[linux-2.6-block.git] / fs / btrfs / transaction.c

/*
 * 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 "ctree.h"
#include "disk-io.h"
#include "transaction.h"

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

static void put_transaction(struct btrfs_transaction *transaction)
{
	WARN_ON(transaction->use_count == 0);
	transaction->use_count--;
	if (transaction->use_count == 0) {
		WARN_ON(total_trans == 0);
		total_trans--;
		list_del_init(&transaction->list);
		memset(transaction, 0, sizeof(*transaction));
		kmem_cache_free(btrfs_transaction_cachep, transaction);
	}
}

static int join_transaction(struct btrfs_root *root)
{
	struct btrfs_transaction *cur_trans;
	cur_trans = root->fs_info->running_transaction;
	if (!cur_trans) {
		cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
					     GFP_NOFS);
		total_trans++;
		BUG_ON(!cur_trans);
		root->fs_info->generation++;
		root->fs_info->running_transaction = cur_trans;
		cur_trans->num_writers = 0;
		cur_trans->transid = root->fs_info->generation;
		init_waitqueue_head(&cur_trans->writer_wait);
		init_waitqueue_head(&cur_trans->commit_wait);
		cur_trans->in_commit = 0;
		cur_trans->use_count = 1;
		cur_trans->commit_done = 0;
		cur_trans->start_time = get_seconds();
		list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
		init_bit_radix(&cur_trans->dirty_pages);
	}
	cur_trans->num_writers++;
	return 0;
}

struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
						   int num_blocks)
{
	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) {
		WARN_ON(root == root->fs_info->extent_root);
		WARN_ON(root->ref_cows != 1);
		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);
			root->commit_root = root->node;
			get_bh(root->node);
		} else {
			WARN_ON(1);
		}
	}
	root->last_trans = running_trans_id;
	h->transid = running_trans_id;
	h->transaction = root->fs_info->running_transaction;
	h->blocks_reserved = num_blocks;
	h->blocks_used = 0;
	h->block_group = NULL;
	root->fs_info->running_transaction->use_count++;
	mutex_unlock(&root->fs_info->trans_mutex);
	return h;
}

int btrfs_end_transaction(struct btrfs_trans_handle *trans,
			  struct btrfs_root *root)
{
	struct btrfs_transaction *cur_trans;

	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);
	put_transaction(cur_trans);
	mutex_unlock(&root->fs_info->trans_mutex);
	memset(trans, 0, sizeof(*trans));
	kmem_cache_free(btrfs_trans_handle_cachep, trans);
	return 0;
}


int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
				     struct btrfs_root *root)
{
	unsigned long gang[16];
	int ret;
	int i;
	int err;
	int werr = 0;
	struct page *page;
	struct radix_tree_root *dirty_pages;
	struct inode *btree_inode = root->fs_info->btree_inode;

	if (!trans || !trans->transaction) {
		return filemap_write_and_wait(btree_inode->i_mapping);
	}
	dirty_pages = &trans->transaction->dirty_pages;
	while(1) {
		ret = find_first_radix_bit(dirty_pages, gang,
					   0, ARRAY_SIZE(gang));
		if (!ret)
			break;
		for (i = 0; i < ret; i++) {
			/* FIXME EIO */
			clear_radix_bit(dirty_pages, gang[i]);
			page = find_lock_page(btree_inode->i_mapping,
					      gang[i]);
			if (!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)
{
	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;

	btrfs_write_dirty_block_groups(trans, extent_root);
	while(1) {
		old_extent_block = btrfs_root_blocknr(&extent_root->root_item);
		if (old_extent_block == bh_blocknr(extent_root->node))
			break;
		btrfs_set_root_blocknr(&extent_root->root_item,
				       bh_blocknr(extent_root->node));
		ret = btrfs_update_root(trans, tree_root,
					&extent_root->root_key,
					&extent_root->root_item);
		BUG_ON(ret);
		btrfs_write_dirty_block_groups(trans, extent_root);
	}
	return 0;
}

static 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);
		if (commit->commit_done)
			break;
		mutex_unlock(&root->fs_info->trans_mutex);
		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_root *root;
};

int btrfs_add_dead_root(struct btrfs_root *root, struct list_head *dead_list)
{
	struct dirty_root *dirty;

	dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
	if (!dirty)
		return -ENOMEM;
	dirty->root = root;
	list_add(&dirty->list, dead_list);
	return 0;
}

static 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 = 0;
	u32 refs;

	while(1) {
		ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
						 ARRAY_SIZE(gang),
						 BTRFS_ROOT_TRANS_TAG);
		if (ret == 0)
			break;
		for (i = 0; i < ret; i++) {
			root = gang[i];
			radix_tree_tag_clear(radix,
				     (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);
				root->commit_root = NULL;
				continue;
			}
			dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
			BUG_ON(!dirty);
			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;
			root->commit_root = NULL;

			root->root_key.offset = root->fs_info->generation;
			btrfs_set_root_blocknr(&root->root_item,
					       bh_blocknr(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);
			if (refs == 1) {
				list_add(&dirty->list, list);
			} else {
				WARN_ON(1);
				kfree(dirty->root);
				kfree(dirty);
			}
		}
	}
	return err;
}

static int drop_dirty_roots(struct btrfs_root *tree_root,
			    struct list_head *list)
{
	struct dirty_root *dirty;
	struct btrfs_trans_handle *trans;
	int ret = 0;
	int err;

	while(!list_empty(list)) {
		mutex_lock(&tree_root->fs_info->fs_mutex);
		dirty = list_entry(list->next, struct dirty_root, list);
		list_del_init(&dirty->list);

		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;
			ret = btrfs_end_transaction(trans, tree_root);
			BUG_ON(ret);
		}
		BUG_ON(ret);
		ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
		if (ret)
			break;
		ret = btrfs_end_transaction(trans, tree_root);
		BUG_ON(ret);

		kfree(dirty->root);
		kfree(dirty);
		mutex_unlock(&tree_root->fs_info->fs_mutex);
		btrfs_btree_balance_dirty(tree_root);
	}
	return ret;
}

int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
			     struct btrfs_root *root)
{
	int ret = 0;
	struct btrfs_transaction *cur_trans;
	struct btrfs_transaction *prev_trans = NULL;
	struct list_head dirty_fs_roots;
	struct radix_tree_root pinned_copy;
	DEFINE_WAIT(wait);

	init_bit_radix(&pinned_copy);
	INIT_LIST_HEAD(&dirty_fs_roots);

	mutex_lock(&root->fs_info->trans_mutex);
	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);
		put_transaction(cur_trans);
		mutex_lock(&root->fs_info->fs_mutex);
		return 0;
	}
	trans->transaction->in_commit = 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);
			put_transaction(prev_trans);

			mutex_lock(&root->fs_info->fs_mutex);
			mutex_lock(&root->fs_info->trans_mutex);
		}
	}
	while (trans->transaction->num_writers > 1) {
		WARN_ON(cur_trans != trans->transaction);
		prepare_to_wait(&trans->transaction->writer_wait, &wait,
				TASK_UNINTERRUPTIBLE);
		if (trans->transaction->num_writers <= 1)
			break;
		mutex_unlock(&root->fs_info->fs_mutex);
		mutex_unlock(&root->fs_info->trans_mutex);
		schedule();
		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);
	WARN_ON(cur_trans != trans->transaction);
	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;
	root->fs_info->running_transaction = NULL;
	btrfs_set_super_generation(&root->fs_info->super_copy,
				   cur_trans->transid);
	btrfs_set_super_root(&root->fs_info->super_copy,
			     bh_blocknr(root->fs_info->tree_root->node));
	memcpy(root->fs_info->disk_super, &root->fs_info->super_copy,
	       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);
	BUG_ON(ret);
	write_ctree_super(trans, root);
	mutex_lock(&root->fs_info->fs_mutex);
	btrfs_finish_extent_commit(trans, root, &pinned_copy);
	mutex_lock(&root->fs_info->trans_mutex);
	cur_trans->commit_done = 1;
	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);

	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;
}

void btrfs_transaction_cleaner(struct work_struct *work)
{
	struct btrfs_fs_info *fs_info = container_of(work,
						     struct btrfs_fs_info,
						     trans_work.work);

	struct btrfs_root *root = fs_info->tree_root;
	struct btrfs_transaction *cur;
	struct btrfs_trans_handle *trans;
	struct list_head dirty_roots;
	unsigned long now;
	unsigned long delay = HZ * 30;
	int ret;

	INIT_LIST_HEAD(&dirty_roots);
	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);
	trans = btrfs_start_transaction(root, 1);
	ret = btrfs_commit_transaction(trans, root);
out:
	mutex_unlock(&root->fs_info->fs_mutex);

	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);
	}
	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");
}

void __exit btrfs_exit_transaction_sys(void)
{
	destroy_workqueue(trans_wq);
}
Commit	Line	Data
6cbd5570 CM	1	/*
	2	* Copyright (C) 2007 Oracle. All rights reserved.
	3	*
	4	* This program is free software; you can redistribute it and/or
	5	* modify it under the terms of the GNU General Public
	6	* License v2 as published by the Free Software Foundation.
	7	*
	8	* This program is distributed in the hope that it will be useful,
	9	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	10	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	11	* General Public License for more details.
	12	*
	13	* You should have received a copy of the GNU General Public
	14	* License along with this program; if not, write to the
	15	* Free Software Foundation, Inc., 59 Temple Place - Suite 330,
	16	* Boston, MA 021110-1307, USA.
	17	*/
	18
79154b1b	19	#include <linux/fs.h>
34088780	20	#include <linux/sched.h>
79154b1b CM	21	#include "ctree.h"
	22	#include "disk-io.h"
	23	#include "transaction.h"
	24
78fae27e	25	static int total_trans = 0;
2c90e5d6 CM	26	extern struct kmem_cache *btrfs_trans_handle_cachep;
	27	extern struct kmem_cache *btrfs_transaction_cachep;
	28
08607c1b CM	29	static struct workqueue_struct *trans_wq;
08607c1b CM	30
0f7d52f4 CM	31	#define BTRFS_ROOT_TRANS_TAG 0
0f7d52f4 CM	32
79154b1b CM	33	static void put_transaction(struct btrfs_transaction *transaction)
79154b1b CM	34	{
2c90e5d6	35	WARN_ON(transaction->use_count == 0);
79154b1b	36	transaction->use_count--;
78fae27e CM	37	if (transaction->use_count == 0) {
	38	WARN_ON(total_trans == 0);
	39	total_trans--;
8fd17795	40	list_del_init(&transaction->list);
2c90e5d6 CM	41	memset(transaction, 0, sizeof(*transaction));
2c90e5d6 CM	42	kmem_cache_free(btrfs_transaction_cachep, transaction);
78fae27e	43	}
79154b1b CM	44	}
	45
	46	static int join_transaction(struct btrfs_root *root)
	47	{
	48	struct btrfs_transaction *cur_trans;
	49	cur_trans = root->fs_info->running_transaction;
	50	if (!cur_trans) {
2c90e5d6 CM	51	cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
2c90e5d6 CM	52	GFP_NOFS);
78fae27e	53	total_trans++;
79154b1b	54	BUG_ON(!cur_trans);
0f7d52f4	55	root->fs_info->generation++;
79154b1b CM	56	root->fs_info->running_transaction = cur_trans;
79154b1b CM	57	cur_trans->num_writers = 0;
0f7d52f4	58	cur_trans->transid = root->fs_info->generation;
79154b1b CM	59	init_waitqueue_head(&cur_trans->writer_wait);
	60	init_waitqueue_head(&cur_trans->commit_wait);
	61	cur_trans->in_commit = 0;
d5719762	62	cur_trans->use_count = 1;
79154b1b	63	cur_trans->commit_done = 0;
08607c1b	64	cur_trans->start_time = get_seconds();
8fd17795	65	list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
7c4452b9	66	init_bit_radix(&cur_trans->dirty_pages);
79154b1b CM	67	}
	68	cur_trans->num_writers++;
	69	return 0;
	70	}
	71
	72	struct btrfs_trans_handle btrfs_start_transaction(struct btrfs_root root,
	73	int num_blocks)
	74	{
2c90e5d6 CM	75	struct btrfs_trans_handle *h =
2c90e5d6 CM	76	kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
79154b1b	77	int ret;
0f7d52f4	78	u64 running_trans_id;
79154b1b CM	79
	80	mutex_lock(&root->fs_info->trans_mutex);
	81	ret = join_transaction(root);
	82	BUG_ON(ret);
0f7d52f4 CM	83	running_trans_id = root->fs_info->running_transaction->transid;
	84
	85	if (root != root->fs_info->tree_root && root->last_trans <
	86	running_trans_id) {
ccd467d6 CM	87	WARN_ON(root == root->fs_info->extent_root);
ccd467d6 CM	88	WARN_ON(root->ref_cows != 1);
5eda7b5e CM	89	if (root->root_item.refs != 0) {
	90	radix_tree_tag_set(&root->fs_info->fs_roots_radix,
	91	(unsigned long)root->root_key.objectid,
	92	BTRFS_ROOT_TRANS_TAG);
	93	root->commit_root = root->node;
	94	get_bh(root->node);
	95	} else {
	96	WARN_ON(1);
	97	}
0f7d52f4 CM	98	}
	99	root->last_trans = running_trans_id;
	100	h->transid = running_trans_id;
79154b1b CM	101	h->transaction = root->fs_info->running_transaction;
	102	h->blocks_reserved = num_blocks;
	103	h->blocks_used = 0;
31f3c99b	104	h->block_group = NULL;
79154b1b CM	105	root->fs_info->running_transaction->use_count++;
	106	mutex_unlock(&root->fs_info->trans_mutex);
	107	return h;
	108	}
	109
	110	int btrfs_end_transaction(struct btrfs_trans_handle *trans,
	111	struct btrfs_root *root)
	112	{
	113	struct btrfs_transaction *cur_trans;
d6e4a428	114
79154b1b CM	115	mutex_lock(&root->fs_info->trans_mutex);
79154b1b CM	116	cur_trans = root->fs_info->running_transaction;
ccd467d6	117	WARN_ON(cur_trans != trans->transaction);
d5719762	118	WARN_ON(cur_trans->num_writers < 1);
ccd467d6	119	cur_trans->num_writers--;
79154b1b CM	120	if (waitqueue_active(&cur_trans->writer_wait))
79154b1b CM	121	wake_up(&cur_trans->writer_wait);
79154b1b CM	122	put_transaction(cur_trans);
79154b1b CM	123	mutex_unlock(&root->fs_info->trans_mutex);
d6025579	124	memset(trans, 0, sizeof(*trans));
2c90e5d6	125	kmem_cache_free(btrfs_trans_handle_cachep, trans);
79154b1b CM	126	return 0;
	127	}
	128
	129
	130	int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
	131	struct btrfs_root *root)
	132	{
7c4452b9 CM	133	unsigned long gang[16];
	134	int ret;
	135	int i;
	136	int err;
	137	int werr = 0;
	138	struct page *page;
	139	struct radix_tree_root *dirty_pages;
	140	struct inode *btree_inode = root->fs_info->btree_inode;
	141
	142	if (!trans \|\| !trans->transaction) {
	143	return filemap_write_and_wait(btree_inode->i_mapping);
	144	}
	145	dirty_pages = &trans->transaction->dirty_pages;
	146	while(1) {
e37c9e69 CM	147	ret = find_first_radix_bit(dirty_pages, gang,
e37c9e69 CM	148	0, ARRAY_SIZE(gang));
7c4452b9 CM	149	if (!ret)
	150	break;
	151	for (i = 0; i < ret; i++) {
	152	/* FIXME EIO */
	153	clear_radix_bit(dirty_pages, gang[i]);
	154	page = find_lock_page(btree_inode->i_mapping,
	155	gang[i]);
	156	if (!page)
	157	continue;
	158	err = write_one_page(page, 0);
	159	if (err)
	160	werr = err;
	161	page_cache_release(page);
	162	}
	163	}
	164	err = filemap_fdatawait(btree_inode->i_mapping);
	165	if (err)
	166	werr = err;
	167	return werr;
79154b1b CM	168	}
	169
	170	int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
	171	struct btrfs_root *root)
	172	{
	173	int ret;
	174	u64 old_extent_block;
	175	struct btrfs_fs_info *fs_info = root->fs_info;
	176	struct btrfs_root *tree_root = fs_info->tree_root;
	177	struct btrfs_root *extent_root = fs_info->extent_root;
79154b1b	178
9078a3e1	179	btrfs_write_dirty_block_groups(trans, extent_root);
79154b1b CM	180	while(1) {
79154b1b CM	181	old_extent_block = btrfs_root_blocknr(&extent_root->root_item);
7eccb903	182	if (old_extent_block == bh_blocknr(extent_root->node))
79154b1b CM	183	break;
79154b1b CM	184	btrfs_set_root_blocknr(&extent_root->root_item,
7eccb903	185	bh_blocknr(extent_root->node));
79154b1b CM	186	ret = btrfs_update_root(trans, tree_root,
	187	&extent_root->root_key,
	188	&extent_root->root_item);
	189	BUG_ON(ret);
9078a3e1	190	btrfs_write_dirty_block_groups(trans, extent_root);
79154b1b CM	191	}
	192	return 0;
	193	}
	194
	195	static int wait_for_commit(struct btrfs_root *root,
	196	struct btrfs_transaction *commit)
	197	{
	198	DEFINE_WAIT(wait);
ccd467d6	199	mutex_lock(&root->fs_info->trans_mutex);
79154b1b CM	200	while(!commit->commit_done) {
	201	prepare_to_wait(&commit->commit_wait, &wait,
	202	TASK_UNINTERRUPTIBLE);
	203	if (commit->commit_done)
	204	break;
	205	mutex_unlock(&root->fs_info->trans_mutex);
	206	schedule();
	207	mutex_lock(&root->fs_info->trans_mutex);
	208	}
ccd467d6	209	mutex_unlock(&root->fs_info->trans_mutex);
79154b1b CM	210	finish_wait(&commit->commit_wait, &wait);
	211	return 0;
	212	}
	213
0f7d52f4 CM	214	struct dirty_root {
0f7d52f4 CM	215	struct list_head list;
0f7d52f4 CM	216	struct btrfs_root *root;
	217	};
	218
5eda7b5e CM	219	int btrfs_add_dead_root(struct btrfs_root root, struct list_head dead_list)
	220	{
	221	struct dirty_root *dirty;
	222
	223	dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
	224	if (!dirty)
	225	return -ENOMEM;
5eda7b5e	226	dirty->root = root;
5eda7b5e CM	227	list_add(&dirty->list, dead_list);
	228	return 0;
	229	}
	230
35b7e476 CM	231	static int add_dirty_roots(struct btrfs_trans_handle *trans,
	232	struct radix_tree_root *radix,
	233	struct list_head *list)
0f7d52f4 CM	234	{
	235	struct dirty_root *dirty;
	236	struct btrfs_root *gang[8];
	237	struct btrfs_root *root;
	238	int i;
	239	int ret;
54aa1f4d	240	int err = 0;
5eda7b5e	241	u32 refs;
54aa1f4d	242
0f7d52f4 CM	243	while(1) {
	244	ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
	245	ARRAY_SIZE(gang),
	246	BTRFS_ROOT_TRANS_TAG);
	247	if (ret == 0)
	248	break;
	249	for (i = 0; i < ret; i++) {
	250	root = gang[i];
2619ba1f CM	251	radix_tree_tag_clear(radix,
	252	(unsigned long)root->root_key.objectid,
	253	BTRFS_ROOT_TRANS_TAG);
0f7d52f4	254	if (root->commit_root == root->node) {
7eccb903	255	WARN_ON(bh_blocknr(root->node) !=
0f7d52f4 CM	256	btrfs_root_blocknr(&root->root_item));
	257	brelse(root->commit_root);
	258	root->commit_root = NULL;
	259	continue;
	260	}
	261	dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
	262	BUG_ON(!dirty);
9f3a7427 CM	263	dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
	264	BUG_ON(!dirty->root);
	265
	266	memset(&root->root_item.drop_progress, 0,
	267	sizeof(struct btrfs_disk_key));
	268	root->root_item.drop_level = 0;
	269
	270	memcpy(dirty->root, root, sizeof(*root));
	271	dirty->root->node = root->commit_root;
0f7d52f4	272	root->commit_root = NULL;
5eda7b5e	273
0f7d52f4 CM	274	root->root_key.offset = root->fs_info->generation;
0f7d52f4 CM	275	btrfs_set_root_blocknr(&root->root_item,
7eccb903	276	bh_blocknr(root->node));
0f7d52f4 CM	277	err = btrfs_insert_root(trans, root->fs_info->tree_root,
	278	&root->root_key,
	279	&root->root_item);
54aa1f4d CM	280	if (err)
54aa1f4d CM	281	break;
9f3a7427 CM	282
	283	refs = btrfs_root_refs(&dirty->root->root_item);
	284	btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
5eda7b5e	285	err = btrfs_update_root(trans, root->fs_info->tree_root,
9f3a7427 CM	286	&dirty->root->root_key,
9f3a7427 CM	287	&dirty->root->root_item);
5eda7b5e CM	288
5eda7b5e CM	289	BUG_ON(err);
9f3a7427	290	if (refs == 1) {
5eda7b5e	291	list_add(&dirty->list, list);
9f3a7427 CM	292	} else {
	293	WARN_ON(1);
	294	kfree(dirty->root);
5eda7b5e	295	kfree(dirty);
9f3a7427	296	}
0f7d52f4 CM	297	}
0f7d52f4 CM	298	}
54aa1f4d	299	return err;
0f7d52f4 CM	300	}
0f7d52f4 CM	301
35b7e476 CM	302	static int drop_dirty_roots(struct btrfs_root *tree_root,
35b7e476 CM	303	struct list_head *list)
0f7d52f4 CM	304	{
	305	struct dirty_root *dirty;
	306	struct btrfs_trans_handle *trans;
54aa1f4d	307	int ret = 0;
9f3a7427 CM	308	int err;
9f3a7427 CM	309
0f7d52f4	310	while(!list_empty(list)) {
facda1e7	311	mutex_lock(&tree_root->fs_info->fs_mutex);
0f7d52f4 CM	312	dirty = list_entry(list->next, struct dirty_root, list);
0f7d52f4 CM	313	list_del_init(&dirty->list);
5eda7b5e	314
9f3a7427 CM	315	while(1) {
	316	trans = btrfs_start_transaction(tree_root, 1);
	317	ret = btrfs_drop_snapshot(trans, dirty->root);
	318	if (ret != -EAGAIN) {
	319	break;
	320	}
	321	err = btrfs_update_root(trans,
	322	tree_root,
	323	&dirty->root->root_key,
	324	&dirty->root->root_item);
	325	if (err)
	326	ret = err;
	327	ret = btrfs_end_transaction(trans, tree_root);
	328	BUG_ON(ret);
	329	}
0f7d52f4	330	BUG_ON(ret);
9f3a7427	331	ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
54aa1f4d CM	332	if (ret)
54aa1f4d CM	333	break;
0f7d52f4 CM	334	ret = btrfs_end_transaction(trans, tree_root);
0f7d52f4 CM	335	BUG_ON(ret);
5eda7b5e	336
9f3a7427	337	kfree(dirty->root);
0f7d52f4	338	kfree(dirty);
facda1e7	339	mutex_unlock(&tree_root->fs_info->fs_mutex);
8c2383c3	340	btrfs_btree_balance_dirty(tree_root);
0f7d52f4	341	}
54aa1f4d	342	return ret;
0f7d52f4 CM	343	}
0f7d52f4 CM	344
79154b1b CM	345	int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
	346	struct btrfs_root *root)
	347	{
	348	int ret = 0;
79154b1b	349	struct btrfs_transaction *cur_trans;
8fd17795	350	struct btrfs_transaction *prev_trans = NULL;
0f7d52f4	351	struct list_head dirty_fs_roots;
ccd467d6	352	struct radix_tree_root pinned_copy;
79154b1b CM	353	DEFINE_WAIT(wait);
79154b1b CM	354
ccd467d6	355	init_bit_radix(&pinned_copy);
0f7d52f4	356	INIT_LIST_HEAD(&dirty_fs_roots);
d6e4a428	357
79154b1b CM	358	mutex_lock(&root->fs_info->trans_mutex);
	359	if (trans->transaction->in_commit) {
	360	cur_trans = trans->transaction;
	361	trans->transaction->use_count++;
ccd467d6	362	mutex_unlock(&root->fs_info->trans_mutex);
79154b1b	363	btrfs_end_transaction(trans, root);
ccd467d6 CM	364
ccd467d6 CM	365	mutex_unlock(&root->fs_info->fs_mutex);
79154b1b CM	366	ret = wait_for_commit(root, cur_trans);
	367	BUG_ON(ret);
	368	put_transaction(cur_trans);
ccd467d6	369	mutex_lock(&root->fs_info->fs_mutex);
79154b1b CM	370	return 0;
79154b1b CM	371	}
2c90e5d6	372	trans->transaction->in_commit = 1;
ccd467d6 CM	373	cur_trans = trans->transaction;
	374	if (cur_trans->list.prev != &root->fs_info->trans_list) {
	375	prev_trans = list_entry(cur_trans->list.prev,
	376	struct btrfs_transaction, list);
	377	if (!prev_trans->commit_done) {
	378	prev_trans->use_count++;
	379	mutex_unlock(&root->fs_info->fs_mutex);
	380	mutex_unlock(&root->fs_info->trans_mutex);
	381
	382	wait_for_commit(root, prev_trans);
	383	put_transaction(prev_trans);
	384
	385	mutex_lock(&root->fs_info->fs_mutex);
	386	mutex_lock(&root->fs_info->trans_mutex);
	387	}
	388	}
79154b1b	389	while (trans->transaction->num_writers > 1) {
2c90e5d6	390	WARN_ON(cur_trans != trans->transaction);
79154b1b CM	391	prepare_to_wait(&trans->transaction->writer_wait, &wait,
	392	TASK_UNINTERRUPTIBLE);
	393	if (trans->transaction->num_writers <= 1)
	394	break;
ccd467d6	395	mutex_unlock(&root->fs_info->fs_mutex);
79154b1b CM	396	mutex_unlock(&root->fs_info->trans_mutex);
79154b1b CM	397	schedule();
ccd467d6	398	mutex_lock(&root->fs_info->fs_mutex);
79154b1b	399	mutex_lock(&root->fs_info->trans_mutex);
2c90e5d6	400	finish_wait(&trans->transaction->writer_wait, &wait);
79154b1b CM	401	}
79154b1b CM	402	finish_wait(&trans->transaction->writer_wait, &wait);
2c90e5d6	403	WARN_ON(cur_trans != trans->transaction);
54aa1f4d CM	404	ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
	405	&dirty_fs_roots);
	406	BUG_ON(ret);
	407
79154b1b CM	408	ret = btrfs_commit_tree_roots(trans, root);
79154b1b CM	409	BUG_ON(ret);
54aa1f4d	410
78fae27e CM	411	cur_trans = root->fs_info->running_transaction;
78fae27e CM	412	root->fs_info->running_transaction = NULL;
4b52dff6 CM	413	btrfs_set_super_generation(&root->fs_info->super_copy,
	414	cur_trans->transid);
	415	btrfs_set_super_root(&root->fs_info->super_copy,
	416	bh_blocknr(root->fs_info->tree_root->node));
	417	memcpy(root->fs_info->disk_super, &root->fs_info->super_copy,
	418	sizeof(root->fs_info->super_copy));
ccd467d6 CM	419
	420	btrfs_copy_pinned(root, &pinned_copy);
	421
78fae27e	422	mutex_unlock(&root->fs_info->trans_mutex);
8fd17795	423	mutex_unlock(&root->fs_info->fs_mutex);
79154b1b CM	424	ret = btrfs_write_and_wait_transaction(trans, root);
79154b1b CM	425	BUG_ON(ret);
79154b1b	426	write_ctree_super(trans, root);
8fd17795	427	mutex_lock(&root->fs_info->fs_mutex);
ccd467d6	428	btrfs_finish_extent_commit(trans, root, &pinned_copy);
78fae27e	429	mutex_lock(&root->fs_info->trans_mutex);
2c90e5d6 CM	430	cur_trans->commit_done = 1;
2c90e5d6 CM	431	wake_up(&cur_trans->commit_wait);
78fae27e	432	put_transaction(cur_trans);
79154b1b	433	put_transaction(cur_trans);
facda1e7 CM	434	if (root->fs_info->closing)
	435	list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
	436	else
	437	list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
78fae27e	438	mutex_unlock(&root->fs_info->trans_mutex);
2c90e5d6	439	kmem_cache_free(btrfs_trans_handle_cachep, trans);
79154b1b	440
facda1e7 CM	441	if (root->fs_info->closing) {
	442	mutex_unlock(&root->fs_info->fs_mutex);
	443	drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
	444	mutex_lock(&root->fs_info->fs_mutex);
	445	}
79154b1b CM	446	return ret;
	447	}
	448
08607c1b CM	449	void btrfs_transaction_cleaner(struct work_struct *work)
	450	{
	451	struct btrfs_fs_info *fs_info = container_of(work,
	452	struct btrfs_fs_info,
	453	trans_work.work);
	454
	455	struct btrfs_root *root = fs_info->tree_root;
	456	struct btrfs_transaction *cur;
	457	struct btrfs_trans_handle *trans;
facda1e7	458	struct list_head dirty_roots;
08607c1b CM	459	unsigned long now;
	460	unsigned long delay = HZ * 30;
	461	int ret;
	462
facda1e7	463	INIT_LIST_HEAD(&dirty_roots);
08607c1b CM	464	mutex_lock(&root->fs_info->fs_mutex);
	465	mutex_lock(&root->fs_info->trans_mutex);
	466	cur = root->fs_info->running_transaction;
	467	if (!cur) {
	468	mutex_unlock(&root->fs_info->trans_mutex);
	469	goto out;
	470	}
	471	now = get_seconds();
	472	if (now < cur->start_time \|\| now - cur->start_time < 30) {
	473	mutex_unlock(&root->fs_info->trans_mutex);
	474	delay = HZ * 5;
	475	goto out;
	476	}
	477	mutex_unlock(&root->fs_info->trans_mutex);
08607c1b CM	478	trans = btrfs_start_transaction(root, 1);
	479	ret = btrfs_commit_transaction(trans, root);
	480	out:
	481	mutex_unlock(&root->fs_info->fs_mutex);
ad693af6 CM	482
	483	mutex_lock(&root->fs_info->trans_mutex);
	484	list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
	485	mutex_unlock(&root->fs_info->trans_mutex);
	486
	487	if (!list_empty(&dirty_roots)) {
	488	drop_dirty_roots(root, &dirty_roots);
	489	}
08607c1b CM	490	btrfs_transaction_queue_work(root, delay);
	491	}
	492
	493	void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
	494	{
	495	queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
	496	}
	497
	498	void btrfs_transaction_flush_work(struct btrfs_root *root)
	499	{
	500	cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work);
	501	flush_workqueue(trans_wq);
	502	}
	503
	504	void __init btrfs_init_transaction_sys(void)
	505	{
	506	trans_wq = create_workqueue("btrfs");
	507	}
	508
	509	void __exit btrfs_exit_transaction_sys(void)
	510	{
	511	destroy_workqueue(trans_wq);
	512	}
	513