?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
driver core: platform.c: fix checkpatch errors and warnings
[linux-2.6.git] / fs / btrfs / free-space-cache.c
blob1f84fc09c1a8e74c5d4b063160e60bdbded44417
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 printk(KERN_INFO "Old style space inode found, converting.\n");
108 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
109 BTRFS_INODE_NODATACOW;
110 block_group->disk_cache_state = BTRFS_DC_CLEAR;
111 }
112
113 if (!block_group->iref) {
114 block_group->inode = igrab(inode);
115 block_group->iref = 1;
116 }
117 spin_unlock(&block_group->lock);
118
119 return inode;
120 }
121
122 int __create_free_space_inode(struct btrfs_root *root,
123 struct btrfs_trans_handle *trans,
124 struct btrfs_path *path, u64 ino, u64 offset)
125 {
126 struct btrfs_key key;
127 struct btrfs_disk_key disk_key;
128 struct btrfs_free_space_header *header;
129 struct btrfs_inode_item *inode_item;
130 struct extent_buffer *leaf;
131 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
132 int ret;
133
134 ret = btrfs_insert_empty_inode(trans, root, path, ino);
135 if (ret)
136 return ret;
137
138 /* We inline crc's for the free disk space cache */
139 if (ino != BTRFS_FREE_INO_OBJECTID)
140 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
141
142 leaf = path->nodes[0];
143 inode_item = btrfs_item_ptr(leaf, path->slots[0],
144 struct btrfs_inode_item);
145 btrfs_item_key(leaf, &disk_key, path->slots[0]);
146 memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
147 sizeof(*inode_item));
148 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
149 btrfs_set_inode_size(leaf, inode_item, 0);
150 btrfs_set_inode_nbytes(leaf, inode_item, 0);
151 btrfs_set_inode_uid(leaf, inode_item, 0);
152 btrfs_set_inode_gid(leaf, inode_item, 0);
153 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
154 btrfs_set_inode_flags(leaf, inode_item, flags);
155 btrfs_set_inode_nlink(leaf, inode_item, 1);
156 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
157 btrfs_set_inode_block_group(leaf, inode_item, offset);
158 btrfs_mark_buffer_dirty(leaf);
159 btrfs_release_path(path);
160
161 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
162 key.offset = offset;
163 key.type = 0;
164
165 ret = btrfs_insert_empty_item(trans, root, path, &key,
166 sizeof(struct btrfs_free_space_header));
167 if (ret < 0) {
168 btrfs_release_path(path);
169 return ret;
170 }
171 leaf = path->nodes[0];
172 header = btrfs_item_ptr(leaf, path->slots[0],
173 struct btrfs_free_space_header);
174 memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
175 btrfs_set_free_space_key(leaf, header, &disk_key);
176 btrfs_mark_buffer_dirty(leaf);
177 btrfs_release_path(path);
178
179 return 0;
180 }
181
182 int create_free_space_inode(struct btrfs_root *root,
183 struct btrfs_trans_handle *trans,
184 struct btrfs_block_group_cache *block_group,
185 struct btrfs_path *path)
186 {
187 int ret;
188 u64 ino;
189
190 ret = btrfs_find_free_objectid(root, &ino);
191 if (ret < 0)
192 return ret;
193
194 return __create_free_space_inode(root, trans, path, ino,
195 block_group->key.objectid);
196 }
197
198 int btrfs_truncate_free_space_cache(struct btrfs_root *root,
199 struct btrfs_trans_handle *trans,
200 struct btrfs_path *path,
201 struct inode *inode)
202 {
203 struct btrfs_block_rsv *rsv;
204 u64 needed_bytes;
205 loff_t oldsize;
206 int ret = 0;
207
208 rsv = trans->block_rsv;
209 trans->block_rsv = &root->fs_info->global_block_rsv;
210
211 /* 1 for slack space, 1 for updating the inode */
212 needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
213 btrfs_calc_trans_metadata_size(root, 1);
214
215 spin_lock(&trans->block_rsv->lock);
216 if (trans->block_rsv->reserved < needed_bytes) {
217 spin_unlock(&trans->block_rsv->lock);
218 trans->block_rsv = rsv;
219 return -ENOSPC;
220 }
221 spin_unlock(&trans->block_rsv->lock);
222
223 oldsize = i_size_read(inode);
224 btrfs_i_size_write(inode, 0);
225 truncate_pagecache(inode, oldsize, 0);
226
227 /*
228 * We don't need an orphan item because truncating the free space cache
229 * will never be split across transactions.
230 */
231 ret = btrfs_truncate_inode_items(trans, root, inode,
232 0, BTRFS_EXTENT_DATA_KEY);
233
234 if (ret) {
235 trans->block_rsv = rsv;
236 btrfs_abort_transaction(trans, root, ret);
237 return ret;
238 }
239
240 ret = btrfs_update_inode(trans, root, inode);
241 if (ret)
242 btrfs_abort_transaction(trans, root, ret);
243 trans->block_rsv = rsv;
244
245 return ret;
246 }
247
248 static int readahead_cache(struct inode *inode)
249 {
250 struct file_ra_state *ra;
251 unsigned long last_index;
252
253 ra = kzalloc(sizeof(*ra), GFP_NOFS);
254 if (!ra)
255 return -ENOMEM;
256
257 file_ra_state_init(ra, inode->i_mapping);
258 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
259
260 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
261
262 kfree(ra);
263
264 return 0;
265 }
266
267 struct io_ctl {
268 void *cur, *orig;
269 struct page *page;
270 struct page **pages;
271 struct btrfs_root *root;
272 unsigned long size;
273 int index;
274 int num_pages;
275 unsigned check_crcs:1;
276 };
277
278 static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode,
279 struct btrfs_root *root)
280 {
281 memset(io_ctl, 0, sizeof(struct io_ctl));
282 io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
283 PAGE_CACHE_SHIFT;
284 io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages,
285 GFP_NOFS);
286 if (!io_ctl->pages)
287 return -ENOMEM;
288 io_ctl->root = root;
289 if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
290 io_ctl->check_crcs = 1;
291 return 0;
292 }
293
294 static void io_ctl_free(struct io_ctl *io_ctl)
295 {
296 kfree(io_ctl->pages);
297 }
298
299 static void io_ctl_unmap_page(struct io_ctl *io_ctl)
300 {
301 if (io_ctl->cur) {
302 kunmap(io_ctl->page);
303 io_ctl->cur = NULL;
304 io_ctl->orig = NULL;
305 }
306 }
307
308 static void io_ctl_map_page(struct io_ctl *io_ctl, int clear)
309 {
310 BUG_ON(io_ctl->index >= io_ctl->num_pages);
311 io_ctl->page = io_ctl->pages[io_ctl->index++];
312 io_ctl->cur = kmap(io_ctl->page);
313 io_ctl->orig = io_ctl->cur;
314 io_ctl->size = PAGE_CACHE_SIZE;
315 if (clear)
316 memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
317 }
318
319 static void io_ctl_drop_pages(struct io_ctl *io_ctl)
320 {
321 int i;
322
323 io_ctl_unmap_page(io_ctl);
324
325 for (i = 0; i < io_ctl->num_pages; i++) {
326 if (io_ctl->pages[i]) {
327 ClearPageChecked(io_ctl->pages[i]);
328 unlock_page(io_ctl->pages[i]);
329 page_cache_release(io_ctl->pages[i]);
330 }
331 }
332 }
333
334 static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode,
335 int uptodate)
336 {
337 struct page *page;
338 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
339 int i;
340
341 for (i = 0; i < io_ctl->num_pages; i++) {
342 page = find_or_create_page(inode->i_mapping, i, mask);
343 if (!page) {
344 io_ctl_drop_pages(io_ctl);
345 return -ENOMEM;
346 }
347 io_ctl->pages[i] = page;
348 if (uptodate && !PageUptodate(page)) {
349 btrfs_readpage(NULL, page);
350 lock_page(page);
351 if (!PageUptodate(page)) {
352 printk(KERN_ERR "btrfs: error reading free "
353 "space cache\n");
354 io_ctl_drop_pages(io_ctl);
355 return -EIO;
356 }
357 }
358 }
359
360 for (i = 0; i < io_ctl->num_pages; i++) {
361 clear_page_dirty_for_io(io_ctl->pages[i]);
362 set_page_extent_mapped(io_ctl->pages[i]);
363 }
364
365 return 0;
366 }
367
368 static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation)
369 {
370 __le64 *val;
371
372 io_ctl_map_page(io_ctl, 1);
373
374 /*
375 * Skip the csum areas. If we don't check crcs then we just have a
376 * 64bit chunk at the front of the first page.
377 */
378 if (io_ctl->check_crcs) {
379 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
380 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
381 } else {
382 io_ctl->cur += sizeof(u64);
383 io_ctl->size -= sizeof(u64) * 2;
384 }
385
386 val = io_ctl->cur;
387 *val = cpu_to_le64(generation);
388 io_ctl->cur += sizeof(u64);
389 }
390
391 static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation)
392 {
393 __le64 *gen;
394
395 /*
396 * Skip the crc area. If we don't check crcs then we just have a 64bit
397 * chunk at the front of the first page.
398 */
399 if (io_ctl->check_crcs) {
400 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
401 io_ctl->size -= sizeof(u64) +
402 (sizeof(u32) * io_ctl->num_pages);
403 } else {
404 io_ctl->cur += sizeof(u64);
405 io_ctl->size -= sizeof(u64) * 2;
406 }
407
408 gen = io_ctl->cur;
409 if (le64_to_cpu(*gen) != generation) {
410 printk_ratelimited(KERN_ERR "btrfs: space cache generation "
411 "(%Lu) does not match inode (%Lu)\n", *gen,
412 generation);
413 io_ctl_unmap_page(io_ctl);
414 return -EIO;
415 }
416 io_ctl->cur += sizeof(u64);
417 return 0;
418 }
419
420 static void io_ctl_set_crc(struct io_ctl *io_ctl, int index)
421 {
422 u32 *tmp;
423 u32 crc = ~(u32)0;
424 unsigned offset = 0;
425
426 if (!io_ctl->check_crcs) {
427 io_ctl_unmap_page(io_ctl);
428 return;
429 }
430
431 if (index == 0)
432 offset = sizeof(u32) * io_ctl->num_pages;
433
434 crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc,
435 PAGE_CACHE_SIZE - offset);
436 btrfs_csum_final(crc, (char *)&crc);
437 io_ctl_unmap_page(io_ctl);
438 tmp = kmap(io_ctl->pages[0]);
439 tmp += index;
440 *tmp = crc;
441 kunmap(io_ctl->pages[0]);
442 }
443
444 static int io_ctl_check_crc(struct io_ctl *io_ctl, int index)
445 {
446 u32 *tmp, val;
447 u32 crc = ~(u32)0;
448 unsigned offset = 0;
449
450 if (!io_ctl->check_crcs) {
451 io_ctl_map_page(io_ctl, 0);
452 return 0;
453 }
454
455 if (index == 0)
456 offset = sizeof(u32) * io_ctl->num_pages;
457
458 tmp = kmap(io_ctl->pages[0]);
459 tmp += index;
460 val = *tmp;
461 kunmap(io_ctl->pages[0]);
462
463 io_ctl_map_page(io_ctl, 0);
464 crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc,
465 PAGE_CACHE_SIZE - offset);
466 btrfs_csum_final(crc, (char *)&crc);
467 if (val != crc) {
468 printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free "
469 "space cache\n");
470 io_ctl_unmap_page(io_ctl);
471 return -EIO;
472 }
473
474 return 0;
475 }
476
477 static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes,
478 void *bitmap)
479 {
480 struct btrfs_free_space_entry *entry;
481
482 if (!io_ctl->cur)
483 return -ENOSPC;
484
485 entry = io_ctl->cur;
486 entry->offset = cpu_to_le64(offset);
487 entry->bytes = cpu_to_le64(bytes);
488 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
489 BTRFS_FREE_SPACE_EXTENT;
490 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
491 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
492
493 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
494 return 0;
495
496 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
497
498 /* No more pages to map */
499 if (io_ctl->index >= io_ctl->num_pages)
500 return 0;
501
502 /* map the next page */
503 io_ctl_map_page(io_ctl, 1);
504 return 0;
505 }
506
507 static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap)
508 {
509 if (!io_ctl->cur)
510 return -ENOSPC;
511
512 /*
513 * If we aren't at the start of the current page, unmap this one and
514 * map the next one if there is any left.
515 */
516 if (io_ctl->cur != io_ctl->orig) {
517 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
518 if (io_ctl->index >= io_ctl->num_pages)
519 return -ENOSPC;
520 io_ctl_map_page(io_ctl, 0);
521 }
522
523 memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
524 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
525 if (io_ctl->index < io_ctl->num_pages)
526 io_ctl_map_page(io_ctl, 0);
527 return 0;
528 }
529
530 static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl)
531 {
532 /*
533 * If we're not on the boundary we know we've modified the page and we
534 * need to crc the page.
535 */
536 if (io_ctl->cur != io_ctl->orig)
537 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
538 else
539 io_ctl_unmap_page(io_ctl);
540
541 while (io_ctl->index < io_ctl->num_pages) {
542 io_ctl_map_page(io_ctl, 1);
543 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
544 }
545 }
546
547 static int io_ctl_read_entry(struct io_ctl *io_ctl,
548 struct btrfs_free_space *entry, u8 *type)
549 {
550 struct btrfs_free_space_entry *e;
551 int ret;
552
553 if (!io_ctl->cur) {
554 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
555 if (ret)
556 return ret;
557 }
558
559 e = io_ctl->cur;
560 entry->offset = le64_to_cpu(e->offset);
561 entry->bytes = le64_to_cpu(e->bytes);
562 *type = e->type;
563 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
564 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
565
566 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
567 return 0;
568
569 io_ctl_unmap_page(io_ctl);
570
571 return 0;
572 }
573
574 static int io_ctl_read_bitmap(struct io_ctl *io_ctl,
575 struct btrfs_free_space *entry)
576 {
577 int ret;
578
579 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
580 if (ret)
581 return ret;
582
583 memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
584 io_ctl_unmap_page(io_ctl);
585
586 return 0;
587 }
588
589 /*
590 * Since we attach pinned extents after the fact we can have contiguous sections
591 * of free space that are split up in entries. This poses a problem with the
592 * tree logging stuff since it could have allocated across what appears to be 2
593 * entries since we would have merged the entries when adding the pinned extents
594 * back to the free space cache. So run through the space cache that we just
595 * loaded and merge contiguous entries. This will make the log replay stuff not
596 * blow up and it will make for nicer allocator behavior.
597 */
598 static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
599 {
600 struct btrfs_free_space *e, *prev = NULL;
601 struct rb_node *n;
602
603 again:
604 spin_lock(&ctl->tree_lock);
605 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
606 e = rb_entry(n, struct btrfs_free_space, offset_index);
607 if (!prev)
608 goto next;
609 if (e->bitmap || prev->bitmap)
610 goto next;
611 if (prev->offset + prev->bytes == e->offset) {
612 unlink_free_space(ctl, prev);
613 unlink_free_space(ctl, e);
614 prev->bytes += e->bytes;
615 kmem_cache_free(btrfs_free_space_cachep, e);
616 link_free_space(ctl, prev);
617 prev = NULL;
618 spin_unlock(&ctl->tree_lock);
619 goto again;
620 }
621 next:
622 prev = e;
623 }
624 spin_unlock(&ctl->tree_lock);
625 }
626
627 int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
628 struct btrfs_free_space_ctl *ctl,
629 struct btrfs_path *path, u64 offset)
630 {
631 struct btrfs_free_space_header *header;
632 struct extent_buffer *leaf;
633 struct io_ctl io_ctl;
634 struct btrfs_key key;
635 struct btrfs_free_space *e, *n;
636 struct list_head bitmaps;
637 u64 num_entries;
638 u64 num_bitmaps;
639 u64 generation;
640 u8 type;
641 int ret = 0;
642
643 INIT_LIST_HEAD(&bitmaps);
644
645 /* Nothing in the space cache, goodbye */
646 if (!i_size_read(inode))
647 return 0;
648
649 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
650 key.offset = offset;
651 key.type = 0;
652
653 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
654 if (ret < 0)
655 return 0;
656 else if (ret > 0) {
657 btrfs_release_path(path);
658 return 0;
659 }
660
661 ret = -1;
662
663 leaf = path->nodes[0];
664 header = btrfs_item_ptr(leaf, path->slots[0],
665 struct btrfs_free_space_header);
666 num_entries = btrfs_free_space_entries(leaf, header);
667 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
668 generation = btrfs_free_space_generation(leaf, header);
669 btrfs_release_path(path);
670
671 if (BTRFS_I(inode)->generation != generation) {
672 printk(KERN_ERR "btrfs: free space inode generation (%llu) did"
673 " not match free space cache generation (%llu)\n",
674 (unsigned long long)BTRFS_I(inode)->generation,
675 (unsigned long long)generation);
676 return 0;
677 }
678
679 if (!num_entries)
680 return 0;
681
682 ret = io_ctl_init(&io_ctl, inode, root);
683 if (ret)
684 return ret;
685
686 ret = readahead_cache(inode);
687 if (ret)
688 goto out;
689
690 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
691 if (ret)
692 goto out;
693
694 ret = io_ctl_check_crc(&io_ctl, 0);
695 if (ret)
696 goto free_cache;
697
698 ret = io_ctl_check_generation(&io_ctl, generation);
699 if (ret)
700 goto free_cache;
701
702 while (num_entries) {
703 e = kmem_cache_zalloc(btrfs_free_space_cachep,
704 GFP_NOFS);
705 if (!e)
706 goto free_cache;
707
708 ret = io_ctl_read_entry(&io_ctl, e, &type);
709 if (ret) {
710 kmem_cache_free(btrfs_free_space_cachep, e);
711 goto free_cache;
712 }
713
714 if (!e->bytes) {
715 kmem_cache_free(btrfs_free_space_cachep, e);
716 goto free_cache;
717 }
718
719 if (type == BTRFS_FREE_SPACE_EXTENT) {
720 spin_lock(&ctl->tree_lock);
721 ret = link_free_space(ctl, e);
722 spin_unlock(&ctl->tree_lock);
723 if (ret) {
724 printk(KERN_ERR "Duplicate entries in "
725 "free space cache, dumping\n");
726 kmem_cache_free(btrfs_free_space_cachep, e);
727 goto free_cache;
728 }
729 } else {
730 BUG_ON(!num_bitmaps);
731 num_bitmaps--;
732 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
733 if (!e->bitmap) {
734 kmem_cache_free(
735 btrfs_free_space_cachep, e);
736 goto free_cache;
737 }
738 spin_lock(&ctl->tree_lock);
739 ret = link_free_space(ctl, e);
740 ctl->total_bitmaps++;
741 ctl->op->recalc_thresholds(ctl);
742 spin_unlock(&ctl->tree_lock);
743 if (ret) {
744 printk(KERN_ERR "Duplicate entries in "
745 "free space cache, dumping\n");
746 kmem_cache_free(btrfs_free_space_cachep, e);
747 goto free_cache;
748 }
749 list_add_tail(&e->list, &bitmaps);
750 }
751
752 num_entries--;
753 }
754
755 io_ctl_unmap_page(&io_ctl);
756
757 /*
758 * We add the bitmaps at the end of the entries in order that
759 * the bitmap entries are added to the cache.
760 */
761 list_for_each_entry_safe(e, n, &bitmaps, list) {
762 list_del_init(&e->list);
763 ret = io_ctl_read_bitmap(&io_ctl, e);
764 if (ret)
765 goto free_cache;
766 }
767
768 io_ctl_drop_pages(&io_ctl);
769 merge_space_tree(ctl);
770 ret = 1;
771 out:
772 io_ctl_free(&io_ctl);
773 return ret;
774 free_cache:
775 io_ctl_drop_pages(&io_ctl);
776 __btrfs_remove_free_space_cache(ctl);
777 goto out;
778 }
779
780 int load_free_space_cache(struct btrfs_fs_info *fs_info,
781 struct btrfs_block_group_cache *block_group)
782 {
783 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
784 struct btrfs_root *root = fs_info->tree_root;
785 struct inode *inode;
786 struct btrfs_path *path;
787 int ret = 0;
788 bool matched;
789 u64 used = btrfs_block_group_used(&block_group->item);
790
791 /*
792 * If this block group has been marked to be cleared for one reason or
793 * another then we can't trust the on disk cache, so just return.
794 */
795 spin_lock(&block_group->lock);
796 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
797 spin_unlock(&block_group->lock);
798 return 0;
799 }
800 spin_unlock(&block_group->lock);
801
802 path = btrfs_alloc_path();
803 if (!path)
804 return 0;
805 path->search_commit_root = 1;
806 path->skip_locking = 1;
807
808 inode = lookup_free_space_inode(root, block_group, path);
809 if (IS_ERR(inode)) {
810 btrfs_free_path(path);
811 return 0;
812 }
813
814 /* We may have converted the inode and made the cache invalid. */
815 spin_lock(&block_group->lock);
816 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
817 spin_unlock(&block_group->lock);
818 btrfs_free_path(path);
819 goto out;
820 }
821 spin_unlock(&block_group->lock);
822
823 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
824 path, block_group->key.objectid);
825 btrfs_free_path(path);
826 if (ret <= 0)
827 goto out;
828
829 spin_lock(&ctl->tree_lock);
830 matched = (ctl->free_space == (block_group->key.offset - used -
831 block_group->bytes_super));
832 spin_unlock(&ctl->tree_lock);
833
834 if (!matched) {
835 __btrfs_remove_free_space_cache(ctl);
836 printk(KERN_ERR "block group %llu has an wrong amount of free "
837 "space\n", block_group->key.objectid);
838 ret = -1;
839 }
840 out:
841 if (ret < 0) {
842 /* This cache is bogus, make sure it gets cleared */
843 spin_lock(&block_group->lock);
844 block_group->disk_cache_state = BTRFS_DC_CLEAR;
845 spin_unlock(&block_group->lock);
846 ret = 0;
847
848 printk(KERN_ERR "btrfs: failed to load free space cache "
849 "for block group %llu\n", block_group->key.objectid);
850 }
851
852 iput(inode);
853 return ret;
854 }
855
856 /**
857 * __btrfs_write_out_cache - write out cached info to an inode
858 * @root - the root the inode belongs to
859 * @ctl - the free space cache we are going to write out
860 * @block_group - the block_group for this cache if it belongs to a block_group
861 * @trans - the trans handle
862 * @path - the path to use
863 * @offset - the offset for the key we'll insert
864 *
865 * This function writes out a free space cache struct to disk for quick recovery
866 * on mount. This will return 0 if it was successfull in writing the cache out,
867 * and -1 if it was not.
868 */
869 int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
870 struct btrfs_free_space_ctl *ctl,
871 struct btrfs_block_group_cache *block_group,
872 struct btrfs_trans_handle *trans,
873 struct btrfs_path *path, u64 offset)
874 {
875 struct btrfs_free_space_header *header;
876 struct extent_buffer *leaf;
877 struct rb_node *node;
878 struct list_head *pos, *n;
879 struct extent_state *cached_state = NULL;
880 struct btrfs_free_cluster *cluster = NULL;
881 struct extent_io_tree *unpin = NULL;
882 struct io_ctl io_ctl;
883 struct list_head bitmap_list;
884 struct btrfs_key key;
885 u64 start, extent_start, extent_end, len;
886 int entries = 0;
887 int bitmaps = 0;
888 int ret;
889 int err = -1;
890
891 INIT_LIST_HEAD(&bitmap_list);
892
893 if (!i_size_read(inode))
894 return -1;
895
896 ret = io_ctl_init(&io_ctl, inode, root);
897 if (ret)
898 return -1;
899
900 /* Get the cluster for this block_group if it exists */
901 if (block_group && !list_empty(&block_group->cluster_list))
902 cluster = list_entry(block_group->cluster_list.next,
903 struct btrfs_free_cluster,
904 block_group_list);
905
906 /* Lock all pages first so we can lock the extent safely. */
907 io_ctl_prepare_pages(&io_ctl, inode, 0);
908
909 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
910 0, &cached_state);
911
912 node = rb_first(&ctl->free_space_offset);
913 if (!node && cluster) {
914 node = rb_first(&cluster->root);
915 cluster = NULL;
916 }
917
918 /* Make sure we can fit our crcs into the first page */
919 if (io_ctl.check_crcs &&
920 (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) {
921 WARN_ON(1);
922 goto out_nospc;
923 }
924
925 io_ctl_set_generation(&io_ctl, trans->transid);
926
927 /* Write out the extent entries */
928 while (node) {
929 struct btrfs_free_space *e;
930
931 e = rb_entry(node, struct btrfs_free_space, offset_index);
932 entries++;
933
934 ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes,
935 e->bitmap);
936 if (ret)
937 goto out_nospc;
938
939 if (e->bitmap) {
940 list_add_tail(&e->list, &bitmap_list);
941 bitmaps++;
942 }
943 node = rb_next(node);
944 if (!node && cluster) {
945 node = rb_first(&cluster->root);
946 cluster = NULL;
947 }
948 }
949
950 /*
951 * We want to add any pinned extents to our free space cache
952 * so we don't leak the space
953 */
954
955 /*
956 * We shouldn't have switched the pinned extents yet so this is the
957 * right one
958 */
959 unpin = root->fs_info->pinned_extents;
960
961 if (block_group)
962 start = block_group->key.objectid;
963
964 while (block_group && (start < block_group->key.objectid +
965 block_group->key.offset)) {
966 ret = find_first_extent_bit(unpin, start,
967 &extent_start, &extent_end,
968 EXTENT_DIRTY, NULL);
969 if (ret) {
970 ret = 0;
971 break;
972 }
973
974 /* This pinned extent is out of our range */
975 if (extent_start >= block_group->key.objectid +
976 block_group->key.offset)
977 break;
978
979 extent_start = max(extent_start, start);
980 extent_end = min(block_group->key.objectid +
981 block_group->key.offset, extent_end + 1);
982 len = extent_end - extent_start;
983
984 entries++;
985 ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL);
986 if (ret)
987 goto out_nospc;
988
989 start = extent_end;
990 }
991
992 /* Write out the bitmaps */
993 list_for_each_safe(pos, n, &bitmap_list) {
994 struct btrfs_free_space *entry =
995 list_entry(pos, struct btrfs_free_space, list);
996
997 ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap);
998 if (ret)
999 goto out_nospc;
1000 list_del_init(&entry->list);
1001 }
1002
1003 /* Zero out the rest of the pages just to make sure */
1004 io_ctl_zero_remaining_pages(&io_ctl);
1005
1006 ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
1007 0, i_size_read(inode), &cached_state);
1008 io_ctl_drop_pages(&io_ctl);
1009 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1010 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1011
1012 if (ret)
1013 goto out;
1014
1015
1016 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1017
1018 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1019 key.offset = offset;
1020 key.type = 0;
1021
1022 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1023 if (ret < 0) {
1024 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1025 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1026 GFP_NOFS);
1027 goto out;
1028 }
1029 leaf = path->nodes[0];
1030 if (ret > 0) {
1031 struct btrfs_key found_key;
1032 BUG_ON(!path->slots[0]);
1033 path->slots[0]--;
1034 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1035 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1036 found_key.offset != offset) {
1037 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1038 inode->i_size - 1,
1039 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1040 NULL, GFP_NOFS);
1041 btrfs_release_path(path);
1042 goto out;
1043 }
1044 }
1045
1046 BTRFS_I(inode)->generation = trans->transid;
1047 header = btrfs_item_ptr(leaf, path->slots[0],
1048 struct btrfs_free_space_header);
1049 btrfs_set_free_space_entries(leaf, header, entries);
1050 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1051 btrfs_set_free_space_generation(leaf, header, trans->transid);
1052 btrfs_mark_buffer_dirty(leaf);
1053 btrfs_release_path(path);
1054
1055 err = 0;
1056 out:
1057 io_ctl_free(&io_ctl);
1058 if (err) {
1059 invalidate_inode_pages2(inode->i_mapping);
1060 BTRFS_I(inode)->generation = 0;
1061 }
1062 btrfs_update_inode(trans, root, inode);
1063 return err;
1064
1065 out_nospc:
1066 list_for_each_safe(pos, n, &bitmap_list) {
1067 struct btrfs_free_space *entry =
1068 list_entry(pos, struct btrfs_free_space, list);
1069 list_del_init(&entry->list);
1070 }
1071 io_ctl_drop_pages(&io_ctl);
1072 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1073 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1074 goto out;
1075 }
1076
1077 int btrfs_write_out_cache(struct btrfs_root *root,
1078 struct btrfs_trans_handle *trans,
1079 struct btrfs_block_group_cache *block_group,
1080 struct btrfs_path *path)
1081 {
1082 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1083 struct inode *inode;
1084 int ret = 0;
1085
1086 root = root->fs_info->tree_root;
1087
1088 spin_lock(&block_group->lock);
1089 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1090 spin_unlock(&block_group->lock);
1091 return 0;
1092 }
1093 spin_unlock(&block_group->lock);
1094
1095 inode = lookup_free_space_inode(root, block_group, path);
1096 if (IS_ERR(inode))
1097 return 0;
1098
1099 ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
1100 path, block_group->key.objectid);
1101 if (ret) {
1102 spin_lock(&block_group->lock);
1103 block_group->disk_cache_state = BTRFS_DC_ERROR;
1104 spin_unlock(&block_group->lock);
1105 ret = 0;
1106 #ifdef DEBUG
1107 printk(KERN_ERR "btrfs: failed to write free space cache "
1108 "for block group %llu\n", block_group->key.objectid);
1109 #endif
1110 }
1111
1112 iput(inode);
1113 return ret;
1114 }
1115
1116 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1117 u64 offset)
1118 {
1119 BUG_ON(offset < bitmap_start);
1120 offset -= bitmap_start;
1121 return (unsigned long)(div_u64(offset, unit));
1122 }
1123
1124 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1125 {
1126 return (unsigned long)(div_u64(bytes, unit));
1127 }
1128
1129 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1130 u64 offset)
1131 {
1132 u64 bitmap_start;
1133 u64 bytes_per_bitmap;
1134
1135 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1136 bitmap_start = offset - ctl->start;
1137 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1138 bitmap_start *= bytes_per_bitmap;
1139 bitmap_start += ctl->start;
1140
1141 return bitmap_start;
1142 }
1143
1144 static int tree_insert_offset(struct rb_root *root, u64 offset,
1145 struct rb_node *node, int bitmap)
1146 {
1147 struct rb_node **p = &root->rb_node;
1148 struct rb_node *parent = NULL;
1149 struct btrfs_free_space *info;
1150
1151 while (*p) {
1152 parent = *p;
1153 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1154
1155 if (offset < info->offset) {
1156 p = &(*p)->rb_left;
1157 } else if (offset > info->offset) {
1158 p = &(*p)->rb_right;
1159 } else {
1160 /*
1161 * we could have a bitmap entry and an extent entry
1162 * share the same offset. If this is the case, we want
1163 * the extent entry to always be found first if we do a
1164 * linear search through the tree, since we want to have
1165 * the quickest allocation time, and allocating from an
1166 * extent is faster than allocating from a bitmap. So
1167 * if we're inserting a bitmap and we find an entry at
1168 * this offset, we want to go right, or after this entry
1169 * logically. If we are inserting an extent and we've
1170 * found a bitmap, we want to go left, or before
1171 * logically.
1172 */
1173 if (bitmap) {
1174 if (info->bitmap) {
1175 WARN_ON_ONCE(1);
1176 return -EEXIST;
1177 }
1178 p = &(*p)->rb_right;
1179 } else {
1180 if (!info->bitmap) {
1181 WARN_ON_ONCE(1);
1182 return -EEXIST;
1183 }
1184 p = &(*p)->rb_left;
1185 }
1186 }
1187 }
1188
1189 rb_link_node(node, parent, p);
1190 rb_insert_color(node, root);
1191
1192 return 0;
1193 }
1194
1195 /*
1196 * searches the tree for the given offset.
1197 *
1198 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1199 * want a section that has at least bytes size and comes at or after the given
1200 * offset.
1201 */
1202 static struct btrfs_free_space *
1203 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1204 u64 offset, int bitmap_only, int fuzzy)
1205 {
1206 struct rb_node *n = ctl->free_space_offset.rb_node;
1207 struct btrfs_free_space *entry, *prev = NULL;
1208
1209 /* find entry that is closest to the 'offset' */
1210 while (1) {
1211 if (!n) {
1212 entry = NULL;
1213 break;
1214 }
1215
1216 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1217 prev = entry;
1218
1219 if (offset < entry->offset)
1220 n = n->rb_left;
1221 else if (offset > entry->offset)
1222 n = n->rb_right;
1223 else
1224 break;
1225 }
1226
1227 if (bitmap_only) {
1228 if (!entry)
1229 return NULL;
1230 if (entry->bitmap)
1231 return entry;
1232
1233 /*
1234 * bitmap entry and extent entry may share same offset,
1235 * in that case, bitmap entry comes after extent entry.
1236 */
1237 n = rb_next(n);
1238 if (!n)
1239 return NULL;
1240 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1241 if (entry->offset != offset)
1242 return NULL;
1243
1244 WARN_ON(!entry->bitmap);
1245 return entry;
1246 } else if (entry) {
1247 if (entry->bitmap) {
1248 /*
1249 * if previous extent entry covers the offset,
1250 * we should return it instead of the bitmap entry
1251 */
1252 n = rb_prev(&entry->offset_index);
1253 if (n) {
1254 prev = rb_entry(n, struct btrfs_free_space,
1255 offset_index);
1256 if (!prev->bitmap &&
1257 prev->offset + prev->bytes > offset)
1258 entry = prev;
1259 }
1260 }
1261 return entry;
1262 }
1263
1264 if (!prev)
1265 return NULL;
1266
1267 /* find last entry before the 'offset' */
1268 entry = prev;
1269 if (entry->offset > offset) {
1270 n = rb_prev(&entry->offset_index);
1271 if (n) {
1272 entry = rb_entry(n, struct btrfs_free_space,
1273 offset_index);
1274 BUG_ON(entry->offset > offset);
1275 } else {
1276 if (fuzzy)
1277 return entry;
1278 else
1279 return NULL;
1280 }
1281 }
1282
1283 if (entry->bitmap) {
1284 n = rb_prev(&entry->offset_index);
1285 if (n) {
1286 prev = rb_entry(n, struct btrfs_free_space,
1287 offset_index);
1288 if (!prev->bitmap &&
1289 prev->offset + prev->bytes > offset)
1290 return prev;
1291 }
1292 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1293 return entry;
1294 } else if (entry->offset + entry->bytes > offset)
1295 return entry;
1296
1297 if (!fuzzy)
1298 return NULL;
1299
1300 while (1) {
1301 if (entry->bitmap) {
1302 if (entry->offset + BITS_PER_BITMAP *
1303 ctl->unit > offset)
1304 break;
1305 } else {
1306 if (entry->offset + entry->bytes > offset)
1307 break;
1308 }
1309
1310 n = rb_next(&entry->offset_index);
1311 if (!n)
1312 return NULL;
1313 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1314 }
1315 return entry;
1316 }
1317
1318 static inline void
1319 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1320 struct btrfs_free_space *info)
1321 {
1322 rb_erase(&info->offset_index, &ctl->free_space_offset);
1323 ctl->free_extents--;
1324 }
1325
1326 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1327 struct btrfs_free_space *info)
1328 {
1329 __unlink_free_space(ctl, info);
1330 ctl->free_space -= info->bytes;
1331 }
1332
1333 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1334 struct btrfs_free_space *info)
1335 {
1336 int ret = 0;
1337
1338 BUG_ON(!info->bitmap && !info->bytes);
1339 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1340 &info->offset_index, (info->bitmap != NULL));
1341 if (ret)
1342 return ret;
1343
1344 ctl->free_space += info->bytes;
1345 ctl->free_extents++;
1346 return ret;
1347 }
1348
1349 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1350 {
1351 struct btrfs_block_group_cache *block_group = ctl->private;
1352 u64 max_bytes;
1353 u64 bitmap_bytes;
1354 u64 extent_bytes;
1355 u64 size = block_group->key.offset;
1356 u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1357 int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1358
1359 max_bitmaps = max(max_bitmaps, 1);
1360
1361 BUG_ON(ctl->total_bitmaps > max_bitmaps);
1362
1363 /*
1364 * The goal is to keep the total amount of memory used per 1gb of space
1365 * at or below 32k, so we need to adjust how much memory we allow to be
1366 * used by extent based free space tracking
1367 */
1368 if (size < 1024 * 1024 * 1024)
1369 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1370 else
1371 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1372 div64_u64(size, 1024 * 1024 * 1024);
1373
1374 /*
1375 * we want to account for 1 more bitmap than what we have so we can make
1376 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1377 * we add more bitmaps.
1378 */
1379 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1380
1381 if (bitmap_bytes >= max_bytes) {
1382 ctl->extents_thresh = 0;
1383 return;
1384 }
1385
1386 /*
1387 * we want the extent entry threshold to always be at most 1/2 the maxw
1388 * bytes we can have, or whatever is less than that.
1389 */
1390 extent_bytes = max_bytes - bitmap_bytes;
1391 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1392
1393 ctl->extents_thresh =
1394 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1395 }
1396
1397 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1398 struct btrfs_free_space *info,
1399 u64 offset, u64 bytes)
1400 {
1401 unsigned long start, count;
1402
1403 start = offset_to_bit(info->offset, ctl->unit, offset);
1404 count = bytes_to_bits(bytes, ctl->unit);
1405 BUG_ON(start + count > BITS_PER_BITMAP);
1406
1407 bitmap_clear(info->bitmap, start, count);
1408
1409 info->bytes -= bytes;
1410 }
1411
1412 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1413 struct btrfs_free_space *info, u64 offset,
1414 u64 bytes)
1415 {
1416 __bitmap_clear_bits(ctl, info, offset, bytes);
1417 ctl->free_space -= bytes;
1418 }
1419
1420 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1421 struct btrfs_free_space *info, u64 offset,
1422 u64 bytes)
1423 {
1424 unsigned long start, count;
1425
1426 start = offset_to_bit(info->offset, ctl->unit, offset);
1427 count = bytes_to_bits(bytes, ctl->unit);
1428 BUG_ON(start + count > BITS_PER_BITMAP);
1429
1430 bitmap_set(info->bitmap, start, count);
1431
1432 info->bytes += bytes;
1433 ctl->free_space += bytes;
1434 }
1435
1436 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1437 struct btrfs_free_space *bitmap_info, u64 *offset,
1438 u64 *bytes)
1439 {
1440 unsigned long found_bits = 0;
1441 unsigned long bits, i;
1442 unsigned long next_zero;
1443
1444 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1445 max_t(u64, *offset, bitmap_info->offset));
1446 bits = bytes_to_bits(*bytes, ctl->unit);
1447
1448 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1449 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1450 BITS_PER_BITMAP, i);
1451 if ((next_zero - i) >= bits) {
1452 found_bits = next_zero - i;
1453 break;
1454 }
1455 i = next_zero;
1456 }
1457
1458 if (found_bits) {
1459 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1460 *bytes = (u64)(found_bits) * ctl->unit;
1461 return 0;
1462 }
1463
1464 return -1;
1465 }
1466
1467 static struct btrfs_free_space *
1468 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1469 unsigned long align)
1470 {
1471 struct btrfs_free_space *entry;
1472 struct rb_node *node;
1473 u64 ctl_off;
1474 u64 tmp;
1475 u64 align_off;
1476 int ret;
1477
1478 if (!ctl->free_space_offset.rb_node)
1479 return NULL;
1480
1481 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1482 if (!entry)
1483 return NULL;
1484
1485 for (node = &entry->offset_index; node; node = rb_next(node)) {
1486 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1487 if (entry->bytes < *bytes)
1488 continue;
1489
1490 /* make sure the space returned is big enough
1491 * to match our requested alignment
1492 */
1493 if (*bytes >= align) {
1494 ctl_off = entry->offset - ctl->start;
1495 tmp = ctl_off + align - 1;;
1496 do_div(tmp, align);
1497 tmp = tmp * align + ctl->start;
1498 align_off = tmp - entry->offset;
1499 } else {
1500 align_off = 0;
1501 tmp = entry->offset;
1502 }
1503
1504 if (entry->bytes < *bytes + align_off)
1505 continue;
1506
1507 if (entry->bitmap) {
1508 ret = search_bitmap(ctl, entry, &tmp, bytes);
1509 if (!ret) {
1510 *offset = tmp;
1511 return entry;
1512 }
1513 continue;
1514 }
1515
1516 *offset = tmp;
1517 *bytes = entry->bytes - align_off;
1518 return entry;
1519 }
1520
1521 return NULL;
1522 }
1523
1524 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1525 struct btrfs_free_space *info, u64 offset)
1526 {
1527 info->offset = offset_to_bitmap(ctl, offset);
1528 info->bytes = 0;
1529 INIT_LIST_HEAD(&info->list);
1530 link_free_space(ctl, info);
1531 ctl->total_bitmaps++;
1532
1533 ctl->op->recalc_thresholds(ctl);
1534 }
1535
1536 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1537 struct btrfs_free_space *bitmap_info)
1538 {
1539 unlink_free_space(ctl, bitmap_info);
1540 kfree(bitmap_info->bitmap);
1541 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1542 ctl->total_bitmaps--;
1543 ctl->op->recalc_thresholds(ctl);
1544 }
1545
1546 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1547 struct btrfs_free_space *bitmap_info,
1548 u64 *offset, u64 *bytes)
1549 {
1550 u64 end;
1551 u64 search_start, search_bytes;
1552 int ret;
1553
1554 again:
1555 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1556
1557 /*
1558 * We need to search for bits in this bitmap. We could only cover some
1559 * of the extent in this bitmap thanks to how we add space, so we need
1560 * to search for as much as it as we can and clear that amount, and then
1561 * go searching for the next bit.
1562 */
1563 search_start = *offset;
1564 search_bytes = ctl->unit;
1565 search_bytes = min(search_bytes, end - search_start + 1);
1566 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1567 BUG_ON(ret < 0 || search_start != *offset);
1568
1569 /* We may have found more bits than what we need */
1570 search_bytes = min(search_bytes, *bytes);
1571
1572 /* Cannot clear past the end of the bitmap */
1573 search_bytes = min(search_bytes, end - search_start + 1);
1574
1575 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1576 *offset += search_bytes;
1577 *bytes -= search_bytes;
1578
1579 if (*bytes) {
1580 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1581 if (!bitmap_info->bytes)
1582 free_bitmap(ctl, bitmap_info);
1583
1584 /*
1585 * no entry after this bitmap, but we still have bytes to
1586 * remove, so something has gone wrong.
1587 */
1588 if (!next)
1589 return -EINVAL;
1590
1591 bitmap_info = rb_entry(next, struct btrfs_free_space,
1592 offset_index);
1593
1594 /*
1595 * if the next entry isn't a bitmap we need to return to let the
1596 * extent stuff do its work.
1597 */
1598 if (!bitmap_info->bitmap)
1599 return -EAGAIN;
1600
1601 /*
1602 * Ok the next item is a bitmap, but it may not actually hold
1603 * the information for the rest of this free space stuff, so
1604 * look for it, and if we don't find it return so we can try
1605 * everything over again.
1606 */
1607 search_start = *offset;
1608 search_bytes = ctl->unit;
1609 ret = search_bitmap(ctl, bitmap_info, &search_start,
1610 &search_bytes);
1611 if (ret < 0 || search_start != *offset)
1612 return -EAGAIN;
1613
1614 goto again;
1615 } else if (!bitmap_info->bytes)
1616 free_bitmap(ctl, bitmap_info);
1617
1618 return 0;
1619 }
1620
1621 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1622 struct btrfs_free_space *info, u64 offset,
1623 u64 bytes)
1624 {
1625 u64 bytes_to_set = 0;
1626 u64 end;
1627
1628 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1629
1630 bytes_to_set = min(end - offset, bytes);
1631
1632 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1633
1634 return bytes_to_set;
1635
1636 }
1637
1638 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1639 struct btrfs_free_space *info)
1640 {
1641 struct btrfs_block_group_cache *block_group = ctl->private;
1642
1643 /*
1644 * If we are below the extents threshold then we can add this as an
1645 * extent, and don't have to deal with the bitmap
1646 */
1647 if (ctl->free_extents < ctl->extents_thresh) {
1648 /*
1649 * If this block group has some small extents we don't want to
1650 * use up all of our free slots in the cache with them, we want
1651 * to reserve them to larger extents, however if we have plent
1652 * of cache left then go ahead an dadd them, no sense in adding
1653 * the overhead of a bitmap if we don't have to.
1654 */
1655 if (info->bytes <= block_group->sectorsize * 4) {
1656 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1657 return false;
1658 } else {
1659 return false;
1660 }
1661 }
1662
1663 /*
1664 * The original block groups from mkfs can be really small, like 8
1665 * megabytes, so don't bother with a bitmap for those entries. However
1666 * some block groups can be smaller than what a bitmap would cover but
1667 * are still large enough that they could overflow the 32k memory limit,
1668 * so allow those block groups to still be allowed to have a bitmap
1669 * entry.
1670 */
1671 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
1672 return false;
1673
1674 return true;
1675 }
1676
1677 static struct btrfs_free_space_op free_space_op = {
1678 .recalc_thresholds = recalculate_thresholds,
1679 .use_bitmap = use_bitmap,
1680 };
1681
1682 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1683 struct btrfs_free_space *info)
1684 {
1685 struct btrfs_free_space *bitmap_info;
1686 struct btrfs_block_group_cache *block_group = NULL;
1687 int added = 0;
1688 u64 bytes, offset, bytes_added;
1689 int ret;
1690
1691 bytes = info->bytes;
1692 offset = info->offset;
1693
1694 if (!ctl->op->use_bitmap(ctl, info))
1695 return 0;
1696
1697 if (ctl->op == &free_space_op)
1698 block_group = ctl->private;
1699 again:
1700 /*
1701 * Since we link bitmaps right into the cluster we need to see if we
1702 * have a cluster here, and if so and it has our bitmap we need to add
1703 * the free space to that bitmap.
1704 */
1705 if (block_group && !list_empty(&block_group->cluster_list)) {
1706 struct btrfs_free_cluster *cluster;
1707 struct rb_node *node;
1708 struct btrfs_free_space *entry;
1709
1710 cluster = list_entry(block_group->cluster_list.next,
1711 struct btrfs_free_cluster,
1712 block_group_list);
1713 spin_lock(&cluster->lock);
1714 node = rb_first(&cluster->root);
1715 if (!node) {
1716 spin_unlock(&cluster->lock);
1717 goto no_cluster_bitmap;
1718 }
1719
1720 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1721 if (!entry->bitmap) {
1722 spin_unlock(&cluster->lock);
1723 goto no_cluster_bitmap;
1724 }
1725
1726 if (entry->offset == offset_to_bitmap(ctl, offset)) {
1727 bytes_added = add_bytes_to_bitmap(ctl, entry,
1728 offset, bytes);
1729 bytes -= bytes_added;
1730 offset += bytes_added;
1731 }
1732 spin_unlock(&cluster->lock);
1733 if (!bytes) {
1734 ret = 1;
1735 goto out;
1736 }
1737 }
1738
1739 no_cluster_bitmap:
1740 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1741 1, 0);
1742 if (!bitmap_info) {
1743 BUG_ON(added);
1744 goto new_bitmap;
1745 }
1746
1747 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
1748 bytes -= bytes_added;
1749 offset += bytes_added;
1750 added = 0;
1751
1752 if (!bytes) {
1753 ret = 1;
1754 goto out;
1755 } else
1756 goto again;
1757
1758 new_bitmap:
1759 if (info && info->bitmap) {
1760 add_new_bitmap(ctl, info, offset);
1761 added = 1;
1762 info = NULL;
1763 goto again;
1764 } else {
1765 spin_unlock(&ctl->tree_lock);
1766
1767 /* no pre-allocated info, allocate a new one */
1768 if (!info) {
1769 info = kmem_cache_zalloc(btrfs_free_space_cachep,
1770 GFP_NOFS);
1771 if (!info) {
1772 spin_lock(&ctl->tree_lock);
1773 ret = -ENOMEM;
1774 goto out;
1775 }
1776 }
1777
1778 /* allocate the bitmap */
1779 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1780 spin_lock(&ctl->tree_lock);
1781 if (!info->bitmap) {
1782 ret = -ENOMEM;
1783 goto out;
1784 }
1785 goto again;
1786 }
1787
1788 out:
1789 if (info) {
1790 if (info->bitmap)
1791 kfree(info->bitmap);
1792 kmem_cache_free(btrfs_free_space_cachep, info);
1793 }
1794
1795 return ret;
1796 }
1797
1798 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1799 struct btrfs_free_space *info, bool update_stat)
1800 {
1801 struct btrfs_free_space *left_info;
1802 struct btrfs_free_space *right_info;
1803 bool merged = false;
1804 u64 offset = info->offset;
1805 u64 bytes = info->bytes;
1806
1807 /*
1808 * first we want to see if there is free space adjacent to the range we
1809 * are adding, if there is remove that struct and add a new one to
1810 * cover the entire range
1811 */
1812 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1813 if (right_info && rb_prev(&right_info->offset_index))
1814 left_info = rb_entry(rb_prev(&right_info->offset_index),
1815 struct btrfs_free_space, offset_index);
1816 else
1817 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
1818
1819 if (right_info && !right_info->bitmap) {
1820 if (update_stat)
1821 unlink_free_space(ctl, right_info);
1822 else
1823 __unlink_free_space(ctl, right_info);
1824 info->bytes += right_info->bytes;
1825 kmem_cache_free(btrfs_free_space_cachep, right_info);
1826 merged = true;
1827 }
1828
1829 if (left_info && !left_info->bitmap &&
1830 left_info->offset + left_info->bytes == offset) {
1831 if (update_stat)
1832 unlink_free_space(ctl, left_info);
1833 else
1834 __unlink_free_space(ctl, left_info);
1835 info->offset = left_info->offset;
1836 info->bytes += left_info->bytes;
1837 kmem_cache_free(btrfs_free_space_cachep, left_info);
1838 merged = true;
1839 }
1840
1841 return merged;
1842 }
1843
1844 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
1845 u64 offset, u64 bytes)
1846 {
1847 struct btrfs_free_space *info;
1848 int ret = 0;
1849
1850 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1851 if (!info)
1852 return -ENOMEM;
1853
1854 info->offset = offset;
1855 info->bytes = bytes;
1856
1857 spin_lock(&ctl->tree_lock);
1858
1859 if (try_merge_free_space(ctl, info, true))
1860 goto link;
1861
1862 /*
1863 * There was no extent directly to the left or right of this new
1864 * extent then we know we're going to have to allocate a new extent, so
1865 * before we do that see if we need to drop this into a bitmap
1866 */
1867 ret = insert_into_bitmap(ctl, info);
1868 if (ret < 0) {
1869 goto out;
1870 } else if (ret) {
1871 ret = 0;
1872 goto out;
1873 }
1874 link:
1875 ret = link_free_space(ctl, info);
1876 if (ret)
1877 kmem_cache_free(btrfs_free_space_cachep, info);
1878 out:
1879 spin_unlock(&ctl->tree_lock);
1880
1881 if (ret) {
1882 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
1883 BUG_ON(ret == -EEXIST);
1884 }
1885
1886 return ret;
1887 }
1888
1889 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1890 u64 offset, u64 bytes)
1891 {
1892 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1893 struct btrfs_free_space *info;
1894 int ret;
1895 bool re_search = false;
1896
1897 spin_lock(&ctl->tree_lock);
1898
1899 again:
1900 ret = 0;
1901 if (!bytes)
1902 goto out_lock;
1903
1904 info = tree_search_offset(ctl, offset, 0, 0);
1905 if (!info) {
1906 /*
1907 * oops didn't find an extent that matched the space we wanted
1908 * to remove, look for a bitmap instead
1909 */
1910 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1911 1, 0);
1912 if (!info) {
1913 /*
1914 * If we found a partial bit of our free space in a
1915 * bitmap but then couldn't find the other part this may
1916 * be a problem, so WARN about it.
1917 */
1918 WARN_ON(re_search);
1919 goto out_lock;
1920 }
1921 }
1922
1923 re_search = false;
1924 if (!info->bitmap) {
1925 unlink_free_space(ctl, info);
1926 if (offset == info->offset) {
1927 u64 to_free = min(bytes, info->bytes);
1928
1929 info->bytes -= to_free;
1930 info->offset += to_free;
1931 if (info->bytes) {
1932 ret = link_free_space(ctl, info);
1933 WARN_ON(ret);
1934 } else {
1935 kmem_cache_free(btrfs_free_space_cachep, info);
1936 }
1937
1938 offset += to_free;
1939 bytes -= to_free;
1940 goto again;
1941 } else {
1942 u64 old_end = info->bytes + info->offset;
1943
1944 info->bytes = offset - info->offset;
1945 ret = link_free_space(ctl, info);
1946 WARN_ON(ret);
1947 if (ret)
1948 goto out_lock;
1949
1950 /* Not enough bytes in this entry to satisfy us */
1951 if (old_end < offset + bytes) {
1952 bytes -= old_end - offset;
1953 offset = old_end;
1954 goto again;
1955 } else if (old_end == offset + bytes) {
1956 /* all done */
1957 goto out_lock;
1958 }
1959 spin_unlock(&ctl->tree_lock);
1960
1961 ret = btrfs_add_free_space(block_group, offset + bytes,
1962 old_end - (offset + bytes));
1963 WARN_ON(ret);
1964 goto out;
1965 }
1966 }
1967
1968 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
1969 if (ret == -EAGAIN) {
1970 re_search = true;
1971 goto again;
1972 }
1973 BUG_ON(ret); /* logic error */
1974 out_lock:
1975 spin_unlock(&ctl->tree_lock);
1976 out:
1977 return ret;
1978 }
1979
1980 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
1981 u64 bytes)
1982 {
1983 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1984 struct btrfs_free_space *info;
1985 struct rb_node *n;
1986 int count = 0;
1987
1988 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
1989 info = rb_entry(n, struct btrfs_free_space, offset_index);
1990 if (info->bytes >= bytes && !block_group->ro)
1991 count++;
1992 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1993 (unsigned long long)info->offset,
1994 (unsigned long long)info->bytes,
1995 (info->bitmap) ? "yes" : "no");
1996 }
1997 printk(KERN_INFO "block group has cluster?: %s\n",
1998 list_empty(&block_group->cluster_list) ? "no" : "yes");
1999 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
2000 "\n", count);
2001 }
2002
2003 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2004 {
2005 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2006
2007 spin_lock_init(&ctl->tree_lock);
2008 ctl->unit = block_group->sectorsize;
2009 ctl->start = block_group->key.objectid;
2010 ctl->private = block_group;
2011 ctl->op = &free_space_op;
2012
2013 /*
2014 * we only want to have 32k of ram per block group for keeping
2015 * track of free space, and if we pass 1/2 of that we want to
2016 * start converting things over to using bitmaps
2017 */
2018 ctl->extents_thresh = ((1024 * 32) / 2) /
2019 sizeof(struct btrfs_free_space);
2020 }
2021
2022 /*
2023 * for a given cluster, put all of its extents back into the free
2024 * space cache. If the block group passed doesn't match the block group
2025 * pointed to by the cluster, someone else raced in and freed the
2026 * cluster already. In that case, we just return without changing anything
2027 */
2028 static int
2029 __btrfs_return_cluster_to_free_space(
2030 struct btrfs_block_group_cache *block_group,
2031 struct btrfs_free_cluster *cluster)
2032 {
2033 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2034 struct btrfs_free_space *entry;
2035 struct rb_node *node;
2036
2037 spin_lock(&cluster->lock);
2038 if (cluster->block_group != block_group)
2039 goto out;
2040
2041 cluster->block_group = NULL;
2042 cluster->window_start = 0;
2043 list_del_init(&cluster->block_group_list);
2044
2045 node = rb_first(&cluster->root);
2046 while (node) {
2047 bool bitmap;
2048
2049 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2050 node = rb_next(&entry->offset_index);
2051 rb_erase(&entry->offset_index, &cluster->root);
2052
2053 bitmap = (entry->bitmap != NULL);
2054 if (!bitmap)
2055 try_merge_free_space(ctl, entry, false);
2056 tree_insert_offset(&ctl->free_space_offset,
2057 entry->offset, &entry->offset_index, bitmap);
2058 }
2059 cluster->root = RB_ROOT;
2060
2061 out:
2062 spin_unlock(&cluster->lock);
2063 btrfs_put_block_group(block_group);
2064 return 0;
2065 }
2066
2067 void __btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl *ctl)
2068 {
2069 struct btrfs_free_space *info;
2070 struct rb_node *node;
2071
2072 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2073 info = rb_entry(node, struct btrfs_free_space, offset_index);
2074 if (!info->bitmap) {
2075 unlink_free_space(ctl, info);
2076 kmem_cache_free(btrfs_free_space_cachep, info);
2077 } else {
2078 free_bitmap(ctl, info);
2079 }
2080 if (need_resched()) {
2081 spin_unlock(&ctl->tree_lock);
2082 cond_resched();
2083 spin_lock(&ctl->tree_lock);
2084 }
2085 }
2086 }
2087
2088 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2089 {
2090 spin_lock(&ctl->tree_lock);
2091 __btrfs_remove_free_space_cache_locked(ctl);
2092 spin_unlock(&ctl->tree_lock);
2093 }
2094
2095 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2096 {
2097 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2098 struct btrfs_free_cluster *cluster;
2099 struct list_head *head;
2100
2101 spin_lock(&ctl->tree_lock);
2102 while ((head = block_group->cluster_list.next) !=
2103 &block_group->cluster_list) {
2104 cluster = list_entry(head, struct btrfs_free_cluster,
2105 block_group_list);
2106
2107 WARN_ON(cluster->block_group != block_group);
2108 __btrfs_return_cluster_to_free_space(block_group, cluster);
2109 if (need_resched()) {
2110 spin_unlock(&ctl->tree_lock);
2111 cond_resched();
2112 spin_lock(&ctl->tree_lock);
2113 }
2114 }
2115 __btrfs_remove_free_space_cache_locked(ctl);
2116 spin_unlock(&ctl->tree_lock);
2117
2118 }
2119
2120 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2121 u64 offset, u64 bytes, u64 empty_size)
2122 {
2123 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2124 struct btrfs_free_space *entry = NULL;
2125 u64 bytes_search = bytes + empty_size;
2126 u64 ret = 0;
2127 u64 align_gap = 0;
2128 u64 align_gap_len = 0;
2129
2130 spin_lock(&ctl->tree_lock);
2131 entry = find_free_space(ctl, &offset, &bytes_search,
2132 block_group->full_stripe_len);
2133 if (!entry)
2134 goto out;
2135
2136 ret = offset;
2137 if (entry->bitmap) {
2138 bitmap_clear_bits(ctl, entry, offset, bytes);
2139 if (!entry->bytes)
2140 free_bitmap(ctl, entry);
2141 } else {
2142
2143 unlink_free_space(ctl, entry);
2144 align_gap_len = offset - entry->offset;
2145 align_gap = entry->offset;
2146
2147 entry->offset = offset + bytes;
2148 WARN_ON(entry->bytes < bytes + align_gap_len);
2149
2150 entry->bytes -= bytes + align_gap_len;
2151 if (!entry->bytes)
2152 kmem_cache_free(btrfs_free_space_cachep, entry);
2153 else
2154 link_free_space(ctl, entry);
2155 }
2156
2157 out:
2158 spin_unlock(&ctl->tree_lock);
2159
2160 if (align_gap_len)
2161 __btrfs_add_free_space(ctl, align_gap, align_gap_len);
2162 return ret;
2163 }
2164
2165 /*
2166 * given a cluster, put all of its extents back into the free space
2167 * cache. If a block group is passed, this function will only free
2168 * a cluster that belongs to the passed block group.
2169 *
2170 * Otherwise, it'll get a reference on the block group pointed to by the
2171 * cluster and remove the cluster from it.
2172 */
2173 int btrfs_return_cluster_to_free_space(
2174 struct btrfs_block_group_cache *block_group,
2175 struct btrfs_free_cluster *cluster)
2176 {
2177 struct btrfs_free_space_ctl *ctl;
2178 int ret;
2179
2180 /* first, get a safe pointer to the block group */
2181 spin_lock(&cluster->lock);
2182 if (!block_group) {
2183 block_group = cluster->block_group;
2184 if (!block_group) {
2185 spin_unlock(&cluster->lock);
2186 return 0;
2187 }
2188 } else if (cluster->block_group != block_group) {
2189 /* someone else has already freed it don't redo their work */
2190 spin_unlock(&cluster->lock);
2191 return 0;
2192 }
2193 atomic_inc(&block_group->count);
2194 spin_unlock(&cluster->lock);
2195
2196 ctl = block_group->free_space_ctl;
2197
2198 /* now return any extents the cluster had on it */
2199 spin_lock(&ctl->tree_lock);
2200 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2201 spin_unlock(&ctl->tree_lock);
2202
2203 /* finally drop our ref */
2204 btrfs_put_block_group(block_group);
2205 return ret;
2206 }
2207
2208 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2209 struct btrfs_free_cluster *cluster,
2210 struct btrfs_free_space *entry,
2211 u64 bytes, u64 min_start)
2212 {
2213 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2214 int err;
2215 u64 search_start = cluster->window_start;
2216 u64 search_bytes = bytes;
2217 u64 ret = 0;
2218
2219 search_start = min_start;
2220 search_bytes = bytes;
2221
2222 err = search_bitmap(ctl, entry, &search_start, &search_bytes);
2223 if (err)
2224 return 0;
2225
2226 ret = search_start;
2227 __bitmap_clear_bits(ctl, entry, ret, bytes);
2228
2229 return ret;
2230 }
2231
2232 /*
2233 * given a cluster, try to allocate 'bytes' from it, returns 0
2234 * if it couldn't find anything suitably large, or a logical disk offset
2235 * if things worked out
2236 */
2237 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2238 struct btrfs_free_cluster *cluster, u64 bytes,
2239 u64 min_start)
2240 {
2241 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2242 struct btrfs_free_space *entry = NULL;
2243 struct rb_node *node;
2244 u64 ret = 0;
2245
2246 spin_lock(&cluster->lock);
2247 if (bytes > cluster->max_size)
2248 goto out;
2249
2250 if (cluster->block_group != block_group)
2251 goto out;
2252
2253 node = rb_first(&cluster->root);
2254 if (!node)
2255 goto out;
2256
2257 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2258 while(1) {
2259 if (entry->bytes < bytes ||
2260 (!entry->bitmap && entry->offset < min_start)) {
2261 node = rb_next(&entry->offset_index);
2262 if (!node)
2263 break;
2264 entry = rb_entry(node, struct btrfs_free_space,
2265 offset_index);
2266 continue;
2267 }
2268
2269 if (entry->bitmap) {
2270 ret = btrfs_alloc_from_bitmap(block_group,
2271 cluster, entry, bytes,
2272 cluster->window_start);
2273 if (ret == 0) {
2274 node = rb_next(&entry->offset_index);
2275 if (!node)
2276 break;
2277 entry = rb_entry(node, struct btrfs_free_space,
2278 offset_index);
2279 continue;
2280 }
2281 cluster->window_start += bytes;
2282 } else {
2283 ret = entry->offset;
2284
2285 entry->offset += bytes;
2286 entry->bytes -= bytes;
2287 }
2288
2289 if (entry->bytes == 0)
2290 rb_erase(&entry->offset_index, &cluster->root);
2291 break;
2292 }
2293 out:
2294 spin_unlock(&cluster->lock);
2295
2296 if (!ret)
2297 return 0;
2298
2299 spin_lock(&ctl->tree_lock);
2300
2301 ctl->free_space -= bytes;
2302 if (entry->bytes == 0) {
2303 ctl->free_extents--;
2304 if (entry->bitmap) {
2305 kfree(entry->bitmap);
2306 ctl->total_bitmaps--;
2307 ctl->op->recalc_thresholds(ctl);
2308 }
2309 kmem_cache_free(btrfs_free_space_cachep, entry);
2310 }
2311
2312 spin_unlock(&ctl->tree_lock);
2313
2314 return ret;
2315 }
2316
2317 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2318 struct btrfs_free_space *entry,
2319 struct btrfs_free_cluster *cluster,
2320 u64 offset, u64 bytes,
2321 u64 cont1_bytes, u64 min_bytes)
2322 {
2323 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2324 unsigned long next_zero;
2325 unsigned long i;
2326 unsigned long want_bits;
2327 unsigned long min_bits;
2328 unsigned long found_bits;
2329 unsigned long start = 0;
2330 unsigned long total_found = 0;
2331 int ret;
2332
2333 i = offset_to_bit(entry->offset, ctl->unit,
2334 max_t(u64, offset, entry->offset));
2335 want_bits = bytes_to_bits(bytes, ctl->unit);
2336 min_bits = bytes_to_bits(min_bytes, ctl->unit);
2337
2338 again:
2339 found_bits = 0;
2340 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2341 next_zero = find_next_zero_bit(entry->bitmap,
2342 BITS_PER_BITMAP, i);
2343 if (next_zero - i >= min_bits) {
2344 found_bits = next_zero - i;
2345 break;
2346 }
2347 i = next_zero;
2348 }
2349
2350 if (!found_bits)
2351 return -ENOSPC;
2352
2353 if (!total_found) {
2354 start = i;
2355 cluster->max_size = 0;
2356 }
2357
2358 total_found += found_bits;
2359
2360 if (cluster->max_size < found_bits * ctl->unit)
2361 cluster->max_size = found_bits * ctl->unit;
2362
2363 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2364 i = next_zero + 1;
2365 goto again;
2366 }
2367
2368 cluster->window_start = start * ctl->unit + entry->offset;
2369 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2370 ret = tree_insert_offset(&cluster->root, entry->offset,
2371 &entry->offset_index, 1);
2372 BUG_ON(ret); /* -EEXIST; Logic error */
2373
2374 trace_btrfs_setup_cluster(block_group, cluster,
2375 total_found * ctl->unit, 1);
2376 return 0;
2377 }
2378
2379 /*
2380 * This searches the block group for just extents to fill the cluster with.
2381 * Try to find a cluster with at least bytes total bytes, at least one
2382 * extent of cont1_bytes, and other clusters of at least min_bytes.
2383 */
2384 static noinline int
2385 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2386 struct btrfs_free_cluster *cluster,
2387 struct list_head *bitmaps, u64 offset, u64 bytes,
2388 u64 cont1_bytes, u64 min_bytes)
2389 {
2390 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2391 struct btrfs_free_space *first = NULL;
2392 struct btrfs_free_space *entry = NULL;
2393 struct btrfs_free_space *last;
2394 struct rb_node *node;
2395 u64 window_start;
2396 u64 window_free;
2397 u64 max_extent;
2398 u64 total_size = 0;
2399
2400 entry = tree_search_offset(ctl, offset, 0, 1);
2401 if (!entry)
2402 return -ENOSPC;
2403
2404 /*
2405 * We don't want bitmaps, so just move along until we find a normal
2406 * extent entry.
2407 */
2408 while (entry->bitmap || entry->bytes < min_bytes) {
2409 if (entry->bitmap && list_empty(&entry->list))
2410 list_add_tail(&entry->list, bitmaps);
2411 node = rb_next(&entry->offset_index);
2412 if (!node)
2413 return -ENOSPC;
2414 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2415 }
2416
2417 window_start = entry->offset;
2418 window_free = entry->bytes;
2419 max_extent = entry->bytes;
2420 first = entry;
2421 last = entry;
2422
2423 for (node = rb_next(&entry->offset_index); node;
2424 node = rb_next(&entry->offset_index)) {
2425 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2426
2427 if (entry->bitmap) {
2428 if (list_empty(&entry->list))
2429 list_add_tail(&entry->list, bitmaps);
2430 continue;
2431 }
2432
2433 if (entry->bytes < min_bytes)
2434 continue;
2435
2436 last = entry;
2437 window_free += entry->bytes;
2438 if (entry->bytes > max_extent)
2439 max_extent = entry->bytes;
2440 }
2441
2442 if (window_free < bytes || max_extent < cont1_bytes)
2443 return -ENOSPC;
2444
2445 cluster->window_start = first->offset;
2446
2447 node = &first->offset_index;
2448
2449 /*
2450 * now we've found our entries, pull them out of the free space
2451 * cache and put them into the cluster rbtree
2452 */
2453 do {
2454 int ret;
2455
2456 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2457 node = rb_next(&entry->offset_index);
2458 if (entry->bitmap || entry->bytes < min_bytes)
2459 continue;
2460
2461 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2462 ret = tree_insert_offset(&cluster->root, entry->offset,
2463 &entry->offset_index, 0);
2464 total_size += entry->bytes;
2465 BUG_ON(ret); /* -EEXIST; Logic error */
2466 } while (node && entry != last);
2467
2468 cluster->max_size = max_extent;
2469 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2470 return 0;
2471 }
2472
2473 /*
2474 * This specifically looks for bitmaps that may work in the cluster, we assume
2475 * that we have already failed to find extents that will work.
2476 */
2477 static noinline int
2478 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2479 struct btrfs_free_cluster *cluster,
2480 struct list_head *bitmaps, u64 offset, u64 bytes,
2481 u64 cont1_bytes, u64 min_bytes)
2482 {
2483 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2484 struct btrfs_free_space *entry;
2485 int ret = -ENOSPC;
2486 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2487
2488 if (ctl->total_bitmaps == 0)
2489 return -ENOSPC;
2490
2491 /*
2492 * The bitmap that covers offset won't be in the list unless offset
2493 * is just its start offset.
2494 */
2495 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2496 if (entry->offset != bitmap_offset) {
2497 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2498 if (entry && list_empty(&entry->list))
2499 list_add(&entry->list, bitmaps);
2500 }
2501
2502 list_for_each_entry(entry, bitmaps, list) {
2503 if (entry->bytes < bytes)
2504 continue;
2505 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2506 bytes, cont1_bytes, min_bytes);
2507 if (!ret)
2508 return 0;
2509 }
2510
2511 /*
2512 * The bitmaps list has all the bitmaps that record free space
2513 * starting after offset, so no more search is required.
2514 */
2515 return -ENOSPC;
2516 }
2517
2518 /*
2519 * here we try to find a cluster of blocks in a block group. The goal
2520 * is to find at least bytes+empty_size.
2521 * We might not find them all in one contiguous area.
2522 *
2523 * returns zero and sets up cluster if things worked out, otherwise
2524 * it returns -enospc
2525 */
2526 int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
2527 struct btrfs_root *root,
2528 struct btrfs_block_group_cache *block_group,
2529 struct btrfs_free_cluster *cluster,
2530 u64 offset, u64 bytes, u64 empty_size)
2531 {
2532 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2533 struct btrfs_free_space *entry, *tmp;
2534 LIST_HEAD(bitmaps);
2535 u64 min_bytes;
2536 u64 cont1_bytes;
2537 int ret;
2538
2539 /*
2540 * Choose the minimum extent size we'll require for this
2541 * cluster. For SSD_SPREAD, don't allow any fragmentation.
2542 * For metadata, allow allocates with smaller extents. For
2543 * data, keep it dense.
2544 */
2545 if (btrfs_test_opt(root, SSD_SPREAD)) {
2546 cont1_bytes = min_bytes = bytes + empty_size;
2547 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2548 cont1_bytes = bytes;
2549 min_bytes = block_group->sectorsize;
2550 } else {
2551 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
2552 min_bytes = block_group->sectorsize;
2553 }
2554
2555 spin_lock(&ctl->tree_lock);
2556
2557 /*
2558 * If we know we don't have enough space to make a cluster don't even
2559 * bother doing all the work to try and find one.
2560 */
2561 if (ctl->free_space < bytes) {
2562 spin_unlock(&ctl->tree_lock);
2563 return -ENOSPC;
2564 }
2565
2566 spin_lock(&cluster->lock);
2567
2568 /* someone already found a cluster, hooray */
2569 if (cluster->block_group) {
2570 ret = 0;
2571 goto out;
2572 }
2573
2574 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
2575 min_bytes);
2576
2577 INIT_LIST_HEAD(&bitmaps);
2578 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2579 bytes + empty_size,
2580 cont1_bytes, min_bytes);
2581 if (ret)
2582 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2583 offset, bytes + empty_size,
2584 cont1_bytes, min_bytes);
2585
2586 /* Clear our temporary list */
2587 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
2588 list_del_init(&entry->list);
2589
2590 if (!ret) {
2591 atomic_inc(&block_group->count);
2592 list_add_tail(&cluster->block_group_list,
2593 &block_group->cluster_list);
2594 cluster->block_group = block_group;
2595 } else {
2596 trace_btrfs_failed_cluster_setup(block_group);
2597 }
2598 out:
2599 spin_unlock(&cluster->lock);
2600 spin_unlock(&ctl->tree_lock);
2601
2602 return ret;
2603 }
2604
2605 /*
2606 * simple code to zero out a cluster
2607 */
2608 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2609 {
2610 spin_lock_init(&cluster->lock);
2611 spin_lock_init(&cluster->refill_lock);
2612 cluster->root = RB_ROOT;
2613 cluster->max_size = 0;
2614 INIT_LIST_HEAD(&cluster->block_group_list);
2615 cluster->block_group = NULL;
2616 }
2617
2618 static int do_trimming(struct btrfs_block_group_cache *block_group,
2619 u64 *total_trimmed, u64 start, u64 bytes,
2620 u64 reserved_start, u64 reserved_bytes)
2621 {
2622 struct btrfs_space_info *space_info = block_group->space_info;
2623 struct btrfs_fs_info *fs_info = block_group->fs_info;
2624 int ret;
2625 int update = 0;
2626 u64 trimmed = 0;
2627
2628 spin_lock(&space_info->lock);
2629 spin_lock(&block_group->lock);
2630 if (!block_group->ro) {
2631 block_group->reserved += reserved_bytes;
2632 space_info->bytes_reserved += reserved_bytes;
2633 update = 1;
2634 }
2635 spin_unlock(&block_group->lock);
2636 spin_unlock(&space_info->lock);
2637
2638 ret = btrfs_error_discard_extent(fs_info->extent_root,
2639 start, bytes, &trimmed);
2640 if (!ret)
2641 *total_trimmed += trimmed;
2642
2643 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
2644
2645 if (update) {
2646 spin_lock(&space_info->lock);
2647 spin_lock(&block_group->lock);
2648 if (block_group->ro)
2649 space_info->bytes_readonly += reserved_bytes;
2650 block_group->reserved -= reserved_bytes;
2651 space_info->bytes_reserved -= reserved_bytes;
2652 spin_unlock(&space_info->lock);
2653 spin_unlock(&block_group->lock);
2654 }
2655
2656 return ret;
2657 }
2658
2659 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
2660 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2661 {
2662 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2663 struct btrfs_free_space *entry;
2664 struct rb_node *node;
2665 int ret = 0;
2666 u64 extent_start;
2667 u64 extent_bytes;
2668 u64 bytes;
2669
2670 while (start < end) {
2671 spin_lock(&ctl->tree_lock);
2672
2673 if (ctl->free_space < minlen) {
2674 spin_unlock(&ctl->tree_lock);
2675 break;
2676 }
2677
2678 entry = tree_search_offset(ctl, start, 0, 1);
2679 if (!entry) {
2680 spin_unlock(&ctl->tree_lock);
2681 break;
2682 }
2683
2684 /* skip bitmaps */
2685 while (entry->bitmap) {
2686 node = rb_next(&entry->offset_index);
2687 if (!node) {
2688 spin_unlock(&ctl->tree_lock);
2689 goto out;
2690 }
2691 entry = rb_entry(node, struct btrfs_free_space,
2692 offset_index);
2693 }
2694
2695 if (entry->offset >= end) {
2696 spin_unlock(&ctl->tree_lock);
2697 break;
2698 }
2699
2700 extent_start = entry->offset;
2701 extent_bytes = entry->bytes;
2702 start = max(start, extent_start);
2703 bytes = min(extent_start + extent_bytes, end) - start;
2704 if (bytes < minlen) {
2705 spin_unlock(&ctl->tree_lock);
2706 goto next;
2707 }
2708
2709 unlink_free_space(ctl, entry);
2710 kmem_cache_free(btrfs_free_space_cachep, entry);
2711
2712 spin_unlock(&ctl->tree_lock);
2713
2714 ret = do_trimming(block_group, total_trimmed, start, bytes,
2715 extent_start, extent_bytes);
2716 if (ret)
2717 break;
2718 next:
2719 start += bytes;
2720
2721 if (fatal_signal_pending(current)) {
2722 ret = -ERESTARTSYS;
2723 break;
2724 }
2725
2726 cond_resched();
2727 }
2728 out:
2729 return ret;
2730 }
2731
2732 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
2733 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2734 {
2735 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2736 struct btrfs_free_space *entry;
2737 int ret = 0;
2738 int ret2;
2739 u64 bytes;
2740 u64 offset = offset_to_bitmap(ctl, start);
2741
2742 while (offset < end) {
2743 bool next_bitmap = false;
2744
2745 spin_lock(&ctl->tree_lock);
2746
2747 if (ctl->free_space < minlen) {
2748 spin_unlock(&ctl->tree_lock);
2749 break;
2750 }
2751
2752 entry = tree_search_offset(ctl, offset, 1, 0);
2753 if (!entry) {
2754 spin_unlock(&ctl->tree_lock);
2755 next_bitmap = true;
2756 goto next;
2757 }
2758
2759 bytes = minlen;
2760 ret2 = search_bitmap(ctl, entry, &start, &bytes);
2761 if (ret2 || start >= end) {
2762 spin_unlock(&ctl->tree_lock);
2763 next_bitmap = true;
2764 goto next;
2765 }
2766
2767 bytes = min(bytes, end - start);
2768 if (bytes < minlen) {
2769 spin_unlock(&ctl->tree_lock);
2770 goto next;
2771 }
2772
2773 bitmap_clear_bits(ctl, entry, start, bytes);
2774 if (entry->bytes == 0)
2775 free_bitmap(ctl, entry);
2776
2777 spin_unlock(&ctl->tree_lock);
2778
2779 ret = do_trimming(block_group, total_trimmed, start, bytes,
2780 start, bytes);
2781 if (ret)
2782 break;
2783 next:
2784 if (next_bitmap) {
2785 offset += BITS_PER_BITMAP * ctl->unit;
2786 } else {
2787 start += bytes;
2788 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
2789 offset += BITS_PER_BITMAP * ctl->unit;
2790 }
2791
2792 if (fatal_signal_pending(current)) {
2793 ret = -ERESTARTSYS;
2794 break;
2795 }
2796
2797 cond_resched();
2798 }
2799
2800 return ret;
2801 }
2802
2803 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2804 u64 *trimmed, u64 start, u64 end, u64 minlen)
2805 {
2806 int ret;
2807
2808 *trimmed = 0;
2809
2810 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
2811 if (ret)
2812 return ret;
2813
2814 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
2815
2816 return ret;
2817 }
2818
2819 /*
2820 * Find the left-most item in the cache tree, and then return the
2821 * smallest inode number in the item.
2822 *
2823 * Note: the returned inode number may not be the smallest one in
2824 * the tree, if the left-most item is a bitmap.
2825 */
2826 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
2827 {
2828 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
2829 struct btrfs_free_space *entry = NULL;
2830 u64 ino = 0;
2831
2832 spin_lock(&ctl->tree_lock);
2833
2834 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
2835 goto out;
2836
2837 entry = rb_entry(rb_first(&ctl->free_space_offset),
2838 struct btrfs_free_space, offset_index);
2839
2840 if (!entry->bitmap) {
2841 ino = entry->offset;
2842
2843 unlink_free_space(ctl, entry);
2844 entry->offset++;
2845 entry->bytes--;
2846 if (!entry->bytes)
2847 kmem_cache_free(btrfs_free_space_cachep, entry);
2848 else
2849 link_free_space(ctl, entry);
2850 } else {
2851 u64 offset = 0;
2852 u64 count = 1;
2853 int ret;
2854
2855 ret = search_bitmap(ctl, entry, &offset, &count);
2856 /* Logic error; Should be empty if it can't find anything */
2857 BUG_ON(ret);
2858
2859 ino = offset;
2860 bitmap_clear_bits(ctl, entry, offset, 1);
2861 if (entry->bytes == 0)
2862 free_bitmap(ctl, entry);
2863 }
2864 out:
2865 spin_unlock(&ctl->tree_lock);
2866
2867 return ino;
2868 }
2869
2870 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
2871 struct btrfs_path *path)
2872 {
2873 struct inode *inode = NULL;
2874
2875 spin_lock(&root->cache_lock);
2876 if (root->cache_inode)
2877 inode = igrab(root->cache_inode);
2878 spin_unlock(&root->cache_lock);
2879 if (inode)
2880 return inode;
2881
2882 inode = __lookup_free_space_inode(root, path, 0);
2883 if (IS_ERR(inode))
2884 return inode;
2885
2886 spin_lock(&root->cache_lock);
2887 if (!btrfs_fs_closing(root->fs_info))
2888 root->cache_inode = igrab(inode);
2889 spin_unlock(&root->cache_lock);
2890
2891 return inode;
2892 }
2893
2894 int create_free_ino_inode(struct btrfs_root *root,
2895 struct btrfs_trans_handle *trans,
2896 struct btrfs_path *path)
2897 {
2898 return __create_free_space_inode(root, trans, path,
2899 BTRFS_FREE_INO_OBJECTID, 0);
2900 }
2901
2902 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2903 {
2904 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2905 struct btrfs_path *path;
2906 struct inode *inode;
2907 int ret = 0;
2908 u64 root_gen = btrfs_root_generation(&root->root_item);
2909
2910 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2911 return 0;
2912
2913 /*
2914 * If we're unmounting then just return, since this does a search on the
2915 * normal root and not the commit root and we could deadlock.
2916 */
2917 if (btrfs_fs_closing(fs_info))
2918 return 0;
2919
2920 path = btrfs_alloc_path();
2921 if (!path)
2922 return 0;
2923
2924 inode = lookup_free_ino_inode(root, path);
2925 if (IS_ERR(inode))
2926 goto out;
2927
2928 if (root_gen != BTRFS_I(inode)->generation)
2929 goto out_put;
2930
2931 ret = __load_free_space_cache(root, inode, ctl, path, 0);
2932
2933 if (ret < 0)
2934 printk(KERN_ERR "btrfs: failed to load free ino cache for "
2935 "root %llu\n", root->root_key.objectid);
2936 out_put:
2937 iput(inode);
2938 out:
2939 btrfs_free_path(path);
2940 return ret;
2941 }
2942
2943 int btrfs_write_out_ino_cache(struct btrfs_root *root,
2944 struct btrfs_trans_handle *trans,
2945 struct btrfs_path *path)
2946 {
2947 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2948 struct inode *inode;
2949 int ret;
2950
2951 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2952 return 0;
2953
2954 inode = lookup_free_ino_inode(root, path);
2955 if (IS_ERR(inode))
2956 return 0;
2957
2958 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
2959 if (ret) {
2960 btrfs_delalloc_release_metadata(inode, inode->i_size);
2961 #ifdef DEBUG
2962 printk(KERN_ERR "btrfs: failed to write free ino cache "
2963 "for root %llu\n", root->root_key.objectid);
2964 #endif
2965 }
2966
2967 iput(inode);
2968 return ret;
2969 }
Linus' kernel tree