]> git.karo-electronics.de Git - mv-sheeva.git/blobdiff - fs/btrfs/disk-io.c
Btrfs: Fix extent_buffer and extent_state leaks
[mv-sheeva.git] / fs / btrfs / disk-io.c
index bb133104e2e91213cafddb63263db15b923f18b2..09f4e694624db3ab3d7f0d30f9bdc691090e739a 100644 (file)
-#include <linux/module.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/fs.h>
 #include <linux/blkdev.h>
-#include <linux/crypto.h>
+#include <linux/crc32c.h>
 #include <linux/scatterlist.h>
 #include <linux/swap.h>
+#include <linux/radix-tree.h>
+#include <linux/writeback.h>
+#include <linux/buffer_head.h> // for block_sync_page
 #include "ctree.h"
 #include "disk-io.h"
 #include "transaction.h"
+#include "btrfs_inode.h"
 
-#define PATTERN 0xDEADBEEFUL
-static inline void check_pattern(struct buffer_head *buf)
+#if 0
+static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
 {
-       if (buf->b_private != (void *)PATTERN)
-               WARN_ON(1);
+       if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
+               printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
+                      (unsigned long long)extent_buffer_blocknr(buf),
+                      (unsigned long long)btrfs_header_blocknr(buf));
+               return 1;
+       }
+       return 0;
 }
+#endif
 
-static inline void set_pattern(struct buffer_head *buf)
+struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
+                                           u64 blocknr)
 {
-       buf->b_private = (void *)PATTERN;
+       struct inode *btree_inode = root->fs_info->btree_inode;
+       struct extent_buffer *eb;
+       eb = find_extent_buffer(&BTRFS_I(btree_inode)->extent_tree,
+                                  blocknr * root->sectorsize,
+                                  root->sectorsize, GFP_NOFS);
+       if (eb)
+               eb->alloc_addr = (unsigned long)__builtin_return_address(0);
+       return eb;
 }
 
-static int check_tree_block(struct btrfs_root *root, struct buffer_head *buf)
+struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
+                                                u64 blocknr)
 {
-       struct btrfs_node *node = btrfs_buffer_node(buf);
-       if (buf->b_blocknr != btrfs_header_blocknr(&node->header)) {
-               BUG();
-       }
-       if (root->node && btrfs_header_parentid(&node->header) !=
-           btrfs_header_parentid(btrfs_buffer_header(root->node))) {
-               BUG();
-       }
-       return 0;
+       struct inode *btree_inode = root->fs_info->btree_inode;
+       struct extent_buffer *eb;
+       eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->extent_tree,
+                                  blocknr * root->sectorsize,
+                                  root->sectorsize, GFP_NOFS);
+       eb->alloc_addr = (unsigned long)__builtin_return_address(0);
+       return eb;
 }
 
-struct buffer_head *btrfs_find_tree_block(struct btrfs_root *root, u64 blocknr)
+struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
+                                   size_t page_offset, u64 start, u64 end,
+                                   int create)
 {
-       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;
-
-       page = find_lock_page(mapping, index);
-       if (!page)
-               return NULL;
-
-       if (!page_has_buffers(page))
-               goto out_unlock;
+       struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
+       struct extent_map *em;
+       int ret;
 
-       head = page_buffers(page);
-       bh = head;
-       do {
-               if (buffer_mapped(bh) && bh->b_blocknr == blocknr) {
-                       ret = bh;
-                       get_bh(bh);
-                       goto out_unlock;
-               }
-               bh = bh->b_this_page;
-       } while (bh != head);
-out_unlock:
-       unlock_page(page);
-       if (ret) {
-               touch_buffer(ret);
-               check_pattern(ret);
+again:
+       em = lookup_extent_mapping(em_tree, start, end);
+       if (em) {
+               goto out;
        }
-       page_cache_release(page);
-       return ret;
+       em = alloc_extent_map(GFP_NOFS);
+       if (!em) {
+               em = ERR_PTR(-ENOMEM);
+               goto out;
+       }
+       em->start = 0;
+       em->end = (i_size_read(inode) & ~((u64)PAGE_CACHE_SIZE -1)) - 1;
+       em->block_start = 0;
+       em->block_end = em->end;
+       em->bdev = inode->i_sb->s_bdev;
+       ret = add_extent_mapping(em_tree, em);
+       if (ret == -EEXIST) {
+               free_extent_map(em);
+               em = NULL;
+               goto again;
+       } else if (ret) {
+               em = ERR_PTR(ret);
+       }
+out:
+       return em;
 }
 
-struct buffer_head *btrfs_find_create_tree_block(struct btrfs_root *root,
-                                                u64 blocknr)
+static int btree_writepage(struct page *page, struct writeback_control *wbc)
 {
-       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;
-       u64 first_block = index << (PAGE_CACHE_SHIFT - blockbits);
-
-       page = grab_cache_page(mapping, index);
-       if (!page)
-               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)) {
-                       bh->b_bdev = root->fs_info->sb->s_bdev;
-                       bh->b_blocknr = first_block;
-                       set_buffer_mapped(bh);
-                       set_pattern(bh);
-               }
-               if (bh->b_blocknr == 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;
+       struct extent_map_tree *tree;
+       tree = &BTRFS_I(page->mapping->host)->extent_tree;
+       return extent_write_full_page(tree, page, btree_get_extent, wbc);
+}
+int btree_readpage(struct file *file, struct page *page)
+{
+       struct extent_map_tree *tree;
+       tree = &BTRFS_I(page->mapping->host)->extent_tree;
+       return extent_read_full_page(tree, page, btree_get_extent);
 }
 
-static sector_t max_block(struct block_device *bdev)
+static int btree_releasepage(struct page *page, gfp_t unused_gfp_flags)
 {
-       sector_t retval = ~((sector_t)0);
-       loff_t sz = i_size_read(bdev->bd_inode);
+       struct extent_map_tree *tree;
+       int ret;
 
-       if (sz) {
-               unsigned int size = block_size(bdev);
-               unsigned int sizebits = blksize_bits(size);
-               retval = (sz >> sizebits);
+       BUG_ON(page->private != 1);
+       tree = &BTRFS_I(page->mapping->host)->extent_tree;
+       ret = try_release_extent_mapping(tree, page);
+       if (ret == 1) {
+               ClearPagePrivate(page);
+               set_page_private(page, 0);
+               page_cache_release(page);
        }
-       return retval;
+       return ret;
 }
 
-static int btree_get_block(struct inode *inode, sector_t iblock,
-                          struct buffer_head *bh, int create)
+static void btree_invalidatepage(struct page *page, unsigned long offset)
 {
-       if (iblock >= max_block(inode->i_sb->s_bdev)) {
-               if (create)
-                       return -EIO;
-
-               /*
-                * for reads, we're just trying to fill a partial page.
-                * return a hole, they will have to call get_block again
-                * before they can fill it, and they will get -EIO at that
-                * time
-                */
-               return 0;
-       }
-       bh->b_bdev = inode->i_sb->s_bdev;
-       bh->b_blocknr = iblock;
-       set_buffer_mapped(bh);
-       return 0;
+       struct extent_map_tree *tree;
+       tree = &BTRFS_I(page->mapping->host)->extent_tree;
+       extent_invalidatepage(tree, page, offset);
+       btree_releasepage(page, GFP_NOFS);
 }
 
 int btrfs_csum_data(struct btrfs_root * root, char *data, size_t len,
                    char *result)
 {
-       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("sha256 digest failed\n");
-       }
-       return ret;
+       return 0;
+#if 0
+       u32 crc;
+       crc = crc32c(0, data, len);
+       memcpy(result, &crc, BTRFS_CRC32_SIZE);
+       return 0;
+#endif
 }
-static int csum_tree_block(struct btrfs_root *root, struct buffer_head *bh,
+
+#if 0
+static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
                           int verify)
 {
-       char result[BTRFS_CSUM_SIZE];
+       return 0;
+       char result[BTRFS_CRC32_SIZE];
        int ret;
        struct btrfs_node *node;
 
-       return 0;
        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_CSUM_SIZE)) {
-                       printk("checksum verify failed on %lu\n",
-                              bh->b_blocknr);
+               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_CSUM_SIZE);
+               memcpy(node->header.csum, result, BTRFS_CRC32_SIZE);
        }
        return 0;
 }
+#endif
 
+#if 0
 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 {
-#if 0
        struct buffer_head *bh;
-       struct btrfs_root *root = btrfs_sb(page->mapping->host->i_sb);
+       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,
@@ -208,82 +196,102 @@ static int btree_writepage(struct page *page, struct writeback_control *wbc)
                        csum_tree_block(root, bh, 0);
                bh = bh->b_this_page;
        } while (bh != head);
-#endif
        return block_write_full_page(page, btree_get_block, wbc);
 }
-
-static int btree_readpage(struct file * file, struct page * page)
-{
-       return block_read_full_page(page, btree_get_block);
-}
+#endif
 
 static struct address_space_operations btree_aops = {
        .readpage       = btree_readpage,
        .writepage      = btree_writepage,
+       .releasepage    = btree_releasepage,
+       .invalidatepage = btree_invalidatepage,
        .sync_page      = block_sync_page,
 };
 
-struct buffer_head *read_tree_block(struct btrfs_root *root, u64 blocknr)
+int readahead_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;
-       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;
-               csum_tree_block(root, bh, 1);
-               set_pattern(bh);
-       } else {
-               unlock_buffer(bh);
-       }
-       if (check_tree_block(root, bh))
-               BUG();
-       return bh;
-fail:
-       brelse(bh);
-       return NULL;
+       struct extent_buffer *buf = NULL;
+       struct inode *btree_inode = root->fs_info->btree_inode;
+       int ret = 0;
 
+       buf = btrfs_find_create_tree_block(root, blocknr);
+       if (!buf)
+               return 0;
+       read_extent_buffer_pages(&BTRFS_I(btree_inode)->extent_tree,
+                                buf, 0);
+       free_extent_buffer(buf);
+       return ret;
 }
 
-int dirty_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
-                    struct buffer_head *buf)
+struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 blocknr)
 {
-       WARN_ON(atomic_read(&buf->b_count) == 0);
-       mark_buffer_dirty(buf);
-       return 0;
+       struct extent_buffer *buf = NULL;
+       struct inode *btree_inode = root->fs_info->btree_inode;
+
+       buf = btrfs_find_create_tree_block(root, blocknr);
+       if (!buf)
+               return NULL;
+       read_extent_buffer_pages(&BTRFS_I(btree_inode)->extent_tree,
+                                buf, 1);
+       buf->alloc_addr = (unsigned long)__builtin_return_address(0);
+       return buf;
 }
 
 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
-                    struct buffer_head *buf)
+                    struct extent_buffer *buf)
+{
+       struct inode *btree_inode = root->fs_info->btree_inode;
+       clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->extent_tree, buf);
+       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)->extent_tree,
+                                       buf);
+       return 0;
+}
+
+int set_tree_block_dirty(struct btrfs_root *root, struct extent_buffer *buf)
 {
-       WARN_ON(atomic_read(&buf->b_count) == 0);
-       clear_buffer_dirty(buf);
+       struct inode *btree_inode = root->fs_info->btree_inode;
+       set_extent_buffer_dirty(&BTRFS_I(btree_inode)->extent_tree, buf);
        return 0;
 }
 
-static int __setup_root(struct btrfs_super_block *super,
+static int __setup_root(int blocksize,
                        struct btrfs_root *root,
                        struct btrfs_fs_info *fs_info,
                        u64 objectid)
 {
        root->node = NULL;
+       root->inode = NULL;
        root->commit_root = NULL;
-       root->blocksize = btrfs_super_blocksize(super);
+       root->sectorsize = blocksize;
+       root->nodesize = blocksize;
+       root->leafsize = blocksize;
        root->ref_cows = 0;
        root->fs_info = fs_info;
+       root->objectid = objectid;
+       root->last_trans = 0;
+       root->highest_inode = 0;
+       root->last_inode_alloc = 0;
+       root->name = NULL;
        memset(&root->root_key, 0, sizeof(root->root_key));
        memset(&root->root_item, 0, sizeof(root->root_item));
+       memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
+       memset(&root->root_kobj, 0, sizeof(root->root_kobj));
+       init_completion(&root->kobj_unregister);
+       init_rwsem(&root->snap_sem);
+       root->defrag_running = 0;
+       root->defrag_level = 0;
+       root->root_key.objectid = objectid;
        return 0;
 }
 
-static int find_and_setup_root(struct btrfs_super_block *super,
+static int find_and_setup_root(int blocksize,
                               struct btrfs_root *tree_root,
                               struct btrfs_fs_info *fs_info,
                               u64 objectid,
@@ -291,7 +299,7 @@ static int find_and_setup_root(struct btrfs_super_block *super,
 {
        int ret;
 
-       __setup_root(super, root, fs_info, objectid);
+       __setup_root(blocksize, root, fs_info, objectid);
        ret = btrfs_find_last_root(tree_root, objectid,
                                   &root->root_item, &root->root_key);
        BUG_ON(ret);
@@ -302,108 +310,273 @@ static int find_and_setup_root(struct btrfs_super_block *super,
        return 0;
 }
 
-struct btrfs_root *open_ctree(struct super_block *sb,
-                             struct buffer_head *sb_buffer,
-                             struct btrfs_super_block *disk_super)
+struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
+                                              struct btrfs_key *location)
+{
+       struct btrfs_root *root;
+       struct btrfs_root *tree_root = fs_info->tree_root;
+       struct btrfs_path *path;
+       struct extent_buffer *l;
+       u64 highest_inode;
+       int ret = 0;
+
+       root = kzalloc(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;
+       }
+
+       __setup_root(fs_info->sb->s_blocksize, root, fs_info,
+                    location->objectid);
+
+       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 = path->nodes[0];
+       read_extent_buffer(l, &root->root_item,
+              btrfs_item_ptr_offset(l, path->slots[0]),
+              sizeof(root->root_item));
+       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 = btrfs_find_highest_inode(root, &highest_inode);
+       if (ret == 0) {
+               root->highest_inode = highest_inode;
+               root->last_inode_alloc = highest_inode;
+       }
+       return root;
+}
+
+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 *root;
+       int ret;
+
+       root = radix_tree_lookup(&fs_info->fs_roots_radix,
+                                (unsigned long)location->objectid);
+       if (root)
+               return root;
+
+       root = btrfs_read_fs_root_no_radix(fs_info, location);
+       if (IS_ERR(root))
+               return root;
+       ret = radix_tree_insert(&fs_info->fs_roots_radix,
+                               (unsigned long)root->root_key.objectid,
+                               root);
+       if (ret) {
+               free_extent_buffer(root->node);
+               kfree(root);
+               return ERR_PTR(ret);
+       }
+
+       ret = btrfs_set_root_name(root, name, namelen);
+       if (ret) {
+               free_extent_buffer(root->node);
+               kfree(root);
+               return ERR_PTR(ret);
+       }
+
+       ret = btrfs_sysfs_add_root(root);
+       if (ret) {
+               free_extent_buffer(root->node);
+               kfree(root->name);
+               kfree(root);
+               return ERR_PTR(ret);
+       }
+
+       ret = btrfs_find_dead_roots(fs_info->tree_root,
+                                   root->root_key.objectid, root);
+       BUG_ON(ret);
+
+       return root;
+}
+
+struct btrfs_root *open_ctree(struct super_block *sb)
 {
-       struct btrfs_root *root = kmalloc(sizeof(struct btrfs_root),
-                                         GFP_NOFS);
        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_root *inode_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 (!btrfs_super_root(disk_super)) {
-               return NULL;
+       if (!extent_root || !tree_root || !fs_info) {
+               err = -ENOMEM;
+               goto fail;
        }
        init_bit_radix(&fs_info->pinned_radix);
        init_bit_radix(&fs_info->pending_del_radix);
-       sb_set_blocksize(sb, sb_buffer->b_size);
+       init_bit_radix(&fs_info->extent_ins_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);
+       memset(&fs_info->super_kobj, 0, sizeof(fs_info->super_kobj));
+       init_completion(&fs_info->kobj_unregister);
+       sb_set_blocksize(sb, 4096);
        fs_info->running_transaction = NULL;
-       fs_info->fs_root = root;
+       fs_info->last_trans_committed = 0;
        fs_info->tree_root = tree_root;
        fs_info->extent_root = extent_root;
-       fs_info->inode_root = inode_root;
-       fs_info->last_inode_alloc = 0;
-       fs_info->last_inode_alloc_dirid = 0;
-       fs_info->disk_super = disk_super;
        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;
+       extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
+                            fs_info->btree_inode->i_mapping,
+                            GFP_NOFS);
+       extent_map_tree_init(&fs_info->free_space_cache,
+                            fs_info->btree_inode->i_mapping, GFP_NOFS);
+       fs_info->do_barriers = 1;
+       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("sha256", 0, CRYPTO_ALG_ASYNC);
-       spin_lock_init(&fs_info->hash_lock);
-
-       if (!fs_info->hash_tfm || IS_ERR(fs_info->hash_tfm)) {
-               printk("failed to allocate sha256 hash\n");
-               return NULL;
-       }
 
        mutex_init(&fs_info->trans_mutex);
        mutex_init(&fs_info->fs_mutex);
-       memset(&fs_info->current_insert, 0, sizeof(fs_info->current_insert));
-       memset(&fs_info->last_insert, 0, sizeof(fs_info->last_insert));
 
-       __setup_root(disk_super, tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
+       __setup_root(sb->s_blocksize, tree_root,
+                    fs_info, BTRFS_ROOT_TREE_OBJECTID);
 
-       fs_info->sb_buffer = read_tree_block(tree_root, sb_buffer->b_blocknr);
+       fs_info->sb_buffer = read_tree_block(tree_root,
+                                            BTRFS_SUPER_INFO_OFFSET /
+                                            sb->s_blocksize);
 
-       if (!fs_info->sb_buffer) {
-printk("failed2\n");
-               return NULL;
-       }
-       brelse(sb_buffer);
-       sb_buffer = NULL;
-       disk_super = (struct btrfs_super_block *)fs_info->sb_buffer->b_data;
-       fs_info->disk_super = disk_super;
+       if (!fs_info->sb_buffer)
+               goto fail_iput;
+
+       read_extent_buffer(fs_info->sb_buffer, &fs_info->super_copy, 0,
+                          sizeof(fs_info->super_copy));
+
+       read_extent_buffer(fs_info->sb_buffer, fs_info->fsid,
+                          (unsigned long)btrfs_super_fsid(fs_info->sb_buffer),
+                          BTRFS_FSID_SIZE);
+       disk_super = &fs_info->super_copy;
+       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);
+
+
+       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));
-       BUG_ON(!tree_root->node);
+       if (!tree_root->node)
+               goto fail_sb_buffer;
 
-       ret = find_and_setup_root(disk_super, tree_root, fs_info,
+       mutex_lock(&fs_info->fs_mutex);
+       ret = find_and_setup_root(sb->s_blocksize, tree_root, fs_info,
                                  BTRFS_EXTENT_TREE_OBJECTID, extent_root);
-       BUG_ON(ret);
+       if (ret) {
+               mutex_unlock(&fs_info->fs_mutex);
+               goto fail_tree_root;
+       }
 
-       ret = find_and_setup_root(disk_super, tree_root, fs_info,
-                                 BTRFS_INODE_MAP_OBJECTID, inode_root);
-       BUG_ON(ret);
+       btrfs_read_block_groups(extent_root);
 
-       ret = find_and_setup_root(disk_super, tree_root, fs_info,
-                                 BTRFS_FS_TREE_OBJECTID, root);
-       BUG_ON(ret);
-       root->commit_root = root->node;
-       get_bh(root->node);
-       root->ref_cows = 1;
-       root->fs_info->generation = root->root_key.offset + 1;
-       return root;
+       fs_info->generation = btrfs_super_generation(disk_super) + 1;
+       mutex_unlock(&fs_info->fs_mutex);
+       return tree_root;
+
+fail_tree_root:
+       free_extent_buffer(tree_root->node);
+fail_sb_buffer:
+       free_extent_buffer(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 buffer_head *bh = root->fs_info->sb_buffer;
-       btrfs_set_super_root(root->fs_info->disk_super,
-                            root->fs_info->tree_root->node->b_blocknr);
-       lock_buffer(bh);
-       clear_buffer_dirty(bh);
-       csum_tree_block(root, bh, 0);
-       bh->b_end_io = end_buffer_write_sync;
-       get_bh(bh);
-       submit_bh(WRITE, bh);
-       wait_on_buffer(bh);
-       if (!buffer_uptodate(bh)) {
-               WARN_ON(1);
-               return -EIO;
+       int ret;
+       struct extent_buffer *super = root->fs_info->sb_buffer;
+       struct inode *btree_inode = root->fs_info->btree_inode;
+
+       set_extent_buffer_dirty(&BTRFS_I(btree_inode)->extent_tree, super);
+       ret = sync_page_range_nolock(btree_inode, btree_inode->i_mapping,
+                                    super->start, super->len);
+       return ret;
+}
+
+int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
+{
+       radix_tree_delete(&fs_info->fs_roots_radix,
+                         (unsigned long)root->root_key.objectid);
+       btrfs_sysfs_del_root(root);
+       if (root->inode)
+               iput(root->inode);
+       if (root->node)
+               free_extent_buffer(root->node);
+       if (root->commit_root)
+               free_extent_buffer(root->commit_root);
+       if (root->name)
+               kfree(root->name);
+       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++)
+                       btrfs_free_fs_root(fs_info, gang[i]);
        }
        return 0;
 }
@@ -412,43 +585,69 @@ int close_ctree(struct btrfs_root *root)
 {
        int ret;
        struct btrfs_trans_handle *trans;
+       struct btrfs_fs_info *fs_info = root->fs_info;
 
+       fs_info->closing = 1;
+       btrfs_transaction_flush_work(root);
+       mutex_lock(&fs_info->fs_mutex);
+       btrfs_defrag_dirty_roots(root->fs_info);
        trans = btrfs_start_transaction(root, 1);
-       btrfs_commit_transaction(trans, root);
+       ret = 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(NULL, root);
+       mutex_unlock(&fs_info->fs_mutex);
 
-       if (root->node)
-               btrfs_block_release(root, root->node);
-       if (root->fs_info->extent_root->node)
-               btrfs_block_release(root->fs_info->extent_root,
-                                   root->fs_info->extent_root->node);
-       if (root->fs_info->inode_root->node)
-               btrfs_block_release(root->fs_info->inode_root,
-                                   root->fs_info->inode_root->node);
-       if (root->fs_info->tree_root->node)
-               btrfs_block_release(root->fs_info->tree_root,
-                                   root->fs_info->tree_root->node);
-       btrfs_block_release(root, root->commit_root);
-       btrfs_block_release(root, root->fs_info->sb_buffer);
-       crypto_free_hash(root->fs_info->hash_tfm);
-       truncate_inode_pages(root->fs_info->btree_inode->i_mapping, 0);
-       iput(root->fs_info->btree_inode);
-       kfree(root->fs_info->extent_root);
-       kfree(root->fs_info->inode_root);
-       kfree(root->fs_info->tree_root);
-       kfree(root->fs_info);
-       kfree(root);
+       if (fs_info->extent_root->node)
+               free_extent_buffer(fs_info->extent_root->node);
+
+       if (fs_info->tree_root->node)
+               free_extent_buffer(fs_info->tree_root->node);
+
+       free_extent_buffer(fs_info->sb_buffer);
+       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 buffer_head *buf)
+int btrfs_buffer_uptodate(struct extent_buffer *buf)
 {
-       check_pattern(buf);
-       brelse(buf);
+       struct inode *btree_inode = buf->first_page->mapping->host;
+       return extent_buffer_uptodate(&BTRFS_I(btree_inode)->extent_tree, buf);
 }
 
+int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
+{
+       struct inode *btree_inode = buf->first_page->mapping->host;
+       return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->extent_tree,
+                                         buf);
+}
+
+void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
+{
+       struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
+       u64 transid = btrfs_header_generation(buf);
+       struct inode *btree_inode = root->fs_info->btree_inode;
+
+       if (transid != root->fs_info->generation) {
+               printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
+                       (unsigned long long)extent_buffer_blocknr(buf),
+                       transid, root->fs_info->generation);
+               WARN_ON(1);
+       }
+       set_extent_buffer_dirty(&BTRFS_I(btree_inode)->extent_tree, buf);
+}
+
+void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
+{
+       balance_dirty_pages_ratelimited_nr(
+                       root->fs_info->btree_inode->i_mapping, nr);
+}