[linux-2.6-block.git] / fs / btrfs / inode-map.c

// SPDX-License-Identifier: GPL-2.0
/*
 * Copyright (C) 2007 Oracle.  All rights reserved.
 */

#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(fs_info, 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 = READA_FORWARD;

	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 */
	down_read(&fs_info->commit_root_sem);

	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->ino_cache_progress = last;
				up_read(&fs_info->commit_root_sem);
				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(fs_info, ctl, last + 1,
					       key.objectid - last - 1);
			wake_up(&root->ino_cache_wait);
		}

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

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

	spin_lock(&root->ino_cache_lock);
	root->ino_cache_state = BTRFS_CACHE_FINISHED;
	spin_unlock(&root->ino_cache_lock);

	root->ino_cache_progress = (u64)-1;
	btrfs_unpin_free_ino(root);
out:
	wake_up(&root->ino_cache_wait);
	up_read(&fs_info->commit_root_sem);

	btrfs_free_path(path);

	return ret;
}

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

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

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

	root->ino_cache_state = BTRFS_CACHE_STARTED;
	spin_unlock(&root->ino_cache_lock);

	ret = load_free_ino_cache(fs_info, root);
	if (ret == 1) {
		spin_lock(&root->ino_cache_lock);
		root->ino_cache_state = BTRFS_CACHE_FINISHED;
		spin_unlock(&root->ino_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(fs_info, ctl, objectid,
				       BTRFS_LAST_FREE_OBJECTID - objectid + 1);
	}

	tsk = kthread_run(caching_kthread, root, "btrfs-ino-cache-%llu",
			  root->root_key.objectid);
	if (IS_ERR(tsk)) {
		btrfs_warn(fs_info, "failed to start inode caching task");
		btrfs_clear_pending_and_info(fs_info, INODE_MAP_CACHE,
					     "disabling inode map caching");
	}
}

int btrfs_find_free_ino(struct btrfs_root *root, u64 *objectid)
{
	if (!btrfs_test_opt(root->fs_info, 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->ino_cache_wait,
		   root->ino_cache_state == BTRFS_CACHE_FINISHED ||
		   root->free_ino_ctl->free_space > 0);

	if (root->ino_cache_state == 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_fs_info *fs_info = root->fs_info;
	struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;

	if (!btrfs_test_opt(fs_info, INODE_MAP_CACHE))
		return;
again:
	if (root->ino_cache_state == BTRFS_CACHE_FINISHED) {
		__btrfs_add_free_space(fs_info, pinned, objectid, 1);
	} else {
		down_write(&fs_info->commit_root_sem);
		spin_lock(&root->ino_cache_lock);
		if (root->ino_cache_state == BTRFS_CACHE_FINISHED) {
			spin_unlock(&root->ino_cache_lock);
			up_write(&fs_info->commit_root_sem);
			goto again;
		}
		spin_unlock(&root->ino_cache_lock);

		start_caching(root);

		__btrfs_add_free_space(fs_info, pinned, objectid, 1);

		up_write(&fs_info->commit_root_sem);
	}
}

/*
 * When a transaction is committed, we'll move those inode numbers which are
 * smaller than root->ino_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_info->commit_root_sem 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;
	spinlock_t *rbroot_lock = &root->free_ino_pinned->tree_lock;
	struct btrfs_free_space *info;
	struct rb_node *n;
	u64 count;

	if (!btrfs_test_opt(root->fs_info, INODE_MAP_CACHE))
		return;

	while (1) {
		spin_lock(rbroot_lock);
		n = rb_first(rbroot);
		if (!n) {
			spin_unlock(rbroot_lock);
			break;
		}

		info = rb_entry(n, struct btrfs_free_space, offset_index);
		BUG_ON(info->bitmap); /* Logic error */

		if (info->offset > root->ino_cache_progress)
			count = 0;
		else
			count = min(root->ino_cache_progress - info->offset + 1,
				    info->bytes);

		rb_erase(&info->offset_index, rbroot);
		spin_unlock(rbroot_lock);
		if (count)
			__btrfs_add_free_space(root->fs_info, ctl,
					       info->offset, count);
		kmem_cache_free(btrfs_free_space_cachep, info);
	}
}

#define INIT_THRESHOLD	((SZ_32K / 2) / sizeof(struct btrfs_free_space))
#define INODES_PER_BITMAP (PAGE_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_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 const 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 const 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;
	INIT_LIST_HEAD(&ctl->trimming_ranges);
	mutex_init(&ctl->cache_writeout_mutex);

	/*
	 * 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_fs_info *fs_info = root->fs_info;
	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	struct btrfs_path *path;
	struct inode *inode;
	struct btrfs_block_rsv *rsv;
	struct extent_changeset *data_reserved = NULL;
	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(fs_info, INODE_MAP_CACHE))
		return 0;

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

	rsv = trans->block_rsv;
	trans->block_rsv = &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(fs_info, 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(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, ret);
		goto out_put;
	}

	if (i_size_read(inode) > 0) {
		ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
		if (ret) {
			if (ret != -ENOSPC)
				btrfs_abort_transaction(trans, ret);
			goto out_put;
		}
	}

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

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

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

	ret = btrfs_delalloc_reserve_space(inode, &data_reserved, 0, 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_extents(BTRFS_I(inode), prealloc, true);
		goto out_put;
	}

	ret = btrfs_write_out_ino_cache(root, trans, path, inode);
	btrfs_delalloc_release_extents(BTRFS_I(inode), prealloc, false);
out_put:
	iput(inode);
out_release:
	trace_btrfs_space_reservation(fs_info, "ino_cache", trans->transid,
				      trans->bytes_reserved, 0);
	btrfs_block_rsv_release(fs_info, trans->block_rsv,
				trans->bytes_reserved);
out:
	trans->block_rsv = rsv;
	trans->bytes_reserved = num_bytes;

	btrfs_free_path(path);
	extent_changeset_free(data_reserved);
	return ret;
}

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_LAST_FREE_OBJECTID)) {
		btrfs_warn(root->fs_info,
			   "the objectid of root %llu reaches its highest value",
			   root->root_key.objectid);
		ret = -ENOSPC;
		goto out;
	}

	*objectid = ++root->highest_objectid;
	ret = 0;
out:
	mutex_unlock(&root->objectid_mutex);
	return ret;
}
Commit	Line	Data
c1d7c514	1	// SPDX-License-Identifier: GPL-2.0
6cbd5570 CM	2	/*
6cbd5570 CM	3	* Copyright (C) 2007 Oracle. All rights reserved.
6cbd5570 CM	4	*/
6cbd5570 CM	5
581bb050 LZ	6	#include <linux/kthread.h>
	7	#include <linux/pagemap.h>
	8
9f5fae2f CM	9	#include "ctree.h"
9f5fae2f CM	10	#include "disk-io.h"
581bb050 LZ	11	#include "free-space-cache.h"
581bb050 LZ	12	#include "inode-map.h"
9f5fae2f CM	13	#include "transaction.h"
9f5fae2f CM	14
581bb050 LZ	15	static int caching_kthread(void *data)
	16	{
	17	struct btrfs_root *root = data;
	18	struct btrfs_fs_info *fs_info = root->fs_info;
	19	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	20	struct btrfs_key key;
	21	struct btrfs_path *path;
	22	struct extent_buffer *leaf;
	23	u64 last = (u64)-1;
	24	int slot;
	25	int ret;
	26
0b246afa	27	if (!btrfs_test_opt(fs_info, INODE_MAP_CACHE))
4b9465cb CM	28	return 0;
4b9465cb CM	29
581bb050 LZ	30	path = btrfs_alloc_path();
	31	if (!path)
	32	return -ENOMEM;
	33
	34	/* Since the commit root is read-only, we can safely skip locking. */
	35	path->skip_locking = 1;
	36	path->search_commit_root = 1;
e4058b54	37	path->reada = READA_FORWARD;
581bb050 LZ	38
	39	key.objectid = BTRFS_FIRST_FREE_OBJECTID;
	40	key.offset = 0;
	41	key.type = BTRFS_INODE_ITEM_KEY;
	42	again:
	43	/* need to make sure the commit_root doesn't disappear */
9e351cc8	44	down_read(&fs_info->commit_root_sem);
581bb050 LZ	45
	46	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
	47	if (ret < 0)
	48	goto out;
	49
	50	while (1) {
7841cb28	51	if (btrfs_fs_closing(fs_info))
581bb050 LZ	52	goto out;
	53
	54	leaf = path->nodes[0];
	55	slot = path->slots[0];
a47d6b70	56	if (slot >= btrfs_header_nritems(leaf)) {
581bb050 LZ	57	ret = btrfs_next_leaf(root, path);
	58	if (ret < 0)
	59	goto out;
	60	else if (ret > 0)
	61	break;
	62
	63	if (need_resched() \|\|
	64	btrfs_transaction_in_commit(fs_info)) {
	65	leaf = path->nodes[0];
	66
fae7f21c	67	if (WARN_ON(btrfs_header_nritems(leaf) == 0))
581bb050	68	break;
581bb050 LZ	69
	70	/*
	71	* Save the key so we can advances forward
	72	* in the next search.
	73	*/
	74	btrfs_item_key_to_cpu(leaf, &key, 0);
945d8962	75	btrfs_release_path(path);
57cdc8db	76	root->ino_cache_progress = last;
9e351cc8	77	up_read(&fs_info->commit_root_sem);
581bb050 LZ	78	schedule_timeout(1);
	79	goto again;
	80	} else
	81	continue;
	82	}
	83
	84	btrfs_item_key_to_cpu(leaf, &key, slot);
	85
	86	if (key.type != BTRFS_INODE_ITEM_KEY)
	87	goto next;
	88
a47d6b70	89	if (key.objectid >= root->highest_objectid)
581bb050 LZ	90	break;
	91
	92	if (last != (u64)-1 && last + 1 != key.objectid) {
ab8d0fc4	93	__btrfs_add_free_space(fs_info, ctl, last + 1,
581bb050	94	key.objectid - last - 1);
57cdc8db	95	wake_up(&root->ino_cache_wait);
581bb050 LZ	96	}
	97
	98	last = key.objectid;
	99	next:
	100	path->slots[0]++;
	101	}
	102
a47d6b70	103	if (last < root->highest_objectid - 1) {
ab8d0fc4	104	__btrfs_add_free_space(fs_info, ctl, last + 1,
a47d6b70	105	root->highest_objectid - last - 1);
581bb050 LZ	106	}
581bb050 LZ	107
57cdc8db DS	108	spin_lock(&root->ino_cache_lock);
	109	root->ino_cache_state = BTRFS_CACHE_FINISHED;
	110	spin_unlock(&root->ino_cache_lock);
581bb050	111
57cdc8db	112	root->ino_cache_progress = (u64)-1;
581bb050 LZ	113	btrfs_unpin_free_ino(root);
581bb050 LZ	114	out:
57cdc8db	115	wake_up(&root->ino_cache_wait);
9e351cc8	116	up_read(&fs_info->commit_root_sem);
581bb050 LZ	117
	118	btrfs_free_path(path);
	119
	120	return ret;
	121	}
	122
	123	static void start_caching(struct btrfs_root *root)
	124	{
ab8d0fc4	125	struct btrfs_fs_info *fs_info = root->fs_info;
a47d6b70	126	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
581bb050	127	struct task_struct *tsk;
82d5902d	128	int ret;
a47d6b70	129	u64 objectid;
581bb050	130
ab8d0fc4	131	if (!btrfs_test_opt(fs_info, INODE_MAP_CACHE))
4b9465cb CM	132	return;
4b9465cb CM	133
57cdc8db DS	134	spin_lock(&root->ino_cache_lock);
	135	if (root->ino_cache_state != BTRFS_CACHE_NO) {
	136	spin_unlock(&root->ino_cache_lock);
581bb050 LZ	137	return;
	138	}
	139
57cdc8db DS	140	root->ino_cache_state = BTRFS_CACHE_STARTED;
57cdc8db DS	141	spin_unlock(&root->ino_cache_lock);
581bb050	142
ab8d0fc4	143	ret = load_free_ino_cache(fs_info, root);
82d5902d	144	if (ret == 1) {
57cdc8db DS	145	spin_lock(&root->ino_cache_lock);
	146	root->ino_cache_state = BTRFS_CACHE_FINISHED;
	147	spin_unlock(&root->ino_cache_lock);
82d5902d LZ	148	return;
	149	}
	150
a47d6b70 LZ	151	/*
	152	* It can be quite time-consuming to fill the cache by searching
	153	* through the extent tree, and this can keep ino allocation path
	154	* waiting. Therefore at start we quickly find out the highest
	155	* inode number and we know we can use inode numbers which fall in
	156	* [highest_ino + 1, BTRFS_LAST_FREE_OBJECTID].
	157	*/
	158	ret = btrfs_find_free_objectid(root, &objectid);
	159	if (!ret && objectid <= BTRFS_LAST_FREE_OBJECTID) {
ab8d0fc4	160	__btrfs_add_free_space(fs_info, ctl, objectid,
a47d6b70 LZ	161	BTRFS_LAST_FREE_OBJECTID - objectid + 1);
	162	}
	163
67a77eb1	164	tsk = kthread_run(caching_kthread, root, "btrfs-ino-cache-%llu",
581bb050	165	root->root_key.objectid);
e60efa84	166	if (IS_ERR(tsk)) {
ab8d0fc4 JM	167	btrfs_warn(fs_info, "failed to start inode caching task");
ab8d0fc4 JM	168	btrfs_clear_pending_and_info(fs_info, INODE_MAP_CACHE,
0b246afa	169	"disabling inode map caching");
e60efa84	170	}
581bb050 LZ	171	}
	172
	173	int btrfs_find_free_ino(struct btrfs_root root, u64 objectid)
	174	{
3cdde224	175	if (!btrfs_test_opt(root->fs_info, INODE_MAP_CACHE))
4b9465cb CM	176	return btrfs_find_free_objectid(root, objectid);
4b9465cb CM	177
581bb050 LZ	178	again:
	179	*objectid = btrfs_find_ino_for_alloc(root);
	180
	181	if (*objectid != 0)
	182	return 0;
	183
	184	start_caching(root);
	185
57cdc8db DS	186	wait_event(root->ino_cache_wait,
57cdc8db DS	187	root->ino_cache_state == BTRFS_CACHE_FINISHED \|\|
581bb050 LZ	188	root->free_ino_ctl->free_space > 0);
581bb050 LZ	189
57cdc8db	190	if (root->ino_cache_state == BTRFS_CACHE_FINISHED &&
581bb050 LZ	191	root->free_ino_ctl->free_space == 0)
	192	return -ENOSPC;
	193	else
	194	goto again;
	195	}
	196
	197	void btrfs_return_ino(struct btrfs_root *root, u64 objectid)
	198	{
ab8d0fc4	199	struct btrfs_fs_info *fs_info = root->fs_info;
581bb050	200	struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;
4b9465cb	201
ab8d0fc4	202	if (!btrfs_test_opt(fs_info, INODE_MAP_CACHE))
4b9465cb	203	return;
581bb050	204	again:
57cdc8db	205	if (root->ino_cache_state == BTRFS_CACHE_FINISHED) {
ab8d0fc4	206	__btrfs_add_free_space(fs_info, pinned, objectid, 1);
581bb050	207	} else {
ab8d0fc4	208	down_write(&fs_info->commit_root_sem);
57cdc8db DS	209	spin_lock(&root->ino_cache_lock);
	210	if (root->ino_cache_state == BTRFS_CACHE_FINISHED) {
	211	spin_unlock(&root->ino_cache_lock);
ab8d0fc4	212	up_write(&fs_info->commit_root_sem);
581bb050 LZ	213	goto again;
581bb050 LZ	214	}
57cdc8db	215	spin_unlock(&root->ino_cache_lock);
581bb050 LZ	216
	217	start_caching(root);
	218
ab8d0fc4	219	__btrfs_add_free_space(fs_info, pinned, objectid, 1);
581bb050	220
ab8d0fc4	221	up_write(&fs_info->commit_root_sem);
581bb050 LZ	222	}
	223	}
	224
	225	/*
57cdc8db DS	226	* When a transaction is committed, we'll move those inode numbers which are
	227	* smaller than root->ino_cache_progress from pinned tree to free_ino tree, and
	228	* others will just be dropped, because the commit root we were searching has
	229	* changed.
581bb050	230	*
9e351cc8	231	* Must be called with root->fs_info->commit_root_sem held
581bb050 LZ	232	*/
	233	void btrfs_unpin_free_ino(struct btrfs_root *root)
	234	{
	235	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	236	struct rb_root *rbroot = &root->free_ino_pinned->free_space_offset;
ae9d8f17	237	spinlock_t *rbroot_lock = &root->free_ino_pinned->tree_lock;
581bb050 LZ	238	struct btrfs_free_space *info;
	239	struct rb_node *n;
	240	u64 count;
	241
3cdde224	242	if (!btrfs_test_opt(root->fs_info, INODE_MAP_CACHE))
4b9465cb CM	243	return;
4b9465cb CM	244
581bb050	245	while (1) {
ae9d8f17	246	spin_lock(rbroot_lock);
581bb050	247	n = rb_first(rbroot);
ae9d8f17 FM	248	if (!n) {
ae9d8f17 FM	249	spin_unlock(rbroot_lock);
581bb050	250	break;
ae9d8f17	251	}
581bb050 LZ	252
581bb050 LZ	253	info = rb_entry(n, struct btrfs_free_space, offset_index);
79787eaa	254	BUG_ON(info->bitmap); /* Logic error */
581bb050	255
57cdc8db	256	if (info->offset > root->ino_cache_progress)
bc931c0e	257	count = 0;
581bb050	258	else
bc931c0e GU	259	count = min(root->ino_cache_progress - info->offset + 1,
bc931c0e GU	260	info->bytes);
581bb050	261
581bb050	262	rb_erase(&info->offset_index, rbroot);
ae9d8f17	263	spin_unlock(rbroot_lock);
bc931c0e	264	if (count)
ab8d0fc4 JM	265	__btrfs_add_free_space(root->fs_info, ctl,
ab8d0fc4 JM	266	info->offset, count);
c3f4a168	267	kmem_cache_free(btrfs_free_space_cachep, info);
581bb050 LZ	268	}
	269	}
	270
ee22184b	271	#define INIT_THRESHOLD ((SZ_32K / 2) / sizeof(struct btrfs_free_space))
09cbfeaf	272	#define INODES_PER_BITMAP (PAGE_SIZE * 8)
581bb050 LZ	273
	274	/*
	275	* The goal is to keep the memory used by the free_ino tree won't
	276	* exceed the memory if we use bitmaps only.
	277	*/
	278	static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
	279	{
	280	struct btrfs_free_space *info;
	281	struct rb_node *n;
	282	int max_ino;
	283	int max_bitmaps;
	284
	285	n = rb_last(&ctl->free_space_offset);
	286	if (!n) {
	287	ctl->extents_thresh = INIT_THRESHOLD;
	288	return;
	289	}
	290	info = rb_entry(n, struct btrfs_free_space, offset_index);
	291
	292	/*
	293	* Find the maximum inode number in the filesystem. Note we
	294	* ignore the fact that this can be a bitmap, because we are
	295	* not doing precise calculation.
	296	*/
	297	max_ino = info->bytes - 1;
	298
	299	max_bitmaps = ALIGN(max_ino, INODES_PER_BITMAP) / INODES_PER_BITMAP;
	300	if (max_bitmaps <= ctl->total_bitmaps) {
	301	ctl->extents_thresh = 0;
	302	return;
	303	}
	304
	305	ctl->extents_thresh = (max_bitmaps - ctl->total_bitmaps) *
09cbfeaf	306	PAGE_SIZE / sizeof(*info);
581bb050 LZ	307	}
	308
	309	/*
	310	* We don't fall back to bitmap, if we are below the extents threshold
	311	* or this chunk of inode numbers is a big one.
	312	*/
	313	static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
	314	struct btrfs_free_space *info)
	315	{
	316	if (ctl->free_extents < ctl->extents_thresh \|\|
	317	info->bytes > INODES_PER_BITMAP / 10)
	318	return false;
	319
	320	return true;
	321	}
	322
20e5506b	323	static const struct btrfs_free_space_op free_ino_op = {
581bb050 LZ	324	.recalc_thresholds = recalculate_thresholds,
	325	.use_bitmap = use_bitmap,
	326	};
	327
	328	static void pinned_recalc_thresholds(struct btrfs_free_space_ctl *ctl)
	329	{
	330	}
	331
	332	static bool pinned_use_bitmap(struct btrfs_free_space_ctl *ctl,
	333	struct btrfs_free_space *info)
	334	{
	335	/*
	336	* We always use extents for two reasons:
	337	*
	338	* - The pinned tree is only used during the process of caching
	339	* work.
	340	* - Make code simpler. See btrfs_unpin_free_ino().
	341	*/
	342	return false;
	343	}
	344
20e5506b	345	static const struct btrfs_free_space_op pinned_free_ino_op = {
581bb050 LZ	346	.recalc_thresholds = pinned_recalc_thresholds,
	347	.use_bitmap = pinned_use_bitmap,
	348	};
	349
	350	void btrfs_init_free_ino_ctl(struct btrfs_root *root)
	351	{
	352	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	353	struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;
	354
	355	spin_lock_init(&ctl->tree_lock);
	356	ctl->unit = 1;
	357	ctl->start = 0;
	358	ctl->private = NULL;
	359	ctl->op = &free_ino_op;
55507ce3 FM	360	INIT_LIST_HEAD(&ctl->trimming_ranges);
55507ce3 FM	361	mutex_init(&ctl->cache_writeout_mutex);
581bb050 LZ	362
	363	/*
	364	* Initially we allow to use 16K of ram to cache chunks of
	365	* inode numbers before we resort to bitmaps. This is somewhat
	366	* arbitrary, but it will be adjusted in runtime.
	367	*/
	368	ctl->extents_thresh = INIT_THRESHOLD;
	369
	370	spin_lock_init(&pinned->tree_lock);
	371	pinned->unit = 1;
	372	pinned->start = 0;
	373	pinned->private = NULL;
	374	pinned->extents_thresh = 0;
	375	pinned->op = &pinned_free_ino_op;
	376	}
	377
82d5902d LZ	378	int btrfs_save_ino_cache(struct btrfs_root *root,
	379	struct btrfs_trans_handle *trans)
	380	{
0b246afa	381	struct btrfs_fs_info *fs_info = root->fs_info;
82d5902d LZ	382	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	383	struct btrfs_path *path;
	384	struct inode *inode;
ba38eb4d	385	struct btrfs_block_rsv *rsv;
364ecf36	386	struct extent_changeset *data_reserved = NULL;
ba38eb4d	387	u64 num_bytes;
82d5902d LZ	388	u64 alloc_hint = 0;
	389	int ret;
	390	int prealloc;
	391	bool retry = false;
	392
ca456ae2	393	/* only fs tree and subvol/snap needs ino cache */
	394	if (root->root_key.objectid != BTRFS_FS_TREE_OBJECTID &&
	395	(root->root_key.objectid < BTRFS_FIRST_FREE_OBJECTID \|\|
	396	root->root_key.objectid > BTRFS_LAST_FREE_OBJECTID))
	397	return 0;
	398
d132a538	399	/* Don't save inode cache if we are deleting this root */
69e9c6c6	400	if (btrfs_root_refs(&root->root_item) == 0)
d132a538 JB	401	return 0;
d132a538 JB	402
0b246afa	403	if (!btrfs_test_opt(fs_info, INODE_MAP_CACHE))
4b9465cb CM	404	return 0;
4b9465cb CM	405
82d5902d LZ	406	path = btrfs_alloc_path();
	407	if (!path)
	408	return -ENOMEM;
4b9465cb	409
ba38eb4d	410	rsv = trans->block_rsv;
0b246afa	411	trans->block_rsv = &fs_info->trans_block_rsv;
ba38eb4d MX	412
	413	num_bytes = trans->bytes_reserved;
	414	/*
	415	* 1 item for inode item insertion if need
7b61cd92 MX	416	* 4 items for inode item update (in the worst case)
7b61cd92 MX	417	* 1 items for slack space if we need do truncation
ba38eb4d MX	418	* 1 item for free space object
	419	* 3 items for pre-allocation
	420	*/
0b246afa	421	trans->bytes_reserved = btrfs_calc_trans_metadata_size(fs_info, 10);
08e007d2 MX	422	ret = btrfs_block_rsv_add(root, trans->block_rsv,
	423	trans->bytes_reserved,
	424	BTRFS_RESERVE_NO_FLUSH);
ba38eb4d MX	425	if (ret)
ba38eb4d MX	426	goto out;
0b246afa JM	427	trace_btrfs_space_reservation(fs_info, "ino_cache", trans->transid,
0b246afa JM	428	trans->bytes_reserved, 1);
82d5902d LZ	429	again:
82d5902d LZ	430	inode = lookup_free_ino_inode(root, path);
79787eaa	431	if (IS_ERR(inode) && (PTR_ERR(inode) != -ENOENT \|\| retry)) {
82d5902d	432	ret = PTR_ERR(inode);
ba38eb4d	433	goto out_release;
82d5902d LZ	434	}
	435
	436	if (IS_ERR(inode)) {
79787eaa	437	BUG_ON(retry); /* Logic error */
82d5902d LZ	438	retry = true;
	439
	440	ret = create_free_ino_inode(root, trans, path);
	441	if (ret)
ba38eb4d	442	goto out_release;
82d5902d LZ	443	goto again;
	444	}
	445
	446	BTRFS_I(inode)->generation = 0;
	447	ret = btrfs_update_inode(trans, root, inode);
79787eaa	448	if (ret) {
66642832	449	btrfs_abort_transaction(trans, ret);
79787eaa JM	450	goto out_put;
79787eaa JM	451	}
82d5902d LZ	452
82d5902d LZ	453	if (i_size_read(inode) > 0) {
77ab86bf	454	ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
79787eaa	455	if (ret) {
7cfa9e51	456	if (ret != -ENOSPC)
66642832	457	btrfs_abort_transaction(trans, ret);
82d5902d	458	goto out_put;
79787eaa	459	}
82d5902d LZ	460	}
82d5902d LZ	461
57cdc8db DS	462	spin_lock(&root->ino_cache_lock);
57cdc8db DS	463	if (root->ino_cache_state != BTRFS_CACHE_FINISHED) {
82d5902d	464	ret = -1;
57cdc8db	465	spin_unlock(&root->ino_cache_lock);
82d5902d LZ	466	goto out_put;
82d5902d LZ	467	}
57cdc8db	468	spin_unlock(&root->ino_cache_lock);
82d5902d LZ	469
	470	spin_lock(&ctl->tree_lock);
	471	prealloc = sizeof(struct btrfs_free_space) * ctl->free_extents;
09cbfeaf KS	472	prealloc = ALIGN(prealloc, PAGE_SIZE);
09cbfeaf KS	473	prealloc += ctl->total_bitmaps * PAGE_SIZE;
82d5902d LZ	474	spin_unlock(&ctl->tree_lock);
	475
	476	/* Just to make sure we have enough space */
09cbfeaf	477	prealloc += 8 * PAGE_SIZE;
82d5902d	478
364ecf36	479	ret = btrfs_delalloc_reserve_space(inode, &data_reserved, 0, prealloc);
82d5902d LZ	480	if (ret)
	481	goto out_put;
	482
	483	ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, prealloc,
	484	prealloc, prealloc, &alloc_hint);
c09544e0	485	if (ret) {
43b18595	486	btrfs_delalloc_release_extents(BTRFS_I(inode), prealloc, true);
82d5902d	487	goto out_put;
c09544e0	488	}
82d5902d	489
53645a91	490	ret = btrfs_write_out_ino_cache(root, trans, path, inode);
43b18595	491	btrfs_delalloc_release_extents(BTRFS_I(inode), prealloc, false);
82d5902d LZ	492	out_put:
82d5902d LZ	493	iput(inode);
ba38eb4d	494	out_release:
0b246afa JM	495	trace_btrfs_space_reservation(fs_info, "ino_cache", trans->transid,
0b246afa JM	496	trans->bytes_reserved, 0);
2ff7e61e JM	497	btrfs_block_rsv_release(fs_info, trans->block_rsv,
2ff7e61e JM	498	trans->bytes_reserved);
82d5902d	499	out:
ba38eb4d MX	500	trans->block_rsv = rsv;
ba38eb4d MX	501	trans->bytes_reserved = num_bytes;
82d5902d LZ	502
82d5902d LZ	503	btrfs_free_path(path);
364ecf36	504	extent_changeset_free(data_reserved);
82d5902d LZ	505	return ret;
	506	}
	507
f32e48e9	508	int btrfs_find_highest_objectid(struct btrfs_root root, u64 objectid)
5be6f7f1 CM	509	{
	510	struct btrfs_path *path;
	511	int ret;
5f39d397	512	struct extent_buffer *l;
5be6f7f1	513	struct btrfs_key search_key;
5f39d397	514	struct btrfs_key found_key;
5be6f7f1 CM	515	int slot;
	516
	517	path = btrfs_alloc_path();
db5b493a TI	518	if (!path)
db5b493a TI	519	return -ENOMEM;
5be6f7f1	520
6527cdbe ZY	521	search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
6527cdbe ZY	522	search_key.type = -1;
5be6f7f1 CM	523	search_key.offset = (u64)-1;
	524	ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
	525	if (ret < 0)
	526	goto error;
79787eaa	527	BUG_ON(ret == 0); /* Corruption */
5be6f7f1 CM	528	if (path->slots[0] > 0) {
5be6f7f1 CM	529	slot = path->slots[0] - 1;
5f39d397 CM	530	l = path->nodes[0];
5f39d397 CM	531	btrfs_item_key_to_cpu(l, &found_key, slot);
13a8a7c8 YZ	532	*objectid = max_t(u64, found_key.objectid,
13a8a7c8 YZ	533	BTRFS_FIRST_FREE_OBJECTID - 1);
5be6f7f1	534	} else {
13a8a7c8	535	*objectid = BTRFS_FIRST_FREE_OBJECTID - 1;
5be6f7f1 CM	536	}
	537	ret = 0;
	538	error:
	539	btrfs_free_path(path);
	540	return ret;
	541	}
	542
581bb050	543	int btrfs_find_free_objectid(struct btrfs_root root, u64 objectid)
9f5fae2f	544	{
9f5fae2f	545	int ret;
a2135011	546	mutex_lock(&root->objectid_mutex);
9f5fae2f	547
13a8a7c8	548	if (unlikely(root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
3c1d84b7 ST	549	btrfs_warn(root->fs_info,
	550	"the objectid of root %llu reaches its highest value",
	551	root->root_key.objectid);
13a8a7c8 YZ	552	ret = -ENOSPC;
13a8a7c8 YZ	553	goto out;
9f5fae2f	554	}
13a8a7c8 YZ	555
	556	*objectid = ++root->highest_objectid;
	557	ret = 0;
	558	out:
a2135011	559	mutex_unlock(&root->objectid_mutex);
9f5fae2f CM	560	return ret;
9f5fae2f CM	561	}