[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;
	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_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;
			radix_tree_tag_clear(&info->fs_roots_radix,
				     (unsigned long)root->root_key.objectid,
				     BTRFS_ROOT_DEFRAG_TAG);
		}
	}
	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;
26b8003f CM	112	h->alloc_exclude_nr = 0;
26b8003f CM	113	h->alloc_exclude_start = 0;
79154b1b CM	114	root->fs_info->running_transaction->use_count++;
	115	mutex_unlock(&root->fs_info->trans_mutex);
	116	return h;
	117	}
	118
	119	int btrfs_end_transaction(struct btrfs_trans_handle *trans,
	120	struct btrfs_root *root)
	121	{
	122	struct btrfs_transaction *cur_trans;
d6e4a428	123
79154b1b CM	124	mutex_lock(&root->fs_info->trans_mutex);
79154b1b CM	125	cur_trans = root->fs_info->running_transaction;
ccd467d6	126	WARN_ON(cur_trans != trans->transaction);
d5719762	127	WARN_ON(cur_trans->num_writers < 1);
ccd467d6	128	cur_trans->num_writers--;
79154b1b CM	129	if (waitqueue_active(&cur_trans->writer_wait))
79154b1b CM	130	wake_up(&cur_trans->writer_wait);
79154b1b CM	131	put_transaction(cur_trans);
79154b1b CM	132	mutex_unlock(&root->fs_info->trans_mutex);
d6025579	133	memset(trans, 0, sizeof(*trans));
2c90e5d6	134	kmem_cache_free(btrfs_trans_handle_cachep, trans);
79154b1b CM	135	return 0;
	136	}
	137
	138
	139	int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
	140	struct btrfs_root *root)
	141	{
7c4452b9 CM	142	unsigned long gang[16];
	143	int ret;
	144	int i;
	145	int err;
	146	int werr = 0;
	147	struct page *page;
	148	struct radix_tree_root *dirty_pages;
	149	struct inode *btree_inode = root->fs_info->btree_inode;
	150
	151	if (!trans \|\| !trans->transaction) {
	152	return filemap_write_and_wait(btree_inode->i_mapping);
	153	}
	154	dirty_pages = &trans->transaction->dirty_pages;
	155	while(1) {
e37c9e69 CM	156	ret = find_first_radix_bit(dirty_pages, gang,
e37c9e69 CM	157	0, ARRAY_SIZE(gang));
7c4452b9 CM	158	if (!ret)
	159	break;
	160	for (i = 0; i < ret; i++) {
	161	/* FIXME EIO */
	162	clear_radix_bit(dirty_pages, gang[i]);
	163	page = find_lock_page(btree_inode->i_mapping,
	164	gang[i]);
	165	if (!page)
	166	continue;
6702ed49 CM	167	if (PageWriteback(page)) {
	168	if (PageDirty(page))
	169	wait_on_page_writeback(page);
	170	else {
	171	unlock_page(page);
	172	page_cache_release(page);
	173	continue;
	174	}
	175	}
7c4452b9 CM	176	err = write_one_page(page, 0);
	177	if (err)
	178	werr = err;
	179	page_cache_release(page);
	180	}
	181	}
	182	err = filemap_fdatawait(btree_inode->i_mapping);
	183	if (err)
	184	werr = err;
	185	return werr;
79154b1b CM	186	}
	187
	188	int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
	189	struct btrfs_root *root)
	190	{
	191	int ret;
	192	u64 old_extent_block;
	193	struct btrfs_fs_info *fs_info = root->fs_info;
	194	struct btrfs_root *tree_root = fs_info->tree_root;
	195	struct btrfs_root *extent_root = fs_info->extent_root;
79154b1b	196
9078a3e1	197	btrfs_write_dirty_block_groups(trans, extent_root);
79154b1b CM	198	while(1) {
79154b1b CM	199	old_extent_block = btrfs_root_blocknr(&extent_root->root_item);
7eccb903	200	if (old_extent_block == bh_blocknr(extent_root->node))
79154b1b CM	201	break;
79154b1b CM	202	btrfs_set_root_blocknr(&extent_root->root_item,
7eccb903	203	bh_blocknr(extent_root->node));
79154b1b CM	204	ret = btrfs_update_root(trans, tree_root,
	205	&extent_root->root_key,
	206	&extent_root->root_item);
	207	BUG_ON(ret);
9078a3e1	208	btrfs_write_dirty_block_groups(trans, extent_root);
79154b1b CM	209	}
	210	return 0;
	211	}
	212
	213	static int wait_for_commit(struct btrfs_root *root,
	214	struct btrfs_transaction *commit)
	215	{
	216	DEFINE_WAIT(wait);
ccd467d6	217	mutex_lock(&root->fs_info->trans_mutex);
79154b1b CM	218	while(!commit->commit_done) {
	219	prepare_to_wait(&commit->commit_wait, &wait,
	220	TASK_UNINTERRUPTIBLE);
	221	if (commit->commit_done)
	222	break;
	223	mutex_unlock(&root->fs_info->trans_mutex);
	224	schedule();
	225	mutex_lock(&root->fs_info->trans_mutex);
	226	}
ccd467d6	227	mutex_unlock(&root->fs_info->trans_mutex);
79154b1b CM	228	finish_wait(&commit->commit_wait, &wait);
	229	return 0;
	230	}
	231
0f7d52f4 CM	232	struct dirty_root {
0f7d52f4 CM	233	struct list_head list;
0f7d52f4 CM	234	struct btrfs_root *root;
	235	};
	236
5eda7b5e CM	237	int btrfs_add_dead_root(struct btrfs_root root, struct list_head dead_list)
	238	{
	239	struct dirty_root *dirty;
	240
	241	dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
	242	if (!dirty)
	243	return -ENOMEM;
5eda7b5e	244	dirty->root = root;
5eda7b5e CM	245	list_add(&dirty->list, dead_list);
	246	return 0;
	247	}
	248
35b7e476 CM	249	static int add_dirty_roots(struct btrfs_trans_handle *trans,
	250	struct radix_tree_root *radix,
	251	struct list_head *list)
0f7d52f4 CM	252	{
	253	struct dirty_root *dirty;
	254	struct btrfs_root *gang[8];
	255	struct btrfs_root *root;
	256	int i;
	257	int ret;
54aa1f4d	258	int err = 0;
5eda7b5e	259	u32 refs;
54aa1f4d	260
0f7d52f4 CM	261	while(1) {
	262	ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
	263	ARRAY_SIZE(gang),
	264	BTRFS_ROOT_TRANS_TAG);
	265	if (ret == 0)
	266	break;
	267	for (i = 0; i < ret; i++) {
	268	root = gang[i];
2619ba1f CM	269	radix_tree_tag_clear(radix,
	270	(unsigned long)root->root_key.objectid,
	271	BTRFS_ROOT_TRANS_TAG);
0f7d52f4	272	if (root->commit_root == root->node) {
7eccb903	273	WARN_ON(bh_blocknr(root->node) !=
0f7d52f4 CM	274	btrfs_root_blocknr(&root->root_item));
	275	brelse(root->commit_root);
	276	root->commit_root = NULL;
	277	continue;
	278	}
	279	dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
	280	BUG_ON(!dirty);
9f3a7427 CM	281	dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
	282	BUG_ON(!dirty->root);
	283
	284	memset(&root->root_item.drop_progress, 0,
	285	sizeof(struct btrfs_disk_key));
	286	root->root_item.drop_level = 0;
	287
	288	memcpy(dirty->root, root, sizeof(*root));
	289	dirty->root->node = root->commit_root;
0f7d52f4	290	root->commit_root = NULL;
5eda7b5e	291
0f7d52f4 CM	292	root->root_key.offset = root->fs_info->generation;
0f7d52f4 CM	293	btrfs_set_root_blocknr(&root->root_item,
7eccb903	294	bh_blocknr(root->node));
0f7d52f4 CM	295	err = btrfs_insert_root(trans, root->fs_info->tree_root,
	296	&root->root_key,
	297	&root->root_item);
54aa1f4d CM	298	if (err)
54aa1f4d CM	299	break;
9f3a7427 CM	300
	301	refs = btrfs_root_refs(&dirty->root->root_item);
	302	btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
5eda7b5e	303	err = btrfs_update_root(trans, root->fs_info->tree_root,
9f3a7427 CM	304	&dirty->root->root_key,
9f3a7427 CM	305	&dirty->root->root_item);
5eda7b5e CM	306
5eda7b5e CM	307	BUG_ON(err);
9f3a7427	308	if (refs == 1) {
5eda7b5e	309	list_add(&dirty->list, list);
9f3a7427 CM	310	} else {
	311	WARN_ON(1);
	312	kfree(dirty->root);
5eda7b5e	313	kfree(dirty);
9f3a7427	314	}
0f7d52f4 CM	315	}
0f7d52f4 CM	316	}
54aa1f4d	317	return err;
0f7d52f4 CM	318	}
0f7d52f4 CM	319
6702ed49 CM	320	int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
	321	{
	322	struct btrfs_root *gang[1];
	323	struct btrfs_root *root;
	324	struct btrfs_root *tree_root = info->tree_root;
	325	struct btrfs_trans_handle *trans;
	326	int i;
	327	int ret;
	328	int err = 0;
	329	u64 last = 0;
	330
	331	trans = btrfs_start_transaction(tree_root, 1);
	332	while(1) {
	333	ret = radix_tree_gang_lookup_tag(&info->fs_roots_radix,
	334	(void **)gang, last,
	335	ARRAY_SIZE(gang),
	336	BTRFS_ROOT_DEFRAG_TAG);
	337	if (ret == 0)
	338	break;
	339	for (i = 0; i < ret; i++) {
	340	root = gang[i];
	341	last = root->root_key.objectid + 1;
	342	radix_tree_tag_clear(&info->fs_roots_radix,
	343	(unsigned long)root->root_key.objectid,
	344	BTRFS_ROOT_DEFRAG_TAG);
	345	if (root->defrag_running)
	346	continue;
	347
	348	while (1) {
	349	mutex_lock(&root->fs_info->trans_mutex);
	350	record_root_in_trans(root);
	351	mutex_unlock(&root->fs_info->trans_mutex);
	352
	353	root->defrag_running = 1;
	354	err = btrfs_defrag_leaves(trans, root, 1);
	355	btrfs_end_transaction(trans, tree_root);
	356	mutex_unlock(&info->fs_mutex);
	357
	358	btrfs_btree_balance_dirty(root);
f4468e94	359	cond_resched();
6702ed49 CM	360
	361	mutex_lock(&info->fs_mutex);
	362	trans = btrfs_start_transaction(tree_root, 1);
	363	if (err != -EAGAIN)
	364	break;
	365	}
	366	root->defrag_running = 0;
409eb95d CM	367	radix_tree_tag_clear(&info->fs_roots_radix,
	368	(unsigned long)root->root_key.objectid,
	369	BTRFS_ROOT_DEFRAG_TAG);
6702ed49 CM	370	}
	371	}
	372	btrfs_end_transaction(trans, tree_root);
	373	return err;
	374	}
	375
35b7e476 CM	376	static int drop_dirty_roots(struct btrfs_root *tree_root,
35b7e476 CM	377	struct list_head *list)
0f7d52f4 CM	378	{
	379	struct dirty_root *dirty;
	380	struct btrfs_trans_handle *trans;
54aa1f4d	381	int ret = 0;
9f3a7427 CM	382	int err;
9f3a7427 CM	383
0f7d52f4	384	while(!list_empty(list)) {
facda1e7	385	mutex_lock(&tree_root->fs_info->fs_mutex);
0f7d52f4 CM	386	dirty = list_entry(list->next, struct dirty_root, list);
0f7d52f4 CM	387	list_del_init(&dirty->list);
5eda7b5e	388
9f3a7427 CM	389	while(1) {
	390	trans = btrfs_start_transaction(tree_root, 1);
	391	ret = btrfs_drop_snapshot(trans, dirty->root);
	392	if (ret != -EAGAIN) {
	393	break;
	394	}
	395	err = btrfs_update_root(trans,
	396	tree_root,
	397	&dirty->root->root_key,
	398	&dirty->root->root_item);
	399	if (err)
	400	ret = err;
	401	ret = btrfs_end_transaction(trans, tree_root);
	402	BUG_ON(ret);
f4468e94 CM	403	mutex_unlock(&tree_root->fs_info->fs_mutex);
	404
	405	btrfs_btree_balance_dirty(tree_root);
	406	schedule();
	407
	408	mutex_lock(&tree_root->fs_info->fs_mutex);
9f3a7427	409	}
0f7d52f4	410	BUG_ON(ret);
9f3a7427	411	ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
54aa1f4d CM	412	if (ret)
54aa1f4d CM	413	break;
0f7d52f4 CM	414	ret = btrfs_end_transaction(trans, tree_root);
0f7d52f4 CM	415	BUG_ON(ret);
5eda7b5e	416
9f3a7427	417	kfree(dirty->root);
0f7d52f4	418	kfree(dirty);
facda1e7	419	mutex_unlock(&tree_root->fs_info->fs_mutex);
8c2383c3	420	btrfs_btree_balance_dirty(tree_root);
f4468e94	421	schedule();
0f7d52f4	422	}
54aa1f4d	423	return ret;
0f7d52f4 CM	424	}
0f7d52f4 CM	425
79154b1b CM	426	int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
	427	struct btrfs_root *root)
	428	{
	429	int ret = 0;
79154b1b	430	struct btrfs_transaction *cur_trans;
8fd17795	431	struct btrfs_transaction *prev_trans = NULL;
0f7d52f4	432	struct list_head dirty_fs_roots;
ccd467d6	433	struct radix_tree_root pinned_copy;
79154b1b CM	434	DEFINE_WAIT(wait);
79154b1b CM	435
ccd467d6	436	init_bit_radix(&pinned_copy);
0f7d52f4	437	INIT_LIST_HEAD(&dirty_fs_roots);
d6e4a428	438
79154b1b CM	439	mutex_lock(&root->fs_info->trans_mutex);
	440	if (trans->transaction->in_commit) {
	441	cur_trans = trans->transaction;
	442	trans->transaction->use_count++;
ccd467d6	443	mutex_unlock(&root->fs_info->trans_mutex);
79154b1b	444	btrfs_end_transaction(trans, root);
ccd467d6 CM	445
ccd467d6 CM	446	mutex_unlock(&root->fs_info->fs_mutex);
79154b1b CM	447	ret = wait_for_commit(root, cur_trans);
	448	BUG_ON(ret);
	449	put_transaction(cur_trans);
ccd467d6	450	mutex_lock(&root->fs_info->fs_mutex);
79154b1b CM	451	return 0;
79154b1b CM	452	}
2c90e5d6	453	trans->transaction->in_commit = 1;
ccd467d6 CM	454	cur_trans = trans->transaction;
	455	if (cur_trans->list.prev != &root->fs_info->trans_list) {
	456	prev_trans = list_entry(cur_trans->list.prev,
	457	struct btrfs_transaction, list);
	458	if (!prev_trans->commit_done) {
	459	prev_trans->use_count++;
	460	mutex_unlock(&root->fs_info->fs_mutex);
	461	mutex_unlock(&root->fs_info->trans_mutex);
	462
	463	wait_for_commit(root, prev_trans);
	464	put_transaction(prev_trans);
	465
	466	mutex_lock(&root->fs_info->fs_mutex);
	467	mutex_lock(&root->fs_info->trans_mutex);
	468	}
	469	}
79154b1b	470	while (trans->transaction->num_writers > 1) {
2c90e5d6	471	WARN_ON(cur_trans != trans->transaction);
79154b1b CM	472	prepare_to_wait(&trans->transaction->writer_wait, &wait,
	473	TASK_UNINTERRUPTIBLE);
	474	if (trans->transaction->num_writers <= 1)
	475	break;
ccd467d6	476	mutex_unlock(&root->fs_info->fs_mutex);
79154b1b CM	477	mutex_unlock(&root->fs_info->trans_mutex);
79154b1b CM	478	schedule();
ccd467d6	479	mutex_lock(&root->fs_info->fs_mutex);
79154b1b	480	mutex_lock(&root->fs_info->trans_mutex);
2c90e5d6	481	finish_wait(&trans->transaction->writer_wait, &wait);
79154b1b CM	482	}
79154b1b CM	483	finish_wait(&trans->transaction->writer_wait, &wait);
2c90e5d6	484	WARN_ON(cur_trans != trans->transaction);
54aa1f4d CM	485	ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
	486	&dirty_fs_roots);
	487	BUG_ON(ret);
	488
79154b1b CM	489	ret = btrfs_commit_tree_roots(trans, root);
79154b1b CM	490	BUG_ON(ret);
54aa1f4d	491
78fae27e CM	492	cur_trans = root->fs_info->running_transaction;
78fae27e CM	493	root->fs_info->running_transaction = NULL;
4b52dff6 CM	494	btrfs_set_super_generation(&root->fs_info->super_copy,
	495	cur_trans->transid);
	496	btrfs_set_super_root(&root->fs_info->super_copy,
	497	bh_blocknr(root->fs_info->tree_root->node));
	498	memcpy(root->fs_info->disk_super, &root->fs_info->super_copy,
	499	sizeof(root->fs_info->super_copy));
ccd467d6 CM	500
	501	btrfs_copy_pinned(root, &pinned_copy);
	502
78fae27e	503	mutex_unlock(&root->fs_info->trans_mutex);
8fd17795	504	mutex_unlock(&root->fs_info->fs_mutex);
79154b1b CM	505	ret = btrfs_write_and_wait_transaction(trans, root);
79154b1b CM	506	BUG_ON(ret);
79154b1b	507	write_ctree_super(trans, root);
8fd17795	508	mutex_lock(&root->fs_info->fs_mutex);
ccd467d6	509	btrfs_finish_extent_commit(trans, root, &pinned_copy);
78fae27e	510	mutex_lock(&root->fs_info->trans_mutex);
2c90e5d6 CM	511	cur_trans->commit_done = 1;
2c90e5d6 CM	512	wake_up(&cur_trans->commit_wait);
78fae27e	513	put_transaction(cur_trans);
79154b1b	514	put_transaction(cur_trans);
facda1e7 CM	515	if (root->fs_info->closing)
	516	list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
	517	else
	518	list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
78fae27e	519	mutex_unlock(&root->fs_info->trans_mutex);
2c90e5d6	520	kmem_cache_free(btrfs_trans_handle_cachep, trans);
79154b1b	521
facda1e7 CM	522	if (root->fs_info->closing) {
	523	mutex_unlock(&root->fs_info->fs_mutex);
	524	drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
	525	mutex_lock(&root->fs_info->fs_mutex);
	526	}
79154b1b CM	527	return ret;
	528	}
	529
08607c1b CM	530	void btrfs_transaction_cleaner(struct work_struct *work)
	531	{
	532	struct btrfs_fs_info *fs_info = container_of(work,
	533	struct btrfs_fs_info,
	534	trans_work.work);
	535
	536	struct btrfs_root *root = fs_info->tree_root;
	537	struct btrfs_transaction *cur;
	538	struct btrfs_trans_handle *trans;
facda1e7	539	struct list_head dirty_roots;
08607c1b CM	540	unsigned long now;
	541	unsigned long delay = HZ * 30;
	542	int ret;
	543
facda1e7	544	INIT_LIST_HEAD(&dirty_roots);
08607c1b CM	545	mutex_lock(&root->fs_info->fs_mutex);
	546	mutex_lock(&root->fs_info->trans_mutex);
	547	cur = root->fs_info->running_transaction;
	548	if (!cur) {
	549	mutex_unlock(&root->fs_info->trans_mutex);
	550	goto out;
	551	}
	552	now = get_seconds();
	553	if (now < cur->start_time \|\| now - cur->start_time < 30) {
	554	mutex_unlock(&root->fs_info->trans_mutex);
	555	delay = HZ * 5;
	556	goto out;
	557	}
	558	mutex_unlock(&root->fs_info->trans_mutex);
6702ed49	559	btrfs_defrag_dirty_roots(root->fs_info);
08607c1b CM	560	trans = btrfs_start_transaction(root, 1);
	561	ret = btrfs_commit_transaction(trans, root);
	562	out:
	563	mutex_unlock(&root->fs_info->fs_mutex);
ad693af6 CM	564
	565	mutex_lock(&root->fs_info->trans_mutex);
	566	list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
	567	mutex_unlock(&root->fs_info->trans_mutex);
	568
	569	if (!list_empty(&dirty_roots)) {
	570	drop_dirty_roots(root, &dirty_roots);
	571	}
08607c1b CM	572	btrfs_transaction_queue_work(root, delay);
	573	}
	574
	575	void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
	576	{
	577	queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
	578	}
	579
	580	void btrfs_transaction_flush_work(struct btrfs_root *root)
	581	{
	582	cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work);
	583	flush_workqueue(trans_wq);
	584	}
	585
	586	void __init btrfs_init_transaction_sys(void)
	587	{
	588	trans_wq = create_workqueue("btrfs");
	589	}
	590
	591	void __exit btrfs_exit_transaction_sys(void)
	592	{
	593	destroy_workqueue(trans_wq);
	594	}
	595