projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: log ram bytes properly
[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 int __create_free_space_inode(struct btrfs_root *root,
124 struct btrfs_trans_handle *trans,
125 struct btrfs_path *path, u64 ino, u64 offset)
126 {
127 struct btrfs_key key;
128 struct btrfs_disk_key disk_key;
129 struct btrfs_free_space_header *header;
130 struct btrfs_inode_item *inode_item;
131 struct extent_buffer *leaf;
132 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
133 int ret;
134
135 ret = btrfs_insert_empty_inode(trans, root, path, ino);
136 if (ret)
137 return ret;
138
139 /* We inline crc's for the free disk space cache */
140 if (ino != BTRFS_FREE_INO_OBJECTID)
141 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
142
143 leaf = path->nodes[0];
144 inode_item = btrfs_item_ptr(leaf, path->slots[0],
145 struct btrfs_inode_item);
146 btrfs_item_key(leaf, &disk_key, path->slots[0]);
147 memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
148 sizeof(*inode_item));
149 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
150 btrfs_set_inode_size(leaf, inode_item, 0);
151 btrfs_set_inode_nbytes(leaf, inode_item, 0);
152 btrfs_set_inode_uid(leaf, inode_item, 0);
153 btrfs_set_inode_gid(leaf, inode_item, 0);
154 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
155 btrfs_set_inode_flags(leaf, inode_item, flags);
156 btrfs_set_inode_nlink(leaf, inode_item, 1);
157 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
158 btrfs_set_inode_block_group(leaf, inode_item, offset);
159 btrfs_mark_buffer_dirty(leaf);
160 btrfs_release_path(path);
161
162 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
163 key.offset = offset;
164 key.type = 0;
165
166 ret = btrfs_insert_empty_item(trans, root, path, &key,
167 sizeof(struct btrfs_free_space_header));
168 if (ret < 0) {
169 btrfs_release_path(path);
170 return ret;
171 }
172 leaf = path->nodes[0];
173 header = btrfs_item_ptr(leaf, path->slots[0],
174 struct btrfs_free_space_header);
175 memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
176 btrfs_set_free_space_key(leaf, header, &disk_key);
177 btrfs_mark_buffer_dirty(leaf);
178 btrfs_release_path(path);
179
180 return 0;
181 }
182
183 int create_free_space_inode(struct btrfs_root *root,
184 struct btrfs_trans_handle *trans,
185 struct btrfs_block_group_cache *block_group,
186 struct btrfs_path *path)
187 {
188 int ret;
189 u64 ino;
190
191 ret = btrfs_find_free_objectid(root, &ino);
192 if (ret < 0)
193 return ret;
194
195 return __create_free_space_inode(root, trans, path, ino,
196 block_group->key.objectid);
197 }
198
199 int btrfs_truncate_free_space_cache(struct btrfs_root *root,
200 struct btrfs_trans_handle *trans,
201 struct btrfs_path *path,
202 struct inode *inode)
203 {
204 struct btrfs_block_rsv *rsv;
205 u64 needed_bytes;
206 loff_t oldsize;
207 int ret = 0;
208
209 rsv = trans->block_rsv;
210 trans->block_rsv = &root->fs_info->global_block_rsv;
211
212 /* 1 for slack space, 1 for updating the inode */
213 needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
214 btrfs_calc_trans_metadata_size(root, 1);
215
216 spin_lock(&trans->block_rsv->lock);
217 if (trans->block_rsv->reserved < needed_bytes) {
218 spin_unlock(&trans->block_rsv->lock);
219 trans->block_rsv = rsv;
220 return -ENOSPC;
221 }
222 spin_unlock(&trans->block_rsv->lock);
223
224 oldsize = i_size_read(inode);
225 btrfs_i_size_write(inode, 0);
226 truncate_pagecache(inode, oldsize, 0);
227
228 /*
229 * We don't need an orphan item because truncating the free space cache
230 * will never be split across transactions.
231 */
232 ret = btrfs_truncate_inode_items(trans, root, inode,
233 0, BTRFS_EXTENT_DATA_KEY);
234
235 if (ret) {
236 trans->block_rsv = rsv;
237 btrfs_abort_transaction(trans, root, ret);
238 return ret;
239 }
240
241 ret = btrfs_update_inode(trans, root, inode);
242 if (ret)
243 btrfs_abort_transaction(trans, root, ret);
244 trans->block_rsv = rsv;
245
246 return ret;
247 }
248
249 static int readahead_cache(struct inode *inode)
250 {
251 struct file_ra_state *ra;
252 unsigned long last_index;
253
254 ra = kzalloc(sizeof(*ra), GFP_NOFS);
255 if (!ra)
256 return -ENOMEM;
257
258 file_ra_state_init(ra, inode->i_mapping);
259 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
260
261 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
262
263 kfree(ra);
264
265 return 0;
266 }
267
268 struct io_ctl {
269 void *cur, *orig;
270 struct page *page;
271 struct page **pages;
272 struct btrfs_root *root;
273 unsigned long size;
274 int index;
275 int num_pages;
276 unsigned check_crcs:1;
277 };
278
279 static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode,
280 struct btrfs_root *root)
281 {
282 memset(io_ctl, 0, sizeof(struct io_ctl));
283 io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
284 PAGE_CACHE_SHIFT;
285 io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages,
286 GFP_NOFS);
287 if (!io_ctl->pages)
288 return -ENOMEM;
289 io_ctl->root = root;
290 if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
291 io_ctl->check_crcs = 1;
292 return 0;
293 }
294
295 static void io_ctl_free(struct io_ctl *io_ctl)
296 {
297 kfree(io_ctl->pages);
298 }
299
300 static void io_ctl_unmap_page(struct io_ctl *io_ctl)
301 {
302 if (io_ctl->cur) {
303 kunmap(io_ctl->page);
304 io_ctl->cur = NULL;
305 io_ctl->orig = NULL;
306 }
307 }
308
309 static void io_ctl_map_page(struct io_ctl *io_ctl, int clear)
310 {
311 BUG_ON(io_ctl->index >= io_ctl->num_pages);
312 io_ctl->page = io_ctl->pages[io_ctl->index++];
313 io_ctl->cur = kmap(io_ctl->page);
314 io_ctl->orig = io_ctl->cur;
315 io_ctl->size = PAGE_CACHE_SIZE;
316 if (clear)
317 memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
318 }
319
320 static void io_ctl_drop_pages(struct io_ctl *io_ctl)
321 {
322 int i;
323
324 io_ctl_unmap_page(io_ctl);
325
326 for (i = 0; i < io_ctl->num_pages; i++) {
327 if (io_ctl->pages[i]) {
328 ClearPageChecked(io_ctl->pages[i]);
329 unlock_page(io_ctl->pages[i]);
330 page_cache_release(io_ctl->pages[i]);
331 }
332 }
333 }
334
335 static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode,
336 int uptodate)
337 {
338 struct page *page;
339 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
340 int i;
341
342 for (i = 0; i < io_ctl->num_pages; i++) {
343 page = find_or_create_page(inode->i_mapping, i, mask);
344 if (!page) {
345 io_ctl_drop_pages(io_ctl);
346 return -ENOMEM;
347 }
348 io_ctl->pages[i] = page;
349 if (uptodate && !PageUptodate(page)) {
350 btrfs_readpage(NULL, page);
351 lock_page(page);
352 if (!PageUptodate(page)) {
353 printk(KERN_ERR "btrfs: error reading free "
354 "space cache\n");
355 io_ctl_drop_pages(io_ctl);
356 return -EIO;
357 }
358 }
359 }
360
361 for (i = 0; i < io_ctl->num_pages; i++) {
362 clear_page_dirty_for_io(io_ctl->pages[i]);
363 set_page_extent_mapped(io_ctl->pages[i]);
364 }
365
366 return 0;
367 }
368
369 static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation)
370 {
371 __le64 *val;
372
373 io_ctl_map_page(io_ctl, 1);
374
375 /*
376 * Skip the csum areas. If we don't check crcs then we just have a
377 * 64bit chunk at the front of the first page.
378 */
379 if (io_ctl->check_crcs) {
380 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
381 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
382 } else {
383 io_ctl->cur += sizeof(u64);
384 io_ctl->size -= sizeof(u64) * 2;
385 }
386
387 val = io_ctl->cur;
388 *val = cpu_to_le64(generation);
389 io_ctl->cur += sizeof(u64);
390 }
391
392 static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation)
393 {
394 __le64 *gen;
395
396 /*
397 * Skip the crc area. If we don't check crcs then we just have a 64bit
398 * chunk at the front of the first page.
399 */
400 if (io_ctl->check_crcs) {
401 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
402 io_ctl->size -= sizeof(u64) +
403 (sizeof(u32) * io_ctl->num_pages);
404 } else {
405 io_ctl->cur += sizeof(u64);
406 io_ctl->size -= sizeof(u64) * 2;
407 }
408
409 gen = io_ctl->cur;
410 if (le64_to_cpu(*gen) != generation) {
411 printk_ratelimited(KERN_ERR "btrfs: space cache generation "
412 "(%Lu) does not match inode (%Lu)\n", *gen,
413 generation);
414 io_ctl_unmap_page(io_ctl);
415 return -EIO;
416 }
417 io_ctl->cur += sizeof(u64);
418 return 0;
419 }
420
421 static void io_ctl_set_crc(struct io_ctl *io_ctl, int index)
422 {
423 u32 *tmp;
424 u32 crc = ~(u32)0;
425 unsigned offset = 0;
426
427 if (!io_ctl->check_crcs) {
428 io_ctl_unmap_page(io_ctl);
429 return;
430 }
431
432 if (index == 0)
433 offset = sizeof(u32) * io_ctl->num_pages;
434
435 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
436 PAGE_CACHE_SIZE - offset);
437 btrfs_csum_final(crc, (char *)&crc);
438 io_ctl_unmap_page(io_ctl);
439 tmp = kmap(io_ctl->pages[0]);
440 tmp += index;
441 *tmp = crc;
442 kunmap(io_ctl->pages[0]);
443 }
444
445 static int io_ctl_check_crc(struct io_ctl *io_ctl, int index)
446 {
447 u32 *tmp, val;
448 u32 crc = ~(u32)0;
449 unsigned offset = 0;
450
451 if (!io_ctl->check_crcs) {
452 io_ctl_map_page(io_ctl, 0);
453 return 0;
454 }
455
456 if (index == 0)
457 offset = sizeof(u32) * io_ctl->num_pages;
458
459 tmp = kmap(io_ctl->pages[0]);
460 tmp += index;
461 val = *tmp;
462 kunmap(io_ctl->pages[0]);
463
464 io_ctl_map_page(io_ctl, 0);
465 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
466 PAGE_CACHE_SIZE - offset);
467 btrfs_csum_final(crc, (char *)&crc);
468 if (val != crc) {
469 printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free "
470 "space cache\n");
471 io_ctl_unmap_page(io_ctl);
472 return -EIO;
473 }
474
475 return 0;
476 }
477
478 static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes,
479 void *bitmap)
480 {
481 struct btrfs_free_space_entry *entry;
482
483 if (!io_ctl->cur)
484 return -ENOSPC;
485
486 entry = io_ctl->cur;
487 entry->offset = cpu_to_le64(offset);
488 entry->bytes = cpu_to_le64(bytes);
489 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
490 BTRFS_FREE_SPACE_EXTENT;
491 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
492 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
493
494 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
495 return 0;
496
497 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
498
499 /* No more pages to map */
500 if (io_ctl->index >= io_ctl->num_pages)
501 return 0;
502
503 /* map the next page */
504 io_ctl_map_page(io_ctl, 1);
505 return 0;
506 }
507
508 static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap)
509 {
510 if (!io_ctl->cur)
511 return -ENOSPC;
512
513 /*
514 * If we aren't at the start of the current page, unmap this one and
515 * map the next one if there is any left.
516 */
517 if (io_ctl->cur != io_ctl->orig) {
518 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
519 if (io_ctl->index >= io_ctl->num_pages)
520 return -ENOSPC;
521 io_ctl_map_page(io_ctl, 0);
522 }
523
524 memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
525 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
526 if (io_ctl->index < io_ctl->num_pages)
527 io_ctl_map_page(io_ctl, 0);
528 return 0;
529 }
530
531 static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl)
532 {
533 /*
534 * If we're not on the boundary we know we've modified the page and we
535 * need to crc the page.
536 */
537 if (io_ctl->cur != io_ctl->orig)
538 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
539 else
540 io_ctl_unmap_page(io_ctl);
541
542 while (io_ctl->index < io_ctl->num_pages) {
543 io_ctl_map_page(io_ctl, 1);
544 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
545 }
546 }
547
548 static int io_ctl_read_entry(struct io_ctl *io_ctl,
549 struct btrfs_free_space *entry, u8 *type)
550 {
551 struct btrfs_free_space_entry *e;
552 int ret;
553
554 if (!io_ctl->cur) {
555 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
556 if (ret)
557 return ret;
558 }
559
560 e = io_ctl->cur;
561 entry->offset = le64_to_cpu(e->offset);
562 entry->bytes = le64_to_cpu(e->bytes);
563 *type = e->type;
564 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
565 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
566
567 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
568 return 0;
569
570 io_ctl_unmap_page(io_ctl);
571
572 return 0;
573 }
574
575 static int io_ctl_read_bitmap(struct io_ctl *io_ctl,
576 struct btrfs_free_space *entry)
577 {
578 int ret;
579
580 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
581 if (ret)
582 return ret;
583
584 memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
585 io_ctl_unmap_page(io_ctl);
586
587 return 0;
588 }
589
590 /*
591 * Since we attach pinned extents after the fact we can have contiguous sections
592 * of free space that are split up in entries. This poses a problem with the
593 * tree logging stuff since it could have allocated across what appears to be 2
594 * entries since we would have merged the entries when adding the pinned extents
595 * back to the free space cache. So run through the space cache that we just
596 * loaded and merge contiguous entries. This will make the log replay stuff not
597 * blow up and it will make for nicer allocator behavior.
598 */
599 static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
600 {
601 struct btrfs_free_space *e, *prev = NULL;
602 struct rb_node *n;
603
604 again:
605 spin_lock(&ctl->tree_lock);
606 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
607 e = rb_entry(n, struct btrfs_free_space, offset_index);
608 if (!prev)
609 goto next;
610 if (e->bitmap || prev->bitmap)
611 goto next;
612 if (prev->offset + prev->bytes == e->offset) {
613 unlink_free_space(ctl, prev);
614 unlink_free_space(ctl, e);
615 prev->bytes += e->bytes;
616 kmem_cache_free(btrfs_free_space_cachep, e);
617 link_free_space(ctl, prev);
618 prev = NULL;
619 spin_unlock(&ctl->tree_lock);
620 goto again;
621 }
622 next:
623 prev = e;
624 }
625 spin_unlock(&ctl->tree_lock);
626 }
627
628 int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
629 struct btrfs_free_space_ctl *ctl,
630 struct btrfs_path *path, u64 offset)
631 {
632 struct btrfs_free_space_header *header;
633 struct extent_buffer *leaf;
634 struct io_ctl io_ctl;
635 struct btrfs_key key;
636 struct btrfs_free_space *e, *n;
637 struct list_head bitmaps;
638 u64 num_entries;
639 u64 num_bitmaps;
640 u64 generation;
641 u8 type;
642 int ret = 0;
643
644 INIT_LIST_HEAD(&bitmaps);
645
646 /* Nothing in the space cache, goodbye */
647 if (!i_size_read(inode))
648 return 0;
649
650 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
651 key.offset = offset;
652 key.type = 0;
653
654 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
655 if (ret < 0)
656 return 0;
657 else if (ret > 0) {
658 btrfs_release_path(path);
659 return 0;
660 }
661
662 ret = -1;
663
664 leaf = path->nodes[0];
665 header = btrfs_item_ptr(leaf, path->slots[0],
666 struct btrfs_free_space_header);
667 num_entries = btrfs_free_space_entries(leaf, header);
668 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
669 generation = btrfs_free_space_generation(leaf, header);
670 btrfs_release_path(path);
671
672 if (BTRFS_I(inode)->generation != generation) {
673 btrfs_err(root->fs_info,
674 "free space inode generation (%llu) "
675 "did not match free space cache generation (%llu)",
676 (unsigned long long)BTRFS_I(inode)->generation,
677 (unsigned long long)generation);
678 return 0;
679 }
680
681 if (!num_entries)
682 return 0;
683
684 ret = io_ctl_init(&io_ctl, inode, root);
685 if (ret)
686 return ret;
687
688 ret = readahead_cache(inode);
689 if (ret)
690 goto out;
691
692 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
693 if (ret)
694 goto out;
695
696 ret = io_ctl_check_crc(&io_ctl, 0);
697 if (ret)
698 goto free_cache;
699
700 ret = io_ctl_check_generation(&io_ctl, generation);
701 if (ret)
702 goto free_cache;
703
704 while (num_entries) {
705 e = kmem_cache_zalloc(btrfs_free_space_cachep,
706 GFP_NOFS);
707 if (!e)
708 goto free_cache;
709
710 ret = io_ctl_read_entry(&io_ctl, e, &type);
711 if (ret) {
712 kmem_cache_free(btrfs_free_space_cachep, e);
713 goto free_cache;
714 }
715
716 if (!e->bytes) {
717 kmem_cache_free(btrfs_free_space_cachep, e);
718 goto free_cache;
719 }
720
721 if (type == BTRFS_FREE_SPACE_EXTENT) {
722 spin_lock(&ctl->tree_lock);
723 ret = link_free_space(ctl, e);
724 spin_unlock(&ctl->tree_lock);
725 if (ret) {
726 btrfs_err(root->fs_info,
727 "Duplicate entries in free space cache, dumping");
728 kmem_cache_free(btrfs_free_space_cachep, e);
729 goto free_cache;
730 }
731 } else {
732 BUG_ON(!num_bitmaps);
733 num_bitmaps--;
734 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
735 if (!e->bitmap) {
736 kmem_cache_free(
737 btrfs_free_space_cachep, e);
738 goto free_cache;
739 }
740 spin_lock(&ctl->tree_lock);
741 ret = link_free_space(ctl, e);
742 ctl->total_bitmaps++;
743 ctl->op->recalc_thresholds(ctl);
744 spin_unlock(&ctl->tree_lock);
745 if (ret) {
746 btrfs_err(root->fs_info,
747 "Duplicate entries in free space cache, dumping");
748 kmem_cache_free(btrfs_free_space_cachep, e);
749 goto free_cache;
750 }
751 list_add_tail(&e->list, &bitmaps);
752 }
753
754 num_entries--;
755 }
756
757 io_ctl_unmap_page(&io_ctl);
758
759 /*
760 * We add the bitmaps at the end of the entries in order that
761 * the bitmap entries are added to the cache.
762 */
763 list_for_each_entry_safe(e, n, &bitmaps, list) {
764 list_del_init(&e->list);
765 ret = io_ctl_read_bitmap(&io_ctl, e);
766 if (ret)
767 goto free_cache;
768 }
769
770 io_ctl_drop_pages(&io_ctl);
771 merge_space_tree(ctl);
772 ret = 1;
773 out:
774 io_ctl_free(&io_ctl);
775 return ret;
776 free_cache:
777 io_ctl_drop_pages(&io_ctl);
778 __btrfs_remove_free_space_cache(ctl);
779 goto out;
780 }
781
782 int load_free_space_cache(struct btrfs_fs_info *fs_info,
783 struct btrfs_block_group_cache *block_group)
784 {
785 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
786 struct btrfs_root *root = fs_info->tree_root;
787 struct inode *inode;
788 struct btrfs_path *path;
789 int ret = 0;
790 bool matched;
791 u64 used = btrfs_block_group_used(&block_group->item);
792
793 /*
794 * If this block group has been marked to be cleared for one reason or
795 * another then we can't trust the on disk cache, so just return.
796 */
797 spin_lock(&block_group->lock);
798 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
799 spin_unlock(&block_group->lock);
800 return 0;
801 }
802 spin_unlock(&block_group->lock);
803
804 path = btrfs_alloc_path();
805 if (!path)
806 return 0;
807 path->search_commit_root = 1;
808 path->skip_locking = 1;
809
810 inode = lookup_free_space_inode(root, block_group, path);
811 if (IS_ERR(inode)) {
812 btrfs_free_path(path);
813 return 0;
814 }
815
816 /* We may have converted the inode and made the cache invalid. */
817 spin_lock(&block_group->lock);
818 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
819 spin_unlock(&block_group->lock);
820 btrfs_free_path(path);
821 goto out;
822 }
823 spin_unlock(&block_group->lock);
824
825 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
826 path, block_group->key.objectid);
827 btrfs_free_path(path);
828 if (ret <= 0)
829 goto out;
830
831 spin_lock(&ctl->tree_lock);
832 matched = (ctl->free_space == (block_group->key.offset - used -
833 block_group->bytes_super));
834 spin_unlock(&ctl->tree_lock);
835
836 if (!matched) {
837 __btrfs_remove_free_space_cache(ctl);
838 btrfs_err(fs_info, "block group %llu has wrong amount of free space",
839 block_group->key.objectid);
840 ret = -1;
841 }
842 out:
843 if (ret < 0) {
844 /* This cache is bogus, make sure it gets cleared */
845 spin_lock(&block_group->lock);
846 block_group->disk_cache_state = BTRFS_DC_CLEAR;
847 spin_unlock(&block_group->lock);
848 ret = 0;
849
850 btrfs_err(fs_info, "failed to load free space cache for block group %llu",
851 block_group->key.objectid);
852 }
853
854 iput(inode);
855 return ret;
856 }
857
858 /**
859 * __btrfs_write_out_cache - write out cached info to an inode
860 * @root - the root the inode belongs to
861 * @ctl - the free space cache we are going to write out
862 * @block_group - the block_group for this cache if it belongs to a block_group
863 * @trans - the trans handle
864 * @path - the path to use
865 * @offset - the offset for the key we'll insert
866 *
867 * This function writes out a free space cache struct to disk for quick recovery
868 * on mount. This will return 0 if it was successfull in writing the cache out,
869 * and -1 if it was not.
870 */
871 int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
872 struct btrfs_free_space_ctl *ctl,
873 struct btrfs_block_group_cache *block_group,
874 struct btrfs_trans_handle *trans,
875 struct btrfs_path *path, u64 offset)
876 {
877 struct btrfs_free_space_header *header;
878 struct extent_buffer *leaf;
879 struct rb_node *node;
880 struct list_head *pos, *n;
881 struct extent_state *cached_state = NULL;
882 struct btrfs_free_cluster *cluster = NULL;
883 struct extent_io_tree *unpin = NULL;
884 struct io_ctl io_ctl;
885 struct list_head bitmap_list;
886 struct btrfs_key key;
887 u64 start, extent_start, extent_end, len;
888 int entries = 0;
889 int bitmaps = 0;
890 int ret;
891 int err = -1;
892
893 INIT_LIST_HEAD(&bitmap_list);
894
895 if (!i_size_read(inode))
896 return -1;
897
898 ret = io_ctl_init(&io_ctl, inode, root);
899 if (ret)
900 return -1;
901
902 /* Get the cluster for this block_group if it exists */
903 if (block_group && !list_empty(&block_group->cluster_list))
904 cluster = list_entry(block_group->cluster_list.next,
905 struct btrfs_free_cluster,
906 block_group_list);
907
908 /* Lock all pages first so we can lock the extent safely. */
909 io_ctl_prepare_pages(&io_ctl, inode, 0);
910
911 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
912 0, &cached_state);
913
914 node = rb_first(&ctl->free_space_offset);
915 if (!node && cluster) {
916 node = rb_first(&cluster->root);
917 cluster = NULL;
918 }
919
920 /* Make sure we can fit our crcs into the first page */
921 if (io_ctl.check_crcs &&
922 (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) {
923 WARN_ON(1);
924 goto out_nospc;
925 }
926
927 io_ctl_set_generation(&io_ctl, trans->transid);
928
929 /* Write out the extent entries */
930 while (node) {
931 struct btrfs_free_space *e;
932
933 e = rb_entry(node, struct btrfs_free_space, offset_index);
934 entries++;
935
936 ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes,
937 e->bitmap);
938 if (ret)
939 goto out_nospc;
940
941 if (e->bitmap) {
942 list_add_tail(&e->list, &bitmap_list);
943 bitmaps++;
944 }
945 node = rb_next(node);
946 if (!node && cluster) {
947 node = rb_first(&cluster->root);
948 cluster = NULL;
949 }
950 }
951
952 /*
953 * We want to add any pinned extents to our free space cache
954 * so we don't leak the space
955 */
956
957 /*
958 * We shouldn't have switched the pinned extents yet so this is the
959 * right one
960 */
961 unpin = root->fs_info->pinned_extents;
962
963 if (block_group)
964 start = block_group->key.objectid;
965
966 while (block_group && (start < block_group->key.objectid +
967 block_group->key.offset)) {
968 ret = find_first_extent_bit(unpin, start,
969 &extent_start, &extent_end,
970 EXTENT_DIRTY, NULL);
971 if (ret) {
972 ret = 0;
973 break;
974 }
975
976 /* This pinned extent is out of our range */
977 if (extent_start >= block_group->key.objectid +
978 block_group->key.offset)
979 break;
980
981 extent_start = max(extent_start, start);
982 extent_end = min(block_group->key.objectid +
983 block_group->key.offset, extent_end + 1);
984 len = extent_end - extent_start;
985
986 entries++;
987 ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL);
988 if (ret)
989 goto out_nospc;
990
991 start = extent_end;
992 }
993
994 /* Write out the bitmaps */
995 list_for_each_safe(pos, n, &bitmap_list) {
996 struct btrfs_free_space *entry =
997 list_entry(pos, struct btrfs_free_space, list);
998
999 ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap);
1000 if (ret)
1001 goto out_nospc;
1002 list_del_init(&entry->list);
1003 }
1004
1005 /* Zero out the rest of the pages just to make sure */
1006 io_ctl_zero_remaining_pages(&io_ctl);
1007
1008 ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
1009 0, i_size_read(inode), &cached_state);
1010 io_ctl_drop_pages(&io_ctl);
1011 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1012 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1013
1014 if (ret)
1015 goto out;
1016
1017
1018 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1019
1020 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1021 key.offset = offset;
1022 key.type = 0;
1023
1024 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1025 if (ret < 0) {
1026 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1027 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1028 GFP_NOFS);
1029 goto out;
1030 }
1031 leaf = path->nodes[0];
1032 if (ret > 0) {
1033 struct btrfs_key found_key;
1034 BUG_ON(!path->slots[0]);
1035 path->slots[0]--;
1036 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1037 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1038 found_key.offset != offset) {
1039 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1040 inode->i_size - 1,
1041 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1042 NULL, GFP_NOFS);
1043 btrfs_release_path(path);
1044 goto out;
1045 }
1046 }
1047
1048 BTRFS_I(inode)->generation = trans->transid;
1049 header = btrfs_item_ptr(leaf, path->slots[0],
1050 struct btrfs_free_space_header);
1051 btrfs_set_free_space_entries(leaf, header, entries);
1052 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1053 btrfs_set_free_space_generation(leaf, header, trans->transid);
1054 btrfs_mark_buffer_dirty(leaf);
1055 btrfs_release_path(path);
1056
1057 err = 0;
1058 out:
1059 io_ctl_free(&io_ctl);
1060 if (err) {
1061 invalidate_inode_pages2(inode->i_mapping);
1062 BTRFS_I(inode)->generation = 0;
1063 }
1064 btrfs_update_inode(trans, root, inode);
1065 return err;
1066
1067 out_nospc:
1068 list_for_each_safe(pos, n, &bitmap_list) {
1069 struct btrfs_free_space *entry =
1070 list_entry(pos, struct btrfs_free_space, list);
1071 list_del_init(&entry->list);
1072 }
1073 io_ctl_drop_pages(&io_ctl);
1074 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1075 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1076 goto out;
1077 }
1078
1079 int btrfs_write_out_cache(struct btrfs_root *root,
1080 struct btrfs_trans_handle *trans,
1081 struct btrfs_block_group_cache *block_group,
1082 struct btrfs_path *path)
1083 {
1084 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1085 struct inode *inode;
1086 int ret = 0;
1087
1088 root = root->fs_info->tree_root;
1089
1090 spin_lock(&block_group->lock);
1091 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1092 spin_unlock(&block_group->lock);
1093 return 0;
1094 }
1095 spin_unlock(&block_group->lock);
1096
1097 inode = lookup_free_space_inode(root, block_group, path);
1098 if (IS_ERR(inode))
1099 return 0;
1100
1101 ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
1102 path, block_group->key.objectid);
1103 if (ret) {
1104 spin_lock(&block_group->lock);
1105 block_group->disk_cache_state = BTRFS_DC_ERROR;
1106 spin_unlock(&block_group->lock);
1107 ret = 0;
1108 #ifdef DEBUG
1109 btrfs_err(root->fs_info,
1110 "failed to write free space cache for block group %llu",
1111 block_group->key.objectid);
1112 #endif
1113 }
1114
1115 iput(inode);
1116 return ret;
1117 }
1118
1119 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1120 u64 offset)
1121 {
1122 BUG_ON(offset < bitmap_start);
1123 offset -= bitmap_start;
1124 return (unsigned long)(div_u64(offset, unit));
1125 }
1126
1127 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1128 {
1129 return (unsigned long)(div_u64(bytes, unit));
1130 }
1131
1132 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1133 u64 offset)
1134 {
1135 u64 bitmap_start;
1136 u64 bytes_per_bitmap;
1137
1138 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1139 bitmap_start = offset - ctl->start;
1140 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1141 bitmap_start *= bytes_per_bitmap;
1142 bitmap_start += ctl->start;
1143
1144 return bitmap_start;
1145 }
1146
1147 static int tree_insert_offset(struct rb_root *root, u64 offset,
1148 struct rb_node *node, int bitmap)
1149 {
1150 struct rb_node **p = &root->rb_node;
1151 struct rb_node *parent = NULL;
1152 struct btrfs_free_space *info;
1153
1154 while (*p) {
1155 parent = *p;
1156 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1157
1158 if (offset < info->offset) {
1159 p = &(*p)->rb_left;
1160 } else if (offset > info->offset) {
1161 p = &(*p)->rb_right;
1162 } else {
1163 /*
1164 * we could have a bitmap entry and an extent entry
1165 * share the same offset. If this is the case, we want
1166 * the extent entry to always be found first if we do a
1167 * linear search through the tree, since we want to have
1168 * the quickest allocation time, and allocating from an
1169 * extent is faster than allocating from a bitmap. So
1170 * if we're inserting a bitmap and we find an entry at
1171 * this offset, we want to go right, or after this entry
1172 * logically. If we are inserting an extent and we've
1173 * found a bitmap, we want to go left, or before
1174 * logically.
1175 */
1176 if (bitmap) {
1177 if (info->bitmap) {
1178 WARN_ON_ONCE(1);
1179 return -EEXIST;
1180 }
1181 p = &(*p)->rb_right;
1182 } else {
1183 if (!info->bitmap) {
1184 WARN_ON_ONCE(1);
1185 return -EEXIST;
1186 }
1187 p = &(*p)->rb_left;
1188 }
1189 }
1190 }
1191
1192 rb_link_node(node, parent, p);
1193 rb_insert_color(node, root);
1194
1195 return 0;
1196 }
1197
1198 /*
1199 * searches the tree for the given offset.
1200 *
1201 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1202 * want a section that has at least bytes size and comes at or after the given
1203 * offset.
1204 */
1205 static struct btrfs_free_space *
1206 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1207 u64 offset, int bitmap_only, int fuzzy)
1208 {
1209 struct rb_node *n = ctl->free_space_offset.rb_node;
1210 struct btrfs_free_space *entry, *prev = NULL;
1211
1212 /* find entry that is closest to the 'offset' */
1213 while (1) {
1214 if (!n) {
1215 entry = NULL;
1216 break;
1217 }
1218
1219 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1220 prev = entry;
1221
1222 if (offset < entry->offset)
1223 n = n->rb_left;
1224 else if (offset > entry->offset)
1225 n = n->rb_right;
1226 else
1227 break;
1228 }
1229
1230 if (bitmap_only) {
1231 if (!entry)
1232 return NULL;
1233 if (entry->bitmap)
1234 return entry;
1235
1236 /*
1237 * bitmap entry and extent entry may share same offset,
1238 * in that case, bitmap entry comes after extent entry.
1239 */
1240 n = rb_next(n);
1241 if (!n)
1242 return NULL;
1243 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1244 if (entry->offset != offset)
1245 return NULL;
1246
1247 WARN_ON(!entry->bitmap);
1248 return entry;
1249 } else if (entry) {
1250 if (entry->bitmap) {
1251 /*
1252 * if previous extent entry covers the offset,
1253 * we should return it instead of the bitmap entry
1254 */
1255 n = rb_prev(&entry->offset_index);
1256 if (n) {
1257 prev = rb_entry(n, struct btrfs_free_space,
1258 offset_index);
1259 if (!prev->bitmap &&
1260 prev->offset + prev->bytes > offset)
1261 entry = prev;
1262 }
1263 }
1264 return entry;
1265 }
1266
1267 if (!prev)
1268 return NULL;
1269
1270 /* find last entry before the 'offset' */
1271 entry = prev;
1272 if (entry->offset > offset) {
1273 n = rb_prev(&entry->offset_index);
1274 if (n) {
1275 entry = rb_entry(n, struct btrfs_free_space,
1276 offset_index);
1277 BUG_ON(entry->offset > offset);
1278 } else {
1279 if (fuzzy)
1280 return entry;
1281 else
1282 return NULL;
1283 }
1284 }
1285
1286 if (entry->bitmap) {
1287 n = rb_prev(&entry->offset_index);
1288 if (n) {
1289 prev = rb_entry(n, struct btrfs_free_space,
1290 offset_index);
1291 if (!prev->bitmap &&
1292 prev->offset + prev->bytes > offset)
1293 return prev;
1294 }
1295 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1296 return entry;
1297 } else if (entry->offset + entry->bytes > offset)
1298 return entry;
1299
1300 if (!fuzzy)
1301 return NULL;
1302
1303 while (1) {
1304 if (entry->bitmap) {
1305 if (entry->offset + BITS_PER_BITMAP *
1306 ctl->unit > offset)
1307 break;
1308 } else {
1309 if (entry->offset + entry->bytes > offset)
1310 break;
1311 }
1312
1313 n = rb_next(&entry->offset_index);
1314 if (!n)
1315 return NULL;
1316 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1317 }
1318 return entry;
1319 }
1320
1321 static inline void
1322 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1323 struct btrfs_free_space *info)
1324 {
1325 rb_erase(&info->offset_index, &ctl->free_space_offset);
1326 ctl->free_extents--;
1327 }
1328
1329 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1330 struct btrfs_free_space *info)
1331 {
1332 __unlink_free_space(ctl, info);
1333 ctl->free_space -= info->bytes;
1334 }
1335
1336 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1337 struct btrfs_free_space *info)
1338 {
1339 int ret = 0;
1340
1341 BUG_ON(!info->bitmap && !info->bytes);
1342 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1343 &info->offset_index, (info->bitmap != NULL));
1344 if (ret)
1345 return ret;
1346
1347 ctl->free_space += info->bytes;
1348 ctl->free_extents++;
1349 return ret;
1350 }
1351
1352 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1353 {
1354 struct btrfs_block_group_cache *block_group = ctl->private;
1355 u64 max_bytes;
1356 u64 bitmap_bytes;
1357 u64 extent_bytes;
1358 u64 size = block_group->key.offset;
1359 u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1360 int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1361
1362 max_bitmaps = max(max_bitmaps, 1);
1363
1364 BUG_ON(ctl->total_bitmaps > max_bitmaps);
1365
1366 /*
1367 * The goal is to keep the total amount of memory used per 1gb of space
1368 * at or below 32k, so we need to adjust how much memory we allow to be
1369 * used by extent based free space tracking
1370 */
1371 if (size < 1024 * 1024 * 1024)
1372 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1373 else
1374 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1375 div64_u64(size, 1024 * 1024 * 1024);
1376
1377 /*
1378 * we want to account for 1 more bitmap than what we have so we can make
1379 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1380 * we add more bitmaps.
1381 */
1382 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1383
1384 if (bitmap_bytes >= max_bytes) {
1385 ctl->extents_thresh = 0;
1386 return;
1387 }
1388
1389 /*
1390 * we want the extent entry threshold to always be at most 1/2 the maxw
1391 * bytes we can have, or whatever is less than that.
1392 */
1393 extent_bytes = max_bytes - bitmap_bytes;
1394 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1395
1396 ctl->extents_thresh =
1397 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1398 }
1399
1400 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1401 struct btrfs_free_space *info,
1402 u64 offset, u64 bytes)
1403 {
1404 unsigned long start, count;
1405
1406 start = offset_to_bit(info->offset, ctl->unit, offset);
1407 count = bytes_to_bits(bytes, ctl->unit);
1408 BUG_ON(start + count > BITS_PER_BITMAP);
1409
1410 bitmap_clear(info->bitmap, start, count);
1411
1412 info->bytes -= bytes;
1413 }
1414
1415 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1416 struct btrfs_free_space *info, u64 offset,
1417 u64 bytes)
1418 {
1419 __bitmap_clear_bits(ctl, info, offset, bytes);
1420 ctl->free_space -= bytes;
1421 }
1422
1423 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1424 struct btrfs_free_space *info, u64 offset,
1425 u64 bytes)
1426 {
1427 unsigned long start, count;
1428
1429 start = offset_to_bit(info->offset, ctl->unit, offset);
1430 count = bytes_to_bits(bytes, ctl->unit);
1431 BUG_ON(start + count > BITS_PER_BITMAP);
1432
1433 bitmap_set(info->bitmap, start, count);
1434
1435 info->bytes += bytes;
1436 ctl->free_space += bytes;
1437 }
1438
1439 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1440 struct btrfs_free_space *bitmap_info, u64 *offset,
1441 u64 *bytes)
1442 {
1443 unsigned long found_bits = 0;
1444 unsigned long bits, i;
1445 unsigned long next_zero;
1446
1447 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1448 max_t(u64, *offset, bitmap_info->offset));
1449 bits = bytes_to_bits(*bytes, ctl->unit);
1450
1451 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1452 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1453 BITS_PER_BITMAP, i);
1454 if ((next_zero - i) >= bits) {
1455 found_bits = next_zero - i;
1456 break;
1457 }
1458 i = next_zero;
1459 }
1460
1461 if (found_bits) {
1462 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1463 *bytes = (u64)(found_bits) * ctl->unit;
1464 return 0;
1465 }
1466
1467 return -1;
1468 }
1469
1470 static struct btrfs_free_space *
1471 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1472 unsigned long align)
1473 {
1474 struct btrfs_free_space *entry;
1475 struct rb_node *node;
1476 u64 ctl_off;
1477 u64 tmp;
1478 u64 align_off;
1479 int ret;
1480
1481 if (!ctl->free_space_offset.rb_node)
1482 return NULL;
1483
1484 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1485 if (!entry)
1486 return NULL;
1487
1488 for (node = &entry->offset_index; node; node = rb_next(node)) {
1489 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1490 if (entry->bytes < *bytes)
1491 continue;
1492
1493 /* make sure the space returned is big enough
1494 * to match our requested alignment
1495 */
1496 if (*bytes >= align) {
1497 ctl_off = entry->offset - ctl->start;
1498 tmp = ctl_off + align - 1;;
1499 do_div(tmp, align);
1500 tmp = tmp * align + ctl->start;
1501 align_off = tmp - entry->offset;
1502 } else {
1503 align_off = 0;
1504 tmp = entry->offset;
1505 }
1506
1507 if (entry->bytes < *bytes + align_off)
1508 continue;
1509
1510 if (entry->bitmap) {
1511 ret = search_bitmap(ctl, entry, &tmp, bytes);
1512 if (!ret) {
1513 *offset = tmp;
1514 return entry;
1515 }
1516 continue;
1517 }
1518
1519 *offset = tmp;
1520 *bytes = entry->bytes - align_off;
1521 return entry;
1522 }
1523
1524 return NULL;
1525 }
1526
1527 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1528 struct btrfs_free_space *info, u64 offset)
1529 {
1530 info->offset = offset_to_bitmap(ctl, offset);
1531 info->bytes = 0;
1532 INIT_LIST_HEAD(&info->list);
1533 link_free_space(ctl, info);
1534 ctl->total_bitmaps++;
1535
1536 ctl->op->recalc_thresholds(ctl);
1537 }
1538
1539 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1540 struct btrfs_free_space *bitmap_info)
1541 {
1542 unlink_free_space(ctl, bitmap_info);
1543 kfree(bitmap_info->bitmap);
1544 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1545 ctl->total_bitmaps--;
1546 ctl->op->recalc_thresholds(ctl);
1547 }
1548
1549 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1550 struct btrfs_free_space *bitmap_info,
1551 u64 *offset, u64 *bytes)
1552 {
1553 u64 end;
1554 u64 search_start, search_bytes;
1555 int ret;
1556
1557 again:
1558 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1559
1560 /*
1561 * We need to search for bits in this bitmap. We could only cover some
1562 * of the extent in this bitmap thanks to how we add space, so we need
1563 * to search for as much as it as we can and clear that amount, and then
1564 * go searching for the next bit.
1565 */
1566 search_start = *offset;
1567 search_bytes = ctl->unit;
1568 search_bytes = min(search_bytes, end - search_start + 1);
1569 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1570 BUG_ON(ret < 0 || search_start != *offset);
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 BUG_ON(ret); /* logic error */
1977 out_lock:
1978 spin_unlock(&ctl->tree_lock);
1979 out:
1980 return ret;
1981 }
1982
1983 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
1984 u64 bytes)
1985 {
1986 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1987 struct btrfs_free_space *info;
1988 struct rb_node *n;
1989 int count = 0;
1990
1991 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
1992 info = rb_entry(n, struct btrfs_free_space, offset_index);
1993 if (info->bytes >= bytes && !block_group->ro)
1994 count++;
1995 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1996 (unsigned long long)info->offset,
1997 (unsigned long long)info->bytes,
1998 (info->bitmap) ? "yes" : "no");
1999 }
2000 printk(KERN_INFO "block group has cluster?: %s\n",
2001 list_empty(&block_group->cluster_list) ? "no" : "yes");
2002 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
2003 "\n", count);
2004 }
2005
2006 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2007 {
2008 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2009
2010 spin_lock_init(&ctl->tree_lock);
2011 ctl->unit = block_group->sectorsize;
2012 ctl->start = block_group->key.objectid;
2013 ctl->private = block_group;
2014 ctl->op = &free_space_op;
2015
2016 /*
2017 * we only want to have 32k of ram per block group for keeping
2018 * track of free space, and if we pass 1/2 of that we want to
2019 * start converting things over to using bitmaps
2020 */
2021 ctl->extents_thresh = ((1024 * 32) / 2) /
2022 sizeof(struct btrfs_free_space);
2023 }
2024
2025 /*
2026 * for a given cluster, put all of its extents back into the free
2027 * space cache. If the block group passed doesn't match the block group
2028 * pointed to by the cluster, someone else raced in and freed the
2029 * cluster already. In that case, we just return without changing anything
2030 */
2031 static int
2032 __btrfs_return_cluster_to_free_space(
2033 struct btrfs_block_group_cache *block_group,
2034 struct btrfs_free_cluster *cluster)
2035 {
2036 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2037 struct btrfs_free_space *entry;
2038 struct rb_node *node;
2039
2040 spin_lock(&cluster->lock);
2041 if (cluster->block_group != block_group)
2042 goto out;
2043
2044 cluster->block_group = NULL;
2045 cluster->window_start = 0;
2046 list_del_init(&cluster->block_group_list);
2047
2048 node = rb_first(&cluster->root);
2049 while (node) {
2050 bool bitmap;
2051
2052 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2053 node = rb_next(&entry->offset_index);
2054 rb_erase(&entry->offset_index, &cluster->root);
2055
2056 bitmap = (entry->bitmap != NULL);
2057 if (!bitmap)
2058 try_merge_free_space(ctl, entry, false);
2059 tree_insert_offset(&ctl->free_space_offset,
2060 entry->offset, &entry->offset_index, bitmap);
2061 }
2062 cluster->root = RB_ROOT;
2063
2064 out:
2065 spin_unlock(&cluster->lock);
2066 btrfs_put_block_group(block_group);
2067 return 0;
2068 }
2069
2070 void __btrfs_remove_free_space_cache_locked(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 0.159024 seconds and 5 git commands to generate.