add(trans, new_inode);
+                       ret = btrfs_orphan_add(trans, new_dentry->d_inode);
                        if (ret)
                                goto out_fail;
                }
+
        }
        ret = btrfs_set_inode_index(new_dir, old_inode, &index);
        if (ret)
                goto out_fail;
 
-       ret = btrfs_add_link(trans, new_dentry, old_inode, 1, index);
+       ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
+                            old_inode, new_dentry->d_name.name,
+                            new_dentry->d_name.len, 1, index);
        if (ret)
                goto out_fail;
 
@@ -3577,19 +3595,7 @@ static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
                inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
                inode->i_fop = &btrfs_file_operations;
                inode->i_op = &btrfs_file_inode_operations;
-               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);
-               extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
-                                    inode->i_mapping, GFP_NOFS);
-               INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
-               mutex_init(&BTRFS_I(inode)->csum_mutex);
-               mutex_init(&BTRFS_I(inode)->extent_mutex);
-               BTRFS_I(inode)->delalloc_bytes = 0;
-               BTRFS_I(inode)->disk_i_size = 0;
-               inode->i_mapping->writeback_index = 0;
                BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
-               btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
        }
        dir->i_sb->s_dirt = 1;
        btrfs_update_inode_block_group(trans, inode);
@@ -3691,6 +3697,7 @@ static struct file_operations btrfs_dir_file_operations = {
        .compat_ioctl   = btrfs_ioctl,
 #endif
        .release        = btrfs_release_file,
+       .fsync          = btrfs_sync_file,
 };
 
 static struct extent_io_ops btrfs_extent_io_ops = {
index 3672669..e3984f9 100644 (file)
@@ -202,8 +202,9 @@ again:
                memcpy(&found_key, &key, sizeof(key));
                key.offset++;
                btrfs_release_path(root, path);
-               dead_root = btrfs_read_fs_root_no_radix(root->fs_info,
-                                                       &found_key);
+               dead_root =
+                       btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
+                                                   &found_key);
                if (IS_ERR(dead_root)) {
                        ret = PTR_ERR(dead_root);
                        goto err;
index eff3ad7..49c4f5b 100644 (file)
@@ -25,6 +25,7 @@
 #include "transaction.h"
 #include "locking.h"
 #include "ref-cache.h"
+#include "tree-log.h"
 
 static int total_trans = 0;
 extern struct kmem_cache *btrfs_trans_handle_cachep;
@@ -57,6 +58,7 @@ static noinline int join_transaction(struct btrfs_root *root)
                root->fs_info->generation++;
                root->fs_info->last_alloc = 0;
                root->fs_info->last_data_alloc = 0;
+               root->fs_info->last_log_alloc = 0;
                cur_trans->num_writers = 1;
                cur_trans->num_joined = 0;
                cur_trans->transid = root->fs_info->generation;
@@ -83,7 +85,7 @@ static noinline int join_transaction(struct btrfs_root *root)
        return 0;
 }
 
-static noinline int record_root_in_trans(struct btrfs_root *root)
+noinline int btrfs_record_root_in_trans(struct btrfs_root *root)
 {
        struct btrfs_dirty_root *dirty;
        u64 running_trans_id = root->fs_info->running_transaction->transid;
@@ -151,7 +153,7 @@ static void wait_current_trans(struct btrfs_root *root)
        }
 }
 
-struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
+static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
                                             int num_blocks, int wait)
 {
        struct btrfs_trans_handle *h =
@@ -164,7 +166,7 @@ struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
        ret = join_transaction(root);
        BUG_ON(ret);
 
-       record_root_in_trans(root);
+       btrfs_record_root_in_trans(root);
        h->transid = root->fs_info->running_transaction->transid;
        h->transaction = root->fs_info->running_transaction;
        h->blocks_reserved = num_blocks;
@@ -456,6 +458,8 @@ static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
                        BUG_ON(!root->ref_tree);
                        dirty = root->dirty_root;
 
+                       btrfs_free_log(trans, root);
+
                        if (root->commit_root == root->node) {
                                WARN_ON(root->node->start !=
                                        btrfs_root_bytenr(&root->root_item));
@@ -600,7 +604,7 @@ static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
                num_bytes -= btrfs_root_used(&dirty->root->root_item);
                bytes_used = btrfs_root_used(&root->root_item);
                if (num_bytes) {
-                       record_root_in_trans(root);
+                       btrfs_record_root_in_trans(root);
                        btrfs_set_root_used(&root->root_item,
                                            bytes_used - num_bytes);
                }
@@ -745,7 +749,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
        int ret;
 
        INIT_LIST_HEAD(&dirty_fs_roots);
-
        mutex_lock(&root->fs_info->trans_mutex);
        if (trans->transaction->in_commit) {
                cur_trans = trans->transaction;
@@ -821,10 +824,30 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 
        WARN_ON(cur_trans != trans->transaction);
 
+       /* btrfs_commit_tree_roots is responsible for getting the
+        * various roots consistent with each other.  Every pointer
+        * in the tree of tree roots has to point to the most up to date
+        * root for every subvolume and other tree.  So, we have to keep
+        * the tree logging code from jumping in and changing any
+        * of the trees.
+        *
+        * At this point in the commit, there can't be any tree-log
+        * writers, but a little lower down we drop the trans mutex
+        * and let new people in.  By holding the tree_log_mutex
+        * from now until after the super is written, we avoid races
+        * with the tree-log code.
+        */
+       mutex_lock(&root->fs_info->tree_log_mutex);
+
        ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
                              &dirty_fs_roots);
        BUG_ON(ret);
 
+       /* add_dirty_roots gets rid of all the tree log roots, it is now
+        * safe to free the root of tree log roots
+        */
+       btrfs_free_log_root_tree(trans, root->fs_info);
+
        ret = btrfs_commit_tree_roots(trans, root);
        BUG_ON(ret);
 
@@ -843,6 +866,12 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
                                   chunk_root->node->start);
        btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
                                         btrfs_header_level(chunk_root->node));
+
+       if (!root->fs_info->log_root_recovering) {
+               btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
+               btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
+       }
+
        memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
               sizeof(root->fs_info->super_copy));
 
@@ -857,6 +886,12 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
        BUG_ON(ret);
        write_ctree_super(trans, root);
 
+       /*
+        * the super is written, we can safely allow the tree-loggers
+        * to go about their business
+        */
+       mutex_unlock(&root->fs_info->tree_log_mutex);
+
        btrfs_finish_extent_commit(trans, root, pinned_copy);
        mutex_lock(&root->fs_info->trans_mutex);
 
index 598baa3..cc63650 100644 (file)
@@ -98,4 +98,5 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
                                   struct btrfs_root *root);
 void btrfs_throttle(struct btrfs_root *root);
+int btrfs_record_root_in_trans(struct btrfs_root *root);
 #endif
index cc2650b..b3bb5bb 100644 (file)
@@ -81,12 +81,12 @@ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
                memcpy(&key, &root->defrag_progress, sizeof(key));
        }
 
-       path->lowest_level = 1;
        path->keep_locks = 1;
        if (cache_only)
                min_trans = root->defrag_trans_start;
 
-       ret = btrfs_search_forward(root, &key, path, cache_only, min_trans);
+       ret = btrfs_search_forward(root, &key, NULL, path,
+                                  cache_only, min_trans);
        if (ret < 0)
                goto out;
        if (ret > 0) {
diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c
new file mode 100644 (file)
index 0000000..d1ce831
--- /dev/null
@@ -0,0 +1,2804 @@
+/*
+ * Copyright (C) 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/sched.h>
+#include "ctree.h"
+#include "transaction.h"
+#include "disk-io.h"
+#include "locking.h"
+#include "print-tree.h"
+#include "compat.h"
+
+/* magic values for the inode_only field in btrfs_log_inode:
+ *
+ * LOG_INODE_ALL means to log everything
+ * LOG_INODE_EXISTS means to log just enough to recreate the inode
+ * during log replay
+ */
+#define LOG_INODE_ALL 0
+#define LOG_INODE_EXISTS 1
+
+/*
+ * 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
+ * we find in the log are created in the subvolume.
+ *
+ * The last stage is to deal with directories and links and extents
+ * and all the other fun semantics
+ */
+#define LOG_WALK_PIN_ONLY 0
+#define LOG_WALK_REPLAY_INODES 1
+#define LOG_WALK_REPLAY_ALL 2
+
+static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
+                            struct btrfs_root *root, struct inode *inode,
+                            int inode_only);
+
+/*
+ * tree logging is a special write ahead log used to make sure that
+ * fsyncs and O_SYNCs can happen without doing full tree commits.
+ *
+ * Full tree commits are expensive because they require commonly
+ * modified blocks to be recowed, creating many dirty pages in the
+ * extent tree an 4x-6x higher write load than ext3.
+ *
+ * Instead of doing a tree commit on every fsync, we use the
+ * key ranges and transaction ids to find items for a given file or directory
+ * that have changed in this transaction.  Those items are copied into
+ * a special tree (one per subvolume root), that tree is written to disk
+ * and then the fsync is considered complete.
+ *
+ * After a crash, items are copied out of the log-tree back into the
+ * subvolume tree.  Any file data extents found are recorded in the extent
+ * allocation tree, and the log-tree freed.
+ *
+ * The log tree is read three times, once to pin down all the extents it is
+ * using in ram and once, once to create all the inodes logged in the tree
+ * and once to do all the other items.
+ */
+
+/*
+ * 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).
+ */
+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,
+                                     BTRFS_TREE_LOG_OBJECTID,
+                                     0, 0, 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->nblocks = cpu_to_le64(1);
+       inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
+
+       btrfs_set_root_bytenr(&root_item, leaf->start);
+       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
+ */
+static int start_log_trans(struct btrfs_trans_handle *trans,
+                          struct btrfs_root *root)
+{
+       int ret;
+       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 (!root->log_root) {
+               ret = btrfs_add_log_tree(trans, root);
+               BUG_ON(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;
+}
+
+/*
+ * returns 0 if there was a log transaction running and we were able
+ * to join, or returns -ENOENT if there were not transactions
+ * in progress
+ */
+static int join_running_log_trans(struct btrfs_root *root)
+{
+       int ret = -ENOENT;
+
+       smp_mb();
+       if (!root->log_root)
+               return -ENOENT;
+
+       mutex_lock(&root->fs_info->tree_log_mutex);
+       if (root->log_root) {
+               ret = 0;
+               atomic_inc(&root->fs_info->tree_log_writers);
+               root->fs_info->tree_log_batch++;
+       }
+       mutex_unlock(&root->fs_info->tree_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)
+{
+       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);
+       return 0;
+}
+
+
+/*
+ * the walk control struct is used to pass state down the chain when
+ * processing the log tree.  The stage field tells us which part
+ * of the log tree processing we are currently doing.  The others
+ * are state fields used for that specific part
+ */
+struct walk_control {
+       /* should we free the extent on disk when done?  This is used
+        * at transaction commit time while freeing a log tree
+        */
+       int free;
+
+       /* should we write out the extent buffer?  This is used
+        * while flushing the log tree to disk during a sync
+        */
+       int write;
+
+       /* should we wait for the extent buffer io to finish?  Also used
+        * while flushing the log tree to disk for a sync
+        */
+       int wait;
+
+       /* pin only walk, we record which extents on disk belong to the
+        * log trees
+        */
+       int pin;
+
+       /* what stage of the replay code we're currently in */
+       int stage;
+
+       /* the root we are currently replaying */
+       struct btrfs_root *replay_dest;
+
+       /* the trans handle for the current replay */
+       struct btrfs_trans_handle *trans;
+
+       /* the function that gets used to process blocks we find in the
+        * tree.  Note the extent_buffer might not be up to date when it is
+        * passed in, and it must be checked or read if you need the data
+        * inside it
+        */
+       int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
+                           struct walk_control *wc, u64 gen);
+};
+
+/*
+ * process_func used to pin down extents, write them or wait on them
+ */
+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->alloc_mutex);
+               btrfs_update_pinned_extents(log->fs_info->extent_root,
+                                           eb->start, eb->len, 1);
+               mutex_unlock(&log->fs_info->alloc_mutex);
+       }
+
+       if (btrfs_buffer_uptodate(eb, gen)) {
+               if (wc->write)
+                       btrfs_write_tree_block(eb);
+               if (wc->wait)
+                       btrfs_wait_tree_block_writeback(eb);
+       }
+       return 0;
+}
+
+/*
+ * Item overwrite used by replay and tree logging.  eb, slot and key all refer
+ * to the src data we are copying out.
+ *
+ * root is the tree we are copying into, and path is a scratch
+ * path for use in this function (it should be released on entry and
+ * will be released on exit).
+ *
+ * If the key is already in the destination tree the existing item is
+ * overwritten.  If the existing item isn't big enough, it is extended.
+ * If it is too large, it is truncated.
+ *
+ * If the key isn't in the destination yet, a new item is inserted.
+ */
+static noinline int overwrite_item(struct btrfs_trans_handle *trans,
+                                  struct btrfs_root *root,
+                                  struct btrfs_path *path,
+                                  struct extent_buffer *eb, int slot,
+                                  struct btrfs_key *key)
+{
+       int ret;
+       u32 item_size;
+       u64 saved_i_size = 0;
+       int save_old_i_size = 0;
+       unsigned long src_ptr;
+       unsigned long dst_ptr;
+       int overwrite_root = 0;
+
+       if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
+               overwrite_root = 1;
+
+       item_size = btrfs_item_size_nr(eb, slot);
+       src_ptr = btrfs_item_ptr_offset(eb, slot);
+
+       /* look for the key in the destination tree */
+       ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
+       if (ret == 0) {
+               char *src_copy;
+               char *dst_copy;
+               u32 dst_size = btrfs_item_size_nr(path->nodes[0],
+                                                 path->slots[0]);
+               if (dst_size != item_size)
+                       goto insert;
+
+               if (item_size == 0) {
+                       btrfs_release_path(root, path);
+                       return 0;
+               }
+               dst_copy = kmalloc(item_size, GFP_NOFS);
+               src_copy = kmalloc(item_size, GFP_NOFS);
+
+               read_extent_buffer(eb, src_copy, src_ptr, item_size);
+
+               dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
+               read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
+                                  item_size);
+               ret = memcmp(dst_copy, src_copy, item_size);
+
+               kfree(dst_copy);
+               kfree(src_copy);
+               /*
+                * they have the same contents, just return, this saves
+                * us from cowing blocks in the destination tree and doing
+                * extra writes that may not have been done by a previous
+                * sync
+                */
+               if (ret == 0) {
+                       btrfs_release_path(root, path);
+                       return 0;
+               }
+
+       }
+insert:
+       btrfs_release_path(root, path);
+       /* try to insert the key into the destination tree */
+       ret = btrfs_insert_empty_item(trans, root, path,
+                                     key, item_size);
+
+       /* make sure any existing item is the correct size */
+       if (ret == -EEXIST) {
+               u32 found_size;
+               found_size = btrfs_item_size_nr(path->nodes[0],
+                                               path->slots[0]);
+               if (found_size > item_size) {
+                       btrfs_truncate_item(trans, root, path, item_size, 1);
+               } else if (found_size < item_size) {
+                       ret = btrfs_del_item(trans, root,
+                                            path);
+                       BUG_ON(ret);
+
+                       btrfs_release_path(root, path);
+                       ret = btrfs_insert_empty_item(trans,
+                                 root, path, key, item_size);
+                       BUG_ON(ret);
+               }
+       } else if (ret) {
+               BUG();
+       }
+       dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
+                                       path->slots[0]);
+
+       /* don't overwrite an existing inode if the generation number
+        * was logged as zero.  This is done when the tree logging code
+        * is just logging an inode to make sure it exists after recovery.
+        *
+        * Also, don't overwrite i_size on directories during replay.
+        * log replay inserts and removes directory items based on the
+        * state of the tree found in the subvolume, and i_size is modified
+        * as it goes
+        */
+       if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
+               struct btrfs_inode_item *src_item;
+               struct btrfs_inode_item *dst_item;
+
+               src_item = (struct btrfs_inode_item *)src_ptr;
+               dst_item = (struct btrfs_inode_item *)dst_ptr;
+
+               if (btrfs_inode_generation(eb, src_item) == 0)
+                       goto no_copy;
+
+               if (overwrite_root &&
+                   S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
+                   S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
+                       save_old_i_size = 1;
+                       saved_i_size = btrfs_inode_size(path->nodes[0],
+                                                       dst_item);
+               }
+       }
+
+       copy_extent_buffer(path->nodes[0], eb, dst_ptr,
+                          src_ptr, item_size);
+
+       if (save_old_i_size) {
+               struct btrfs_inode_item *dst_item;
+               dst_item = (struct btrfs_inode_item *)dst_ptr;
+               btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
+       }
+
+       /* make sure the generation is filled in */
+       if (key->type == BTRFS_INODE_ITEM_KEY) {
+               struct btrfs_inode_item *dst_item;
+               dst_item = (struct btrfs_inode_item *)dst_ptr;
+               if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
+                       btrfs_set_inode_generation(path->nodes[0], dst_item,
+                                                  trans->transid);
+               }
+       }
+no_copy:
+       btrfs_mark_buffer_dirty(path->nodes[0]);
+       btrfs_release_path(root, path);
+       return 0;
+}
+
+/*
+ * simple helper to read an inode off the disk from a given root
+ * This can only be called for subvolume roots and not for the log
+ */
+static noinline struct inode *read_one_inode(struct btrfs_root *root,
+                                            u64 objectid)
+{
+       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)) {
+               iput(inode);
+               inode = NULL;
+       }
+       return inode;
+}
+
+/* replays a single extent in 'eb' at 'slot' with 'key' into the
+ * subvolume 'root'.  path is released on entry and should be released
+ * on exit.
+ *
+ * extents in the log tree have not been allocated out of the extent
+ * tree yet.  So, this completes the allocation, taking a reference
+ * as required if the extent already exists or creating a new extent
+ * if it isn't in the extent allocation tree yet.
+ *
+ * The extent is inserted into the file, dropping any existing extents
+ * from the file that overlap the new one.
+ */
+static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
+                                     struct btrfs_root *root,
+                                     struct btrfs_path *path,
+                                     struct extent_buffer *eb, int slot,
+                                     struct btrfs_key *key)
+{
+       int found_type;
+       u64 mask = root->sectorsize - 1;
+       u64 extent_end;
+       u64 alloc_hint;
+       u64 start = key->offset;
+       struct btrfs_file_extent_item *item;
+       struct inode *inode = NULL;
+       unsigned long size;
+       int ret = 0;
+
+       item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
+       found_type = btrfs_file_extent_type(eb, item);
+
+       if (found_type == BTRFS_FILE_EXTENT_REG)
+               extent_end = start + btrfs_file_extent_num_bytes(eb, item);
+       else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
+               size = btrfs_file_extent_inline_len(eb,
+                                                   btrfs_item_nr(eb, slot));
+               extent_end = (start + size + mask) & ~mask;
+       } else {
+               ret = 0;
+               goto out;
+       }
+
+       inode = read_one_inode(root, key->objectid);
+       if (!inode) {
+               ret = -EIO;
+               goto out;
+       }
+
+       /*
+        * first check to see if we already have this extent in the
+        * 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,
+                                      start, 0);
+
+       if (ret == 0 && found_type == BTRFS_FILE_EXTENT_REG) {
+               struct btrfs_file_extent_item cmp1;
+               struct btrfs_file_extent_item cmp2;
+               struct btrfs_file_extent_item *existing;
+               struct extent_buffer *leaf;
+
+               leaf = path->nodes[0];
+               existing = btrfs_item_ptr(leaf, path->slots[0],
+                                         struct btrfs_file_extent_item);
+
+               read_extent_buffer(eb, &cmp1, (unsigned long)item,
+                                  sizeof(cmp1));
+               read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
+                                  sizeof(cmp2));
+
+               /*
+                * we already have a pointer to this exact extent,
+                * we don't have to do anything
+                */
+               if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
+                       btrfs_release_path(root, path);
+                       goto out;
+               }
+       }
+       btrfs_release_path(root, path);
+
+       /* drop any overlapping extents */
+       ret = btrfs_drop_extents(trans, root, inode,
+                        start, extent_end, start, &alloc_hint);
+       BUG_ON(ret);
+
+       BUG_ON(ret);
+       if (found_type == BTRFS_FILE_EXTENT_REG) {
+               struct btrfs_key ins;
+
+               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;
+
+               /* insert the extent pointer in the file */
+               ret = overwrite_item(trans, root, path, eb, slot, key);
+               BUG_ON(ret);
+
+               /*
+                * is this extent already allocated in the extent
+                * allocation tree?  If so, just add a reference
+                */
+               ret = btrfs_lookup_extent(root, path, ins.objectid, ins.offset);
+               btrfs_release_path(root, path);
+               if (ret == 0) {
+                       ret = btrfs_inc_extent_ref(trans, root,
+                                  ins.objectid, ins.offset,
+                                  root->root_key.objectid,
+                                  trans->transid, key->objectid, start);
+               } else {
+                       /*
+                        * insert the extent pointer in the extent
+                        * allocation tree
+                        */
+                       ret = btrfs_alloc_logged_extent(trans, root,
+                                               root->root_key.objectid,
+                                               trans->transid, key->objectid,
+                                               start, &ins);
+                       BUG_ON(ret);
+               }
+       } 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_blocks, update it here */
+       inode->i_blocks += (extent_end - start) >> 9;
+       btrfs_update_inode(trans, root, inode);
+out:
+       if (inode)
+               iput(inode);
+       return ret;
+}
+
+/*
+ * when cleaning up conflicts between the directory names in the
+ * subvolume, directory names in the log and directory names in the
+ * inode back references, we may have to unlink inodes from directories.
+ *
+ * This is a helper function to do the unlink of a specific directory
+ * item
+ */
+static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
+                                     struct btrfs_root *root,
+                                     struct btrfs_path *path,
+                                     struct inode *dir,
+                                     struct btrfs_dir_item *di)
+{
+       struct inode *inode;
+       char *name;
+       int name_len;
+       struct extent_buffer *leaf;
+       struct btrfs_key location;
+       int ret;
+
+       leaf = path->nodes[0];
+
+       btrfs_dir_item_key_to_cpu(leaf, di, &location);
+       name_len = btrfs_dir_name_len(leaf, di);
+       name = kmalloc(name_len, GFP_NOFS);
+       read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
+       btrfs_release_path(root, path);
+
+       inode = read_one_inode(root, location.objectid);
+       BUG_ON(!inode);
+
+       btrfs_inc_nlink(inode);
+       ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
+       kfree(name);
+
+       iput(inode);
+       return ret;
+}
+
+/*
+ * helper function to see if a given name and sequence number found
+ * in an inode back reference are already in a directory and correctly
+ * point to this inode
+ */
+static noinline int inode_in_dir(struct btrfs_root *root,
+                                struct btrfs_path *path,
+                                u64 dirid, u64 objectid, u64 index,
+                                const char *name, int name_len)
+{
+       struct btrfs_dir_item *di;
+       struct btrfs_key location;
+       int match = 0;
+
+       di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
+                                        index, name, name_len, 0);
+       if (di && !IS_ERR(di)) {
+               btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
+               if (location.objectid != objectid)
+                       goto out;
+       } else
+               goto out;
+       btrfs_release_path(root, path);
+
+       di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
+       if (di && !IS_ERR(di)) {
+               btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
+               if (location.objectid != objectid)
+                       goto out;
+       } else
+               goto out;
+       match = 1;
+out:
+       btrfs_release_path(root, path);
+       return match;
+}
+
+/*
+ * helper function to check a log tree for a named back reference in
+ * an inode.  This is used to decide if a back reference that is
+ * found in the subvolume conflicts with what we find in the log.
+ *
+ * inode backreferences may have multiple refs in a single item,
+ * during replay we process one reference at a time, and we don't
+ * want to delete valid links to a file from the subvolume if that
+ * link is also in the log.
+ */
+static noinline int backref_in_log(struct btrfs_root *log,
+                                  struct btrfs_key *key,
+                                  char *name, int namelen)
+{
+       struct btrfs_path *path;
+       struct btrfs_inode_ref *ref;
+       unsigned long ptr;
+       unsigned long ptr_end;
+       unsigned long name_ptr;
+       int found_name_len;
+       int item_size;
+       int ret;
+       int match = 0;
+
+       path = btrfs_alloc_path();
+       ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
+       if (ret != 0)
+               goto out;
+
+       item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
+       ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
+       ptr_end = ptr + item_size;
+       while (ptr < ptr_end) {
+               ref = (struct btrfs_inode_ref *)ptr;
+               found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
+               if (found_name_len == namelen) {
+                       name_ptr = (unsigned long)(ref + 1);
+                       ret = memcmp_extent_buffer(path->nodes[0], name,
+                                                  name_ptr, namelen);
+                       if (ret == 0) {
+                               match = 1;
+                               goto out;
+                       }
+               }
+               ptr = (unsigned long)(ref + 1) + found_name_len;
+       }
+out:
+       btrfs_free_path(path);
+       return match;
+}
+
+
+/*
+ * replay one inode back reference item found in the log tree.
+ * eb, slot and key refer to the buffer and key found in the log tree.
+ * root is the destination we are replaying into, and path is for temp
+ * use by this function.  (it should be released on return).
+ */
+static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
+                                 struct btrfs_root *root,
+                                 struct btrfs_root *log,
+                                 struct btrfs_path *path,
+                                 struct extent_buffer *eb, int slot,
+                                 struct btrfs_key *key)
+{
+       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;
+
+       /*
+        * it is possible that we didn't log all the parent directories
+        * for a given inode.  If we don't find the dir, just don't
+        * copy the back ref in.  The link count fixup code will take
+        * care of the rest
+        */
+       dir = read_one_inode(root, key->offset);
+       if (!dir)
+               return -ENOENT;
+
+       inode = read_one_inode(root, key->objectid);
+       BUG_ON(!dir);
+
+       ref_ptr = btrfs_item_ptr_offset(eb, slot);
+       ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
+
+again:
+       ref = (struct btrfs_inode_ref *)ref_ptr;
+
+       namelen = btrfs_inode_ref_name_len(eb, ref);
+       name = kmalloc(namelen, GFP_NOFS);
+       BUG_ON(!name);
+
+       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,
+                        btrfs_inode_ref_index(eb, ref),
+                        name, namelen)) {
+               goto out;
+       }
+
+       /*
+        * look for a conflicting back reference in the metadata.
+        * if we find one we have to unlink that name of the file
+        * before we add our new link.  Later on, we overwrite any
+        * existing back reference, and we don't want to create
+        * dangling pointers in the directory.
+        */
+conflict_again:
+       ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
+       if (ret == 0) {
+               char *victim_name;
+               int victim_name_len;
+               struct btrfs_inode_ref *victim_ref;
+               unsigned long ptr;
+               unsigned long ptr_end;
+               struct extent_buffer *leaf = path->nodes[0];
+
+               /* are we trying to overwrite a back ref for the root directory
+                * if so, just jump out, we're done
+                */
+               if (key->objectid == key->offset)
+                       goto out_nowrite;
+
+               /* check all the names in this back reference to see
+                * if they are in the log.  if so, we allow them to stay
+                * otherwise they must be unlinked as a conflict
+                */
+               ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
+               ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
+               while(ptr < ptr_end) {
+                       victim_ref = (struct btrfs_inode_ref *)ptr;
+                       victim_name_len = btrfs_inode_ref_name_len(leaf,
+                                                                  victim_ref);
+                       victim_name = kmalloc(victim_name_len, GFP_NOFS);
+                       BUG_ON(!victim_name);
+
+                       read_extent_buffer(leaf, victim_name,
+                                          (unsigned long)(victim_ref + 1),
+                                          victim_name_len);
+
+                       if (!backref_in_log(log, key, victim_name,
+                                           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);
+                               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);
+       }
+       btrfs_release_path(root, path);
+
+       /* insert our name */
+       ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
+                            btrfs_inode_ref_index(eb, ref));
+       BUG_ON(ret);
+
+       btrfs_update_inode(trans, root, inode);
+
+out:
+       ref_ptr = (unsigned long)(ref + 1) + namelen;
+       kfree(name);
+       if (ref_ptr < ref_end)
+               goto again;
+
+       /* finally write the back reference in the inode */
+       ret = overwrite_item(trans, root, path, eb, slot, key);
+       BUG_ON(ret);
+
+out_nowrite:
+       btrfs_release_path(root, 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)
+{
+       int ret;
+       u32 item_size = btrfs_item_size_nr(eb, slot);
+       u64 cur_offset;
+       unsigned long file_bytes;
+       struct btrfs_ordered_sum *sums;
+       struct btrfs_sector_sum *sector_sum;
+       struct inode *inode;
+       unsigned long ptr;
+
+       file_bytes = (item_size / BTRFS_CRC32_SIZE) * root->sectorsize;
+       inode = read_one_inode(root, key->objectid);
+       if (!inode) {
+               return -EIO;
+       }
+
+       sums = kzalloc(btrfs_ordered_sum_size(root, file_bytes), GFP_NOFS);
+       if (!sums) {
+               iput(inode);
+               return -ENOMEM;
+       }
+
+       INIT_LIST_HEAD(&sums->list);
+       sums->len = file_bytes;
+       sums->file_offset = key->offset;
+
+       /*
+        * 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->offset = cur_offset;
+               read_extent_buffer(eb, &sector_sum->sum, ptr, BTRFS_CRC32_SIZE);
+               sector_sum++;
+               item_size -= BTRFS_CRC32_SIZE;
+               ptr += BTRFS_CRC32_SIZE;
+               cur_offset += root->sectorsize;
+       }
+
+       /* let btrfs_csum_file_blocks add them into the file */
+       ret = btrfs_csum_file_blocks(trans, root, inode, sums);
+       BUG_ON(ret);
+       kfree(sums);
+       iput(inode);
+
+       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
+ * lots of complexity to the log code, we just scan the backrefs
+ * for any file that has been through replay.
+ *
+ * The scan will update the link count on the inode to reflect the
+ * number of back refs found.  If it goes down to zero, the iput
+ * will free the inode.
+ */
+static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
+                                          struct btrfs_root *root,
+                                          struct inode *inode)
+{
+       struct btrfs_path *path;
+       int ret;
+       struct btrfs_key key;
+       u64 nlink = 0;
+       unsigned long ptr;
+       unsigned long ptr_end;
+       int name_len;
+
+       key.objectid = inode->i_ino;
+       key.type = BTRFS_INODE_REF_KEY;
+       key.offset = (u64)-1;
+
+       path = btrfs_alloc_path();
+
+       while(1) {
+               ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+               if (ret < 0)
+                       break;
+               if (ret > 0) {
+                       if (path->slots[0] == 0)
+                               break;
+                       path->slots[0]--;
+               }
+               btrfs_item_key_to_cpu(path->nodes[0], &key,
+                                     path->slots[0]);
+               if (key.objectid != inode->i_ino ||
+                   key.type != BTRFS_INODE_REF_KEY)
+                       break;
+               ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
+               ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
+                                                  path->slots[0]);
+               while(ptr < ptr_end) {
+                       struct btrfs_inode_ref *ref;
+
+                       ref = (struct btrfs_inode_ref *)ptr;
+                       name_len = btrfs_inode_ref_name_len(path->nodes[0],
+                                                           ref);
+                       ptr = (unsigned long)(ref + 1) + name_len;
+                       nlink++;
+               }
+
+               if (key.offset == 0)
+                       break;
+               key.offset--;
+               btrfs_release_path(root, path);
+       }
+       btrfs_free_path(path);
+       if (nlink != inode->i_nlink) {
+               inode->i_nlink = nlink;
+               btrfs_update_inode(trans, root, inode);
+       }
+
+       return 0;
+}
+
+static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
+                                           struct btrfs_root *root,
+                                           struct btrfs_path *path)
+{
+       int ret;
+       struct btrfs_key key;
+       struct inode *inode;
+
+       key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
+       key.type = BTRFS_ORPHAN_ITEM_KEY;
+       key.offset = (u64)-1;
+       while(1) {
+               ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+               if (ret < 0)
+                       break;
+
+               if (ret == 1) {
+                       if (path->slots[0] == 0)
+                               break;
+                       path->slots[0]--;
+               }
+
+               btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+               if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
+                   key.type != BTRFS_ORPHAN_ITEM_KEY)
+                       break;
+
+               ret = btrfs_del_item(trans, root, path);
+               BUG_ON(ret);
+
+               btrfs_release_path(root, path);
+               inode = read_one_inode(root, key.offset);
+               BUG_ON(!inode);
+
+               ret = fixup_inode_link_count(trans, root, inode);
+               BUG_ON(ret);
+
+               iput(inode);
+
+               if (key.offset == 0)
+                       break;
+               key.offset--;
+       }
+       btrfs_release_path(root, path);
+       return 0;
+}
+
+
+/*
+ * record a given inode in the fixup dir so we can check its link
+ * count when replay is done.  The link count is incremented here
+ * so the inode won't go away until we check it
+ */
+static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
+                                     struct btrfs_root *root,
+                                     struct btrfs_path *path,
+                                     u64 objectid)
+{
+       struct btrfs_key key;
+       int ret = 0;
+       struct inode *inode;
+
+       inode = read_one_inode(root, objectid);
+       BUG_ON(!inode);
+
+       key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
+       btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
+       key.offset = objectid;
+
+       ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
+
+       btrfs_release_path(root, path);
+       if (ret == 0) {
+               btrfs_inc_nlink(inode);
+               btrfs_update_inode(trans, root, inode);
+       } else if (ret == -EEXIST) {
+               ret = 0;
+       } else {
+               BUG();
+       }
+       iput(inode);
+
+       return ret;
+}
+
+/*
+ * when replaying the log for a directory, we only insert names
+ * for inodes that actually exist.  This means an fsync on a directory
+ * does not implicitly fsync all the new files in it
+ */
+static noinline int insert_one_name(struct btrfs_trans_handle *trans,
+                                   struct btrfs_root *root,
+                                   struct btrfs_path *path,
+                                   u64 dirid, u64 index,
+                                   char *name, int name_len, u8 type,
+                                   struct btrfs_key *location)
+{
+       struct inode *inode;
+       struct inode *dir;
+       int ret;
+
+       inode = read_one_inode(root, location->objectid);
+       if (!inode)
+               return -ENOENT;
+
+       dir = read_one_inode(root, dirid);
+       if (!dir) {
+               iput(inode);
+               return -EIO;
+       }
+       ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
+
+       /* FIXME, put inode into FIXUP list */
+
+       iput(inode);
+       iput(dir);
+       return ret;
+}
+
+/*
+ * take a single entry in a log directory item and replay it into
+ * the subvolume.
+ *
+ * if a conflicting item exists in the subdirectory already,
+ * the inode it points to is unlinked and put into the link count
+ * fix up tree.
+ *
+ * If a name from the log points to a file or directory that does
+ * not exist in the FS, it is skipped.  fsyncs on directories
+ * do not force down inodes inside that directory, just changes to the
+ * names or unlinks in a directory.
+ */
+static noinline int replay_one_name(struct btrfs_trans_handle *trans,
+                                   struct btrfs_root *root,
+                                   struct btrfs_path *path,
+                                   struct extent_buffer *eb,
+                                   struct btrfs_dir_item *di,
+                                   struct btrfs_key *key)
+{
+       char *name;
+       int name_len;
+       struct btrfs_dir_item *dst_di;
+       struct btrfs_key found_key;
+       struct btrfs_key log_key;
+       struct inode *dir;
+       struct inode *inode;
+       u8 log_type;
+       int ret;
+
+       dir = read_one_inode(root, key->objectid);
+       BUG_ON(!dir);
+
+       name_len = btrfs_dir_name_len(eb, di);
+       name = kmalloc(name_len, GFP_NOFS);
+       log_type = btrfs_dir_type(eb, di);
+       read_extent_buffer(eb, name, (unsigned long)(di + 1),
+                  name_len);
+
+       btrfs_dir_item_key_to_cpu(eb, di, &log_key);
+       if (key->type == BTRFS_DIR_ITEM_KEY) {
+               dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
+                                      name, name_len, 1);
+       }
+       else if (key->type == BTRFS_DIR_INDEX_KEY) {
+               dst_di = btrfs_lookup_dir_index_item(trans, root, path,
+                                                    key->objectid,
+                                                    key->offset, name,
+                                                    name_len, 1);
+       } else {
+               BUG();
+       }
+       if (!dst_di || IS_ERR(dst_di)) {
+               /* we need a sequence number to insert, so we only
+                * do inserts for the BTRFS_DIR_INDEX_KEY types
+                */
+               if (key->type != BTRFS_DIR_INDEX_KEY)
+                       goto out;
+               goto insert;
+       }
+
+       btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
+       /* the existing item matches the logged item */
+       if (found_key.objectid == log_key.objectid &&
+           found_key.type == log_key.type &&
+           found_key.offset == log_key.offset &&
+           btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
+               goto out;
+       }
+
+       /*
+        * don't drop the conflicting directory entry if the inode
+        * for the new entry doesn't exist
+        */
+       inode = read_one_inode(root, log_key.objectid);
+       if (!inode)
+               goto out;
+
+       iput(inode);
+       ret = drop_one_dir_item(trans, root, path, dir, dst_di);
+       BUG_ON(ret);
+
+       if (key->type == BTRFS_DIR_INDEX_KEY)
+               goto insert;
+out:
+       btrfs_release_path(root, path);
+       kfree(name);
+       iput(dir);
+       return 0;
+
+insert:
+       btrfs_release_path(root, path);
+       ret = insert_one_name(trans, root, path, key->objectid, key->offset,
+                             name, name_len, log_type, &log_key);
+
+       if (ret && ret != -ENOENT)
+               BUG();
+       goto out;
+}
+
+/*
+ * find all the names in a directory item and reconcile them into
+ * the subvolume.  Only BTRFS_DIR_ITEM_KEY types will have more than
+ * one name in a directory item, but the same code gets used for
+ * both directory index types
+ */
+static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
+                                       struct btrfs_root *root,
+                                       struct btrfs_path *path,
+                                       struct extent_buffer *eb, int slot,
+                                       struct btrfs_key *key)
+{
+       int ret;
+       u32 item_size = btrfs_item_size_nr(eb, slot);
+       struct btrfs_dir_item *di;
+       int name_len;
+       unsigned long ptr;
+       unsigned long ptr_end;
+
+       ptr = btrfs_item_ptr_offset(eb, slot);
+       ptr_end = ptr + item_size;
+       while(ptr < ptr_end) {
+               di = (struct btrfs_dir_item *)ptr;
+               name_len = btrfs_dir_name_len(eb, di);
+               ret = replay_one_name(trans, root, path, eb, di, key);
+               BUG_ON(ret);
+               ptr = (unsigned long)(di + 1);
+               ptr += name_len;
+       }
+       return 0;
+}
+
+/*
+ * directory replay has two parts.  There are the standard directory
+ * items in the log copied from the subvolume, and range items
+ * created in the log while the subvolume was logged.
+ *
+ * The range items tell us which parts of the key space the log
+ * is authoritative for.  During replay, if a key in the subvolume
+ * directory is in a logged range item, but not actually in the log
+ * that means it was deleted from the directory before the fsync
+ * and should be removed.
+ */
+static noinline int find_dir_range(struct btrfs_root *root,
+                                  struct btrfs_path *path,
+                                  u64 dirid, int key_type,
+                                  u64 *start_ret, u64 *end_ret)
+{
+       struct btrfs_key key;
+       u64 found_end;
+       struct btrfs_dir_log_item *item;
+       int ret;
+       int nritems;
+
+       if (*start_ret == (u64)-1)
+               return 1;
+
+       key.objectid = dirid;
+       key.type = key_type;
+       key.offset = *start_ret;
+
+       ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+       if (ret < 0)
+               goto out;
+       if (ret > 0) {
+               if (path->slots[0] == 0)
+                       goto out;
+               path->slots[0]--;
+       }
+       if (ret != 0)
+               btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+       if (key.type != key_type || key.objectid != dirid) {
+               ret = 1;
+               goto next;
+       }
+       item = btrfs_item_ptr(path->nodes[0], path->slots[0],
+                             struct btrfs_dir_log_item);
+       found_end = btrfs_dir_log_end(path->nodes[0], item);
+
+       if (*start_ret >= key.offset && *start_ret <= found_end) {
+               ret = 0;
+               *start_ret = key.offset;
+               *end_ret = found_end;
+               goto out;
+       }
+       ret = 1;
+next:
+       /* check the next slot in the tree to see if it is a valid item */
+       nritems = btrfs_header_nritems(path->nodes[0]);
+       if (path->slots[0] >= nritems) {
+               ret = btrfs_next_leaf(root, path);
+               if (ret)
+                       goto out;
+       } else {
+               path->slots[0]++;
+       }
+
+       btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+       if (key.type != key_type || key.objectid != dirid) {
+               ret = 1;
+               goto out;
+       }
+       item = btrfs_item_ptr(path->nodes[0], path->slots[0],
+                             struct btrfs_dir_log_item);
+       found_end = btrfs_dir_log_end(path->nodes[0], item);
+       *start_ret = key.offset;
+       *end_ret = found_end;
+       ret = 0;
+out:
+       btrfs_release_path(root, path);
+       return ret;
+}
+
+/*
+ * this looks for a given directory item in the log.  If the directory
+ * item is not in the log, the item is removed and the inode it points
+ * to is unlinked
+ */
+static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
+                                     struct btrfs_root *root,
+                                     struct btrfs_root *log,
+                                     struct btrfs_path *path,
+                                     struct btrfs_path *log_path,
+                                     struct inode *dir,
+                                     struct btrfs_key *dir_key)
+{
+       int ret;
+       struct extent_buffer *eb;
+       int slot;
+       u32 item_size;
+       struct btrfs_dir_item *di;
+       struct btrfs_dir_item *log_di;
+       int name_len;
+       unsigned long ptr;
+       unsigned long ptr_end;
+       char *name;
+       struct inode *inode;
+       struct btrfs_key location;
+
+again:
+       eb = path->nodes[0];
+       slot = path->slots[0];
+       item_size = btrfs_item_size_nr(eb, slot);
+       ptr = btrfs_item_ptr_offset(eb, slot);
+       ptr_end = ptr + item_size;
+       while(ptr < ptr_end) {
+               di = (struct btrfs_dir_item *)ptr;
+               name_len = btrfs_dir_name_len(eb, di);
+               name = kmalloc(name_len, GFP_NOFS);
+               if (!name) {
+                       ret = -ENOMEM;
+                       goto out;
+               }
+               read_extent_buffer(eb, name, (unsigned long)(di + 1),
+                                 name_len);
+               log_di = NULL;
+               if (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) {
+                       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)) {
+                       btrfs_dir_item_key_to_cpu(eb, di, &location);
+                       btrfs_release_path(root, path);
+                       btrfs_release_path(log, log_path);
+                       inode = read_one_inode(root, location.objectid);
+                       BUG_ON(!inode);
+
+                       ret = link_to_fixup_dir(trans, root,
+                                               path, location.objectid);
+                       BUG_ON(ret);
+                       btrfs_inc_nlink(inode);
+                       ret = btrfs_unlink_inode(trans, root, dir, inode,
+                                                name, name_len);
+                       BUG_ON(ret);
+                       kfree(name);
+                       iput(inode);
+
+                       /* there might still be more names under this key
+                        * check and repeat if required
+                        */
+                       ret = btrfs_search_slot(NULL, root, dir_key, path,
+                                               0, 0);
+                       if (ret == 0)
+                               goto again;
+                       ret = 0;
+                       goto out;
+               }
+               btrfs_release_path(log, log_path);
+               kfree(name);
+
+               ptr = (unsigned long)(di + 1);
+               ptr += name_len;
+       }
+       ret = 0;
+out:
+       btrfs_release_path(root, path);
+       btrfs_release_path(log, log_path);
+       return ret;
+}
+
+/*
+ * deletion replay happens before we copy any new directory items
+ * out of the log or out of backreferences from inodes.  It
+ * scans the log to find ranges of keys that log is authoritative for,
+ * and then scans the directory to find items in those ranges that are
+ * not present in the log.
+ *
+ * Anything we don't find in the log is unlinked and removed from the
+ * directory.
+ */
+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 range_start;
+       u64 range_end;
+       int key_type = BTRFS_DIR_LOG_ITEM_KEY;
+       int ret = 0;
+       struct btrfs_key dir_key;
+       struct btrfs_key found_key;
+       struct btrfs_path *log_path;
+       struct inode *dir;
+
+       dir_key.objectid = dirid;
+       dir_key.type = BTRFS_DIR_ITEM_KEY;
+       log_path = btrfs_alloc_path();
+       if (!log_path)
+               return -ENOMEM;
+
+       dir = read_one_inode(root, dirid);
+       /* it isn't an error if the inode isn't there, that can happen
+        * because we replay the deletes before we copy in the inode item
+        * from the log
+        */
+       if (!dir) {
+               btrfs_free_path(log_path);
+               return 0;
+       }
+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;
+
+               dir_key.offset = range_start;
+               while(1) {
+                       int nritems;
+                       ret = btrfs_search_slot(NULL, root, &dir_key, path,
+                                               0, 0);
+                       if (ret < 0)
+                               goto out;
+
+                       nritems = btrfs_header_nritems(path->nodes[0]);
+                       if (path->slots[0] >= nritems) {
+                               ret = btrfs_next_leaf(root, path);
+                               if (ret)
+                                       break;
+                       }
+                       btrfs_item_key_to_cpu(path->nodes[0], &found_key,
+                                             path->slots[0]);
+                       if (found_key.objectid != dirid ||
+                           found_key.type != dir_key.type)
+                               goto next_type;
+
+                       if (found_key.offset > range_end)
+                               break;
+
+                       ret = check_item_in_log(trans, root, log, path,
+                                               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);
+               if (range_end == (u64)-1)
+                       break;
+               range_start = range_end + 1;
+       }
+
+next_type:
+       ret = 0;
+       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);
+               goto again;
+       }
+out:
+       btrfs_release_path(root, path);
+       btrfs_free_path(log_path);
+       iput(dir);
+       return ret;
+}
+
+/*
+ * the process_func used to replay items from the log tree.  This
+ * gets called in two different stages.  The first stage just looks
+ * for inodes and makes sure they are all copied into the subvolume.
+ *
+ * The second stage copies all the other item types from the log into
+ * the subvolume.  The two stage approach is slower, but gets rid of
+ * lots of complexity around inodes referencing other inodes that exist
+ * only in the log (references come from either directory items or inode
+ * back refs).
+ */
+static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
+                            struct walk_control *wc, u64 gen)
+{
+       int nritems;
+       struct btrfs_path *path;
+       struct btrfs_root *root = wc->replay_dest;
+       struct btrfs_key key;
+       u32 item_size;
+       int level;
+       int i;
+       int ret;
+
+       btrfs_read_buffer(eb, gen);
+
+       level = btrfs_header_level(eb);
+
+       if (level != 0)
+               return 0;
+
+       path = btrfs_alloc_path();
+       BUG_ON(!path);
+
+       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;
+
+                       inode_item = btrfs_item_ptr(eb, i,
+                                           struct btrfs_inode_item);
+                       mode = btrfs_inode_mode(eb, inode_item);
+                       if (S_ISDIR(mode)) {
+                               ret = replay_dir_deletes(wc->trans,
+                                        root, log, path, key.objectid);
+                               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
+                        */
+                       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);
+                               BUG_ON(ret);
+                               iput(inode);
+                       }
+                       ret = link_to_fixup_dir(wc->trans, root,
+                                               path, key.objectid);
+                       BUG_ON(ret);
+               }
+               if (wc->stage < LOG_WALK_REPLAY_ALL)
+                       continue;
+
+               /* these keys are simply copied */
+               if (key.type == BTRFS_XATTR_ITEM_KEY) {
+                       ret = overwrite_item(wc->trans, root, path,
+                                            eb, i, &key);
+                       BUG_ON(ret);
+               } else if (key.type == BTRFS_INODE_REF_KEY) {
+                       ret = add_inode_ref(wc->trans, root, log, path,
+                                           eb, i, &key);
+                       BUG_ON(ret && ret != -ENOENT);
+               } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
+                       ret = replay_one_extent(wc->trans, root, path,
+                                               eb, i, &key);
+                       BUG_ON(ret);
+               } else if (key.type == BTRFS_CSUM_ITEM_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,
+                                                 eb, i, &key);
+                       BUG_ON(ret);
+               }
+       }
+       btrfs_free_path(path);
+       return 0;
+}
+
+static int noinline walk_down_log_tree(struct btrfs_trans_handle *trans,
+                                  struct btrfs_root *root,
+                                  struct btrfs_path *path, int *level,
+                                  struct walk_control *wc)
+{
+       u64 root_owner;
+       u64 root_gen;
+       u64 bytenr;
+       u64 ptr_gen;
+       struct extent_buffer *next;
+       struct extent_buffer *cur;
+       struct extent_buffer *parent;
+       u32 blocksize;
+       int ret = 0;
+
+       WARN_ON(*level < 0);
+       WARN_ON(*level >= BTRFS_MAX_LEVEL);
+
+       while(*level > 0) {
+               WARN_ON(*level < 0);
+               WARN_ON(*level >= BTRFS_MAX_LEVEL);
+               cur = path->nodes[*level];
+
+               if (btrfs_header_level(cur) != *level)
+                       WARN_ON(1);
+
+               if (path->slots[*level] >=
+                   btrfs_header_nritems(cur))
+                       break;
+
+               bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
+               ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
+               blocksize = btrfs_level_size(root, *level - 1);
+
+               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 (*level == 1) {
+                       path->slots[*level]++;
+                       if (wc->free) {
+                               btrfs_read_buffer(next, ptr_gen);
+
+                               btrfs_tree_lock(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_extent(trans, root, bytenr,
+                                                       blocksize, root_owner,
+                                                       root_gen, 0, 0, 1);
+                               BUG_ON(ret);
+                       }
+                       free_extent_buffer(next);
+                       continue;
+               }
+               btrfs_read_buffer(next, ptr_gen);
+
+               WARN_ON(*level <= 0);
+               if (path->nodes[*level-1])
+                       free_extent_buffer(path->nodes[*level-1]);
+               path->nodes[*level-1] = next;
+               *level = btrfs_header_level(next);
+               path->slots[*level] = 0;
+               cond_resched();
+       }
+       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_extent(trans, root, bytenr, blocksize,
+                                         root_owner, root_gen, 0, 0, 1);
+               BUG_ON(ret);
+       }
+       free_extent_buffer(path->nodes[*level]);
+       path->nodes[*level] = NULL;
+       *level += 1;
+
+       cond_resched();
+       return 0;
+}
+
+static int noinline walk_up_log_tree(struct btrfs_trans_handle *trans,
+                                struct btrfs_root *root,
+                                struct btrfs_path *path, int *level,
+                                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];
+                       path->slots[i]++;
+                       *level = i;
+                       WARN_ON(*level == 0);
+                       return 0;
+               } else {
+                       if (path->nodes[*level] == root->node) {
+                               root_owner = root->root_key.objectid;
+                               root_gen =
+                                  btrfs_header_generation(path->nodes[*level]);
+                       } else {
+                               struct extent_buffer *node;
+                               node = path->nodes[*level + 1];
+                               root_owner = btrfs_header_owner(node);
+                               root_gen = btrfs_header_generation(node);
+                       }
+                       wc->process_func(root, path->nodes[*level], wc,
+                                btrfs_header_generation(path->nodes[*level]));
+                       if (wc->free) {
+                               struct extent_buffer *next;
+
+                               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_extent(trans, root,
+                                               path->nodes[*level]->start,
+                                               path->nodes[*level]->len,
+                                               root_owner, root_gen, 0, 0, 1);
+                               BUG_ON(ret);
+                       }
+                       free_extent_buffer(path->nodes[*level]);
+                       path->nodes[*level] = NULL;
+                       *level = i + 1;
+               }
+       }
+       return 1;
+}
+
+/*
+ * drop the reference count on the tree rooted at 'snap'.  This traverses
+ * the tree freeing any blocks that have a ref count of zero after being
+ * decremented.
+ */
+static int walk_log_tree(struct btrfs_trans_handle *trans,
+                        struct btrfs_root *log, struct walk_control *wc)
+{
+       int ret = 0;
+       int wret;
+       int level;
+       struct btrfs_path *path;
+       int i;
+       int orig_level;
+
+       path = btrfs_alloc_path();
+       BUG_ON(!path);
+
+       level = btrfs_header_level(log->node);
+       orig_level = level;
+       path->nodes[level] = log->node;
+       extent_buffer_get(log->node);
+       path->slots[level] = 0;
+
+       while(1) {
+               wret = walk_down_log_tree(trans, log, path, &level, wc);
+               if (wret > 0)
+                       break;
+               if (wret < 0)
+                       ret = wret;
+
+               wret = walk_up_log_tree(trans, log, path, &level, wc);
+               if (wret > 0)
+                       break;
+               if (wret < 0)
+                       ret = wret;
+       }
+
+       /* was the root node processed? if not, catch it here */
+       if (path->nodes[orig_level]) {
+               wc->process_func(log, path->nodes[orig_level], wc,
+                        btrfs_header_generation(path->nodes[orig_level]));
+               if (wc->free) {
+                       struct extent_buffer *next;
+
+                       next = path->nodes[orig_level];
+
+                       btrfs_tree_lock(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_extent(trans, log,
+                                               next->start, next->len,
+                                               log->root_key.objectid,
+                                               btrfs_header_generation(next),
+                                               0, 0, 1);
+                       BUG_ON(ret);
+               }
+       }
+
+       for (i = 0; i <= orig_level; i++) {
+               if (path->nodes[i]) {
+                       free_extent_buffer(path->nodes[i]);
+                       path->nodes[i] = NULL;
+               }
+       }
+       btrfs_free_path(path);
+       if (wc->free)
+               free_extent_buffer(log->node);
+       return ret;
+}
+
+int wait_log_commit(struct btrfs_root *log)
+{
+       DEFINE_WAIT(wait);
+       u64 transid = log->fs_info->tree_log_transid;
+
+       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))
+                       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));
+       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
+ */
+int btrfs_sync_log(struct btrfs_trans_handle *trans,
+                  struct btrfs_root *root)
+{
+       int ret;
+       unsigned long batch;
+       struct btrfs_root *log = root->log_root;
+       struct walk_control wc = {
+               .write = 1,
+               .process_func = process_one_buffer
+       };
+
+       mutex_lock(&log->fs_info->tree_log_mutex);
+       if (atomic_read(&log->fs_info->tree_log_commit)) {
+               wait_log_commit(log);
+               goto out;
+       }
+       atomic_set(&log->fs_info->tree_log_commit, 1);
+
+       while(1) {
+               mutex_unlock(&log->fs_info->tree_log_mutex);
+               schedule_timeout_uninterruptible(1);
+               mutex_lock(&log->fs_info->tree_log_mutex);
+               batch = log->fs_info->tree_log_batch;
+
+               while(atomic_read(&log->fs_info->tree_log_writers)) {
+                       DEFINE_WAIT(wait);
+                       prepare_to_wait(&log->fs_info->tree_log_wait, &wait,
+                                       TASK_UNINTERRUPTIBLE);
+                       batch = log->fs_info->tree_log_batch;
+                       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);
+               }
+               if (batch == log->fs_info->tree_log_batch)
+                       break;
+       }
+       ret = walk_log_tree(trans, log, &wc);
+       BUG_ON(ret);
+
+       ret = walk_log_tree(trans, log->fs_info->log_root_tree, &wc);
+       BUG_ON(ret);
+
+       wc.wait = 1;
+
+       ret = walk_log_tree(trans, log, &wc);
+       BUG_ON(ret);
+
+       ret = walk_log_tree(trans, log->fs_info->log_root_tree, &wc);
+       BUG_ON(ret);
+
+       btrfs_set_super_log_root(&root->fs_info->super_for_commit,
+                                log->fs_info->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));
+
+       write_ctree_super(trans, log->fs_info->tree_root);
+       log->fs_info->tree_log_transid++;
+       log->fs_info->tree_log_batch = 0;
+       atomic_set(&log->fs_info->tree_log_commit, 0);
+       smp_mb();
+       if (waitqueue_active(&log->fs_info->tree_log_wait))
+               wake_up(&log->fs_info->tree_log_wait);
+out:
+       mutex_unlock(&log->fs_info->tree_log_mutex);
+       return 0;
+
+}
+
+/*
+ * 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)
+{
+       int ret;
+       struct btrfs_root *log;
+       struct key;
+       struct walk_control wc = {
+               .free = 1,
+               .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);
+
+       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;
+}
+
+/*
+ * 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)
+{
+       u64 bytenr = btrfs_root_bytenr(&log->root_item);
+       int ret;
+
+       if (log->node->start == bytenr)
+               return 0;
+
+       btrfs_set_root_bytenr(&log->root_item, log->node->start);
+       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;
+}
+
+/*
+ * If both a file and directory are logged, and unlinks or renames are
+ * mixed in, we have a few interesting corners:
+ *
+ * create file X in dir Y
+ * link file X to X.link in dir Y
+ * fsync file X
+ * unlink file X but leave X.link
+ * fsync dir Y
+ *
+ * After a crash we would expect only X.link to exist.  But file X
+ * didn't get fsync'd again so the log has back refs for X and X.link.
+ *
+ * We solve this by removing directory entries and inode backrefs from the
+ * log when a file that was logged in the current transaction is
+ * unlinked.  Any later fsync will include the updated log entries, and
+ * we'll be able to reconstruct the proper directory items from backrefs.
+ *
+ * This optimizations allows us to avoid relogging the entire inode
+ * or the entire directory.
+ */
+int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
+                                struct btrfs_root *root,
+                                const char *name, int name_len,
+                                struct inode *dir, u64 index)
+{
+       struct btrfs_root *log;
+       struct btrfs_dir_item *di;
+       struct btrfs_path *path;
+       int ret;
+       int bytes_del = 0;
+
+       ret = join_running_log_trans(root);
+       if (ret)
+               return 0;
+
+       mutex_lock(&BTRFS_I(dir)->log_mutex);
+
+       log = root->log_root;
+       path = btrfs_alloc_path();
+       di = btrfs_lookup_dir_item(trans, log, path, dir->i_ino,
+                                  name, name_len, -1);
+       if (di && !IS_ERR(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,
+                                        index, name, name_len, -1);
+       if (di && !IS_ERR(di)) {
+               ret = btrfs_delete_one_dir_name(trans, log, path, di);
+               bytes_del += name_len;
+               BUG_ON(ret);
+       }
+
+       /* update the directory size in the log to reflect the names
+        * we have removed
+        */
+       if (bytes_del) {
+               struct btrfs_key key;
+
+               key.objectid = dir->i_ino;
+               key.offset = 0;
+               key.type = BTRFS_INODE_ITEM_KEY;
+               btrfs_release_path(log, path);
+
+               ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
+               if (ret == 0) {
+                       struct btrfs_inode_item *item;
+                       u64 i_size;
+
+                       item = btrfs_item_ptr(path->nodes[0], path->slots[0],
+                                             struct btrfs_inode_item);
+                       i_size = btrfs_inode_size(path->nodes[0], item);
+                       if (i_size > bytes_del)
+                               i_size -= bytes_del;
+                       else
+                               i_size = 0;
+                       btrfs_set_inode_size(path->nodes[0], item, i_size);
+                       btrfs_mark_buffer_dirty(path->nodes[0]);
+               } else
+                       ret = 0;
+               btrfs_release_path(log, path);
+       }
+
+       btrfs_free_path(path);
+       mutex_unlock(&BTRFS_I(dir)->log_mutex);
+       end_log_trans(root);
+
+       return 0;
+}
+
+/* see comments for btrfs_del_dir_entries_in_log */
+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)
+{
+       struct btrfs_root *log;
+       u64 index;
+       int ret;
+
+       ret = join_running_log_trans(root);
+       if (ret)
+               return 0;
+       log = root->log_root;
+       mutex_lock(&BTRFS_I(inode)->log_mutex);
+
+       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);
+
+       if (ret == 0 || ret == -ENOENT)
+               return 0;
+       return ret;
+}
+
+/*
+ * creates a range item in the log for 'dirid'.  first_offset and
+ * last_offset tell us which parts of the key space the log should
+ * be considered authoritative for.
+ */
+static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
+                                      struct btrfs_root *log,
+                                      struct btrfs_path *path,
+                                      int key_type, u64 dirid,
+                                      u64 first_offset, u64 last_offset)
+{
+       int ret;
+       struct btrfs_key key;
+       struct btrfs_dir_log_item *item;
+
+       key.objectid = dirid;
+       key.offset = first_offset;
+       if (key_type == BTRFS_DIR_ITEM_KEY)
+               key.type = BTRFS_DIR_LOG_ITEM_KEY;
+       else
+               key.type = BTRFS_DIR_LOG_INDEX_KEY;
+       ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
+       BUG_ON(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);
+       return 0;
+}
+
+/*
+ * log all the items included in the current transaction for a given
+ * directory.  This also creates the range items in the log tree required
+ * to replay anything deleted before the fsync
+ */
+static noinline int log_dir_items(struct btrfs_trans_handle *trans,
+                         struct btrfs_root *root, struct inode *inode,
+                         struct btrfs_path *path,
+                         struct btrfs_path *dst_path, int key_type,
+                         u64 min_offset, u64 *last_offset_ret)
+{
+       struct btrfs_key min_key;
+       struct btrfs_key max_key;
+       struct btrfs_root *log = root->log_root;
+       struct extent_buffer *src;
+       int ret;
+       int i;
+       int nritems;
+       u64 first_offset = min_offset;
+       u64 last_offset = (u64)-1;
+
+       log = root->log_root;
+       max_key.objectid = inode->i_ino;
+       max_key.offset = (u64)-1;
+       max_key.type = key_type;
+
+       min_key.objectid = inode->i_ino;
+       min_key.type = key_type;
+       min_key.offset = min_offset;
+
+       path->keep_locks = 1;
+
+       ret = btrfs_search_forward(root, &min_key, &max_key,
+                                  path, 0, trans->transid);
+
+       /*
+        * 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;
+               min_key.type = key_type;
+               min_key.offset = (u64)-1;
+               btrfs_release_path(root, path);
+               ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
+               if (ret < 0) {
+                       btrfs_release_path(root, path);
+                       return ret;
+               }
+               ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
+
+               /* if ret == 0 there are items for this type,
+                * create a range to tell us the last key of this type.
+                * otherwise, there are no items in this directory after
+                * *min_offset, and we create a range to indicate that.
+                */
+               if (ret == 0) {
+                       struct btrfs_key tmp;
+                       btrfs_item_key_to_cpu(path->nodes[0], &tmp,
+                                             path->slots[0]);
+                       if (key_type == tmp.type) {
+                               first_offset = max(min_offset, tmp.offset) + 1;
+                       }
+               }
+               goto done;
+       }
+
+       /* go backward to find any previous key */
+       ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
+       if (ret == 0) {
+               struct btrfs_key tmp;
+               btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
+               if (key_type == tmp.type) {
+                       first_offset = tmp.offset;
+                       ret = overwrite_item(trans, log, dst_path,
+                                            path->nodes[0], path->slots[0],
+                                            &tmp);
+               }
+       }
+       btrfs_release_path(root, path);
+
+       /* find the first key from this transaction again */
+       ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
+       if (ret != 0) {
+               WARN_ON(1);
+               goto done;
+       }
+
+       /*
+        * we have a block from this transaction, log every item in it
+        * from our directory
+        */
+       while(1) {
+               struct btrfs_key tmp;
+               src = path->nodes[0];
+               nritems = btrfs_header_nritems(src);
+               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)
+                               goto done;
+                       ret = overwrite_item(trans, log, dst_path, src, i,
+                                            &min_key);
+                       BUG_ON(ret);
+               }
+               path->slots[0] = nritems;
+
+               /*
+                * look ahead to the next item and see if it is also
+                * from this directory and from this transaction
+                */
+               ret = btrfs_next_leaf(root, path);
+               if (ret == 1) {
+                       last_offset = (u64)-1;
+                       goto done;
+               }
+               btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
+               if (tmp.objectid != inode->i_ino || tmp.type != key_type) {
+                       last_offset = (u64)-1;
+                       goto done;
+               }
+               if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
+                       ret = overwrite_item(trans, log, dst_path,
+                                            path->nodes[0], path->slots[0],
+                                            &tmp);
+
+                       BUG_ON(ret);
+                       last_offset = tmp.offset;
+                       goto done;
+               }
+       }
+done:
+       *last_offset_ret = last_offset;
+       btrfs_release_path(root, path);
+       btrfs_release_path(log, 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;
+}
+
+/*
+ * logging directories is very similar to logging inodes, We find all the items
+ * from the current transaction and write them to the log.
+ *
+ * The recovery code scans the directory in the subvolume, and if it finds a
+ * key in the range logged that is not present in the log tree, then it means
+ * that dir entry was unlinked during the transaction.
+ *
+ * In order for that scan to work, we must include one key smaller than
+ * the smallest logged by this transaction and one key larger than the largest
+ * key logged by this transaction.
+ */
+static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
+                         struct btrfs_root *root, struct inode *inode,
+                         struct btrfs_path *path,
+                         struct btrfs_path *dst_path)
+{
+       u64 min_key;
+       u64 max_key;
+       int ret;
+       int key_type = BTRFS_DIR_ITEM_KEY;
+
+again:
+       min_key = 0;
+       max_key = 0;
+       while(1) {
+               ret = log_dir_items(trans, root, inode, path,
+                                   dst_path, key_type, min_key,
+                                   &max_key);
+               BUG_ON(ret);
+               if (max_key == (u64)-1)
+                       break;
+               min_key = max_key + 1;
+       }
+
+       if (key_type == BTRFS_DIR_ITEM_KEY) {
+               key_type = BTRFS_DIR_INDEX_KEY;
+               goto again;
+       }
+       return 0;
+}
+
+/*
+ * a helper function to drop items from the log before we relog an
+ * inode.  max_key_type indicates the highest item type to remove.
+ * This cannot be run for file data extents because it does not
+ * free the extents they point to.
+ */
+static int drop_objectid_items(struct btrfs_trans_handle *trans,
+                                 struct btrfs_root *log,
+                                 struct btrfs_path *path,
+                                 u64 objectid, int max_key_type)
+{
+       int ret;
+       struct btrfs_key key;
+       struct btrfs_key found_key;
+
+       key.objectid = objectid;
+       key.type = max_key_type;
+       key.offset = (u64)-1;
+
+       while(1) {
+               ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
+
+               if (ret != 1)
+                       break;
+
+               if (path->slots[0] == 0)
+                       break;
+
+               path->slots[0]--;
+               btrfs_item_key_to_cpu(path->nodes[0], &found_key,
+                                     path->slots[0]);
+
+               if (found_key.objectid != objectid)
+                       break;
+
+               ret = btrfs_del_item(trans, log, path);
+               BUG_ON(ret);
+               btrfs_release_path(log, path);
+       }
+       btrfs_release_path(log, path);
+       return 0;
+}
+
+/* log a single inode in the tree log.
+ * At least one parent directory for this inode must exist in the tree
+ * or be logged already.
+ *
+ * Any items from this inode changed by the current transaction are copied
+ * to the log tree.  An extra reference is taken on any extents in this
+ * file, allowing us to avoid a whole pile of corner cases around logging
+ * blocks that have been removed from the tree.
+ *
+ * See LOG_INODE_ALL and related defines for a description of what inode_only
+ * does.
+ *
+ * This handles both files and directories.
+ */
+static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
+                            struct btrfs_root *root, struct inode *inode,
+                            int inode_only)
+{
+       struct btrfs_path *path;
+       struct btrfs_path *dst_path;
+       struct btrfs_key min_key;
+       struct btrfs_key max_key;
+       struct btrfs_root *log = root->log_root;
+       unsigned long src_offset;
+       unsigned long dst_offset;
+       struct extent_buffer *src;
+       struct btrfs_file_extent_item *extent;
+       struct btrfs_inode_item *inode_item;
+       u32 size;
+       int ret;
+
+       log = root->log_root;
+
+       path = btrfs_alloc_path();
+       dst_path = btrfs_alloc_path();
+
+       min_key.objectid = inode->i_ino;
+       min_key.type = BTRFS_INODE_ITEM_KEY;
+       min_key.offset = 0;
+
+       max_key.objectid = inode->i_ino;
+       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);
+
+       /*
+        * a brute force approach to making sure we get the most uptodate
+        * copies of everything.
+        */
+       if (S_ISDIR(inode->i_mode)) {
+               int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
+
+               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);
+       } else {
+               ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
+       }
+       BUG_ON(ret);
+       path->keep_locks = 1;
+
+       while(1) {
+               ret = btrfs_search_forward(root, &min_key, &max_key,
+                                          path, 0, trans->transid);
+               if (ret != 0)
+                       break;
+
+               if (min_key.objectid != inode->i_ino)
+                       break;
+               if (min_key.type > max_key.type)
+                       break;
+
+               src = path->nodes[0];
+               size = btrfs_item_size_nr(src, path->slots[0]);
+               ret = btrfs_insert_empty_item(trans, log, dst_path, &min_key,
+                                             size);
+               if (ret)
+                       BUG();
+
+               dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
+                                                  dst_path->slots[0]);
+
+               src_offset = btrfs_item_ptr_offset(src, path->slots[0]);
+
+               copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
+                                  src_offset, size);
+
+               if (inode_only == LOG_INODE_EXISTS &&
+                   min_key.type == BTRFS_INODE_ITEM_KEY) {
+                       inode_item = btrfs_item_ptr(dst_path->nodes[0],
+                                                   dst_path->slots[0],
+                                                   struct btrfs_inode_item);
+                       btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
+
+                       /* set the generation to zero so the recover code
+                        * can tell the difference between an logging
+                        * just to say 'this inode exists' and a logging
+                        * to say 'update this inode with these values'
+                        */
+                       btrfs_set_inode_generation(dst_path->nodes[0],
+                                                  inode_item, 0);
+               }
+               /* take a reference on file data extents so that truncates
+                * or deletes of this inode don't have to relog the inode
+                * again
+                */
+               if (btrfs_key_type(&min_key) == BTRFS_EXTENT_DATA_KEY) {
+                       int found_type;
+                       extent = btrfs_item_ptr(src, path->slots[0],
+                                               struct btrfs_file_extent_item);
+
+                       found_type = btrfs_file_extent_type(src, extent);
+                       if (found_type == BTRFS_FILE_EXTENT_REG) {
+                               u64 ds = btrfs_file_extent_disk_bytenr(src,
+                                                                  extent);
+                               u64 dl = btrfs_file_extent_disk_num_bytes(src,
+                                                                     extent);
+                               /* ds == 0 is a hole */
+                               if (ds != 0) {
+                                       ret = btrfs_inc_extent_ref(trans, log,
+                                                  ds, dl,
+                                                  log->root_key.objectid,
+                                                  0,
+                                                  inode->i_ino,
+                                                  min_key.offset);
+                                       BUG_ON(ret);
+                               }
+                       }
+               }
+
+               btrfs_mark_buffer_dirty(dst_path->nodes[0]);
+               btrfs_release_path(root, path);
+               btrfs_release_path(log, dst_path);
+
+               if (min_key.offset < (u64)-1)
+                       min_key.offset++;
+               else if (min_key.type < (u8)-1)
+                       min_key.type++;
+               else if (min_key.objectid < (u64)-1)
+                       min_key.objectid++;
+               else
+                       break;
+       }
+       if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
+               btrfs_release_path(root, path);
+               btrfs_release_path(log, dst_path);
+               ret = log_directory_changes(trans, root, inode, path, dst_path);
+               BUG_ON(ret);
+       }
+       mutex_unlock(&BTRFS_I(inode)->log_mutex);
+
+       btrfs_free_path(path);
+       btrfs_free_path(dst_path);
+
+       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);
+out:
+       return 0;
+}
+
+int btrfs_log_inode(struct btrfs_trans_handle *trans,
+                   struct btrfs_root *root, struct inode *inode,
+                   int inode_only)
+{
+       int ret;
+
+       start_log_trans(trans, root);
+       ret = __btrfs_log_inode(trans, root, inode, inode_only);
+       end_log_trans(root);
+       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 inode_only = LOG_INODE_ALL;
+       struct super_block *sb;
+       int ret;
+
+       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)
+                       break;
+
+               if (BTRFS_I(dentry->d_inode)->generation <=
+                   root->fs_info->last_trans_committed)
+                       break;
+       }
+       end_log_trans(root);
+       return 0;
+}
+
+/*
+ * it is not safe to log dentry if the chunk root has added new
+ * chunks.  This returns 0 if the dentry was logged, and 1 otherwise.
+ * If this returns 1, you must commit the transaction to safely get your
+ * data on disk.
+ */
+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);
+}
+
+/*
+ * should be called during mount to recover any replay any log trees
+ * from the FS
+ */
+int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
+{
+       int ret;
+       struct btrfs_path *path;
+       struct btrfs_trans_handle *trans;
+       struct btrfs_key key;
+       struct btrfs_key found_key;
+       struct btrfs_key tmp_key;
+       struct btrfs_root *log;
+       struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
+       struct walk_control wc = {
+               .process_func = process_one_buffer,
+               .stage = 0,
+       };
+
+       fs_info->log_root_recovering = 1;
+       path = btrfs_alloc_path();
+       BUG_ON(!path);
+
+       trans = btrfs_start_transaction(fs_info->tree_root, 1);
+
+       wc.trans = trans;
+       wc.pin = 1;
+
+       walk_log_tree(trans, log_root_tree, &wc);
+
+again:
+       key.objectid = BTRFS_TREE_LOG_OBJECTID;
+       key.offset = (u64)-1;
+       btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
+
+       while(1) {
+               ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
+               if (ret < 0)
+                       break;
+               if (ret > 0) {
+                       if (path->slots[0] == 0)
+                               break;
+                       path->slots[0]--;
+               }
+               btrfs_item_key_to_cpu(path->nodes[0], &found_key,
+                                     path->slots[0]);
+               btrfs_release_path(log_root_tree, 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);
+
+
+               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(!wc.replay_dest);
+
+               btrfs_record_root_in_trans(wc.replay_dest);
+               ret = walk_log_tree(trans, log, &wc);
+               BUG_ON(ret);
+
+               if (wc.stage == LOG_WALK_REPLAY_ALL) {
+                       ret = fixup_inode_link_counts(trans, wc.replay_dest,
+                                                     path);
+                       BUG_ON(ret);
+               }
+
+               key.offset = found_key.offset - 1;
+               free_extent_buffer(log->node);
+               kfree(log);
+
+               if (found_key.offset == 0)
+                       break;
+       }
+       btrfs_release_path(log_root_tree, path);
+
+       /* step one is to pin it all, step two is to replay just inodes */
+       if (wc.pin) {
+               wc.pin = 0;
+               wc.process_func = replay_one_buffer;
+               wc.stage = LOG_WALK_REPLAY_INODES;
+               goto again;
+       }
+       /* step three is to replay everything */
+       if (wc.stage < LOG_WALK_REPLAY_ALL) {
+               wc.stage++;
+               goto again;
+       }
+
+       btrfs_free_path(path);
+
+       free_extent_buffer(log_root_tree->node);
+       log_root_tree->log_root = NULL;
+       fs_info->log_root_recovering = 0;
+
+       /* step 4: commit the transaction, which also unpins the blocks */
+       btrfs_commit_transaction(trans, fs_info->tree_root);
+
+       kfree(log_root_tree);
+       return 0;
+}
diff --git a/fs/btrfs/tree-log.h b/fs/btrfs/tree-log.h
new file mode 100644 (file)
index 0000000..b9409b3
--- /dev/null
@@ -0,0 +1,41 @@
+/*
+ * Copyright (C) 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef __TREE_LOG_
+#define __TREE_LOG_
+
+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,
+                                struct inode *dir, u64 index);
+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);
+#endif