?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'linux-next' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad...
[zen-stable.git] / fs / btrfs / free-space-cache.c
blobec23d43d0c357870a14bf6e79a0ce5baeb93a43d
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
37 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
38 struct btrfs_path *path,
39 u64 offset)
40 {
41 struct btrfs_key key;
42 struct btrfs_key location;
43 struct btrfs_disk_key disk_key;
44 struct btrfs_free_space_header *header;
45 struct extent_buffer *leaf;
46 struct inode *inode = NULL;
47 int ret;
48
49 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
50 key.offset = offset;
51 key.type = 0;
52
53 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
54 if (ret < 0)
55 return ERR_PTR(ret);
56 if (ret > 0) {
57 btrfs_release_path(path);
58 return ERR_PTR(-ENOENT);
59 }
60
61 leaf = path->nodes[0];
62 header = btrfs_item_ptr(leaf, path->slots[0],
63 struct btrfs_free_space_header);
64 btrfs_free_space_key(leaf, header, &disk_key);
65 btrfs_disk_key_to_cpu(&location, &disk_key);
66 btrfs_release_path(path);
67
68 inode = btrfs_iget(root->fs_info->sb, &location, root, NULL);
69 if (!inode)
70 return ERR_PTR(-ENOENT);
71 if (IS_ERR(inode))
72 return inode;
73 if (is_bad_inode(inode)) {
74 iput(inode);
75 return ERR_PTR(-ENOENT);
76 }
77
78 inode->i_mapping->flags &= ~__GFP_FS;
79
80 return inode;
81 }
82
83 struct inode *lookup_free_space_inode(struct btrfs_root *root,
84 struct btrfs_block_group_cache
85 *block_group, struct btrfs_path *path)
86 {
87 struct inode *inode = NULL;
88 u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
89
90 spin_lock(&block_group->lock);
91 if (block_group->inode)
92 inode = igrab(block_group->inode);
93 spin_unlock(&block_group->lock);
94 if (inode)
95 return inode;
96
97 inode = __lookup_free_space_inode(root, path,
98 block_group->key.objectid);
99 if (IS_ERR(inode))
100 return inode;
101
102 spin_lock(&block_group->lock);
103 if (!((BTRFS_I(inode)->flags & flags) == flags)) {
104 printk(KERN_INFO "Old style space inode found, converting.\n");
105 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
106 BTRFS_INODE_NODATACOW;
107 block_group->disk_cache_state = BTRFS_DC_CLEAR;
108 }
109
110 if (!block_group->iref) {
111 block_group->inode = igrab(inode);
112 block_group->iref = 1;
113 }
114 spin_unlock(&block_group->lock);
115
116 return inode;
117 }
118
119 int __create_free_space_inode(struct btrfs_root *root,
120 struct btrfs_trans_handle *trans,
121 struct btrfs_path *path, u64 ino, u64 offset)
122 {
123 struct btrfs_key key;
124 struct btrfs_disk_key disk_key;
125 struct btrfs_free_space_header *header;
126 struct btrfs_inode_item *inode_item;
127 struct extent_buffer *leaf;
128 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
129 int ret;
130
131 ret = btrfs_insert_empty_inode(trans, root, path, ino);
132 if (ret)
133 return ret;
134
135 /* We inline crc's for the free disk space cache */
136 if (ino != BTRFS_FREE_INO_OBJECTID)
137 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
138
139 leaf = path->nodes[0];
140 inode_item = btrfs_item_ptr(leaf, path->slots[0],
141 struct btrfs_inode_item);
142 btrfs_item_key(leaf, &disk_key, path->slots[0]);
143 memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
144 sizeof(*inode_item));
145 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
146 btrfs_set_inode_size(leaf, inode_item, 0);
147 btrfs_set_inode_nbytes(leaf, inode_item, 0);
148 btrfs_set_inode_uid(leaf, inode_item, 0);
149 btrfs_set_inode_gid(leaf, inode_item, 0);
150 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
151 btrfs_set_inode_flags(leaf, inode_item, flags);
152 btrfs_set_inode_nlink(leaf, inode_item, 1);
153 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
154 btrfs_set_inode_block_group(leaf, inode_item, offset);
155 btrfs_mark_buffer_dirty(leaf);
156 btrfs_release_path(path);
157
158 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
159 key.offset = offset;
160 key.type = 0;
161
162 ret = btrfs_insert_empty_item(trans, root, path, &key,
163 sizeof(struct btrfs_free_space_header));
164 if (ret < 0) {
165 btrfs_release_path(path);
166 return ret;
167 }
168 leaf = path->nodes[0];
169 header = btrfs_item_ptr(leaf, path->slots[0],
170 struct btrfs_free_space_header);
171 memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
172 btrfs_set_free_space_key(leaf, header, &disk_key);
173 btrfs_mark_buffer_dirty(leaf);
174 btrfs_release_path(path);
175
176 return 0;
177 }
178
179 int create_free_space_inode(struct btrfs_root *root,
180 struct btrfs_trans_handle *trans,
181 struct btrfs_block_group_cache *block_group,
182 struct btrfs_path *path)
183 {
184 int ret;
185 u64 ino;
186
187 ret = btrfs_find_free_objectid(root, &ino);
188 if (ret < 0)
189 return ret;
190
191 return __create_free_space_inode(root, trans, path, ino,
192 block_group->key.objectid);
193 }
194
195 int btrfs_truncate_free_space_cache(struct btrfs_root *root,
196 struct btrfs_trans_handle *trans,
197 struct btrfs_path *path,
198 struct inode *inode)
199 {
200 struct btrfs_block_rsv *rsv;
201 u64 needed_bytes;
202 loff_t oldsize;
203 int ret = 0;
204
205 rsv = trans->block_rsv;
206 trans->block_rsv = &root->fs_info->global_block_rsv;
207
208 /* 1 for slack space, 1 for updating the inode */
209 needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
210 btrfs_calc_trans_metadata_size(root, 1);
211
212 spin_lock(&trans->block_rsv->lock);
213 if (trans->block_rsv->reserved < needed_bytes) {
214 spin_unlock(&trans->block_rsv->lock);
215 trans->block_rsv = rsv;
216 return -ENOSPC;
217 }
218 spin_unlock(&trans->block_rsv->lock);
219
220 oldsize = i_size_read(inode);
221 btrfs_i_size_write(inode, 0);
222 truncate_pagecache(inode, oldsize, 0);
223
224 /*
225 * We don't need an orphan item because truncating the free space cache
226 * will never be split across transactions.
227 */
228 ret = btrfs_truncate_inode_items(trans, root, inode,
229 0, BTRFS_EXTENT_DATA_KEY);
230
231 if (ret) {
232 trans->block_rsv = rsv;
233 WARN_ON(1);
234 return ret;
235 }
236
237 ret = btrfs_update_inode(trans, root, inode);
238 trans->block_rsv = rsv;
239
240 return ret;
241 }
242
243 static int readahead_cache(struct inode *inode)
244 {
245 struct file_ra_state *ra;
246 unsigned long last_index;
247
248 ra = kzalloc(sizeof(*ra), GFP_NOFS);
249 if (!ra)
250 return -ENOMEM;
251
252 file_ra_state_init(ra, inode->i_mapping);
253 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
254
255 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
256
257 kfree(ra);
258
259 return 0;
260 }
261
262 struct io_ctl {
263 void *cur, *orig;
264 struct page *page;
265 struct page **pages;
266 struct btrfs_root *root;
267 unsigned long size;
268 int index;
269 int num_pages;
270 unsigned check_crcs:1;
271 };
272
273 static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode,
274 struct btrfs_root *root)
275 {
276 memset(io_ctl, 0, sizeof(struct io_ctl));
277 io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
278 PAGE_CACHE_SHIFT;
279 io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages,
280 GFP_NOFS);
281 if (!io_ctl->pages)
282 return -ENOMEM;
283 io_ctl->root = root;
284 if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
285 io_ctl->check_crcs = 1;
286 return 0;
287 }
288
289 static void io_ctl_free(struct io_ctl *io_ctl)
290 {
291 kfree(io_ctl->pages);
292 }
293
294 static void io_ctl_unmap_page(struct io_ctl *io_ctl)
295 {
296 if (io_ctl->cur) {
297 kunmap(io_ctl->page);
298 io_ctl->cur = NULL;
299 io_ctl->orig = NULL;
300 }
301 }
302
303 static void io_ctl_map_page(struct io_ctl *io_ctl, int clear)
304 {
305 WARN_ON(io_ctl->cur);
306 BUG_ON(io_ctl->index >= io_ctl->num_pages);
307 io_ctl->page = io_ctl->pages[io_ctl->index++];
308 io_ctl->cur = kmap(io_ctl->page);
309 io_ctl->orig = io_ctl->cur;
310 io_ctl->size = PAGE_CACHE_SIZE;
311 if (clear)
312 memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
313 }
314
315 static void io_ctl_drop_pages(struct io_ctl *io_ctl)
316 {
317 int i;
318
319 io_ctl_unmap_page(io_ctl);
320
321 for (i = 0; i < io_ctl->num_pages; i++) {
322 ClearPageChecked(io_ctl->pages[i]);
323 unlock_page(io_ctl->pages[i]);
324 page_cache_release(io_ctl->pages[i]);
325 }
326 }
327
328 static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode,
329 int uptodate)
330 {
331 struct page *page;
332 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
333 int i;
334
335 for (i = 0; i < io_ctl->num_pages; i++) {
336 page = find_or_create_page(inode->i_mapping, i, mask);
337 if (!page) {
338 io_ctl_drop_pages(io_ctl);
339 return -ENOMEM;
340 }
341 io_ctl->pages[i] = page;
342 if (uptodate && !PageUptodate(page)) {
343 btrfs_readpage(NULL, page);
344 lock_page(page);
345 if (!PageUptodate(page)) {
346 printk(KERN_ERR "btrfs: error reading free "
347 "space cache\n");
348 io_ctl_drop_pages(io_ctl);
349 return -EIO;
350 }
351 }
352 }
353
354 for (i = 0; i < io_ctl->num_pages; i++) {
355 clear_page_dirty_for_io(io_ctl->pages[i]);
356 set_page_extent_mapped(io_ctl->pages[i]);
357 }
358
359 return 0;
360 }
361
362 static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation)
363 {
364 u64 *val;
365
366 io_ctl_map_page(io_ctl, 1);
367
368 /*
369 * Skip the csum areas. If we don't check crcs then we just have a
370 * 64bit chunk at the front of the first page.
371 */
372 if (io_ctl->check_crcs) {
373 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
374 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
375 } else {
376 io_ctl->cur += sizeof(u64);
377 io_ctl->size -= sizeof(u64) * 2;
378 }
379
380 val = io_ctl->cur;
381 *val = cpu_to_le64(generation);
382 io_ctl->cur += sizeof(u64);
383 }
384
385 static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation)
386 {
387 u64 *gen;
388
389 /*
390 * Skip the crc area. If we don't check crcs then we just have a 64bit
391 * chunk at the front of the first page.
392 */
393 if (io_ctl->check_crcs) {
394 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
395 io_ctl->size -= sizeof(u64) +
396 (sizeof(u32) * io_ctl->num_pages);
397 } else {
398 io_ctl->cur += sizeof(u64);
399 io_ctl->size -= sizeof(u64) * 2;
400 }
401
402 gen = io_ctl->cur;
403 if (le64_to_cpu(*gen) != generation) {
404 printk_ratelimited(KERN_ERR "btrfs: space cache generation "
405 "(%Lu) does not match inode (%Lu)\n", *gen,
406 generation);
407 io_ctl_unmap_page(io_ctl);
408 return -EIO;
409 }
410 io_ctl->cur += sizeof(u64);
411 return 0;
412 }
413
414 static void io_ctl_set_crc(struct io_ctl *io_ctl, int index)
415 {
416 u32 *tmp;
417 u32 crc = ~(u32)0;
418 unsigned offset = 0;
419
420 if (!io_ctl->check_crcs) {
421 io_ctl_unmap_page(io_ctl);
422 return;
423 }
424
425 if (index == 0)
426 offset = sizeof(u32) * io_ctl->num_pages;;
427
428 crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc,
429 PAGE_CACHE_SIZE - offset);
430 btrfs_csum_final(crc, (char *)&crc);
431 io_ctl_unmap_page(io_ctl);
432 tmp = kmap(io_ctl->pages[0]);
433 tmp += index;
434 *tmp = crc;
435 kunmap(io_ctl->pages[0]);
436 }
437
438 static int io_ctl_check_crc(struct io_ctl *io_ctl, int index)
439 {
440 u32 *tmp, val;
441 u32 crc = ~(u32)0;
442 unsigned offset = 0;
443
444 if (!io_ctl->check_crcs) {
445 io_ctl_map_page(io_ctl, 0);
446 return 0;
447 }
448
449 if (index == 0)
450 offset = sizeof(u32) * io_ctl->num_pages;
451
452 tmp = kmap(io_ctl->pages[0]);
453 tmp += index;
454 val = *tmp;
455 kunmap(io_ctl->pages[0]);
456
457 io_ctl_map_page(io_ctl, 0);
458 crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc,
459 PAGE_CACHE_SIZE - offset);
460 btrfs_csum_final(crc, (char *)&crc);
461 if (val != crc) {
462 printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free "
463 "space cache\n");
464 io_ctl_unmap_page(io_ctl);
465 return -EIO;
466 }
467
468 return 0;
469 }
470
471 static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes,
472 void *bitmap)
473 {
474 struct btrfs_free_space_entry *entry;
475
476 if (!io_ctl->cur)
477 return -ENOSPC;
478
479 entry = io_ctl->cur;
480 entry->offset = cpu_to_le64(offset);
481 entry->bytes = cpu_to_le64(bytes);
482 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
483 BTRFS_FREE_SPACE_EXTENT;
484 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
485 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
486
487 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
488 return 0;
489
490 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
491
492 /* No more pages to map */
493 if (io_ctl->index >= io_ctl->num_pages)
494 return 0;
495
496 /* map the next page */
497 io_ctl_map_page(io_ctl, 1);
498 return 0;
499 }
500
501 static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap)
502 {
503 if (!io_ctl->cur)
504 return -ENOSPC;
505
506 /*
507 * If we aren't at the start of the current page, unmap this one and
508 * map the next one if there is any left.
509 */
510 if (io_ctl->cur != io_ctl->orig) {
511 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
512 if (io_ctl->index >= io_ctl->num_pages)
513 return -ENOSPC;
514 io_ctl_map_page(io_ctl, 0);
515 }
516
517 memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
518 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
519 if (io_ctl->index < io_ctl->num_pages)
520 io_ctl_map_page(io_ctl, 0);
521 return 0;
522 }
523
524 static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl)
525 {
526 /*
527 * If we're not on the boundary we know we've modified the page and we
528 * need to crc the page.
529 */
530 if (io_ctl->cur != io_ctl->orig)
531 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
532 else
533 io_ctl_unmap_page(io_ctl);
534
535 while (io_ctl->index < io_ctl->num_pages) {
536 io_ctl_map_page(io_ctl, 1);
537 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
538 }
539 }
540
541 static int io_ctl_read_entry(struct io_ctl *io_ctl,
542 struct btrfs_free_space *entry, u8 *type)
543 {
544 struct btrfs_free_space_entry *e;
545 int ret;
546
547 if (!io_ctl->cur) {
548 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
549 if (ret)
550 return ret;
551 }
552
553 e = io_ctl->cur;
554 entry->offset = le64_to_cpu(e->offset);
555 entry->bytes = le64_to_cpu(e->bytes);
556 *type = e->type;
557 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
558 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
559
560 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
561 return 0;
562
563 io_ctl_unmap_page(io_ctl);
564
565 return 0;
566 }
567
568 static int io_ctl_read_bitmap(struct io_ctl *io_ctl,
569 struct btrfs_free_space *entry)
570 {
571 int ret;
572
573 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
574 if (ret)
575 return ret;
576
577 memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
578 io_ctl_unmap_page(io_ctl);
579
580 return 0;
581 }
582
583 int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
584 struct btrfs_free_space_ctl *ctl,
585 struct btrfs_path *path, u64 offset)
586 {
587 struct btrfs_free_space_header *header;
588 struct extent_buffer *leaf;
589 struct io_ctl io_ctl;
590 struct btrfs_key key;
591 struct btrfs_free_space *e, *n;
592 struct list_head bitmaps;
593 u64 num_entries;
594 u64 num_bitmaps;
595 u64 generation;
596 u8 type;
597 int ret = 0;
598
599 INIT_LIST_HEAD(&bitmaps);
600
601 /* Nothing in the space cache, goodbye */
602 if (!i_size_read(inode))
603 return 0;
604
605 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
606 key.offset = offset;
607 key.type = 0;
608
609 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
610 if (ret < 0)
611 return 0;
612 else if (ret > 0) {
613 btrfs_release_path(path);
614 return 0;
615 }
616
617 ret = -1;
618
619 leaf = path->nodes[0];
620 header = btrfs_item_ptr(leaf, path->slots[0],
621 struct btrfs_free_space_header);
622 num_entries = btrfs_free_space_entries(leaf, header);
623 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
624 generation = btrfs_free_space_generation(leaf, header);
625 btrfs_release_path(path);
626
627 if (BTRFS_I(inode)->generation != generation) {
628 printk(KERN_ERR "btrfs: free space inode generation (%llu) did"
629 " not match free space cache generation (%llu)\n",
630 (unsigned long long)BTRFS_I(inode)->generation,
631 (unsigned long long)generation);
632 return 0;
633 }
634
635 if (!num_entries)
636 return 0;
637
638 io_ctl_init(&io_ctl, inode, root);
639 ret = readahead_cache(inode);
640 if (ret)
641 goto out;
642
643 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
644 if (ret)
645 goto out;
646
647 ret = io_ctl_check_crc(&io_ctl, 0);
648 if (ret)
649 goto free_cache;
650
651 ret = io_ctl_check_generation(&io_ctl, generation);
652 if (ret)
653 goto free_cache;
654
655 while (num_entries) {
656 e = kmem_cache_zalloc(btrfs_free_space_cachep,
657 GFP_NOFS);
658 if (!e)
659 goto free_cache;
660
661 ret = io_ctl_read_entry(&io_ctl, e, &type);
662 if (ret) {
663 kmem_cache_free(btrfs_free_space_cachep, e);
664 goto free_cache;
665 }
666
667 if (!e->bytes) {
668 kmem_cache_free(btrfs_free_space_cachep, e);
669 goto free_cache;
670 }
671
672 if (type == BTRFS_FREE_SPACE_EXTENT) {
673 spin_lock(&ctl->tree_lock);
674 ret = link_free_space(ctl, e);
675 spin_unlock(&ctl->tree_lock);
676 if (ret) {
677 printk(KERN_ERR "Duplicate entries in "
678 "free space cache, dumping\n");
679 kmem_cache_free(btrfs_free_space_cachep, e);
680 goto free_cache;
681 }
682 } else {
683 BUG_ON(!num_bitmaps);
684 num_bitmaps--;
685 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
686 if (!e->bitmap) {
687 kmem_cache_free(
688 btrfs_free_space_cachep, e);
689 goto free_cache;
690 }
691 spin_lock(&ctl->tree_lock);
692 ret = link_free_space(ctl, e);
693 ctl->total_bitmaps++;
694 ctl->op->recalc_thresholds(ctl);
695 spin_unlock(&ctl->tree_lock);
696 if (ret) {
697 printk(KERN_ERR "Duplicate entries in "
698 "free space cache, dumping\n");
699 kmem_cache_free(btrfs_free_space_cachep, e);
700 goto free_cache;
701 }
702 list_add_tail(&e->list, &bitmaps);
703 }
704
705 num_entries--;
706 }
707
708 io_ctl_unmap_page(&io_ctl);
709
710 /*
711 * We add the bitmaps at the end of the entries in order that
712 * the bitmap entries are added to the cache.
713 */
714 list_for_each_entry_safe(e, n, &bitmaps, list) {
715 list_del_init(&e->list);
716 ret = io_ctl_read_bitmap(&io_ctl, e);
717 if (ret)
718 goto free_cache;
719 }
720
721 io_ctl_drop_pages(&io_ctl);
722 ret = 1;
723 out:
724 io_ctl_free(&io_ctl);
725 return ret;
726 free_cache:
727 io_ctl_drop_pages(&io_ctl);
728 __btrfs_remove_free_space_cache(ctl);
729 goto out;
730 }
731
732 int load_free_space_cache(struct btrfs_fs_info *fs_info,
733 struct btrfs_block_group_cache *block_group)
734 {
735 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
736 struct btrfs_root *root = fs_info->tree_root;
737 struct inode *inode;
738 struct btrfs_path *path;
739 int ret = 0;
740 bool matched;
741 u64 used = btrfs_block_group_used(&block_group->item);
742
743 /*
744 * If we're unmounting then just return, since this does a search on the
745 * normal root and not the commit root and we could deadlock.
746 */
747 if (btrfs_fs_closing(fs_info))
748 return 0;
749
750 /*
751 * If this block group has been marked to be cleared for one reason or
752 * another then we can't trust the on disk cache, so just return.
753 */
754 spin_lock(&block_group->lock);
755 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
756 spin_unlock(&block_group->lock);
757 return 0;
758 }
759 spin_unlock(&block_group->lock);
760
761 path = btrfs_alloc_path();
762 if (!path)
763 return 0;
764
765 inode = lookup_free_space_inode(root, block_group, path);
766 if (IS_ERR(inode)) {
767 btrfs_free_path(path);
768 return 0;
769 }
770
771 /* We may have converted the inode and made the cache invalid. */
772 spin_lock(&block_group->lock);
773 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
774 spin_unlock(&block_group->lock);
775 goto out;
776 }
777 spin_unlock(&block_group->lock);
778
779 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
780 path, block_group->key.objectid);
781 btrfs_free_path(path);
782 if (ret <= 0)
783 goto out;
784
785 spin_lock(&ctl->tree_lock);
786 matched = (ctl->free_space == (block_group->key.offset - used -
787 block_group->bytes_super));
788 spin_unlock(&ctl->tree_lock);
789
790 if (!matched) {
791 __btrfs_remove_free_space_cache(ctl);
792 printk(KERN_ERR "block group %llu has an wrong amount of free "
793 "space\n", block_group->key.objectid);
794 ret = -1;
795 }
796 out:
797 if (ret < 0) {
798 /* This cache is bogus, make sure it gets cleared */
799 spin_lock(&block_group->lock);
800 block_group->disk_cache_state = BTRFS_DC_CLEAR;
801 spin_unlock(&block_group->lock);
802 ret = 0;
803
804 printk(KERN_ERR "btrfs: failed to load free space cache "
805 "for block group %llu\n", block_group->key.objectid);
806 }
807
808 iput(inode);
809 return ret;
810 }
811
812 /**
813 * __btrfs_write_out_cache - write out cached info to an inode
814 * @root - the root the inode belongs to
815 * @ctl - the free space cache we are going to write out
816 * @block_group - the block_group for this cache if it belongs to a block_group
817 * @trans - the trans handle
818 * @path - the path to use
819 * @offset - the offset for the key we'll insert
820 *
821 * This function writes out a free space cache struct to disk for quick recovery
822 * on mount. This will return 0 if it was successfull in writing the cache out,
823 * and -1 if it was not.
824 */
825 int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
826 struct btrfs_free_space_ctl *ctl,
827 struct btrfs_block_group_cache *block_group,
828 struct btrfs_trans_handle *trans,
829 struct btrfs_path *path, u64 offset)
830 {
831 struct btrfs_free_space_header *header;
832 struct extent_buffer *leaf;
833 struct rb_node *node;
834 struct list_head *pos, *n;
835 struct extent_state *cached_state = NULL;
836 struct btrfs_free_cluster *cluster = NULL;
837 struct extent_io_tree *unpin = NULL;
838 struct io_ctl io_ctl;
839 struct list_head bitmap_list;
840 struct btrfs_key key;
841 u64 start, end, len;
842 int entries = 0;
843 int bitmaps = 0;
844 int ret;
845 int err = -1;
846
847 INIT_LIST_HEAD(&bitmap_list);
848
849 if (!i_size_read(inode))
850 return -1;
851
852 io_ctl_init(&io_ctl, inode, root);
853
854 /* Get the cluster for this block_group if it exists */
855 if (block_group && !list_empty(&block_group->cluster_list))
856 cluster = list_entry(block_group->cluster_list.next,
857 struct btrfs_free_cluster,
858 block_group_list);
859
860 /*
861 * We shouldn't have switched the pinned extents yet so this is the
862 * right one
863 */
864 unpin = root->fs_info->pinned_extents;
865
866 /* Lock all pages first so we can lock the extent safely. */
867 io_ctl_prepare_pages(&io_ctl, inode, 0);
868
869 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
870 0, &cached_state, GFP_NOFS);
871
872 /*
873 * When searching for pinned extents, we need to start at our start
874 * offset.
875 */
876 if (block_group)
877 start = block_group->key.objectid;
878
879 node = rb_first(&ctl->free_space_offset);
880 if (!node && cluster) {
881 node = rb_first(&cluster->root);
882 cluster = NULL;
883 }
884
885 /* Make sure we can fit our crcs into the first page */
886 if (io_ctl.check_crcs &&
887 (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) {
888 WARN_ON(1);
889 goto out_nospc;
890 }
891
892 io_ctl_set_generation(&io_ctl, trans->transid);
893
894 /* Write out the extent entries */
895 while (node) {
896 struct btrfs_free_space *e;
897
898 e = rb_entry(node, struct btrfs_free_space, offset_index);
899 entries++;
900
901 ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes,
902 e->bitmap);
903 if (ret)
904 goto out_nospc;
905
906 if (e->bitmap) {
907 list_add_tail(&e->list, &bitmap_list);
908 bitmaps++;
909 }
910 node = rb_next(node);
911 if (!node && cluster) {
912 node = rb_first(&cluster->root);
913 cluster = NULL;
914 }
915 }
916
917 /*
918 * We want to add any pinned extents to our free space cache
919 * so we don't leak the space
920 */
921 while (block_group && (start < block_group->key.objectid +
922 block_group->key.offset)) {
923 ret = find_first_extent_bit(unpin, start, &start, &end,
924 EXTENT_DIRTY);
925 if (ret) {
926 ret = 0;
927 break;
928 }
929
930 /* This pinned extent is out of our range */
931 if (start >= block_group->key.objectid +
932 block_group->key.offset)
933 break;
934
935 len = block_group->key.objectid +
936 block_group->key.offset - start;
937 len = min(len, end + 1 - start);
938
939 entries++;
940 ret = io_ctl_add_entry(&io_ctl, start, len, NULL);
941 if (ret)
942 goto out_nospc;
943
944 start = end + 1;
945 }
946
947 /* Write out the bitmaps */
948 list_for_each_safe(pos, n, &bitmap_list) {
949 struct btrfs_free_space *entry =
950 list_entry(pos, struct btrfs_free_space, list);
951
952 ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap);
953 if (ret)
954 goto out_nospc;
955 list_del_init(&entry->list);
956 }
957
958 /* Zero out the rest of the pages just to make sure */
959 io_ctl_zero_remaining_pages(&io_ctl);
960
961 ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
962 0, i_size_read(inode), &cached_state);
963 io_ctl_drop_pages(&io_ctl);
964 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
965 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
966
967 if (ret)
968 goto out;
969
970
971 ret = filemap_write_and_wait(inode->i_mapping);
972 if (ret)
973 goto out;
974
975 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
976 key.offset = offset;
977 key.type = 0;
978
979 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
980 if (ret < 0) {
981 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
982 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
983 GFP_NOFS);
984 goto out;
985 }
986 leaf = path->nodes[0];
987 if (ret > 0) {
988 struct btrfs_key found_key;
989 BUG_ON(!path->slots[0]);
990 path->slots[0]--;
991 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
992 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
993 found_key.offset != offset) {
994 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
995 inode->i_size - 1,
996 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
997 NULL, GFP_NOFS);
998 btrfs_release_path(path);
999 goto out;
1000 }
1001 }
1002
1003 BTRFS_I(inode)->generation = trans->transid;
1004 header = btrfs_item_ptr(leaf, path->slots[0],
1005 struct btrfs_free_space_header);
1006 btrfs_set_free_space_entries(leaf, header, entries);
1007 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1008 btrfs_set_free_space_generation(leaf, header, trans->transid);
1009 btrfs_mark_buffer_dirty(leaf);
1010 btrfs_release_path(path);
1011
1012 err = 0;
1013 out:
1014 io_ctl_free(&io_ctl);
1015 if (err) {
1016 invalidate_inode_pages2(inode->i_mapping);
1017 BTRFS_I(inode)->generation = 0;
1018 }
1019 btrfs_update_inode(trans, root, inode);
1020 return err;
1021
1022 out_nospc:
1023 list_for_each_safe(pos, n, &bitmap_list) {
1024 struct btrfs_free_space *entry =
1025 list_entry(pos, struct btrfs_free_space, list);
1026 list_del_init(&entry->list);
1027 }
1028 io_ctl_drop_pages(&io_ctl);
1029 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1030 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1031 goto out;
1032 }
1033
1034 int btrfs_write_out_cache(struct btrfs_root *root,
1035 struct btrfs_trans_handle *trans,
1036 struct btrfs_block_group_cache *block_group,
1037 struct btrfs_path *path)
1038 {
1039 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1040 struct inode *inode;
1041 int ret = 0;
1042
1043 root = root->fs_info->tree_root;
1044
1045 spin_lock(&block_group->lock);
1046 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1047 spin_unlock(&block_group->lock);
1048 return 0;
1049 }
1050 spin_unlock(&block_group->lock);
1051
1052 inode = lookup_free_space_inode(root, block_group, path);
1053 if (IS_ERR(inode))
1054 return 0;
1055
1056 ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
1057 path, block_group->key.objectid);
1058 if (ret) {
1059 spin_lock(&block_group->lock);
1060 block_group->disk_cache_state = BTRFS_DC_ERROR;
1061 spin_unlock(&block_group->lock);
1062 ret = 0;
1063 #ifdef DEBUG
1064 printk(KERN_ERR "btrfs: failed to write free space cace "
1065 "for block group %llu\n", block_group->key.objectid);
1066 #endif
1067 }
1068
1069 iput(inode);
1070 return ret;
1071 }
1072
1073 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1074 u64 offset)
1075 {
1076 BUG_ON(offset < bitmap_start);
1077 offset -= bitmap_start;
1078 return (unsigned long)(div_u64(offset, unit));
1079 }
1080
1081 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1082 {
1083 return (unsigned long)(div_u64(bytes, unit));
1084 }
1085
1086 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1087 u64 offset)
1088 {
1089 u64 bitmap_start;
1090 u64 bytes_per_bitmap;
1091
1092 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1093 bitmap_start = offset - ctl->start;
1094 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1095 bitmap_start *= bytes_per_bitmap;
1096 bitmap_start += ctl->start;
1097
1098 return bitmap_start;
1099 }
1100
1101 static int tree_insert_offset(struct rb_root *root, u64 offset,
1102 struct rb_node *node, int bitmap)
1103 {
1104 struct rb_node **p = &root->rb_node;
1105 struct rb_node *parent = NULL;
1106 struct btrfs_free_space *info;
1107
1108 while (*p) {
1109 parent = *p;
1110 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1111
1112 if (offset < info->offset) {
1113 p = &(*p)->rb_left;
1114 } else if (offset > info->offset) {
1115 p = &(*p)->rb_right;
1116 } else {
1117 /*
1118 * we could have a bitmap entry and an extent entry
1119 * share the same offset. If this is the case, we want
1120 * the extent entry to always be found first if we do a
1121 * linear search through the tree, since we want to have
1122 * the quickest allocation time, and allocating from an
1123 * extent is faster than allocating from a bitmap. So
1124 * if we're inserting a bitmap and we find an entry at
1125 * this offset, we want to go right, or after this entry
1126 * logically. If we are inserting an extent and we've
1127 * found a bitmap, we want to go left, or before
1128 * logically.
1129 */
1130 if (bitmap) {
1131 if (info->bitmap) {
1132 WARN_ON_ONCE(1);
1133 return -EEXIST;
1134 }
1135 p = &(*p)->rb_right;
1136 } else {
1137 if (!info->bitmap) {
1138 WARN_ON_ONCE(1);
1139 return -EEXIST;
1140 }
1141 p = &(*p)->rb_left;
1142 }
1143 }
1144 }
1145
1146 rb_link_node(node, parent, p);
1147 rb_insert_color(node, root);
1148
1149 return 0;
1150 }
1151
1152 /*
1153 * searches the tree for the given offset.
1154 *
1155 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1156 * want a section that has at least bytes size and comes at or after the given
1157 * offset.
1158 */
1159 static struct btrfs_free_space *
1160 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1161 u64 offset, int bitmap_only, int fuzzy)
1162 {
1163 struct rb_node *n = ctl->free_space_offset.rb_node;
1164 struct btrfs_free_space *entry, *prev = NULL;
1165
1166 /* find entry that is closest to the 'offset' */
1167 while (1) {
1168 if (!n) {
1169 entry = NULL;
1170 break;
1171 }
1172
1173 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1174 prev = entry;
1175
1176 if (offset < entry->offset)
1177 n = n->rb_left;
1178 else if (offset > entry->offset)
1179 n = n->rb_right;
1180 else
1181 break;
1182 }
1183
1184 if (bitmap_only) {
1185 if (!entry)
1186 return NULL;
1187 if (entry->bitmap)
1188 return entry;
1189
1190 /*
1191 * bitmap entry and extent entry may share same offset,
1192 * in that case, bitmap entry comes after extent entry.
1193 */
1194 n = rb_next(n);
1195 if (!n)
1196 return NULL;
1197 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1198 if (entry->offset != offset)
1199 return NULL;
1200
1201 WARN_ON(!entry->bitmap);
1202 return entry;
1203 } else if (entry) {
1204 if (entry->bitmap) {
1205 /*
1206 * if previous extent entry covers the offset,
1207 * we should return it instead of the bitmap entry
1208 */
1209 n = &entry->offset_index;
1210 while (1) {
1211 n = rb_prev(n);
1212 if (!n)
1213 break;
1214 prev = rb_entry(n, struct btrfs_free_space,
1215 offset_index);
1216 if (!prev->bitmap) {
1217 if (prev->offset + prev->bytes > offset)
1218 entry = prev;
1219 break;
1220 }
1221 }
1222 }
1223 return entry;
1224 }
1225
1226 if (!prev)
1227 return NULL;
1228
1229 /* find last entry before the 'offset' */
1230 entry = prev;
1231 if (entry->offset > offset) {
1232 n = rb_prev(&entry->offset_index);
1233 if (n) {
1234 entry = rb_entry(n, struct btrfs_free_space,
1235 offset_index);
1236 BUG_ON(entry->offset > offset);
1237 } else {
1238 if (fuzzy)
1239 return entry;
1240 else
1241 return NULL;
1242 }
1243 }
1244
1245 if (entry->bitmap) {
1246 n = &entry->offset_index;
1247 while (1) {
1248 n = rb_prev(n);
1249 if (!n)
1250 break;
1251 prev = rb_entry(n, struct btrfs_free_space,
1252 offset_index);
1253 if (!prev->bitmap) {
1254 if (prev->offset + prev->bytes > offset)
1255 return prev;
1256 break;
1257 }
1258 }
1259 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1260 return entry;
1261 } else if (entry->offset + entry->bytes > offset)
1262 return entry;
1263
1264 if (!fuzzy)
1265 return NULL;
1266
1267 while (1) {
1268 if (entry->bitmap) {
1269 if (entry->offset + BITS_PER_BITMAP *
1270 ctl->unit > offset)
1271 break;
1272 } else {
1273 if (entry->offset + entry->bytes > offset)
1274 break;
1275 }
1276
1277 n = rb_next(&entry->offset_index);
1278 if (!n)
1279 return NULL;
1280 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1281 }
1282 return entry;
1283 }
1284
1285 static inline void
1286 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1287 struct btrfs_free_space *info)
1288 {
1289 rb_erase(&info->offset_index, &ctl->free_space_offset);
1290 ctl->free_extents--;
1291 }
1292
1293 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1294 struct btrfs_free_space *info)
1295 {
1296 __unlink_free_space(ctl, info);
1297 ctl->free_space -= info->bytes;
1298 }
1299
1300 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1301 struct btrfs_free_space *info)
1302 {
1303 int ret = 0;
1304
1305 BUG_ON(!info->bitmap && !info->bytes);
1306 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1307 &info->offset_index, (info->bitmap != NULL));
1308 if (ret)
1309 return ret;
1310
1311 ctl->free_space += info->bytes;
1312 ctl->free_extents++;
1313 return ret;
1314 }
1315
1316 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1317 {
1318 struct btrfs_block_group_cache *block_group = ctl->private;
1319 u64 max_bytes;
1320 u64 bitmap_bytes;
1321 u64 extent_bytes;
1322 u64 size = block_group->key.offset;
1323 u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize;
1324 int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1325
1326 BUG_ON(ctl->total_bitmaps > max_bitmaps);
1327
1328 /*
1329 * The goal is to keep the total amount of memory used per 1gb of space
1330 * at or below 32k, so we need to adjust how much memory we allow to be
1331 * used by extent based free space tracking
1332 */
1333 if (size < 1024 * 1024 * 1024)
1334 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1335 else
1336 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1337 div64_u64(size, 1024 * 1024 * 1024);
1338
1339 /*
1340 * we want to account for 1 more bitmap than what we have so we can make
1341 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1342 * we add more bitmaps.
1343 */
1344 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1345
1346 if (bitmap_bytes >= max_bytes) {
1347 ctl->extents_thresh = 0;
1348 return;
1349 }
1350
1351 /*
1352 * we want the extent entry threshold to always be at most 1/2 the maxw
1353 * bytes we can have, or whatever is less than that.
1354 */
1355 extent_bytes = max_bytes - bitmap_bytes;
1356 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1357
1358 ctl->extents_thresh =
1359 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1360 }
1361
1362 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1363 struct btrfs_free_space *info,
1364 u64 offset, u64 bytes)
1365 {
1366 unsigned long start, count;
1367
1368 start = offset_to_bit(info->offset, ctl->unit, offset);
1369 count = bytes_to_bits(bytes, ctl->unit);
1370 BUG_ON(start + count > BITS_PER_BITMAP);
1371
1372 bitmap_clear(info->bitmap, start, count);
1373
1374 info->bytes -= bytes;
1375 }
1376
1377 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1378 struct btrfs_free_space *info, u64 offset,
1379 u64 bytes)
1380 {
1381 __bitmap_clear_bits(ctl, info, offset, bytes);
1382 ctl->free_space -= bytes;
1383 }
1384
1385 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1386 struct btrfs_free_space *info, u64 offset,
1387 u64 bytes)
1388 {
1389 unsigned long start, count;
1390
1391 start = offset_to_bit(info->offset, ctl->unit, offset);
1392 count = bytes_to_bits(bytes, ctl->unit);
1393 BUG_ON(start + count > BITS_PER_BITMAP);
1394
1395 bitmap_set(info->bitmap, start, count);
1396
1397 info->bytes += bytes;
1398 ctl->free_space += bytes;
1399 }
1400
1401 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1402 struct btrfs_free_space *bitmap_info, u64 *offset,
1403 u64 *bytes)
1404 {
1405 unsigned long found_bits = 0;
1406 unsigned long bits, i;
1407 unsigned long next_zero;
1408
1409 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1410 max_t(u64, *offset, bitmap_info->offset));
1411 bits = bytes_to_bits(*bytes, ctl->unit);
1412
1413 for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i);
1414 i < BITS_PER_BITMAP;
1415 i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) {
1416 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1417 BITS_PER_BITMAP, i);
1418 if ((next_zero - i) >= bits) {
1419 found_bits = next_zero - i;
1420 break;
1421 }
1422 i = next_zero;
1423 }
1424
1425 if (found_bits) {
1426 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1427 *bytes = (u64)(found_bits) * ctl->unit;
1428 return 0;
1429 }
1430
1431 return -1;
1432 }
1433
1434 static struct btrfs_free_space *
1435 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes)
1436 {
1437 struct btrfs_free_space *entry;
1438 struct rb_node *node;
1439 int ret;
1440
1441 if (!ctl->free_space_offset.rb_node)
1442 return NULL;
1443
1444 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1445 if (!entry)
1446 return NULL;
1447
1448 for (node = &entry->offset_index; node; node = rb_next(node)) {
1449 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1450 if (entry->bytes < *bytes)
1451 continue;
1452
1453 if (entry->bitmap) {
1454 ret = search_bitmap(ctl, entry, offset, bytes);
1455 if (!ret)
1456 return entry;
1457 continue;
1458 }
1459
1460 *offset = entry->offset;
1461 *bytes = entry->bytes;
1462 return entry;
1463 }
1464
1465 return NULL;
1466 }
1467
1468 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1469 struct btrfs_free_space *info, u64 offset)
1470 {
1471 info->offset = offset_to_bitmap(ctl, offset);
1472 info->bytes = 0;
1473 INIT_LIST_HEAD(&info->list);
1474 link_free_space(ctl, info);
1475 ctl->total_bitmaps++;
1476
1477 ctl->op->recalc_thresholds(ctl);
1478 }
1479
1480 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1481 struct btrfs_free_space *bitmap_info)
1482 {
1483 unlink_free_space(ctl, bitmap_info);
1484 kfree(bitmap_info->bitmap);
1485 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1486 ctl->total_bitmaps--;
1487 ctl->op->recalc_thresholds(ctl);
1488 }
1489
1490 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1491 struct btrfs_free_space *bitmap_info,
1492 u64 *offset, u64 *bytes)
1493 {
1494 u64 end;
1495 u64 search_start, search_bytes;
1496 int ret;
1497
1498 again:
1499 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1500
1501 /*
1502 * XXX - this can go away after a few releases.
1503 *
1504 * since the only user of btrfs_remove_free_space is the tree logging
1505 * stuff, and the only way to test that is under crash conditions, we
1506 * want to have this debug stuff here just in case somethings not
1507 * working. Search the bitmap for the space we are trying to use to
1508 * make sure its actually there. If its not there then we need to stop
1509 * because something has gone wrong.
1510 */
1511 search_start = *offset;
1512 search_bytes = *bytes;
1513 search_bytes = min(search_bytes, end - search_start + 1);
1514 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1515 BUG_ON(ret < 0 || search_start != *offset);
1516
1517 if (*offset > bitmap_info->offset && *offset + *bytes > end) {
1518 bitmap_clear_bits(ctl, bitmap_info, *offset, end - *offset + 1);
1519 *bytes -= end - *offset + 1;
1520 *offset = end + 1;
1521 } else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) {
1522 bitmap_clear_bits(ctl, bitmap_info, *offset, *bytes);
1523 *bytes = 0;
1524 }
1525
1526 if (*bytes) {
1527 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1528 if (!bitmap_info->bytes)
1529 free_bitmap(ctl, bitmap_info);
1530
1531 /*
1532 * no entry after this bitmap, but we still have bytes to
1533 * remove, so something has gone wrong.
1534 */
1535 if (!next)
1536 return -EINVAL;
1537
1538 bitmap_info = rb_entry(next, struct btrfs_free_space,
1539 offset_index);
1540
1541 /*
1542 * if the next entry isn't a bitmap we need to return to let the
1543 * extent stuff do its work.
1544 */
1545 if (!bitmap_info->bitmap)
1546 return -EAGAIN;
1547
1548 /*
1549 * Ok the next item is a bitmap, but it may not actually hold
1550 * the information for the rest of this free space stuff, so
1551 * look for it, and if we don't find it return so we can try
1552 * everything over again.
1553 */
1554 search_start = *offset;
1555 search_bytes = *bytes;
1556 ret = search_bitmap(ctl, bitmap_info, &search_start,
1557 &search_bytes);
1558 if (ret < 0 || search_start != *offset)
1559 return -EAGAIN;
1560
1561 goto again;
1562 } else if (!bitmap_info->bytes)
1563 free_bitmap(ctl, bitmap_info);
1564
1565 return 0;
1566 }
1567
1568 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1569 struct btrfs_free_space *info, u64 offset,
1570 u64 bytes)
1571 {
1572 u64 bytes_to_set = 0;
1573 u64 end;
1574
1575 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1576
1577 bytes_to_set = min(end - offset, bytes);
1578
1579 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1580
1581 return bytes_to_set;
1582
1583 }
1584
1585 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1586 struct btrfs_free_space *info)
1587 {
1588 struct btrfs_block_group_cache *block_group = ctl->private;
1589
1590 /*
1591 * If we are below the extents threshold then we can add this as an
1592 * extent, and don't have to deal with the bitmap
1593 */
1594 if (ctl->free_extents < ctl->extents_thresh) {
1595 /*
1596 * If this block group has some small extents we don't want to
1597 * use up all of our free slots in the cache with them, we want
1598 * to reserve them to larger extents, however if we have plent
1599 * of cache left then go ahead an dadd them, no sense in adding
1600 * the overhead of a bitmap if we don't have to.
1601 */
1602 if (info->bytes <= block_group->sectorsize * 4) {
1603 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1604 return false;
1605 } else {
1606 return false;
1607 }
1608 }
1609
1610 /*
1611 * some block groups are so tiny they can't be enveloped by a bitmap, so
1612 * don't even bother to create a bitmap for this
1613 */
1614 if (BITS_PER_BITMAP * block_group->sectorsize >
1615 block_group->key.offset)
1616 return false;
1617
1618 return true;
1619 }
1620
1621 static struct btrfs_free_space_op free_space_op = {
1622 .recalc_thresholds = recalculate_thresholds,
1623 .use_bitmap = use_bitmap,
1624 };
1625
1626 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1627 struct btrfs_free_space *info)
1628 {
1629 struct btrfs_free_space *bitmap_info;
1630 struct btrfs_block_group_cache *block_group = NULL;
1631 int added = 0;
1632 u64 bytes, offset, bytes_added;
1633 int ret;
1634
1635 bytes = info->bytes;
1636 offset = info->offset;
1637
1638 if (!ctl->op->use_bitmap(ctl, info))
1639 return 0;
1640
1641 if (ctl->op == &free_space_op)
1642 block_group = ctl->private;
1643 again:
1644 /*
1645 * Since we link bitmaps right into the cluster we need to see if we
1646 * have a cluster here, and if so and it has our bitmap we need to add
1647 * the free space to that bitmap.
1648 */
1649 if (block_group && !list_empty(&block_group->cluster_list)) {
1650 struct btrfs_free_cluster *cluster;
1651 struct rb_node *node;
1652 struct btrfs_free_space *entry;
1653
1654 cluster = list_entry(block_group->cluster_list.next,
1655 struct btrfs_free_cluster,
1656 block_group_list);
1657 spin_lock(&cluster->lock);
1658 node = rb_first(&cluster->root);
1659 if (!node) {
1660 spin_unlock(&cluster->lock);
1661 goto no_cluster_bitmap;
1662 }
1663
1664 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1665 if (!entry->bitmap) {
1666 spin_unlock(&cluster->lock);
1667 goto no_cluster_bitmap;
1668 }
1669
1670 if (entry->offset == offset_to_bitmap(ctl, offset)) {
1671 bytes_added = add_bytes_to_bitmap(ctl, entry,
1672 offset, bytes);
1673 bytes -= bytes_added;
1674 offset += bytes_added;
1675 }
1676 spin_unlock(&cluster->lock);
1677 if (!bytes) {
1678 ret = 1;
1679 goto out;
1680 }
1681 }
1682
1683 no_cluster_bitmap:
1684 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1685 1, 0);
1686 if (!bitmap_info) {
1687 BUG_ON(added);
1688 goto new_bitmap;
1689 }
1690
1691 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
1692 bytes -= bytes_added;
1693 offset += bytes_added;
1694 added = 0;
1695
1696 if (!bytes) {
1697 ret = 1;
1698 goto out;
1699 } else
1700 goto again;
1701
1702 new_bitmap:
1703 if (info && info->bitmap) {
1704 add_new_bitmap(ctl, info, offset);
1705 added = 1;
1706 info = NULL;
1707 goto again;
1708 } else {
1709 spin_unlock(&ctl->tree_lock);
1710
1711 /* no pre-allocated info, allocate a new one */
1712 if (!info) {
1713 info = kmem_cache_zalloc(btrfs_free_space_cachep,
1714 GFP_NOFS);
1715 if (!info) {
1716 spin_lock(&ctl->tree_lock);
1717 ret = -ENOMEM;
1718 goto out;
1719 }
1720 }
1721
1722 /* allocate the bitmap */
1723 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1724 spin_lock(&ctl->tree_lock);
1725 if (!info->bitmap) {
1726 ret = -ENOMEM;
1727 goto out;
1728 }
1729 goto again;
1730 }
1731
1732 out:
1733 if (info) {
1734 if (info->bitmap)
1735 kfree(info->bitmap);
1736 kmem_cache_free(btrfs_free_space_cachep, info);
1737 }
1738
1739 return ret;
1740 }
1741
1742 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1743 struct btrfs_free_space *info, bool update_stat)
1744 {
1745 struct btrfs_free_space *left_info;
1746 struct btrfs_free_space *right_info;
1747 bool merged = false;
1748 u64 offset = info->offset;
1749 u64 bytes = info->bytes;
1750
1751 /*
1752 * first we want to see if there is free space adjacent to the range we
1753 * are adding, if there is remove that struct and add a new one to
1754 * cover the entire range
1755 */
1756 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1757 if (right_info && rb_prev(&right_info->offset_index))
1758 left_info = rb_entry(rb_prev(&right_info->offset_index),
1759 struct btrfs_free_space, offset_index);
1760 else
1761 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
1762
1763 if (right_info && !right_info->bitmap) {
1764 if (update_stat)
1765 unlink_free_space(ctl, right_info);
1766 else
1767 __unlink_free_space(ctl, right_info);
1768 info->bytes += right_info->bytes;
1769 kmem_cache_free(btrfs_free_space_cachep, right_info);
1770 merged = true;
1771 }
1772
1773 if (left_info && !left_info->bitmap &&
1774 left_info->offset + left_info->bytes == offset) {
1775 if (update_stat)
1776 unlink_free_space(ctl, left_info);
1777 else
1778 __unlink_free_space(ctl, left_info);
1779 info->offset = left_info->offset;
1780 info->bytes += left_info->bytes;
1781 kmem_cache_free(btrfs_free_space_cachep, left_info);
1782 merged = true;
1783 }
1784
1785 return merged;
1786 }
1787
1788 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
1789 u64 offset, u64 bytes)
1790 {
1791 struct btrfs_free_space *info;
1792 int ret = 0;
1793
1794 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1795 if (!info)
1796 return -ENOMEM;
1797
1798 info->offset = offset;
1799 info->bytes = bytes;
1800
1801 spin_lock(&ctl->tree_lock);
1802
1803 if (try_merge_free_space(ctl, info, true))
1804 goto link;
1805
1806 /*
1807 * There was no extent directly to the left or right of this new
1808 * extent then we know we're going to have to allocate a new extent, so
1809 * before we do that see if we need to drop this into a bitmap
1810 */
1811 ret = insert_into_bitmap(ctl, info);
1812 if (ret < 0) {
1813 goto out;
1814 } else if (ret) {
1815 ret = 0;
1816 goto out;
1817 }
1818 link:
1819 ret = link_free_space(ctl, info);
1820 if (ret)
1821 kmem_cache_free(btrfs_free_space_cachep, info);
1822 out:
1823 spin_unlock(&ctl->tree_lock);
1824
1825 if (ret) {
1826 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
1827 BUG_ON(ret == -EEXIST);
1828 }
1829
1830 return ret;
1831 }
1832
1833 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1834 u64 offset, u64 bytes)
1835 {
1836 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1837 struct btrfs_free_space *info;
1838 struct btrfs_free_space *next_info = NULL;
1839 int ret = 0;
1840
1841 spin_lock(&ctl->tree_lock);
1842
1843 again:
1844 info = tree_search_offset(ctl, offset, 0, 0);
1845 if (!info) {
1846 /*
1847 * oops didn't find an extent that matched the space we wanted
1848 * to remove, look for a bitmap instead
1849 */
1850 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1851 1, 0);
1852 if (!info) {
1853 /* the tree logging code might be calling us before we
1854 * have fully loaded the free space rbtree for this
1855 * block group. So it is possible the entry won't
1856 * be in the rbtree yet at all. The caching code
1857 * will make sure not to put it in the rbtree if
1858 * the logging code has pinned it.
1859 */
1860 goto out_lock;
1861 }
1862 }
1863
1864 if (info->bytes < bytes && rb_next(&info->offset_index)) {
1865 u64 end;
1866 next_info = rb_entry(rb_next(&info->offset_index),
1867 struct btrfs_free_space,
1868 offset_index);
1869
1870 if (next_info->bitmap)
1871 end = next_info->offset +
1872 BITS_PER_BITMAP * ctl->unit - 1;
1873 else
1874 end = next_info->offset + next_info->bytes;
1875
1876 if (next_info->bytes < bytes ||
1877 next_info->offset > offset || offset > end) {
1878 printk(KERN_CRIT "Found free space at %llu, size %llu,"
1879 " trying to use %llu\n",
1880 (unsigned long long)info->offset,
1881 (unsigned long long)info->bytes,
1882 (unsigned long long)bytes);
1883 WARN_ON(1);
1884 ret = -EINVAL;
1885 goto out_lock;
1886 }
1887
1888 info = next_info;
1889 }
1890
1891 if (info->bytes == bytes) {
1892 unlink_free_space(ctl, info);
1893 if (info->bitmap) {
1894 kfree(info->bitmap);
1895 ctl->total_bitmaps--;
1896 }
1897 kmem_cache_free(btrfs_free_space_cachep, info);
1898 ret = 0;
1899 goto out_lock;
1900 }
1901
1902 if (!info->bitmap && info->offset == offset) {
1903 unlink_free_space(ctl, info);
1904 info->offset += bytes;
1905 info->bytes -= bytes;
1906 ret = link_free_space(ctl, info);
1907 WARN_ON(ret);
1908 goto out_lock;
1909 }
1910
1911 if (!info->bitmap && info->offset <= offset &&
1912 info->offset + info->bytes >= offset + bytes) {
1913 u64 old_start = info->offset;
1914 /*
1915 * we're freeing space in the middle of the info,
1916 * this can happen during tree log replay
1917 *
1918 * first unlink the old info and then
1919 * insert it again after the hole we're creating
1920 */
1921 unlink_free_space(ctl, info);
1922 if (offset + bytes < info->offset + info->bytes) {
1923 u64 old_end = info->offset + info->bytes;
1924
1925 info->offset = offset + bytes;
1926 info->bytes = old_end - info->offset;
1927 ret = link_free_space(ctl, info);
1928 WARN_ON(ret);
1929 if (ret)
1930 goto out_lock;
1931 } else {
1932 /* the hole we're creating ends at the end
1933 * of the info struct, just free the info
1934 */
1935 kmem_cache_free(btrfs_free_space_cachep, info);
1936 }
1937 spin_unlock(&ctl->tree_lock);
1938
1939 /* step two, insert a new info struct to cover
1940 * anything before the hole
1941 */
1942 ret = btrfs_add_free_space(block_group, old_start,
1943 offset - old_start);
1944 WARN_ON(ret);
1945 goto out;
1946 }
1947
1948 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
1949 if (ret == -EAGAIN)
1950 goto again;
1951 BUG_ON(ret);
1952 out_lock:
1953 spin_unlock(&ctl->tree_lock);
1954 out:
1955 return ret;
1956 }
1957
1958 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
1959 u64 bytes)
1960 {
1961 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1962 struct btrfs_free_space *info;
1963 struct rb_node *n;
1964 int count = 0;
1965
1966 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
1967 info = rb_entry(n, struct btrfs_free_space, offset_index);
1968 if (info->bytes >= bytes)
1969 count++;
1970 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1971 (unsigned long long)info->offset,
1972 (unsigned long long)info->bytes,
1973 (info->bitmap) ? "yes" : "no");
1974 }
1975 printk(KERN_INFO "block group has cluster?: %s\n",
1976 list_empty(&block_group->cluster_list) ? "no" : "yes");
1977 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
1978 "\n", count);
1979 }
1980
1981 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
1982 {
1983 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1984
1985 spin_lock_init(&ctl->tree_lock);
1986 ctl->unit = block_group->sectorsize;
1987 ctl->start = block_group->key.objectid;
1988 ctl->private = block_group;
1989 ctl->op = &free_space_op;
1990
1991 /*
1992 * we only want to have 32k of ram per block group for keeping
1993 * track of free space, and if we pass 1/2 of that we want to
1994 * start converting things over to using bitmaps
1995 */
1996 ctl->extents_thresh = ((1024 * 32) / 2) /
1997 sizeof(struct btrfs_free_space);
1998 }
1999
2000 /*
2001 * for a given cluster, put all of its extents back into the free
2002 * space cache. If the block group passed doesn't match the block group
2003 * pointed to by the cluster, someone else raced in and freed the
2004 * cluster already. In that case, we just return without changing anything
2005 */
2006 static int
2007 __btrfs_return_cluster_to_free_space(
2008 struct btrfs_block_group_cache *block_group,
2009 struct btrfs_free_cluster *cluster)
2010 {
2011 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2012 struct btrfs_free_space *entry;
2013 struct rb_node *node;
2014
2015 spin_lock(&cluster->lock);
2016 if (cluster->block_group != block_group)
2017 goto out;
2018
2019 cluster->block_group = NULL;
2020 cluster->window_start = 0;
2021 list_del_init(&cluster->block_group_list);
2022
2023 node = rb_first(&cluster->root);
2024 while (node) {
2025 bool bitmap;
2026
2027 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2028 node = rb_next(&entry->offset_index);
2029 rb_erase(&entry->offset_index, &cluster->root);
2030
2031 bitmap = (entry->bitmap != NULL);
2032 if (!bitmap)
2033 try_merge_free_space(ctl, entry, false);
2034 tree_insert_offset(&ctl->free_space_offset,
2035 entry->offset, &entry->offset_index, bitmap);
2036 }
2037 cluster->root = RB_ROOT;
2038
2039 out:
2040 spin_unlock(&cluster->lock);
2041 btrfs_put_block_group(block_group);
2042 return 0;
2043 }
2044
2045 void __btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl *ctl)
2046 {
2047 struct btrfs_free_space *info;
2048 struct rb_node *node;
2049
2050 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2051 info = rb_entry(node, struct btrfs_free_space, offset_index);
2052 if (!info->bitmap) {
2053 unlink_free_space(ctl, info);
2054 kmem_cache_free(btrfs_free_space_cachep, info);
2055 } else {
2056 free_bitmap(ctl, info);
2057 }
2058 if (need_resched()) {
2059 spin_unlock(&ctl->tree_lock);
2060 cond_resched();
2061 spin_lock(&ctl->tree_lock);
2062 }
2063 }
2064 }
2065
2066 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2067 {
2068 spin_lock(&ctl->tree_lock);
2069 __btrfs_remove_free_space_cache_locked(ctl);
2070 spin_unlock(&ctl->tree_lock);
2071 }
2072
2073 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2074 {
2075 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2076 struct btrfs_free_cluster *cluster;
2077 struct list_head *head;
2078
2079 spin_lock(&ctl->tree_lock);
2080 while ((head = block_group->cluster_list.next) !=
2081 &block_group->cluster_list) {
2082 cluster = list_entry(head, struct btrfs_free_cluster,
2083 block_group_list);
2084
2085 WARN_ON(cluster->block_group != block_group);
2086 __btrfs_return_cluster_to_free_space(block_group, cluster);
2087 if (need_resched()) {
2088 spin_unlock(&ctl->tree_lock);
2089 cond_resched();
2090 spin_lock(&ctl->tree_lock);
2091 }
2092 }
2093 __btrfs_remove_free_space_cache_locked(ctl);
2094 spin_unlock(&ctl->tree_lock);
2095
2096 }
2097
2098 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2099 u64 offset, u64 bytes, u64 empty_size)
2100 {
2101 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2102 struct btrfs_free_space *entry = NULL;
2103 u64 bytes_search = bytes + empty_size;
2104 u64 ret = 0;
2105
2106 spin_lock(&ctl->tree_lock);
2107 entry = find_free_space(ctl, &offset, &bytes_search);
2108 if (!entry)
2109 goto out;
2110
2111 ret = offset;
2112 if (entry->bitmap) {
2113 bitmap_clear_bits(ctl, entry, offset, bytes);
2114 if (!entry->bytes)
2115 free_bitmap(ctl, entry);
2116 } else {
2117 unlink_free_space(ctl, entry);
2118 entry->offset += bytes;
2119 entry->bytes -= bytes;
2120 if (!entry->bytes)
2121 kmem_cache_free(btrfs_free_space_cachep, entry);
2122 else
2123 link_free_space(ctl, entry);
2124 }
2125
2126 out:
2127 spin_unlock(&ctl->tree_lock);
2128
2129 return ret;
2130 }
2131
2132 /*
2133 * given a cluster, put all of its extents back into the free space
2134 * cache. If a block group is passed, this function will only free
2135 * a cluster that belongs to the passed block group.
2136 *
2137 * Otherwise, it'll get a reference on the block group pointed to by the
2138 * cluster and remove the cluster from it.
2139 */
2140 int btrfs_return_cluster_to_free_space(
2141 struct btrfs_block_group_cache *block_group,
2142 struct btrfs_free_cluster *cluster)
2143 {
2144 struct btrfs_free_space_ctl *ctl;
2145 int ret;
2146
2147 /* first, get a safe pointer to the block group */
2148 spin_lock(&cluster->lock);
2149 if (!block_group) {
2150 block_group = cluster->block_group;
2151 if (!block_group) {
2152 spin_unlock(&cluster->lock);
2153 return 0;
2154 }
2155 } else if (cluster->block_group != block_group) {
2156 /* someone else has already freed it don't redo their work */
2157 spin_unlock(&cluster->lock);
2158 return 0;
2159 }
2160 atomic_inc(&block_group->count);
2161 spin_unlock(&cluster->lock);
2162
2163 ctl = block_group->free_space_ctl;
2164
2165 /* now return any extents the cluster had on it */
2166 spin_lock(&ctl->tree_lock);
2167 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2168 spin_unlock(&ctl->tree_lock);
2169
2170 /* finally drop our ref */
2171 btrfs_put_block_group(block_group);
2172 return ret;
2173 }
2174
2175 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2176 struct btrfs_free_cluster *cluster,
2177 struct btrfs_free_space *entry,
2178 u64 bytes, u64 min_start)
2179 {
2180 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2181 int err;
2182 u64 search_start = cluster->window_start;
2183 u64 search_bytes = bytes;
2184 u64 ret = 0;
2185
2186 search_start = min_start;
2187 search_bytes = bytes;
2188
2189 err = search_bitmap(ctl, entry, &search_start, &search_bytes);
2190 if (err)
2191 return 0;
2192
2193 ret = search_start;
2194 __bitmap_clear_bits(ctl, entry, ret, bytes);
2195
2196 return ret;
2197 }
2198
2199 /*
2200 * given a cluster, try to allocate 'bytes' from it, returns 0
2201 * if it couldn't find anything suitably large, or a logical disk offset
2202 * if things worked out
2203 */
2204 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2205 struct btrfs_free_cluster *cluster, u64 bytes,
2206 u64 min_start)
2207 {
2208 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2209 struct btrfs_free_space *entry = NULL;
2210 struct rb_node *node;
2211 u64 ret = 0;
2212
2213 spin_lock(&cluster->lock);
2214 if (bytes > cluster->max_size)
2215 goto out;
2216
2217 if (cluster->block_group != block_group)
2218 goto out;
2219
2220 node = rb_first(&cluster->root);
2221 if (!node)
2222 goto out;
2223
2224 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2225 while(1) {
2226 if (entry->bytes < bytes ||
2227 (!entry->bitmap && entry->offset < min_start)) {
2228 node = rb_next(&entry->offset_index);
2229 if (!node)
2230 break;
2231 entry = rb_entry(node, struct btrfs_free_space,
2232 offset_index);
2233 continue;
2234 }
2235
2236 if (entry->bitmap) {
2237 ret = btrfs_alloc_from_bitmap(block_group,
2238 cluster, entry, bytes,
2239 min_start);
2240 if (ret == 0) {
2241 node = rb_next(&entry->offset_index);
2242 if (!node)
2243 break;
2244 entry = rb_entry(node, struct btrfs_free_space,
2245 offset_index);
2246 continue;
2247 }
2248 } else {
2249 ret = entry->offset;
2250
2251 entry->offset += bytes;
2252 entry->bytes -= bytes;
2253 }
2254
2255 if (entry->bytes == 0)
2256 rb_erase(&entry->offset_index, &cluster->root);
2257 break;
2258 }
2259 out:
2260 spin_unlock(&cluster->lock);
2261
2262 if (!ret)
2263 return 0;
2264
2265 spin_lock(&ctl->tree_lock);
2266
2267 ctl->free_space -= bytes;
2268 if (entry->bytes == 0) {
2269 ctl->free_extents--;
2270 if (entry->bitmap) {
2271 kfree(entry->bitmap);
2272 ctl->total_bitmaps--;
2273 ctl->op->recalc_thresholds(ctl);
2274 }
2275 kmem_cache_free(btrfs_free_space_cachep, entry);
2276 }
2277
2278 spin_unlock(&ctl->tree_lock);
2279
2280 return ret;
2281 }
2282
2283 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2284 struct btrfs_free_space *entry,
2285 struct btrfs_free_cluster *cluster,
2286 u64 offset, u64 bytes, u64 min_bytes)
2287 {
2288 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2289 unsigned long next_zero;
2290 unsigned long i;
2291 unsigned long search_bits;
2292 unsigned long total_bits;
2293 unsigned long found_bits;
2294 unsigned long start = 0;
2295 unsigned long total_found = 0;
2296 int ret;
2297 bool found = false;
2298
2299 i = offset_to_bit(entry->offset, block_group->sectorsize,
2300 max_t(u64, offset, entry->offset));
2301 search_bits = bytes_to_bits(bytes, block_group->sectorsize);
2302 total_bits = bytes_to_bits(min_bytes, block_group->sectorsize);
2303
2304 again:
2305 found_bits = 0;
2306 for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i);
2307 i < BITS_PER_BITMAP;
2308 i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) {
2309 next_zero = find_next_zero_bit(entry->bitmap,
2310 BITS_PER_BITMAP, i);
2311 if (next_zero - i >= search_bits) {
2312 found_bits = next_zero - i;
2313 break;
2314 }
2315 i = next_zero;
2316 }
2317
2318 if (!found_bits)
2319 return -ENOSPC;
2320
2321 if (!found) {
2322 start = i;
2323 cluster->max_size = 0;
2324 found = true;
2325 }
2326
2327 total_found += found_bits;
2328
2329 if (cluster->max_size < found_bits * block_group->sectorsize)
2330 cluster->max_size = found_bits * block_group->sectorsize;
2331
2332 if (total_found < total_bits) {
2333 i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, next_zero);
2334 if (i - start > total_bits * 2) {
2335 total_found = 0;
2336 cluster->max_size = 0;
2337 found = false;
2338 }
2339 goto again;
2340 }
2341
2342 cluster->window_start = start * block_group->sectorsize +
2343 entry->offset;
2344 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2345 ret = tree_insert_offset(&cluster->root, entry->offset,
2346 &entry->offset_index, 1);
2347 BUG_ON(ret);
2348
2349 return 0;
2350 }
2351
2352 /*
2353 * This searches the block group for just extents to fill the cluster with.
2354 */
2355 static noinline int
2356 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2357 struct btrfs_free_cluster *cluster,
2358 struct list_head *bitmaps, u64 offset, u64 bytes,
2359 u64 min_bytes)
2360 {
2361 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2362 struct btrfs_free_space *first = NULL;
2363 struct btrfs_free_space *entry = NULL;
2364 struct btrfs_free_space *prev = NULL;
2365 struct btrfs_free_space *last;
2366 struct rb_node *node;
2367 u64 window_start;
2368 u64 window_free;
2369 u64 max_extent;
2370 u64 max_gap = 128 * 1024;
2371
2372 entry = tree_search_offset(ctl, offset, 0, 1);
2373 if (!entry)
2374 return -ENOSPC;
2375
2376 /*
2377 * We don't want bitmaps, so just move along until we find a normal
2378 * extent entry.
2379 */
2380 while (entry->bitmap) {
2381 if (list_empty(&entry->list))
2382 list_add_tail(&entry->list, bitmaps);
2383 node = rb_next(&entry->offset_index);
2384 if (!node)
2385 return -ENOSPC;
2386 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2387 }
2388
2389 window_start = entry->offset;
2390 window_free = entry->bytes;
2391 max_extent = entry->bytes;
2392 first = entry;
2393 last = entry;
2394 prev = entry;
2395
2396 while (window_free <= min_bytes) {
2397 node = rb_next(&entry->offset_index);
2398 if (!node)
2399 return -ENOSPC;
2400 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2401
2402 if (entry->bitmap) {
2403 if (list_empty(&entry->list))
2404 list_add_tail(&entry->list, bitmaps);
2405 continue;
2406 }
2407
2408 /*
2409 * we haven't filled the empty size and the window is
2410 * very large. reset and try again
2411 */
2412 if (entry->offset - (prev->offset + prev->bytes) > max_gap ||
2413 entry->offset - window_start > (min_bytes * 2)) {
2414 first = entry;
2415 window_start = entry->offset;
2416 window_free = entry->bytes;
2417 last = entry;
2418 max_extent = entry->bytes;
2419 } else {
2420 last = entry;
2421 window_free += entry->bytes;
2422 if (entry->bytes > max_extent)
2423 max_extent = entry->bytes;
2424 }
2425 prev = entry;
2426 }
2427
2428 cluster->window_start = first->offset;
2429
2430 node = &first->offset_index;
2431
2432 /*
2433 * now we've found our entries, pull them out of the free space
2434 * cache and put them into the cluster rbtree
2435 */
2436 do {
2437 int ret;
2438
2439 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2440 node = rb_next(&entry->offset_index);
2441 if (entry->bitmap)
2442 continue;
2443
2444 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2445 ret = tree_insert_offset(&cluster->root, entry->offset,
2446 &entry->offset_index, 0);
2447 BUG_ON(ret);
2448 } while (node && entry != last);
2449
2450 cluster->max_size = max_extent;
2451
2452 return 0;
2453 }
2454
2455 /*
2456 * This specifically looks for bitmaps that may work in the cluster, we assume
2457 * that we have already failed to find extents that will work.
2458 */
2459 static noinline int
2460 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2461 struct btrfs_free_cluster *cluster,
2462 struct list_head *bitmaps, u64 offset, u64 bytes,
2463 u64 min_bytes)
2464 {
2465 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2466 struct btrfs_free_space *entry;
2467 int ret = -ENOSPC;
2468 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2469
2470 if (ctl->total_bitmaps == 0)
2471 return -ENOSPC;
2472
2473 /*
2474 * The bitmap that covers offset won't be in the list unless offset
2475 * is just its start offset.
2476 */
2477 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2478 if (entry->offset != bitmap_offset) {
2479 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2480 if (entry && list_empty(&entry->list))
2481 list_add(&entry->list, bitmaps);
2482 }
2483
2484 list_for_each_entry(entry, bitmaps, list) {
2485 if (entry->bytes < min_bytes)
2486 continue;
2487 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2488 bytes, min_bytes);
2489 if (!ret)
2490 return 0;
2491 }
2492
2493 /*
2494 * The bitmaps list has all the bitmaps that record free space
2495 * starting after offset, so no more search is required.
2496 */
2497 return -ENOSPC;
2498 }
2499
2500 /*
2501 * here we try to find a cluster of blocks in a block group. The goal
2502 * is to find at least bytes free and up to empty_size + bytes free.
2503 * We might not find them all in one contiguous area.
2504 *
2505 * returns zero and sets up cluster if things worked out, otherwise
2506 * it returns -enospc
2507 */
2508 int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
2509 struct btrfs_root *root,
2510 struct btrfs_block_group_cache *block_group,
2511 struct btrfs_free_cluster *cluster,
2512 u64 offset, u64 bytes, u64 empty_size)
2513 {
2514 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2515 struct btrfs_free_space *entry, *tmp;
2516 LIST_HEAD(bitmaps);
2517 u64 min_bytes;
2518 int ret;
2519
2520 /* for metadata, allow allocates with more holes */
2521 if (btrfs_test_opt(root, SSD_SPREAD)) {
2522 min_bytes = bytes + empty_size;
2523 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2524 /*
2525 * we want to do larger allocations when we are
2526 * flushing out the delayed refs, it helps prevent
2527 * making more work as we go along.
2528 */
2529 if (trans->transaction->delayed_refs.flushing)
2530 min_bytes = max(bytes, (bytes + empty_size) >> 1);
2531 else
2532 min_bytes = max(bytes, (bytes + empty_size) >> 4);
2533 } else
2534 min_bytes = max(bytes, (bytes + empty_size) >> 2);
2535
2536 spin_lock(&ctl->tree_lock);
2537
2538 /*
2539 * If we know we don't have enough space to make a cluster don't even
2540 * bother doing all the work to try and find one.
2541 */
2542 if (ctl->free_space < min_bytes) {
2543 spin_unlock(&ctl->tree_lock);
2544 return -ENOSPC;
2545 }
2546
2547 spin_lock(&cluster->lock);
2548
2549 /* someone already found a cluster, hooray */
2550 if (cluster->block_group) {
2551 ret = 0;
2552 goto out;
2553 }
2554
2555 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2556 bytes, min_bytes);
2557 if (ret)
2558 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2559 offset, bytes, min_bytes);
2560
2561 /* Clear our temporary list */
2562 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
2563 list_del_init(&entry->list);
2564
2565 if (!ret) {
2566 atomic_inc(&block_group->count);
2567 list_add_tail(&cluster->block_group_list,
2568 &block_group->cluster_list);
2569 cluster->block_group = block_group;
2570 }
2571 out:
2572 spin_unlock(&cluster->lock);
2573 spin_unlock(&ctl->tree_lock);
2574
2575 return ret;
2576 }
2577
2578 /*
2579 * simple code to zero out a cluster
2580 */
2581 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2582 {
2583 spin_lock_init(&cluster->lock);
2584 spin_lock_init(&cluster->refill_lock);
2585 cluster->root = RB_ROOT;
2586 cluster->max_size = 0;
2587 INIT_LIST_HEAD(&cluster->block_group_list);
2588 cluster->block_group = NULL;
2589 }
2590
2591 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2592 u64 *trimmed, u64 start, u64 end, u64 minlen)
2593 {
2594 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2595 struct btrfs_free_space *entry = NULL;
2596 struct btrfs_fs_info *fs_info = block_group->fs_info;
2597 u64 bytes = 0;
2598 u64 actually_trimmed;
2599 int ret = 0;
2600
2601 *trimmed = 0;
2602
2603 while (start < end) {
2604 spin_lock(&ctl->tree_lock);
2605
2606 if (ctl->free_space < minlen) {
2607 spin_unlock(&ctl->tree_lock);
2608 break;
2609 }
2610
2611 entry = tree_search_offset(ctl, start, 0, 1);
2612 if (!entry)
2613 entry = tree_search_offset(ctl,
2614 offset_to_bitmap(ctl, start),
2615 1, 1);
2616
2617 if (!entry || entry->offset >= end) {
2618 spin_unlock(&ctl->tree_lock);
2619 break;
2620 }
2621
2622 if (entry->bitmap) {
2623 ret = search_bitmap(ctl, entry, &start, &bytes);
2624 if (!ret) {
2625 if (start >= end) {
2626 spin_unlock(&ctl->tree_lock);
2627 break;
2628 }
2629 bytes = min(bytes, end - start);
2630 bitmap_clear_bits(ctl, entry, start, bytes);
2631 if (entry->bytes == 0)
2632 free_bitmap(ctl, entry);
2633 } else {
2634 start = entry->offset + BITS_PER_BITMAP *
2635 block_group->sectorsize;
2636 spin_unlock(&ctl->tree_lock);
2637 ret = 0;
2638 continue;
2639 }
2640 } else {
2641 start = entry->offset;
2642 bytes = min(entry->bytes, end - start);
2643 unlink_free_space(ctl, entry);
2644 kmem_cache_free(btrfs_free_space_cachep, entry);
2645 }
2646
2647 spin_unlock(&ctl->tree_lock);
2648
2649 if (bytes >= minlen) {
2650 struct btrfs_space_info *space_info;
2651 int update = 0;
2652
2653 space_info = block_group->space_info;
2654 spin_lock(&space_info->lock);
2655 spin_lock(&block_group->lock);
2656 if (!block_group->ro) {
2657 block_group->reserved += bytes;
2658 space_info->bytes_reserved += bytes;
2659 update = 1;
2660 }
2661 spin_unlock(&block_group->lock);
2662 spin_unlock(&space_info->lock);
2663
2664 ret = btrfs_error_discard_extent(fs_info->extent_root,
2665 start,
2666 bytes,
2667 &actually_trimmed);
2668
2669 btrfs_add_free_space(block_group, start, bytes);
2670 if (update) {
2671 spin_lock(&space_info->lock);
2672 spin_lock(&block_group->lock);
2673 if (block_group->ro)
2674 space_info->bytes_readonly += bytes;
2675 block_group->reserved -= bytes;
2676 space_info->bytes_reserved -= bytes;
2677 spin_unlock(&space_info->lock);
2678 spin_unlock(&block_group->lock);
2679 }
2680
2681 if (ret)
2682 break;
2683 *trimmed += actually_trimmed;
2684 }
2685 start += bytes;
2686 bytes = 0;
2687
2688 if (fatal_signal_pending(current)) {
2689 ret = -ERESTARTSYS;
2690 break;
2691 }
2692
2693 cond_resched();
2694 }
2695
2696 return ret;
2697 }
2698
2699 /*
2700 * Find the left-most item in the cache tree, and then return the
2701 * smallest inode number in the item.
2702 *
2703 * Note: the returned inode number may not be the smallest one in
2704 * the tree, if the left-most item is a bitmap.
2705 */
2706 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
2707 {
2708 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
2709 struct btrfs_free_space *entry = NULL;
2710 u64 ino = 0;
2711
2712 spin_lock(&ctl->tree_lock);
2713
2714 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
2715 goto out;
2716
2717 entry = rb_entry(rb_first(&ctl->free_space_offset),
2718 struct btrfs_free_space, offset_index);
2719
2720 if (!entry->bitmap) {
2721 ino = entry->offset;
2722
2723 unlink_free_space(ctl, entry);
2724 entry->offset++;
2725 entry->bytes--;
2726 if (!entry->bytes)
2727 kmem_cache_free(btrfs_free_space_cachep, entry);
2728 else
2729 link_free_space(ctl, entry);
2730 } else {
2731 u64 offset = 0;
2732 u64 count = 1;
2733 int ret;
2734
2735 ret = search_bitmap(ctl, entry, &offset, &count);
2736 BUG_ON(ret);
2737
2738 ino = offset;
2739 bitmap_clear_bits(ctl, entry, offset, 1);
2740 if (entry->bytes == 0)
2741 free_bitmap(ctl, entry);
2742 }
2743 out:
2744 spin_unlock(&ctl->tree_lock);
2745
2746 return ino;
2747 }
2748
2749 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
2750 struct btrfs_path *path)
2751 {
2752 struct inode *inode = NULL;
2753
2754 spin_lock(&root->cache_lock);
2755 if (root->cache_inode)
2756 inode = igrab(root->cache_inode);
2757 spin_unlock(&root->cache_lock);
2758 if (inode)
2759 return inode;
2760
2761 inode = __lookup_free_space_inode(root, path, 0);
2762 if (IS_ERR(inode))
2763 return inode;
2764
2765 spin_lock(&root->cache_lock);
2766 if (!btrfs_fs_closing(root->fs_info))
2767 root->cache_inode = igrab(inode);
2768 spin_unlock(&root->cache_lock);
2769
2770 return inode;
2771 }
2772
2773 int create_free_ino_inode(struct btrfs_root *root,
2774 struct btrfs_trans_handle *trans,
2775 struct btrfs_path *path)
2776 {
2777 return __create_free_space_inode(root, trans, path,
2778 BTRFS_FREE_INO_OBJECTID, 0);
2779 }
2780
2781 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2782 {
2783 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2784 struct btrfs_path *path;
2785 struct inode *inode;
2786 int ret = 0;
2787 u64 root_gen = btrfs_root_generation(&root->root_item);
2788
2789 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2790 return 0;
2791
2792 /*
2793 * If we're unmounting then just return, since this does a search on the
2794 * normal root and not the commit root and we could deadlock.
2795 */
2796 if (btrfs_fs_closing(fs_info))
2797 return 0;
2798
2799 path = btrfs_alloc_path();
2800 if (!path)
2801 return 0;
2802
2803 inode = lookup_free_ino_inode(root, path);
2804 if (IS_ERR(inode))
2805 goto out;
2806
2807 if (root_gen != BTRFS_I(inode)->generation)
2808 goto out_put;
2809
2810 ret = __load_free_space_cache(root, inode, ctl, path, 0);
2811
2812 if (ret < 0)
2813 printk(KERN_ERR "btrfs: failed to load free ino cache for "
2814 "root %llu\n", root->root_key.objectid);
2815 out_put:
2816 iput(inode);
2817 out:
2818 btrfs_free_path(path);
2819 return ret;
2820 }
2821
2822 int btrfs_write_out_ino_cache(struct btrfs_root *root,
2823 struct btrfs_trans_handle *trans,
2824 struct btrfs_path *path)
2825 {
2826 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2827 struct inode *inode;
2828 int ret;
2829
2830 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2831 return 0;
2832
2833 inode = lookup_free_ino_inode(root, path);
2834 if (IS_ERR(inode))
2835 return 0;
2836
2837 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
2838 if (ret) {
2839 btrfs_delalloc_release_metadata(inode, inode->i_size);
2840 #ifdef DEBUG
2841 printk(KERN_ERR "btrfs: failed to write free ino cache "
2842 "for root %llu\n", root->root_key.objectid);
2843 #endif
2844 }
2845
2846 iput(inode);
2847 return ret;
2848 }