[mirror_ubuntu-jammy-kernel.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
#define BTRFS_ROOT_DEFRAG_TAG 1

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 = 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->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);
	} else {
		cur_trans->num_writers++;
		cur_trans->num_joined++;
	}

	return 0;
}

static 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;
			get_bh(root->node);
		} else {
			WARN_ON(1);
		}
		root->last_trans = running_trans_id;
	}
	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;

	mutex_lock(&root->fs_info->trans_mutex);
	ret = join_transaction(root);
	BUG_ON(ret);

	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);
	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;
			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)
{
	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;
}

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;

	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);
		btrfs_end_transaction(trans, root);
		mutex_unlock(&info->fs_mutex);

		btrfs_btree_balance_dirty(root);
		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 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 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);
			mutex_unlock(&tree_root->fs_info->fs_mutex);

			btrfs_btree_balance_dirty(tree_root);
			schedule();

			mutex_lock(&tree_root->fs_info->fs_mutex);
		}
		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);
		schedule();
	}
	return ret;
}

int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
			     struct btrfs_root *root)
{
	unsigned long joined = 0;
	unsigned long timeout = 1;
	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);
	int ret;

	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);

		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;
	}
	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);

			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(&cur_trans->writer_wait, &wait,
				TASK_UNINTERRUPTIBLE);

		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_timeout(timeout);

		mutex_lock(&root->fs_info->fs_mutex);
		mutex_lock(&root->fs_info->trans_mutex);
		finish_wait(&cur_trans->writer_wait, &wait);
	} while (cur_trans->num_writers > 1 ||
		 (cur_trans->num_joined != joined));

	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;
	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);

	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;
}
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;
	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");
}

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	31	#define BTRFS_ROOT_TRANS_TAG 0
6702ed49	32	#define BTRFS_ROOT_DEFRAG_TAG 1
0f7d52f4	33
79154b1b CM	34	static void put_transaction(struct btrfs_transaction *transaction)
79154b1b CM	35	{
2c90e5d6	36	WARN_ON(transaction->use_count == 0);
79154b1b	37	transaction->use_count--;
78fae27e CM	38	if (transaction->use_count == 0) {
	39	WARN_ON(total_trans == 0);
	40	total_trans--;
8fd17795	41	list_del_init(&transaction->list);
2c90e5d6 CM	42	memset(transaction, 0, sizeof(*transaction));
2c90e5d6 CM	43	kmem_cache_free(btrfs_transaction_cachep, transaction);
78fae27e	44	}
79154b1b CM	45	}
	46
	47	static int join_transaction(struct btrfs_root *root)
	48	{
	49	struct btrfs_transaction *cur_trans;
	50	cur_trans = root->fs_info->running_transaction;
	51	if (!cur_trans) {
2c90e5d6 CM	52	cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
2c90e5d6 CM	53	GFP_NOFS);
78fae27e	54	total_trans++;
79154b1b	55	BUG_ON(!cur_trans);
0f7d52f4	56	root->fs_info->generation++;
79154b1b	57	root->fs_info->running_transaction = cur_trans;
15ee9bc7 JB	58	cur_trans->num_writers = 1;
15ee9bc7 JB	59	cur_trans->num_joined = 0;
0f7d52f4	60	cur_trans->transid = root->fs_info->generation;
79154b1b CM	61	init_waitqueue_head(&cur_trans->writer_wait);
	62	init_waitqueue_head(&cur_trans->commit_wait);
	63	cur_trans->in_commit = 0;
d5719762	64	cur_trans->use_count = 1;
79154b1b	65	cur_trans->commit_done = 0;
08607c1b	66	cur_trans->start_time = get_seconds();
8fd17795	67	list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
7c4452b9	68	init_bit_radix(&cur_trans->dirty_pages);
15ee9bc7 JB	69	} else {
	70	cur_trans->num_writers++;
	71	cur_trans->num_joined++;
79154b1b	72	}
15ee9bc7	73
79154b1b CM	74	return 0;
	75	}
	76
6702ed49 CM	77	static int record_root_in_trans(struct btrfs_root *root)
	78	{
	79	u64 running_trans_id = root->fs_info->running_transaction->transid;
	80	if (root->ref_cows && root->last_trans < running_trans_id) {
	81	WARN_ON(root == root->fs_info->extent_root);
	82	if (root->root_item.refs != 0) {
	83	radix_tree_tag_set(&root->fs_info->fs_roots_radix,
	84	(unsigned long)root->root_key.objectid,
	85	BTRFS_ROOT_TRANS_TAG);
	86	radix_tree_tag_set(&root->fs_info->fs_roots_radix,
	87	(unsigned long)root->root_key.objectid,
	88	BTRFS_ROOT_DEFRAG_TAG);
	89	root->commit_root = root->node;
	90	get_bh(root->node);
	91	} else {
	92	WARN_ON(1);
	93	}
	94	root->last_trans = running_trans_id;
	95	}
	96	return 0;
	97	}
	98
79154b1b CM	99	struct btrfs_trans_handle btrfs_start_transaction(struct btrfs_root root,
	100	int num_blocks)
	101	{
2c90e5d6 CM	102	struct btrfs_trans_handle *h =
2c90e5d6 CM	103	kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
79154b1b CM	104	int ret;
	105
	106	mutex_lock(&root->fs_info->trans_mutex);
	107	ret = join_transaction(root);
	108	BUG_ON(ret);
0f7d52f4	109
6702ed49 CM	110	record_root_in_trans(root);
6702ed49 CM	111	h->transid = root->fs_info->running_transaction->transid;
79154b1b CM	112	h->transaction = root->fs_info->running_transaction;
	113	h->blocks_reserved = num_blocks;
	114	h->blocks_used = 0;
31f3c99b	115	h->block_group = NULL;
26b8003f CM	116	h->alloc_exclude_nr = 0;
26b8003f CM	117	h->alloc_exclude_start = 0;
79154b1b CM	118	root->fs_info->running_transaction->use_count++;
	119	mutex_unlock(&root->fs_info->trans_mutex);
	120	return h;
	121	}
	122
	123	int btrfs_end_transaction(struct btrfs_trans_handle *trans,
	124	struct btrfs_root *root)
	125	{
	126	struct btrfs_transaction *cur_trans;
d6e4a428	127
79154b1b CM	128	mutex_lock(&root->fs_info->trans_mutex);
79154b1b CM	129	cur_trans = root->fs_info->running_transaction;
ccd467d6	130	WARN_ON(cur_trans != trans->transaction);
d5719762	131	WARN_ON(cur_trans->num_writers < 1);
ccd467d6	132	cur_trans->num_writers--;
79154b1b CM	133	if (waitqueue_active(&cur_trans->writer_wait))
79154b1b CM	134	wake_up(&cur_trans->writer_wait);
79154b1b CM	135	put_transaction(cur_trans);
79154b1b CM	136	mutex_unlock(&root->fs_info->trans_mutex);
d6025579	137	memset(trans, 0, sizeof(*trans));
2c90e5d6	138	kmem_cache_free(btrfs_trans_handle_cachep, trans);
79154b1b CM	139	return 0;
	140	}
	141
	142
	143	int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
	144	struct btrfs_root *root)
	145	{
7c4452b9 CM	146	unsigned long gang[16];
	147	int ret;
	148	int i;
	149	int err;
	150	int werr = 0;
	151	struct page *page;
	152	struct radix_tree_root *dirty_pages;
	153	struct inode *btree_inode = root->fs_info->btree_inode;
	154
	155	if (!trans \|\| !trans->transaction) {
	156	return filemap_write_and_wait(btree_inode->i_mapping);
	157	}
	158	dirty_pages = &trans->transaction->dirty_pages;
	159	while(1) {
e37c9e69 CM	160	ret = find_first_radix_bit(dirty_pages, gang,
e37c9e69 CM	161	0, ARRAY_SIZE(gang));
7c4452b9 CM	162	if (!ret)
	163	break;
	164	for (i = 0; i < ret; i++) {
	165	/* FIXME EIO */
	166	clear_radix_bit(dirty_pages, gang[i]);
	167	page = find_lock_page(btree_inode->i_mapping,
	168	gang[i]);
	169	if (!page)
	170	continue;
6702ed49 CM	171	if (PageWriteback(page)) {
	172	if (PageDirty(page))
	173	wait_on_page_writeback(page);
	174	else {
	175	unlock_page(page);
	176	page_cache_release(page);
	177	continue;
	178	}
	179	}
7c4452b9 CM	180	err = write_one_page(page, 0);
	181	if (err)
	182	werr = err;
	183	page_cache_release(page);
	184	}
	185	}
	186	err = filemap_fdatawait(btree_inode->i_mapping);
	187	if (err)
	188	werr = err;
	189	return werr;
79154b1b CM	190	}
	191
	192	int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
	193	struct btrfs_root *root)
	194	{
	195	int ret;
	196	u64 old_extent_block;
	197	struct btrfs_fs_info *fs_info = root->fs_info;
	198	struct btrfs_root *tree_root = fs_info->tree_root;
	199	struct btrfs_root *extent_root = fs_info->extent_root;
79154b1b	200
9078a3e1	201	btrfs_write_dirty_block_groups(trans, extent_root);
79154b1b CM	202	while(1) {
79154b1b CM	203	old_extent_block = btrfs_root_blocknr(&extent_root->root_item);
7eccb903	204	if (old_extent_block == bh_blocknr(extent_root->node))
79154b1b CM	205	break;
79154b1b CM	206	btrfs_set_root_blocknr(&extent_root->root_item,
7eccb903	207	bh_blocknr(extent_root->node));
79154b1b CM	208	ret = btrfs_update_root(trans, tree_root,
	209	&extent_root->root_key,
	210	&extent_root->root_item);
	211	BUG_ON(ret);
9078a3e1	212	btrfs_write_dirty_block_groups(trans, extent_root);
79154b1b CM	213	}
	214	return 0;
	215	}
	216
	217	static int wait_for_commit(struct btrfs_root *root,
	218	struct btrfs_transaction *commit)
	219	{
	220	DEFINE_WAIT(wait);
ccd467d6	221	mutex_lock(&root->fs_info->trans_mutex);
79154b1b CM	222	while(!commit->commit_done) {
	223	prepare_to_wait(&commit->commit_wait, &wait,
	224	TASK_UNINTERRUPTIBLE);
	225	if (commit->commit_done)
	226	break;
	227	mutex_unlock(&root->fs_info->trans_mutex);
	228	schedule();
	229	mutex_lock(&root->fs_info->trans_mutex);
	230	}
ccd467d6	231	mutex_unlock(&root->fs_info->trans_mutex);
79154b1b CM	232	finish_wait(&commit->commit_wait, &wait);
	233	return 0;
	234	}
	235
0f7d52f4 CM	236	struct dirty_root {
0f7d52f4 CM	237	struct list_head list;
0f7d52f4 CM	238	struct btrfs_root *root;
	239	};
	240
5eda7b5e CM	241	int btrfs_add_dead_root(struct btrfs_root root, struct list_head dead_list)
	242	{
	243	struct dirty_root *dirty;
	244
	245	dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
	246	if (!dirty)
	247	return -ENOMEM;
5eda7b5e	248	dirty->root = root;
5eda7b5e CM	249	list_add(&dirty->list, dead_list);
	250	return 0;
	251	}
	252
35b7e476 CM	253	static int add_dirty_roots(struct btrfs_trans_handle *trans,
	254	struct radix_tree_root *radix,
	255	struct list_head *list)
0f7d52f4 CM	256	{
	257	struct dirty_root *dirty;
	258	struct btrfs_root *gang[8];
	259	struct btrfs_root *root;
	260	int i;
	261	int ret;
54aa1f4d	262	int err = 0;
5eda7b5e	263	u32 refs;
54aa1f4d	264
0f7d52f4 CM	265	while(1) {
	266	ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
	267	ARRAY_SIZE(gang),
	268	BTRFS_ROOT_TRANS_TAG);
	269	if (ret == 0)
	270	break;
	271	for (i = 0; i < ret; i++) {
	272	root = gang[i];
2619ba1f CM	273	radix_tree_tag_clear(radix,
	274	(unsigned long)root->root_key.objectid,
	275	BTRFS_ROOT_TRANS_TAG);
0f7d52f4	276	if (root->commit_root == root->node) {
7eccb903	277	WARN_ON(bh_blocknr(root->node) !=
0f7d52f4 CM	278	btrfs_root_blocknr(&root->root_item));
	279	brelse(root->commit_root);
	280	root->commit_root = NULL;
	281	continue;
	282	}
	283	dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
	284	BUG_ON(!dirty);
9f3a7427 CM	285	dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
	286	BUG_ON(!dirty->root);
	287
	288	memset(&root->root_item.drop_progress, 0,
	289	sizeof(struct btrfs_disk_key));
	290	root->root_item.drop_level = 0;
	291
	292	memcpy(dirty->root, root, sizeof(*root));
	293	dirty->root->node = root->commit_root;
0f7d52f4	294	root->commit_root = NULL;
5eda7b5e	295
0f7d52f4 CM	296	root->root_key.offset = root->fs_info->generation;
0f7d52f4 CM	297	btrfs_set_root_blocknr(&root->root_item,
7eccb903	298	bh_blocknr(root->node));
0f7d52f4 CM	299	err = btrfs_insert_root(trans, root->fs_info->tree_root,
	300	&root->root_key,
	301	&root->root_item);
54aa1f4d CM	302	if (err)
54aa1f4d CM	303	break;
9f3a7427 CM	304
	305	refs = btrfs_root_refs(&dirty->root->root_item);
	306	btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
5eda7b5e	307	err = btrfs_update_root(trans, root->fs_info->tree_root,
9f3a7427 CM	308	&dirty->root->root_key,
9f3a7427 CM	309	&dirty->root->root_item);
5eda7b5e CM	310
5eda7b5e CM	311	BUG_ON(err);
9f3a7427	312	if (refs == 1) {
5eda7b5e	313	list_add(&dirty->list, list);
9f3a7427 CM	314	} else {
	315	WARN_ON(1);
	316	kfree(dirty->root);
5eda7b5e	317	kfree(dirty);
9f3a7427	318	}
0f7d52f4 CM	319	}
0f7d52f4 CM	320	}
54aa1f4d	321	return err;
0f7d52f4 CM	322	}
0f7d52f4 CM	323
e9d0b13b CM	324	int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
	325	{
	326	struct btrfs_fs_info *info = root->fs_info;
	327	int ret;
	328	struct btrfs_trans_handle *trans;
	329
	330	if (root->defrag_running)
	331	return 0;
	332
	333	trans = btrfs_start_transaction(root, 1);
	334	while (1) {
	335	root->defrag_running = 1;
	336	ret = btrfs_defrag_leaves(trans, root, cacheonly);
	337	btrfs_end_transaction(trans, root);
	338	mutex_unlock(&info->fs_mutex);
	339
	340	btrfs_btree_balance_dirty(root);
	341	cond_resched();
	342
	343	mutex_lock(&info->fs_mutex);
	344	trans = btrfs_start_transaction(root, 1);
	345	if (ret != -EAGAIN)
	346	break;
	347	}
	348	root->defrag_running = 0;
	349	radix_tree_tag_clear(&info->fs_roots_radix,
	350	(unsigned long)root->root_key.objectid,
	351	BTRFS_ROOT_DEFRAG_TAG);
	352	btrfs_end_transaction(trans, root);
	353	return 0;
	354	}
	355
6702ed49 CM	356	int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
	357	{
	358	struct btrfs_root *gang[1];
	359	struct btrfs_root *root;
6702ed49 CM	360	int i;
	361	int ret;
	362	int err = 0;
	363	u64 last = 0;
	364
6702ed49 CM	365	while(1) {
	366	ret = radix_tree_gang_lookup_tag(&info->fs_roots_radix,
	367	(void **)gang, last,
	368	ARRAY_SIZE(gang),
	369	BTRFS_ROOT_DEFRAG_TAG);
	370	if (ret == 0)
	371	break;
	372	for (i = 0; i < ret; i++) {
	373	root = gang[i];
	374	last = root->root_key.objectid + 1;
e9d0b13b	375	btrfs_defrag_root(root, 1);
6702ed49 CM	376	}
6702ed49 CM	377	}
e9d0b13b	378	btrfs_defrag_root(info->extent_root, 1);
6702ed49 CM	379	return err;
	380	}
	381
35b7e476 CM	382	static int drop_dirty_roots(struct btrfs_root *tree_root,
35b7e476 CM	383	struct list_head *list)
0f7d52f4 CM	384	{
	385	struct dirty_root *dirty;
	386	struct btrfs_trans_handle *trans;
54aa1f4d	387	int ret = 0;
9f3a7427 CM	388	int err;
9f3a7427 CM	389
0f7d52f4	390	while(!list_empty(list)) {
facda1e7	391	mutex_lock(&tree_root->fs_info->fs_mutex);
0f7d52f4 CM	392	dirty = list_entry(list->next, struct dirty_root, list);
0f7d52f4 CM	393	list_del_init(&dirty->list);
5eda7b5e	394
9f3a7427 CM	395	while(1) {
	396	trans = btrfs_start_transaction(tree_root, 1);
	397	ret = btrfs_drop_snapshot(trans, dirty->root);
	398	if (ret != -EAGAIN) {
	399	break;
	400	}
	401	err = btrfs_update_root(trans,
	402	tree_root,
	403	&dirty->root->root_key,
	404	&dirty->root->root_item);
	405	if (err)
	406	ret = err;
	407	ret = btrfs_end_transaction(trans, tree_root);
	408	BUG_ON(ret);
f4468e94 CM	409	mutex_unlock(&tree_root->fs_info->fs_mutex);
	410
	411	btrfs_btree_balance_dirty(tree_root);
	412	schedule();
	413
	414	mutex_lock(&tree_root->fs_info->fs_mutex);
9f3a7427	415	}
0f7d52f4	416	BUG_ON(ret);
9f3a7427	417	ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
54aa1f4d CM	418	if (ret)
54aa1f4d CM	419	break;
0f7d52f4 CM	420	ret = btrfs_end_transaction(trans, tree_root);
0f7d52f4 CM	421	BUG_ON(ret);
5eda7b5e	422
9f3a7427	423	kfree(dirty->root);
0f7d52f4	424	kfree(dirty);
facda1e7	425	mutex_unlock(&tree_root->fs_info->fs_mutex);
8c2383c3	426	btrfs_btree_balance_dirty(tree_root);
f4468e94	427	schedule();
0f7d52f4	428	}
54aa1f4d	429	return ret;
0f7d52f4 CM	430	}
0f7d52f4 CM	431
79154b1b CM	432	int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
	433	struct btrfs_root *root)
	434	{
15ee9bc7 JB	435	unsigned long joined = 0;
15ee9bc7 JB	436	unsigned long timeout = 1;
79154b1b	437	struct btrfs_transaction *cur_trans;
8fd17795	438	struct btrfs_transaction *prev_trans = NULL;
0f7d52f4	439	struct list_head dirty_fs_roots;
ccd467d6	440	struct radix_tree_root pinned_copy;
79154b1b	441	DEFINE_WAIT(wait);
15ee9bc7	442	int ret;
79154b1b	443
ccd467d6	444	init_bit_radix(&pinned_copy);
0f7d52f4	445	INIT_LIST_HEAD(&dirty_fs_roots);
d6e4a428	446
79154b1b CM	447	mutex_lock(&root->fs_info->trans_mutex);
	448	if (trans->transaction->in_commit) {
	449	cur_trans = trans->transaction;
	450	trans->transaction->use_count++;
ccd467d6	451	mutex_unlock(&root->fs_info->trans_mutex);
79154b1b	452	btrfs_end_transaction(trans, root);
ccd467d6 CM	453
ccd467d6 CM	454	mutex_unlock(&root->fs_info->fs_mutex);
79154b1b CM	455	ret = wait_for_commit(root, cur_trans);
79154b1b CM	456	BUG_ON(ret);
15ee9bc7 JB	457
15ee9bc7 JB	458	mutex_lock(&root->fs_info->trans_mutex);
79154b1b	459	put_transaction(cur_trans);
15ee9bc7 JB	460	mutex_unlock(&root->fs_info->trans_mutex);
15ee9bc7 JB	461
ccd467d6	462	mutex_lock(&root->fs_info->fs_mutex);
79154b1b CM	463	return 0;
79154b1b CM	464	}
2c90e5d6	465	trans->transaction->in_commit = 1;
ccd467d6 CM	466	cur_trans = trans->transaction;
	467	if (cur_trans->list.prev != &root->fs_info->trans_list) {
	468	prev_trans = list_entry(cur_trans->list.prev,
	469	struct btrfs_transaction, list);
	470	if (!prev_trans->commit_done) {
	471	prev_trans->use_count++;
	472	mutex_unlock(&root->fs_info->fs_mutex);
	473	mutex_unlock(&root->fs_info->trans_mutex);
	474
	475	wait_for_commit(root, prev_trans);
ccd467d6 CM	476
	477	mutex_lock(&root->fs_info->fs_mutex);
	478	mutex_lock(&root->fs_info->trans_mutex);
15ee9bc7	479	put_transaction(prev_trans);
ccd467d6 CM	480	}
ccd467d6 CM	481	}
15ee9bc7 JB	482
	483	do {
	484	joined = cur_trans->num_joined;
2c90e5d6	485	WARN_ON(cur_trans != trans->transaction);
15ee9bc7	486	prepare_to_wait(&cur_trans->writer_wait, &wait,
79154b1b	487	TASK_UNINTERRUPTIBLE);
15ee9bc7 JB	488
	489	if (cur_trans->num_writers > 1)
	490	timeout = MAX_SCHEDULE_TIMEOUT;
	491	else
	492	timeout = 1;
	493
ccd467d6	494	mutex_unlock(&root->fs_info->fs_mutex);
79154b1b	495	mutex_unlock(&root->fs_info->trans_mutex);
15ee9bc7 JB	496
	497	schedule_timeout(timeout);
	498
ccd467d6	499	mutex_lock(&root->fs_info->fs_mutex);
79154b1b	500	mutex_lock(&root->fs_info->trans_mutex);
15ee9bc7 JB	501	finish_wait(&cur_trans->writer_wait, &wait);
	502	} while (cur_trans->num_writers > 1 \|\|
	503	(cur_trans->num_joined != joined));
	504
2c90e5d6	505	WARN_ON(cur_trans != trans->transaction);
54aa1f4d CM	506	ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
	507	&dirty_fs_roots);
	508	BUG_ON(ret);
	509
79154b1b CM	510	ret = btrfs_commit_tree_roots(trans, root);
79154b1b CM	511	BUG_ON(ret);
54aa1f4d	512
78fae27e CM	513	cur_trans = root->fs_info->running_transaction;
78fae27e CM	514	root->fs_info->running_transaction = NULL;
4b52dff6 CM	515	btrfs_set_super_generation(&root->fs_info->super_copy,
	516	cur_trans->transid);
	517	btrfs_set_super_root(&root->fs_info->super_copy,
	518	bh_blocknr(root->fs_info->tree_root->node));
	519	memcpy(root->fs_info->disk_super, &root->fs_info->super_copy,
	520	sizeof(root->fs_info->super_copy));
ccd467d6 CM	521
	522	btrfs_copy_pinned(root, &pinned_copy);
	523
78fae27e	524	mutex_unlock(&root->fs_info->trans_mutex);
8fd17795	525	mutex_unlock(&root->fs_info->fs_mutex);
79154b1b CM	526	ret = btrfs_write_and_wait_transaction(trans, root);
79154b1b CM	527	BUG_ON(ret);
79154b1b	528	write_ctree_super(trans, root);
8fd17795	529	mutex_lock(&root->fs_info->fs_mutex);
ccd467d6	530	btrfs_finish_extent_commit(trans, root, &pinned_copy);
78fae27e	531	mutex_lock(&root->fs_info->trans_mutex);
2c90e5d6	532	cur_trans->commit_done = 1;
15ee9bc7	533	root->fs_info->last_trans_committed = cur_trans->transid;
2c90e5d6	534	wake_up(&cur_trans->commit_wait);
78fae27e	535	put_transaction(cur_trans);
79154b1b	536	put_transaction(cur_trans);
facda1e7 CM	537	if (root->fs_info->closing)
	538	list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
	539	else
	540	list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
78fae27e	541	mutex_unlock(&root->fs_info->trans_mutex);
2c90e5d6	542	kmem_cache_free(btrfs_trans_handle_cachep, trans);
79154b1b	543
facda1e7 CM	544	if (root->fs_info->closing) {
	545	mutex_unlock(&root->fs_info->fs_mutex);
	546	drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
	547	mutex_lock(&root->fs_info->fs_mutex);
	548	}
79154b1b CM	549	return ret;
	550	}
	551
e9d0b13b CM	552	int btrfs_clean_old_snapshots(struct btrfs_root *root)
	553	{
	554	struct list_head dirty_roots;
	555	INIT_LIST_HEAD(&dirty_roots);
	556
	557	mutex_lock(&root->fs_info->trans_mutex);
	558	list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
	559	mutex_unlock(&root->fs_info->trans_mutex);
	560
	561	if (!list_empty(&dirty_roots)) {
	562	drop_dirty_roots(root, &dirty_roots);
	563	}
	564	return 0;
	565	}
08607c1b CM	566	void btrfs_transaction_cleaner(struct work_struct *work)
	567	{
	568	struct btrfs_fs_info *fs_info = container_of(work,
	569	struct btrfs_fs_info,
	570	trans_work.work);
	571
	572	struct btrfs_root *root = fs_info->tree_root;
	573	struct btrfs_transaction *cur;
	574	struct btrfs_trans_handle *trans;
	575	unsigned long now;
	576	unsigned long delay = HZ * 30;
	577	int ret;
	578
08607c1b CM	579	mutex_lock(&root->fs_info->fs_mutex);
	580	mutex_lock(&root->fs_info->trans_mutex);
	581	cur = root->fs_info->running_transaction;
	582	if (!cur) {
	583	mutex_unlock(&root->fs_info->trans_mutex);
	584	goto out;
	585	}
	586	now = get_seconds();
	587	if (now < cur->start_time \|\| now - cur->start_time < 30) {
	588	mutex_unlock(&root->fs_info->trans_mutex);
	589	delay = HZ * 5;
	590	goto out;
	591	}
	592	mutex_unlock(&root->fs_info->trans_mutex);
6702ed49	593	btrfs_defrag_dirty_roots(root->fs_info);
08607c1b CM	594	trans = btrfs_start_transaction(root, 1);
	595	ret = btrfs_commit_transaction(trans, root);
	596	out:
	597	mutex_unlock(&root->fs_info->fs_mutex);
e9d0b13b	598	btrfs_clean_old_snapshots(root);
08607c1b CM	599	btrfs_transaction_queue_work(root, delay);
	600	}
	601
	602	void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
	603	{
	604	queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
	605	}
	606
	607	void btrfs_transaction_flush_work(struct btrfs_root *root)
	608	{
	609	cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work);
	610	flush_workqueue(trans_wq);
	611	}
	612
	613	void __init btrfs_init_transaction_sys(void)
	614	{
	615	trans_wq = create_workqueue("btrfs");
	616	}
	617
	618	void __exit btrfs_exit_transaction_sys(void)
	619	{
	620	destroy_workqueue(trans_wq);
	621	}
	622