[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
#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 = 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;
}

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;
	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_dirty_roots(struct btrfs_fs_info *info)
{
	struct btrfs_root *gang[1];
	struct btrfs_root *root;
	struct btrfs_root *tree_root = info->tree_root;
	struct btrfs_trans_handle *trans;
	int i;
	int ret;
	int err = 0;
	u64 last = 0;

	trans = btrfs_start_transaction(tree_root, 1);
	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;
			radix_tree_tag_clear(&info->fs_roots_radix,
				     (unsigned long)root->root_key.objectid,
				     BTRFS_ROOT_DEFRAG_TAG);
			if (root->defrag_running)
				continue;

			while (1) {
				mutex_lock(&root->fs_info->trans_mutex);
				record_root_in_trans(root);
				mutex_unlock(&root->fs_info->trans_mutex);

				root->defrag_running = 1;
				err = btrfs_defrag_leaves(trans, root, 1);
				btrfs_end_transaction(trans, tree_root);
				mutex_unlock(&info->fs_mutex);

				btrfs_btree_balance_dirty(root);
				cond_resched();

				mutex_lock(&info->fs_mutex);
				trans = btrfs_start_transaction(tree_root, 1);
				if (err != -EAGAIN)
					break;
			}
			root->defrag_running = 0;
		}
	}
	btrfs_end_transaction(trans, tree_root);
	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)
{
	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);
	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);

	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	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 CM	57	root->fs_info->running_transaction = cur_trans;
79154b1b CM	58	cur_trans->num_writers = 0;
0f7d52f4	59	cur_trans->transid = root->fs_info->generation;
79154b1b CM	60	init_waitqueue_head(&cur_trans->writer_wait);
	61	init_waitqueue_head(&cur_trans->commit_wait);
	62	cur_trans->in_commit = 0;
d5719762	63	cur_trans->use_count = 1;
79154b1b	64	cur_trans->commit_done = 0;
08607c1b	65	cur_trans->start_time = get_seconds();
8fd17795	66	list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
7c4452b9	67	init_bit_radix(&cur_trans->dirty_pages);
79154b1b CM	68	}
	69	cur_trans->num_writers++;
	70	return 0;
	71	}
	72
6702ed49 CM	73	static int record_root_in_trans(struct btrfs_root *root)
	74	{
	75	u64 running_trans_id = root->fs_info->running_transaction->transid;
	76	if (root->ref_cows && root->last_trans < running_trans_id) {
	77	WARN_ON(root == root->fs_info->extent_root);
	78	if (root->root_item.refs != 0) {
	79	radix_tree_tag_set(&root->fs_info->fs_roots_radix,
	80	(unsigned long)root->root_key.objectid,
	81	BTRFS_ROOT_TRANS_TAG);
	82	radix_tree_tag_set(&root->fs_info->fs_roots_radix,
	83	(unsigned long)root->root_key.objectid,
	84	BTRFS_ROOT_DEFRAG_TAG);
	85	root->commit_root = root->node;
	86	get_bh(root->node);
	87	} else {
	88	WARN_ON(1);
	89	}
	90	root->last_trans = running_trans_id;
	91	}
	92	return 0;
	93	}
	94
79154b1b CM	95	struct btrfs_trans_handle btrfs_start_transaction(struct btrfs_root root,
	96	int num_blocks)
	97	{
2c90e5d6 CM	98	struct btrfs_trans_handle *h =
2c90e5d6 CM	99	kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
79154b1b CM	100	int ret;
	101
	102	mutex_lock(&root->fs_info->trans_mutex);
	103	ret = join_transaction(root);
	104	BUG_ON(ret);
0f7d52f4	105
6702ed49 CM	106	record_root_in_trans(root);
6702ed49 CM	107	h->transid = root->fs_info->running_transaction->transid;
79154b1b CM	108	h->transaction = root->fs_info->running_transaction;
	109	h->blocks_reserved = num_blocks;
	110	h->blocks_used = 0;
31f3c99b	111	h->block_group = NULL;
79154b1b CM	112	root->fs_info->running_transaction->use_count++;
	113	mutex_unlock(&root->fs_info->trans_mutex);
	114	return h;
	115	}
	116
	117	int btrfs_end_transaction(struct btrfs_trans_handle *trans,
	118	struct btrfs_root *root)
	119	{
	120	struct btrfs_transaction *cur_trans;
d6e4a428	121
79154b1b CM	122	mutex_lock(&root->fs_info->trans_mutex);
79154b1b CM	123	cur_trans = root->fs_info->running_transaction;
ccd467d6	124	WARN_ON(cur_trans != trans->transaction);
d5719762	125	WARN_ON(cur_trans->num_writers < 1);
ccd467d6	126	cur_trans->num_writers--;
79154b1b CM	127	if (waitqueue_active(&cur_trans->writer_wait))
79154b1b CM	128	wake_up(&cur_trans->writer_wait);
79154b1b CM	129	put_transaction(cur_trans);
79154b1b CM	130	mutex_unlock(&root->fs_info->trans_mutex);
d6025579	131	memset(trans, 0, sizeof(*trans));
2c90e5d6	132	kmem_cache_free(btrfs_trans_handle_cachep, trans);
79154b1b CM	133	return 0;
	134	}
	135
	136
	137	int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
	138	struct btrfs_root *root)
	139	{
7c4452b9 CM	140	unsigned long gang[16];
	141	int ret;
	142	int i;
	143	int err;
	144	int werr = 0;
	145	struct page *page;
	146	struct radix_tree_root *dirty_pages;
	147	struct inode *btree_inode = root->fs_info->btree_inode;
	148
	149	if (!trans \|\| !trans->transaction) {
	150	return filemap_write_and_wait(btree_inode->i_mapping);
	151	}
	152	dirty_pages = &trans->transaction->dirty_pages;
	153	while(1) {
e37c9e69 CM	154	ret = find_first_radix_bit(dirty_pages, gang,
e37c9e69 CM	155	0, ARRAY_SIZE(gang));
7c4452b9 CM	156	if (!ret)
	157	break;
	158	for (i = 0; i < ret; i++) {
	159	/* FIXME EIO */
	160	clear_radix_bit(dirty_pages, gang[i]);
	161	page = find_lock_page(btree_inode->i_mapping,
	162	gang[i]);
	163	if (!page)
	164	continue;
6702ed49 CM	165	if (PageWriteback(page)) {
	166	if (PageDirty(page))
	167	wait_on_page_writeback(page);
	168	else {
	169	unlock_page(page);
	170	page_cache_release(page);
	171	continue;
	172	}
	173	}
7c4452b9 CM	174	err = write_one_page(page, 0);
	175	if (err)
	176	werr = err;
	177	page_cache_release(page);
	178	}
	179	}
	180	err = filemap_fdatawait(btree_inode->i_mapping);
	181	if (err)
	182	werr = err;
	183	return werr;
79154b1b CM	184	}
	185
	186	int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
	187	struct btrfs_root *root)
	188	{
	189	int ret;
	190	u64 old_extent_block;
	191	struct btrfs_fs_info *fs_info = root->fs_info;
	192	struct btrfs_root *tree_root = fs_info->tree_root;
	193	struct btrfs_root *extent_root = fs_info->extent_root;
79154b1b	194
9078a3e1	195	btrfs_write_dirty_block_groups(trans, extent_root);
79154b1b CM	196	while(1) {
79154b1b CM	197	old_extent_block = btrfs_root_blocknr(&extent_root->root_item);
7eccb903	198	if (old_extent_block == bh_blocknr(extent_root->node))
79154b1b CM	199	break;
79154b1b CM	200	btrfs_set_root_blocknr(&extent_root->root_item,
7eccb903	201	bh_blocknr(extent_root->node));
79154b1b CM	202	ret = btrfs_update_root(trans, tree_root,
	203	&extent_root->root_key,
	204	&extent_root->root_item);
	205	BUG_ON(ret);
9078a3e1	206	btrfs_write_dirty_block_groups(trans, extent_root);
79154b1b CM	207	}
	208	return 0;
	209	}
	210
	211	static int wait_for_commit(struct btrfs_root *root,
	212	struct btrfs_transaction *commit)
	213	{
	214	DEFINE_WAIT(wait);
ccd467d6	215	mutex_lock(&root->fs_info->trans_mutex);
79154b1b CM	216	while(!commit->commit_done) {
	217	prepare_to_wait(&commit->commit_wait, &wait,
	218	TASK_UNINTERRUPTIBLE);
	219	if (commit->commit_done)
	220	break;
	221	mutex_unlock(&root->fs_info->trans_mutex);
	222	schedule();
	223	mutex_lock(&root->fs_info->trans_mutex);
	224	}
ccd467d6	225	mutex_unlock(&root->fs_info->trans_mutex);
79154b1b CM	226	finish_wait(&commit->commit_wait, &wait);
	227	return 0;
	228	}
	229
0f7d52f4 CM	230	struct dirty_root {
0f7d52f4 CM	231	struct list_head list;
0f7d52f4 CM	232	struct btrfs_root *root;
	233	};
	234
5eda7b5e CM	235	int btrfs_add_dead_root(struct btrfs_root root, struct list_head dead_list)
	236	{
	237	struct dirty_root *dirty;
	238
	239	dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
	240	if (!dirty)
	241	return -ENOMEM;
5eda7b5e	242	dirty->root = root;
5eda7b5e CM	243	list_add(&dirty->list, dead_list);
	244	return 0;
	245	}
	246
35b7e476 CM	247	static int add_dirty_roots(struct btrfs_trans_handle *trans,
	248	struct radix_tree_root *radix,
	249	struct list_head *list)
0f7d52f4 CM	250	{
	251	struct dirty_root *dirty;
	252	struct btrfs_root *gang[8];
	253	struct btrfs_root *root;
	254	int i;
	255	int ret;
54aa1f4d	256	int err = 0;
5eda7b5e	257	u32 refs;
54aa1f4d	258
0f7d52f4 CM	259	while(1) {
	260	ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
	261	ARRAY_SIZE(gang),
	262	BTRFS_ROOT_TRANS_TAG);
	263	if (ret == 0)
	264	break;
	265	for (i = 0; i < ret; i++) {
	266	root = gang[i];
2619ba1f CM	267	radix_tree_tag_clear(radix,
	268	(unsigned long)root->root_key.objectid,
	269	BTRFS_ROOT_TRANS_TAG);
0f7d52f4	270	if (root->commit_root == root->node) {
7eccb903	271	WARN_ON(bh_blocknr(root->node) !=
0f7d52f4 CM	272	btrfs_root_blocknr(&root->root_item));
	273	brelse(root->commit_root);
	274	root->commit_root = NULL;
	275	continue;
	276	}
	277	dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
	278	BUG_ON(!dirty);
9f3a7427 CM	279	dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
	280	BUG_ON(!dirty->root);
	281
	282	memset(&root->root_item.drop_progress, 0,
	283	sizeof(struct btrfs_disk_key));
	284	root->root_item.drop_level = 0;
	285
	286	memcpy(dirty->root, root, sizeof(*root));
	287	dirty->root->node = root->commit_root;
0f7d52f4	288	root->commit_root = NULL;
5eda7b5e	289
0f7d52f4 CM	290	root->root_key.offset = root->fs_info->generation;
0f7d52f4 CM	291	btrfs_set_root_blocknr(&root->root_item,
7eccb903	292	bh_blocknr(root->node));
0f7d52f4 CM	293	err = btrfs_insert_root(trans, root->fs_info->tree_root,
	294	&root->root_key,
	295	&root->root_item);
54aa1f4d CM	296	if (err)
54aa1f4d CM	297	break;
9f3a7427 CM	298
	299	refs = btrfs_root_refs(&dirty->root->root_item);
	300	btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
5eda7b5e	301	err = btrfs_update_root(trans, root->fs_info->tree_root,
9f3a7427 CM	302	&dirty->root->root_key,
9f3a7427 CM	303	&dirty->root->root_item);
5eda7b5e CM	304
5eda7b5e CM	305	BUG_ON(err);
9f3a7427	306	if (refs == 1) {
5eda7b5e	307	list_add(&dirty->list, list);
9f3a7427 CM	308	} else {
	309	WARN_ON(1);
	310	kfree(dirty->root);
5eda7b5e	311	kfree(dirty);
9f3a7427	312	}
0f7d52f4 CM	313	}
0f7d52f4 CM	314	}
54aa1f4d	315	return err;
0f7d52f4 CM	316	}
0f7d52f4 CM	317
6702ed49 CM	318	int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
	319	{
	320	struct btrfs_root *gang[1];
	321	struct btrfs_root *root;
	322	struct btrfs_root *tree_root = info->tree_root;
	323	struct btrfs_trans_handle *trans;
	324	int i;
	325	int ret;
	326	int err = 0;
	327	u64 last = 0;
	328
	329	trans = btrfs_start_transaction(tree_root, 1);
	330	while(1) {
	331	ret = radix_tree_gang_lookup_tag(&info->fs_roots_radix,
	332	(void **)gang, last,
	333	ARRAY_SIZE(gang),
	334	BTRFS_ROOT_DEFRAG_TAG);
	335	if (ret == 0)
	336	break;
	337	for (i = 0; i < ret; i++) {
	338	root = gang[i];
	339	last = root->root_key.objectid + 1;
	340	radix_tree_tag_clear(&info->fs_roots_radix,
	341	(unsigned long)root->root_key.objectid,
	342	BTRFS_ROOT_DEFRAG_TAG);
	343	if (root->defrag_running)
	344	continue;
	345
	346	while (1) {
	347	mutex_lock(&root->fs_info->trans_mutex);
	348	record_root_in_trans(root);
	349	mutex_unlock(&root->fs_info->trans_mutex);
	350
	351	root->defrag_running = 1;
	352	err = btrfs_defrag_leaves(trans, root, 1);
	353	btrfs_end_transaction(trans, tree_root);
	354	mutex_unlock(&info->fs_mutex);
	355
	356	btrfs_btree_balance_dirty(root);
f4468e94	357	cond_resched();
6702ed49 CM	358
	359	mutex_lock(&info->fs_mutex);
	360	trans = btrfs_start_transaction(tree_root, 1);
	361	if (err != -EAGAIN)
	362	break;
	363	}
	364	root->defrag_running = 0;
	365	}
	366	}
	367	btrfs_end_transaction(trans, tree_root);
	368	return err;
	369	}
	370
35b7e476 CM	371	static int drop_dirty_roots(struct btrfs_root *tree_root,
35b7e476 CM	372	struct list_head *list)
0f7d52f4 CM	373	{
	374	struct dirty_root *dirty;
	375	struct btrfs_trans_handle *trans;
54aa1f4d	376	int ret = 0;
9f3a7427 CM	377	int err;
9f3a7427 CM	378
0f7d52f4	379	while(!list_empty(list)) {
facda1e7	380	mutex_lock(&tree_root->fs_info->fs_mutex);
0f7d52f4 CM	381	dirty = list_entry(list->next, struct dirty_root, list);
0f7d52f4 CM	382	list_del_init(&dirty->list);
5eda7b5e	383
9f3a7427 CM	384	while(1) {
	385	trans = btrfs_start_transaction(tree_root, 1);
	386	ret = btrfs_drop_snapshot(trans, dirty->root);
	387	if (ret != -EAGAIN) {
	388	break;
	389	}
	390	err = btrfs_update_root(trans,
	391	tree_root,
	392	&dirty->root->root_key,
	393	&dirty->root->root_item);
	394	if (err)
	395	ret = err;
	396	ret = btrfs_end_transaction(trans, tree_root);
	397	BUG_ON(ret);
f4468e94 CM	398	mutex_unlock(&tree_root->fs_info->fs_mutex);
	399
	400	btrfs_btree_balance_dirty(tree_root);
	401	schedule();
	402
	403	mutex_lock(&tree_root->fs_info->fs_mutex);
9f3a7427	404	}
0f7d52f4	405	BUG_ON(ret);
9f3a7427	406	ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
54aa1f4d CM	407	if (ret)
54aa1f4d CM	408	break;
0f7d52f4 CM	409	ret = btrfs_end_transaction(trans, tree_root);
0f7d52f4 CM	410	BUG_ON(ret);
5eda7b5e	411
9f3a7427	412	kfree(dirty->root);
0f7d52f4	413	kfree(dirty);
facda1e7	414	mutex_unlock(&tree_root->fs_info->fs_mutex);
8c2383c3	415	btrfs_btree_balance_dirty(tree_root);
f4468e94	416	schedule();
0f7d52f4	417	}
54aa1f4d	418	return ret;
0f7d52f4 CM	419	}
0f7d52f4 CM	420
79154b1b CM	421	int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
	422	struct btrfs_root *root)
	423	{
	424	int ret = 0;
79154b1b	425	struct btrfs_transaction *cur_trans;
8fd17795	426	struct btrfs_transaction *prev_trans = NULL;
0f7d52f4	427	struct list_head dirty_fs_roots;
ccd467d6	428	struct radix_tree_root pinned_copy;
79154b1b CM	429	DEFINE_WAIT(wait);
79154b1b CM	430
ccd467d6	431	init_bit_radix(&pinned_copy);
0f7d52f4	432	INIT_LIST_HEAD(&dirty_fs_roots);
d6e4a428	433
79154b1b CM	434	mutex_lock(&root->fs_info->trans_mutex);
	435	if (trans->transaction->in_commit) {
	436	cur_trans = trans->transaction;
	437	trans->transaction->use_count++;
ccd467d6	438	mutex_unlock(&root->fs_info->trans_mutex);
79154b1b	439	btrfs_end_transaction(trans, root);
ccd467d6 CM	440
ccd467d6 CM	441	mutex_unlock(&root->fs_info->fs_mutex);
79154b1b CM	442	ret = wait_for_commit(root, cur_trans);
	443	BUG_ON(ret);
	444	put_transaction(cur_trans);
ccd467d6	445	mutex_lock(&root->fs_info->fs_mutex);
79154b1b CM	446	return 0;
79154b1b CM	447	}
2c90e5d6	448	trans->transaction->in_commit = 1;
ccd467d6 CM	449	cur_trans = trans->transaction;
	450	if (cur_trans->list.prev != &root->fs_info->trans_list) {
	451	prev_trans = list_entry(cur_trans->list.prev,
	452	struct btrfs_transaction, list);
	453	if (!prev_trans->commit_done) {
	454	prev_trans->use_count++;
	455	mutex_unlock(&root->fs_info->fs_mutex);
	456	mutex_unlock(&root->fs_info->trans_mutex);
	457
	458	wait_for_commit(root, prev_trans);
	459	put_transaction(prev_trans);
	460
	461	mutex_lock(&root->fs_info->fs_mutex);
	462	mutex_lock(&root->fs_info->trans_mutex);
	463	}
	464	}
79154b1b	465	while (trans->transaction->num_writers > 1) {
2c90e5d6	466	WARN_ON(cur_trans != trans->transaction);
79154b1b CM	467	prepare_to_wait(&trans->transaction->writer_wait, &wait,
	468	TASK_UNINTERRUPTIBLE);
	469	if (trans->transaction->num_writers <= 1)
	470	break;
ccd467d6	471	mutex_unlock(&root->fs_info->fs_mutex);
79154b1b CM	472	mutex_unlock(&root->fs_info->trans_mutex);
79154b1b CM	473	schedule();
ccd467d6	474	mutex_lock(&root->fs_info->fs_mutex);
79154b1b	475	mutex_lock(&root->fs_info->trans_mutex);
2c90e5d6	476	finish_wait(&trans->transaction->writer_wait, &wait);
79154b1b CM	477	}
79154b1b CM	478	finish_wait(&trans->transaction->writer_wait, &wait);
2c90e5d6	479	WARN_ON(cur_trans != trans->transaction);
54aa1f4d CM	480	ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
	481	&dirty_fs_roots);
	482	BUG_ON(ret);
	483
79154b1b CM	484	ret = btrfs_commit_tree_roots(trans, root);
79154b1b CM	485	BUG_ON(ret);
54aa1f4d	486
78fae27e CM	487	cur_trans = root->fs_info->running_transaction;
78fae27e CM	488	root->fs_info->running_transaction = NULL;
4b52dff6 CM	489	btrfs_set_super_generation(&root->fs_info->super_copy,
	490	cur_trans->transid);
	491	btrfs_set_super_root(&root->fs_info->super_copy,
	492	bh_blocknr(root->fs_info->tree_root->node));
	493	memcpy(root->fs_info->disk_super, &root->fs_info->super_copy,
	494	sizeof(root->fs_info->super_copy));
ccd467d6 CM	495
	496	btrfs_copy_pinned(root, &pinned_copy);
	497
78fae27e	498	mutex_unlock(&root->fs_info->trans_mutex);
8fd17795	499	mutex_unlock(&root->fs_info->fs_mutex);
79154b1b CM	500	ret = btrfs_write_and_wait_transaction(trans, root);
79154b1b CM	501	BUG_ON(ret);
79154b1b	502	write_ctree_super(trans, root);
8fd17795	503	mutex_lock(&root->fs_info->fs_mutex);
ccd467d6	504	btrfs_finish_extent_commit(trans, root, &pinned_copy);
78fae27e	505	mutex_lock(&root->fs_info->trans_mutex);
2c90e5d6 CM	506	cur_trans->commit_done = 1;
2c90e5d6 CM	507	wake_up(&cur_trans->commit_wait);
78fae27e	508	put_transaction(cur_trans);
79154b1b	509	put_transaction(cur_trans);
facda1e7 CM	510	if (root->fs_info->closing)
	511	list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
	512	else
	513	list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
78fae27e	514	mutex_unlock(&root->fs_info->trans_mutex);
2c90e5d6	515	kmem_cache_free(btrfs_trans_handle_cachep, trans);
79154b1b	516
facda1e7 CM	517	if (root->fs_info->closing) {
	518	mutex_unlock(&root->fs_info->fs_mutex);
	519	drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
	520	mutex_lock(&root->fs_info->fs_mutex);
	521	}
79154b1b CM	522	return ret;
	523	}
	524
08607c1b CM	525	void btrfs_transaction_cleaner(struct work_struct *work)
	526	{
	527	struct btrfs_fs_info *fs_info = container_of(work,
	528	struct btrfs_fs_info,
	529	trans_work.work);
	530
	531	struct btrfs_root *root = fs_info->tree_root;
	532	struct btrfs_transaction *cur;
	533	struct btrfs_trans_handle *trans;
facda1e7	534	struct list_head dirty_roots;
08607c1b CM	535	unsigned long now;
	536	unsigned long delay = HZ * 30;
	537	int ret;
	538
facda1e7	539	INIT_LIST_HEAD(&dirty_roots);
08607c1b CM	540	mutex_lock(&root->fs_info->fs_mutex);
	541	mutex_lock(&root->fs_info->trans_mutex);
	542	cur = root->fs_info->running_transaction;
	543	if (!cur) {
	544	mutex_unlock(&root->fs_info->trans_mutex);
	545	goto out;
	546	}
	547	now = get_seconds();
	548	if (now < cur->start_time \|\| now - cur->start_time < 30) {
	549	mutex_unlock(&root->fs_info->trans_mutex);
	550	delay = HZ * 5;
	551	goto out;
	552	}
	553	mutex_unlock(&root->fs_info->trans_mutex);
6702ed49	554	btrfs_defrag_dirty_roots(root->fs_info);
08607c1b CM	555	trans = btrfs_start_transaction(root, 1);
	556	ret = btrfs_commit_transaction(trans, root);
	557	out:
	558	mutex_unlock(&root->fs_info->fs_mutex);
ad693af6 CM	559
	560	mutex_lock(&root->fs_info->trans_mutex);
	561	list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
	562	mutex_unlock(&root->fs_info->trans_mutex);
	563
	564	if (!list_empty(&dirty_roots)) {
	565	drop_dirty_roots(root, &dirty_roots);
	566	}
08607c1b CM	567	btrfs_transaction_queue_work(root, delay);
	568	}
	569
	570	void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
	571	{
	572	queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
	573	}
	574
	575	void btrfs_transaction_flush_work(struct btrfs_root *root)
	576	{
	577	cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work);
	578	flush_workqueue(trans_wq);
	579	}
	580
	581	void __init btrfs_init_transaction_sys(void)
	582	{
	583	trans_wq = create_workqueue("btrfs");
	584	}
	585
	586	void __exit btrfs_exit_transaction_sys(void)
	587	{
	588	destroy_workqueue(trans_wq);
	589	}
	590