2 * Copyright (C) 2007 Oracle. All rights reserved.
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.
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.
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.
19 #include <linux/delay.h>
20 #include <linux/kthread.h>
21 #include <linux/pagemap.h>
25 #include "free-space-cache.h"
26 #include "inode-map.h"
27 #include "transaction.h"
29 static int caching_kthread(void *data)
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;
35 struct btrfs_path *path;
36 struct extent_buffer *leaf;
41 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
44 path = btrfs_alloc_path();
48 /* Since the commit root is read-only, we can safely skip locking. */
49 path->skip_locking = 1;
50 path->search_commit_root = 1;
53 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
55 key.type = BTRFS_INODE_ITEM_KEY;
57 /* need to make sure the commit_root doesn't disappear */
58 mutex_lock(&root->fs_commit_mutex);
60 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
65 if (btrfs_fs_closing(fs_info))
68 leaf = path->nodes[0];
69 slot = path->slots[0];
70 if (slot >= btrfs_header_nritems(leaf)) {
71 ret = btrfs_next_leaf(root, path);
78 btrfs_transaction_in_commit(fs_info)) {
79 leaf = path->nodes[0];
81 if (WARN_ON(btrfs_header_nritems(leaf) == 0))
85 * Save the key so we can advances forward
88 btrfs_item_key_to_cpu(leaf, &key, 0);
89 btrfs_release_path(path);
90 root->cache_progress = last;
91 mutex_unlock(&root->fs_commit_mutex);
98 btrfs_item_key_to_cpu(leaf, &key, slot);
100 if (key.type != BTRFS_INODE_ITEM_KEY)
103 if (key.objectid >= root->highest_objectid)
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);
117 if (last < root->highest_objectid - 1) {
118 __btrfs_add_free_space(ctl, last + 1,
119 root->highest_objectid - last - 1);
122 spin_lock(&root->cache_lock);
123 root->cached = BTRFS_CACHE_FINISHED;
124 spin_unlock(&root->cache_lock);
126 root->cache_progress = (u64)-1;
127 btrfs_unpin_free_ino(root);
129 wake_up(&root->cache_wait);
130 mutex_unlock(&root->fs_commit_mutex);
132 btrfs_free_path(path);
137 static void start_caching(struct btrfs_root *root)
139 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
140 struct task_struct *tsk;
144 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
147 spin_lock(&root->cache_lock);
148 if (root->cached != BTRFS_CACHE_NO) {
149 spin_unlock(&root->cache_lock);
153 root->cached = BTRFS_CACHE_STARTED;
154 spin_unlock(&root->cache_lock);
156 ret = load_free_ino_cache(root->fs_info, root);
158 spin_lock(&root->cache_lock);
159 root->cached = BTRFS_CACHE_FINISHED;
160 spin_unlock(&root->cache_lock);
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].
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);
177 tsk = kthread_run(caching_kthread, root, "btrfs-ino-cache-%llu\n",
178 root->root_key.objectid);
179 BUG_ON(IS_ERR(tsk)); /* -ENOMEM */
182 int btrfs_find_free_ino(struct btrfs_root *root, u64 *objectid)
184 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
185 return btrfs_find_free_objectid(root, objectid);
188 *objectid = btrfs_find_ino_for_alloc(root);
195 wait_event(root->cache_wait,
196 root->cached == BTRFS_CACHE_FINISHED ||
197 root->free_ino_ctl->free_space > 0);
199 if (root->cached == BTRFS_CACHE_FINISHED &&
200 root->free_ino_ctl->free_space == 0)
206 void btrfs_return_ino(struct btrfs_root *root, u64 objectid)
208 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
209 struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;
211 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
215 if (root->cached == BTRFS_CACHE_FINISHED) {
216 __btrfs_add_free_space(ctl, objectid, 1);
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.
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);
233 spin_unlock(&root->cache_lock);
237 if (objectid <= root->cache_progress ||
238 objectid >= root->highest_objectid)
239 __btrfs_add_free_space(ctl, objectid, 1);
241 __btrfs_add_free_space(pinned, objectid, 1);
243 mutex_unlock(&root->fs_commit_mutex);
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.
253 * Must be called with root->fs_commit_mutex held
255 void btrfs_unpin_free_ino(struct btrfs_root *root)
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;
263 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
267 n = rb_first(rbroot);
271 info = rb_entry(n, struct btrfs_free_space, offset_index);
272 BUG_ON(info->bitmap); /* Logic error */
274 if (info->offset > root->cache_progress)
276 else if (info->offset + info->bytes > root->cache_progress)
277 count = root->cache_progress - info->offset + 1;
281 __btrfs_add_free_space(ctl, info->offset, count);
283 rb_erase(&info->offset_index, rbroot);
288 #define INIT_THRESHOLD (((1024 * 32) / 2) / sizeof(struct btrfs_free_space))
289 #define INODES_PER_BITMAP (PAGE_CACHE_SIZE * 8)
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.
295 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
297 struct btrfs_free_space *info;
302 n = rb_last(&ctl->free_space_offset);
304 ctl->extents_thresh = INIT_THRESHOLD;
307 info = rb_entry(n, struct btrfs_free_space, offset_index);
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.
314 max_ino = info->bytes - 1;
316 max_bitmaps = ALIGN(max_ino, INODES_PER_BITMAP) / INODES_PER_BITMAP;
317 if (max_bitmaps <= ctl->total_bitmaps) {
318 ctl->extents_thresh = 0;
322 ctl->extents_thresh = (max_bitmaps - ctl->total_bitmaps) *
323 PAGE_CACHE_SIZE / sizeof(*info);
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.
330 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
331 struct btrfs_free_space *info)
333 if (ctl->free_extents < ctl->extents_thresh ||
334 info->bytes > INODES_PER_BITMAP / 10)
340 static struct btrfs_free_space_op free_ino_op = {
341 .recalc_thresholds = recalculate_thresholds,
342 .use_bitmap = use_bitmap,
345 static void pinned_recalc_thresholds(struct btrfs_free_space_ctl *ctl)
349 static bool pinned_use_bitmap(struct btrfs_free_space_ctl *ctl,
350 struct btrfs_free_space *info)
353 * We always use extents for two reasons:
355 * - The pinned tree is only used during the process of caching
357 * - Make code simpler. See btrfs_unpin_free_ino().
362 static struct btrfs_free_space_op pinned_free_ino_op = {
363 .recalc_thresholds = pinned_recalc_thresholds,
364 .use_bitmap = pinned_use_bitmap,
367 void btrfs_init_free_ino_ctl(struct btrfs_root *root)
369 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
370 struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;
372 spin_lock_init(&ctl->tree_lock);
376 ctl->op = &free_ino_op;
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.
383 ctl->extents_thresh = INIT_THRESHOLD;
385 spin_lock_init(&pinned->tree_lock);
388 pinned->private = NULL;
389 pinned->extents_thresh = 0;
390 pinned->op = &pinned_free_ino_op;
393 int btrfs_save_ino_cache(struct btrfs_root *root,
394 struct btrfs_trans_handle *trans)
396 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
397 struct btrfs_path *path;
399 struct btrfs_block_rsv *rsv;
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))
412 /* Don't save inode cache if we are deleting this root */
413 if (btrfs_root_refs(&root->root_item) == 0)
416 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
419 path = btrfs_alloc_path();
423 rsv = trans->block_rsv;
424 trans->block_rsv = &root->fs_info->trans_block_rsv;
426 num_bytes = trans->bytes_reserved;
428 * 1 item for inode item insertion if need
429 * 4 items for inode item update (in the worst case)
430 * 1 items for slack space if we need do truncation
431 * 1 item for free space object
432 * 3 items for pre-allocation
434 trans->bytes_reserved = btrfs_calc_trans_metadata_size(root, 10);
435 ret = btrfs_block_rsv_add(root, trans->block_rsv,
436 trans->bytes_reserved,
437 BTRFS_RESERVE_NO_FLUSH);
440 trace_btrfs_space_reservation(root->fs_info, "ino_cache",
441 trans->transid, trans->bytes_reserved, 1);
443 inode = lookup_free_ino_inode(root, path);
444 if (IS_ERR(inode) && (PTR_ERR(inode) != -ENOENT || retry)) {
445 ret = PTR_ERR(inode);
450 BUG_ON(retry); /* Logic error */
453 ret = create_free_ino_inode(root, trans, path);
459 BTRFS_I(inode)->generation = 0;
460 ret = btrfs_update_inode(trans, root, inode);
462 btrfs_abort_transaction(trans, root, ret);
466 if (i_size_read(inode) > 0) {
467 ret = btrfs_truncate_free_space_cache(root, trans, inode);
470 btrfs_abort_transaction(trans, root, ret);
475 spin_lock(&root->cache_lock);
476 if (root->cached != BTRFS_CACHE_FINISHED) {
478 spin_unlock(&root->cache_lock);
481 spin_unlock(&root->cache_lock);
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);
489 /* Just to make sure we have enough space */
490 prealloc += 8 * PAGE_CACHE_SIZE;
492 ret = btrfs_delalloc_reserve_space(inode, prealloc);
496 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, prealloc,
497 prealloc, prealloc, &alloc_hint);
499 btrfs_delalloc_release_space(inode, prealloc);
502 btrfs_free_reserved_data_space(inode, prealloc);
504 ret = btrfs_write_out_ino_cache(root, trans, path, inode);
508 trace_btrfs_space_reservation(root->fs_info, "ino_cache",
509 trans->transid, trans->bytes_reserved, 0);
510 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
512 trans->block_rsv = rsv;
513 trans->bytes_reserved = num_bytes;
515 btrfs_free_path(path);
519 static int btrfs_find_highest_objectid(struct btrfs_root *root, u64 *objectid)
521 struct btrfs_path *path;
523 struct extent_buffer *l;
524 struct btrfs_key search_key;
525 struct btrfs_key found_key;
528 path = btrfs_alloc_path();
532 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
533 search_key.type = -1;
534 search_key.offset = (u64)-1;
535 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
538 BUG_ON(ret == 0); /* Corruption */
539 if (path->slots[0] > 0) {
540 slot = path->slots[0] - 1;
542 btrfs_item_key_to_cpu(l, &found_key, slot);
543 *objectid = max_t(u64, found_key.objectid,
544 BTRFS_FIRST_FREE_OBJECTID - 1);
546 *objectid = BTRFS_FIRST_FREE_OBJECTID - 1;
550 btrfs_free_path(path);
554 int btrfs_find_free_objectid(struct btrfs_root *root, u64 *objectid)
557 mutex_lock(&root->objectid_mutex);
559 if (unlikely(root->highest_objectid < BTRFS_FIRST_FREE_OBJECTID)) {
560 ret = btrfs_find_highest_objectid(root,
561 &root->highest_objectid);
566 if (unlikely(root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
571 *objectid = ++root->highest_objectid;
574 mutex_unlock(&root->objectid_mutex);