[linux.git] / fs / btrfs / inode-map.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/delay.h>
#include <linux/kthread.h>
#include <linux/pagemap.h>

#include "ctree.h"
#include "disk-io.h"
#include "free-space-cache.h"
#include "inode-map.h"
#include "transaction.h"

static int caching_kthread(void *data)
{
	struct btrfs_root *root = data;
	struct btrfs_fs_info *fs_info = root->fs_info;
	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	struct btrfs_key key;
	struct btrfs_path *path;
	struct extent_buffer *leaf;
	u64 last = (u64)-1;
	int slot;
	int ret;

	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
		return 0;

	path = btrfs_alloc_path();
	if (!path)
		return -ENOMEM;

	/* Since the commit root is read-only, we can safely skip locking. */
	path->skip_locking = 1;
	path->search_commit_root = 1;
	path->reada = 2;

	key.objectid = BTRFS_FIRST_FREE_OBJECTID;
	key.offset = 0;
	key.type = BTRFS_INODE_ITEM_KEY;
again:
	/* need to make sure the commit_root doesn't disappear */
	mutex_lock(&root->fs_commit_mutex);

	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
	if (ret < 0)
		goto out;

	while (1) {
		if (btrfs_fs_closing(fs_info))
			goto out;

		leaf = path->nodes[0];
		slot = path->slots[0];
		if (slot >= btrfs_header_nritems(leaf)) {
			ret = btrfs_next_leaf(root, path);
			if (ret < 0)
				goto out;
			else if (ret > 0)
				break;

			if (need_resched() ||
			    btrfs_transaction_in_commit(fs_info)) {
				leaf = path->nodes[0];

				if (btrfs_header_nritems(leaf) == 0) {
					WARN_ON(1);
					break;
				}

				/*
				 * Save the key so we can advances forward
				 * in the next search.
				 */
				btrfs_item_key_to_cpu(leaf, &key, 0);
				btrfs_release_path(path);
				root->cache_progress = last;
				mutex_unlock(&root->fs_commit_mutex);
				schedule_timeout(1);
				goto again;
			} else
				continue;
		}

		btrfs_item_key_to_cpu(leaf, &key, slot);

		if (key.type != BTRFS_INODE_ITEM_KEY)
			goto next;

		if (key.objectid >= root->highest_objectid)
			break;

		if (last != (u64)-1 && last + 1 != key.objectid) {
			__btrfs_add_free_space(ctl, last + 1,
					       key.objectid - last - 1);
			wake_up(&root->cache_wait);
		}

		last = key.objectid;
next:
		path->slots[0]++;
	}

	if (last < root->highest_objectid - 1) {
		__btrfs_add_free_space(ctl, last + 1,
				       root->highest_objectid - last - 1);
	}

	spin_lock(&root->cache_lock);
	root->cached = BTRFS_CACHE_FINISHED;
	spin_unlock(&root->cache_lock);

	root->cache_progress = (u64)-1;
	btrfs_unpin_free_ino(root);
out:
	wake_up(&root->cache_wait);
	mutex_unlock(&root->fs_commit_mutex);

	btrfs_free_path(path);

	return ret;
}

static void start_caching(struct btrfs_root *root)
{
	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	struct task_struct *tsk;
	int ret;
	u64 objectid;

	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
		return;

	spin_lock(&root->cache_lock);
	if (root->cached != BTRFS_CACHE_NO) {
		spin_unlock(&root->cache_lock);
		return;
	}

	root->cached = BTRFS_CACHE_STARTED;
	spin_unlock(&root->cache_lock);

	ret = load_free_ino_cache(root->fs_info, root);
	if (ret == 1) {
		spin_lock(&root->cache_lock);
		root->cached = BTRFS_CACHE_FINISHED;
		spin_unlock(&root->cache_lock);
		return;
	}

	/*
	 * It can be quite time-consuming to fill the cache by searching
	 * through the extent tree, and this can keep ino allocation path
	 * waiting. Therefore at start we quickly find out the highest
	 * inode number and we know we can use inode numbers which fall in
	 * [highest_ino + 1, BTRFS_LAST_FREE_OBJECTID].
	 */
	ret = btrfs_find_free_objectid(root, &objectid);
	if (!ret && objectid <= BTRFS_LAST_FREE_OBJECTID) {
		__btrfs_add_free_space(ctl, objectid,
				       BTRFS_LAST_FREE_OBJECTID - objectid + 1);
	}

	tsk = kthread_run(caching_kthread, root, "btrfs-ino-cache-%llu\n",
			  root->root_key.objectid);
	BUG_ON(IS_ERR(tsk));
}

int btrfs_find_free_ino(struct btrfs_root *root, u64 *objectid)
{
	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
		return btrfs_find_free_objectid(root, objectid);

again:
	*objectid = btrfs_find_ino_for_alloc(root);

	if (*objectid != 0)
		return 0;

	start_caching(root);

	wait_event(root->cache_wait,
		   root->cached == BTRFS_CACHE_FINISHED ||
		   root->free_ino_ctl->free_space > 0);

	if (root->cached == BTRFS_CACHE_FINISHED &&
	    root->free_ino_ctl->free_space == 0)
		return -ENOSPC;
	else
		goto again;
}

void btrfs_return_ino(struct btrfs_root *root, u64 objectid)
{
	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;

	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
		return;

again:
	if (root->cached == BTRFS_CACHE_FINISHED) {
		__btrfs_add_free_space(ctl, objectid, 1);
	} else {
		/*
		 * If we are in the process of caching free ino chunks,
		 * to avoid adding the same inode number to the free_ino
		 * tree twice due to cross transaction, we'll leave it
		 * in the pinned tree until a transaction is committed
		 * or the caching work is done.
		 */

		mutex_lock(&root->fs_commit_mutex);
		spin_lock(&root->cache_lock);
		if (root->cached == BTRFS_CACHE_FINISHED) {
			spin_unlock(&root->cache_lock);
			mutex_unlock(&root->fs_commit_mutex);
			goto again;
		}
		spin_unlock(&root->cache_lock);

		start_caching(root);

		if (objectid <= root->cache_progress ||
		    objectid > root->highest_objectid)
			__btrfs_add_free_space(ctl, objectid, 1);
		else
			__btrfs_add_free_space(pinned, objectid, 1);

		mutex_unlock(&root->fs_commit_mutex);
	}
}

/*
 * When a transaction is committed, we'll move those inode numbers which
 * are smaller than root->cache_progress from pinned tree to free_ino tree,
 * and others will just be dropped, because the commit root we were
 * searching has changed.
 *
 * Must be called with root->fs_commit_mutex held
 */
void btrfs_unpin_free_ino(struct btrfs_root *root)
{
	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	struct rb_root *rbroot = &root->free_ino_pinned->free_space_offset;
	struct btrfs_free_space *info;
	struct rb_node *n;
	u64 count;

	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
		return;

	while (1) {
		n = rb_first(rbroot);
		if (!n)
			break;

		info = rb_entry(n, struct btrfs_free_space, offset_index);
		BUG_ON(info->bitmap);

		if (info->offset > root->cache_progress)
			goto free;
		else if (info->offset + info->bytes > root->cache_progress)
			count = root->cache_progress - info->offset + 1;
		else
			count = info->bytes;

		__btrfs_add_free_space(ctl, info->offset, count);
free:
		rb_erase(&info->offset_index, rbroot);
		kfree(info);
	}
}

#define INIT_THRESHOLD	(((1024 * 32) / 2) / sizeof(struct btrfs_free_space))
#define INODES_PER_BITMAP (PAGE_CACHE_SIZE * 8)

/*
 * The goal is to keep the memory used by the free_ino tree won't
 * exceed the memory if we use bitmaps only.
 */
static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
{
	struct btrfs_free_space *info;
	struct rb_node *n;
	int max_ino;
	int max_bitmaps;

	n = rb_last(&ctl->free_space_offset);
	if (!n) {
		ctl->extents_thresh = INIT_THRESHOLD;
		return;
	}
	info = rb_entry(n, struct btrfs_free_space, offset_index);

	/*
	 * Find the maximum inode number in the filesystem. Note we
	 * ignore the fact that this can be a bitmap, because we are
	 * not doing precise calculation.
	 */
	max_ino = info->bytes - 1;

	max_bitmaps = ALIGN(max_ino, INODES_PER_BITMAP) / INODES_PER_BITMAP;
	if (max_bitmaps <= ctl->total_bitmaps) {
		ctl->extents_thresh = 0;
		return;
	}

	ctl->extents_thresh = (max_bitmaps - ctl->total_bitmaps) *
				PAGE_CACHE_SIZE / sizeof(*info);
}

/*
 * We don't fall back to bitmap, if we are below the extents threshold
 * or this chunk of inode numbers is a big one.
 */
static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
		       struct btrfs_free_space *info)
{
	if (ctl->free_extents < ctl->extents_thresh ||
	    info->bytes > INODES_PER_BITMAP / 10)
		return false;

	return true;
}

static struct btrfs_free_space_op free_ino_op = {
	.recalc_thresholds	= recalculate_thresholds,
	.use_bitmap		= use_bitmap,
};

static void pinned_recalc_thresholds(struct btrfs_free_space_ctl *ctl)
{
}

static bool pinned_use_bitmap(struct btrfs_free_space_ctl *ctl,
			      struct btrfs_free_space *info)
{
	/*
	 * We always use extents for two reasons:
	 *
	 * - The pinned tree is only used during the process of caching
	 *   work.
	 * - Make code simpler. See btrfs_unpin_free_ino().
	 */
	return false;
}

static struct btrfs_free_space_op pinned_free_ino_op = {
	.recalc_thresholds	= pinned_recalc_thresholds,
	.use_bitmap		= pinned_use_bitmap,
};

void btrfs_init_free_ino_ctl(struct btrfs_root *root)
{
	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;

	spin_lock_init(&ctl->tree_lock);
	ctl->unit = 1;
	ctl->start = 0;
	ctl->private = NULL;
	ctl->op = &free_ino_op;

	/*
	 * Initially we allow to use 16K of ram to cache chunks of
	 * inode numbers before we resort to bitmaps. This is somewhat
	 * arbitrary, but it will be adjusted in runtime.
	 */
	ctl->extents_thresh = INIT_THRESHOLD;

	spin_lock_init(&pinned->tree_lock);
	pinned->unit = 1;
	pinned->start = 0;
	pinned->private = NULL;
	pinned->extents_thresh = 0;
	pinned->op = &pinned_free_ino_op;
}

int btrfs_save_ino_cache(struct btrfs_root *root,
			 struct btrfs_trans_handle *trans)
{
	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	struct btrfs_path *path;
	struct inode *inode;
	struct btrfs_block_rsv *rsv;
	u64 num_bytes;
	u64 alloc_hint = 0;
	int ret;
	int prealloc;
	bool retry = false;

	/* only fs tree and subvol/snap needs ino cache */
	if (root->root_key.objectid != BTRFS_FS_TREE_OBJECTID &&
	    (root->root_key.objectid < BTRFS_FIRST_FREE_OBJECTID ||
	     root->root_key.objectid > BTRFS_LAST_FREE_OBJECTID))
		return 0;

	/* Don't save inode cache if we are deleting this root */
	if (btrfs_root_refs(&root->root_item) == 0 &&
	    root != root->fs_info->tree_root)
		return 0;

	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
		return 0;

	path = btrfs_alloc_path();
	if (!path)
		return -ENOMEM;

	rsv = trans->block_rsv;
	trans->block_rsv = &root->fs_info->trans_block_rsv;

	num_bytes = trans->bytes_reserved;
	/*
	 * 1 item for inode item insertion if need
	 * 3 items for inode item update (in the worst case)
	 * 1 item for free space object
	 * 3 items for pre-allocation
	 */
	trans->bytes_reserved = btrfs_calc_trans_metadata_size(root, 8);
	ret = btrfs_block_rsv_add_noflush(root, trans->block_rsv,
					  trans->bytes_reserved);
	if (ret)
		goto out;
again:
	inode = lookup_free_ino_inode(root, path);
	if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
		ret = PTR_ERR(inode);
		goto out_release;
	}

	if (IS_ERR(inode)) {
		BUG_ON(retry);
		retry = true;

		ret = create_free_ino_inode(root, trans, path);
		if (ret)
			goto out_release;
		goto again;
	}

	BTRFS_I(inode)->generation = 0;
	ret = btrfs_update_inode(trans, root, inode);
	WARN_ON(ret);

	if (i_size_read(inode) > 0) {
		ret = btrfs_truncate_free_space_cache(root, trans, path, inode);
		if (ret)
			goto out_put;
	}

	spin_lock(&root->cache_lock);
	if (root->cached != BTRFS_CACHE_FINISHED) {
		ret = -1;
		spin_unlock(&root->cache_lock);
		goto out_put;
	}
	spin_unlock(&root->cache_lock);

	spin_lock(&ctl->tree_lock);
	prealloc = sizeof(struct btrfs_free_space) * ctl->free_extents;
	prealloc = ALIGN(prealloc, PAGE_CACHE_SIZE);
	prealloc += ctl->total_bitmaps * PAGE_CACHE_SIZE;
	spin_unlock(&ctl->tree_lock);

	/* Just to make sure we have enough space */
	prealloc += 8 * PAGE_CACHE_SIZE;

	ret = btrfs_delalloc_reserve_space(inode, prealloc);
	if (ret)
		goto out_put;

	ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, prealloc,
					      prealloc, prealloc, &alloc_hint);
	if (ret) {
		btrfs_delalloc_release_space(inode, prealloc);
		goto out_put;
	}
	btrfs_free_reserved_data_space(inode, prealloc);

	ret = btrfs_write_out_ino_cache(root, trans, path);
out_put:
	iput(inode);
out_release:
	btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
out:
	trans->block_rsv = rsv;
	trans->bytes_reserved = num_bytes;

	btrfs_free_path(path);
	return ret;
}

static int btrfs_find_highest_objectid(struct btrfs_root *root, u64 *objectid)
{
	struct btrfs_path *path;
	int ret;
	struct extent_buffer *l;
	struct btrfs_key search_key;
	struct btrfs_key found_key;
	int slot;

	path = btrfs_alloc_path();
	if (!path)
		return -ENOMEM;

	search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
	search_key.type = -1;
	search_key.offset = (u64)-1;
	ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
	if (ret < 0)
		goto error;
	BUG_ON(ret == 0);
	if (path->slots[0] > 0) {
		slot = path->slots[0] - 1;
		l = path->nodes[0];
		btrfs_item_key_to_cpu(l, &found_key, slot);
		*objectid = max_t(u64, found_key.objectid,
				  BTRFS_FIRST_FREE_OBJECTID - 1);
	} else {
		*objectid = BTRFS_FIRST_FREE_OBJECTID - 1;
	}
	ret = 0;
error:
	btrfs_free_path(path);
	return ret;
}

int btrfs_find_free_objectid(struct btrfs_root *root, u64 *objectid)
{
	int ret;
	mutex_lock(&root->objectid_mutex);

	if (unlikely(root->highest_objectid < BTRFS_FIRST_FREE_OBJECTID)) {
		ret = btrfs_find_highest_objectid(root,
						  &root->highest_objectid);
		if (ret)
			goto out;
	}

	if (unlikely(root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
		ret = -ENOSPC;
		goto out;
	}

	*objectid = ++root->highest_objectid;
	ret = 0;
out:
	mutex_unlock(&root->objectid_mutex);
	return ret;
}
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
581bb050 LZ	19	#include <linux/delay.h>
	20	#include <linux/kthread.h>
	21	#include <linux/pagemap.h>
	22
9f5fae2f CM	23	#include "ctree.h"
9f5fae2f CM	24	#include "disk-io.h"
581bb050 LZ	25	#include "free-space-cache.h"
581bb050 LZ	26	#include "inode-map.h"
9f5fae2f CM	27	#include "transaction.h"
9f5fae2f CM	28
581bb050 LZ	29	static int caching_kthread(void *data)
	30	{
	31	struct btrfs_root *root = data;
	32	struct btrfs_fs_info *fs_info = root->fs_info;
	33	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	34	struct btrfs_key key;
	35	struct btrfs_path *path;
	36	struct extent_buffer *leaf;
	37	u64 last = (u64)-1;
	38	int slot;
	39	int ret;
	40
4b9465cb CM	41	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
	42	return 0;
	43
581bb050 LZ	44	path = btrfs_alloc_path();
	45	if (!path)
	46	return -ENOMEM;
	47
	48	/* Since the commit root is read-only, we can safely skip locking. */
	49	path->skip_locking = 1;
	50	path->search_commit_root = 1;
	51	path->reada = 2;
	52
	53	key.objectid = BTRFS_FIRST_FREE_OBJECTID;
	54	key.offset = 0;
	55	key.type = BTRFS_INODE_ITEM_KEY;
	56	again:
	57	/* need to make sure the commit_root doesn't disappear */
	58	mutex_lock(&root->fs_commit_mutex);
	59
	60	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
	61	if (ret < 0)
	62	goto out;
	63
	64	while (1) {
7841cb28	65	if (btrfs_fs_closing(fs_info))
581bb050 LZ	66	goto out;
	67
	68	leaf = path->nodes[0];
	69	slot = path->slots[0];
a47d6b70	70	if (slot >= btrfs_header_nritems(leaf)) {
581bb050 LZ	71	ret = btrfs_next_leaf(root, path);
	72	if (ret < 0)
	73	goto out;
	74	else if (ret > 0)
	75	break;
	76
	77	if (need_resched() \|\|
	78	btrfs_transaction_in_commit(fs_info)) {
	79	leaf = path->nodes[0];
	80
	81	if (btrfs_header_nritems(leaf) == 0) {
	82	WARN_ON(1);
	83	break;
	84	}
	85
	86	/*
	87	* Save the key so we can advances forward
	88	* in the next search.
	89	*/
	90	btrfs_item_key_to_cpu(leaf, &key, 0);
945d8962	91	btrfs_release_path(path);
581bb050 LZ	92	root->cache_progress = last;
	93	mutex_unlock(&root->fs_commit_mutex);
	94	schedule_timeout(1);
	95	goto again;
	96	} else
	97	continue;
	98	}
	99
	100	btrfs_item_key_to_cpu(leaf, &key, slot);
	101
	102	if (key.type != BTRFS_INODE_ITEM_KEY)
	103	goto next;
	104
a47d6b70	105	if (key.objectid >= root->highest_objectid)
581bb050 LZ	106	break;
	107
	108	if (last != (u64)-1 && last + 1 != key.objectid) {
	109	__btrfs_add_free_space(ctl, last + 1,
	110	key.objectid - last - 1);
	111	wake_up(&root->cache_wait);
	112	}
	113
	114	last = key.objectid;
	115	next:
	116	path->slots[0]++;
	117	}
	118
a47d6b70	119	if (last < root->highest_objectid - 1) {
581bb050	120	__btrfs_add_free_space(ctl, last + 1,
a47d6b70	121	root->highest_objectid - last - 1);
581bb050 LZ	122	}
	123
	124	spin_lock(&root->cache_lock);
	125	root->cached = BTRFS_CACHE_FINISHED;
	126	spin_unlock(&root->cache_lock);
	127
	128	root->cache_progress = (u64)-1;
	129	btrfs_unpin_free_ino(root);
	130	out:
	131	wake_up(&root->cache_wait);
	132	mutex_unlock(&root->fs_commit_mutex);
	133
	134	btrfs_free_path(path);
	135
	136	return ret;
	137	}
	138
	139	static void start_caching(struct btrfs_root *root)
	140	{
a47d6b70	141	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
581bb050	142	struct task_struct *tsk;
82d5902d	143	int ret;
a47d6b70	144	u64 objectid;
581bb050	145
4b9465cb CM	146	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
	147	return;
	148
581bb050 LZ	149	spin_lock(&root->cache_lock);
	150	if (root->cached != BTRFS_CACHE_NO) {
	151	spin_unlock(&root->cache_lock);
	152	return;
	153	}
	154
	155	root->cached = BTRFS_CACHE_STARTED;
	156	spin_unlock(&root->cache_lock);
	157
82d5902d LZ	158	ret = load_free_ino_cache(root->fs_info, root);
	159	if (ret == 1) {
	160	spin_lock(&root->cache_lock);
	161	root->cached = BTRFS_CACHE_FINISHED;
	162	spin_unlock(&root->cache_lock);
	163	return;
	164	}
	165
a47d6b70 LZ	166	/*
	167	* It can be quite time-consuming to fill the cache by searching
	168	* through the extent tree, and this can keep ino allocation path
	169	* waiting. Therefore at start we quickly find out the highest
	170	* inode number and we know we can use inode numbers which fall in
	171	* [highest_ino + 1, BTRFS_LAST_FREE_OBJECTID].
	172	*/
	173	ret = btrfs_find_free_objectid(root, &objectid);
	174	if (!ret && objectid <= BTRFS_LAST_FREE_OBJECTID) {
	175	__btrfs_add_free_space(ctl, objectid,
	176	BTRFS_LAST_FREE_OBJECTID - objectid + 1);
	177	}
	178
581bb050 LZ	179	tsk = kthread_run(caching_kthread, root, "btrfs-ino-cache-%llu\n",
	180	root->root_key.objectid);
	181	BUG_ON(IS_ERR(tsk));
	182	}
	183
	184	int btrfs_find_free_ino(struct btrfs_root root, u64 objectid)
	185	{
4b9465cb CM	186	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
	187	return btrfs_find_free_objectid(root, objectid);
	188
581bb050 LZ	189	again:
	190	*objectid = btrfs_find_ino_for_alloc(root);
	191
	192	if (*objectid != 0)
	193	return 0;
	194
	195	start_caching(root);
	196
	197	wait_event(root->cache_wait,
	198	root->cached == BTRFS_CACHE_FINISHED \|\|
	199	root->free_ino_ctl->free_space > 0);
	200
	201	if (root->cached == BTRFS_CACHE_FINISHED &&
	202	root->free_ino_ctl->free_space == 0)
	203	return -ENOSPC;
	204	else
	205	goto again;
	206	}
	207
	208	void btrfs_return_ino(struct btrfs_root *root, u64 objectid)
	209	{
	210	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	211	struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;
4b9465cb CM	212
	213	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
	214	return;
	215
581bb050 LZ	216	again:
	217	if (root->cached == BTRFS_CACHE_FINISHED) {
	218	__btrfs_add_free_space(ctl, objectid, 1);
	219	} else {
	220	/*
	221	* If we are in the process of caching free ino chunks,
	222	* to avoid adding the same inode number to the free_ino
	223	* tree twice due to cross transaction, we'll leave it
	224	* in the pinned tree until a transaction is committed
	225	* or the caching work is done.
	226	*/
	227
	228	mutex_lock(&root->fs_commit_mutex);
	229	spin_lock(&root->cache_lock);
	230	if (root->cached == BTRFS_CACHE_FINISHED) {
	231	spin_unlock(&root->cache_lock);
	232	mutex_unlock(&root->fs_commit_mutex);
	233	goto again;
	234	}
	235	spin_unlock(&root->cache_lock);
	236
	237	start_caching(root);
	238
a47d6b70 LZ	239	if (objectid <= root->cache_progress \|\|
a47d6b70 LZ	240	objectid > root->highest_objectid)
581bb050 LZ	241	__btrfs_add_free_space(ctl, objectid, 1);
	242	else
	243	__btrfs_add_free_space(pinned, objectid, 1);
	244
	245	mutex_unlock(&root->fs_commit_mutex);
	246	}
	247	}
	248
	249	/*
	250	* When a transaction is committed, we'll move those inode numbers which
	251	* are smaller than root->cache_progress from pinned tree to free_ino tree,
	252	* and others will just be dropped, because the commit root we were
	253	* searching has changed.
	254	*
	255	* Must be called with root->fs_commit_mutex held
	256	*/
	257	void btrfs_unpin_free_ino(struct btrfs_root *root)
	258	{
	259	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	260	struct rb_root *rbroot = &root->free_ino_pinned->free_space_offset;
	261	struct btrfs_free_space *info;
	262	struct rb_node *n;
	263	u64 count;
	264
4b9465cb CM	265	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
	266	return;
	267
581bb050 LZ	268	while (1) {
	269	n = rb_first(rbroot);
	270	if (!n)
	271	break;
	272
	273	info = rb_entry(n, struct btrfs_free_space, offset_index);
	274	BUG_ON(info->bitmap);
	275
	276	if (info->offset > root->cache_progress)
	277	goto free;
	278	else if (info->offset + info->bytes > root->cache_progress)
	279	count = root->cache_progress - info->offset + 1;
	280	else
	281	count = info->bytes;
	282
	283	__btrfs_add_free_space(ctl, info->offset, count);
	284	free:
	285	rb_erase(&info->offset_index, rbroot);
	286	kfree(info);
	287	}
	288	}
	289
	290	#define INIT_THRESHOLD (((1024 * 32) / 2) / sizeof(struct btrfs_free_space))
	291	#define INODES_PER_BITMAP (PAGE_CACHE_SIZE * 8)
	292
	293	/*
	294	* The goal is to keep the memory used by the free_ino tree won't
	295	* exceed the memory if we use bitmaps only.
	296	*/
	297	static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
	298	{
	299	struct btrfs_free_space *info;
	300	struct rb_node *n;
	301	int max_ino;
	302	int max_bitmaps;
	303
	304	n = rb_last(&ctl->free_space_offset);
	305	if (!n) {
	306	ctl->extents_thresh = INIT_THRESHOLD;
	307	return;
	308	}
	309	info = rb_entry(n, struct btrfs_free_space, offset_index);
	310
	311	/*
	312	* Find the maximum inode number in the filesystem. Note we
	313	* ignore the fact that this can be a bitmap, because we are
	314	* not doing precise calculation.
	315	*/
	316	max_ino = info->bytes - 1;
	317
	318	max_bitmaps = ALIGN(max_ino, INODES_PER_BITMAP) / INODES_PER_BITMAP;
	319	if (max_bitmaps <= ctl->total_bitmaps) {
	320	ctl->extents_thresh = 0;
	321	return;
	322	}
	323
	324	ctl->extents_thresh = (max_bitmaps - ctl->total_bitmaps) *
	325	PAGE_CACHE_SIZE / sizeof(*info);
	326	}
	327
	328	/*
	329	* We don't fall back to bitmap, if we are below the extents threshold
	330	* or this chunk of inode numbers is a big one.
	331	*/
332	static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
333	struct btrfs_free_space *info)
334	{
335	if (ctl->free_extents < ctl->extents_thresh \|\|
336	info->bytes > INODES_PER_BITMAP / 10)
337	return false;
338
339	return true;
340	}
341
342	static struct btrfs_free_space_op free_ino_op = {
343	.recalc_thresholds = recalculate_thresholds,
344	.use_bitmap = use_bitmap,
345	};
346
347	static void pinned_recalc_thresholds(struct btrfs_free_space_ctl *ctl)
348	{
349	}
350
351	static bool pinned_use_bitmap(struct btrfs_free_space_ctl *ctl,
352	struct btrfs_free_space *info)
353	{
354	/*
355	* We always use extents for two reasons:
356	*
357	* - The pinned tree is only used during the process of caching
358	* work.
359	* - Make code simpler. See btrfs_unpin_free_ino().
360	*/
361	return false;
362	}
363
364	static struct btrfs_free_space_op pinned_free_ino_op = {
365	.recalc_thresholds = pinned_recalc_thresholds,
366	.use_bitmap = pinned_use_bitmap,
367	};
368
369	void btrfs_init_free_ino_ctl(struct btrfs_root *root)
370	{
371	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
372	struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;
373
374	spin_lock_init(&ctl->tree_lock);
375	ctl->unit = 1;
376	ctl->start = 0;
377	ctl->private = NULL;
378	ctl->op = &free_ino_op;
379
380	/*
381	* Initially we allow to use 16K of ram to cache chunks of
382	* inode numbers before we resort to bitmaps. This is somewhat
383	* arbitrary, but it will be adjusted in runtime.
384	*/
385	ctl->extents_thresh = INIT_THRESHOLD;
386
387	spin_lock_init(&pinned->tree_lock);
388	pinned->unit = 1;
389	pinned->start = 0;
390	pinned->private = NULL;
391	pinned->extents_thresh = 0;
392	pinned->op = &pinned_free_ino_op;
393	}
394
82d5902d LZ	395	int btrfs_save_ino_cache(struct btrfs_root *root,
	396	struct btrfs_trans_handle *trans)
	397	{
	398	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	399	struct btrfs_path *path;
	400	struct inode *inode;
ba38eb4d MX	401	struct btrfs_block_rsv *rsv;
ba38eb4d MX	402	u64 num_bytes;
82d5902d LZ	403	u64 alloc_hint = 0;
	404	int ret;
	405	int prealloc;
	406	bool retry = false;
	407
ca456ae2	408	/* only fs tree and subvol/snap needs ino cache */
	409	if (root->root_key.objectid != BTRFS_FS_TREE_OBJECTID &&
	410	(root->root_key.objectid < BTRFS_FIRST_FREE_OBJECTID \|\|
	411	root->root_key.objectid > BTRFS_LAST_FREE_OBJECTID))
	412	return 0;
	413
d132a538 JB	414	/* Don't save inode cache if we are deleting this root */
	415	if (btrfs_root_refs(&root->root_item) == 0 &&
	416	root != root->fs_info->tree_root)
	417	return 0;
	418
4b9465cb CM	419	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
	420	return 0;
	421
82d5902d LZ	422	path = btrfs_alloc_path();
	423	if (!path)
	424	return -ENOMEM;
4b9465cb	425
ba38eb4d MX	426	rsv = trans->block_rsv;
	427	trans->block_rsv = &root->fs_info->trans_block_rsv;
	428
	429	num_bytes = trans->bytes_reserved;
	430	/*
	431	* 1 item for inode item insertion if need
	432	* 3 items for inode item update (in the worst case)
	433	* 1 item for free space object
	434	* 3 items for pre-allocation
	435	*/
	436	trans->bytes_reserved = btrfs_calc_trans_metadata_size(root, 8);
	437	ret = btrfs_block_rsv_add_noflush(root, trans->block_rsv,
	438	trans->bytes_reserved);
	439	if (ret)
	440	goto out;
82d5902d LZ	441	again:
	442	inode = lookup_free_ino_inode(root, path);
	443	if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
	444	ret = PTR_ERR(inode);
ba38eb4d	445	goto out_release;
82d5902d LZ	446	}
	447
	448	if (IS_ERR(inode)) {
	449	BUG_ON(retry);
	450	retry = true;
	451
	452	ret = create_free_ino_inode(root, trans, path);
	453	if (ret)
ba38eb4d	454	goto out_release;
82d5902d LZ	455	goto again;
	456	}
	457
	458	BTRFS_I(inode)->generation = 0;
	459	ret = btrfs_update_inode(trans, root, inode);
	460	WARN_ON(ret);
	461
	462	if (i_size_read(inode) > 0) {
	463	ret = btrfs_truncate_free_space_cache(root, trans, path, inode);
	464	if (ret)
	465	goto out_put;
	466	}
	467
	468	spin_lock(&root->cache_lock);
	469	if (root->cached != BTRFS_CACHE_FINISHED) {
	470	ret = -1;
	471	spin_unlock(&root->cache_lock);
	472	goto out_put;
	473	}
	474	spin_unlock(&root->cache_lock);
	475
	476	spin_lock(&ctl->tree_lock);
	477	prealloc = sizeof(struct btrfs_free_space) * ctl->free_extents;
	478	prealloc = ALIGN(prealloc, PAGE_CACHE_SIZE);
	479	prealloc += ctl->total_bitmaps * PAGE_CACHE_SIZE;
	480	spin_unlock(&ctl->tree_lock);
	481
	482	/* Just to make sure we have enough space */
	483	prealloc += 8 * PAGE_CACHE_SIZE;
	484
c09544e0	485	ret = btrfs_delalloc_reserve_space(inode, prealloc);
82d5902d LZ	486	if (ret)
	487	goto out_put;
	488
	489	ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, prealloc,
	490	prealloc, prealloc, &alloc_hint);
c09544e0 JB	491	if (ret) {
c09544e0 JB	492	btrfs_delalloc_release_space(inode, prealloc);
82d5902d	493	goto out_put;
c09544e0	494	}
82d5902d LZ	495	btrfs_free_reserved_data_space(inode, prealloc);
82d5902d LZ	496
ba38eb4d	497	ret = btrfs_write_out_ino_cache(root, trans, path);
82d5902d LZ	498	out_put:
82d5902d LZ	499	iput(inode);
ba38eb4d MX	500	out_release:
ba38eb4d MX	501	btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
82d5902d	502	out:
ba38eb4d MX	503	trans->block_rsv = rsv;
ba38eb4d MX	504	trans->bytes_reserved = num_bytes;
82d5902d LZ	505
	506	btrfs_free_path(path);
	507	return ret;
	508	}
	509
581bb050	510	static int btrfs_find_highest_objectid(struct btrfs_root root, u64 objectid)
5be6f7f1 CM	511	{
	512	struct btrfs_path *path;
	513	int ret;
5f39d397	514	struct extent_buffer *l;
5be6f7f1	515	struct btrfs_key search_key;
5f39d397	516	struct btrfs_key found_key;
5be6f7f1 CM	517	int slot;
	518
	519	path = btrfs_alloc_path();
db5b493a TI	520	if (!path)
db5b493a TI	521	return -ENOMEM;
5be6f7f1	522
6527cdbe ZY	523	search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
6527cdbe ZY	524	search_key.type = -1;
5be6f7f1 CM	525	search_key.offset = (u64)-1;
	526	ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
	527	if (ret < 0)
	528	goto error;
	529	BUG_ON(ret == 0);
	530	if (path->slots[0] > 0) {
	531	slot = path->slots[0] - 1;
5f39d397 CM	532	l = path->nodes[0];
5f39d397 CM	533	btrfs_item_key_to_cpu(l, &found_key, slot);
13a8a7c8 YZ	534	*objectid = max_t(u64, found_key.objectid,
13a8a7c8 YZ	535	BTRFS_FIRST_FREE_OBJECTID - 1);
5be6f7f1	536	} else {
13a8a7c8	537	*objectid = BTRFS_FIRST_FREE_OBJECTID - 1;
5be6f7f1 CM	538	}
	539	ret = 0;
	540	error:
	541	btrfs_free_path(path);
	542	return ret;
	543	}
	544
581bb050	545	int btrfs_find_free_objectid(struct btrfs_root root, u64 objectid)
9f5fae2f	546	{
9f5fae2f	547	int ret;
a2135011	548	mutex_lock(&root->objectid_mutex);
9f5fae2f	549
13a8a7c8	550	if (unlikely(root->highest_objectid < BTRFS_FIRST_FREE_OBJECTID)) {
581bb050 LZ	551	ret = btrfs_find_highest_objectid(root,
581bb050 LZ	552	&root->highest_objectid);
13a8a7c8 YZ	553	if (ret)
	554	goto out;
	555	}
5b21f2ed	556
13a8a7c8 YZ	557	if (unlikely(root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
	558	ret = -ENOSPC;
	559	goto out;
9f5fae2f	560	}
13a8a7c8 YZ	561
	562	*objectid = ++root->highest_objectid;
	563	ret = 0;
	564	out:
a2135011	565	mutex_unlock(&root->objectid_mutex);
9f5fae2f CM	566	return ret;
9f5fae2f CM	567	}