]> git.karo-electronics.de Git - karo-tx-linux.git/blobdiff - fs/btrfs/transaction.c
Btrfs: Throttle file_write when data=ordered is flushing the inode
[karo-tx-linux.git] / fs / btrfs / transaction.c
index 8a2545f6a5b6370438b64db9c267d7b84d1ed222..c85cb48d95ee3dcd2aa1c8dbb1ad46971202563a 100644 (file)
@@ -1,5 +1,25 @@
-#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/sched.h>
+#include <linux/writeback.h>
+#include <linux/pagemap.h>
 #include "ctree.h"
 #include "disk-io.h"
 #include "transaction.h"
@@ -8,14 +28,15 @@ static int total_trans = 0;
 extern struct kmem_cache *btrfs_trans_handle_cachep;
 extern struct kmem_cache *btrfs_transaction_cachep;
 
+static struct workqueue_struct *trans_wq;
+
 #define BTRFS_ROOT_TRANS_TAG 0
+#define BTRFS_ROOT_DEFRAG_TAG 1
 
-#define TRANS_MAGIC 0xE1E10E
-static void put_transaction(struct btrfs_transaction *transaction)
+static noinline void put_transaction(struct btrfs_transaction *transaction)
 {
        WARN_ON(transaction->use_count == 0);
        transaction->use_count--;
-       WARN_ON(transaction->magic != TRANS_MAGIC);
        if (transaction->use_count == 0) {
                WARN_ON(total_trans == 0);
                total_trans--;
@@ -25,7 +46,7 @@ static void put_transaction(struct btrfs_transaction *transaction)
        }
 }
 
-static int join_transaction(struct btrfs_root *root)
+static noinline int join_transaction(struct btrfs_root *root)
 {
        struct btrfs_transaction *cur_trans;
        cur_trans = root->fs_info->running_transaction;
@@ -36,17 +57,50 @@ static int join_transaction(struct btrfs_root *root)
                BUG_ON(!cur_trans);
                root->fs_info->generation++;
                root->fs_info->running_transaction = cur_trans;
-               cur_trans->num_writers = 0;
+               root->fs_info->last_alloc = 0;
+               root->fs_info->last_data_alloc = 0;
+               cur_trans->num_writers = 1;
+               cur_trans->num_joined = 0;
                cur_trans->transid = root->fs_info->generation;
                init_waitqueue_head(&cur_trans->writer_wait);
                init_waitqueue_head(&cur_trans->commit_wait);
-               cur_trans->magic = TRANS_MAGIC;
                cur_trans->in_commit = 0;
                cur_trans->use_count = 1;
                cur_trans->commit_done = 0;
+               cur_trans->start_time = get_seconds();
+               INIT_LIST_HEAD(&cur_trans->pending_snapshots);
                list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
+               btrfs_ordered_inode_tree_init(&cur_trans->ordered_inode_tree);
+               extent_io_tree_init(&cur_trans->dirty_pages,
+                                    root->fs_info->btree_inode->i_mapping,
+                                    GFP_NOFS);
+       } else {
+               cur_trans->num_writers++;
+               cur_trans->num_joined++;
+       }
+
+       return 0;
+}
+
+static noinline int record_root_in_trans(struct btrfs_root *root)
+{
+       u64 running_trans_id = root->fs_info->running_transaction->transid;
+       if (root->ref_cows && root->last_trans < running_trans_id) {
+               WARN_ON(root == root->fs_info->extent_root);
+               if (root->root_item.refs != 0) {
+                       radix_tree_tag_set(&root->fs_info->fs_roots_radix,
+                                  (unsigned long)root->root_key.objectid,
+                                  BTRFS_ROOT_TRANS_TAG);
+                       radix_tree_tag_set(&root->fs_info->fs_roots_radix,
+                                  (unsigned long)root->root_key.objectid,
+                                  BTRFS_ROOT_DEFRAG_TAG);
+                       root->commit_root = root->node;
+                       extent_buffer_get(root->node);
+               } else {
+                       WARN_ON(1);
+               }
+               root->last_trans = running_trans_id;
        }
-       cur_trans->num_writers++;
        return 0;
 }
 
@@ -56,29 +110,21 @@ struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
        struct btrfs_trans_handle *h =
                kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
        int ret;
-       u64 running_trans_id;
 
        mutex_lock(&root->fs_info->trans_mutex);
        ret = join_transaction(root);
        BUG_ON(ret);
-       running_trans_id = root->fs_info->running_transaction->transid;
 
-       if (root != root->fs_info->tree_root && root->last_trans <
-           running_trans_id) {
-               radix_tree_tag_set(&root->fs_info->fs_roots_radix,
-                                  (unsigned long)root->root_key.objectid,
-                                  BTRFS_ROOT_TRANS_TAG);
-               root->commit_root = root->node;
-               get_bh(root->node);
-       }
-       root->last_trans = running_trans_id;
-       h->transid = running_trans_id;
+       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;
        h->blocks_used = 0;
+       h->block_group = NULL;
+       h->alloc_exclude_nr = 0;
+       h->alloc_exclude_start = 0;
        root->fs_info->running_transaction->use_count++;
        mutex_unlock(&root->fs_info->trans_mutex);
-       h->magic = h->magic2 = TRANS_MAGIC;
        return h;
 }
 
@@ -87,14 +133,13 @@ int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 {
        struct btrfs_transaction *cur_trans;
 
-       WARN_ON(trans->magic != TRANS_MAGIC);
-       WARN_ON(trans->magic2 != TRANS_MAGIC);
        mutex_lock(&root->fs_info->trans_mutex);
        cur_trans = root->fs_info->running_transaction;
+       WARN_ON(cur_trans != trans->transaction);
        WARN_ON(cur_trans->num_writers < 1);
+       cur_trans->num_writers--;
        if (waitqueue_active(&cur_trans->writer_wait))
                wake_up(&cur_trans->writer_wait);
-       cur_trans->num_writers--;
        put_transaction(cur_trans);
        mutex_unlock(&root->fs_info->trans_mutex);
        memset(trans, 0, sizeof(*trans));
@@ -106,45 +151,98 @@ int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
                                     struct btrfs_root *root)
 {
-       filemap_write_and_wait(root->fs_info->btree_inode->i_mapping);
-       return 0;
+       int ret;
+       int err;
+       int werr = 0;
+       struct extent_io_tree *dirty_pages;
+       struct page *page;
+       struct inode *btree_inode = root->fs_info->btree_inode;
+       u64 start;
+       u64 end;
+       unsigned long index;
+
+       if (!trans || !trans->transaction) {
+               return filemap_write_and_wait(btree_inode->i_mapping);
+       }
+       dirty_pages = &trans->transaction->dirty_pages;
+       while(1) {
+               ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
+                                           EXTENT_DIRTY);
+               if (ret)
+                       break;
+               clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
+               while(start <= end) {
+                       index = start >> PAGE_CACHE_SHIFT;
+                       start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
+                       page = find_lock_page(btree_inode->i_mapping, index);
+                       if (!page)
+                               continue;
+                       if (PageWriteback(page)) {
+                               if (PageDirty(page))
+                                       wait_on_page_writeback(page);
+                               else {
+                                       unlock_page(page);
+                                       page_cache_release(page);
+                                       continue;
+                               }
+                       }
+                       err = write_one_page(page, 0);
+                       if (err)
+                               werr = err;
+                       page_cache_release(page);
+               }
+       }
+       err = filemap_fdatawait(btree_inode->i_mapping);
+       if (err)
+               werr = err;
+       return werr;
 }
 
-int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
-                           struct btrfs_root *root)
+static int update_cowonly_root(struct btrfs_trans_handle *trans,
+                              struct btrfs_root *root)
 {
        int ret;
-       u64 old_extent_block;
-       struct btrfs_fs_info *fs_info = root->fs_info;
-       struct btrfs_root *tree_root = fs_info->tree_root;
-       struct btrfs_root *extent_root = fs_info->extent_root;
-       struct btrfs_root *dev_root = fs_info->dev_root;
+       u64 old_root_bytenr;
+       struct btrfs_root *tree_root = root->fs_info->tree_root;
 
-       if (btrfs_super_device_root(fs_info->disk_super) !=
-           bh_blocknr(dev_root->node)) {
-               btrfs_set_super_device_root(fs_info->disk_super,
-                                           bh_blocknr(dev_root->node));
-       }
-       btrfs_write_dirty_block_groups(trans, extent_root);
+       btrfs_write_dirty_block_groups(trans, root);
        while(1) {
-               old_extent_block = btrfs_root_blocknr(&extent_root->root_item);
-               if (old_extent_block == bh_blocknr(extent_root->node))
+               old_root_bytenr = btrfs_root_bytenr(&root->root_item);
+               if (old_root_bytenr == root->node->start)
                        break;
-               btrfs_set_root_blocknr(&extent_root->root_item,
-                                      bh_blocknr(extent_root->node));
+               btrfs_set_root_bytenr(&root->root_item,
+                                      root->node->start);
+               btrfs_set_root_level(&root->root_item,
+                                    btrfs_header_level(root->node));
                ret = btrfs_update_root(trans, tree_root,
-                                       &extent_root->root_key,
-                                       &extent_root->root_item);
+                                       &root->root_key,
+                                       &root->root_item);
                BUG_ON(ret);
-               btrfs_write_dirty_block_groups(trans, extent_root);
+               btrfs_write_dirty_block_groups(trans, root);
        }
        return 0;
 }
 
-static int wait_for_commit(struct btrfs_root *root,
-                          struct btrfs_transaction *commit)
+int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
+                           struct btrfs_root *root)
+{
+       struct btrfs_fs_info *fs_info = root->fs_info;
+       struct list_head *next;
+
+       while(!list_empty(&fs_info->dirty_cowonly_roots)) {
+               next = fs_info->dirty_cowonly_roots.next;
+               list_del_init(next);
+               root = list_entry(next, struct btrfs_root, dirty_list);
+               update_cowonly_root(trans, root);
+       }
+       return 0;
+}
+
+static noinline int wait_for_commit(struct btrfs_root *root,
+                                   struct btrfs_transaction *commit)
 {
        DEFINE_WAIT(wait);
+       mutex_lock(&root->fs_info->trans_mutex);
        while(!commit->commit_done) {
                prepare_to_wait(&commit->commit_wait, &wait,
                                TASK_UNINTERRUPTIBLE);
@@ -154,26 +252,44 @@ static int wait_for_commit(struct btrfs_root *root,
                schedule();
                mutex_lock(&root->fs_info->trans_mutex);
        }
+       mutex_unlock(&root->fs_info->trans_mutex);
        finish_wait(&commit->commit_wait, &wait);
        return 0;
 }
 
 struct dirty_root {
        struct list_head list;
-       struct btrfs_key snap_key;
-       struct buffer_head *commit_root;
        struct btrfs_root *root;
+       struct btrfs_root *latest_root;
 };
 
-int add_dirty_roots(struct btrfs_trans_handle *trans,
-                   struct radix_tree_root *radix, struct list_head *list)
+int btrfs_add_dead_root(struct btrfs_root *root,
+                       struct btrfs_root *latest,
+                       struct list_head *dead_list)
+{
+       struct dirty_root *dirty;
+
+       dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
+       if (!dirty)
+               return -ENOMEM;
+       dirty->root = root;
+       dirty->latest_root = latest;
+       list_add(&dirty->list, dead_list);
+       return 0;
+}
+
+static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
+                                   struct radix_tree_root *radix,
+                                   struct list_head *list)
 {
        struct dirty_root *dirty;
        struct btrfs_root *gang[8];
        struct btrfs_root *root;
        int i;
        int ret;
-       int err;
+       int err = 0;
+       u32 refs;
+
        while(1) {
                ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
                                                 ARRAY_SIZE(gang),
@@ -186,51 +302,328 @@ int add_dirty_roots(struct btrfs_trans_handle *trans,
                                     (unsigned long)root->root_key.objectid,
                                     BTRFS_ROOT_TRANS_TAG);
                        if (root->commit_root == root->node) {
-                               WARN_ON(bh_blocknr(root->node) !=
-                                       btrfs_root_blocknr(&root->root_item));
-                               brelse(root->commit_root);
+                               WARN_ON(root->node->start !=
+                                       btrfs_root_bytenr(&root->root_item));
+                               free_extent_buffer(root->commit_root);
                                root->commit_root = NULL;
+
+                               /* make sure to update the root on disk
+                                * so we get any updates to the block used
+                                * counts
+                                */
+                               err = btrfs_update_root(trans,
+                                               root->fs_info->tree_root,
+                                               &root->root_key,
+                                               &root->root_item);
                                continue;
                        }
                        dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
                        BUG_ON(!dirty);
-                       memcpy(&dirty->snap_key, &root->root_key,
-                              sizeof(root->root_key));
-                       dirty->commit_root = root->commit_root;
+                       dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
+                       BUG_ON(!dirty->root);
+
+                       memset(&root->root_item.drop_progress, 0,
+                              sizeof(struct btrfs_disk_key));
+                       root->root_item.drop_level = 0;
+
+                       memcpy(dirty->root, root, sizeof(*root));
+                       dirty->root->node = root->commit_root;
+                       dirty->latest_root = root;
                        root->commit_root = NULL;
-                       dirty->root = root;
+
                        root->root_key.offset = root->fs_info->generation;
-                       btrfs_set_root_blocknr(&root->root_item,
-                                              bh_blocknr(root->node));
+                       btrfs_set_root_bytenr(&root->root_item,
+                                             root->node->start);
+                       btrfs_set_root_level(&root->root_item,
+                                            btrfs_header_level(root->node));
                        err = btrfs_insert_root(trans, root->fs_info->tree_root,
                                                &root->root_key,
                                                &root->root_item);
+                       if (err)
+                               break;
+
+                       refs = btrfs_root_refs(&dirty->root->root_item);
+                       btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
+                       err = btrfs_update_root(trans, root->fs_info->tree_root,
+                                               &dirty->root->root_key,
+                                               &dirty->root->root_item);
+
                        BUG_ON(err);
-                       list_add(&dirty->list, list);
+                       if (refs == 1) {
+                               list_add(&dirty->list, list);
+                       } else {
+                               WARN_ON(1);
+                               kfree(dirty->root);
+                               kfree(dirty);
+                       }
                }
        }
+       return err;
+}
+
+int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
+{
+       struct btrfs_fs_info *info = root->fs_info;
+       int ret;
+       struct btrfs_trans_handle *trans;
+       unsigned long nr;
+
+       if (root->defrag_running)
+               return 0;
+       trans = btrfs_start_transaction(root, 1);
+       while (1) {
+               root->defrag_running = 1;
+               ret = btrfs_defrag_leaves(trans, root, cacheonly);
+               nr = trans->blocks_used;
+               btrfs_end_transaction(trans, root);
+               mutex_unlock(&info->fs_mutex);
+               btrfs_btree_balance_dirty(info->tree_root, nr);
+               cond_resched();
+
+               mutex_lock(&info->fs_mutex);
+               trans = btrfs_start_transaction(root, 1);
+               if (ret != -EAGAIN)
+                       break;
+       }
+       root->defrag_running = 0;
+       radix_tree_tag_clear(&info->fs_roots_radix,
+                    (unsigned long)root->root_key.objectid,
+                    BTRFS_ROOT_DEFRAG_TAG);
+       btrfs_end_transaction(trans, root);
        return 0;
 }
 
-int drop_dirty_roots(struct btrfs_root *tree_root, struct list_head *list)
+int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
+{
+       struct btrfs_root *gang[1];
+       struct btrfs_root *root;
+       int i;
+       int ret;
+       int err = 0;
+       u64 last = 0;
+
+       while(1) {
+               ret = radix_tree_gang_lookup_tag(&info->fs_roots_radix,
+                                                (void **)gang, last,
+                                                ARRAY_SIZE(gang),
+                                                BTRFS_ROOT_DEFRAG_TAG);
+               if (ret == 0)
+                       break;
+               for (i = 0; i < ret; i++) {
+                       root = gang[i];
+                       last = root->root_key.objectid + 1;
+                       btrfs_defrag_root(root, 1);
+               }
+       }
+       btrfs_defrag_root(info->extent_root, 1);
+       return err;
+}
+
+static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
+                                    struct list_head *list)
 {
        struct dirty_root *dirty;
        struct btrfs_trans_handle *trans;
-       int ret;
+       unsigned long nr;
+       u64 num_bytes;
+       u64 bytes_used;
+       int ret = 0;
+       int err;
 
        while(!list_empty(list)) {
+               struct btrfs_root *root;
+
+               mutex_lock(&tree_root->fs_info->fs_mutex);
                dirty = list_entry(list->next, struct dirty_root, list);
                list_del_init(&dirty->list);
-               trans = btrfs_start_transaction(tree_root, 1);
-               ret = btrfs_drop_snapshot(trans, dirty->root,
-                                         dirty->commit_root);
-               BUG_ON(ret);
 
-               ret = btrfs_del_root(trans, tree_root, &dirty->snap_key);
+               num_bytes = btrfs_root_used(&dirty->root->root_item);
+               root = dirty->latest_root;
+               root->fs_info->throttles++;
+
+               while(1) {
+                       trans = btrfs_start_transaction(tree_root, 1);
+                       ret = btrfs_drop_snapshot(trans, dirty->root);
+                       if (ret != -EAGAIN) {
+                               break;
+                       }
+
+                       err = btrfs_update_root(trans,
+                                       tree_root,
+                                       &dirty->root->root_key,
+                                       &dirty->root->root_item);
+                       if (err)
+                               ret = err;
+                       nr = trans->blocks_used;
+                       ret = btrfs_end_transaction(trans, tree_root);
+                       BUG_ON(ret);
+                       mutex_unlock(&tree_root->fs_info->fs_mutex);
+                       btrfs_btree_balance_dirty(tree_root, nr);
+                       cond_resched();
+                       mutex_lock(&tree_root->fs_info->fs_mutex);
+               }
                BUG_ON(ret);
+               root->fs_info->throttles--;
+
+               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_set_root_used(&root->root_item,
+                                           bytes_used - num_bytes);
+               }
+               ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
+               if (ret) {
+                       BUG();
+                       break;
+               }
+               nr = trans->blocks_used;
                ret = btrfs_end_transaction(trans, tree_root);
                BUG_ON(ret);
+
+               free_extent_buffer(dirty->root->node);
+               kfree(dirty->root);
                kfree(dirty);
+               mutex_unlock(&tree_root->fs_info->fs_mutex);
+
+               btrfs_btree_balance_dirty(tree_root, nr);
+               cond_resched();
+       }
+       return ret;
+}
+
+int btrfs_write_ordered_inodes(struct btrfs_trans_handle *trans,
+                               struct btrfs_root *root)
+{
+       struct btrfs_transaction *cur_trans = trans->transaction;
+       struct inode *inode;
+       u64 root_objectid = 0;
+       u64 objectid = 0;
+       int ret;
+
+       root->fs_info->throttles++;
+       while(1) {
+               ret = btrfs_find_first_ordered_inode(
+                               &cur_trans->ordered_inode_tree,
+                               &root_objectid, &objectid, &inode);
+               if (!ret)
+                       break;
+
+               mutex_unlock(&root->fs_info->trans_mutex);
+               mutex_unlock(&root->fs_info->fs_mutex);
+
+               if (S_ISREG(inode->i_mode)) {
+                       atomic_inc(&BTRFS_I(inode)->ordered_writeback);
+                       filemap_fdatawrite(inode->i_mapping);
+                       atomic_dec(&BTRFS_I(inode)->ordered_writeback);
+               }
+               iput(inode);
+
+               mutex_lock(&root->fs_info->fs_mutex);
+               mutex_lock(&root->fs_info->trans_mutex);
+       }
+       while(1) {
+               root_objectid = 0;
+               objectid = 0;
+               ret = btrfs_find_del_first_ordered_inode(
+                               &cur_trans->ordered_inode_tree,
+                               &root_objectid, &objectid, &inode);
+               if (!ret)
+                       break;
+               mutex_unlock(&root->fs_info->trans_mutex);
+               mutex_unlock(&root->fs_info->fs_mutex);
+
+               if (S_ISREG(inode->i_mode)) {
+                       atomic_inc(&BTRFS_I(inode)->ordered_writeback);
+                       filemap_write_and_wait(inode->i_mapping);
+                       atomic_dec(&BTRFS_I(inode)->ordered_writeback);
+               }
+               atomic_dec(&inode->i_count);
+               iput(inode);
+
+               mutex_lock(&root->fs_info->fs_mutex);
+               mutex_lock(&root->fs_info->trans_mutex);
+       }
+       root->fs_info->throttles--;
+       return 0;
+}
+
+static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
+                                  struct btrfs_fs_info *fs_info,
+                                  struct btrfs_pending_snapshot *pending)
+{
+       struct btrfs_key key;
+       struct btrfs_root_item *new_root_item;
+       struct btrfs_root *tree_root = fs_info->tree_root;
+       struct btrfs_root *root = pending->root;
+       struct extent_buffer *tmp;
+       int ret;
+       u64 objectid;
+
+       new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
+       if (!new_root_item) {
+               ret = -ENOMEM;
+               goto fail;
+       }
+       ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
+       if (ret)
+               goto fail;
+
+       memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
+
+       key.objectid = objectid;
+       key.offset = 1;
+       btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
+
+       extent_buffer_get(root->node);
+       btrfs_cow_block(trans, root, root->node, NULL, 0, &tmp);
+       free_extent_buffer(tmp);
+
+       btrfs_copy_root(trans, root, root->node, &tmp, objectid);
+
+       btrfs_set_root_bytenr(new_root_item, tmp->start);
+       btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
+       ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
+                               new_root_item);
+       free_extent_buffer(tmp);
+       if (ret)
+               goto fail;
+
+       /*
+        * insert the directory item
+        */
+       key.offset = (u64)-1;
+       ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
+                                   pending->name, strlen(pending->name),
+                                   root->fs_info->sb->s_root->d_inode->i_ino,
+                                   &key, BTRFS_FT_DIR);
+
+       if (ret)
+               goto fail;
+
+       ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
+                            pending->name, strlen(pending->name), objectid,
+                            root->fs_info->sb->s_root->d_inode->i_ino);
+fail:
+       kfree(new_root_item);
+       return ret;
+}
+
+static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
+                                            struct btrfs_fs_info *fs_info)
+{
+       struct btrfs_pending_snapshot *pending;
+       struct list_head *head = &trans->transaction->pending_snapshots;
+       int ret;
+
+       while(!list_empty(head)) {
+               pending = list_entry(head->next,
+                                    struct btrfs_pending_snapshot, list);
+               ret = create_pending_snapshot(trans, fs_info, pending);
+               BUG_ON(ret);
+               list_del(&pending->list);
+               kfree(pending->name);
+               kfree(pending);
        }
        return 0;
 }
@@ -238,11 +631,15 @@ int drop_dirty_roots(struct btrfs_root *tree_root, struct list_head *list)
 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
                             struct btrfs_root *root)
 {
-       int ret = 0;
+       unsigned long joined = 0;
+       unsigned long timeout = 1;
        struct btrfs_transaction *cur_trans;
        struct btrfs_transaction *prev_trans = NULL;
+       struct btrfs_root *chunk_root = root->fs_info->chunk_root;
        struct list_head dirty_fs_roots;
+       struct extent_io_tree *pinned_copy;
        DEFINE_WAIT(wait);
+       int ret;
 
        INIT_LIST_HEAD(&dirty_fs_roots);
 
@@ -250,66 +647,213 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
        if (trans->transaction->in_commit) {
                cur_trans = trans->transaction;
                trans->transaction->use_count++;
+               mutex_unlock(&root->fs_info->trans_mutex);
                btrfs_end_transaction(trans, root);
+
+               mutex_unlock(&root->fs_info->fs_mutex);
                ret = wait_for_commit(root, cur_trans);
                BUG_ON(ret);
+
+               mutex_lock(&root->fs_info->trans_mutex);
                put_transaction(cur_trans);
                mutex_unlock(&root->fs_info->trans_mutex);
+
+               mutex_lock(&root->fs_info->fs_mutex);
                return 0;
        }
-       cur_trans = trans->transaction;
+
+       pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
+       if (!pinned_copy)
+               return -ENOMEM;
+
+       extent_io_tree_init(pinned_copy,
+                            root->fs_info->btree_inode->i_mapping, GFP_NOFS);
+
        trans->transaction->in_commit = 1;
-       while (trans->transaction->num_writers > 1) {
+       cur_trans = trans->transaction;
+       if (cur_trans->list.prev != &root->fs_info->trans_list) {
+               prev_trans = list_entry(cur_trans->list.prev,
+                                       struct btrfs_transaction, list);
+               if (!prev_trans->commit_done) {
+                       prev_trans->use_count++;
+                       mutex_unlock(&root->fs_info->fs_mutex);
+                       mutex_unlock(&root->fs_info->trans_mutex);
+
+                       wait_for_commit(root, prev_trans);
+
+                       mutex_lock(&root->fs_info->fs_mutex);
+                       mutex_lock(&root->fs_info->trans_mutex);
+                       put_transaction(prev_trans);
+               }
+       }
+
+       do {
+               joined = cur_trans->num_joined;
                WARN_ON(cur_trans != trans->transaction);
-               prepare_to_wait(&trans->transaction->writer_wait, &wait,
+               prepare_to_wait(&cur_trans->writer_wait, &wait,
                                TASK_UNINTERRUPTIBLE);
-               if (trans->transaction->num_writers <= 1)
-                       break;
+
+               if (cur_trans->num_writers > 1)
+                       timeout = MAX_SCHEDULE_TIMEOUT;
+               else
+                       timeout = 1;
+
+               mutex_unlock(&root->fs_info->fs_mutex);
                mutex_unlock(&root->fs_info->trans_mutex);
-               schedule();
+
+               schedule_timeout(timeout);
+
+               mutex_lock(&root->fs_info->fs_mutex);
                mutex_lock(&root->fs_info->trans_mutex);
-               finish_wait(&trans->transaction->writer_wait, &wait);
-       }
-       finish_wait(&trans->transaction->writer_wait, &wait);
+               finish_wait(&cur_trans->writer_wait, &wait);
+               ret = btrfs_write_ordered_inodes(trans, root);
+
+       } while (cur_trans->num_writers > 1 ||
+                (cur_trans->num_joined != joined));
+
+       ret = create_pending_snapshots(trans, root->fs_info);
+       BUG_ON(ret);
+
        WARN_ON(cur_trans != trans->transaction);
-       add_dirty_roots(trans, &root->fs_info->fs_roots_radix, &dirty_fs_roots);
+
+       ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
+                             &dirty_fs_roots);
+       BUG_ON(ret);
+
        ret = btrfs_commit_tree_roots(trans, root);
        BUG_ON(ret);
+
        cur_trans = root->fs_info->running_transaction;
+       spin_lock(&root->fs_info->new_trans_lock);
        root->fs_info->running_transaction = NULL;
-       if (cur_trans->list.prev != &root->fs_info->trans_list) {
-               prev_trans = list_entry(cur_trans->list.prev,
-                                       struct btrfs_transaction, list);
-               if (prev_trans->commit_done)
-                       prev_trans = NULL;
-               else
-                       prev_trans->use_count++;
-       }
+       spin_unlock(&root->fs_info->new_trans_lock);
+       btrfs_set_super_generation(&root->fs_info->super_copy,
+                                  cur_trans->transid);
+       btrfs_set_super_root(&root->fs_info->super_copy,
+                            root->fs_info->tree_root->node->start);
+       btrfs_set_super_root_level(&root->fs_info->super_copy,
+                          btrfs_header_level(root->fs_info->tree_root->node));
+
+       btrfs_set_super_chunk_root(&root->fs_info->super_copy,
+                                  chunk_root->node->start);
+       btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
+                                        btrfs_header_level(chunk_root->node));
+       write_extent_buffer(root->fs_info->sb_buffer,
+                           &root->fs_info->super_copy, 0,
+                           sizeof(root->fs_info->super_copy));
+
+       btrfs_copy_pinned(root, pinned_copy);
+
        mutex_unlock(&root->fs_info->trans_mutex);
        mutex_unlock(&root->fs_info->fs_mutex);
        ret = btrfs_write_and_wait_transaction(trans, root);
-       if (prev_trans) {
-               mutex_lock(&root->fs_info->trans_mutex);
-               wait_for_commit(root, prev_trans);
-               put_transaction(prev_trans);
-               mutex_unlock(&root->fs_info->trans_mutex);
-       }
-       btrfs_set_super_generation(root->fs_info->disk_super,
-                                  cur_trans->transid);
        BUG_ON(ret);
        write_ctree_super(trans, root);
 
        mutex_lock(&root->fs_info->fs_mutex);
-       btrfs_finish_extent_commit(trans, root);
+       btrfs_finish_extent_commit(trans, root, pinned_copy);
        mutex_lock(&root->fs_info->trans_mutex);
+
+       kfree(pinned_copy);
+
        cur_trans->commit_done = 1;
+       root->fs_info->last_trans_committed = cur_trans->transid;
        wake_up(&cur_trans->commit_wait);
        put_transaction(cur_trans);
        put_transaction(cur_trans);
+
+       if (root->fs_info->closing)
+               list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
+       else
+               list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
+
        mutex_unlock(&root->fs_info->trans_mutex);
        kmem_cache_free(btrfs_trans_handle_cachep, trans);
 
-       drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
+       if (root->fs_info->closing) {
+               mutex_unlock(&root->fs_info->fs_mutex);
+               drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
+               mutex_lock(&root->fs_info->fs_mutex);
+       }
        return ret;
 }
 
+int btrfs_clean_old_snapshots(struct btrfs_root *root)
+{
+       struct list_head dirty_roots;
+       INIT_LIST_HEAD(&dirty_roots);
+
+       mutex_lock(&root->fs_info->trans_mutex);
+       list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
+       mutex_unlock(&root->fs_info->trans_mutex);
+
+       if (!list_empty(&dirty_roots)) {
+               drop_dirty_roots(root, &dirty_roots);
+       }
+       return 0;
+}
+#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
+void btrfs_transaction_cleaner(void *p)
+#else
+void btrfs_transaction_cleaner(struct work_struct *work)
+#endif
+{
+#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
+       struct btrfs_fs_info *fs_info = p;
+#else
+       struct btrfs_fs_info *fs_info = container_of(work,
+                                                    struct btrfs_fs_info,
+                                                    trans_work.work);
+
+#endif
+       struct btrfs_root *root = fs_info->tree_root;
+       struct btrfs_transaction *cur;
+       struct btrfs_trans_handle *trans;
+       unsigned long now;
+       unsigned long delay = HZ * 30;
+       int ret;
+
+       mutex_lock(&root->fs_info->fs_mutex);
+       mutex_lock(&root->fs_info->trans_mutex);
+       cur = root->fs_info->running_transaction;
+       if (!cur) {
+               mutex_unlock(&root->fs_info->trans_mutex);
+               goto out;
+       }
+       now = get_seconds();
+       if (now < cur->start_time || now - cur->start_time < 30) {
+               mutex_unlock(&root->fs_info->trans_mutex);
+               delay = HZ * 5;
+               goto out;
+       }
+       mutex_unlock(&root->fs_info->trans_mutex);
+       btrfs_defrag_dirty_roots(root->fs_info);
+       trans = btrfs_start_transaction(root, 1);
+       ret = btrfs_commit_transaction(trans, root);
+out:
+       mutex_unlock(&root->fs_info->fs_mutex);
+       btrfs_clean_old_snapshots(root);
+       btrfs_transaction_queue_work(root, delay);
+}
+
+void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
+{
+       queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
+}
+
+void btrfs_transaction_flush_work(struct btrfs_root *root)
+{
+       cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work);
+       flush_workqueue(trans_wq);
+}
+
+void __init btrfs_init_transaction_sys(void)
+{
+       trans_wq = create_workqueue("btrfs-transaction");
+}
+
+void btrfs_exit_transaction_sys(void)
+{
+       destroy_workqueue(trans_wq);
+}
+