Btrfs: i386 fixes from axboe
[linux-2.6-block.git] / fs / btrfs / disk-io.c
index 25ce07908ee3f89cb94679932a16d33963785f26..b0ebba6f2e467b119d3cdaa4d662f12b56928d4a 100644 (file)
-#define _XOPEN_SOURCE 500
-#include <stdio.h>
-#include <stdlib.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <fcntl.h>
-#include <unistd.h>
-#include "kerncompat.h"
-#include "radix-tree.h"
+/*
+ * Copyright (C) 2007 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/module.h>
+#include <linux/fs.h>
+#include <linux/blkdev.h>
+#include <linux/crypto.h>
+#include <linux/scatterlist.h>
+#include <linux/swap.h>
+#include <linux/radix-tree.h>
+#include <linux/writeback.h>
 #include "ctree.h"
 #include "disk-io.h"
 #include "transaction.h"
+#include "btrfs_inode.h"
 
-static int allocated_blocks = 0;
-int cache_max = 10000;
-
-static int check_tree_block(struct btrfs_root *root, struct btrfs_buffer *buf)
+u64 bh_blocknr(struct buffer_head *bh)
 {
-       if (buf->blocknr != btrfs_header_blocknr(&buf->node.header))
-               BUG();
-       if (root->node && btrfs_header_parentid(&buf->node.header) !=
-           btrfs_header_parentid(&root->node->node.header))
-               BUG();
-       return 0;
+       return bh->b_blocknr;
 }
 
-static int free_some_buffers(struct btrfs_root *root)
+static int check_tree_block(struct btrfs_root *root, struct buffer_head *buf)
 {
-       struct list_head *node, *next;
-       struct btrfs_buffer *b;
-       if (root->cache_size < cache_max)
-               return 0;
-       list_for_each_safe(node, next, &root->cache) {
-               b = list_entry(node, struct btrfs_buffer, cache);
-               if (b->count == 1) {
-                       BUG_ON(!list_empty(&b->dirty));
-                       list_del_init(&b->cache);
-                       btrfs_block_release(root, b);
-                       if (root->cache_size < cache_max)
-                               break;
-               }
+       struct btrfs_node *node = btrfs_buffer_node(buf);
+       if (bh_blocknr(buf) != btrfs_header_blocknr(&node->header)) {
+               printk(KERN_CRIT "bh_blocknr(buf) is %llu, header is %llu\n",
+                      (unsigned long long)bh_blocknr(buf),
+                      (unsigned long long)btrfs_header_blocknr(&node->header));
+               return 1;
        }
        return 0;
 }
 
-struct btrfs_buffer *alloc_tree_block(struct btrfs_root *root, u64 blocknr)
+struct buffer_head *btrfs_find_tree_block(struct btrfs_root *root, u64 blocknr)
 {
-       struct btrfs_buffer *buf;
-       int ret;
+       struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
+       int blockbits = root->fs_info->sb->s_blocksize_bits;
+       unsigned long index = blocknr >> (PAGE_CACHE_SHIFT - blockbits);
+       struct page *page;
+       struct buffer_head *bh;
+       struct buffer_head *head;
+       struct buffer_head *ret = NULL;
 
-       buf = malloc(sizeof(struct btrfs_buffer) + root->blocksize);
-       if (!buf)
-               return buf;
-       allocated_blocks++;
-       buf->blocknr = blocknr;
-       buf->count = 2;
-       INIT_LIST_HEAD(&buf->dirty);
-       free_some_buffers(root);
-       radix_tree_preload(GFP_KERNEL);
-       ret = radix_tree_insert(&root->cache_radix, blocknr, buf);
-       radix_tree_preload_end();
-       list_add_tail(&buf->cache, &root->cache);
-       root->cache_size++;
-       if (ret) {
-               free(buf);
+
+       page = find_lock_page(mapping, index);
+       if (!page)
                return NULL;
-       }
-       return buf;
+
+       if (!page_has_buffers(page))
+               goto out_unlock;
+
+       head = page_buffers(page);
+       bh = head;
+       do {
+               if (buffer_mapped(bh) && bh_blocknr(bh) == blocknr) {
+                       ret = bh;
+                       get_bh(bh);
+                       goto out_unlock;
+               }
+               bh = bh->b_this_page;
+       } while (bh != head);
+out_unlock:
+       unlock_page(page);
+       page_cache_release(page);
+       return ret;
 }
 
-struct btrfs_buffer *find_tree_block(struct btrfs_root *root, u64 blocknr)
+int btrfs_map_bh_to_logical(struct btrfs_root *root, struct buffer_head *bh,
+                            u64 logical)
 {
-       struct btrfs_buffer *buf;
-       buf = radix_tree_lookup(&root->cache_radix, blocknr);
-       if (buf) {
-               buf->count++;
+       if (logical == 0) {
+               bh->b_bdev = NULL;
+               bh->b_blocknr = 0;
+               set_buffer_mapped(bh);
        } else {
-               buf = alloc_tree_block(root, blocknr);
-               if (!buf) {
-                       BUG();
-                       return NULL;
-               }
+               map_bh(bh, root->fs_info->sb, logical);
        }
-       return buf;
+       return 0;
 }
 
-struct btrfs_buffer *read_tree_block(struct btrfs_root *root, u64 blocknr)
+struct buffer_head *btrfs_find_create_tree_block(struct btrfs_root *root,
+                                                u64 blocknr)
 {
-       loff_t offset = blocknr * root->blocksize;
-       struct btrfs_buffer *buf;
-       int ret;
+       struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
+       int blockbits = root->fs_info->sb->s_blocksize_bits;
+       unsigned long index = blocknr >> (PAGE_CACHE_SHIFT - blockbits);
+       struct page *page;
+       struct buffer_head *bh;
+       struct buffer_head *head;
+       struct buffer_head *ret = NULL;
+       int err;
+       u64 first_block = index << (PAGE_CACHE_SHIFT - blockbits);
+
+       page = find_or_create_page(mapping, index, GFP_NOFS);
+       if (!page)
+               return NULL;
 
-       buf = radix_tree_lookup(&root->cache_radix, blocknr);
-       if (buf) {
-               buf->count++;
-       } else {
-               buf = alloc_tree_block(root, blocknr);
-               if (!buf)
-                       return NULL;
-               ret = pread(root->fp, &buf->node, root->blocksize, offset);
-               if (ret != root->blocksize) {
-                       free(buf);
-                       return NULL;
+       if (!page_has_buffers(page))
+               create_empty_buffers(page, root->fs_info->sb->s_blocksize, 0);
+       head = page_buffers(page);
+       bh = head;
+       do {
+               if (!buffer_mapped(bh)) {
+                       err = btrfs_map_bh_to_logical(root, bh, first_block);
+                       BUG_ON(err);
                }
-       }
-       if (check_tree_block(root, buf))
-               BUG();
-       return buf;
+               if (bh_blocknr(bh) == blocknr) {
+                       ret = bh;
+                       get_bh(bh);
+                       goto out_unlock;
+               }
+               bh = bh->b_this_page;
+               first_block++;
+       } while (bh != head);
+out_unlock:
+       unlock_page(page);
+       if (ret)
+               touch_buffer(ret);
+       page_cache_release(page);
+       return ret;
 }
 
-int dirty_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
-                    struct btrfs_buffer *buf)
+static int btree_get_block(struct inode *inode, sector_t iblock,
+                          struct buffer_head *bh, int create)
 {
-       if (!list_empty(&buf->dirty))
-               return 0;
-       list_add_tail(&buf->dirty, &root->trans);
-       buf->count++;
-       return 0;
+       int err;
+       struct btrfs_root *root = BTRFS_I(bh->b_page->mapping->host)->root;
+       err = btrfs_map_bh_to_logical(root, bh, iblock);
+       return err;
 }
 
-int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
-                    struct btrfs_buffer *buf)
+int btrfs_csum_data(struct btrfs_root * root, char *data, size_t len,
+                   char *result)
 {
-       if (!list_empty(&buf->dirty)) {
-               list_del_init(&buf->dirty);
-               btrfs_block_release(root, buf);
+       struct scatterlist sg;
+       struct crypto_hash *tfm = root->fs_info->hash_tfm;
+       struct hash_desc desc;
+       int ret;
+
+       desc.tfm = tfm;
+       desc.flags = 0;
+       sg_init_one(&sg, data, len);
+       spin_lock(&root->fs_info->hash_lock);
+       ret = crypto_hash_digest(&desc, &sg, 1, result);
+       spin_unlock(&root->fs_info->hash_lock);
+       if (ret) {
+               printk("digest failed\n");
        }
-       return 0;
+       return ret;
 }
-
-int write_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
-                    struct btrfs_buffer *buf)
+static int csum_tree_block(struct btrfs_root *root, struct buffer_head *bh,
+                          int verify)
 {
-       u64 blocknr = buf->blocknr;
-       loff_t offset = blocknr * root->blocksize;
+       char result[BTRFS_CRC32_SIZE];
        int ret;
+       struct btrfs_node *node;
 
-       if (buf->blocknr != btrfs_header_blocknr(&buf->node.header))
-               BUG();
-       ret = pwrite(root->fp, &buf->node, root->blocksize, offset);
-       if (ret != root->blocksize)
+       ret = btrfs_csum_data(root, bh->b_data + BTRFS_CSUM_SIZE,
+                             bh->b_size - BTRFS_CSUM_SIZE, result);
+       if (ret)
                return ret;
+       if (verify) {
+               if (memcmp(bh->b_data, result, BTRFS_CRC32_SIZE)) {
+                       printk("btrfs: %s checksum verify failed on %llu\n",
+                              root->fs_info->sb->s_id,
+                              (unsigned long long)bh_blocknr(bh));
+                       return 1;
+               }
+       } else {
+               node = btrfs_buffer_node(bh);
+               memcpy(node->header.csum, result, BTRFS_CRC32_SIZE);
+       }
        return 0;
 }
 
-static int __commit_transaction(struct btrfs_trans_handle *trans, struct
-                               btrfs_root *root)
+static int btree_writepage(struct page *page, struct writeback_control *wbc)
 {
-       struct btrfs_buffer *b;
-       int ret = 0;
-       int wret;
-       while(!list_empty(&root->trans)) {
-               b = list_entry(root->trans.next, struct btrfs_buffer, dirty);
-               list_del_init(&b->dirty);
-               wret = write_tree_block(trans, root, b);
-               if (wret)
-                       ret = wret;
-               btrfs_block_release(root, b);
+       struct buffer_head *bh;
+       struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
+       struct buffer_head *head;
+       if (!page_has_buffers(page)) {
+               create_empty_buffers(page, root->fs_info->sb->s_blocksize,
+                                       (1 << BH_Dirty)|(1 << BH_Uptodate));
        }
-       return ret;
+       head = page_buffers(page);
+       bh = head;
+       do {
+               if (buffer_dirty(bh))
+                       csum_tree_block(root, bh, 0);
+               bh = bh->b_this_page;
+       } while (bh != head);
+       return block_write_full_page(page, btree_get_block, wbc);
 }
 
-static int commit_extent_and_tree_roots(struct btrfs_trans_handle *trans,
-                                       struct btrfs_root *tree_root, struct
-                                       btrfs_root *extent_root)
+static int btree_readpage(struct file * file, struct page * page)
 {
-       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(trans, tree_root,
-                                       &extent_root->root_key,
-                                       &extent_root->root_item);
-               BUG_ON(ret);
-       }
-       __commit_transaction(trans, extent_root);
-       __commit_transaction(trans, tree_root);
-       return 0;
+       return block_read_full_page(page, btree_get_block);
 }
 
-int btrfs_commit_transaction(struct btrfs_trans_handle *trans, struct
-                            btrfs_root *root, struct btrfs_super_block *s)
+static struct address_space_operations btree_aops = {
+       .readpage       = btree_readpage,
+       .writepage      = btree_writepage,
+       .sync_page      = block_sync_page,
+};
+
+int readahead_tree_block(struct btrfs_root *root, u64 blocknr)
 {
+       struct buffer_head *bh = NULL;
        int ret = 0;
-       struct btrfs_buffer *snap = root->commit_root;
-       struct btrfs_key snap_key;
 
-       ret = __commit_transaction(trans, root);
-       BUG_ON(ret);
-
-       if (root->commit_root == root->node)
+       bh = btrfs_find_create_tree_block(root, blocknr);
+       if (!bh)
                return 0;
+       if (buffer_uptodate(bh)) {
+               ret = 1;
+               goto done;
+       }
+       if (test_set_buffer_locked(bh)) {
+               ret = 1;
+               goto done;
+       }
+       if (!buffer_uptodate(bh)) {
+               get_bh(bh);
+               bh->b_end_io = end_buffer_read_sync;
+               submit_bh(READ, bh);
+       } else {
+               unlock_buffer(bh);
+               ret = 1;
+       }
+done:
+       brelse(bh);
+       return ret;
+}
 
-       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(trans, root->tree_root, &root->root_key,
-                               &root->root_item);
-       BUG_ON(ret);
-
-       ret = commit_extent_and_tree_roots(trans, root->tree_root,
-                                          root->extent_root);
-       BUG_ON(ret);
-
-       write_ctree_super(trans, root, s);
-       btrfs_finish_extent_commit(trans, root->extent_root);
-       btrfs_finish_extent_commit(trans, root->tree_root);
-
-       root->commit_root = root->node;
-       root->node->count++;
-       ret = btrfs_drop_snapshot(trans, root, snap);
-       BUG_ON(ret);
+struct buffer_head *read_tree_block(struct btrfs_root *root, u64 blocknr)
+{
+       struct buffer_head *bh = NULL;
+
+       bh = btrfs_find_create_tree_block(root, blocknr);
+       if (!bh)
+               return bh;
+       if (buffer_uptodate(bh))
+               goto uptodate;
+       lock_buffer(bh);
+       if (!buffer_uptodate(bh)) {
+               get_bh(bh);
+               bh->b_end_io = end_buffer_read_sync;
+               submit_bh(READ, bh);
+               wait_on_buffer(bh);
+               if (!buffer_uptodate(bh))
+                       goto fail;
+       } else {
+               unlock_buffer(bh);
+       }
+uptodate:
+       if (!buffer_checked(bh)) {
+               csum_tree_block(root, bh, 1);
+               set_buffer_checked(bh);
+       }
+       if (check_tree_block(root, bh))
+               goto fail;
+       return bh;
+fail:
+       brelse(bh);
+       return NULL;
+}
 
-       ret = btrfs_del_root(trans, root->tree_root, &snap_key);
-       BUG_ON(ret);
+int dirty_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+                    struct buffer_head *buf)
+{
+       WARN_ON(atomic_read(&buf->b_count) == 0);
+       mark_buffer_dirty(buf);
+       return 0;
+}
 
-       return ret;
+int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+                    struct buffer_head *buf)
+{
+       WARN_ON(atomic_read(&buf->b_count) == 0);
+       clear_buffer_dirty(buf);
+       return 0;
 }
 
-static int __setup_root(struct btrfs_super_block *super,
-                       struct btrfs_root *root, u64 objectid, int fp)
+static int __setup_root(int blocksize,
+                       struct btrfs_root *root,
+                       struct btrfs_fs_info *fs_info,
+                       u64 objectid)
 {
-       INIT_LIST_HEAD(&root->trans);
-       INIT_LIST_HEAD(&root->cache);
-       root->cache_size = 0;
-       root->fp = fp;
        root->node = NULL;
+       root->inode = NULL;
        root->commit_root = NULL;
-       root->blocksize = btrfs_super_blocksize(super);
+       root->blocksize = blocksize;
        root->ref_cows = 0;
-       memset(&root->current_insert, 0, sizeof(root->current_insert));
-       memset(&root->last_insert, 0, sizeof(root->last_insert));
+       root->fs_info = fs_info;
+       root->objectid = objectid;
+       root->last_trans = 0;
+       root->highest_inode = 0;
+       root->last_inode_alloc = 0;
        memset(&root->root_key, 0, sizeof(root->root_key));
        memset(&root->root_item, 0, sizeof(root->root_item));
+       root->root_key.objectid = objectid;
        return 0;
 }
 
-static int find_and_setup_root(struct btrfs_super_block *super,
-                              struct btrfs_root *tree_root, u64 objectid,
-                              struct btrfs_root *root, int fp)
+static int find_and_setup_root(int blocksize,
+                              struct btrfs_root *tree_root,
+                              struct btrfs_fs_info *fs_info,
+                              u64 objectid,
+                              struct btrfs_root *root)
 {
        int ret;
 
-       __setup_root(super, root, objectid, fp);
+       __setup_root(blocksize, root, fs_info, objectid);
        ret = btrfs_find_last_root(tree_root, objectid,
                                   &root->root_item, &root->root_key);
        BUG_ON(ret);
@@ -258,139 +336,298 @@ static int find_and_setup_root(struct btrfs_super_block *super,
        return 0;
 }
 
-struct btrfs_root *open_ctree(char *filename, struct btrfs_super_block *super)
+struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
+                                     struct btrfs_key *location)
 {
-       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;
+       struct btrfs_root *root;
+       struct btrfs_root *tree_root = fs_info->tree_root;
+       struct btrfs_path *path;
+       struct btrfs_leaf *l;
+       u64 highest_inode;
+       int ret = 0;
 
-       root->extent_root = extent_root;
-       root->tree_root = tree_root;
+       root = radix_tree_lookup(&fs_info->fs_roots_radix,
+                                (unsigned long)location->objectid);
+       if (root)
+               return root;
+       root = kmalloc(sizeof(*root), GFP_NOFS);
+       if (!root)
+               return ERR_PTR(-ENOMEM);
+       if (location->offset == (u64)-1) {
+               ret = find_and_setup_root(fs_info->sb->s_blocksize,
+                                         fs_info->tree_root, fs_info,
+                                         location->objectid, root);
+               if (ret) {
+                       kfree(root);
+                       return ERR_PTR(ret);
+               }
+               goto insert;
+       }
 
-       extent_root->extent_root = extent_root;
-       extent_root->tree_root = tree_root;
+       __setup_root(fs_info->sb->s_blocksize, root, fs_info,
+                    location->objectid);
 
-       tree_root->extent_root = extent_root;
-       tree_root->tree_root = tree_root;
+       path = btrfs_alloc_path();
+       BUG_ON(!path);
+       ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
+       if (ret != 0) {
+               if (ret > 0)
+                       ret = -ENOENT;
+               goto out;
+       }
+       l = btrfs_buffer_leaf(path->nodes[0]);
+       memcpy(&root->root_item,
+              btrfs_item_ptr(l, path->slots[0], struct btrfs_root_item),
+              sizeof(root->root_item));
+       memcpy(&root->root_key, location, sizeof(*location));
+       ret = 0;
+out:
+       btrfs_release_path(root, path);
+       btrfs_free_path(path);
+       if (ret) {
+               kfree(root);
+               return ERR_PTR(ret);
+       }
+       root->node = read_tree_block(root,
+                                    btrfs_root_blocknr(&root->root_item));
+       BUG_ON(!root->node);
+insert:
+       root->ref_cows = 1;
+       ret = radix_tree_insert(&fs_info->fs_roots_radix,
+                               (unsigned long)root->root_key.objectid,
+                               root);
+       if (ret) {
+               brelse(root->node);
+               kfree(root);
+               return ERR_PTR(ret);
+       }
+       ret = btrfs_find_highest_inode(root, &highest_inode);
+       if (ret == 0) {
+               root->highest_inode = highest_inode;
+               root->last_inode_alloc = highest_inode;
+       }
+       return root;
+}
 
-       fp = open(filename, O_CREAT | O_RDWR, 0600);
-       if (fp < 0) {
-               free(root);
-               return NULL;
+struct btrfs_root *open_ctree(struct super_block *sb)
+{
+       struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
+                                                GFP_NOFS);
+       struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
+                                              GFP_NOFS);
+       struct btrfs_fs_info *fs_info = kmalloc(sizeof(*fs_info),
+                                               GFP_NOFS);
+       int ret;
+       int err = -EIO;
+       struct btrfs_super_block *disk_super;
+
+       if (!extent_root || !tree_root || !fs_info) {
+               err = -ENOMEM;
+               goto fail;
        }
-       INIT_RADIX_TREE(&root->cache_radix, GFP_KERNEL);
-       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);
-       if (ret == 0 || btrfs_super_root(super) == 0) {
-               printf("making new FS!\n");
-               ret = mkfs(fp, 0, 1024);
-               if (ret)
-                       return NULL;
-               ret = pread(fp, super, sizeof(struct btrfs_super_block),
-                            BTRFS_SUPER_INFO_OFFSET);
-               if (ret != sizeof(struct btrfs_super_block))
-                       return NULL;
+       init_bit_radix(&fs_info->pinned_radix);
+       init_bit_radix(&fs_info->pending_del_radix);
+       init_bit_radix(&fs_info->extent_map_radix);
+       INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
+       INIT_RADIX_TREE(&fs_info->block_group_radix, GFP_KERNEL);
+       INIT_RADIX_TREE(&fs_info->block_group_data_radix, GFP_KERNEL);
+       INIT_LIST_HEAD(&fs_info->trans_list);
+       INIT_LIST_HEAD(&fs_info->dead_roots);
+       sb_set_blocksize(sb, 4096);
+       fs_info->running_transaction = NULL;
+       fs_info->tree_root = tree_root;
+       fs_info->extent_root = extent_root;
+       fs_info->sb = sb;
+       fs_info->btree_inode = new_inode(sb);
+       fs_info->btree_inode->i_ino = 1;
+       fs_info->btree_inode->i_nlink = 1;
+       fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
+       fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
+       fs_info->do_barriers = 1;
+       fs_info->extent_tree_insert_nr = 0;
+       fs_info->extent_tree_prealloc_nr = 0;
+       fs_info->closing = 0;
+
+       INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
+       BTRFS_I(fs_info->btree_inode)->root = tree_root;
+       memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
+              sizeof(struct btrfs_key));
+       insert_inode_hash(fs_info->btree_inode);
+       mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
+       fs_info->hash_tfm = crypto_alloc_hash("crc32c", 0, CRYPTO_ALG_ASYNC);
+       spin_lock_init(&fs_info->hash_lock);
+
+       if (!fs_info->hash_tfm || IS_ERR(fs_info->hash_tfm)) {
+               printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
+               err = -ENOMEM;
+               goto fail_iput;
        }
-       BUG_ON(ret < 0);
+       mutex_init(&fs_info->trans_mutex);
+       mutex_init(&fs_info->fs_mutex);
 
-       __setup_root(super, tree_root, BTRFS_ROOT_TREE_OBJECTID, fp);
-       tree_root->node = read_tree_block(tree_root, btrfs_super_root(super));
-       BUG_ON(!tree_root->node);
+       __setup_root(sb->s_blocksize, tree_root,
+                    fs_info, BTRFS_ROOT_TREE_OBJECTID);
 
-       ret = find_and_setup_root(super, tree_root, BTRFS_EXTENT_TREE_OBJECTID,
-                                 extent_root, fp);
-       BUG_ON(ret);
+       fs_info->sb_buffer = read_tree_block(tree_root,
+                                            BTRFS_SUPER_INFO_OFFSET /
+                                            sb->s_blocksize);
 
-       ret = find_and_setup_root(super, tree_root, BTRFS_FS_TREE_OBJECTID,
-                                 root, fp);
-       BUG_ON(ret);
+       if (!fs_info->sb_buffer)
+               goto fail_iput;
+       disk_super = (struct btrfs_super_block *)fs_info->sb_buffer->b_data;
 
-       root->commit_root = root->node;
-       root->node->count++;
-       root->ref_cows = 1;
-       return root;
+       if (!btrfs_super_root(disk_super))
+               goto fail_sb_buffer;
+
+       i_size_write(fs_info->btree_inode,
+                    btrfs_super_total_blocks(disk_super) <<
+                    fs_info->btree_inode->i_blkbits);
+
+       fs_info->disk_super = disk_super;
+
+       if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
+                   sizeof(disk_super->magic))) {
+               printk("btrfs: valid FS not found on %s\n", sb->s_id);
+               goto fail_sb_buffer;
+       }
+       tree_root->node = read_tree_block(tree_root,
+                                         btrfs_super_root(disk_super));
+       if (!tree_root->node)
+               goto fail_sb_buffer;
+
+       mutex_lock(&fs_info->fs_mutex);
+       ret = find_and_setup_root(sb->s_blocksize, tree_root, fs_info,
+                                 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
+       if (ret) {
+               mutex_unlock(&fs_info->fs_mutex);
+               goto fail_tree_root;
+       }
+
+       btrfs_read_block_groups(extent_root);
+
+       fs_info->generation = btrfs_super_generation(disk_super) + 1;
+       mutex_unlock(&fs_info->fs_mutex);
+       return tree_root;
+
+fail_tree_root:
+       btrfs_block_release(tree_root, tree_root->node);
+fail_sb_buffer:
+       btrfs_block_release(tree_root, fs_info->sb_buffer);
+fail_iput:
+       iput(fs_info->btree_inode);
+fail:
+       kfree(extent_root);
+       kfree(tree_root);
+       kfree(fs_info);
+       return ERR_PTR(err);
 }
 
 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
-                     *root, struct btrfs_super_block *s)
+                     *root)
 {
        int ret;
-       btrfs_set_super_root(s, root->tree_root->node->blocknr);
-       ret = pwrite(root->fp, s, sizeof(*s),
-                    BTRFS_SUPER_INFO_OFFSET);
-       if (ret != sizeof(*s)) {
-               fprintf(stderr, "failed to write new super block err %d\n", ret);
-               return ret;
+       struct buffer_head *bh = root->fs_info->sb_buffer;
+
+       btrfs_set_super_root(root->fs_info->disk_super,
+                            bh_blocknr(root->fs_info->tree_root->node));
+       lock_buffer(bh);
+       WARN_ON(atomic_read(&bh->b_count) < 1);
+       clear_buffer_dirty(bh);
+       csum_tree_block(root, bh, 0);
+       bh->b_end_io = end_buffer_write_sync;
+       get_bh(bh);
+       if (root->fs_info->do_barriers)
+               ret = submit_bh(WRITE_BARRIER, bh);
+       else
+               ret = submit_bh(WRITE, bh);
+       if (ret == -EOPNOTSUPP) {
+               set_buffer_uptodate(bh);
+               root->fs_info->do_barriers = 0;
+               ret = submit_bh(WRITE, bh);
+       }
+       wait_on_buffer(bh);
+       if (!buffer_uptodate(bh)) {
+               WARN_ON(1);
+               return -EIO;
        }
        return 0;
 }
 
-static int drop_cache(struct btrfs_root *root)
+static int free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
 {
-       while(!list_empty(&root->cache)) {
-               struct btrfs_buffer *b = list_entry(root->cache.next,
-                                                  struct btrfs_buffer, cache);
-               list_del_init(&b->cache);
-               btrfs_block_release(root, b);
+       radix_tree_delete(&fs_info->fs_roots_radix,
+                         (unsigned long)root->root_key.objectid);
+       if (root->inode)
+               iput(root->inode);
+       if (root->node)
+               brelse(root->node);
+       if (root->commit_root)
+               brelse(root->commit_root);
+       kfree(root);
+       return 0;
+}
+
+static int del_fs_roots(struct btrfs_fs_info *fs_info)
+{
+       int ret;
+       struct btrfs_root *gang[8];
+       int i;
+
+       while(1) {
+               ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
+                                            (void **)gang, 0,
+                                            ARRAY_SIZE(gang));
+               if (!ret)
+                       break;
+               for (i = 0; i < ret; i++)
+                       free_fs_root(fs_info, gang[i]);
        }
        return 0;
 }
-int close_ctree(struct btrfs_root *root, struct btrfs_super_block *s)
+
+int close_ctree(struct btrfs_root *root)
 {
        int ret;
        struct btrfs_trans_handle *trans;
-
-       trans = root->running_transaction;
-       btrfs_commit_transaction(trans, root, s);
-       ret = commit_extent_and_tree_roots(trans, root->tree_root,
-                                          root->extent_root);
+       struct btrfs_fs_info *fs_info = root->fs_info;
+
+       fs_info->closing = 1;
+       btrfs_transaction_flush_work(root);
+       mutex_lock(&fs_info->fs_mutex);
+       trans = btrfs_start_transaction(root, 1);
+       btrfs_commit_transaction(trans, root);
+       /* run commit again to  drop the original snapshot */
+       trans = btrfs_start_transaction(root, 1);
+       btrfs_commit_transaction(trans, root);
+       ret = btrfs_write_and_wait_transaction(NULL, root);
        BUG_ON(ret);
-       write_ctree_super(trans, 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);
+       write_ctree_super(NULL, root);
+       mutex_unlock(&fs_info->fs_mutex);
+
+       if (fs_info->extent_root->node)
+               btrfs_block_release(fs_info->extent_root,
+                                   fs_info->extent_root->node);
+       if (fs_info->tree_root->node)
+               btrfs_block_release(fs_info->tree_root,
+                                   fs_info->tree_root->node);
+       btrfs_block_release(root, fs_info->sb_buffer);
+       crypto_free_hash(fs_info->hash_tfm);
+       truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
+       iput(fs_info->btree_inode);
+
+       btrfs_free_block_groups(root->fs_info);
+       del_fs_roots(fs_info);
+       kfree(fs_info->extent_root);
+       kfree(fs_info->tree_root);
        return 0;
 }
 
-void btrfs_block_release(struct btrfs_root *root, struct btrfs_buffer *buf)
+void btrfs_block_release(struct btrfs_root *root, struct buffer_head *buf)
 {
-       buf->count--;
-       if (buf->count < 0)
-               BUG();
-       if (buf->count == 0) {
-               BUG_ON(!list_empty(&buf->cache));
-               BUG_ON(!list_empty(&buf->dirty));
-               if (!radix_tree_lookup(&root->cache_radix, buf->blocknr))
-                       BUG();
-               radix_tree_delete(&root->cache_radix, buf->blocknr);
-               memset(buf, 0, sizeof(*buf));
-               free(buf);
-               BUG_ON(allocated_blocks == 0);
-               allocated_blocks--;
-               BUG_ON(root->cache_size == 0);
-               root->cache_size--;
-       }
+       brelse(buf);
 }
 
+void btrfs_btree_balance_dirty(struct btrfs_root *root)
+{
+       balance_dirty_pages_ratelimited(root->fs_info->btree_inode->i_mapping);
+}