?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'pm-sleep'
[linux/fpc-iii.git] / fs / btrfs / free-space-cache.c
blob73f3de7a083c2aab0b48e0e0a53f61044dba794e
1 /*
2 * Copyright (C) 2008 Red Hat. 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
19 #include <linux/pagemap.h>
20 #include <linux/sched.h>
21 #include <linux/slab.h>
22 #include <linux/math64.h>
23 #include <linux/ratelimit.h>
24 #include "ctree.h"
25 #include "free-space-cache.h"
26 #include "transaction.h"
27 #include "disk-io.h"
28 #include "extent_io.h"
29 #include "inode-map.h"
30
31 #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8)
32 #define MAX_CACHE_BYTES_PER_GIG (32 * 1024)
33
34 static int link_free_space(struct btrfs_free_space_ctl *ctl,
35 struct btrfs_free_space *info);
36 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
37 struct btrfs_free_space *info);
38
39 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
40 struct btrfs_path *path,
41 u64 offset)
42 {
43 struct btrfs_key key;
44 struct btrfs_key location;
45 struct btrfs_disk_key disk_key;
46 struct btrfs_free_space_header *header;
47 struct extent_buffer *leaf;
48 struct inode *inode = NULL;
49 int ret;
50
51 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
52 key.offset = offset;
53 key.type = 0;
54
55 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
56 if (ret < 0)
57 return ERR_PTR(ret);
58 if (ret > 0) {
59 btrfs_release_path(path);
60 return ERR_PTR(-ENOENT);
61 }
62
63 leaf = path->nodes[0];
64 header = btrfs_item_ptr(leaf, path->slots[0],
65 struct btrfs_free_space_header);
66 btrfs_free_space_key(leaf, header, &disk_key);
67 btrfs_disk_key_to_cpu(&location, &disk_key);
68 btrfs_release_path(path);
69
70 inode = btrfs_iget(root->fs_info->sb, &location, root, NULL);
71 if (!inode)
72 return ERR_PTR(-ENOENT);
73 if (IS_ERR(inode))
74 return inode;
75 if (is_bad_inode(inode)) {
76 iput(inode);
77 return ERR_PTR(-ENOENT);
78 }
79
80 mapping_set_gfp_mask(inode->i_mapping,
81 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
82
83 return inode;
84 }
85
86 struct inode *lookup_free_space_inode(struct btrfs_root *root,
87 struct btrfs_block_group_cache
88 *block_group, struct btrfs_path *path)
89 {
90 struct inode *inode = NULL;
91 u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
92
93 spin_lock(&block_group->lock);
94 if (block_group->inode)
95 inode = igrab(block_group->inode);
96 spin_unlock(&block_group->lock);
97 if (inode)
98 return inode;
99
100 inode = __lookup_free_space_inode(root, path,
101 block_group->key.objectid);
102 if (IS_ERR(inode))
103 return inode;
104
105 spin_lock(&block_group->lock);
106 if (!((BTRFS_I(inode)->flags & flags) == flags)) {
107 btrfs_info(root->fs_info,
108 "Old style space inode found, converting.");
109 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
110 BTRFS_INODE_NODATACOW;
111 block_group->disk_cache_state = BTRFS_DC_CLEAR;
112 }
113
114 if (!block_group->iref) {
115 block_group->inode = igrab(inode);
116 block_group->iref = 1;
117 }
118 spin_unlock(&block_group->lock);
119
120 return inode;
121 }
122
123 static int __create_free_space_inode(struct btrfs_root *root,
124 struct btrfs_trans_handle *trans,
125 struct btrfs_path *path,
126 u64 ino, u64 offset)
127 {
128 struct btrfs_key key;
129 struct btrfs_disk_key disk_key;
130 struct btrfs_free_space_header *header;
131 struct btrfs_inode_item *inode_item;
132 struct extent_buffer *leaf;
133 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
134 int ret;
135
136 ret = btrfs_insert_empty_inode(trans, root, path, ino);
137 if (ret)
138 return ret;
139
140 /* We inline crc's for the free disk space cache */
141 if (ino != BTRFS_FREE_INO_OBJECTID)
142 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
143
144 leaf = path->nodes[0];
145 inode_item = btrfs_item_ptr(leaf, path->slots[0],
146 struct btrfs_inode_item);
147 btrfs_item_key(leaf, &disk_key, path->slots[0]);
148 memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
149 sizeof(*inode_item));
150 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
151 btrfs_set_inode_size(leaf, inode_item, 0);
152 btrfs_set_inode_nbytes(leaf, inode_item, 0);
153 btrfs_set_inode_uid(leaf, inode_item, 0);
154 btrfs_set_inode_gid(leaf, inode_item, 0);
155 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
156 btrfs_set_inode_flags(leaf, inode_item, flags);
157 btrfs_set_inode_nlink(leaf, inode_item, 1);
158 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
159 btrfs_set_inode_block_group(leaf, inode_item, offset);
160 btrfs_mark_buffer_dirty(leaf);
161 btrfs_release_path(path);
162
163 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
164 key.offset = offset;
165 key.type = 0;
166
167 ret = btrfs_insert_empty_item(trans, root, path, &key,
168 sizeof(struct btrfs_free_space_header));
169 if (ret < 0) {
170 btrfs_release_path(path);
171 return ret;
172 }
173 leaf = path->nodes[0];
174 header = btrfs_item_ptr(leaf, path->slots[0],
175 struct btrfs_free_space_header);
176 memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
177 btrfs_set_free_space_key(leaf, header, &disk_key);
178 btrfs_mark_buffer_dirty(leaf);
179 btrfs_release_path(path);
180
181 return 0;
182 }
183
184 int create_free_space_inode(struct btrfs_root *root,
185 struct btrfs_trans_handle *trans,
186 struct btrfs_block_group_cache *block_group,
187 struct btrfs_path *path)
188 {
189 int ret;
190 u64 ino;
191
192 ret = btrfs_find_free_objectid(root, &ino);
193 if (ret < 0)
194 return ret;
195
196 return __create_free_space_inode(root, trans, path, ino,
197 block_group->key.objectid);
198 }
199
200 int btrfs_check_trunc_cache_free_space(struct btrfs_root *root,
201 struct btrfs_block_rsv *rsv)
202 {
203 u64 needed_bytes;
204 int ret;
205
206 /* 1 for slack space, 1 for updating the inode */
207 needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
208 btrfs_calc_trans_metadata_size(root, 1);
209
210 spin_lock(&rsv->lock);
211 if (rsv->reserved < needed_bytes)
212 ret = -ENOSPC;
213 else
214 ret = 0;
215 spin_unlock(&rsv->lock);
216 return ret;
217 }
218
219 int btrfs_truncate_free_space_cache(struct btrfs_root *root,
220 struct btrfs_trans_handle *trans,
221 struct inode *inode)
222 {
223 int ret = 0;
224
225 btrfs_i_size_write(inode, 0);
226 truncate_pagecache(inode, 0);
227
228 /*
229 * We don't need an orphan item because truncating the free space cache
230 * will never be split across transactions.
231 */
232 ret = btrfs_truncate_inode_items(trans, root, inode,
233 0, BTRFS_EXTENT_DATA_KEY);
234 if (ret) {
235 btrfs_abort_transaction(trans, root, ret);
236 return ret;
237 }
238
239 ret = btrfs_update_inode(trans, root, inode);
240 if (ret)
241 btrfs_abort_transaction(trans, root, ret);
242
243 return ret;
244 }
245
246 static int readahead_cache(struct inode *inode)
247 {
248 struct file_ra_state *ra;
249 unsigned long last_index;
250
251 ra = kzalloc(sizeof(*ra), GFP_NOFS);
252 if (!ra)
253 return -ENOMEM;
254
255 file_ra_state_init(ra, inode->i_mapping);
256 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
257
258 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
259
260 kfree(ra);
261
262 return 0;
263 }
264
265 struct io_ctl {
266 void *cur, *orig;
267 struct page *page;
268 struct page **pages;
269 struct btrfs_root *root;
270 unsigned long size;
271 int index;
272 int num_pages;
273 unsigned check_crcs:1;
274 };
275
276 static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode,
277 struct btrfs_root *root)
278 {
279 memset(io_ctl, 0, sizeof(struct io_ctl));
280 io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
281 PAGE_CACHE_SHIFT;
282 io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages,
283 GFP_NOFS);
284 if (!io_ctl->pages)
285 return -ENOMEM;
286 io_ctl->root = root;
287 if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
288 io_ctl->check_crcs = 1;
289 return 0;
290 }
291
292 static void io_ctl_free(struct io_ctl *io_ctl)
293 {
294 kfree(io_ctl->pages);
295 }
296
297 static void io_ctl_unmap_page(struct io_ctl *io_ctl)
298 {
299 if (io_ctl->cur) {
300 kunmap(io_ctl->page);
301 io_ctl->cur = NULL;
302 io_ctl->orig = NULL;
303 }
304 }
305
306 static void io_ctl_map_page(struct io_ctl *io_ctl, int clear)
307 {
308 ASSERT(io_ctl->index < io_ctl->num_pages);
309 io_ctl->page = io_ctl->pages[io_ctl->index++];
310 io_ctl->cur = kmap(io_ctl->page);
311 io_ctl->orig = io_ctl->cur;
312 io_ctl->size = PAGE_CACHE_SIZE;
313 if (clear)
314 memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
315 }
316
317 static void io_ctl_drop_pages(struct io_ctl *io_ctl)
318 {
319 int i;
320
321 io_ctl_unmap_page(io_ctl);
322
323 for (i = 0; i < io_ctl->num_pages; i++) {
324 if (io_ctl->pages[i]) {
325 ClearPageChecked(io_ctl->pages[i]);
326 unlock_page(io_ctl->pages[i]);
327 page_cache_release(io_ctl->pages[i]);
328 }
329 }
330 }
331
332 static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode,
333 int uptodate)
334 {
335 struct page *page;
336 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
337 int i;
338
339 for (i = 0; i < io_ctl->num_pages; i++) {
340 page = find_or_create_page(inode->i_mapping, i, mask);
341 if (!page) {
342 io_ctl_drop_pages(io_ctl);
343 return -ENOMEM;
344 }
345 io_ctl->pages[i] = page;
346 if (uptodate && !PageUptodate(page)) {
347 btrfs_readpage(NULL, page);
348 lock_page(page);
349 if (!PageUptodate(page)) {
350 btrfs_err(BTRFS_I(inode)->root->fs_info,
351 "error reading free space cache");
352 io_ctl_drop_pages(io_ctl);
353 return -EIO;
354 }
355 }
356 }
357
358 for (i = 0; i < io_ctl->num_pages; i++) {
359 clear_page_dirty_for_io(io_ctl->pages[i]);
360 set_page_extent_mapped(io_ctl->pages[i]);
361 }
362
363 return 0;
364 }
365
366 static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation)
367 {
368 __le64 *val;
369
370 io_ctl_map_page(io_ctl, 1);
371
372 /*
373 * Skip the csum areas. If we don't check crcs then we just have a
374 * 64bit chunk at the front of the first page.
375 */
376 if (io_ctl->check_crcs) {
377 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
378 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
379 } else {
380 io_ctl->cur += sizeof(u64);
381 io_ctl->size -= sizeof(u64) * 2;
382 }
383
384 val = io_ctl->cur;
385 *val = cpu_to_le64(generation);
386 io_ctl->cur += sizeof(u64);
387 }
388
389 static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation)
390 {
391 __le64 *gen;
392
393 /*
394 * Skip the crc area. If we don't check crcs then we just have a 64bit
395 * chunk at the front of the first page.
396 */
397 if (io_ctl->check_crcs) {
398 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
399 io_ctl->size -= sizeof(u64) +
400 (sizeof(u32) * io_ctl->num_pages);
401 } else {
402 io_ctl->cur += sizeof(u64);
403 io_ctl->size -= sizeof(u64) * 2;
404 }
405
406 gen = io_ctl->cur;
407 if (le64_to_cpu(*gen) != generation) {
408 printk_ratelimited(KERN_ERR "BTRFS: space cache generation "
409 "(%Lu) does not match inode (%Lu)\n", *gen,
410 generation);
411 io_ctl_unmap_page(io_ctl);
412 return -EIO;
413 }
414 io_ctl->cur += sizeof(u64);
415 return 0;
416 }
417
418 static void io_ctl_set_crc(struct io_ctl *io_ctl, int index)
419 {
420 u32 *tmp;
421 u32 crc = ~(u32)0;
422 unsigned offset = 0;
423
424 if (!io_ctl->check_crcs) {
425 io_ctl_unmap_page(io_ctl);
426 return;
427 }
428
429 if (index == 0)
430 offset = sizeof(u32) * io_ctl->num_pages;
431
432 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
433 PAGE_CACHE_SIZE - offset);
434 btrfs_csum_final(crc, (char *)&crc);
435 io_ctl_unmap_page(io_ctl);
436 tmp = kmap(io_ctl->pages[0]);
437 tmp += index;
438 *tmp = crc;
439 kunmap(io_ctl->pages[0]);
440 }
441
442 static int io_ctl_check_crc(struct io_ctl *io_ctl, int index)
443 {
444 u32 *tmp, val;
445 u32 crc = ~(u32)0;
446 unsigned offset = 0;
447
448 if (!io_ctl->check_crcs) {
449 io_ctl_map_page(io_ctl, 0);
450 return 0;
451 }
452
453 if (index == 0)
454 offset = sizeof(u32) * io_ctl->num_pages;
455
456 tmp = kmap(io_ctl->pages[0]);
457 tmp += index;
458 val = *tmp;
459 kunmap(io_ctl->pages[0]);
460
461 io_ctl_map_page(io_ctl, 0);
462 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
463 PAGE_CACHE_SIZE - offset);
464 btrfs_csum_final(crc, (char *)&crc);
465 if (val != crc) {
466 printk_ratelimited(KERN_ERR "BTRFS: csum mismatch on free "
467 "space cache\n");
468 io_ctl_unmap_page(io_ctl);
469 return -EIO;
470 }
471
472 return 0;
473 }
474
475 static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes,
476 void *bitmap)
477 {
478 struct btrfs_free_space_entry *entry;
479
480 if (!io_ctl->cur)
481 return -ENOSPC;
482
483 entry = io_ctl->cur;
484 entry->offset = cpu_to_le64(offset);
485 entry->bytes = cpu_to_le64(bytes);
486 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
487 BTRFS_FREE_SPACE_EXTENT;
488 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
489 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
490
491 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
492 return 0;
493
494 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
495
496 /* No more pages to map */
497 if (io_ctl->index >= io_ctl->num_pages)
498 return 0;
499
500 /* map the next page */
501 io_ctl_map_page(io_ctl, 1);
502 return 0;
503 }
504
505 static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap)
506 {
507 if (!io_ctl->cur)
508 return -ENOSPC;
509
510 /*
511 * If we aren't at the start of the current page, unmap this one and
512 * map the next one if there is any left.
513 */
514 if (io_ctl->cur != io_ctl->orig) {
515 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
516 if (io_ctl->index >= io_ctl->num_pages)
517 return -ENOSPC;
518 io_ctl_map_page(io_ctl, 0);
519 }
520
521 memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
522 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
523 if (io_ctl->index < io_ctl->num_pages)
524 io_ctl_map_page(io_ctl, 0);
525 return 0;
526 }
527
528 static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl)
529 {
530 /*
531 * If we're not on the boundary we know we've modified the page and we
532 * need to crc the page.
533 */
534 if (io_ctl->cur != io_ctl->orig)
535 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
536 else
537 io_ctl_unmap_page(io_ctl);
538
539 while (io_ctl->index < io_ctl->num_pages) {
540 io_ctl_map_page(io_ctl, 1);
541 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
542 }
543 }
544
545 static int io_ctl_read_entry(struct io_ctl *io_ctl,
546 struct btrfs_free_space *entry, u8 *type)
547 {
548 struct btrfs_free_space_entry *e;
549 int ret;
550
551 if (!io_ctl->cur) {
552 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
553 if (ret)
554 return ret;
555 }
556
557 e = io_ctl->cur;
558 entry->offset = le64_to_cpu(e->offset);
559 entry->bytes = le64_to_cpu(e->bytes);
560 *type = e->type;
561 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
562 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
563
564 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
565 return 0;
566
567 io_ctl_unmap_page(io_ctl);
568
569 return 0;
570 }
571
572 static int io_ctl_read_bitmap(struct io_ctl *io_ctl,
573 struct btrfs_free_space *entry)
574 {
575 int ret;
576
577 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
578 if (ret)
579 return ret;
580
581 memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
582 io_ctl_unmap_page(io_ctl);
583
584 return 0;
585 }
586
587 /*
588 * Since we attach pinned extents after the fact we can have contiguous sections
589 * of free space that are split up in entries. This poses a problem with the
590 * tree logging stuff since it could have allocated across what appears to be 2
591 * entries since we would have merged the entries when adding the pinned extents
592 * back to the free space cache. So run through the space cache that we just
593 * loaded and merge contiguous entries. This will make the log replay stuff not
594 * blow up and it will make for nicer allocator behavior.
595 */
596 static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
597 {
598 struct btrfs_free_space *e, *prev = NULL;
599 struct rb_node *n;
600
601 again:
602 spin_lock(&ctl->tree_lock);
603 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
604 e = rb_entry(n, struct btrfs_free_space, offset_index);
605 if (!prev)
606 goto next;
607 if (e->bitmap || prev->bitmap)
608 goto next;
609 if (prev->offset + prev->bytes == e->offset) {
610 unlink_free_space(ctl, prev);
611 unlink_free_space(ctl, e);
612 prev->bytes += e->bytes;
613 kmem_cache_free(btrfs_free_space_cachep, e);
614 link_free_space(ctl, prev);
615 prev = NULL;
616 spin_unlock(&ctl->tree_lock);
617 goto again;
618 }
619 next:
620 prev = e;
621 }
622 spin_unlock(&ctl->tree_lock);
623 }
624
625 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
626 struct btrfs_free_space_ctl *ctl,
627 struct btrfs_path *path, u64 offset)
628 {
629 struct btrfs_free_space_header *header;
630 struct extent_buffer *leaf;
631 struct io_ctl io_ctl;
632 struct btrfs_key key;
633 struct btrfs_free_space *e, *n;
634 struct list_head bitmaps;
635 u64 num_entries;
636 u64 num_bitmaps;
637 u64 generation;
638 u8 type;
639 int ret = 0;
640
641 INIT_LIST_HEAD(&bitmaps);
642
643 /* Nothing in the space cache, goodbye */
644 if (!i_size_read(inode))
645 return 0;
646
647 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
648 key.offset = offset;
649 key.type = 0;
650
651 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
652 if (ret < 0)
653 return 0;
654 else if (ret > 0) {
655 btrfs_release_path(path);
656 return 0;
657 }
658
659 ret = -1;
660
661 leaf = path->nodes[0];
662 header = btrfs_item_ptr(leaf, path->slots[0],
663 struct btrfs_free_space_header);
664 num_entries = btrfs_free_space_entries(leaf, header);
665 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
666 generation = btrfs_free_space_generation(leaf, header);
667 btrfs_release_path(path);
668
669 if (BTRFS_I(inode)->generation != generation) {
670 btrfs_err(root->fs_info,
671 "free space inode generation (%llu) "
672 "did not match free space cache generation (%llu)",
673 BTRFS_I(inode)->generation, generation);
674 return 0;
675 }
676
677 if (!num_entries)
678 return 0;
679
680 ret = io_ctl_init(&io_ctl, inode, root);
681 if (ret)
682 return ret;
683
684 ret = readahead_cache(inode);
685 if (ret)
686 goto out;
687
688 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
689 if (ret)
690 goto out;
691
692 ret = io_ctl_check_crc(&io_ctl, 0);
693 if (ret)
694 goto free_cache;
695
696 ret = io_ctl_check_generation(&io_ctl, generation);
697 if (ret)
698 goto free_cache;
699
700 while (num_entries) {
701 e = kmem_cache_zalloc(btrfs_free_space_cachep,
702 GFP_NOFS);
703 if (!e)
704 goto free_cache;
705
706 ret = io_ctl_read_entry(&io_ctl, e, &type);
707 if (ret) {
708 kmem_cache_free(btrfs_free_space_cachep, e);
709 goto free_cache;
710 }
711
712 if (!e->bytes) {
713 kmem_cache_free(btrfs_free_space_cachep, e);
714 goto free_cache;
715 }
716
717 if (type == BTRFS_FREE_SPACE_EXTENT) {
718 spin_lock(&ctl->tree_lock);
719 ret = link_free_space(ctl, e);
720 spin_unlock(&ctl->tree_lock);
721 if (ret) {
722 btrfs_err(root->fs_info,
723 "Duplicate entries in free space cache, dumping");
724 kmem_cache_free(btrfs_free_space_cachep, e);
725 goto free_cache;
726 }
727 } else {
728 ASSERT(num_bitmaps);
729 num_bitmaps--;
730 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
731 if (!e->bitmap) {
732 kmem_cache_free(
733 btrfs_free_space_cachep, e);
734 goto free_cache;
735 }
736 spin_lock(&ctl->tree_lock);
737 ret = link_free_space(ctl, e);
738 ctl->total_bitmaps++;
739 ctl->op->recalc_thresholds(ctl);
740 spin_unlock(&ctl->tree_lock);
741 if (ret) {
742 btrfs_err(root->fs_info,
743 "Duplicate entries in free space cache, dumping");
744 kmem_cache_free(btrfs_free_space_cachep, e);
745 goto free_cache;
746 }
747 list_add_tail(&e->list, &bitmaps);
748 }
749
750 num_entries--;
751 }
752
753 io_ctl_unmap_page(&io_ctl);
754
755 /*
756 * We add the bitmaps at the end of the entries in order that
757 * the bitmap entries are added to the cache.
758 */
759 list_for_each_entry_safe(e, n, &bitmaps, list) {
760 list_del_init(&e->list);
761 ret = io_ctl_read_bitmap(&io_ctl, e);
762 if (ret)
763 goto free_cache;
764 }
765
766 io_ctl_drop_pages(&io_ctl);
767 merge_space_tree(ctl);
768 ret = 1;
769 out:
770 io_ctl_free(&io_ctl);
771 return ret;
772 free_cache:
773 io_ctl_drop_pages(&io_ctl);
774 __btrfs_remove_free_space_cache(ctl);
775 goto out;
776 }
777
778 int load_free_space_cache(struct btrfs_fs_info *fs_info,
779 struct btrfs_block_group_cache *block_group)
780 {
781 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
782 struct btrfs_root *root = fs_info->tree_root;
783 struct inode *inode;
784 struct btrfs_path *path;
785 int ret = 0;
786 bool matched;
787 u64 used = btrfs_block_group_used(&block_group->item);
788
789 /*
790 * If this block group has been marked to be cleared for one reason or
791 * another then we can't trust the on disk cache, so just return.
792 */
793 spin_lock(&block_group->lock);
794 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
795 spin_unlock(&block_group->lock);
796 return 0;
797 }
798 spin_unlock(&block_group->lock);
799
800 path = btrfs_alloc_path();
801 if (!path)
802 return 0;
803 path->search_commit_root = 1;
804 path->skip_locking = 1;
805
806 inode = lookup_free_space_inode(root, block_group, path);
807 if (IS_ERR(inode)) {
808 btrfs_free_path(path);
809 return 0;
810 }
811
812 /* We may have converted the inode and made the cache invalid. */
813 spin_lock(&block_group->lock);
814 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
815 spin_unlock(&block_group->lock);
816 btrfs_free_path(path);
817 goto out;
818 }
819 spin_unlock(&block_group->lock);
820
821 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
822 path, block_group->key.objectid);
823 btrfs_free_path(path);
824 if (ret <= 0)
825 goto out;
826
827 spin_lock(&ctl->tree_lock);
828 matched = (ctl->free_space == (block_group->key.offset - used -
829 block_group->bytes_super));
830 spin_unlock(&ctl->tree_lock);
831
832 if (!matched) {
833 __btrfs_remove_free_space_cache(ctl);
834 btrfs_err(fs_info, "block group %llu has wrong amount of free space",
835 block_group->key.objectid);
836 ret = -1;
837 }
838 out:
839 if (ret < 0) {
840 /* This cache is bogus, make sure it gets cleared */
841 spin_lock(&block_group->lock);
842 block_group->disk_cache_state = BTRFS_DC_CLEAR;
843 spin_unlock(&block_group->lock);
844 ret = 0;
845
846 btrfs_err(fs_info, "failed to load free space cache for block group %llu",
847 block_group->key.objectid);
848 }
849
850 iput(inode);
851 return ret;
852 }
853
854 /**
855 * __btrfs_write_out_cache - write out cached info to an inode
856 * @root - the root the inode belongs to
857 * @ctl - the free space cache we are going to write out
858 * @block_group - the block_group for this cache if it belongs to a block_group
859 * @trans - the trans handle
860 * @path - the path to use
861 * @offset - the offset for the key we'll insert
862 *
863 * This function writes out a free space cache struct to disk for quick recovery
864 * on mount. This will return 0 if it was successfull in writing the cache out,
865 * and -1 if it was not.
866 */
867 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
868 struct btrfs_free_space_ctl *ctl,
869 struct btrfs_block_group_cache *block_group,
870 struct btrfs_trans_handle *trans,
871 struct btrfs_path *path, u64 offset)
872 {
873 struct btrfs_free_space_header *header;
874 struct extent_buffer *leaf;
875 struct rb_node *node;
876 struct list_head *pos, *n;
877 struct extent_state *cached_state = NULL;
878 struct btrfs_free_cluster *cluster = NULL;
879 struct extent_io_tree *unpin = NULL;
880 struct io_ctl io_ctl;
881 struct list_head bitmap_list;
882 struct btrfs_key key;
883 u64 start, extent_start, extent_end, len;
884 int entries = 0;
885 int bitmaps = 0;
886 int ret;
887 int err = -1;
888
889 INIT_LIST_HEAD(&bitmap_list);
890
891 if (!i_size_read(inode))
892 return -1;
893
894 ret = io_ctl_init(&io_ctl, inode, root);
895 if (ret)
896 return -1;
897
898 /* Get the cluster for this block_group if it exists */
899 if (block_group && !list_empty(&block_group->cluster_list))
900 cluster = list_entry(block_group->cluster_list.next,
901 struct btrfs_free_cluster,
902 block_group_list);
903
904 /* Lock all pages first so we can lock the extent safely. */
905 io_ctl_prepare_pages(&io_ctl, inode, 0);
906
907 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
908 0, &cached_state);
909
910 node = rb_first(&ctl->free_space_offset);
911 if (!node && cluster) {
912 node = rb_first(&cluster->root);
913 cluster = NULL;
914 }
915
916 /* Make sure we can fit our crcs into the first page */
917 if (io_ctl.check_crcs &&
918 (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE)
919 goto out_nospc;
920
921 io_ctl_set_generation(&io_ctl, trans->transid);
922
923 /* Write out the extent entries */
924 while (node) {
925 struct btrfs_free_space *e;
926
927 e = rb_entry(node, struct btrfs_free_space, offset_index);
928 entries++;
929
930 ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes,
931 e->bitmap);
932 if (ret)
933 goto out_nospc;
934
935 if (e->bitmap) {
936 list_add_tail(&e->list, &bitmap_list);
937 bitmaps++;
938 }
939 node = rb_next(node);
940 if (!node && cluster) {
941 node = rb_first(&cluster->root);
942 cluster = NULL;
943 }
944 }
945
946 /*
947 * We want to add any pinned extents to our free space cache
948 * so we don't leak the space
949 */
950
951 /*
952 * We shouldn't have switched the pinned extents yet so this is the
953 * right one
954 */
955 unpin = root->fs_info->pinned_extents;
956
957 if (block_group)
958 start = block_group->key.objectid;
959
960 while (block_group && (start < block_group->key.objectid +
961 block_group->key.offset)) {
962 ret = find_first_extent_bit(unpin, start,
963 &extent_start, &extent_end,
964 EXTENT_DIRTY, NULL);
965 if (ret) {
966 ret = 0;
967 break;
968 }
969
970 /* This pinned extent is out of our range */
971 if (extent_start >= block_group->key.objectid +
972 block_group->key.offset)
973 break;
974
975 extent_start = max(extent_start, start);
976 extent_end = min(block_group->key.objectid +
977 block_group->key.offset, extent_end + 1);
978 len = extent_end - extent_start;
979
980 entries++;
981 ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL);
982 if (ret)
983 goto out_nospc;
984
985 start = extent_end;
986 }
987
988 /* Write out the bitmaps */
989 list_for_each_safe(pos, n, &bitmap_list) {
990 struct btrfs_free_space *entry =
991 list_entry(pos, struct btrfs_free_space, list);
992
993 ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap);
994 if (ret)
995 goto out_nospc;
996 list_del_init(&entry->list);
997 }
998
999 /* Zero out the rest of the pages just to make sure */
1000 io_ctl_zero_remaining_pages(&io_ctl);
1001
1002 ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
1003 0, i_size_read(inode), &cached_state);
1004 io_ctl_drop_pages(&io_ctl);
1005 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1006 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1007
1008 if (ret)
1009 goto out;
1010
1011 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1012 if (ret) {
1013 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1014 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1015 GFP_NOFS);
1016 goto out;
1017 }
1018
1019 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1020 key.offset = offset;
1021 key.type = 0;
1022
1023 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1024 if (ret < 0) {
1025 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1026 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1027 GFP_NOFS);
1028 goto out;
1029 }
1030 leaf = path->nodes[0];
1031 if (ret > 0) {
1032 struct btrfs_key found_key;
1033 ASSERT(path->slots[0]);
1034 path->slots[0]--;
1035 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1036 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1037 found_key.offset != offset) {
1038 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1039 inode->i_size - 1,
1040 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1041 NULL, GFP_NOFS);
1042 btrfs_release_path(path);
1043 goto out;
1044 }
1045 }
1046
1047 BTRFS_I(inode)->generation = trans->transid;
1048 header = btrfs_item_ptr(leaf, path->slots[0],
1049 struct btrfs_free_space_header);
1050 btrfs_set_free_space_entries(leaf, header, entries);
1051 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1052 btrfs_set_free_space_generation(leaf, header, trans->transid);
1053 btrfs_mark_buffer_dirty(leaf);
1054 btrfs_release_path(path);
1055
1056 err = 0;
1057 out:
1058 io_ctl_free(&io_ctl);
1059 if (err) {
1060 invalidate_inode_pages2(inode->i_mapping);
1061 BTRFS_I(inode)->generation = 0;
1062 }
1063 btrfs_update_inode(trans, root, inode);
1064 return err;
1065
1066 out_nospc:
1067 list_for_each_safe(pos, n, &bitmap_list) {
1068 struct btrfs_free_space *entry =
1069 list_entry(pos, struct btrfs_free_space, list);
1070 list_del_init(&entry->list);
1071 }
1072 io_ctl_drop_pages(&io_ctl);
1073 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1074 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1075 goto out;
1076 }
1077
1078 int btrfs_write_out_cache(struct btrfs_root *root,
1079 struct btrfs_trans_handle *trans,
1080 struct btrfs_block_group_cache *block_group,
1081 struct btrfs_path *path)
1082 {
1083 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1084 struct inode *inode;
1085 int ret = 0;
1086
1087 root = root->fs_info->tree_root;
1088
1089 spin_lock(&block_group->lock);
1090 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1091 spin_unlock(&block_group->lock);
1092 return 0;
1093 }
1094 spin_unlock(&block_group->lock);
1095
1096 inode = lookup_free_space_inode(root, block_group, path);
1097 if (IS_ERR(inode))
1098 return 0;
1099
1100 ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
1101 path, block_group->key.objectid);
1102 if (ret) {
1103 spin_lock(&block_group->lock);
1104 block_group->disk_cache_state = BTRFS_DC_ERROR;
1105 spin_unlock(&block_group->lock);
1106 ret = 0;
1107 #ifdef DEBUG
1108 btrfs_err(root->fs_info,
1109 "failed to write free space cache for block group %llu",
1110 block_group->key.objectid);
1111 #endif
1112 }
1113
1114 iput(inode);
1115 return ret;
1116 }
1117
1118 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1119 u64 offset)
1120 {
1121 ASSERT(offset >= bitmap_start);
1122 offset -= bitmap_start;
1123 return (unsigned long)(div_u64(offset, unit));
1124 }
1125
1126 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1127 {
1128 return (unsigned long)(div_u64(bytes, unit));
1129 }
1130
1131 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1132 u64 offset)
1133 {
1134 u64 bitmap_start;
1135 u64 bytes_per_bitmap;
1136
1137 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1138 bitmap_start = offset - ctl->start;
1139 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1140 bitmap_start *= bytes_per_bitmap;
1141 bitmap_start += ctl->start;
1142
1143 return bitmap_start;
1144 }
1145
1146 static int tree_insert_offset(struct rb_root *root, u64 offset,
1147 struct rb_node *node, int bitmap)
1148 {
1149 struct rb_node **p = &root->rb_node;
1150 struct rb_node *parent = NULL;
1151 struct btrfs_free_space *info;
1152
1153 while (*p) {
1154 parent = *p;
1155 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1156
1157 if (offset < info->offset) {
1158 p = &(*p)->rb_left;
1159 } else if (offset > info->offset) {
1160 p = &(*p)->rb_right;
1161 } else {
1162 /*
1163 * we could have a bitmap entry and an extent entry
1164 * share the same offset. If this is the case, we want
1165 * the extent entry to always be found first if we do a
1166 * linear search through the tree, since we want to have
1167 * the quickest allocation time, and allocating from an
1168 * extent is faster than allocating from a bitmap. So
1169 * if we're inserting a bitmap and we find an entry at
1170 * this offset, we want to go right, or after this entry
1171 * logically. If we are inserting an extent and we've
1172 * found a bitmap, we want to go left, or before
1173 * logically.
1174 */
1175 if (bitmap) {
1176 if (info->bitmap) {
1177 WARN_ON_ONCE(1);
1178 return -EEXIST;
1179 }
1180 p = &(*p)->rb_right;
1181 } else {
1182 if (!info->bitmap) {
1183 WARN_ON_ONCE(1);
1184 return -EEXIST;
1185 }
1186 p = &(*p)->rb_left;
1187 }
1188 }
1189 }
1190
1191 rb_link_node(node, parent, p);
1192 rb_insert_color(node, root);
1193
1194 return 0;
1195 }
1196
1197 /*
1198 * searches the tree for the given offset.
1199 *
1200 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1201 * want a section that has at least bytes size and comes at or after the given
1202 * offset.
1203 */
1204 static struct btrfs_free_space *
1205 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1206 u64 offset, int bitmap_only, int fuzzy)
1207 {
1208 struct rb_node *n = ctl->free_space_offset.rb_node;
1209 struct btrfs_free_space *entry, *prev = NULL;
1210
1211 /* find entry that is closest to the 'offset' */
1212 while (1) {
1213 if (!n) {
1214 entry = NULL;
1215 break;
1216 }
1217
1218 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1219 prev = entry;
1220
1221 if (offset < entry->offset)
1222 n = n->rb_left;
1223 else if (offset > entry->offset)
1224 n = n->rb_right;
1225 else
1226 break;
1227 }
1228
1229 if (bitmap_only) {
1230 if (!entry)
1231 return NULL;
1232 if (entry->bitmap)
1233 return entry;
1234
1235 /*
1236 * bitmap entry and extent entry may share same offset,
1237 * in that case, bitmap entry comes after extent entry.
1238 */
1239 n = rb_next(n);
1240 if (!n)
1241 return NULL;
1242 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1243 if (entry->offset != offset)
1244 return NULL;
1245
1246 WARN_ON(!entry->bitmap);
1247 return entry;
1248 } else if (entry) {
1249 if (entry->bitmap) {
1250 /*
1251 * if previous extent entry covers the offset,
1252 * we should return it instead of the bitmap entry
1253 */
1254 n = rb_prev(&entry->offset_index);
1255 if (n) {
1256 prev = rb_entry(n, struct btrfs_free_space,
1257 offset_index);
1258 if (!prev->bitmap &&
1259 prev->offset + prev->bytes > offset)
1260 entry = prev;
1261 }
1262 }
1263 return entry;
1264 }
1265
1266 if (!prev)
1267 return NULL;
1268
1269 /* find last entry before the 'offset' */
1270 entry = prev;
1271 if (entry->offset > offset) {
1272 n = rb_prev(&entry->offset_index);
1273 if (n) {
1274 entry = rb_entry(n, struct btrfs_free_space,
1275 offset_index);
1276 ASSERT(entry->offset <= offset);
1277 } else {
1278 if (fuzzy)
1279 return entry;
1280 else
1281 return NULL;
1282 }
1283 }
1284
1285 if (entry->bitmap) {
1286 n = rb_prev(&entry->offset_index);
1287 if (n) {
1288 prev = rb_entry(n, struct btrfs_free_space,
1289 offset_index);
1290 if (!prev->bitmap &&
1291 prev->offset + prev->bytes > offset)
1292 return prev;
1293 }
1294 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1295 return entry;
1296 } else if (entry->offset + entry->bytes > offset)
1297 return entry;
1298
1299 if (!fuzzy)
1300 return NULL;
1301
1302 while (1) {
1303 if (entry->bitmap) {
1304 if (entry->offset + BITS_PER_BITMAP *
1305 ctl->unit > offset)
1306 break;
1307 } else {
1308 if (entry->offset + entry->bytes > offset)
1309 break;
1310 }
1311
1312 n = rb_next(&entry->offset_index);
1313 if (!n)
1314 return NULL;
1315 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1316 }
1317 return entry;
1318 }
1319
1320 static inline void
1321 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1322 struct btrfs_free_space *info)
1323 {
1324 rb_erase(&info->offset_index, &ctl->free_space_offset);
1325 ctl->free_extents--;
1326 }
1327
1328 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1329 struct btrfs_free_space *info)
1330 {
1331 __unlink_free_space(ctl, info);
1332 ctl->free_space -= info->bytes;
1333 }
1334
1335 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1336 struct btrfs_free_space *info)
1337 {
1338 int ret = 0;
1339
1340 ASSERT(info->bytes || info->bitmap);
1341 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1342 &info->offset_index, (info->bitmap != NULL));
1343 if (ret)
1344 return ret;
1345
1346 ctl->free_space += info->bytes;
1347 ctl->free_extents++;
1348 return ret;
1349 }
1350
1351 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1352 {
1353 struct btrfs_block_group_cache *block_group = ctl->private;
1354 u64 max_bytes;
1355 u64 bitmap_bytes;
1356 u64 extent_bytes;
1357 u64 size = block_group->key.offset;
1358 u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1359 int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1360
1361 max_bitmaps = max(max_bitmaps, 1);
1362
1363 ASSERT(ctl->total_bitmaps <= max_bitmaps);
1364
1365 /*
1366 * The goal is to keep the total amount of memory used per 1gb of space
1367 * at or below 32k, so we need to adjust how much memory we allow to be
1368 * used by extent based free space tracking
1369 */
1370 if (size < 1024 * 1024 * 1024)
1371 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1372 else
1373 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1374 div64_u64(size, 1024 * 1024 * 1024);
1375
1376 /*
1377 * we want to account for 1 more bitmap than what we have so we can make
1378 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1379 * we add more bitmaps.
1380 */
1381 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1382
1383 if (bitmap_bytes >= max_bytes) {
1384 ctl->extents_thresh = 0;
1385 return;
1386 }
1387
1388 /*
1389 * we want the extent entry threshold to always be at most 1/2 the maxw
1390 * bytes we can have, or whatever is less than that.
1391 */
1392 extent_bytes = max_bytes - bitmap_bytes;
1393 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1394
1395 ctl->extents_thresh =
1396 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1397 }
1398
1399 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1400 struct btrfs_free_space *info,
1401 u64 offset, u64 bytes)
1402 {
1403 unsigned long start, count;
1404
1405 start = offset_to_bit(info->offset, ctl->unit, offset);
1406 count = bytes_to_bits(bytes, ctl->unit);
1407 ASSERT(start + count <= BITS_PER_BITMAP);
1408
1409 bitmap_clear(info->bitmap, start, count);
1410
1411 info->bytes -= bytes;
1412 }
1413
1414 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1415 struct btrfs_free_space *info, u64 offset,
1416 u64 bytes)
1417 {
1418 __bitmap_clear_bits(ctl, info, offset, bytes);
1419 ctl->free_space -= bytes;
1420 }
1421
1422 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1423 struct btrfs_free_space *info, u64 offset,
1424 u64 bytes)
1425 {
1426 unsigned long start, count;
1427
1428 start = offset_to_bit(info->offset, ctl->unit, offset);
1429 count = bytes_to_bits(bytes, ctl->unit);
1430 ASSERT(start + count <= BITS_PER_BITMAP);
1431
1432 bitmap_set(info->bitmap, start, count);
1433
1434 info->bytes += bytes;
1435 ctl->free_space += bytes;
1436 }
1437
1438 /*
1439 * If we can not find suitable extent, we will use bytes to record
1440 * the size of the max extent.
1441 */
1442 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1443 struct btrfs_free_space *bitmap_info, u64 *offset,
1444 u64 *bytes)
1445 {
1446 unsigned long found_bits = 0;
1447 unsigned long max_bits = 0;
1448 unsigned long bits, i;
1449 unsigned long next_zero;
1450 unsigned long extent_bits;
1451
1452 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1453 max_t(u64, *offset, bitmap_info->offset));
1454 bits = bytes_to_bits(*bytes, ctl->unit);
1455
1456 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1457 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1458 BITS_PER_BITMAP, i);
1459 extent_bits = next_zero - i;
1460 if (extent_bits >= bits) {
1461 found_bits = extent_bits;
1462 break;
1463 } else if (extent_bits > max_bits) {
1464 max_bits = extent_bits;
1465 }
1466 i = next_zero;
1467 }
1468
1469 if (found_bits) {
1470 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1471 *bytes = (u64)(found_bits) * ctl->unit;
1472 return 0;
1473 }
1474
1475 *bytes = (u64)(max_bits) * ctl->unit;
1476 return -1;
1477 }
1478
1479 /* Cache the size of the max extent in bytes */
1480 static struct btrfs_free_space *
1481 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1482 unsigned long align, u64 *max_extent_size)
1483 {
1484 struct btrfs_free_space *entry;
1485 struct rb_node *node;
1486 u64 tmp;
1487 u64 align_off;
1488 int ret;
1489
1490 if (!ctl->free_space_offset.rb_node)
1491 goto out;
1492
1493 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1494 if (!entry)
1495 goto out;
1496
1497 for (node = &entry->offset_index; node; node = rb_next(node)) {
1498 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1499 if (entry->bytes < *bytes) {
1500 if (entry->bytes > *max_extent_size)
1501 *max_extent_size = entry->bytes;
1502 continue;
1503 }
1504
1505 /* make sure the space returned is big enough
1506 * to match our requested alignment
1507 */
1508 if (*bytes >= align) {
1509 tmp = entry->offset - ctl->start + align - 1;
1510 do_div(tmp, align);
1511 tmp = tmp * align + ctl->start;
1512 align_off = tmp - entry->offset;
1513 } else {
1514 align_off = 0;
1515 tmp = entry->offset;
1516 }
1517
1518 if (entry->bytes < *bytes + align_off) {
1519 if (entry->bytes > *max_extent_size)
1520 *max_extent_size = entry->bytes;
1521 continue;
1522 }
1523
1524 if (entry->bitmap) {
1525 u64 size = *bytes;
1526
1527 ret = search_bitmap(ctl, entry, &tmp, &size);
1528 if (!ret) {
1529 *offset = tmp;
1530 *bytes = size;
1531 return entry;
1532 } else if (size > *max_extent_size) {
1533 *max_extent_size = size;
1534 }
1535 continue;
1536 }
1537
1538 *offset = tmp;
1539 *bytes = entry->bytes - align_off;
1540 return entry;
1541 }
1542 out:
1543 return NULL;
1544 }
1545
1546 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1547 struct btrfs_free_space *info, u64 offset)
1548 {
1549 info->offset = offset_to_bitmap(ctl, offset);
1550 info->bytes = 0;
1551 INIT_LIST_HEAD(&info->list);
1552 link_free_space(ctl, info);
1553 ctl->total_bitmaps++;
1554
1555 ctl->op->recalc_thresholds(ctl);
1556 }
1557
1558 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1559 struct btrfs_free_space *bitmap_info)
1560 {
1561 unlink_free_space(ctl, bitmap_info);
1562 kfree(bitmap_info->bitmap);
1563 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1564 ctl->total_bitmaps--;
1565 ctl->op->recalc_thresholds(ctl);
1566 }
1567
1568 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1569 struct btrfs_free_space *bitmap_info,
1570 u64 *offset, u64 *bytes)
1571 {
1572 u64 end;
1573 u64 search_start, search_bytes;
1574 int ret;
1575
1576 again:
1577 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1578
1579 /*
1580 * We need to search for bits in this bitmap. We could only cover some
1581 * of the extent in this bitmap thanks to how we add space, so we need
1582 * to search for as much as it as we can and clear that amount, and then
1583 * go searching for the next bit.
1584 */
1585 search_start = *offset;
1586 search_bytes = ctl->unit;
1587 search_bytes = min(search_bytes, end - search_start + 1);
1588 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1589 if (ret < 0 || search_start != *offset)
1590 return -EINVAL;
1591
1592 /* We may have found more bits than what we need */
1593 search_bytes = min(search_bytes, *bytes);
1594
1595 /* Cannot clear past the end of the bitmap */
1596 search_bytes = min(search_bytes, end - search_start + 1);
1597
1598 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1599 *offset += search_bytes;
1600 *bytes -= search_bytes;
1601
1602 if (*bytes) {
1603 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1604 if (!bitmap_info->bytes)
1605 free_bitmap(ctl, bitmap_info);
1606
1607 /*
1608 * no entry after this bitmap, but we still have bytes to
1609 * remove, so something has gone wrong.
1610 */
1611 if (!next)
1612 return -EINVAL;
1613
1614 bitmap_info = rb_entry(next, struct btrfs_free_space,
1615 offset_index);
1616
1617 /*
1618 * if the next entry isn't a bitmap we need to return to let the
1619 * extent stuff do its work.
1620 */
1621 if (!bitmap_info->bitmap)
1622 return -EAGAIN;
1623
1624 /*
1625 * Ok the next item is a bitmap, but it may not actually hold
1626 * the information for the rest of this free space stuff, so
1627 * look for it, and if we don't find it return so we can try
1628 * everything over again.
1629 */
1630 search_start = *offset;
1631 search_bytes = ctl->unit;
1632 ret = search_bitmap(ctl, bitmap_info, &search_start,
1633 &search_bytes);
1634 if (ret < 0 || search_start != *offset)
1635 return -EAGAIN;
1636
1637 goto again;
1638 } else if (!bitmap_info->bytes)
1639 free_bitmap(ctl, bitmap_info);
1640
1641 return 0;
1642 }
1643
1644 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1645 struct btrfs_free_space *info, u64 offset,
1646 u64 bytes)
1647 {
1648 u64 bytes_to_set = 0;
1649 u64 end;
1650
1651 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1652
1653 bytes_to_set = min(end - offset, bytes);
1654
1655 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1656
1657 return bytes_to_set;
1658
1659 }
1660
1661 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1662 struct btrfs_free_space *info)
1663 {
1664 struct btrfs_block_group_cache *block_group = ctl->private;
1665
1666 /*
1667 * If we are below the extents threshold then we can add this as an
1668 * extent, and don't have to deal with the bitmap
1669 */
1670 if (ctl->free_extents < ctl->extents_thresh) {
1671 /*
1672 * If this block group has some small extents we don't want to
1673 * use up all of our free slots in the cache with them, we want
1674 * to reserve them to larger extents, however if we have plent
1675 * of cache left then go ahead an dadd them, no sense in adding
1676 * the overhead of a bitmap if we don't have to.
1677 */
1678 if (info->bytes <= block_group->sectorsize * 4) {
1679 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1680 return false;
1681 } else {
1682 return false;
1683 }
1684 }
1685
1686 /*
1687 * The original block groups from mkfs can be really small, like 8
1688 * megabytes, so don't bother with a bitmap for those entries. However
1689 * some block groups can be smaller than what a bitmap would cover but
1690 * are still large enough that they could overflow the 32k memory limit,
1691 * so allow those block groups to still be allowed to have a bitmap
1692 * entry.
1693 */
1694 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
1695 return false;
1696
1697 return true;
1698 }
1699
1700 static struct btrfs_free_space_op free_space_op = {
1701 .recalc_thresholds = recalculate_thresholds,
1702 .use_bitmap = use_bitmap,
1703 };
1704
1705 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1706 struct btrfs_free_space *info)
1707 {
1708 struct btrfs_free_space *bitmap_info;
1709 struct btrfs_block_group_cache *block_group = NULL;
1710 int added = 0;
1711 u64 bytes, offset, bytes_added;
1712 int ret;
1713
1714 bytes = info->bytes;
1715 offset = info->offset;
1716
1717 if (!ctl->op->use_bitmap(ctl, info))
1718 return 0;
1719
1720 if (ctl->op == &free_space_op)
1721 block_group = ctl->private;
1722 again:
1723 /*
1724 * Since we link bitmaps right into the cluster we need to see if we
1725 * have a cluster here, and if so and it has our bitmap we need to add
1726 * the free space to that bitmap.
1727 */
1728 if (block_group && !list_empty(&block_group->cluster_list)) {
1729 struct btrfs_free_cluster *cluster;
1730 struct rb_node *node;
1731 struct btrfs_free_space *entry;
1732
1733 cluster = list_entry(block_group->cluster_list.next,
1734 struct btrfs_free_cluster,
1735 block_group_list);
1736 spin_lock(&cluster->lock);
1737 node = rb_first(&cluster->root);
1738 if (!node) {
1739 spin_unlock(&cluster->lock);
1740 goto no_cluster_bitmap;
1741 }
1742
1743 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1744 if (!entry->bitmap) {
1745 spin_unlock(&cluster->lock);
1746 goto no_cluster_bitmap;
1747 }
1748
1749 if (entry->offset == offset_to_bitmap(ctl, offset)) {
1750 bytes_added = add_bytes_to_bitmap(ctl, entry,
1751 offset, bytes);
1752 bytes -= bytes_added;
1753 offset += bytes_added;
1754 }
1755 spin_unlock(&cluster->lock);
1756 if (!bytes) {
1757 ret = 1;
1758 goto out;
1759 }
1760 }
1761
1762 no_cluster_bitmap:
1763 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1764 1, 0);
1765 if (!bitmap_info) {
1766 ASSERT(added == 0);
1767 goto new_bitmap;
1768 }
1769
1770 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
1771 bytes -= bytes_added;
1772 offset += bytes_added;
1773 added = 0;
1774
1775 if (!bytes) {
1776 ret = 1;
1777 goto out;
1778 } else
1779 goto again;
1780
1781 new_bitmap:
1782 if (info && info->bitmap) {
1783 add_new_bitmap(ctl, info, offset);
1784 added = 1;
1785 info = NULL;
1786 goto again;
1787 } else {
1788 spin_unlock(&ctl->tree_lock);
1789
1790 /* no pre-allocated info, allocate a new one */
1791 if (!info) {
1792 info = kmem_cache_zalloc(btrfs_free_space_cachep,
1793 GFP_NOFS);
1794 if (!info) {
1795 spin_lock(&ctl->tree_lock);
1796 ret = -ENOMEM;
1797 goto out;
1798 }
1799 }
1800
1801 /* allocate the bitmap */
1802 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1803 spin_lock(&ctl->tree_lock);
1804 if (!info->bitmap) {
1805 ret = -ENOMEM;
1806 goto out;
1807 }
1808 goto again;
1809 }
1810
1811 out:
1812 if (info) {
1813 if (info->bitmap)
1814 kfree(info->bitmap);
1815 kmem_cache_free(btrfs_free_space_cachep, info);
1816 }
1817
1818 return ret;
1819 }
1820
1821 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1822 struct btrfs_free_space *info, bool update_stat)
1823 {
1824 struct btrfs_free_space *left_info;
1825 struct btrfs_free_space *right_info;
1826 bool merged = false;
1827 u64 offset = info->offset;
1828 u64 bytes = info->bytes;
1829
1830 /*
1831 * first we want to see if there is free space adjacent to the range we
1832 * are adding, if there is remove that struct and add a new one to
1833 * cover the entire range
1834 */
1835 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1836 if (right_info && rb_prev(&right_info->offset_index))
1837 left_info = rb_entry(rb_prev(&right_info->offset_index),
1838 struct btrfs_free_space, offset_index);
1839 else
1840 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
1841
1842 if (right_info && !right_info->bitmap) {
1843 if (update_stat)
1844 unlink_free_space(ctl, right_info);
1845 else
1846 __unlink_free_space(ctl, right_info);
1847 info->bytes += right_info->bytes;
1848 kmem_cache_free(btrfs_free_space_cachep, right_info);
1849 merged = true;
1850 }
1851
1852 if (left_info && !left_info->bitmap &&
1853 left_info->offset + left_info->bytes == offset) {
1854 if (update_stat)
1855 unlink_free_space(ctl, left_info);
1856 else
1857 __unlink_free_space(ctl, left_info);
1858 info->offset = left_info->offset;
1859 info->bytes += left_info->bytes;
1860 kmem_cache_free(btrfs_free_space_cachep, left_info);
1861 merged = true;
1862 }
1863
1864 return merged;
1865 }
1866
1867 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
1868 u64 offset, u64 bytes)
1869 {
1870 struct btrfs_free_space *info;
1871 int ret = 0;
1872
1873 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1874 if (!info)
1875 return -ENOMEM;
1876
1877 info->offset = offset;
1878 info->bytes = bytes;
1879
1880 spin_lock(&ctl->tree_lock);
1881
1882 if (try_merge_free_space(ctl, info, true))
1883 goto link;
1884
1885 /*
1886 * There was no extent directly to the left or right of this new
1887 * extent then we know we're going to have to allocate a new extent, so
1888 * before we do that see if we need to drop this into a bitmap
1889 */
1890 ret = insert_into_bitmap(ctl, info);
1891 if (ret < 0) {
1892 goto out;
1893 } else if (ret) {
1894 ret = 0;
1895 goto out;
1896 }
1897 link:
1898 ret = link_free_space(ctl, info);
1899 if (ret)
1900 kmem_cache_free(btrfs_free_space_cachep, info);
1901 out:
1902 spin_unlock(&ctl->tree_lock);
1903
1904 if (ret) {
1905 printk(KERN_CRIT "BTRFS: unable to add free space :%d\n", ret);
1906 ASSERT(ret != -EEXIST);
1907 }
1908
1909 return ret;
1910 }
1911
1912 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1913 u64 offset, u64 bytes)
1914 {
1915 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1916 struct btrfs_free_space *info;
1917 int ret;
1918 bool re_search = false;
1919
1920 spin_lock(&ctl->tree_lock);
1921
1922 again:
1923 ret = 0;
1924 if (!bytes)
1925 goto out_lock;
1926
1927 info = tree_search_offset(ctl, offset, 0, 0);
1928 if (!info) {
1929 /*
1930 * oops didn't find an extent that matched the space we wanted
1931 * to remove, look for a bitmap instead
1932 */
1933 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1934 1, 0);
1935 if (!info) {
1936 /*
1937 * If we found a partial bit of our free space in a
1938 * bitmap but then couldn't find the other part this may
1939 * be a problem, so WARN about it.
1940 */
1941 WARN_ON(re_search);
1942 goto out_lock;
1943 }
1944 }
1945
1946 re_search = false;
1947 if (!info->bitmap) {
1948 unlink_free_space(ctl, info);
1949 if (offset == info->offset) {
1950 u64 to_free = min(bytes, info->bytes);
1951
1952 info->bytes -= to_free;
1953 info->offset += to_free;
1954 if (info->bytes) {
1955 ret = link_free_space(ctl, info);
1956 WARN_ON(ret);
1957 } else {
1958 kmem_cache_free(btrfs_free_space_cachep, info);
1959 }
1960
1961 offset += to_free;
1962 bytes -= to_free;
1963 goto again;
1964 } else {
1965 u64 old_end = info->bytes + info->offset;
1966
1967 info->bytes = offset - info->offset;
1968 ret = link_free_space(ctl, info);
1969 WARN_ON(ret);
1970 if (ret)
1971 goto out_lock;
1972
1973 /* Not enough bytes in this entry to satisfy us */
1974 if (old_end < offset + bytes) {
1975 bytes -= old_end - offset;
1976 offset = old_end;
1977 goto again;
1978 } else if (old_end == offset + bytes) {
1979 /* all done */
1980 goto out_lock;
1981 }
1982 spin_unlock(&ctl->tree_lock);
1983
1984 ret = btrfs_add_free_space(block_group, offset + bytes,
1985 old_end - (offset + bytes));
1986 WARN_ON(ret);
1987 goto out;
1988 }
1989 }
1990
1991 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
1992 if (ret == -EAGAIN) {
1993 re_search = true;
1994 goto again;
1995 }
1996 out_lock:
1997 spin_unlock(&ctl->tree_lock);
1998 out:
1999 return ret;
2000 }
2001
2002 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
2003 u64 bytes)
2004 {
2005 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2006 struct btrfs_free_space *info;
2007 struct rb_node *n;
2008 int count = 0;
2009
2010 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2011 info = rb_entry(n, struct btrfs_free_space, offset_index);
2012 if (info->bytes >= bytes && !block_group->ro)
2013 count++;
2014 btrfs_crit(block_group->fs_info,
2015 "entry offset %llu, bytes %llu, bitmap %s",
2016 info->offset, info->bytes,
2017 (info->bitmap) ? "yes" : "no");
2018 }
2019 btrfs_info(block_group->fs_info, "block group has cluster?: %s",
2020 list_empty(&block_group->cluster_list) ? "no" : "yes");
2021 btrfs_info(block_group->fs_info,
2022 "%d blocks of free space at or bigger than bytes is", count);
2023 }
2024
2025 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2026 {
2027 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2028
2029 spin_lock_init(&ctl->tree_lock);
2030 ctl->unit = block_group->sectorsize;
2031 ctl->start = block_group->key.objectid;
2032 ctl->private = block_group;
2033 ctl->op = &free_space_op;
2034
2035 /*
2036 * we only want to have 32k of ram per block group for keeping
2037 * track of free space, and if we pass 1/2 of that we want to
2038 * start converting things over to using bitmaps
2039 */
2040 ctl->extents_thresh = ((1024 * 32) / 2) /
2041 sizeof(struct btrfs_free_space);
2042 }
2043
2044 /*
2045 * for a given cluster, put all of its extents back into the free
2046 * space cache. If the block group passed doesn't match the block group
2047 * pointed to by the cluster, someone else raced in and freed the
2048 * cluster already. In that case, we just return without changing anything
2049 */
2050 static int
2051 __btrfs_return_cluster_to_free_space(
2052 struct btrfs_block_group_cache *block_group,
2053 struct btrfs_free_cluster *cluster)
2054 {
2055 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2056 struct btrfs_free_space *entry;
2057 struct rb_node *node;
2058
2059 spin_lock(&cluster->lock);
2060 if (cluster->block_group != block_group)
2061 goto out;
2062
2063 cluster->block_group = NULL;
2064 cluster->window_start = 0;
2065 list_del_init(&cluster->block_group_list);
2066
2067 node = rb_first(&cluster->root);
2068 while (node) {
2069 bool bitmap;
2070
2071 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2072 node = rb_next(&entry->offset_index);
2073 rb_erase(&entry->offset_index, &cluster->root);
2074
2075 bitmap = (entry->bitmap != NULL);
2076 if (!bitmap)
2077 try_merge_free_space(ctl, entry, false);
2078 tree_insert_offset(&ctl->free_space_offset,
2079 entry->offset, &entry->offset_index, bitmap);
2080 }
2081 cluster->root = RB_ROOT;
2082
2083 out:
2084 spin_unlock(&cluster->lock);
2085 btrfs_put_block_group(block_group);
2086 return 0;
2087 }
2088
2089 static void __btrfs_remove_free_space_cache_locked(
2090 struct btrfs_free_space_ctl *ctl)
2091 {
2092 struct btrfs_free_space *info;
2093 struct rb_node *node;
2094
2095 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2096 info = rb_entry(node, struct btrfs_free_space, offset_index);
2097 if (!info->bitmap) {
2098 unlink_free_space(ctl, info);
2099 kmem_cache_free(btrfs_free_space_cachep, info);
2100 } else {
2101 free_bitmap(ctl, info);
2102 }
2103 if (need_resched()) {
2104 spin_unlock(&ctl->tree_lock);
2105 cond_resched();
2106 spin_lock(&ctl->tree_lock);
2107 }
2108 }
2109 }
2110
2111 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2112 {
2113 spin_lock(&ctl->tree_lock);
2114 __btrfs_remove_free_space_cache_locked(ctl);
2115 spin_unlock(&ctl->tree_lock);
2116 }
2117
2118 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2119 {
2120 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2121 struct btrfs_free_cluster *cluster;
2122 struct list_head *head;
2123
2124 spin_lock(&ctl->tree_lock);
2125 while ((head = block_group->cluster_list.next) !=
2126 &block_group->cluster_list) {
2127 cluster = list_entry(head, struct btrfs_free_cluster,
2128 block_group_list);
2129
2130 WARN_ON(cluster->block_group != block_group);
2131 __btrfs_return_cluster_to_free_space(block_group, cluster);
2132 if (need_resched()) {
2133 spin_unlock(&ctl->tree_lock);
2134 cond_resched();
2135 spin_lock(&ctl->tree_lock);
2136 }
2137 }
2138 __btrfs_remove_free_space_cache_locked(ctl);
2139 spin_unlock(&ctl->tree_lock);
2140
2141 }
2142
2143 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2144 u64 offset, u64 bytes, u64 empty_size,
2145 u64 *max_extent_size)
2146 {
2147 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2148 struct btrfs_free_space *entry = NULL;
2149 u64 bytes_search = bytes + empty_size;
2150 u64 ret = 0;
2151 u64 align_gap = 0;
2152 u64 align_gap_len = 0;
2153
2154 spin_lock(&ctl->tree_lock);
2155 entry = find_free_space(ctl, &offset, &bytes_search,
2156 block_group->full_stripe_len, max_extent_size);
2157 if (!entry)
2158 goto out;
2159
2160 ret = offset;
2161 if (entry->bitmap) {
2162 bitmap_clear_bits(ctl, entry, offset, bytes);
2163 if (!entry->bytes)
2164 free_bitmap(ctl, entry);
2165 } else {
2166 unlink_free_space(ctl, entry);
2167 align_gap_len = offset - entry->offset;
2168 align_gap = entry->offset;
2169
2170 entry->offset = offset + bytes;
2171 WARN_ON(entry->bytes < bytes + align_gap_len);
2172
2173 entry->bytes -= bytes + align_gap_len;
2174 if (!entry->bytes)
2175 kmem_cache_free(btrfs_free_space_cachep, entry);
2176 else
2177 link_free_space(ctl, entry);
2178 }
2179 out:
2180 spin_unlock(&ctl->tree_lock);
2181
2182 if (align_gap_len)
2183 __btrfs_add_free_space(ctl, align_gap, align_gap_len);
2184 return ret;
2185 }
2186
2187 /*
2188 * given a cluster, put all of its extents back into the free space
2189 * cache. If a block group is passed, this function will only free
2190 * a cluster that belongs to the passed block group.
2191 *
2192 * Otherwise, it'll get a reference on the block group pointed to by the
2193 * cluster and remove the cluster from it.
2194 */
2195 int btrfs_return_cluster_to_free_space(
2196 struct btrfs_block_group_cache *block_group,
2197 struct btrfs_free_cluster *cluster)
2198 {
2199 struct btrfs_free_space_ctl *ctl;
2200 int ret;
2201
2202 /* first, get a safe pointer to the block group */
2203 spin_lock(&cluster->lock);
2204 if (!block_group) {
2205 block_group = cluster->block_group;
2206 if (!block_group) {
2207 spin_unlock(&cluster->lock);
2208 return 0;
2209 }
2210 } else if (cluster->block_group != block_group) {
2211 /* someone else has already freed it don't redo their work */
2212 spin_unlock(&cluster->lock);
2213 return 0;
2214 }
2215 atomic_inc(&block_group->count);
2216 spin_unlock(&cluster->lock);
2217
2218 ctl = block_group->free_space_ctl;
2219
2220 /* now return any extents the cluster had on it */
2221 spin_lock(&ctl->tree_lock);
2222 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2223 spin_unlock(&ctl->tree_lock);
2224
2225 /* finally drop our ref */
2226 btrfs_put_block_group(block_group);
2227 return ret;
2228 }
2229
2230 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2231 struct btrfs_free_cluster *cluster,
2232 struct btrfs_free_space *entry,
2233 u64 bytes, u64 min_start,
2234 u64 *max_extent_size)
2235 {
2236 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2237 int err;
2238 u64 search_start = cluster->window_start;
2239 u64 search_bytes = bytes;
2240 u64 ret = 0;
2241
2242 search_start = min_start;
2243 search_bytes = bytes;
2244
2245 err = search_bitmap(ctl, entry, &search_start, &search_bytes);
2246 if (err) {
2247 if (search_bytes > *max_extent_size)
2248 *max_extent_size = search_bytes;
2249 return 0;
2250 }
2251
2252 ret = search_start;
2253 __bitmap_clear_bits(ctl, entry, ret, bytes);
2254
2255 return ret;
2256 }
2257
2258 /*
2259 * given a cluster, try to allocate 'bytes' from it, returns 0
2260 * if it couldn't find anything suitably large, or a logical disk offset
2261 * if things worked out
2262 */
2263 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2264 struct btrfs_free_cluster *cluster, u64 bytes,
2265 u64 min_start, u64 *max_extent_size)
2266 {
2267 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2268 struct btrfs_free_space *entry = NULL;
2269 struct rb_node *node;
2270 u64 ret = 0;
2271
2272 spin_lock(&cluster->lock);
2273 if (bytes > cluster->max_size)
2274 goto out;
2275
2276 if (cluster->block_group != block_group)
2277 goto out;
2278
2279 node = rb_first(&cluster->root);
2280 if (!node)
2281 goto out;
2282
2283 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2284 while (1) {
2285 if (entry->bytes < bytes && entry->bytes > *max_extent_size)
2286 *max_extent_size = entry->bytes;
2287
2288 if (entry->bytes < bytes ||
2289 (!entry->bitmap && entry->offset < min_start)) {
2290 node = rb_next(&entry->offset_index);
2291 if (!node)
2292 break;
2293 entry = rb_entry(node, struct btrfs_free_space,
2294 offset_index);
2295 continue;
2296 }
2297
2298 if (entry->bitmap) {
2299 ret = btrfs_alloc_from_bitmap(block_group,
2300 cluster, entry, bytes,
2301 cluster->window_start,
2302 max_extent_size);
2303 if (ret == 0) {
2304 node = rb_next(&entry->offset_index);
2305 if (!node)
2306 break;
2307 entry = rb_entry(node, struct btrfs_free_space,
2308 offset_index);
2309 continue;
2310 }
2311 cluster->window_start += bytes;
2312 } else {
2313 ret = entry->offset;
2314
2315 entry->offset += bytes;
2316 entry->bytes -= bytes;
2317 }
2318
2319 if (entry->bytes == 0)
2320 rb_erase(&entry->offset_index, &cluster->root);
2321 break;
2322 }
2323 out:
2324 spin_unlock(&cluster->lock);
2325
2326 if (!ret)
2327 return 0;
2328
2329 spin_lock(&ctl->tree_lock);
2330
2331 ctl->free_space -= bytes;
2332 if (entry->bytes == 0) {
2333 ctl->free_extents--;
2334 if (entry->bitmap) {
2335 kfree(entry->bitmap);
2336 ctl->total_bitmaps--;
2337 ctl->op->recalc_thresholds(ctl);
2338 }
2339 kmem_cache_free(btrfs_free_space_cachep, entry);
2340 }
2341
2342 spin_unlock(&ctl->tree_lock);
2343
2344 return ret;
2345 }
2346
2347 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2348 struct btrfs_free_space *entry,
2349 struct btrfs_free_cluster *cluster,
2350 u64 offset, u64 bytes,
2351 u64 cont1_bytes, u64 min_bytes)
2352 {
2353 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2354 unsigned long next_zero;
2355 unsigned long i;
2356 unsigned long want_bits;
2357 unsigned long min_bits;
2358 unsigned long found_bits;
2359 unsigned long start = 0;
2360 unsigned long total_found = 0;
2361 int ret;
2362
2363 i = offset_to_bit(entry->offset, ctl->unit,
2364 max_t(u64, offset, entry->offset));
2365 want_bits = bytes_to_bits(bytes, ctl->unit);
2366 min_bits = bytes_to_bits(min_bytes, ctl->unit);
2367
2368 again:
2369 found_bits = 0;
2370 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2371 next_zero = find_next_zero_bit(entry->bitmap,
2372 BITS_PER_BITMAP, i);
2373 if (next_zero - i >= min_bits) {
2374 found_bits = next_zero - i;
2375 break;
2376 }
2377 i = next_zero;
2378 }
2379
2380 if (!found_bits)
2381 return -ENOSPC;
2382
2383 if (!total_found) {
2384 start = i;
2385 cluster->max_size = 0;
2386 }
2387
2388 total_found += found_bits;
2389
2390 if (cluster->max_size < found_bits * ctl->unit)
2391 cluster->max_size = found_bits * ctl->unit;
2392
2393 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2394 i = next_zero + 1;
2395 goto again;
2396 }
2397
2398 cluster->window_start = start * ctl->unit + entry->offset;
2399 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2400 ret = tree_insert_offset(&cluster->root, entry->offset,
2401 &entry->offset_index, 1);
2402 ASSERT(!ret); /* -EEXIST; Logic error */
2403
2404 trace_btrfs_setup_cluster(block_group, cluster,
2405 total_found * ctl->unit, 1);
2406 return 0;
2407 }
2408
2409 /*
2410 * This searches the block group for just extents to fill the cluster with.
2411 * Try to find a cluster with at least bytes total bytes, at least one
2412 * extent of cont1_bytes, and other clusters of at least min_bytes.
2413 */
2414 static noinline int
2415 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2416 struct btrfs_free_cluster *cluster,
2417 struct list_head *bitmaps, u64 offset, u64 bytes,
2418 u64 cont1_bytes, u64 min_bytes)
2419 {
2420 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2421 struct btrfs_free_space *first = NULL;
2422 struct btrfs_free_space *entry = NULL;
2423 struct btrfs_free_space *last;
2424 struct rb_node *node;
2425 u64 window_free;
2426 u64 max_extent;
2427 u64 total_size = 0;
2428
2429 entry = tree_search_offset(ctl, offset, 0, 1);
2430 if (!entry)
2431 return -ENOSPC;
2432
2433 /*
2434 * We don't want bitmaps, so just move along until we find a normal
2435 * extent entry.
2436 */
2437 while (entry->bitmap || entry->bytes < min_bytes) {
2438 if (entry->bitmap && list_empty(&entry->list))
2439 list_add_tail(&entry->list, bitmaps);
2440 node = rb_next(&entry->offset_index);
2441 if (!node)
2442 return -ENOSPC;
2443 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2444 }
2445
2446 window_free = entry->bytes;
2447 max_extent = entry->bytes;
2448 first = entry;
2449 last = entry;
2450
2451 for (node = rb_next(&entry->offset_index); node;
2452 node = rb_next(&entry->offset_index)) {
2453 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2454
2455 if (entry->bitmap) {
2456 if (list_empty(&entry->list))
2457 list_add_tail(&entry->list, bitmaps);
2458 continue;
2459 }
2460
2461 if (entry->bytes < min_bytes)
2462 continue;
2463
2464 last = entry;
2465 window_free += entry->bytes;
2466 if (entry->bytes > max_extent)
2467 max_extent = entry->bytes;
2468 }
2469
2470 if (window_free < bytes || max_extent < cont1_bytes)
2471 return -ENOSPC;
2472
2473 cluster->window_start = first->offset;
2474
2475 node = &first->offset_index;
2476
2477 /*
2478 * now we've found our entries, pull them out of the free space
2479 * cache and put them into the cluster rbtree
2480 */
2481 do {
2482 int ret;
2483
2484 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2485 node = rb_next(&entry->offset_index);
2486 if (entry->bitmap || entry->bytes < min_bytes)
2487 continue;
2488
2489 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2490 ret = tree_insert_offset(&cluster->root, entry->offset,
2491 &entry->offset_index, 0);
2492 total_size += entry->bytes;
2493 ASSERT(!ret); /* -EEXIST; Logic error */
2494 } while (node && entry != last);
2495
2496 cluster->max_size = max_extent;
2497 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2498 return 0;
2499 }
2500
2501 /*
2502 * This specifically looks for bitmaps that may work in the cluster, we assume
2503 * that we have already failed to find extents that will work.
2504 */
2505 static noinline int
2506 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2507 struct btrfs_free_cluster *cluster,
2508 struct list_head *bitmaps, u64 offset, u64 bytes,
2509 u64 cont1_bytes, u64 min_bytes)
2510 {
2511 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2512 struct btrfs_free_space *entry;
2513 int ret = -ENOSPC;
2514 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2515
2516 if (ctl->total_bitmaps == 0)
2517 return -ENOSPC;
2518
2519 /*
2520 * The bitmap that covers offset won't be in the list unless offset
2521 * is just its start offset.
2522 */
2523 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2524 if (entry->offset != bitmap_offset) {
2525 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2526 if (entry && list_empty(&entry->list))
2527 list_add(&entry->list, bitmaps);
2528 }
2529
2530 list_for_each_entry(entry, bitmaps, list) {
2531 if (entry->bytes < bytes)
2532 continue;
2533 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2534 bytes, cont1_bytes, min_bytes);
2535 if (!ret)
2536 return 0;
2537 }
2538
2539 /*
2540 * The bitmaps list has all the bitmaps that record free space
2541 * starting after offset, so no more search is required.
2542 */
2543 return -ENOSPC;
2544 }
2545
2546 /*
2547 * here we try to find a cluster of blocks in a block group. The goal
2548 * is to find at least bytes+empty_size.
2549 * We might not find them all in one contiguous area.
2550 *
2551 * returns zero and sets up cluster if things worked out, otherwise
2552 * it returns -enospc
2553 */
2554 int btrfs_find_space_cluster(struct btrfs_root *root,
2555 struct btrfs_block_group_cache *block_group,
2556 struct btrfs_free_cluster *cluster,
2557 u64 offset, u64 bytes, u64 empty_size)
2558 {
2559 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2560 struct btrfs_free_space *entry, *tmp;
2561 LIST_HEAD(bitmaps);
2562 u64 min_bytes;
2563 u64 cont1_bytes;
2564 int ret;
2565
2566 /*
2567 * Choose the minimum extent size we'll require for this
2568 * cluster. For SSD_SPREAD, don't allow any fragmentation.
2569 * For metadata, allow allocates with smaller extents. For
2570 * data, keep it dense.
2571 */
2572 if (btrfs_test_opt(root, SSD_SPREAD)) {
2573 cont1_bytes = min_bytes = bytes + empty_size;
2574 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2575 cont1_bytes = bytes;
2576 min_bytes = block_group->sectorsize;
2577 } else {
2578 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
2579 min_bytes = block_group->sectorsize;
2580 }
2581
2582 spin_lock(&ctl->tree_lock);
2583
2584 /*
2585 * If we know we don't have enough space to make a cluster don't even
2586 * bother doing all the work to try and find one.
2587 */
2588 if (ctl->free_space < bytes) {
2589 spin_unlock(&ctl->tree_lock);
2590 return -ENOSPC;
2591 }
2592
2593 spin_lock(&cluster->lock);
2594
2595 /* someone already found a cluster, hooray */
2596 if (cluster->block_group) {
2597 ret = 0;
2598 goto out;
2599 }
2600
2601 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
2602 min_bytes);
2603
2604 INIT_LIST_HEAD(&bitmaps);
2605 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2606 bytes + empty_size,
2607 cont1_bytes, min_bytes);
2608 if (ret)
2609 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2610 offset, bytes + empty_size,
2611 cont1_bytes, min_bytes);
2612
2613 /* Clear our temporary list */
2614 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
2615 list_del_init(&entry->list);
2616
2617 if (!ret) {
2618 atomic_inc(&block_group->count);
2619 list_add_tail(&cluster->block_group_list,
2620 &block_group->cluster_list);
2621 cluster->block_group = block_group;
2622 } else {
2623 trace_btrfs_failed_cluster_setup(block_group);
2624 }
2625 out:
2626 spin_unlock(&cluster->lock);
2627 spin_unlock(&ctl->tree_lock);
2628
2629 return ret;
2630 }
2631
2632 /*
2633 * simple code to zero out a cluster
2634 */
2635 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2636 {
2637 spin_lock_init(&cluster->lock);
2638 spin_lock_init(&cluster->refill_lock);
2639 cluster->root = RB_ROOT;
2640 cluster->max_size = 0;
2641 INIT_LIST_HEAD(&cluster->block_group_list);
2642 cluster->block_group = NULL;
2643 }
2644
2645 static int do_trimming(struct btrfs_block_group_cache *block_group,
2646 u64 *total_trimmed, u64 start, u64 bytes,
2647 u64 reserved_start, u64 reserved_bytes)
2648 {
2649 struct btrfs_space_info *space_info = block_group->space_info;
2650 struct btrfs_fs_info *fs_info = block_group->fs_info;
2651 int ret;
2652 int update = 0;
2653 u64 trimmed = 0;
2654
2655 spin_lock(&space_info->lock);
2656 spin_lock(&block_group->lock);
2657 if (!block_group->ro) {
2658 block_group->reserved += reserved_bytes;
2659 space_info->bytes_reserved += reserved_bytes;
2660 update = 1;
2661 }
2662 spin_unlock(&block_group->lock);
2663 spin_unlock(&space_info->lock);
2664
2665 ret = btrfs_error_discard_extent(fs_info->extent_root,
2666 start, bytes, &trimmed);
2667 if (!ret)
2668 *total_trimmed += trimmed;
2669
2670 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
2671
2672 if (update) {
2673 spin_lock(&space_info->lock);
2674 spin_lock(&block_group->lock);
2675 if (block_group->ro)
2676 space_info->bytes_readonly += reserved_bytes;
2677 block_group->reserved -= reserved_bytes;
2678 space_info->bytes_reserved -= reserved_bytes;
2679 spin_unlock(&space_info->lock);
2680 spin_unlock(&block_group->lock);
2681 }
2682
2683 return ret;
2684 }
2685
2686 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
2687 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2688 {
2689 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2690 struct btrfs_free_space *entry;
2691 struct rb_node *node;
2692 int ret = 0;
2693 u64 extent_start;
2694 u64 extent_bytes;
2695 u64 bytes;
2696
2697 while (start < end) {
2698 spin_lock(&ctl->tree_lock);
2699
2700 if (ctl->free_space < minlen) {
2701 spin_unlock(&ctl->tree_lock);
2702 break;
2703 }
2704
2705 entry = tree_search_offset(ctl, start, 0, 1);
2706 if (!entry) {
2707 spin_unlock(&ctl->tree_lock);
2708 break;
2709 }
2710
2711 /* skip bitmaps */
2712 while (entry->bitmap) {
2713 node = rb_next(&entry->offset_index);
2714 if (!node) {
2715 spin_unlock(&ctl->tree_lock);
2716 goto out;
2717 }
2718 entry = rb_entry(node, struct btrfs_free_space,
2719 offset_index);
2720 }
2721
2722 if (entry->offset >= end) {
2723 spin_unlock(&ctl->tree_lock);
2724 break;
2725 }
2726
2727 extent_start = entry->offset;
2728 extent_bytes = entry->bytes;
2729 start = max(start, extent_start);
2730 bytes = min(extent_start + extent_bytes, end) - start;
2731 if (bytes < minlen) {
2732 spin_unlock(&ctl->tree_lock);
2733 goto next;
2734 }
2735
2736 unlink_free_space(ctl, entry);
2737 kmem_cache_free(btrfs_free_space_cachep, entry);
2738
2739 spin_unlock(&ctl->tree_lock);
2740
2741 ret = do_trimming(block_group, total_trimmed, start, bytes,
2742 extent_start, extent_bytes);
2743 if (ret)
2744 break;
2745 next:
2746 start += bytes;
2747
2748 if (fatal_signal_pending(current)) {
2749 ret = -ERESTARTSYS;
2750 break;
2751 }
2752
2753 cond_resched();
2754 }
2755 out:
2756 return ret;
2757 }
2758
2759 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
2760 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2761 {
2762 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2763 struct btrfs_free_space *entry;
2764 int ret = 0;
2765 int ret2;
2766 u64 bytes;
2767 u64 offset = offset_to_bitmap(ctl, start);
2768
2769 while (offset < end) {
2770 bool next_bitmap = false;
2771
2772 spin_lock(&ctl->tree_lock);
2773
2774 if (ctl->free_space < minlen) {
2775 spin_unlock(&ctl->tree_lock);
2776 break;
2777 }
2778
2779 entry = tree_search_offset(ctl, offset, 1, 0);
2780 if (!entry) {
2781 spin_unlock(&ctl->tree_lock);
2782 next_bitmap = true;
2783 goto next;
2784 }
2785
2786 bytes = minlen;
2787 ret2 = search_bitmap(ctl, entry, &start, &bytes);
2788 if (ret2 || start >= end) {
2789 spin_unlock(&ctl->tree_lock);
2790 next_bitmap = true;
2791 goto next;
2792 }
2793
2794 bytes = min(bytes, end - start);
2795 if (bytes < minlen) {
2796 spin_unlock(&ctl->tree_lock);
2797 goto next;
2798 }
2799
2800 bitmap_clear_bits(ctl, entry, start, bytes);
2801 if (entry->bytes == 0)
2802 free_bitmap(ctl, entry);
2803
2804 spin_unlock(&ctl->tree_lock);
2805
2806 ret = do_trimming(block_group, total_trimmed, start, bytes,
2807 start, bytes);
2808 if (ret)
2809 break;
2810 next:
2811 if (next_bitmap) {
2812 offset += BITS_PER_BITMAP * ctl->unit;
2813 } else {
2814 start += bytes;
2815 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
2816 offset += BITS_PER_BITMAP * ctl->unit;
2817 }
2818
2819 if (fatal_signal_pending(current)) {
2820 ret = -ERESTARTSYS;
2821 break;
2822 }
2823
2824 cond_resched();
2825 }
2826
2827 return ret;
2828 }
2829
2830 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2831 u64 *trimmed, u64 start, u64 end, u64 minlen)
2832 {
2833 int ret;
2834
2835 *trimmed = 0;
2836
2837 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
2838 if (ret)
2839 return ret;
2840
2841 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
2842
2843 return ret;
2844 }
2845
2846 /*
2847 * Find the left-most item in the cache tree, and then return the
2848 * smallest inode number in the item.
2849 *
2850 * Note: the returned inode number may not be the smallest one in
2851 * the tree, if the left-most item is a bitmap.
2852 */
2853 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
2854 {
2855 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
2856 struct btrfs_free_space *entry = NULL;
2857 u64 ino = 0;
2858
2859 spin_lock(&ctl->tree_lock);
2860
2861 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
2862 goto out;
2863
2864 entry = rb_entry(rb_first(&ctl->free_space_offset),
2865 struct btrfs_free_space, offset_index);
2866
2867 if (!entry->bitmap) {
2868 ino = entry->offset;
2869
2870 unlink_free_space(ctl, entry);
2871 entry->offset++;
2872 entry->bytes--;
2873 if (!entry->bytes)
2874 kmem_cache_free(btrfs_free_space_cachep, entry);
2875 else
2876 link_free_space(ctl, entry);
2877 } else {
2878 u64 offset = 0;
2879 u64 count = 1;
2880 int ret;
2881
2882 ret = search_bitmap(ctl, entry, &offset, &count);
2883 /* Logic error; Should be empty if it can't find anything */
2884 ASSERT(!ret);
2885
2886 ino = offset;
2887 bitmap_clear_bits(ctl, entry, offset, 1);
2888 if (entry->bytes == 0)
2889 free_bitmap(ctl, entry);
2890 }
2891 out:
2892 spin_unlock(&ctl->tree_lock);
2893
2894 return ino;
2895 }
2896
2897 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
2898 struct btrfs_path *path)
2899 {
2900 struct inode *inode = NULL;
2901
2902 spin_lock(&root->cache_lock);
2903 if (root->cache_inode)
2904 inode = igrab(root->cache_inode);
2905 spin_unlock(&root->cache_lock);
2906 if (inode)
2907 return inode;
2908
2909 inode = __lookup_free_space_inode(root, path, 0);
2910 if (IS_ERR(inode))
2911 return inode;
2912
2913 spin_lock(&root->cache_lock);
2914 if (!btrfs_fs_closing(root->fs_info))
2915 root->cache_inode = igrab(inode);
2916 spin_unlock(&root->cache_lock);
2917
2918 return inode;
2919 }
2920
2921 int create_free_ino_inode(struct btrfs_root *root,
2922 struct btrfs_trans_handle *trans,
2923 struct btrfs_path *path)
2924 {
2925 return __create_free_space_inode(root, trans, path,
2926 BTRFS_FREE_INO_OBJECTID, 0);
2927 }
2928
2929 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2930 {
2931 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2932 struct btrfs_path *path;
2933 struct inode *inode;
2934 int ret = 0;
2935 u64 root_gen = btrfs_root_generation(&root->root_item);
2936
2937 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2938 return 0;
2939
2940 /*
2941 * If we're unmounting then just return, since this does a search on the
2942 * normal root and not the commit root and we could deadlock.
2943 */
2944 if (btrfs_fs_closing(fs_info))
2945 return 0;
2946
2947 path = btrfs_alloc_path();
2948 if (!path)
2949 return 0;
2950
2951 inode = lookup_free_ino_inode(root, path);
2952 if (IS_ERR(inode))
2953 goto out;
2954
2955 if (root_gen != BTRFS_I(inode)->generation)
2956 goto out_put;
2957
2958 ret = __load_free_space_cache(root, inode, ctl, path, 0);
2959
2960 if (ret < 0)
2961 btrfs_err(fs_info,
2962 "failed to load free ino cache for root %llu",
2963 root->root_key.objectid);
2964 out_put:
2965 iput(inode);
2966 out:
2967 btrfs_free_path(path);
2968 return ret;
2969 }
2970
2971 int btrfs_write_out_ino_cache(struct btrfs_root *root,
2972 struct btrfs_trans_handle *trans,
2973 struct btrfs_path *path,
2974 struct inode *inode)
2975 {
2976 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2977 int ret;
2978
2979 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2980 return 0;
2981
2982 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
2983 if (ret) {
2984 btrfs_delalloc_release_metadata(inode, inode->i_size);
2985 #ifdef DEBUG
2986 btrfs_err(root->fs_info,
2987 "failed to write free ino cache for root %llu",
2988 root->root_key.objectid);
2989 #endif
2990 }
2991
2992 return ret;
2993 }
2994
2995 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
2996 /*
2997 * Use this if you need to make a bitmap or extent entry specifically, it
2998 * doesn't do any of the merging that add_free_space does, this acts a lot like
2999 * how the free space cache loading stuff works, so you can get really weird
3000 * configurations.
3001 */
3002 int test_add_free_space_entry(struct btrfs_block_group_cache *cache,
3003 u64 offset, u64 bytes, bool bitmap)
3004 {
3005 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3006 struct btrfs_free_space *info = NULL, *bitmap_info;
3007 void *map = NULL;
3008 u64 bytes_added;
3009 int ret;
3010
3011 again:
3012 if (!info) {
3013 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
3014 if (!info)
3015 return -ENOMEM;
3016 }
3017
3018 if (!bitmap) {
3019 spin_lock(&ctl->tree_lock);
3020 info->offset = offset;
3021 info->bytes = bytes;
3022 ret = link_free_space(ctl, info);
3023 spin_unlock(&ctl->tree_lock);
3024 if (ret)
3025 kmem_cache_free(btrfs_free_space_cachep, info);
3026 return ret;
3027 }
3028
3029 if (!map) {
3030 map = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
3031 if (!map) {
3032 kmem_cache_free(btrfs_free_space_cachep, info);
3033 return -ENOMEM;
3034 }
3035 }
3036
3037 spin_lock(&ctl->tree_lock);
3038 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3039 1, 0);
3040 if (!bitmap_info) {
3041 info->bitmap = map;
3042 map = NULL;
3043 add_new_bitmap(ctl, info, offset);
3044 bitmap_info = info;
3045 }
3046
3047 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
3048 bytes -= bytes_added;
3049 offset += bytes_added;
3050 spin_unlock(&ctl->tree_lock);
3051
3052 if (bytes)
3053 goto again;
3054
3055 if (map)
3056 kfree(map);
3057 return 0;
3058 }
3059
3060 /*
3061 * Checks to see if the given range is in the free space cache. This is really
3062 * just used to check the absence of space, so if there is free space in the
3063 * range at all we will return 1.
3064 */
3065 int test_check_exists(struct btrfs_block_group_cache *cache,
3066 u64 offset, u64 bytes)
3067 {
3068 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3069 struct btrfs_free_space *info;
3070 int ret = 0;
3071
3072 spin_lock(&ctl->tree_lock);
3073 info = tree_search_offset(ctl, offset, 0, 0);
3074 if (!info) {
3075 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3076 1, 0);
3077 if (!info)
3078 goto out;
3079 }
3080
3081 have_info:
3082 if (info->bitmap) {
3083 u64 bit_off, bit_bytes;
3084 struct rb_node *n;
3085 struct btrfs_free_space *tmp;
3086
3087 bit_off = offset;
3088 bit_bytes = ctl->unit;
3089 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes);
3090 if (!ret) {
3091 if (bit_off == offset) {
3092 ret = 1;
3093 goto out;
3094 } else if (bit_off > offset &&
3095 offset + bytes > bit_off) {
3096 ret = 1;
3097 goto out;
3098 }
3099 }
3100
3101 n = rb_prev(&info->offset_index);
3102 while (n) {
3103 tmp = rb_entry(n, struct btrfs_free_space,
3104 offset_index);
3105 if (tmp->offset + tmp->bytes < offset)
3106 break;
3107 if (offset + bytes < tmp->offset) {
3108 n = rb_prev(&info->offset_index);
3109 continue;
3110 }
3111 info = tmp;
3112 goto have_info;
3113 }
3114
3115 n = rb_next(&info->offset_index);
3116 while (n) {
3117 tmp = rb_entry(n, struct btrfs_free_space,
3118 offset_index);
3119 if (offset + bytes < tmp->offset)
3120 break;
3121 if (tmp->offset + tmp->bytes < offset) {
3122 n = rb_next(&info->offset_index);
3123 continue;
3124 }
3125 info = tmp;
3126 goto have_info;
3127 }
3128
3129 goto out;
3130 }
3131
3132 if (info->offset == offset) {
3133 ret = 1;
3134 goto out;
3135 }
3136
3137 if (offset > info->offset && offset < info->offset + info->bytes)
3138 ret = 1;
3139 out:
3140 spin_unlock(&ctl->tree_lock);
3141 return ret;
3142 }
3143 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */
Linux with added FPC-III support