projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: remove warn on in free space cache writeout
[deliverable/linux.git] / fs / btrfs / free-space-cache.c
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_truncate_free_space_cache(struct btrfs_root *root,
201 struct btrfs_trans_handle *trans,
202 struct btrfs_path *path,
203 struct inode *inode)
204 {
205 struct btrfs_block_rsv *rsv;
206 u64 needed_bytes;
207 loff_t oldsize;
208 int ret = 0;
209
210 rsv = trans->block_rsv;
211 trans->block_rsv = &root->fs_info->global_block_rsv;
212
213 /* 1 for slack space, 1 for updating the inode */
214 needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
215 btrfs_calc_trans_metadata_size(root, 1);
216
217 spin_lock(&trans->block_rsv->lock);
218 if (trans->block_rsv->reserved < needed_bytes) {
219 spin_unlock(&trans->block_rsv->lock);
220 trans->block_rsv = rsv;
221 return -ENOSPC;
222 }
223 spin_unlock(&trans->block_rsv->lock);
224
225 oldsize = i_size_read(inode);
226 btrfs_i_size_write(inode, 0);
227 truncate_pagecache(inode, oldsize, 0);
228
229 /*
230 * We don't need an orphan item because truncating the free space cache
231 * will never be split across transactions.
232 */
233 ret = btrfs_truncate_inode_items(trans, root, inode,
234 0, BTRFS_EXTENT_DATA_KEY);
235
236 if (ret) {
237 trans->block_rsv = rsv;
238 btrfs_abort_transaction(trans, root, ret);
239 return ret;
240 }
241
242 ret = btrfs_update_inode(trans, root, inode);
243 if (ret)
244 btrfs_abort_transaction(trans, root, ret);
245 trans->block_rsv = rsv;
246
247 return ret;
248 }
249
250 static int readahead_cache(struct inode *inode)
251 {
252 struct file_ra_state *ra;
253 unsigned long last_index;
254
255 ra = kzalloc(sizeof(*ra), GFP_NOFS);
256 if (!ra)
257 return -ENOMEM;
258
259 file_ra_state_init(ra, inode->i_mapping);
260 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
261
262 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
263
264 kfree(ra);
265
266 return 0;
267 }
268
269 struct io_ctl {
270 void *cur, *orig;
271 struct page *page;
272 struct page **pages;
273 struct btrfs_root *root;
274 unsigned long size;
275 int index;
276 int num_pages;
277 unsigned check_crcs:1;
278 };
279
280 static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode,
281 struct btrfs_root *root)
282 {
283 memset(io_ctl, 0, sizeof(struct io_ctl));
284 io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
285 PAGE_CACHE_SHIFT;
286 io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages,
287 GFP_NOFS);
288 if (!io_ctl->pages)
289 return -ENOMEM;
290 io_ctl->root = root;
291 if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
292 io_ctl->check_crcs = 1;
293 return 0;
294 }
295
296 static void io_ctl_free(struct io_ctl *io_ctl)
297 {
298 kfree(io_ctl->pages);
299 }
300
301 static void io_ctl_unmap_page(struct io_ctl *io_ctl)
302 {
303 if (io_ctl->cur) {
304 kunmap(io_ctl->page);
305 io_ctl->cur = NULL;
306 io_ctl->orig = NULL;
307 }
308 }
309
310 static void io_ctl_map_page(struct io_ctl *io_ctl, int clear)
311 {
312 BUG_ON(io_ctl->index >= io_ctl->num_pages);
313 io_ctl->page = io_ctl->pages[io_ctl->index++];
314 io_ctl->cur = kmap(io_ctl->page);
315 io_ctl->orig = io_ctl->cur;
316 io_ctl->size = PAGE_CACHE_SIZE;
317 if (clear)
318 memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
319 }
320
321 static void io_ctl_drop_pages(struct io_ctl *io_ctl)
322 {
323 int i;
324
325 io_ctl_unmap_page(io_ctl);
326
327 for (i = 0; i < io_ctl->num_pages; i++) {
328 if (io_ctl->pages[i]) {
329 ClearPageChecked(io_ctl->pages[i]);
330 unlock_page(io_ctl->pages[i]);
331 page_cache_release(io_ctl->pages[i]);
332 }
333 }
334 }
335
336 static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode,
337 int uptodate)
338 {
339 struct page *page;
340 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
341 int i;
342
343 for (i = 0; i < io_ctl->num_pages; i++) {
344 page = find_or_create_page(inode->i_mapping, i, mask);
345 if (!page) {
346 io_ctl_drop_pages(io_ctl);
347 return -ENOMEM;
348 }
349 io_ctl->pages[i] = page;
350 if (uptodate && !PageUptodate(page)) {
351 btrfs_readpage(NULL, page);
352 lock_page(page);
353 if (!PageUptodate(page)) {
354 printk(KERN_ERR "btrfs: error reading free "
355 "space cache\n");
356 io_ctl_drop_pages(io_ctl);
357 return -EIO;
358 }
359 }
360 }
361
362 for (i = 0; i < io_ctl->num_pages; i++) {
363 clear_page_dirty_for_io(io_ctl->pages[i]);
364 set_page_extent_mapped(io_ctl->pages[i]);
365 }
366
367 return 0;
368 }
369
370 static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation)
371 {
372 __le64 *val;
373
374 io_ctl_map_page(io_ctl, 1);
375
376 /*
377 * Skip the csum areas. If we don't check crcs then we just have a
378 * 64bit chunk at the front of the first page.
379 */
380 if (io_ctl->check_crcs) {
381 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
382 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
383 } else {
384 io_ctl->cur += sizeof(u64);
385 io_ctl->size -= sizeof(u64) * 2;
386 }
387
388 val = io_ctl->cur;
389 *val = cpu_to_le64(generation);
390 io_ctl->cur += sizeof(u64);
391 }
392
393 static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation)
394 {
395 __le64 *gen;
396
397 /*
398 * Skip the crc area. If we don't check crcs then we just have a 64bit
399 * chunk at the front of the first page.
400 */
401 if (io_ctl->check_crcs) {
402 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
403 io_ctl->size -= sizeof(u64) +
404 (sizeof(u32) * io_ctl->num_pages);
405 } else {
406 io_ctl->cur += sizeof(u64);
407 io_ctl->size -= sizeof(u64) * 2;
408 }
409
410 gen = io_ctl->cur;
411 if (le64_to_cpu(*gen) != generation) {
412 printk_ratelimited(KERN_ERR "btrfs: space cache generation "
413 "(%Lu) does not match inode (%Lu)\n", *gen,
414 generation);
415 io_ctl_unmap_page(io_ctl);
416 return -EIO;
417 }
418 io_ctl->cur += sizeof(u64);
419 return 0;
420 }
421
422 static void io_ctl_set_crc(struct io_ctl *io_ctl, int index)
423 {
424 u32 *tmp;
425 u32 crc = ~(u32)0;
426 unsigned offset = 0;
427
428 if (!io_ctl->check_crcs) {
429 io_ctl_unmap_page(io_ctl);
430 return;
431 }
432
433 if (index == 0)
434 offset = sizeof(u32) * io_ctl->num_pages;
435
436 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
437 PAGE_CACHE_SIZE - offset);
438 btrfs_csum_final(crc, (char *)&crc);
439 io_ctl_unmap_page(io_ctl);
440 tmp = kmap(io_ctl->pages[0]);
441 tmp += index;
442 *tmp = crc;
443 kunmap(io_ctl->pages[0]);
444 }
445
446 static int io_ctl_check_crc(struct io_ctl *io_ctl, int index)
447 {
448 u32 *tmp, val;
449 u32 crc = ~(u32)0;
450 unsigned offset = 0;
451
452 if (!io_ctl->check_crcs) {
453 io_ctl_map_page(io_ctl, 0);
454 return 0;
455 }
456
457 if (index == 0)
458 offset = sizeof(u32) * io_ctl->num_pages;
459
460 tmp = kmap(io_ctl->pages[0]);
461 tmp += index;
462 val = *tmp;
463 kunmap(io_ctl->pages[0]);
464
465 io_ctl_map_page(io_ctl, 0);
466 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
467 PAGE_CACHE_SIZE - offset);
468 btrfs_csum_final(crc, (char *)&crc);
469 if (val != crc) {
470 printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free "
471 "space cache\n");
472 io_ctl_unmap_page(io_ctl);
473 return -EIO;
474 }
475
476 return 0;
477 }
478
479 static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes,
480 void *bitmap)
481 {
482 struct btrfs_free_space_entry *entry;
483
484 if (!io_ctl->cur)
485 return -ENOSPC;
486
487 entry = io_ctl->cur;
488 entry->offset = cpu_to_le64(offset);
489 entry->bytes = cpu_to_le64(bytes);
490 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
491 BTRFS_FREE_SPACE_EXTENT;
492 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
493 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
494
495 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
496 return 0;
497
498 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
499
500 /* No more pages to map */
501 if (io_ctl->index >= io_ctl->num_pages)
502 return 0;
503
504 /* map the next page */
505 io_ctl_map_page(io_ctl, 1);
506 return 0;
507 }
508
509 static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap)
510 {
511 if (!io_ctl->cur)
512 return -ENOSPC;
513
514 /*
515 * If we aren't at the start of the current page, unmap this one and
516 * map the next one if there is any left.
517 */
518 if (io_ctl->cur != io_ctl->orig) {
519 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
520 if (io_ctl->index >= io_ctl->num_pages)
521 return -ENOSPC;
522 io_ctl_map_page(io_ctl, 0);
523 }
524
525 memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
526 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
527 if (io_ctl->index < io_ctl->num_pages)
528 io_ctl_map_page(io_ctl, 0);
529 return 0;
530 }
531
532 static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl)
533 {
534 /*
535 * If we're not on the boundary we know we've modified the page and we
536 * need to crc the page.
537 */
538 if (io_ctl->cur != io_ctl->orig)
539 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
540 else
541 io_ctl_unmap_page(io_ctl);
542
543 while (io_ctl->index < io_ctl->num_pages) {
544 io_ctl_map_page(io_ctl, 1);
545 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
546 }
547 }
548
549 static int io_ctl_read_entry(struct io_ctl *io_ctl,
550 struct btrfs_free_space *entry, u8 *type)
551 {
552 struct btrfs_free_space_entry *e;
553 int ret;
554
555 if (!io_ctl->cur) {
556 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
557 if (ret)
558 return ret;
559 }
560
561 e = io_ctl->cur;
562 entry->offset = le64_to_cpu(e->offset);
563 entry->bytes = le64_to_cpu(e->bytes);
564 *type = e->type;
565 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
566 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
567
568 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
569 return 0;
570
571 io_ctl_unmap_page(io_ctl);
572
573 return 0;
574 }
575
576 static int io_ctl_read_bitmap(struct io_ctl *io_ctl,
577 struct btrfs_free_space *entry)
578 {
579 int ret;
580
581 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
582 if (ret)
583 return ret;
584
585 memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
586 io_ctl_unmap_page(io_ctl);
587
588 return 0;
589 }
590
591 /*
592 * Since we attach pinned extents after the fact we can have contiguous sections
593 * of free space that are split up in entries. This poses a problem with the
594 * tree logging stuff since it could have allocated across what appears to be 2
595 * entries since we would have merged the entries when adding the pinned extents
596 * back to the free space cache. So run through the space cache that we just
597 * loaded and merge contiguous entries. This will make the log replay stuff not
598 * blow up and it will make for nicer allocator behavior.
599 */
600 static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
601 {
602 struct btrfs_free_space *e, *prev = NULL;
603 struct rb_node *n;
604
605 again:
606 spin_lock(&ctl->tree_lock);
607 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
608 e = rb_entry(n, struct btrfs_free_space, offset_index);
609 if (!prev)
610 goto next;
611 if (e->bitmap || prev->bitmap)
612 goto next;
613 if (prev->offset + prev->bytes == e->offset) {
614 unlink_free_space(ctl, prev);
615 unlink_free_space(ctl, e);
616 prev->bytes += e->bytes;
617 kmem_cache_free(btrfs_free_space_cachep, e);
618 link_free_space(ctl, prev);
619 prev = NULL;
620 spin_unlock(&ctl->tree_lock);
621 goto again;
622 }
623 next:
624 prev = e;
625 }
626 spin_unlock(&ctl->tree_lock);
627 }
628
629 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
630 struct btrfs_free_space_ctl *ctl,
631 struct btrfs_path *path, u64 offset)
632 {
633 struct btrfs_free_space_header *header;
634 struct extent_buffer *leaf;
635 struct io_ctl io_ctl;
636 struct btrfs_key key;
637 struct btrfs_free_space *e, *n;
638 struct list_head bitmaps;
639 u64 num_entries;
640 u64 num_bitmaps;
641 u64 generation;
642 u8 type;
643 int ret = 0;
644
645 INIT_LIST_HEAD(&bitmaps);
646
647 /* Nothing in the space cache, goodbye */
648 if (!i_size_read(inode))
649 return 0;
650
651 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
652 key.offset = offset;
653 key.type = 0;
654
655 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
656 if (ret < 0)
657 return 0;
658 else if (ret > 0) {
659 btrfs_release_path(path);
660 return 0;
661 }
662
663 ret = -1;
664
665 leaf = path->nodes[0];
666 header = btrfs_item_ptr(leaf, path->slots[0],
667 struct btrfs_free_space_header);
668 num_entries = btrfs_free_space_entries(leaf, header);
669 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
670 generation = btrfs_free_space_generation(leaf, header);
671 btrfs_release_path(path);
672
673 if (BTRFS_I(inode)->generation != generation) {
674 btrfs_err(root->fs_info,
675 "free space inode generation (%llu) "
676 "did not match free space cache generation (%llu)",
677 (unsigned long long)BTRFS_I(inode)->generation,
678 (unsigned long long)generation);
679 return 0;
680 }
681
682 if (!num_entries)
683 return 0;
684
685 ret = io_ctl_init(&io_ctl, inode, root);
686 if (ret)
687 return ret;
688
689 ret = readahead_cache(inode);
690 if (ret)
691 goto out;
692
693 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
694 if (ret)
695 goto out;
696
697 ret = io_ctl_check_crc(&io_ctl, 0);
698 if (ret)
699 goto free_cache;
700
701 ret = io_ctl_check_generation(&io_ctl, generation);
702 if (ret)
703 goto free_cache;
704
705 while (num_entries) {
706 e = kmem_cache_zalloc(btrfs_free_space_cachep,
707 GFP_NOFS);
708 if (!e)
709 goto free_cache;
710
711 ret = io_ctl_read_entry(&io_ctl, e, &type);
712 if (ret) {
713 kmem_cache_free(btrfs_free_space_cachep, e);
714 goto free_cache;
715 }
716
717 if (!e->bytes) {
718 kmem_cache_free(btrfs_free_space_cachep, e);
719 goto free_cache;
720 }
721
722 if (type == BTRFS_FREE_SPACE_EXTENT) {
723 spin_lock(&ctl->tree_lock);
724 ret = link_free_space(ctl, e);
725 spin_unlock(&ctl->tree_lock);
726 if (ret) {
727 btrfs_err(root->fs_info,
728 "Duplicate entries in free space cache, dumping");
729 kmem_cache_free(btrfs_free_space_cachep, e);
730 goto free_cache;
731 }
732 } else {
733 BUG_ON(!num_bitmaps);
734 num_bitmaps--;
735 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
736 if (!e->bitmap) {
737 kmem_cache_free(
738 btrfs_free_space_cachep, e);
739 goto free_cache;
740 }
741 spin_lock(&ctl->tree_lock);
742 ret = link_free_space(ctl, e);
743 ctl->total_bitmaps++;
744 ctl->op->recalc_thresholds(ctl);
745 spin_unlock(&ctl->tree_lock);
746 if (ret) {
747 btrfs_err(root->fs_info,
748 "Duplicate entries in free space cache, dumping");
749 kmem_cache_free(btrfs_free_space_cachep, e);
750 goto free_cache;
751 }
752 list_add_tail(&e->list, &bitmaps);
753 }
754
755 num_entries--;
756 }
757
758 io_ctl_unmap_page(&io_ctl);
759
760 /*
761 * We add the bitmaps at the end of the entries in order that
762 * the bitmap entries are added to the cache.
763 */
764 list_for_each_entry_safe(e, n, &bitmaps, list) {
765 list_del_init(&e->list);
766 ret = io_ctl_read_bitmap(&io_ctl, e);
767 if (ret)
768 goto free_cache;
769 }
770
771 io_ctl_drop_pages(&io_ctl);
772 merge_space_tree(ctl);
773 ret = 1;
774 out:
775 io_ctl_free(&io_ctl);
776 return ret;
777 free_cache:
778 io_ctl_drop_pages(&io_ctl);
779 __btrfs_remove_free_space_cache(ctl);
780 goto out;
781 }
782
783 int load_free_space_cache(struct btrfs_fs_info *fs_info,
784 struct btrfs_block_group_cache *block_group)
785 {
786 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
787 struct btrfs_root *root = fs_info->tree_root;
788 struct inode *inode;
789 struct btrfs_path *path;
790 int ret = 0;
791 bool matched;
792 u64 used = btrfs_block_group_used(&block_group->item);
793
794 /*
795 * If this block group has been marked to be cleared for one reason or
796 * another then we can't trust the on disk cache, so just return.
797 */
798 spin_lock(&block_group->lock);
799 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
800 spin_unlock(&block_group->lock);
801 return 0;
802 }
803 spin_unlock(&block_group->lock);
804
805 path = btrfs_alloc_path();
806 if (!path)
807 return 0;
808 path->search_commit_root = 1;
809 path->skip_locking = 1;
810
811 inode = lookup_free_space_inode(root, block_group, path);
812 if (IS_ERR(inode)) {
813 btrfs_free_path(path);
814 return 0;
815 }
816
817 /* We may have converted the inode and made the cache invalid. */
818 spin_lock(&block_group->lock);
819 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
820 spin_unlock(&block_group->lock);
821 btrfs_free_path(path);
822 goto out;
823 }
824 spin_unlock(&block_group->lock);
825
826 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
827 path, block_group->key.objectid);
828 btrfs_free_path(path);
829 if (ret <= 0)
830 goto out;
831
832 spin_lock(&ctl->tree_lock);
833 matched = (ctl->free_space == (block_group->key.offset - used -
834 block_group->bytes_super));
835 spin_unlock(&ctl->tree_lock);
836
837 if (!matched) {
838 __btrfs_remove_free_space_cache(ctl);
839 btrfs_err(fs_info, "block group %llu has wrong amount of free space",
840 block_group->key.objectid);
841 ret = -1;
842 }
843 out:
844 if (ret < 0) {
845 /* This cache is bogus, make sure it gets cleared */
846 spin_lock(&block_group->lock);
847 block_group->disk_cache_state = BTRFS_DC_CLEAR;
848 spin_unlock(&block_group->lock);
849 ret = 0;
850
851 btrfs_err(fs_info, "failed to load free space cache for block group %llu",
852 block_group->key.objectid);
853 }
854
855 iput(inode);
856 return ret;
857 }
858
859 /**
860 * __btrfs_write_out_cache - write out cached info to an inode
861 * @root - the root the inode belongs to
862 * @ctl - the free space cache we are going to write out
863 * @block_group - the block_group for this cache if it belongs to a block_group
864 * @trans - the trans handle
865 * @path - the path to use
866 * @offset - the offset for the key we'll insert
867 *
868 * This function writes out a free space cache struct to disk for quick recovery
869 * on mount. This will return 0 if it was successfull in writing the cache out,
870 * and -1 if it was not.
871 */
872 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
873 struct btrfs_free_space_ctl *ctl,
874 struct btrfs_block_group_cache *block_group,
875 struct btrfs_trans_handle *trans,
876 struct btrfs_path *path, u64 offset)
877 {
878 struct btrfs_free_space_header *header;
879 struct extent_buffer *leaf;
880 struct rb_node *node;
881 struct list_head *pos, *n;
882 struct extent_state *cached_state = NULL;
883 struct btrfs_free_cluster *cluster = NULL;
884 struct extent_io_tree *unpin = NULL;
885 struct io_ctl io_ctl;
886 struct list_head bitmap_list;
887 struct btrfs_key key;
888 u64 start, extent_start, extent_end, len;
889 int entries = 0;
890 int bitmaps = 0;
891 int ret;
892 int err = -1;
893
894 INIT_LIST_HEAD(&bitmap_list);
895
896 if (!i_size_read(inode))
897 return -1;
898
899 ret = io_ctl_init(&io_ctl, inode, root);
900 if (ret)
901 return -1;
902
903 /* Get the cluster for this block_group if it exists */
904 if (block_group && !list_empty(&block_group->cluster_list))
905 cluster = list_entry(block_group->cluster_list.next,
906 struct btrfs_free_cluster,
907 block_group_list);
908
909 /* Lock all pages first so we can lock the extent safely. */
910 io_ctl_prepare_pages(&io_ctl, inode, 0);
911
912 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
913 0, &cached_state);
914
915 node = rb_first(&ctl->free_space_offset);
916 if (!node && cluster) {
917 node = rb_first(&cluster->root);
918 cluster = NULL;
919 }
920
921 /* Make sure we can fit our crcs into the first page */
922 if (io_ctl.check_crcs &&
923 (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE)
924 goto out_nospc;
925
926 io_ctl_set_generation(&io_ctl, trans->transid);
927
928 /* Write out the extent entries */
929 while (node) {
930 struct btrfs_free_space *e;
931
932 e = rb_entry(node, struct btrfs_free_space, offset_index);
933 entries++;
934
935 ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes,
936 e->bitmap);
937 if (ret)
938 goto out_nospc;
939
940 if (e->bitmap) {
941 list_add_tail(&e->list, &bitmap_list);
942 bitmaps++;
943 }
944 node = rb_next(node);
945 if (!node && cluster) {
946 node = rb_first(&cluster->root);
947 cluster = NULL;
948 }
949 }
950
951 /*
952 * We want to add any pinned extents to our free space cache
953 * so we don't leak the space
954 */
955
956 /*
957 * We shouldn't have switched the pinned extents yet so this is the
958 * right one
959 */
960 unpin = root->fs_info->pinned_extents;
961
962 if (block_group)
963 start = block_group->key.objectid;
964
965 while (block_group && (start < block_group->key.objectid +
966 block_group->key.offset)) {
967 ret = find_first_extent_bit(unpin, start,
968 &extent_start, &extent_end,
969 EXTENT_DIRTY, NULL);
970 if (ret) {
971 ret = 0;
972 break;
973 }
974
975 /* This pinned extent is out of our range */
976 if (extent_start >= block_group->key.objectid +
977 block_group->key.offset)
978 break;
979
980 extent_start = max(extent_start, start);
981 extent_end = min(block_group->key.objectid +
982 block_group->key.offset, extent_end + 1);
983 len = extent_end - extent_start;
984
985 entries++;
986 ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL);
987 if (ret)
988 goto out_nospc;
989
990 start = extent_end;
991 }
992
993 /* Write out the bitmaps */
994 list_for_each_safe(pos, n, &bitmap_list) {
995 struct btrfs_free_space *entry =
996 list_entry(pos, struct btrfs_free_space, list);
997
998 ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap);
999 if (ret)
1000 goto out_nospc;
1001 list_del_init(&entry->list);
1002 }
1003
1004 /* Zero out the rest of the pages just to make sure */
1005 io_ctl_zero_remaining_pages(&io_ctl);
1006
1007 ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
1008 0, i_size_read(inode), &cached_state);
1009 io_ctl_drop_pages(&io_ctl);
1010 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1011 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1012
1013 if (ret)
1014 goto out;
1015
1016
1017 btrfs_wait_ordered_range(inode, 0, (u64)-1);
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 BUG_ON(!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 BUG_ON(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 BUG_ON(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 BUG_ON(!info->bitmap && !info->bytes);
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 BUG_ON(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 BUG_ON(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 BUG_ON(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 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1439 struct btrfs_free_space *bitmap_info, u64 *offset,
1440 u64 *bytes)
1441 {
1442 unsigned long found_bits = 0;
1443 unsigned long bits, i;
1444 unsigned long next_zero;
1445
1446 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1447 max_t(u64, *offset, bitmap_info->offset));
1448 bits = bytes_to_bits(*bytes, ctl->unit);
1449
1450 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1451 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1452 BITS_PER_BITMAP, i);
1453 if ((next_zero - i) >= bits) {
1454 found_bits = next_zero - i;
1455 break;
1456 }
1457 i = next_zero;
1458 }
1459
1460 if (found_bits) {
1461 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1462 *bytes = (u64)(found_bits) * ctl->unit;
1463 return 0;
1464 }
1465
1466 return -1;
1467 }
1468
1469 static struct btrfs_free_space *
1470 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1471 unsigned long align)
1472 {
1473 struct btrfs_free_space *entry;
1474 struct rb_node *node;
1475 u64 ctl_off;
1476 u64 tmp;
1477 u64 align_off;
1478 int ret;
1479
1480 if (!ctl->free_space_offset.rb_node)
1481 return NULL;
1482
1483 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1484 if (!entry)
1485 return NULL;
1486
1487 for (node = &entry->offset_index; node; node = rb_next(node)) {
1488 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1489 if (entry->bytes < *bytes)
1490 continue;
1491
1492 /* make sure the space returned is big enough
1493 * to match our requested alignment
1494 */
1495 if (*bytes >= align) {
1496 ctl_off = entry->offset - ctl->start;
1497 tmp = ctl_off + align - 1;;
1498 do_div(tmp, align);
1499 tmp = tmp * align + ctl->start;
1500 align_off = tmp - entry->offset;
1501 } else {
1502 align_off = 0;
1503 tmp = entry->offset;
1504 }
1505
1506 if (entry->bytes < *bytes + align_off)
1507 continue;
1508
1509 if (entry->bitmap) {
1510 ret = search_bitmap(ctl, entry, &tmp, bytes);
1511 if (!ret) {
1512 *offset = tmp;
1513 return entry;
1514 }
1515 continue;
1516 }
1517
1518 *offset = tmp;
1519 *bytes = entry->bytes - align_off;
1520 return entry;
1521 }
1522
1523 return NULL;
1524 }
1525
1526 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1527 struct btrfs_free_space *info, u64 offset)
1528 {
1529 info->offset = offset_to_bitmap(ctl, offset);
1530 info->bytes = 0;
1531 INIT_LIST_HEAD(&info->list);
1532 link_free_space(ctl, info);
1533 ctl->total_bitmaps++;
1534
1535 ctl->op->recalc_thresholds(ctl);
1536 }
1537
1538 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1539 struct btrfs_free_space *bitmap_info)
1540 {
1541 unlink_free_space(ctl, bitmap_info);
1542 kfree(bitmap_info->bitmap);
1543 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1544 ctl->total_bitmaps--;
1545 ctl->op->recalc_thresholds(ctl);
1546 }
1547
1548 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1549 struct btrfs_free_space *bitmap_info,
1550 u64 *offset, u64 *bytes)
1551 {
1552 u64 end;
1553 u64 search_start, search_bytes;
1554 int ret;
1555
1556 again:
1557 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1558
1559 /*
1560 * We need to search for bits in this bitmap. We could only cover some
1561 * of the extent in this bitmap thanks to how we add space, so we need
1562 * to search for as much as it as we can and clear that amount, and then
1563 * go searching for the next bit.
1564 */
1565 search_start = *offset;
1566 search_bytes = ctl->unit;
1567 search_bytes = min(search_bytes, end - search_start + 1);
1568 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1569 if (ret < 0 || search_start != *offset)
1570 return -EINVAL;
1571
1572 /* We may have found more bits than what we need */
1573 search_bytes = min(search_bytes, *bytes);
1574
1575 /* Cannot clear past the end of the bitmap */
1576 search_bytes = min(search_bytes, end - search_start + 1);
1577
1578 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1579 *offset += search_bytes;
1580 *bytes -= search_bytes;
1581
1582 if (*bytes) {
1583 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1584 if (!bitmap_info->bytes)
1585 free_bitmap(ctl, bitmap_info);
1586
1587 /*
1588 * no entry after this bitmap, but we still have bytes to
1589 * remove, so something has gone wrong.
1590 */
1591 if (!next)
1592 return -EINVAL;
1593
1594 bitmap_info = rb_entry(next, struct btrfs_free_space,
1595 offset_index);
1596
1597 /*
1598 * if the next entry isn't a bitmap we need to return to let the
1599 * extent stuff do its work.
1600 */
1601 if (!bitmap_info->bitmap)
1602 return -EAGAIN;
1603
1604 /*
1605 * Ok the next item is a bitmap, but it may not actually hold
1606 * the information for the rest of this free space stuff, so
1607 * look for it, and if we don't find it return so we can try
1608 * everything over again.
1609 */
1610 search_start = *offset;
1611 search_bytes = ctl->unit;
1612 ret = search_bitmap(ctl, bitmap_info, &search_start,
1613 &search_bytes);
1614 if (ret < 0 || search_start != *offset)
1615 return -EAGAIN;
1616
1617 goto again;
1618 } else if (!bitmap_info->bytes)
1619 free_bitmap(ctl, bitmap_info);
1620
1621 return 0;
1622 }
1623
1624 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1625 struct btrfs_free_space *info, u64 offset,
1626 u64 bytes)
1627 {
1628 u64 bytes_to_set = 0;
1629 u64 end;
1630
1631 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1632
1633 bytes_to_set = min(end - offset, bytes);
1634
1635 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1636
1637 return bytes_to_set;
1638
1639 }
1640
1641 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1642 struct btrfs_free_space *info)
1643 {
1644 struct btrfs_block_group_cache *block_group = ctl->private;
1645
1646 /*
1647 * If we are below the extents threshold then we can add this as an
1648 * extent, and don't have to deal with the bitmap
1649 */
1650 if (ctl->free_extents < ctl->extents_thresh) {
1651 /*
1652 * If this block group has some small extents we don't want to
1653 * use up all of our free slots in the cache with them, we want
1654 * to reserve them to larger extents, however if we have plent
1655 * of cache left then go ahead an dadd them, no sense in adding
1656 * the overhead of a bitmap if we don't have to.
1657 */
1658 if (info->bytes <= block_group->sectorsize * 4) {
1659 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1660 return false;
1661 } else {
1662 return false;
1663 }
1664 }
1665
1666 /*
1667 * The original block groups from mkfs can be really small, like 8
1668 * megabytes, so don't bother with a bitmap for those entries. However
1669 * some block groups can be smaller than what a bitmap would cover but
1670 * are still large enough that they could overflow the 32k memory limit,
1671 * so allow those block groups to still be allowed to have a bitmap
1672 * entry.
1673 */
1674 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
1675 return false;
1676
1677 return true;
1678 }
1679
1680 static struct btrfs_free_space_op free_space_op = {
1681 .recalc_thresholds = recalculate_thresholds,
1682 .use_bitmap = use_bitmap,
1683 };
1684
1685 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1686 struct btrfs_free_space *info)
1687 {
1688 struct btrfs_free_space *bitmap_info;
1689 struct btrfs_block_group_cache *block_group = NULL;
1690 int added = 0;
1691 u64 bytes, offset, bytes_added;
1692 int ret;
1693
1694 bytes = info->bytes;
1695 offset = info->offset;
1696
1697 if (!ctl->op->use_bitmap(ctl, info))
1698 return 0;
1699
1700 if (ctl->op == &free_space_op)
1701 block_group = ctl->private;
1702 again:
1703 /*
1704 * Since we link bitmaps right into the cluster we need to see if we
1705 * have a cluster here, and if so and it has our bitmap we need to add
1706 * the free space to that bitmap.
1707 */
1708 if (block_group && !list_empty(&block_group->cluster_list)) {
1709 struct btrfs_free_cluster *cluster;
1710 struct rb_node *node;
1711 struct btrfs_free_space *entry;
1712
1713 cluster = list_entry(block_group->cluster_list.next,
1714 struct btrfs_free_cluster,
1715 block_group_list);
1716 spin_lock(&cluster->lock);
1717 node = rb_first(&cluster->root);
1718 if (!node) {
1719 spin_unlock(&cluster->lock);
1720 goto no_cluster_bitmap;
1721 }
1722
1723 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1724 if (!entry->bitmap) {
1725 spin_unlock(&cluster->lock);
1726 goto no_cluster_bitmap;
1727 }
1728
1729 if (entry->offset == offset_to_bitmap(ctl, offset)) {
1730 bytes_added = add_bytes_to_bitmap(ctl, entry,
1731 offset, bytes);
1732 bytes -= bytes_added;
1733 offset += bytes_added;
1734 }
1735 spin_unlock(&cluster->lock);
1736 if (!bytes) {
1737 ret = 1;
1738 goto out;
1739 }
1740 }
1741
1742 no_cluster_bitmap:
1743 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1744 1, 0);
1745 if (!bitmap_info) {
1746 BUG_ON(added);
1747 goto new_bitmap;
1748 }
1749
1750 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
1751 bytes -= bytes_added;
1752 offset += bytes_added;
1753 added = 0;
1754
1755 if (!bytes) {
1756 ret = 1;
1757 goto out;
1758 } else
1759 goto again;
1760
1761 new_bitmap:
1762 if (info && info->bitmap) {
1763 add_new_bitmap(ctl, info, offset);
1764 added = 1;
1765 info = NULL;
1766 goto again;
1767 } else {
1768 spin_unlock(&ctl->tree_lock);
1769
1770 /* no pre-allocated info, allocate a new one */
1771 if (!info) {
1772 info = kmem_cache_zalloc(btrfs_free_space_cachep,
1773 GFP_NOFS);
1774 if (!info) {
1775 spin_lock(&ctl->tree_lock);
1776 ret = -ENOMEM;
1777 goto out;
1778 }
1779 }
1780
1781 /* allocate the bitmap */
1782 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1783 spin_lock(&ctl->tree_lock);
1784 if (!info->bitmap) {
1785 ret = -ENOMEM;
1786 goto out;
1787 }
1788 goto again;
1789 }
1790
1791 out:
1792 if (info) {
1793 if (info->bitmap)
1794 kfree(info->bitmap);
1795 kmem_cache_free(btrfs_free_space_cachep, info);
1796 }
1797
1798 return ret;
1799 }
1800
1801 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1802 struct btrfs_free_space *info, bool update_stat)
1803 {
1804 struct btrfs_free_space *left_info;
1805 struct btrfs_free_space *right_info;
1806 bool merged = false;
1807 u64 offset = info->offset;
1808 u64 bytes = info->bytes;
1809
1810 /*
1811 * first we want to see if there is free space adjacent to the range we
1812 * are adding, if there is remove that struct and add a new one to
1813 * cover the entire range
1814 */
1815 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1816 if (right_info && rb_prev(&right_info->offset_index))
1817 left_info = rb_entry(rb_prev(&right_info->offset_index),
1818 struct btrfs_free_space, offset_index);
1819 else
1820 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
1821
1822 if (right_info && !right_info->bitmap) {
1823 if (update_stat)
1824 unlink_free_space(ctl, right_info);
1825 else
1826 __unlink_free_space(ctl, right_info);
1827 info->bytes += right_info->bytes;
1828 kmem_cache_free(btrfs_free_space_cachep, right_info);
1829 merged = true;
1830 }
1831
1832 if (left_info && !left_info->bitmap &&
1833 left_info->offset + left_info->bytes == offset) {
1834 if (update_stat)
1835 unlink_free_space(ctl, left_info);
1836 else
1837 __unlink_free_space(ctl, left_info);
1838 info->offset = left_info->offset;
1839 info->bytes += left_info->bytes;
1840 kmem_cache_free(btrfs_free_space_cachep, left_info);
1841 merged = true;
1842 }
1843
1844 return merged;
1845 }
1846
1847 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
1848 u64 offset, u64 bytes)
1849 {
1850 struct btrfs_free_space *info;
1851 int ret = 0;
1852
1853 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1854 if (!info)
1855 return -ENOMEM;
1856
1857 info->offset = offset;
1858 info->bytes = bytes;
1859
1860 spin_lock(&ctl->tree_lock);
1861
1862 if (try_merge_free_space(ctl, info, true))
1863 goto link;
1864
1865 /*
1866 * There was no extent directly to the left or right of this new
1867 * extent then we know we're going to have to allocate a new extent, so
1868 * before we do that see if we need to drop this into a bitmap
1869 */
1870 ret = insert_into_bitmap(ctl, info);
1871 if (ret < 0) {
1872 goto out;
1873 } else if (ret) {
1874 ret = 0;
1875 goto out;
1876 }
1877 link:
1878 ret = link_free_space(ctl, info);
1879 if (ret)
1880 kmem_cache_free(btrfs_free_space_cachep, info);
1881 out:
1882 spin_unlock(&ctl->tree_lock);
1883
1884 if (ret) {
1885 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
1886 BUG_ON(ret == -EEXIST);
1887 }
1888
1889 return ret;
1890 }
1891
1892 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1893 u64 offset, u64 bytes)
1894 {
1895 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1896 struct btrfs_free_space *info;
1897 int ret;
1898 bool re_search = false;
1899
1900 spin_lock(&ctl->tree_lock);
1901
1902 again:
1903 ret = 0;
1904 if (!bytes)
1905 goto out_lock;
1906
1907 info = tree_search_offset(ctl, offset, 0, 0);
1908 if (!info) {
1909 /*
1910 * oops didn't find an extent that matched the space we wanted
1911 * to remove, look for a bitmap instead
1912 */
1913 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1914 1, 0);
1915 if (!info) {
1916 /*
1917 * If we found a partial bit of our free space in a
1918 * bitmap but then couldn't find the other part this may
1919 * be a problem, so WARN about it.
1920 */
1921 WARN_ON(re_search);
1922 goto out_lock;
1923 }
1924 }
1925
1926 re_search = false;
1927 if (!info->bitmap) {
1928 unlink_free_space(ctl, info);
1929 if (offset == info->offset) {
1930 u64 to_free = min(bytes, info->bytes);
1931
1932 info->bytes -= to_free;
1933 info->offset += to_free;
1934 if (info->bytes) {
1935 ret = link_free_space(ctl, info);
1936 WARN_ON(ret);
1937 } else {
1938 kmem_cache_free(btrfs_free_space_cachep, info);
1939 }
1940
1941 offset += to_free;
1942 bytes -= to_free;
1943 goto again;
1944 } else {
1945 u64 old_end = info->bytes + info->offset;
1946
1947 info->bytes = offset - info->offset;
1948 ret = link_free_space(ctl, info);
1949 WARN_ON(ret);
1950 if (ret)
1951 goto out_lock;
1952
1953 /* Not enough bytes in this entry to satisfy us */
1954 if (old_end < offset + bytes) {
1955 bytes -= old_end - offset;
1956 offset = old_end;
1957 goto again;
1958 } else if (old_end == offset + bytes) {
1959 /* all done */
1960 goto out_lock;
1961 }
1962 spin_unlock(&ctl->tree_lock);
1963
1964 ret = btrfs_add_free_space(block_group, offset + bytes,
1965 old_end - (offset + bytes));
1966 WARN_ON(ret);
1967 goto out;
1968 }
1969 }
1970
1971 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
1972 if (ret == -EAGAIN) {
1973 re_search = true;
1974 goto again;
1975 }
1976 out_lock:
1977 spin_unlock(&ctl->tree_lock);
1978 out:
1979 return ret;
1980 }
1981
1982 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
1983 u64 bytes)
1984 {
1985 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1986 struct btrfs_free_space *info;
1987 struct rb_node *n;
1988 int count = 0;
1989
1990 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
1991 info = rb_entry(n, struct btrfs_free_space, offset_index);
1992 if (info->bytes >= bytes && !block_group->ro)
1993 count++;
1994 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1995 (unsigned long long)info->offset,
1996 (unsigned long long)info->bytes,
1997 (info->bitmap) ? "yes" : "no");
1998 }
1999 printk(KERN_INFO "block group has cluster?: %s\n",
2000 list_empty(&block_group->cluster_list) ? "no" : "yes");
2001 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
2002 "\n", count);
2003 }
2004
2005 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2006 {
2007 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2008
2009 spin_lock_init(&ctl->tree_lock);
2010 ctl->unit = block_group->sectorsize;
2011 ctl->start = block_group->key.objectid;
2012 ctl->private = block_group;
2013 ctl->op = &free_space_op;
2014
2015 /*
2016 * we only want to have 32k of ram per block group for keeping
2017 * track of free space, and if we pass 1/2 of that we want to
2018 * start converting things over to using bitmaps
2019 */
2020 ctl->extents_thresh = ((1024 * 32) / 2) /
2021 sizeof(struct btrfs_free_space);
2022 }
2023
2024 /*
2025 * for a given cluster, put all of its extents back into the free
2026 * space cache. If the block group passed doesn't match the block group
2027 * pointed to by the cluster, someone else raced in and freed the
2028 * cluster already. In that case, we just return without changing anything
2029 */
2030 static int
2031 __btrfs_return_cluster_to_free_space(
2032 struct btrfs_block_group_cache *block_group,
2033 struct btrfs_free_cluster *cluster)
2034 {
2035 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2036 struct btrfs_free_space *entry;
2037 struct rb_node *node;
2038
2039 spin_lock(&cluster->lock);
2040 if (cluster->block_group != block_group)
2041 goto out;
2042
2043 cluster->block_group = NULL;
2044 cluster->window_start = 0;
2045 list_del_init(&cluster->block_group_list);
2046
2047 node = rb_first(&cluster->root);
2048 while (node) {
2049 bool bitmap;
2050
2051 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2052 node = rb_next(&entry->offset_index);
2053 rb_erase(&entry->offset_index, &cluster->root);
2054
2055 bitmap = (entry->bitmap != NULL);
2056 if (!bitmap)
2057 try_merge_free_space(ctl, entry, false);
2058 tree_insert_offset(&ctl->free_space_offset,
2059 entry->offset, &entry->offset_index, bitmap);
2060 }
2061 cluster->root = RB_ROOT;
2062
2063 out:
2064 spin_unlock(&cluster->lock);
2065 btrfs_put_block_group(block_group);
2066 return 0;
2067 }
2068
2069 static void __btrfs_remove_free_space_cache_locked(
2070 struct btrfs_free_space_ctl *ctl)
2071 {
2072 struct btrfs_free_space *info;
2073 struct rb_node *node;
2074
2075 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2076 info = rb_entry(node, struct btrfs_free_space, offset_index);
2077 if (!info->bitmap) {
2078 unlink_free_space(ctl, info);
2079 kmem_cache_free(btrfs_free_space_cachep, info);
2080 } else {
2081 free_bitmap(ctl, info);
2082 }
2083 if (need_resched()) {
2084 spin_unlock(&ctl->tree_lock);
2085 cond_resched();
2086 spin_lock(&ctl->tree_lock);
2087 }
2088 }
2089 }
2090
2091 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2092 {
2093 spin_lock(&ctl->tree_lock);
2094 __btrfs_remove_free_space_cache_locked(ctl);
2095 spin_unlock(&ctl->tree_lock);
2096 }
2097
2098 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2099 {
2100 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2101 struct btrfs_free_cluster *cluster;
2102 struct list_head *head;
2103
2104 spin_lock(&ctl->tree_lock);
2105 while ((head = block_group->cluster_list.next) !=
2106 &block_group->cluster_list) {
2107 cluster = list_entry(head, struct btrfs_free_cluster,
2108 block_group_list);
2109
2110 WARN_ON(cluster->block_group != block_group);
2111 __btrfs_return_cluster_to_free_space(block_group, cluster);
2112 if (need_resched()) {
2113 spin_unlock(&ctl->tree_lock);
2114 cond_resched();
2115 spin_lock(&ctl->tree_lock);
2116 }
2117 }
2118 __btrfs_remove_free_space_cache_locked(ctl);
2119 spin_unlock(&ctl->tree_lock);
2120
2121 }
2122
2123 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2124 u64 offset, u64 bytes, u64 empty_size)
2125 {
2126 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2127 struct btrfs_free_space *entry = NULL;
2128 u64 bytes_search = bytes + empty_size;
2129 u64 ret = 0;
2130 u64 align_gap = 0;
2131 u64 align_gap_len = 0;
2132
2133 spin_lock(&ctl->tree_lock);
2134 entry = find_free_space(ctl, &offset, &bytes_search,
2135 block_group->full_stripe_len);
2136 if (!entry)
2137 goto out;
2138
2139 ret = offset;
2140 if (entry->bitmap) {
2141 bitmap_clear_bits(ctl, entry, offset, bytes);
2142 if (!entry->bytes)
2143 free_bitmap(ctl, entry);
2144 } else {
2145
2146 unlink_free_space(ctl, entry);
2147 align_gap_len = offset - entry->offset;
2148 align_gap = entry->offset;
2149
2150 entry->offset = offset + bytes;
2151 WARN_ON(entry->bytes < bytes + align_gap_len);
2152
2153 entry->bytes -= bytes + align_gap_len;
2154 if (!entry->bytes)
2155 kmem_cache_free(btrfs_free_space_cachep, entry);
2156 else
2157 link_free_space(ctl, entry);
2158 }
2159
2160 out:
2161 spin_unlock(&ctl->tree_lock);
2162
2163 if (align_gap_len)
2164 __btrfs_add_free_space(ctl, align_gap, align_gap_len);
2165 return ret;
2166 }
2167
2168 /*
2169 * given a cluster, put all of its extents back into the free space
2170 * cache. If a block group is passed, this function will only free
2171 * a cluster that belongs to the passed block group.
2172 *
2173 * Otherwise, it'll get a reference on the block group pointed to by the
2174 * cluster and remove the cluster from it.
2175 */
2176 int btrfs_return_cluster_to_free_space(
2177 struct btrfs_block_group_cache *block_group,
2178 struct btrfs_free_cluster *cluster)
2179 {
2180 struct btrfs_free_space_ctl *ctl;
2181 int ret;
2182
2183 /* first, get a safe pointer to the block group */
2184 spin_lock(&cluster->lock);
2185 if (!block_group) {
2186 block_group = cluster->block_group;
2187 if (!block_group) {
2188 spin_unlock(&cluster->lock);
2189 return 0;
2190 }
2191 } else if (cluster->block_group != block_group) {
2192 /* someone else has already freed it don't redo their work */
2193 spin_unlock(&cluster->lock);
2194 return 0;
2195 }
2196 atomic_inc(&block_group->count);
2197 spin_unlock(&cluster->lock);
2198
2199 ctl = block_group->free_space_ctl;
2200
2201 /* now return any extents the cluster had on it */
2202 spin_lock(&ctl->tree_lock);
2203 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2204 spin_unlock(&ctl->tree_lock);
2205
2206 /* finally drop our ref */
2207 btrfs_put_block_group(block_group);
2208 return ret;
2209 }
2210
2211 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2212 struct btrfs_free_cluster *cluster,
2213 struct btrfs_free_space *entry,
2214 u64 bytes, u64 min_start)
2215 {
2216 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2217 int err;
2218 u64 search_start = cluster->window_start;
2219 u64 search_bytes = bytes;
2220 u64 ret = 0;
2221
2222 search_start = min_start;
2223 search_bytes = bytes;
2224
2225 err = search_bitmap(ctl, entry, &search_start, &search_bytes);
2226 if (err)
2227 return 0;
2228
2229 ret = search_start;
2230 __bitmap_clear_bits(ctl, entry, ret, bytes);
2231
2232 return ret;
2233 }
2234
2235 /*
2236 * given a cluster, try to allocate 'bytes' from it, returns 0
2237 * if it couldn't find anything suitably large, or a logical disk offset
2238 * if things worked out
2239 */
2240 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2241 struct btrfs_free_cluster *cluster, u64 bytes,
2242 u64 min_start)
2243 {
2244 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2245 struct btrfs_free_space *entry = NULL;
2246 struct rb_node *node;
2247 u64 ret = 0;
2248
2249 spin_lock(&cluster->lock);
2250 if (bytes > cluster->max_size)
2251 goto out;
2252
2253 if (cluster->block_group != block_group)
2254 goto out;
2255
2256 node = rb_first(&cluster->root);
2257 if (!node)
2258 goto out;
2259
2260 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2261 while(1) {
2262 if (entry->bytes < bytes ||
2263 (!entry->bitmap && entry->offset < min_start)) {
2264 node = rb_next(&entry->offset_index);
2265 if (!node)
2266 break;
2267 entry = rb_entry(node, struct btrfs_free_space,
2268 offset_index);
2269 continue;
2270 }
2271
2272 if (entry->bitmap) {
2273 ret = btrfs_alloc_from_bitmap(block_group,
2274 cluster, entry, bytes,
2275 cluster->window_start);
2276 if (ret == 0) {
2277 node = rb_next(&entry->offset_index);
2278 if (!node)
2279 break;
2280 entry = rb_entry(node, struct btrfs_free_space,
2281 offset_index);
2282 continue;
2283 }
2284 cluster->window_start += bytes;
2285 } else {
2286 ret = entry->offset;
2287
2288 entry->offset += bytes;
2289 entry->bytes -= bytes;
2290 }
2291
2292 if (entry->bytes == 0)
2293 rb_erase(&entry->offset_index, &cluster->root);
2294 break;
2295 }
2296 out:
2297 spin_unlock(&cluster->lock);
2298
2299 if (!ret)
2300 return 0;
2301
2302 spin_lock(&ctl->tree_lock);
2303
2304 ctl->free_space -= bytes;
2305 if (entry->bytes == 0) {
2306 ctl->free_extents--;
2307 if (entry->bitmap) {
2308 kfree(entry->bitmap);
2309 ctl->total_bitmaps--;
2310 ctl->op->recalc_thresholds(ctl);
2311 }
2312 kmem_cache_free(btrfs_free_space_cachep, entry);
2313 }
2314
2315 spin_unlock(&ctl->tree_lock);
2316
2317 return ret;
2318 }
2319
2320 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2321 struct btrfs_free_space *entry,
2322 struct btrfs_free_cluster *cluster,
2323 u64 offset, u64 bytes,
2324 u64 cont1_bytes, u64 min_bytes)
2325 {
2326 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2327 unsigned long next_zero;
2328 unsigned long i;
2329 unsigned long want_bits;
2330 unsigned long min_bits;
2331 unsigned long found_bits;
2332 unsigned long start = 0;
2333 unsigned long total_found = 0;
2334 int ret;
2335
2336 i = offset_to_bit(entry->offset, ctl->unit,
2337 max_t(u64, offset, entry->offset));
2338 want_bits = bytes_to_bits(bytes, ctl->unit);
2339 min_bits = bytes_to_bits(min_bytes, ctl->unit);
2340
2341 again:
2342 found_bits = 0;
2343 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2344 next_zero = find_next_zero_bit(entry->bitmap,
2345 BITS_PER_BITMAP, i);
2346 if (next_zero - i >= min_bits) {
2347 found_bits = next_zero - i;
2348 break;
2349 }
2350 i = next_zero;
2351 }
2352
2353 if (!found_bits)
2354 return -ENOSPC;
2355
2356 if (!total_found) {
2357 start = i;
2358 cluster->max_size = 0;
2359 }
2360
2361 total_found += found_bits;
2362
2363 if (cluster->max_size < found_bits * ctl->unit)
2364 cluster->max_size = found_bits * ctl->unit;
2365
2366 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2367 i = next_zero + 1;
2368 goto again;
2369 }
2370
2371 cluster->window_start = start * ctl->unit + entry->offset;
2372 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2373 ret = tree_insert_offset(&cluster->root, entry->offset,
2374 &entry->offset_index, 1);
2375 BUG_ON(ret); /* -EEXIST; Logic error */
2376
2377 trace_btrfs_setup_cluster(block_group, cluster,
2378 total_found * ctl->unit, 1);
2379 return 0;
2380 }
2381
2382 /*
2383 * This searches the block group for just extents to fill the cluster with.
2384 * Try to find a cluster with at least bytes total bytes, at least one
2385 * extent of cont1_bytes, and other clusters of at least min_bytes.
2386 */
2387 static noinline int
2388 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2389 struct btrfs_free_cluster *cluster,
2390 struct list_head *bitmaps, u64 offset, u64 bytes,
2391 u64 cont1_bytes, u64 min_bytes)
2392 {
2393 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2394 struct btrfs_free_space *first = NULL;
2395 struct btrfs_free_space *entry = NULL;
2396 struct btrfs_free_space *last;
2397 struct rb_node *node;
2398 u64 window_start;
2399 u64 window_free;
2400 u64 max_extent;
2401 u64 total_size = 0;
2402
2403 entry = tree_search_offset(ctl, offset, 0, 1);
2404 if (!entry)
2405 return -ENOSPC;
2406
2407 /*
2408 * We don't want bitmaps, so just move along until we find a normal
2409 * extent entry.
2410 */
2411 while (entry->bitmap || entry->bytes < min_bytes) {
2412 if (entry->bitmap && list_empty(&entry->list))
2413 list_add_tail(&entry->list, bitmaps);
2414 node = rb_next(&entry->offset_index);
2415 if (!node)
2416 return -ENOSPC;
2417 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2418 }
2419
2420 window_start = entry->offset;
2421 window_free = entry->bytes;
2422 max_extent = entry->bytes;
2423 first = entry;
2424 last = entry;
2425
2426 for (node = rb_next(&entry->offset_index); node;
2427 node = rb_next(&entry->offset_index)) {
2428 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2429
2430 if (entry->bitmap) {
2431 if (list_empty(&entry->list))
2432 list_add_tail(&entry->list, bitmaps);
2433 continue;
2434 }
2435
2436 if (entry->bytes < min_bytes)
2437 continue;
2438
2439 last = entry;
2440 window_free += entry->bytes;
2441 if (entry->bytes > max_extent)
2442 max_extent = entry->bytes;
2443 }
2444
2445 if (window_free < bytes || max_extent < cont1_bytes)
2446 return -ENOSPC;
2447
2448 cluster->window_start = first->offset;
2449
2450 node = &first->offset_index;
2451
2452 /*
2453 * now we've found our entries, pull them out of the free space
2454 * cache and put them into the cluster rbtree
2455 */
2456 do {
2457 int ret;
2458
2459 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2460 node = rb_next(&entry->offset_index);
2461 if (entry->bitmap || entry->bytes < min_bytes)
2462 continue;
2463
2464 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2465 ret = tree_insert_offset(&cluster->root, entry->offset,
2466 &entry->offset_index, 0);
2467 total_size += entry->bytes;
2468 BUG_ON(ret); /* -EEXIST; Logic error */
2469 } while (node && entry != last);
2470
2471 cluster->max_size = max_extent;
2472 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2473 return 0;
2474 }
2475
2476 /*
2477 * This specifically looks for bitmaps that may work in the cluster, we assume
2478 * that we have already failed to find extents that will work.
2479 */
2480 static noinline int
2481 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2482 struct btrfs_free_cluster *cluster,
2483 struct list_head *bitmaps, u64 offset, u64 bytes,
2484 u64 cont1_bytes, u64 min_bytes)
2485 {
2486 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2487 struct btrfs_free_space *entry;
2488 int ret = -ENOSPC;
2489 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2490
2491 if (ctl->total_bitmaps == 0)
2492 return -ENOSPC;
2493
2494 /*
2495 * The bitmap that covers offset won't be in the list unless offset
2496 * is just its start offset.
2497 */
2498 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2499 if (entry->offset != bitmap_offset) {
2500 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2501 if (entry && list_empty(&entry->list))
2502 list_add(&entry->list, bitmaps);
2503 }
2504
2505 list_for_each_entry(entry, bitmaps, list) {
2506 if (entry->bytes < bytes)
2507 continue;
2508 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2509 bytes, cont1_bytes, min_bytes);
2510 if (!ret)
2511 return 0;
2512 }
2513
2514 /*
2515 * The bitmaps list has all the bitmaps that record free space
2516 * starting after offset, so no more search is required.
2517 */
2518 return -ENOSPC;
2519 }
2520
2521 /*
2522 * here we try to find a cluster of blocks in a block group. The goal
2523 * is to find at least bytes+empty_size.
2524 * We might not find them all in one contiguous area.
2525 *
2526 * returns zero and sets up cluster if things worked out, otherwise
2527 * it returns -enospc
2528 */
2529 int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
2530 struct btrfs_root *root,
2531 struct btrfs_block_group_cache *block_group,
2532 struct btrfs_free_cluster *cluster,
2533 u64 offset, u64 bytes, u64 empty_size)
2534 {
2535 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2536 struct btrfs_free_space *entry, *tmp;
2537 LIST_HEAD(bitmaps);
2538 u64 min_bytes;
2539 u64 cont1_bytes;
2540 int ret;
2541
2542 /*
2543 * Choose the minimum extent size we'll require for this
2544 * cluster. For SSD_SPREAD, don't allow any fragmentation.
2545 * For metadata, allow allocates with smaller extents. For
2546 * data, keep it dense.
2547 */
2548 if (btrfs_test_opt(root, SSD_SPREAD)) {
2549 cont1_bytes = min_bytes = bytes + empty_size;
2550 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2551 cont1_bytes = bytes;
2552 min_bytes = block_group->sectorsize;
2553 } else {
2554 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
2555 min_bytes = block_group->sectorsize;
2556 }
2557
2558 spin_lock(&ctl->tree_lock);
2559
2560 /*
2561 * If we know we don't have enough space to make a cluster don't even
2562 * bother doing all the work to try and find one.
2563 */
2564 if (ctl->free_space < bytes) {
2565 spin_unlock(&ctl->tree_lock);
2566 return -ENOSPC;
2567 }
2568
2569 spin_lock(&cluster->lock);
2570
2571 /* someone already found a cluster, hooray */
2572 if (cluster->block_group) {
2573 ret = 0;
2574 goto out;
2575 }
2576
2577 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
2578 min_bytes);
2579
2580 INIT_LIST_HEAD(&bitmaps);
2581 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2582 bytes + empty_size,
2583 cont1_bytes, min_bytes);
2584 if (ret)
2585 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2586 offset, bytes + empty_size,
2587 cont1_bytes, min_bytes);
2588
2589 /* Clear our temporary list */
2590 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
2591 list_del_init(&entry->list);
2592
2593 if (!ret) {
2594 atomic_inc(&block_group->count);
2595 list_add_tail(&cluster->block_group_list,
2596 &block_group->cluster_list);
2597 cluster->block_group = block_group;
2598 } else {
2599 trace_btrfs_failed_cluster_setup(block_group);
2600 }
2601 out:
2602 spin_unlock(&cluster->lock);
2603 spin_unlock(&ctl->tree_lock);
2604
2605 return ret;
2606 }
2607
2608 /*
2609 * simple code to zero out a cluster
2610 */
2611 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2612 {
2613 spin_lock_init(&cluster->lock);
2614 spin_lock_init(&cluster->refill_lock);
2615 cluster->root = RB_ROOT;
2616 cluster->max_size = 0;
2617 INIT_LIST_HEAD(&cluster->block_group_list);
2618 cluster->block_group = NULL;
2619 }
2620
2621 static int do_trimming(struct btrfs_block_group_cache *block_group,
2622 u64 *total_trimmed, u64 start, u64 bytes,
2623 u64 reserved_start, u64 reserved_bytes)
2624 {
2625 struct btrfs_space_info *space_info = block_group->space_info;
2626 struct btrfs_fs_info *fs_info = block_group->fs_info;
2627 int ret;
2628 int update = 0;
2629 u64 trimmed = 0;
2630
2631 spin_lock(&space_info->lock);
2632 spin_lock(&block_group->lock);
2633 if (!block_group->ro) {
2634 block_group->reserved += reserved_bytes;
2635 space_info->bytes_reserved += reserved_bytes;
2636 update = 1;
2637 }
2638 spin_unlock(&block_group->lock);
2639 spin_unlock(&space_info->lock);
2640
2641 ret = btrfs_error_discard_extent(fs_info->extent_root,
2642 start, bytes, &trimmed);
2643 if (!ret)
2644 *total_trimmed += trimmed;
2645
2646 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
2647
2648 if (update) {
2649 spin_lock(&space_info->lock);
2650 spin_lock(&block_group->lock);
2651 if (block_group->ro)
2652 space_info->bytes_readonly += reserved_bytes;
2653 block_group->reserved -= reserved_bytes;
2654 space_info->bytes_reserved -= reserved_bytes;
2655 spin_unlock(&space_info->lock);
2656 spin_unlock(&block_group->lock);
2657 }
2658
2659 return ret;
2660 }
2661
2662 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
2663 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2664 {
2665 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2666 struct btrfs_free_space *entry;
2667 struct rb_node *node;
2668 int ret = 0;
2669 u64 extent_start;
2670 u64 extent_bytes;
2671 u64 bytes;
2672
2673 while (start < end) {
2674 spin_lock(&ctl->tree_lock);
2675
2676 if (ctl->free_space < minlen) {
2677 spin_unlock(&ctl->tree_lock);
2678 break;
2679 }
2680
2681 entry = tree_search_offset(ctl, start, 0, 1);
2682 if (!entry) {
2683 spin_unlock(&ctl->tree_lock);
2684 break;
2685 }
2686
2687 /* skip bitmaps */
2688 while (entry->bitmap) {
2689 node = rb_next(&entry->offset_index);
2690 if (!node) {
2691 spin_unlock(&ctl->tree_lock);
2692 goto out;
2693 }
2694 entry = rb_entry(node, struct btrfs_free_space,
2695 offset_index);
2696 }
2697
2698 if (entry->offset >= end) {
2699 spin_unlock(&ctl->tree_lock);
2700 break;
2701 }
2702
2703 extent_start = entry->offset;
2704 extent_bytes = entry->bytes;
2705 start = max(start, extent_start);
2706 bytes = min(extent_start + extent_bytes, end) - start;
2707 if (bytes < minlen) {
2708 spin_unlock(&ctl->tree_lock);
2709 goto next;
2710 }
2711
2712 unlink_free_space(ctl, entry);
2713 kmem_cache_free(btrfs_free_space_cachep, entry);
2714
2715 spin_unlock(&ctl->tree_lock);
2716
2717 ret = do_trimming(block_group, total_trimmed, start, bytes,
2718 extent_start, extent_bytes);
2719 if (ret)
2720 break;
2721 next:
2722 start += bytes;
2723
2724 if (fatal_signal_pending(current)) {
2725 ret = -ERESTARTSYS;
2726 break;
2727 }
2728
2729 cond_resched();
2730 }
2731 out:
2732 return ret;
2733 }
2734
2735 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
2736 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2737 {
2738 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2739 struct btrfs_free_space *entry;
2740 int ret = 0;
2741 int ret2;
2742 u64 bytes;
2743 u64 offset = offset_to_bitmap(ctl, start);
2744
2745 while (offset < end) {
2746 bool next_bitmap = false;
2747
2748 spin_lock(&ctl->tree_lock);
2749
2750 if (ctl->free_space < minlen) {
2751 spin_unlock(&ctl->tree_lock);
2752 break;
2753 }
2754
2755 entry = tree_search_offset(ctl, offset, 1, 0);
2756 if (!entry) {
2757 spin_unlock(&ctl->tree_lock);
2758 next_bitmap = true;
2759 goto next;
2760 }
2761
2762 bytes = minlen;
2763 ret2 = search_bitmap(ctl, entry, &start, &bytes);
2764 if (ret2 || start >= end) {
2765 spin_unlock(&ctl->tree_lock);
2766 next_bitmap = true;
2767 goto next;
2768 }
2769
2770 bytes = min(bytes, end - start);
2771 if (bytes < minlen) {
2772 spin_unlock(&ctl->tree_lock);
2773 goto next;
2774 }
2775
2776 bitmap_clear_bits(ctl, entry, start, bytes);
2777 if (entry->bytes == 0)
2778 free_bitmap(ctl, entry);
2779
2780 spin_unlock(&ctl->tree_lock);
2781
2782 ret = do_trimming(block_group, total_trimmed, start, bytes,
2783 start, bytes);
2784 if (ret)
2785 break;
2786 next:
2787 if (next_bitmap) {
2788 offset += BITS_PER_BITMAP * ctl->unit;
2789 } else {
2790 start += bytes;
2791 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
2792 offset += BITS_PER_BITMAP * ctl->unit;
2793 }
2794
2795 if (fatal_signal_pending(current)) {
2796 ret = -ERESTARTSYS;
2797 break;
2798 }
2799
2800 cond_resched();
2801 }
2802
2803 return ret;
2804 }
2805
2806 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2807 u64 *trimmed, u64 start, u64 end, u64 minlen)
2808 {
2809 int ret;
2810
2811 *trimmed = 0;
2812
2813 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
2814 if (ret)
2815 return ret;
2816
2817 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
2818
2819 return ret;
2820 }
2821
2822 /*
2823 * Find the left-most item in the cache tree, and then return the
2824 * smallest inode number in the item.
2825 *
2826 * Note: the returned inode number may not be the smallest one in
2827 * the tree, if the left-most item is a bitmap.
2828 */
2829 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
2830 {
2831 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
2832 struct btrfs_free_space *entry = NULL;
2833 u64 ino = 0;
2834
2835 spin_lock(&ctl->tree_lock);
2836
2837 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
2838 goto out;
2839
2840 entry = rb_entry(rb_first(&ctl->free_space_offset),
2841 struct btrfs_free_space, offset_index);
2842
2843 if (!entry->bitmap) {
2844 ino = entry->offset;
2845
2846 unlink_free_space(ctl, entry);
2847 entry->offset++;
2848 entry->bytes--;
2849 if (!entry->bytes)
2850 kmem_cache_free(btrfs_free_space_cachep, entry);
2851 else
2852 link_free_space(ctl, entry);
2853 } else {
2854 u64 offset = 0;
2855 u64 count = 1;
2856 int ret;
2857
2858 ret = search_bitmap(ctl, entry, &offset, &count);
2859 /* Logic error; Should be empty if it can't find anything */
2860 BUG_ON(ret);
2861
2862 ino = offset;
2863 bitmap_clear_bits(ctl, entry, offset, 1);
2864 if (entry->bytes == 0)
2865 free_bitmap(ctl, entry);
2866 }
2867 out:
2868 spin_unlock(&ctl->tree_lock);
2869
2870 return ino;
2871 }
2872
2873 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
2874 struct btrfs_path *path)
2875 {
2876 struct inode *inode = NULL;
2877
2878 spin_lock(&root->cache_lock);
2879 if (root->cache_inode)
2880 inode = igrab(root->cache_inode);
2881 spin_unlock(&root->cache_lock);
2882 if (inode)
2883 return inode;
2884
2885 inode = __lookup_free_space_inode(root, path, 0);
2886 if (IS_ERR(inode))
2887 return inode;
2888
2889 spin_lock(&root->cache_lock);
2890 if (!btrfs_fs_closing(root->fs_info))
2891 root->cache_inode = igrab(inode);
2892 spin_unlock(&root->cache_lock);
2893
2894 return inode;
2895 }
2896
2897 int create_free_ino_inode(struct btrfs_root *root,
2898 struct btrfs_trans_handle *trans,
2899 struct btrfs_path *path)
2900 {
2901 return __create_free_space_inode(root, trans, path,
2902 BTRFS_FREE_INO_OBJECTID, 0);
2903 }
2904
2905 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2906 {
2907 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2908 struct btrfs_path *path;
2909 struct inode *inode;
2910 int ret = 0;
2911 u64 root_gen = btrfs_root_generation(&root->root_item);
2912
2913 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2914 return 0;
2915
2916 /*
2917 * If we're unmounting then just return, since this does a search on the
2918 * normal root and not the commit root and we could deadlock.
2919 */
2920 if (btrfs_fs_closing(fs_info))
2921 return 0;
2922
2923 path = btrfs_alloc_path();
2924 if (!path)
2925 return 0;
2926
2927 inode = lookup_free_ino_inode(root, path);
2928 if (IS_ERR(inode))
2929 goto out;
2930
2931 if (root_gen != BTRFS_I(inode)->generation)
2932 goto out_put;
2933
2934 ret = __load_free_space_cache(root, inode, ctl, path, 0);
2935
2936 if (ret < 0)
2937 btrfs_err(fs_info,
2938 "failed to load free ino cache for root %llu",
2939 root->root_key.objectid);
2940 out_put:
2941 iput(inode);
2942 out:
2943 btrfs_free_path(path);
2944 return ret;
2945 }
2946
2947 int btrfs_write_out_ino_cache(struct btrfs_root *root,
2948 struct btrfs_trans_handle *trans,
2949 struct btrfs_path *path)
2950 {
2951 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2952 struct inode *inode;
2953 int ret;
2954
2955 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2956 return 0;
2957
2958 inode = lookup_free_ino_inode(root, path);
2959 if (IS_ERR(inode))
2960 return 0;
2961
2962 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
2963 if (ret) {
2964 btrfs_delalloc_release_metadata(inode, inode->i_size);
2965 #ifdef DEBUG
2966 btrfs_err(root->fs_info,
2967 "failed to write free ino cache for root %llu",
2968 root->root_key.objectid);
2969 #endif
2970 }
2971
2972 iput(inode);
2973 return ret;
2974 }
2975
2976 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
2977 static struct btrfs_block_group_cache *init_test_block_group(void)
2978 {
2979 struct btrfs_block_group_cache *cache;
2980
2981 cache = kzalloc(sizeof(*cache), GFP_NOFS);
2982 if (!cache)
2983 return NULL;
2984 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
2985 GFP_NOFS);
2986 if (!cache->free_space_ctl) {
2987 kfree(cache);
2988 return NULL;
2989 }
2990
2991 cache->key.objectid = 0;
2992 cache->key.offset = 1024 * 1024 * 1024;
2993 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2994 cache->sectorsize = 4096;
2995
2996 spin_lock_init(&cache->lock);
2997 INIT_LIST_HEAD(&cache->list);
2998 INIT_LIST_HEAD(&cache->cluster_list);
2999 INIT_LIST_HEAD(&cache->new_bg_list);
3000
3001 btrfs_init_free_space_ctl(cache);
3002
3003 return cache;
3004 }
3005
3006 /*
3007 * Checks to see if the given range is in the free space cache. This is really
3008 * just used to check the absence of space, so if there is free space in the
3009 * range at all we will return 1.
3010 */
3011 static int check_exists(struct btrfs_block_group_cache *cache, u64 offset,
3012 u64 bytes)
3013 {
3014 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3015 struct btrfs_free_space *info;
3016 int ret = 0;
3017
3018 spin_lock(&ctl->tree_lock);
3019 info = tree_search_offset(ctl, offset, 0, 0);
3020 if (!info) {
3021 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3022 1, 0);
3023 if (!info)
3024 goto out;
3025 }
3026
3027 have_info:
3028 if (info->bitmap) {
3029 u64 bit_off, bit_bytes;
3030 struct rb_node *n;
3031 struct btrfs_free_space *tmp;
3032
3033 bit_off = offset;
3034 bit_bytes = ctl->unit;
3035 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes);
3036 if (!ret) {
3037 if (bit_off == offset) {
3038 ret = 1;
3039 goto out;
3040 } else if (bit_off > offset &&
3041 offset + bytes > bit_off) {
3042 ret = 1;
3043 goto out;
3044 }
3045 }
3046
3047 n = rb_prev(&info->offset_index);
3048 while (n) {
3049 tmp = rb_entry(n, struct btrfs_free_space,
3050 offset_index);
3051 if (tmp->offset + tmp->bytes < offset)
3052 break;
3053 if (offset + bytes < tmp->offset) {
3054 n = rb_prev(&info->offset_index);
3055 continue;
3056 }
3057 info = tmp;
3058 goto have_info;
3059 }
3060
3061 n = rb_next(&info->offset_index);
3062 while (n) {
3063 tmp = rb_entry(n, struct btrfs_free_space,
3064 offset_index);
3065 if (offset + bytes < tmp->offset)
3066 break;
3067 if (tmp->offset + tmp->bytes < offset) {
3068 n = rb_next(&info->offset_index);
3069 continue;
3070 }
3071 info = tmp;
3072 goto have_info;
3073 }
3074
3075 goto out;
3076 }
3077
3078 if (info->offset == offset) {
3079 ret = 1;
3080 goto out;
3081 }
3082
3083 if (offset > info->offset && offset < info->offset + info->bytes)
3084 ret = 1;
3085 out:
3086 spin_unlock(&ctl->tree_lock);
3087 return ret;
3088 }
3089
3090 /*
3091 * Use this if you need to make a bitmap or extent entry specifically, it
3092 * doesn't do any of the merging that add_free_space does, this acts a lot like
3093 * how the free space cache loading stuff works, so you can get really weird
3094 * configurations.
3095 */
3096 static int add_free_space_entry(struct btrfs_block_group_cache *cache,
3097 u64 offset, u64 bytes, bool bitmap)
3098 {
3099 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3100 struct btrfs_free_space *info = NULL, *bitmap_info;
3101 void *map = NULL;
3102 u64 bytes_added;
3103 int ret;
3104
3105 again:
3106 if (!info) {
3107 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
3108 if (!info)
3109 return -ENOMEM;
3110 }
3111
3112 if (!bitmap) {
3113 spin_lock(&ctl->tree_lock);
3114 info->offset = offset;
3115 info->bytes = bytes;
3116 ret = link_free_space(ctl, info);
3117 spin_unlock(&ctl->tree_lock);
3118 if (ret)
3119 kmem_cache_free(btrfs_free_space_cachep, info);
3120 return ret;
3121 }
3122
3123 if (!map) {
3124 map = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
3125 if (!map) {
3126 kmem_cache_free(btrfs_free_space_cachep, info);
3127 return -ENOMEM;
3128 }
3129 }
3130
3131 spin_lock(&ctl->tree_lock);
3132 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3133 1, 0);
3134 if (!bitmap_info) {
3135 info->bitmap = map;
3136 map = NULL;
3137 add_new_bitmap(ctl, info, offset);
3138 bitmap_info = info;
3139 }
3140
3141 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
3142 bytes -= bytes_added;
3143 offset += bytes_added;
3144 spin_unlock(&ctl->tree_lock);
3145
3146 if (bytes)
3147 goto again;
3148
3149 if (map)
3150 kfree(map);
3151 return 0;
3152 }
3153
3154 /*
3155 * This test just does basic sanity checking, making sure we can add an exten
3156 * entry and remove space from either end and the middle, and make sure we can
3157 * remove space that covers adjacent extent entries.
3158 */
3159 static int test_extents(struct btrfs_block_group_cache *cache)
3160 {
3161 int ret = 0;
3162
3163 printk(KERN_ERR "Running extent only tests\n");
3164
3165 /* First just make sure we can remove an entire entry */
3166 ret = btrfs_add_free_space(cache, 0, 4 * 1024 * 1024);
3167 if (ret) {
3168 printk(KERN_ERR "Error adding initial extents %d\n", ret);
3169 return ret;
3170 }
3171
3172 ret = btrfs_remove_free_space(cache, 0, 4 * 1024 * 1024);
3173 if (ret) {
3174 printk(KERN_ERR "Error removing extent %d\n", ret);
3175 return ret;
3176 }
3177
3178 if (check_exists(cache, 0, 4 * 1024 * 1024)) {
3179 printk(KERN_ERR "Full remove left some lingering space\n");
3180 return -1;
3181 }
3182
3183 /* Ok edge and middle cases now */
3184 ret = btrfs_add_free_space(cache, 0, 4 * 1024 * 1024);
3185 if (ret) {
3186 printk(KERN_ERR "Error adding half extent %d\n", ret);
3187 return ret;
3188 }
3189
3190 ret = btrfs_remove_free_space(cache, 3 * 1024 * 1024, 1 * 1024 * 1024);
3191 if (ret) {
3192 printk(KERN_ERR "Error removing tail end %d\n", ret);
3193 return ret;
3194 }
3195
3196 ret = btrfs_remove_free_space(cache, 0, 1 * 1024 * 1024);
3197 if (ret) {
3198 printk(KERN_ERR "Error removing front end %d\n", ret);
3199 return ret;
3200 }
3201
3202 ret = btrfs_remove_free_space(cache, 2 * 1024 * 1024, 4096);
3203 if (ret) {
3204 printk(KERN_ERR "Error removing middle peice %d\n", ret);
3205 return ret;
3206 }
3207
3208 if (check_exists(cache, 0, 1 * 1024 * 1024)) {
3209 printk(KERN_ERR "Still have space at the front\n");
3210 return -1;
3211 }
3212
3213 if (check_exists(cache, 2 * 1024 * 1024, 4096)) {
3214 printk(KERN_ERR "Still have space in the middle\n");
3215 return -1;
3216 }
3217
3218 if (check_exists(cache, 3 * 1024 * 1024, 1 * 1024 * 1024)) {
3219 printk(KERN_ERR "Still have space at the end\n");
3220 return -1;
3221 }
3222
3223 /* Cleanup */
3224 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3225
3226 return 0;
3227 }
3228
3229 static int test_bitmaps(struct btrfs_block_group_cache *cache)
3230 {
3231 u64 next_bitmap_offset;
3232 int ret;
3233
3234 printk(KERN_ERR "Running bitmap only tests\n");
3235
3236 ret = add_free_space_entry(cache, 0, 4 * 1024 * 1024, 1);
3237 if (ret) {
3238 printk(KERN_ERR "Couldn't create a bitmap entry %d\n", ret);
3239 return ret;
3240 }
3241
3242 ret = btrfs_remove_free_space(cache, 0, 4 * 1024 * 1024);
3243 if (ret) {
3244 printk(KERN_ERR "Error removing bitmap full range %d\n", ret);
3245 return ret;
3246 }
3247
3248 if (check_exists(cache, 0, 4 * 1024 * 1024)) {
3249 printk(KERN_ERR "Left some space in bitmap\n");
3250 return -1;
3251 }
3252
3253 ret = add_free_space_entry(cache, 0, 4 * 1024 * 1024, 1);
3254 if (ret) {
3255 printk(KERN_ERR "Couldn't add to our bitmap entry %d\n", ret);
3256 return ret;
3257 }
3258
3259 ret = btrfs_remove_free_space(cache, 1 * 1024 * 1024, 2 * 1024 * 1024);
3260 if (ret) {
3261 printk(KERN_ERR "Couldn't remove middle chunk %d\n", ret);
3262 return ret;
3263 }
3264
3265 /*
3266 * The first bitmap we have starts at offset 0 so the next one is just
3267 * at the end of the first bitmap.
3268 */
3269 next_bitmap_offset = (u64)(BITS_PER_BITMAP * 4096);
3270
3271 /* Test a bit straddling two bitmaps */
3272 ret = add_free_space_entry(cache, next_bitmap_offset -
3273 (2 * 1024 * 1024), 4 * 1024 * 1024, 1);
3274 if (ret) {
3275 printk(KERN_ERR "Couldn't add space that straddles two bitmaps"
3276 " %d\n", ret);
3277 return ret;
3278 }
3279
3280 ret = btrfs_remove_free_space(cache, next_bitmap_offset -
3281 (1 * 1024 * 1024), 2 * 1024 * 1024);
3282 if (ret) {
3283 printk(KERN_ERR "Couldn't remove overlapping space %d\n", ret);
3284 return ret;
3285 }
3286
3287 if (check_exists(cache, next_bitmap_offset - (1 * 1024 * 1024),
3288 2 * 1024 * 1024)) {
3289 printk(KERN_ERR "Left some space when removing overlapping\n");
3290 return -1;
3291 }
3292
3293 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3294
3295 return 0;
3296 }
3297
3298 /* This is the high grade jackassery */
3299 static int test_bitmaps_and_extents(struct btrfs_block_group_cache *cache)
3300 {
3301 u64 bitmap_offset = (u64)(BITS_PER_BITMAP * 4096);
3302 int ret;
3303
3304 printk(KERN_ERR "Running bitmap and extent tests\n");
3305
3306 /*
3307 * First let's do something simple, an extent at the same offset as the
3308 * bitmap, but the free space completely in the extent and then
3309 * completely in the bitmap.
3310 */
3311 ret = add_free_space_entry(cache, 4 * 1024 * 1024, 1 * 1024 * 1024, 1);
3312 if (ret) {
3313 printk(KERN_ERR "Couldn't create bitmap entry %d\n", ret);
3314 return ret;
3315 }
3316
3317 ret = add_free_space_entry(cache, 0, 1 * 1024 * 1024, 0);
3318 if (ret) {
3319 printk(KERN_ERR "Couldn't add extent entry %d\n", ret);
3320 return ret;
3321 }
3322
3323 ret = btrfs_remove_free_space(cache, 0, 1 * 1024 * 1024);
3324 if (ret) {
3325 printk(KERN_ERR "Couldn't remove extent entry %d\n", ret);
3326 return ret;
3327 }
3328
3329 if (check_exists(cache, 0, 1 * 1024 * 1024)) {
3330 printk(KERN_ERR "Left remnants after our remove\n");
3331 return -1;
3332 }
3333
3334 /* Now to add back the extent entry and remove from the bitmap */
3335 ret = add_free_space_entry(cache, 0, 1 * 1024 * 1024, 0);
3336 if (ret) {
3337 printk(KERN_ERR "Couldn't re-add extent entry %d\n", ret);
3338 return ret;
3339 }
3340
3341 ret = btrfs_remove_free_space(cache, 4 * 1024 * 1024, 1 * 1024 * 1024);
3342 if (ret) {
3343 printk(KERN_ERR "Couldn't remove from bitmap %d\n", ret);
3344 return ret;
3345 }
3346
3347 if (check_exists(cache, 4 * 1024 * 1024, 1 * 1024 * 1024)) {
3348 printk(KERN_ERR "Left remnants in the bitmap\n");
3349 return -1;
3350 }
3351
3352 /*
3353 * Ok so a little more evil, extent entry and bitmap at the same offset,
3354 * removing an overlapping chunk.
3355 */
3356 ret = add_free_space_entry(cache, 1 * 1024 * 1024, 4 * 1024 * 1024, 1);
3357 if (ret) {
3358 printk(KERN_ERR "Couldn't add to a bitmap %d\n", ret);
3359 return ret;
3360 }
3361
3362 ret = btrfs_remove_free_space(cache, 512 * 1024, 3 * 1024 * 1024);
3363 if (ret) {
3364 printk(KERN_ERR "Couldn't remove overlapping space %d\n", ret);
3365 return ret;
3366 }
3367
3368 if (check_exists(cache, 512 * 1024, 3 * 1024 * 1024)) {
3369 printk(KERN_ERR "Left over peices after removing "
3370 "overlapping\n");
3371 return -1;
3372 }
3373
3374 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3375
3376 /* Now with the extent entry offset into the bitmap */
3377 ret = add_free_space_entry(cache, 4 * 1024 * 1024, 4 * 1024 * 1024, 1);
3378 if (ret) {
3379 printk(KERN_ERR "Couldn't add space to the bitmap %d\n", ret);
3380 return ret;
3381 }
3382
3383 ret = add_free_space_entry(cache, 2 * 1024 * 1024, 2 * 1024 * 1024, 0);
3384 if (ret) {
3385 printk(KERN_ERR "Couldn't add extent to the cache %d\n", ret);
3386 return ret;
3387 }
3388
3389 ret = btrfs_remove_free_space(cache, 3 * 1024 * 1024, 4 * 1024 * 1024);
3390 if (ret) {
3391 printk(KERN_ERR "Problem removing overlapping space %d\n", ret);
3392 return ret;
3393 }
3394
3395 if (check_exists(cache, 3 * 1024 * 1024, 4 * 1024 * 1024)) {
3396 printk(KERN_ERR "Left something behind when removing space");
3397 return -1;
3398 }
3399
3400 /*
3401 * This has blown up in the past, the extent entry starts before the
3402 * bitmap entry, but we're trying to remove an offset that falls
3403 * completely within the bitmap range and is in both the extent entry
3404 * and the bitmap entry, looks like this
3405 *
3406 * [ extent ]
3407 * [ bitmap ]
3408 * [ del ]
3409 */
3410 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3411 ret = add_free_space_entry(cache, bitmap_offset + 4 * 1024 * 1024,
3412 4 * 1024 * 1024, 1);
3413 if (ret) {
3414 printk(KERN_ERR "Couldn't add bitmap %d\n", ret);
3415 return ret;
3416 }
3417
3418 ret = add_free_space_entry(cache, bitmap_offset - 1 * 1024 * 1024,
3419 5 * 1024 * 1024, 0);
3420 if (ret) {
3421 printk(KERN_ERR "Couldn't add extent entry %d\n", ret);
3422 return ret;
3423 }
3424
3425 ret = btrfs_remove_free_space(cache, bitmap_offset + 1 * 1024 * 1024,
3426 5 * 1024 * 1024);
3427 if (ret) {
3428 printk(KERN_ERR "Failed to free our space %d\n", ret);
3429 return ret;
3430 }
3431
3432 if (check_exists(cache, bitmap_offset + 1 * 1024 * 1024,
3433 5 * 1024 * 1024)) {
3434 printk(KERN_ERR "Left stuff over\n");
3435 return -1;
3436 }
3437
3438 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3439
3440 /*
3441 * This blew up before, we have part of the free space in a bitmap and
3442 * then the entirety of the rest of the space in an extent. This used
3443 * to return -EAGAIN back from btrfs_remove_extent, make sure this
3444 * doesn't happen.
3445 */
3446 ret = add_free_space_entry(cache, 1 * 1024 * 1024, 2 * 1024 * 1024, 1);
3447 if (ret) {
3448 printk(KERN_ERR "Couldn't add bitmap entry %d\n", ret);
3449 return ret;
3450 }
3451
3452 ret = add_free_space_entry(cache, 3 * 1024 * 1024, 1 * 1024 * 1024, 0);
3453 if (ret) {
3454 printk(KERN_ERR "Couldn't add extent entry %d\n", ret);
3455 return ret;
3456 }
3457
3458 ret = btrfs_remove_free_space(cache, 1 * 1024 * 1024, 3 * 1024 * 1024);
3459 if (ret) {
3460 printk(KERN_ERR "Error removing bitmap and extent "
3461 "overlapping %d\n", ret);
3462 return ret;
3463 }
3464
3465 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3466 return 0;
3467 }
3468
3469 void btrfs_test_free_space_cache(void)
3470 {
3471 struct btrfs_block_group_cache *cache;
3472
3473 printk(KERN_ERR "Running btrfs free space cache tests\n");
3474
3475 cache = init_test_block_group();
3476 if (!cache) {
3477 printk(KERN_ERR "Couldn't run the tests\n");
3478 return;
3479 }
3480
3481 if (test_extents(cache))
3482 goto out;
3483 if (test_bitmaps(cache))
3484 goto out;
3485 if (test_bitmaps_and_extents(cache))
3486 goto out;
3487 out:
3488 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3489 kfree(cache->free_space_ctl);
3490 kfree(cache);
3491 printk(KERN_ERR "Free space cache tests finished\n");
3492 }
3493 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */
Linux kernel - Deliverables
This page took 0.231598 seconds and 6 git commands to generate.