Btrfs: reference counts on data extents

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

5 files changed, 82 insertions, 24 deletions

diff --git a/fs/btrfs/TODO b/fs/btrfs/TODO
index 2ca301b289a6..7aec75e9a37e 100644
--- a/fs/btrfs/TODO
+++ b/fs/btrfs/TODO
@@ -8,7 +8,10 @@
 * Add block mapping tree (simple dm layer)
 * Add simple tree locking (semaphore per tree)
 * Make allocator smarter
+* make level a field in header
+* add a block group to struct inode
 * Make directory hashing work on 32 bit
+* Make sure nobh stuff is working properly for cows
 * Do actual block accounting
 * Check compat and incompat flags on the inode
 * Add virtual filesystems, mountable snapshots
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 7b7120d3ab42..1a98952e0faf 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -9,10 +9,11 @@ struct btrfs_transaction;
 
 #define BTRFS_MAGIC "_BtRfS_M"
 
-#define BTRFS_ROOT_TREE_OBJECTID 1
-#define BTRFS_EXTENT_TREE_OBJECTID 2
-#define BTRFS_INODE_MAP_OBJECTID 3
-#define BTRFS_FS_TREE_OBJECTID 4
+#define BTRFS_ROOT_TREE_OBJECTID 1ULL
+#define BTRFS_EXTENT_TREE_OBJECTID 2ULL
+#define BTRFS_INODE_MAP_OBJECTID 3ULL
+#define BTRFS_FS_TREE_OBJECTID 4ULL
+#define BTRFS_FIRST_FREE_OBJECTID 5ULL
 
 /*
  * we can actually store much bigger names, but lets not confuse the rest
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 82f6e9eed1d0..4d4fc48c0a31 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -13,7 +13,7 @@ static int del_pending_extents(struct btrfs_trans_handle *trans, struct
                                btrfs_root *extent_root);
 
 static int inc_block_ref(struct btrfs_trans_handle *trans, struct btrfs_root
-                         *root, u64 blocknr)
+                         *root, u64 blocknr, u64 num_blocks)
 {
         struct btrfs_path path;
         int ret;
@@ -29,7 +29,7 @@ static int inc_block_ref(struct btrfs_trans_handle *trans, struct btrfs_root
         key.objectid = blocknr;
         key.flags = 0;
         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
-        key.offset = 1;
+        key.offset = num_blocks;
         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, &path,
                                 0, 1);
         if (ret != 0)
@@ -48,7 +48,7 @@ static int inc_block_ref(struct btrfs_trans_handle *trans, struct btrfs_root
 }
 
 static int lookup_block_ref(struct btrfs_trans_handle *trans, struct btrfs_root
-                            *root, u64 blocknr, u32 *refs)
+                            *root, u64 blocknr, u64 num_blocks, u32 *refs)
 {
         struct btrfs_path path;
         int ret;
@@ -57,7 +57,7 @@ static int lookup_block_ref(struct btrfs_trans_handle *trans, struct btrfs_root
         struct btrfs_extent_item *item;
         btrfs_init_path(&path);
         key.objectid = blocknr;
-        key.offset = 1;
+        key.offset = num_blocks;
         key.flags = 0;
         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, &path,
@@ -76,17 +76,34 @@ int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 {
         u64 blocknr;
         struct btrfs_node *buf_node;
+        struct btrfs_leaf *buf_leaf;
+        struct btrfs_disk_key *key;
+        struct btrfs_file_extent_item *fi;
         int i;
+        int leaf;
+        int ret;
 
         if (!root->ref_cows)
                 return 0;
         buf_node = btrfs_buffer_node(buf);
-        if (btrfs_is_leaf(buf_node))
-                return 0;
-
+        leaf = btrfs_is_leaf(buf_node);
+        buf_leaf = btrfs_buffer_leaf(buf);
         for (i = 0; i < btrfs_header_nritems(&buf_node->header); i++) {
-                blocknr = btrfs_node_blockptr(buf_node, i);
-                inc_block_ref(trans, root, blocknr);
+                if (leaf) {
+                        key = &buf_leaf->items[i].key;
+                        if (btrfs_disk_key_type(key) != BTRFS_EXTENT_DATA_KEY)
+                                continue;
+                        fi = btrfs_item_ptr(buf_leaf, i,
+                                            struct btrfs_file_extent_item);
+                        ret = inc_block_ref(trans, root,
+                                    btrfs_file_extent_disk_blocknr(fi),
+                                    btrfs_file_extent_disk_num_blocks(fi));
+                        BUG_ON(ret);
+                } else {
+                        blocknr = btrfs_node_blockptr(buf_node, i);
+                        ret = inc_block_ref(trans, root, blocknr, 1);
+                        BUG_ON(ret);
+                }
         }
         return 0;
 }
@@ -469,6 +486,37 @@ struct buffer_head *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
         return buf;
 }
 
+static int drop_leaf_ref(struct btrfs_trans_handle *trans,
+                         struct btrfs_root *root, struct buffer_head *cur)
+{
+        struct btrfs_disk_key *key;
+        struct btrfs_leaf *leaf;
+        struct btrfs_file_extent_item *fi;
+        int i;
+        int nritems;
+        int ret;
+
+        BUG_ON(!btrfs_is_leaf(btrfs_buffer_node(cur)));
+        leaf = btrfs_buffer_leaf(cur);
+        nritems = btrfs_header_nritems(&leaf->header);
+        for (i = 0; i < nritems; i++) {
+                key = &leaf->items[i].key;
+                if (btrfs_disk_key_type(key) != BTRFS_EXTENT_DATA_KEY)
+                        continue;
+                fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
+                /*
+                 * FIXME make sure to insert a trans record that
+                 * repeats the snapshot del on crash
+                 */
+                ret = btrfs_free_extent(trans, root,
+                                        btrfs_file_extent_disk_blocknr(fi),
+                                        btrfs_file_extent_disk_num_blocks(fi),
+                                        0);
+                BUG_ON(ret);
+        }
+        return 0;
+}
+
 /*
  * helper function for drop_snapshot, this walks down the tree dropping ref
  * counts as it goes.
@@ -483,28 +531,33 @@ static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root
         u32 refs;
 
         ret = lookup_block_ref(trans, root, path->nodes[*level]->b_blocknr,
-                               &refs);
+                               1, &refs);
         BUG_ON(ret);
         if (refs > 1)
                 goto out;
         /*
          * walk down to the last node level and free all the leaves
          */
-        while(*level > 0) {
+        while(*level >= 0) {
                 cur = path->nodes[*level];
                 if (path->slots[*level] >=
                     btrfs_header_nritems(btrfs_buffer_header(cur)))
                         break;
+                if (*level == 0) {
+                        ret = drop_leaf_ref(trans, root, cur);
+                        BUG_ON(ret);
+                        break;
+                }
                 blocknr = btrfs_node_blockptr(btrfs_buffer_node(cur),
                                               path->slots[*level]);
-                ret = lookup_block_ref(trans, root, blocknr, &refs);
-                if (refs != 1 || *level == 1) {
+                ret = lookup_block_ref(trans, root, blocknr, 1, &refs);
+                BUG_ON(ret);
+                if (refs != 1) {
                         path->slots[*level]++;
                         ret = btrfs_free_extent(trans, root, blocknr, 1, 1);
                         BUG_ON(ret);
                         continue;
                 }
-                BUG_ON(ret);
                 next = read_tree_block(root, blocknr);
                 if (path->nodes[*level-1])
                         btrfs_block_release(root, path->nodes[*level-1]);
@@ -513,8 +566,8 @@ static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root
                 path->slots[*level] = 0;
         }
 out:
-        ret = btrfs_free_extent(trans, root, path->nodes[*level]->b_blocknr,
-                                1, 1);
+        ret = btrfs_free_extent(trans, root,
+                                path->nodes[*level]->b_blocknr, 1, 1);
         btrfs_block_release(root, path->nodes[*level]);
         path->nodes[*level] = NULL;
         *level += 1;
@@ -544,10 +597,10 @@ static int walk_up_tree(struct btrfs_trans_handle *trans, struct btrfs_root
                         ret = btrfs_free_extent(trans, root,
                                                 path->nodes[*level]->b_blocknr,
                                                 1, 1);
+                        BUG_ON(ret);
                         btrfs_block_release(root, path->nodes[*level]);
                         path->nodes[*level] = NULL;
                         *level = i + 1;
-                        BUG_ON(ret);
                 }
         }
         return 1;
diff --git a/fs/btrfs/inode-map.c b/fs/btrfs/inode-map.c
index ad2d375b830d..1b2c5e043508 100644
--- a/fs/btrfs/inode-map.c
+++ b/fs/btrfs/inode-map.c
@@ -25,6 +25,7 @@ int btrfs_find_free_objectid(struct btrfs_trans_handle *trans,
         if (fs_root->fs_info->last_inode_alloc_dirid == dirid)
                 search_start = fs_root->fs_info->last_inode_alloc;
 
+        search_start = max(search_start, BTRFS_FIRST_FREE_OBJECTID);
         search_key.objectid = search_start;
         search_key.flags = 0;
         btrfs_set_key_type(&search_key, BTRFS_INODE_MAP_ITEM_KEY);
diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
index 2c2883f2856d..487da5a213c0 100644
--- a/fs/btrfs/super.c
+++ b/fs/btrfs/super.c
@@ -700,7 +700,7 @@ out:
 static int btrfs_prepare_write(struct file *file, struct page *page,
                                unsigned from, unsigned to)
 {
-        return block_prepare_write(page, from, to, btrfs_get_block);
+        return nobh_prepare_write(page, from, to, btrfs_get_block);
 }
 
 static void btrfs_write_super(struct super_block *sb)
@@ -721,7 +721,7 @@ static int btrfs_readpages(struct file *file, struct address_space *mapping,
 
 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
 {
-        return block_write_full_page(page, btrfs_get_block, wbc);
+        return nobh_writepage(page, btrfs_get_block, wbc);
 }
 
 static int btrfs_get_sb(struct file_system_type *fs_type,
@@ -768,7 +768,7 @@ static struct address_space_operations btrfs_aops = {
         .writepage      = btrfs_writepage,
         .sync_page      = block_sync_page,
         .prepare_write  = btrfs_prepare_write,
-        .commit_write   = generic_commit_write,
+        .commit_write   = nobh_commit_write,
 };
 
 static struct inode_operations btrfs_file_inode_operations = {