Btrfs: Change the super to point to a tree of trees to enable persistent snapshots

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

1 files changed, 107 insertions, 28 deletions

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index c34c0c60935f..3d4bf6833f2a 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -154,41 +154,96 @@ static int __commit_transaction(struct btrfs_root *root)
         return ret;
 }
 
+static int commit_extent_and_tree_roots(struct btrfs_root *tree_root,
+                                        struct btrfs_root *extent_root)
+{
+        int ret;
+        u64 old_extent_block;
+
+        while(1) {
+                old_extent_block = btrfs_root_blocknr(&extent_root->root_item);
+                if (old_extent_block == extent_root->node->blocknr)
+                        break;
+                btrfs_set_root_blocknr(&extent_root->root_item,
+                                       extent_root->node->blocknr);
+                ret = btrfs_update_root(tree_root,
+                                        &extent_root->root_key,
+                                        &extent_root->root_item);
+                BUG_ON(ret);
+        }
+        __commit_transaction(extent_root);
+        __commit_transaction(tree_root);
+        return 0;
+}
+
 int btrfs_commit_transaction(struct btrfs_root *root,
                              struct btrfs_super_block *s)
 {
         int ret = 0;
+        struct btrfs_buffer *snap = root->commit_root;
+        struct btrfs_key snap_key;
 
         ret = __commit_transaction(root);
-        if (!ret && root != root->extent_root)
-                ret = __commit_transaction(root->extent_root);
         BUG_ON(ret);
-        if (root->commit_root != root->node) {
-                struct btrfs_buffer *snap = root->commit_root;
-                root->commit_root = root->node;
-                root->node->count++;
-                ret = btrfs_drop_snapshot(root, snap);
-                BUG_ON(ret);
-                // btrfs_block_release(root, snap);
-        }
+
+        if (root->commit_root == root->node)
+                return 0;
+
+        memcpy(&snap_key, &root->root_key, sizeof(snap_key));
+        root->root_key.offset++;
+
+        btrfs_set_root_blocknr(&root->root_item, root->node->blocknr);
+        ret = btrfs_insert_root(root->tree_root, &root->root_key,
+                                &root->root_item);
+        BUG_ON(ret);
+
+        ret = commit_extent_and_tree_roots(root->tree_root, root->extent_root);
+        BUG_ON(ret);
+
         write_ctree_super(root, s);
-        btrfs_finish_extent_commit(root);
+        btrfs_finish_extent_commit(root->extent_root);
+        btrfs_finish_extent_commit(root->tree_root);
+
+        root->commit_root = root->node;
+        root->node->count++;
+        ret = btrfs_drop_snapshot(root, snap);
+        BUG_ON(ret);
+
+        ret = btrfs_del_root(root->tree_root, &snap_key);
+        BUG_ON(ret);
+
         return ret;
 }
 
-static int __setup_root(struct btrfs_root *root, struct btrfs_root *extent_root,
-                        struct btrfs_root_info *info, int fp)
+static int __setup_root(struct btrfs_root *root, u64 objectid, int fp)
 {
         INIT_LIST_HEAD(&root->trans);
         INIT_LIST_HEAD(&root->cache);
         root->cache_size = 0;
         root->fp = fp;
         root->node = NULL;
-        root->extent_root = extent_root;
         root->commit_root = NULL;
-        root->node = read_tree_block(root, info->tree_root);
         memset(&root->current_insert, 0, sizeof(root->current_insert));
         memset(&root->last_insert, 0, sizeof(root->last_insert));
+        memset(&root->root_key, 0, sizeof(root->root_key));
+        memset(&root->root_item, 0, sizeof(root->root_item));
+        return 0;
+}
+
+static int find_and_setup_root(struct btrfs_root *tree_root, u64 objectid,
+                        struct btrfs_root *root, int fp)
+{
+        int ret;
+
+        __setup_root(root, objectid, fp);
+        ret = btrfs_find_last_root(tree_root, objectid,
+                                   &root->root_item, &root->root_key);
+        BUG_ON(ret);
+
+        root->node = read_tree_block(root,
+                                     btrfs_root_blocknr(&root->root_item));
+        root->ref_cows = 0;
+        BUG_ON(!root->node);
         return 0;
 }
 
@@ -196,9 +251,19 @@ struct btrfs_root *open_ctree(char *filename, struct btrfs_super_block *super)
 {
         struct btrfs_root *root = malloc(sizeof(struct btrfs_root));
         struct btrfs_root *extent_root = malloc(sizeof(struct btrfs_root));
+        struct btrfs_root *tree_root = malloc(sizeof(struct btrfs_root));
         int fp;
         int ret;
 
+        root->extent_root = extent_root;
+        root->tree_root = tree_root;
+
+        extent_root->extent_root = extent_root;
+        extent_root->tree_root = tree_root;
+
+        tree_root->extent_root = extent_root;
+        tree_root->tree_root = tree_root;
+
         fp = open(filename, O_CREAT | O_RDWR, 0600);
         if (fp < 0) {
                 free(root);
@@ -208,11 +273,14 @@ struct btrfs_root *open_ctree(char *filename, struct btrfs_super_block *super)
         INIT_RADIX_TREE(&root->pinned_radix, GFP_KERNEL);
         INIT_RADIX_TREE(&extent_root->pinned_radix, GFP_KERNEL);
         INIT_RADIX_TREE(&extent_root->cache_radix, GFP_KERNEL);
+        INIT_RADIX_TREE(&tree_root->pinned_radix, GFP_KERNEL);
+        INIT_RADIX_TREE(&tree_root->cache_radix, GFP_KERNEL);
+
         ret = pread(fp, super, sizeof(struct btrfs_super_block),
                      BTRFS_SUPER_INFO_OFFSET(BTRFS_BLOCKSIZE));
-        if (ret == 0 || super->root_info.tree_root == 0) {
+        if (ret == 0 || btrfs_super_root(super) == 0) {
                 printf("making new FS!\n");
-                ret = mkfs(fp);
+                ret = mkfs(fp, 0, BTRFS_BLOCKSIZE);
                 if (ret)
                         return NULL;
                 ret = pread(fp, super, sizeof(struct btrfs_super_block),
@@ -221,24 +289,29 @@ struct btrfs_root *open_ctree(char *filename, struct btrfs_super_block *super)
                         return NULL;
         }
         BUG_ON(ret < 0);
-        __setup_root(root, extent_root, &super->root_info, fp);
-        __setup_root(extent_root, extent_root, &super->extent_info, fp);
+
+        __setup_root(tree_root, BTRFS_ROOT_TREE_OBJECTID, fp);
+        tree_root->node = read_tree_block(tree_root, btrfs_super_root(super));
+        BUG_ON(!tree_root->node);
+
+        ret = find_and_setup_root(tree_root, BTRFS_EXTENT_TREE_OBJECTID,
+                                  extent_root, fp);
+        BUG_ON(ret);
+
+        ret = find_and_setup_root(tree_root, BTRFS_FS_TREE_OBJECTID,
+                                  root, fp);
+        BUG_ON(ret);
+
         root->commit_root = root->node;
         root->node->count++;
+        root->ref_cows = 1;
         return root;
 }
 
-static int __update_root(struct btrfs_root *root, struct btrfs_root_info *info)
-{
-        info->tree_root = root->node->blocknr;
-        return 0;
-}
-
 int write_ctree_super(struct btrfs_root *root, struct btrfs_super_block *s)
 {
         int ret;
-        __update_root(root, &s->root_info);
-        __update_root(root->extent_root, &s->extent_info);
+        btrfs_set_super_root(s, root->tree_root->node->blocknr);
         ret = pwrite(root->fp, s, sizeof(*s),
                      BTRFS_SUPER_INFO_OFFSET(BTRFS_BLOCKSIZE));
         if (ret != sizeof(*s)) {
@@ -260,19 +333,25 @@ static int drop_cache(struct btrfs_root *root)
 }
 int close_ctree(struct btrfs_root *root, struct btrfs_super_block *s)
 {
+        int ret;
         btrfs_commit_transaction(root, s);
-        __commit_transaction(root->extent_root);
+        ret = commit_extent_and_tree_roots(root->tree_root, root->extent_root);
+        BUG_ON(ret);
         write_ctree_super(root, s);
         drop_cache(root->extent_root);
+        drop_cache(root->tree_root);
         drop_cache(root);
         BUG_ON(!list_empty(&root->trans));
         BUG_ON(!list_empty(&root->extent_root->trans));
+        BUG_ON(!list_empty(&root->tree_root->trans));
 
         close(root->fp);
         if (root->node)
                 btrfs_block_release(root, root->node);
         if (root->extent_root->node)
                 btrfs_block_release(root->extent_root, root->extent_root->node);
+        if (root->tree_root->node)
+                btrfs_block_release(root->tree_root, root->tree_root->node);
         btrfs_block_release(root, root->commit_root);
         free(root);
         printf("on close %d blocks are allocated\n", allocated_blocks);