[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 (WARN_ON(btrfs_header_nritems(leaf) == 0))
					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)); /* -ENOMEM */
}

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); /* Logic error */

		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)
		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
	 * 4 items for inode item update (in the worst case)
	 * 1 items for slack space if we need do truncation
	 * 1 item for free space object
	 * 3 items for pre-allocation
	 */
	trans->bytes_reserved = btrfs_calc_trans_metadata_size(root, 10);
	ret = btrfs_block_rsv_add(root, trans->block_rsv,
				  trans->bytes_reserved,
				  BTRFS_RESERVE_NO_FLUSH);
	if (ret)
		goto out;
	trace_btrfs_space_reservation(root->fs_info, "ino_cache",
				      trans->transid, trans->bytes_reserved, 1);
again:
	inode = lookup_free_ino_inode(root, path);
	if (IS_ERR(inode) && (PTR_ERR(inode) != -ENOENT || retry)) {
		ret = PTR_ERR(inode);
		goto out_release;
	}

	if (IS_ERR(inode)) {
		BUG_ON(retry); /* Logic error */
		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);
	if (ret) {
		btrfs_abort_transaction(trans, root, ret);
		goto out_put;
	}

	if (i_size_read(inode) > 0) {
		ret = btrfs_truncate_free_space_cache(root, trans, inode);
		if (ret) {
			if (ret != -ENOSPC)
				btrfs_abort_transaction(trans, root, 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, inode);
out_put:
	iput(inode);
out_release:
	trace_btrfs_space_reservation(root->fs_info, "ino_cache",
				      trans->transid, trans->bytes_reserved, 0);
	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); /* Corruption */
	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
fae7f21c	81	if (WARN_ON(btrfs_header_nritems(leaf) == 0))
581bb050	82	break;
581bb050 LZ	83
	84	/*
	85	* Save the key so we can advances forward
	86	* in the next search.
	87	*/
	88	btrfs_item_key_to_cpu(leaf, &key, 0);
945d8962	89	btrfs_release_path(path);
581bb050 LZ	90	root->cache_progress = last;
	91	mutex_unlock(&root->fs_commit_mutex);
	92	schedule_timeout(1);
	93	goto again;
	94	} else
	95	continue;
	96	}
	97
	98	btrfs_item_key_to_cpu(leaf, &key, slot);
	99
	100	if (key.type != BTRFS_INODE_ITEM_KEY)
	101	goto next;
	102
a47d6b70	103	if (key.objectid >= root->highest_objectid)
581bb050 LZ	104	break;
	105
	106	if (last != (u64)-1 && last + 1 != key.objectid) {
	107	__btrfs_add_free_space(ctl, last + 1,
	108	key.objectid - last - 1);
	109	wake_up(&root->cache_wait);
	110	}
	111
	112	last = key.objectid;
	113	next:
	114	path->slots[0]++;
	115	}
	116
a47d6b70	117	if (last < root->highest_objectid - 1) {
581bb050	118	__btrfs_add_free_space(ctl, last + 1,
a47d6b70	119	root->highest_objectid - last - 1);
581bb050 LZ	120	}
	121
	122	spin_lock(&root->cache_lock);
	123	root->cached = BTRFS_CACHE_FINISHED;
	124	spin_unlock(&root->cache_lock);
	125
	126	root->cache_progress = (u64)-1;
	127	btrfs_unpin_free_ino(root);
	128	out:
	129	wake_up(&root->cache_wait);
	130	mutex_unlock(&root->fs_commit_mutex);
	131
	132	btrfs_free_path(path);
	133
	134	return ret;
	135	}
	136
	137	static void start_caching(struct btrfs_root *root)
	138	{
a47d6b70	139	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
581bb050	140	struct task_struct *tsk;
82d5902d	141	int ret;
a47d6b70	142	u64 objectid;
581bb050	143
4b9465cb CM	144	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
	145	return;
	146
581bb050 LZ	147	spin_lock(&root->cache_lock);
	148	if (root->cached != BTRFS_CACHE_NO) {
	149	spin_unlock(&root->cache_lock);
	150	return;
	151	}
	152
	153	root->cached = BTRFS_CACHE_STARTED;
	154	spin_unlock(&root->cache_lock);
	155
82d5902d LZ	156	ret = load_free_ino_cache(root->fs_info, root);
	157	if (ret == 1) {
	158	spin_lock(&root->cache_lock);
	159	root->cached = BTRFS_CACHE_FINISHED;
	160	spin_unlock(&root->cache_lock);
	161	return;
	162	}
	163
a47d6b70 LZ	164	/*
	165	* It can be quite time-consuming to fill the cache by searching
	166	* through the extent tree, and this can keep ino allocation path
	167	* waiting. Therefore at start we quickly find out the highest
	168	* inode number and we know we can use inode numbers which fall in
	169	* [highest_ino + 1, BTRFS_LAST_FREE_OBJECTID].
	170	*/
	171	ret = btrfs_find_free_objectid(root, &objectid);
	172	if (!ret && objectid <= BTRFS_LAST_FREE_OBJECTID) {
	173	__btrfs_add_free_space(ctl, objectid,
	174	BTRFS_LAST_FREE_OBJECTID - objectid + 1);
	175	}
	176
581bb050 LZ	177	tsk = kthread_run(caching_kthread, root, "btrfs-ino-cache-%llu\n",
581bb050 LZ	178	root->root_key.objectid);
79787eaa	179	BUG_ON(IS_ERR(tsk)); /* -ENOMEM */
581bb050 LZ	180	}
	181
	182	int btrfs_find_free_ino(struct btrfs_root root, u64 objectid)
	183	{
4b9465cb CM	184	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
	185	return btrfs_find_free_objectid(root, objectid);
	186
581bb050 LZ	187	again:
	188	*objectid = btrfs_find_ino_for_alloc(root);
	189
	190	if (*objectid != 0)
	191	return 0;
	192
	193	start_caching(root);
	194
	195	wait_event(root->cache_wait,
	196	root->cached == BTRFS_CACHE_FINISHED \|\|
	197	root->free_ino_ctl->free_space > 0);
	198
	199	if (root->cached == BTRFS_CACHE_FINISHED &&
	200	root->free_ino_ctl->free_space == 0)
	201	return -ENOSPC;
	202	else
	203	goto again;
	204	}
	205
	206	void btrfs_return_ino(struct btrfs_root *root, u64 objectid)
	207	{
	208	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	209	struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;
4b9465cb CM	210
	211	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
	212	return;
	213
581bb050 LZ	214	again:
	215	if (root->cached == BTRFS_CACHE_FINISHED) {
	216	__btrfs_add_free_space(ctl, objectid, 1);
	217	} else {
	218	/*
	219	* If we are in the process of caching free ino chunks,
	220	* to avoid adding the same inode number to the free_ino
	221	* tree twice due to cross transaction, we'll leave it
	222	* in the pinned tree until a transaction is committed
	223	* or the caching work is done.
	224	*/
	225
	226	mutex_lock(&root->fs_commit_mutex);
	227	spin_lock(&root->cache_lock);
	228	if (root->cached == BTRFS_CACHE_FINISHED) {
	229	spin_unlock(&root->cache_lock);
	230	mutex_unlock(&root->fs_commit_mutex);
	231	goto again;
	232	}
	233	spin_unlock(&root->cache_lock);
	234
	235	start_caching(root);
	236
a47d6b70	237	if (objectid <= root->cache_progress \|\|
ff76b056	238	objectid >= root->highest_objectid)
581bb050 LZ	239	__btrfs_add_free_space(ctl, objectid, 1);
	240	else
	241	__btrfs_add_free_space(pinned, objectid, 1);
	242
	243	mutex_unlock(&root->fs_commit_mutex);
	244	}
	245	}
	246
	247	/*
	248	* When a transaction is committed, we'll move those inode numbers which
	249	* are smaller than root->cache_progress from pinned tree to free_ino tree,
	250	* and others will just be dropped, because the commit root we were
	251	* searching has changed.
	252	*
	253	* Must be called with root->fs_commit_mutex held
	254	*/
	255	void btrfs_unpin_free_ino(struct btrfs_root *root)
	256	{
	257	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	258	struct rb_root *rbroot = &root->free_ino_pinned->free_space_offset;
	259	struct btrfs_free_space *info;
	260	struct rb_node *n;
	261	u64 count;
	262
4b9465cb CM	263	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
	264	return;
	265
581bb050 LZ	266	while (1) {
	267	n = rb_first(rbroot);
	268	if (!n)
	269	break;
	270
	271	info = rb_entry(n, struct btrfs_free_space, offset_index);
79787eaa	272	BUG_ON(info->bitmap); /* Logic error */
581bb050 LZ	273
	274	if (info->offset > root->cache_progress)
	275	goto free;
	276	else if (info->offset + info->bytes > root->cache_progress)
	277	count = root->cache_progress - info->offset + 1;
	278	else
	279	count = info->bytes;
	280
	281	__btrfs_add_free_space(ctl, info->offset, count);
	282	free:
	283	rb_erase(&info->offset_index, rbroot);
	284	kfree(info);
	285	}
	286	}
	287
	288	#define INIT_THRESHOLD (((1024 * 32) / 2) / sizeof(struct btrfs_free_space))
	289	#define INODES_PER_BITMAP (PAGE_CACHE_SIZE * 8)
	290
	291	/*
	292	* The goal is to keep the memory used by the free_ino tree won't
	293	* exceed the memory if we use bitmaps only.
	294	*/
	295	static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
	296	{
	297	struct btrfs_free_space *info;
	298	struct rb_node *n;
	299	int max_ino;
	300	int max_bitmaps;
	301
	302	n = rb_last(&ctl->free_space_offset);
	303	if (!n) {
	304	ctl->extents_thresh = INIT_THRESHOLD;
	305	return;
	306	}
	307	info = rb_entry(n, struct btrfs_free_space, offset_index);
	308
	309	/*
	310	* Find the maximum inode number in the filesystem. Note we
	311	* ignore the fact that this can be a bitmap, because we are
	312	* not doing precise calculation.
	313	*/
	314	max_ino = info->bytes - 1;
	315
	316	max_bitmaps = ALIGN(max_ino, INODES_PER_BITMAP) / INODES_PER_BITMAP;
	317	if (max_bitmaps <= ctl->total_bitmaps) {
	318	ctl->extents_thresh = 0;
	319	return;
	320	}
	321
	322	ctl->extents_thresh = (max_bitmaps - ctl->total_bitmaps) *
	323	PAGE_CACHE_SIZE / sizeof(*info);
	324	}
	325
	326	/*
	327	* We don't fall back to bitmap, if we are below the extents threshold
	328	* or this chunk of inode numbers is a big one.
	329	*/
	330	static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
	331	struct btrfs_free_space *info)
	332	{
	333	if (ctl->free_extents < ctl->extents_thresh \|\|
	334	info->bytes > INODES_PER_BITMAP / 10)
	335	return false;
	336
337	return true;
338	}
339
340	static struct btrfs_free_space_op free_ino_op = {
341	.recalc_thresholds = recalculate_thresholds,
342	.use_bitmap = use_bitmap,
343	};
344
345	static void pinned_recalc_thresholds(struct btrfs_free_space_ctl *ctl)
346	{
347	}
348
349	static bool pinned_use_bitmap(struct btrfs_free_space_ctl *ctl,
350	struct btrfs_free_space *info)
351	{
352	/*
353	* We always use extents for two reasons:
354	*
355	* - The pinned tree is only used during the process of caching
356	* work.
357	* - Make code simpler. See btrfs_unpin_free_ino().
358	*/
359	return false;
360	}
361
362	static struct btrfs_free_space_op pinned_free_ino_op = {
363	.recalc_thresholds = pinned_recalc_thresholds,
364	.use_bitmap = pinned_use_bitmap,
365	};
366
367	void btrfs_init_free_ino_ctl(struct btrfs_root *root)
368	{
369	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
370	struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;
371
372	spin_lock_init(&ctl->tree_lock);
373	ctl->unit = 1;
374	ctl->start = 0;
375	ctl->private = NULL;
376	ctl->op = &free_ino_op;
377
378	/*
379	* Initially we allow to use 16K of ram to cache chunks of
380	* inode numbers before we resort to bitmaps. This is somewhat
381	* arbitrary, but it will be adjusted in runtime.
382	*/
383	ctl->extents_thresh = INIT_THRESHOLD;
384
385	spin_lock_init(&pinned->tree_lock);
386	pinned->unit = 1;
387	pinned->start = 0;
388	pinned->private = NULL;
389	pinned->extents_thresh = 0;
390	pinned->op = &pinned_free_ino_op;
391	}
392
82d5902d LZ	393	int btrfs_save_ino_cache(struct btrfs_root *root,
	394	struct btrfs_trans_handle *trans)
	395	{
	396	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	397	struct btrfs_path *path;
	398	struct inode *inode;
ba38eb4d MX	399	struct btrfs_block_rsv *rsv;
ba38eb4d MX	400	u64 num_bytes;
82d5902d LZ	401	u64 alloc_hint = 0;
	402	int ret;
	403	int prealloc;
	404	bool retry = false;
	405
ca456ae2	406	/* only fs tree and subvol/snap needs ino cache */
	407	if (root->root_key.objectid != BTRFS_FS_TREE_OBJECTID &&
	408	(root->root_key.objectid < BTRFS_FIRST_FREE_OBJECTID \|\|
	409	root->root_key.objectid > BTRFS_LAST_FREE_OBJECTID))
	410	return 0;
	411
d132a538	412	/* Don't save inode cache if we are deleting this root */
69e9c6c6	413	if (btrfs_root_refs(&root->root_item) == 0)
d132a538 JB	414	return 0;
d132a538 JB	415
4b9465cb CM	416	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
	417	return 0;
	418
82d5902d LZ	419	path = btrfs_alloc_path();
	420	if (!path)
	421	return -ENOMEM;
4b9465cb	422
ba38eb4d MX	423	rsv = trans->block_rsv;
	424	trans->block_rsv = &root->fs_info->trans_block_rsv;
	425
	426	num_bytes = trans->bytes_reserved;
	427	/*
	428	* 1 item for inode item insertion if need
7b61cd92 MX	429	* 4 items for inode item update (in the worst case)
7b61cd92 MX	430	* 1 items for slack space if we need do truncation
ba38eb4d MX	431	* 1 item for free space object
	432	* 3 items for pre-allocation
	433	*/
7b61cd92	434	trans->bytes_reserved = btrfs_calc_trans_metadata_size(root, 10);
08e007d2 MX	435	ret = btrfs_block_rsv_add(root, trans->block_rsv,
	436	trans->bytes_reserved,
	437	BTRFS_RESERVE_NO_FLUSH);
ba38eb4d MX	438	if (ret)
ba38eb4d MX	439	goto out;
e77266e4	440	trace_btrfs_space_reservation(root->fs_info, "ino_cache",
2bcc0328	441	trans->transid, trans->bytes_reserved, 1);
82d5902d LZ	442	again:
82d5902d LZ	443	inode = lookup_free_ino_inode(root, path);
79787eaa	444	if (IS_ERR(inode) && (PTR_ERR(inode) != -ENOENT \|\| retry)) {
82d5902d	445	ret = PTR_ERR(inode);
ba38eb4d	446	goto out_release;
82d5902d LZ	447	}
	448
	449	if (IS_ERR(inode)) {
79787eaa	450	BUG_ON(retry); /* Logic error */
82d5902d LZ	451	retry = true;
	452
	453	ret = create_free_ino_inode(root, trans, path);
	454	if (ret)
ba38eb4d	455	goto out_release;
82d5902d LZ	456	goto again;
	457	}
	458
	459	BTRFS_I(inode)->generation = 0;
	460	ret = btrfs_update_inode(trans, root, inode);
79787eaa JM	461	if (ret) {
	462	btrfs_abort_transaction(trans, root, ret);
	463	goto out_put;
	464	}
82d5902d LZ	465
82d5902d LZ	466	if (i_size_read(inode) > 0) {
74514323	467	ret = btrfs_truncate_free_space_cache(root, trans, inode);
79787eaa	468	if (ret) {
7cfa9e51 MX	469	if (ret != -ENOSPC)
7cfa9e51 MX	470	btrfs_abort_transaction(trans, root, ret);
82d5902d	471	goto out_put;
79787eaa	472	}
82d5902d LZ	473	}
	474
	475	spin_lock(&root->cache_lock);
	476	if (root->cached != BTRFS_CACHE_FINISHED) {
	477	ret = -1;
	478	spin_unlock(&root->cache_lock);
	479	goto out_put;
	480	}
	481	spin_unlock(&root->cache_lock);
	482
	483	spin_lock(&ctl->tree_lock);
	484	prealloc = sizeof(struct btrfs_free_space) * ctl->free_extents;
	485	prealloc = ALIGN(prealloc, PAGE_CACHE_SIZE);
	486	prealloc += ctl->total_bitmaps * PAGE_CACHE_SIZE;
	487	spin_unlock(&ctl->tree_lock);
	488
	489	/* Just to make sure we have enough space */
	490	prealloc += 8 * PAGE_CACHE_SIZE;
	491
c09544e0	492	ret = btrfs_delalloc_reserve_space(inode, prealloc);
82d5902d LZ	493	if (ret)
	494	goto out_put;
	495
	496	ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, prealloc,
	497	prealloc, prealloc, &alloc_hint);
c09544e0 JB	498	if (ret) {
c09544e0 JB	499	btrfs_delalloc_release_space(inode, prealloc);
82d5902d	500	goto out_put;
c09544e0	501	}
82d5902d LZ	502	btrfs_free_reserved_data_space(inode, prealloc);
82d5902d LZ	503
53645a91	504	ret = btrfs_write_out_ino_cache(root, trans, path, inode);
82d5902d LZ	505	out_put:
82d5902d LZ	506	iput(inode);
ba38eb4d	507	out_release:
e77266e4	508	trace_btrfs_space_reservation(root->fs_info, "ino_cache",
2bcc0328	509	trans->transid, trans->bytes_reserved, 0);
ba38eb4d	510	btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
82d5902d	511	out:
ba38eb4d MX	512	trans->block_rsv = rsv;
ba38eb4d MX	513	trans->bytes_reserved = num_bytes;
82d5902d LZ	514
	515	btrfs_free_path(path);
	516	return ret;
	517	}
	518
581bb050	519	static int btrfs_find_highest_objectid(struct btrfs_root root, u64 objectid)
5be6f7f1 CM	520	{
	521	struct btrfs_path *path;
	522	int ret;
5f39d397	523	struct extent_buffer *l;
5be6f7f1	524	struct btrfs_key search_key;
5f39d397	525	struct btrfs_key found_key;
5be6f7f1 CM	526	int slot;
	527
	528	path = btrfs_alloc_path();
db5b493a TI	529	if (!path)
db5b493a TI	530	return -ENOMEM;
5be6f7f1	531
6527cdbe ZY	532	search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
6527cdbe ZY	533	search_key.type = -1;
5be6f7f1 CM	534	search_key.offset = (u64)-1;
	535	ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
	536	if (ret < 0)
	537	goto error;
79787eaa	538	BUG_ON(ret == 0); /* Corruption */
5be6f7f1 CM	539	if (path->slots[0] > 0) {
5be6f7f1 CM	540	slot = path->slots[0] - 1;
5f39d397 CM	541	l = path->nodes[0];
5f39d397 CM	542	btrfs_item_key_to_cpu(l, &found_key, slot);
13a8a7c8 YZ	543	*objectid = max_t(u64, found_key.objectid,
13a8a7c8 YZ	544	BTRFS_FIRST_FREE_OBJECTID - 1);
5be6f7f1	545	} else {
13a8a7c8	546	*objectid = BTRFS_FIRST_FREE_OBJECTID - 1;
5be6f7f1 CM	547	}
	548	ret = 0;
	549	error:
	550	btrfs_free_path(path);
	551	return ret;
	552	}
	553
581bb050	554	int btrfs_find_free_objectid(struct btrfs_root root, u64 objectid)
9f5fae2f	555	{
9f5fae2f	556	int ret;
a2135011	557	mutex_lock(&root->objectid_mutex);
9f5fae2f	558
13a8a7c8	559	if (unlikely(root->highest_objectid < BTRFS_FIRST_FREE_OBJECTID)) {
581bb050 LZ	560	ret = btrfs_find_highest_objectid(root,
581bb050 LZ	561	&root->highest_objectid);
13a8a7c8 YZ	562	if (ret)
	563	goto out;
	564	}
5b21f2ed	565
13a8a7c8 YZ	566	if (unlikely(root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
	567	ret = -ENOSPC;
	568	goto out;
9f5fae2f	569	}
13a8a7c8 YZ	570
	571	*objectid = ++root->highest_objectid;
	572	ret = 0;
	573	out:
a2135011	574	mutex_unlock(&root->objectid_mutex);
9f5fae2f CM	575	return ret;
9f5fae2f CM	576	}