?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
HID: hiddev: Fix slab-out-of-bounds write in hiddev_ioctl_usage()
[linux/fpc-iii.git] / fs / btrfs / free-space-cache.c
blob05b1b0f99f0bc39b658cf6d4237b1887a30b9308
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 #include "volumes.h"
31
32 #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8)
33 #define MAX_CACHE_BYTES_PER_GIG (32 * 1024)
34
35 struct btrfs_trim_range {
36 u64 start;
37 u64 bytes;
38 struct list_head list;
39 };
40
41 static int link_free_space(struct btrfs_free_space_ctl *ctl,
42 struct btrfs_free_space *info);
43 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
44 struct btrfs_free_space *info);
45
46 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
47 struct btrfs_path *path,
48 u64 offset)
49 {
50 struct btrfs_key key;
51 struct btrfs_key location;
52 struct btrfs_disk_key disk_key;
53 struct btrfs_free_space_header *header;
54 struct extent_buffer *leaf;
55 struct inode *inode = NULL;
56 int ret;
57
58 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
59 key.offset = offset;
60 key.type = 0;
61
62 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
63 if (ret < 0)
64 return ERR_PTR(ret);
65 if (ret > 0) {
66 btrfs_release_path(path);
67 return ERR_PTR(-ENOENT);
68 }
69
70 leaf = path->nodes[0];
71 header = btrfs_item_ptr(leaf, path->slots[0],
72 struct btrfs_free_space_header);
73 btrfs_free_space_key(leaf, header, &disk_key);
74 btrfs_disk_key_to_cpu(&location, &disk_key);
75 btrfs_release_path(path);
76
77 inode = btrfs_iget(root->fs_info->sb, &location, root, NULL);
78 if (!inode)
79 return ERR_PTR(-ENOENT);
80 if (IS_ERR(inode))
81 return inode;
82 if (is_bad_inode(inode)) {
83 iput(inode);
84 return ERR_PTR(-ENOENT);
85 }
86
87 mapping_set_gfp_mask(inode->i_mapping,
88 mapping_gfp_constraint(inode->i_mapping,
89 ~(__GFP_FS | __GFP_HIGHMEM)));
90
91 return inode;
92 }
93
94 struct inode *lookup_free_space_inode(struct btrfs_root *root,
95 struct btrfs_block_group_cache
96 *block_group, struct btrfs_path *path)
97 {
98 struct inode *inode = NULL;
99 u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
100
101 spin_lock(&block_group->lock);
102 if (block_group->inode)
103 inode = igrab(block_group->inode);
104 spin_unlock(&block_group->lock);
105 if (inode)
106 return inode;
107
108 inode = __lookup_free_space_inode(root, path,
109 block_group->key.objectid);
110 if (IS_ERR(inode))
111 return inode;
112
113 spin_lock(&block_group->lock);
114 if (!((BTRFS_I(inode)->flags & flags) == flags)) {
115 btrfs_info(root->fs_info,
116 "Old style space inode found, converting.");
117 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
118 BTRFS_INODE_NODATACOW;
119 block_group->disk_cache_state = BTRFS_DC_CLEAR;
120 }
121
122 if (!block_group->iref) {
123 block_group->inode = igrab(inode);
124 block_group->iref = 1;
125 }
126 spin_unlock(&block_group->lock);
127
128 return inode;
129 }
130
131 static int __create_free_space_inode(struct btrfs_root *root,
132 struct btrfs_trans_handle *trans,
133 struct btrfs_path *path,
134 u64 ino, u64 offset)
135 {
136 struct btrfs_key key;
137 struct btrfs_disk_key disk_key;
138 struct btrfs_free_space_header *header;
139 struct btrfs_inode_item *inode_item;
140 struct extent_buffer *leaf;
141 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
142 int ret;
143
144 ret = btrfs_insert_empty_inode(trans, root, path, ino);
145 if (ret)
146 return ret;
147
148 /* We inline crc's for the free disk space cache */
149 if (ino != BTRFS_FREE_INO_OBJECTID)
150 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
151
152 leaf = path->nodes[0];
153 inode_item = btrfs_item_ptr(leaf, path->slots[0],
154 struct btrfs_inode_item);
155 btrfs_item_key(leaf, &disk_key, path->slots[0]);
156 memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
157 sizeof(*inode_item));
158 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
159 btrfs_set_inode_size(leaf, inode_item, 0);
160 btrfs_set_inode_nbytes(leaf, inode_item, 0);
161 btrfs_set_inode_uid(leaf, inode_item, 0);
162 btrfs_set_inode_gid(leaf, inode_item, 0);
163 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
164 btrfs_set_inode_flags(leaf, inode_item, flags);
165 btrfs_set_inode_nlink(leaf, inode_item, 1);
166 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
167 btrfs_set_inode_block_group(leaf, inode_item, offset);
168 btrfs_mark_buffer_dirty(leaf);
169 btrfs_release_path(path);
170
171 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
172 key.offset = offset;
173 key.type = 0;
174 ret = btrfs_insert_empty_item(trans, root, path, &key,
175 sizeof(struct btrfs_free_space_header));
176 if (ret < 0) {
177 btrfs_release_path(path);
178 return ret;
179 }
180
181 leaf = path->nodes[0];
182 header = btrfs_item_ptr(leaf, path->slots[0],
183 struct btrfs_free_space_header);
184 memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
185 btrfs_set_free_space_key(leaf, header, &disk_key);
186 btrfs_mark_buffer_dirty(leaf);
187 btrfs_release_path(path);
188
189 return 0;
190 }
191
192 int create_free_space_inode(struct btrfs_root *root,
193 struct btrfs_trans_handle *trans,
194 struct btrfs_block_group_cache *block_group,
195 struct btrfs_path *path)
196 {
197 int ret;
198 u64 ino;
199
200 ret = btrfs_find_free_objectid(root, &ino);
201 if (ret < 0)
202 return ret;
203
204 return __create_free_space_inode(root, trans, path, ino,
205 block_group->key.objectid);
206 }
207
208 int btrfs_check_trunc_cache_free_space(struct btrfs_root *root,
209 struct btrfs_block_rsv *rsv)
210 {
211 u64 needed_bytes;
212 int ret;
213
214 /* 1 for slack space, 1 for updating the inode */
215 needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
216 btrfs_calc_trans_metadata_size(root, 1);
217
218 spin_lock(&rsv->lock);
219 if (rsv->reserved < needed_bytes)
220 ret = -ENOSPC;
221 else
222 ret = 0;
223 spin_unlock(&rsv->lock);
224 return ret;
225 }
226
227 int btrfs_truncate_free_space_cache(struct btrfs_root *root,
228 struct btrfs_trans_handle *trans,
229 struct btrfs_block_group_cache *block_group,
230 struct inode *inode)
231 {
232 int ret = 0;
233 struct btrfs_path *path = btrfs_alloc_path();
234 bool locked = false;
235
236 if (!path) {
237 ret = -ENOMEM;
238 goto fail;
239 }
240
241 if (block_group) {
242 locked = true;
243 mutex_lock(&trans->transaction->cache_write_mutex);
244 if (!list_empty(&block_group->io_list)) {
245 list_del_init(&block_group->io_list);
246
247 btrfs_wait_cache_io(root, trans, block_group,
248 &block_group->io_ctl, path,
249 block_group->key.objectid);
250 btrfs_put_block_group(block_group);
251 }
252
253 /*
254 * now that we've truncated the cache away, its no longer
255 * setup or written
256 */
257 spin_lock(&block_group->lock);
258 block_group->disk_cache_state = BTRFS_DC_CLEAR;
259 spin_unlock(&block_group->lock);
260 }
261 btrfs_free_path(path);
262
263 btrfs_i_size_write(inode, 0);
264 truncate_pagecache(inode, 0);
265
266 /*
267 * We don't need an orphan item because truncating the free space cache
268 * will never be split across transactions.
269 * We don't need to check for -EAGAIN because we're a free space
270 * cache inode
271 */
272 ret = btrfs_truncate_inode_items(trans, root, inode,
273 0, BTRFS_EXTENT_DATA_KEY);
274 if (ret)
275 goto fail;
276
277 ret = btrfs_update_inode(trans, root, inode);
278
279 fail:
280 if (locked)
281 mutex_unlock(&trans->transaction->cache_write_mutex);
282 if (ret)
283 btrfs_abort_transaction(trans, root, ret);
284
285 return ret;
286 }
287
288 static int readahead_cache(struct inode *inode)
289 {
290 struct file_ra_state *ra;
291 unsigned long last_index;
292
293 ra = kzalloc(sizeof(*ra), GFP_NOFS);
294 if (!ra)
295 return -ENOMEM;
296
297 file_ra_state_init(ra, inode->i_mapping);
298 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
299
300 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
301
302 kfree(ra);
303
304 return 0;
305 }
306
307 static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode,
308 struct btrfs_root *root, int write)
309 {
310 int num_pages;
311 int check_crcs = 0;
312
313 num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE);
314
315 if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
316 check_crcs = 1;
317
318 /* Make sure we can fit our crcs into the first page */
319 if (write && check_crcs &&
320 (num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE)
321 return -ENOSPC;
322
323 memset(io_ctl, 0, sizeof(struct btrfs_io_ctl));
324
325 io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS);
326 if (!io_ctl->pages)
327 return -ENOMEM;
328
329 io_ctl->num_pages = num_pages;
330 io_ctl->root = root;
331 io_ctl->check_crcs = check_crcs;
332 io_ctl->inode = inode;
333
334 return 0;
335 }
336
337 static void io_ctl_free(struct btrfs_io_ctl *io_ctl)
338 {
339 kfree(io_ctl->pages);
340 io_ctl->pages = NULL;
341 }
342
343 static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl)
344 {
345 if (io_ctl->cur) {
346 io_ctl->cur = NULL;
347 io_ctl->orig = NULL;
348 }
349 }
350
351 static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear)
352 {
353 ASSERT(io_ctl->index < io_ctl->num_pages);
354 io_ctl->page = io_ctl->pages[io_ctl->index++];
355 io_ctl->cur = page_address(io_ctl->page);
356 io_ctl->orig = io_ctl->cur;
357 io_ctl->size = PAGE_CACHE_SIZE;
358 if (clear)
359 memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
360 }
361
362 static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl)
363 {
364 int i;
365
366 io_ctl_unmap_page(io_ctl);
367
368 for (i = 0; i < io_ctl->num_pages; i++) {
369 if (io_ctl->pages[i]) {
370 ClearPageChecked(io_ctl->pages[i]);
371 unlock_page(io_ctl->pages[i]);
372 page_cache_release(io_ctl->pages[i]);
373 }
374 }
375 }
376
377 static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, struct inode *inode,
378 int uptodate)
379 {
380 struct page *page;
381 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
382 int i;
383
384 for (i = 0; i < io_ctl->num_pages; i++) {
385 page = find_or_create_page(inode->i_mapping, i, mask);
386 if (!page) {
387 io_ctl_drop_pages(io_ctl);
388 return -ENOMEM;
389 }
390 io_ctl->pages[i] = page;
391 if (uptodate && !PageUptodate(page)) {
392 btrfs_readpage(NULL, page);
393 lock_page(page);
394 if (page->mapping != inode->i_mapping) {
395 btrfs_err(BTRFS_I(inode)->root->fs_info,
396 "free space cache page truncated");
397 io_ctl_drop_pages(io_ctl);
398 return -EIO;
399 }
400 if (!PageUptodate(page)) {
401 btrfs_err(BTRFS_I(inode)->root->fs_info,
402 "error reading free space cache");
403 io_ctl_drop_pages(io_ctl);
404 return -EIO;
405 }
406 }
407 }
408
409 for (i = 0; i < io_ctl->num_pages; i++) {
410 clear_page_dirty_for_io(io_ctl->pages[i]);
411 set_page_extent_mapped(io_ctl->pages[i]);
412 }
413
414 return 0;
415 }
416
417 static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
418 {
419 __le64 *val;
420
421 io_ctl_map_page(io_ctl, 1);
422
423 /*
424 * Skip the csum areas. If we don't check crcs then we just have a
425 * 64bit chunk at the front of the first page.
426 */
427 if (io_ctl->check_crcs) {
428 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
429 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
430 } else {
431 io_ctl->cur += sizeof(u64);
432 io_ctl->size -= sizeof(u64) * 2;
433 }
434
435 val = io_ctl->cur;
436 *val = cpu_to_le64(generation);
437 io_ctl->cur += sizeof(u64);
438 }
439
440 static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
441 {
442 __le64 *gen;
443
444 /*
445 * Skip the crc area. If we don't check crcs then we just have a 64bit
446 * chunk at the front of the first page.
447 */
448 if (io_ctl->check_crcs) {
449 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
450 io_ctl->size -= sizeof(u64) +
451 (sizeof(u32) * io_ctl->num_pages);
452 } else {
453 io_ctl->cur += sizeof(u64);
454 io_ctl->size -= sizeof(u64) * 2;
455 }
456
457 gen = io_ctl->cur;
458 if (le64_to_cpu(*gen) != generation) {
459 btrfs_err_rl(io_ctl->root->fs_info,
460 "space cache generation (%llu) does not match inode (%llu)",
461 *gen, generation);
462 io_ctl_unmap_page(io_ctl);
463 return -EIO;
464 }
465 io_ctl->cur += sizeof(u64);
466 return 0;
467 }
468
469 static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
470 {
471 u32 *tmp;
472 u32 crc = ~(u32)0;
473 unsigned offset = 0;
474
475 if (!io_ctl->check_crcs) {
476 io_ctl_unmap_page(io_ctl);
477 return;
478 }
479
480 if (index == 0)
481 offset = sizeof(u32) * io_ctl->num_pages;
482
483 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
484 PAGE_CACHE_SIZE - offset);
485 btrfs_csum_final(crc, (char *)&crc);
486 io_ctl_unmap_page(io_ctl);
487 tmp = page_address(io_ctl->pages[0]);
488 tmp += index;
489 *tmp = crc;
490 }
491
492 static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
493 {
494 u32 *tmp, val;
495 u32 crc = ~(u32)0;
496 unsigned offset = 0;
497
498 if (!io_ctl->check_crcs) {
499 io_ctl_map_page(io_ctl, 0);
500 return 0;
501 }
502
503 if (index == 0)
504 offset = sizeof(u32) * io_ctl->num_pages;
505
506 tmp = page_address(io_ctl->pages[0]);
507 tmp += index;
508 val = *tmp;
509
510 io_ctl_map_page(io_ctl, 0);
511 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
512 PAGE_CACHE_SIZE - offset);
513 btrfs_csum_final(crc, (char *)&crc);
514 if (val != crc) {
515 btrfs_err_rl(io_ctl->root->fs_info,
516 "csum mismatch on free space cache");
517 io_ctl_unmap_page(io_ctl);
518 return -EIO;
519 }
520
521 return 0;
522 }
523
524 static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
525 void *bitmap)
526 {
527 struct btrfs_free_space_entry *entry;
528
529 if (!io_ctl->cur)
530 return -ENOSPC;
531
532 entry = io_ctl->cur;
533 entry->offset = cpu_to_le64(offset);
534 entry->bytes = cpu_to_le64(bytes);
535 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
536 BTRFS_FREE_SPACE_EXTENT;
537 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
538 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
539
540 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
541 return 0;
542
543 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
544
545 /* No more pages to map */
546 if (io_ctl->index >= io_ctl->num_pages)
547 return 0;
548
549 /* map the next page */
550 io_ctl_map_page(io_ctl, 1);
551 return 0;
552 }
553
554 static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
555 {
556 if (!io_ctl->cur)
557 return -ENOSPC;
558
559 /*
560 * If we aren't at the start of the current page, unmap this one and
561 * map the next one if there is any left.
562 */
563 if (io_ctl->cur != io_ctl->orig) {
564 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
565 if (io_ctl->index >= io_ctl->num_pages)
566 return -ENOSPC;
567 io_ctl_map_page(io_ctl, 0);
568 }
569
570 memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
571 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
572 if (io_ctl->index < io_ctl->num_pages)
573 io_ctl_map_page(io_ctl, 0);
574 return 0;
575 }
576
577 static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
578 {
579 /*
580 * If we're not on the boundary we know we've modified the page and we
581 * need to crc the page.
582 */
583 if (io_ctl->cur != io_ctl->orig)
584 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
585 else
586 io_ctl_unmap_page(io_ctl);
587
588 while (io_ctl->index < io_ctl->num_pages) {
589 io_ctl_map_page(io_ctl, 1);
590 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
591 }
592 }
593
594 static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
595 struct btrfs_free_space *entry, u8 *type)
596 {
597 struct btrfs_free_space_entry *e;
598 int ret;
599
600 if (!io_ctl->cur) {
601 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
602 if (ret)
603 return ret;
604 }
605
606 e = io_ctl->cur;
607 entry->offset = le64_to_cpu(e->offset);
608 entry->bytes = le64_to_cpu(e->bytes);
609 *type = e->type;
610 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
611 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
612
613 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
614 return 0;
615
616 io_ctl_unmap_page(io_ctl);
617
618 return 0;
619 }
620
621 static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
622 struct btrfs_free_space *entry)
623 {
624 int ret;
625
626 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
627 if (ret)
628 return ret;
629
630 memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
631 io_ctl_unmap_page(io_ctl);
632
633 return 0;
634 }
635
636 /*
637 * Since we attach pinned extents after the fact we can have contiguous sections
638 * of free space that are split up in entries. This poses a problem with the
639 * tree logging stuff since it could have allocated across what appears to be 2
640 * entries since we would have merged the entries when adding the pinned extents
641 * back to the free space cache. So run through the space cache that we just
642 * loaded and merge contiguous entries. This will make the log replay stuff not
643 * blow up and it will make for nicer allocator behavior.
644 */
645 static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
646 {
647 struct btrfs_free_space *e, *prev = NULL;
648 struct rb_node *n;
649
650 again:
651 spin_lock(&ctl->tree_lock);
652 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
653 e = rb_entry(n, struct btrfs_free_space, offset_index);
654 if (!prev)
655 goto next;
656 if (e->bitmap || prev->bitmap)
657 goto next;
658 if (prev->offset + prev->bytes == e->offset) {
659 unlink_free_space(ctl, prev);
660 unlink_free_space(ctl, e);
661 prev->bytes += e->bytes;
662 kmem_cache_free(btrfs_free_space_cachep, e);
663 link_free_space(ctl, prev);
664 prev = NULL;
665 spin_unlock(&ctl->tree_lock);
666 goto again;
667 }
668 next:
669 prev = e;
670 }
671 spin_unlock(&ctl->tree_lock);
672 }
673
674 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
675 struct btrfs_free_space_ctl *ctl,
676 struct btrfs_path *path, u64 offset)
677 {
678 struct btrfs_free_space_header *header;
679 struct extent_buffer *leaf;
680 struct btrfs_io_ctl io_ctl;
681 struct btrfs_key key;
682 struct btrfs_free_space *e, *n;
683 LIST_HEAD(bitmaps);
684 u64 num_entries;
685 u64 num_bitmaps;
686 u64 generation;
687 u8 type;
688 int ret = 0;
689
690 /* Nothing in the space cache, goodbye */
691 if (!i_size_read(inode))
692 return 0;
693
694 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
695 key.offset = offset;
696 key.type = 0;
697
698 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
699 if (ret < 0)
700 return 0;
701 else if (ret > 0) {
702 btrfs_release_path(path);
703 return 0;
704 }
705
706 ret = -1;
707
708 leaf = path->nodes[0];
709 header = btrfs_item_ptr(leaf, path->slots[0],
710 struct btrfs_free_space_header);
711 num_entries = btrfs_free_space_entries(leaf, header);
712 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
713 generation = btrfs_free_space_generation(leaf, header);
714 btrfs_release_path(path);
715
716 if (!BTRFS_I(inode)->generation) {
717 btrfs_info(root->fs_info,
718 "The free space cache file (%llu) is invalid. skip it\n",
719 offset);
720 return 0;
721 }
722
723 if (BTRFS_I(inode)->generation != generation) {
724 btrfs_err(root->fs_info,
725 "free space inode generation (%llu) "
726 "did not match free space cache generation (%llu)",
727 BTRFS_I(inode)->generation, generation);
728 return 0;
729 }
730
731 if (!num_entries)
732 return 0;
733
734 ret = io_ctl_init(&io_ctl, inode, root, 0);
735 if (ret)
736 return ret;
737
738 ret = readahead_cache(inode);
739 if (ret)
740 goto out;
741
742 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
743 if (ret)
744 goto out;
745
746 ret = io_ctl_check_crc(&io_ctl, 0);
747 if (ret)
748 goto free_cache;
749
750 ret = io_ctl_check_generation(&io_ctl, generation);
751 if (ret)
752 goto free_cache;
753
754 while (num_entries) {
755 e = kmem_cache_zalloc(btrfs_free_space_cachep,
756 GFP_NOFS);
757 if (!e)
758 goto free_cache;
759
760 ret = io_ctl_read_entry(&io_ctl, e, &type);
761 if (ret) {
762 kmem_cache_free(btrfs_free_space_cachep, e);
763 goto free_cache;
764 }
765
766 if (!e->bytes) {
767 kmem_cache_free(btrfs_free_space_cachep, e);
768 goto free_cache;
769 }
770
771 if (type == BTRFS_FREE_SPACE_EXTENT) {
772 spin_lock(&ctl->tree_lock);
773 ret = link_free_space(ctl, e);
774 spin_unlock(&ctl->tree_lock);
775 if (ret) {
776 btrfs_err(root->fs_info,
777 "Duplicate entries in free space cache, dumping");
778 kmem_cache_free(btrfs_free_space_cachep, e);
779 goto free_cache;
780 }
781 } else {
782 ASSERT(num_bitmaps);
783 num_bitmaps--;
784 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
785 if (!e->bitmap) {
786 kmem_cache_free(
787 btrfs_free_space_cachep, e);
788 goto free_cache;
789 }
790 spin_lock(&ctl->tree_lock);
791 ret = link_free_space(ctl, e);
792 ctl->total_bitmaps++;
793 ctl->op->recalc_thresholds(ctl);
794 spin_unlock(&ctl->tree_lock);
795 if (ret) {
796 btrfs_err(root->fs_info,
797 "Duplicate entries in free space cache, dumping");
798 kmem_cache_free(btrfs_free_space_cachep, e);
799 goto free_cache;
800 }
801 list_add_tail(&e->list, &bitmaps);
802 }
803
804 num_entries--;
805 }
806
807 io_ctl_unmap_page(&io_ctl);
808
809 /*
810 * We add the bitmaps at the end of the entries in order that
811 * the bitmap entries are added to the cache.
812 */
813 list_for_each_entry_safe(e, n, &bitmaps, list) {
814 list_del_init(&e->list);
815 ret = io_ctl_read_bitmap(&io_ctl, e);
816 if (ret)
817 goto free_cache;
818 }
819
820 io_ctl_drop_pages(&io_ctl);
821 merge_space_tree(ctl);
822 ret = 1;
823 out:
824 io_ctl_free(&io_ctl);
825 return ret;
826 free_cache:
827 io_ctl_drop_pages(&io_ctl);
828 __btrfs_remove_free_space_cache(ctl);
829 goto out;
830 }
831
832 int load_free_space_cache(struct btrfs_fs_info *fs_info,
833 struct btrfs_block_group_cache *block_group)
834 {
835 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
836 struct btrfs_root *root = fs_info->tree_root;
837 struct inode *inode;
838 struct btrfs_path *path;
839 int ret = 0;
840 bool matched;
841 u64 used = btrfs_block_group_used(&block_group->item);
842
843 /*
844 * If this block group has been marked to be cleared for one reason or
845 * another then we can't trust the on disk cache, so just return.
846 */
847 spin_lock(&block_group->lock);
848 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
849 spin_unlock(&block_group->lock);
850 return 0;
851 }
852 spin_unlock(&block_group->lock);
853
854 path = btrfs_alloc_path();
855 if (!path)
856 return 0;
857 path->search_commit_root = 1;
858 path->skip_locking = 1;
859
860 inode = lookup_free_space_inode(root, block_group, path);
861 if (IS_ERR(inode)) {
862 btrfs_free_path(path);
863 return 0;
864 }
865
866 /* We may have converted the inode and made the cache invalid. */
867 spin_lock(&block_group->lock);
868 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
869 spin_unlock(&block_group->lock);
870 btrfs_free_path(path);
871 goto out;
872 }
873 spin_unlock(&block_group->lock);
874
875 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
876 path, block_group->key.objectid);
877 btrfs_free_path(path);
878 if (ret <= 0)
879 goto out;
880
881 spin_lock(&ctl->tree_lock);
882 matched = (ctl->free_space == (block_group->key.offset - used -
883 block_group->bytes_super));
884 spin_unlock(&ctl->tree_lock);
885
886 if (!matched) {
887 __btrfs_remove_free_space_cache(ctl);
888 btrfs_warn(fs_info, "block group %llu has wrong amount of free space",
889 block_group->key.objectid);
890 ret = -1;
891 }
892 out:
893 if (ret < 0) {
894 /* This cache is bogus, make sure it gets cleared */
895 spin_lock(&block_group->lock);
896 block_group->disk_cache_state = BTRFS_DC_CLEAR;
897 spin_unlock(&block_group->lock);
898 ret = 0;
899
900 btrfs_warn(fs_info, "failed to load free space cache for block group %llu, rebuilding it now",
901 block_group->key.objectid);
902 }
903
904 iput(inode);
905 return ret;
906 }
907
908 static noinline_for_stack
909 int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
910 struct btrfs_free_space_ctl *ctl,
911 struct btrfs_block_group_cache *block_group,
912 int *entries, int *bitmaps,
913 struct list_head *bitmap_list)
914 {
915 int ret;
916 struct btrfs_free_cluster *cluster = NULL;
917 struct btrfs_free_cluster *cluster_locked = NULL;
918 struct rb_node *node = rb_first(&ctl->free_space_offset);
919 struct btrfs_trim_range *trim_entry;
920
921 /* Get the cluster for this block_group if it exists */
922 if (block_group && !list_empty(&block_group->cluster_list)) {
923 cluster = list_entry(block_group->cluster_list.next,
924 struct btrfs_free_cluster,
925 block_group_list);
926 }
927
928 if (!node && cluster) {
929 cluster_locked = cluster;
930 spin_lock(&cluster_locked->lock);
931 node = rb_first(&cluster->root);
932 cluster = NULL;
933 }
934
935 /* Write out the extent entries */
936 while (node) {
937 struct btrfs_free_space *e;
938
939 e = rb_entry(node, struct btrfs_free_space, offset_index);
940 *entries += 1;
941
942 ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
943 e->bitmap);
944 if (ret)
945 goto fail;
946
947 if (e->bitmap) {
948 list_add_tail(&e->list, bitmap_list);
949 *bitmaps += 1;
950 }
951 node = rb_next(node);
952 if (!node && cluster) {
953 node = rb_first(&cluster->root);
954 cluster_locked = cluster;
955 spin_lock(&cluster_locked->lock);
956 cluster = NULL;
957 }
958 }
959 if (cluster_locked) {
960 spin_unlock(&cluster_locked->lock);
961 cluster_locked = NULL;
962 }
963
964 /*
965 * Make sure we don't miss any range that was removed from our rbtree
966 * because trimming is running. Otherwise after a umount+mount (or crash
967 * after committing the transaction) we would leak free space and get
968 * an inconsistent free space cache report from fsck.
969 */
970 list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
971 ret = io_ctl_add_entry(io_ctl, trim_entry->start,
972 trim_entry->bytes, NULL);
973 if (ret)
974 goto fail;
975 *entries += 1;
976 }
977
978 return 0;
979 fail:
980 if (cluster_locked)
981 spin_unlock(&cluster_locked->lock);
982 return -ENOSPC;
983 }
984
985 static noinline_for_stack int
986 update_cache_item(struct btrfs_trans_handle *trans,
987 struct btrfs_root *root,
988 struct inode *inode,
989 struct btrfs_path *path, u64 offset,
990 int entries, int bitmaps)
991 {
992 struct btrfs_key key;
993 struct btrfs_free_space_header *header;
994 struct extent_buffer *leaf;
995 int ret;
996
997 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
998 key.offset = offset;
999 key.type = 0;
1000
1001 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1002 if (ret < 0) {
1003 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1004 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1005 GFP_NOFS);
1006 goto fail;
1007 }
1008 leaf = path->nodes[0];
1009 if (ret > 0) {
1010 struct btrfs_key found_key;
1011 ASSERT(path->slots[0]);
1012 path->slots[0]--;
1013 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1014 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1015 found_key.offset != offset) {
1016 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1017 inode->i_size - 1,
1018 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1019 NULL, GFP_NOFS);
1020 btrfs_release_path(path);
1021 goto fail;
1022 }
1023 }
1024
1025 BTRFS_I(inode)->generation = trans->transid;
1026 header = btrfs_item_ptr(leaf, path->slots[0],
1027 struct btrfs_free_space_header);
1028 btrfs_set_free_space_entries(leaf, header, entries);
1029 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1030 btrfs_set_free_space_generation(leaf, header, trans->transid);
1031 btrfs_mark_buffer_dirty(leaf);
1032 btrfs_release_path(path);
1033
1034 return 0;
1035
1036 fail:
1037 return -1;
1038 }
1039
1040 static noinline_for_stack int
1041 write_pinned_extent_entries(struct btrfs_root *root,
1042 struct btrfs_block_group_cache *block_group,
1043 struct btrfs_io_ctl *io_ctl,
1044 int *entries)
1045 {
1046 u64 start, extent_start, extent_end, len;
1047 struct extent_io_tree *unpin = NULL;
1048 int ret;
1049
1050 if (!block_group)
1051 return 0;
1052
1053 /*
1054 * We want to add any pinned extents to our free space cache
1055 * so we don't leak the space
1056 *
1057 * We shouldn't have switched the pinned extents yet so this is the
1058 * right one
1059 */
1060 unpin = root->fs_info->pinned_extents;
1061
1062 start = block_group->key.objectid;
1063
1064 while (start < block_group->key.objectid + block_group->key.offset) {
1065 ret = find_first_extent_bit(unpin, start,
1066 &extent_start, &extent_end,
1067 EXTENT_DIRTY, NULL);
1068 if (ret)
1069 return 0;
1070
1071 /* This pinned extent is out of our range */
1072 if (extent_start >= block_group->key.objectid +
1073 block_group->key.offset)
1074 return 0;
1075
1076 extent_start = max(extent_start, start);
1077 extent_end = min(block_group->key.objectid +
1078 block_group->key.offset, extent_end + 1);
1079 len = extent_end - extent_start;
1080
1081 *entries += 1;
1082 ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1083 if (ret)
1084 return -ENOSPC;
1085
1086 start = extent_end;
1087 }
1088
1089 return 0;
1090 }
1091
1092 static noinline_for_stack int
1093 write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1094 {
1095 struct list_head *pos, *n;
1096 int ret;
1097
1098 /* Write out the bitmaps */
1099 list_for_each_safe(pos, n, bitmap_list) {
1100 struct btrfs_free_space *entry =
1101 list_entry(pos, struct btrfs_free_space, list);
1102
1103 ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1104 if (ret)
1105 return -ENOSPC;
1106 list_del_init(&entry->list);
1107 }
1108
1109 return 0;
1110 }
1111
1112 static int flush_dirty_cache(struct inode *inode)
1113 {
1114 int ret;
1115
1116 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1117 if (ret)
1118 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1119 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1120 GFP_NOFS);
1121
1122 return ret;
1123 }
1124
1125 static void noinline_for_stack
1126 cleanup_bitmap_list(struct list_head *bitmap_list)
1127 {
1128 struct list_head *pos, *n;
1129
1130 list_for_each_safe(pos, n, bitmap_list) {
1131 struct btrfs_free_space *entry =
1132 list_entry(pos, struct btrfs_free_space, list);
1133 list_del_init(&entry->list);
1134 }
1135 }
1136
1137 static void noinline_for_stack
1138 cleanup_write_cache_enospc(struct inode *inode,
1139 struct btrfs_io_ctl *io_ctl,
1140 struct extent_state **cached_state,
1141 struct list_head *bitmap_list)
1142 {
1143 io_ctl_drop_pages(io_ctl);
1144 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1145 i_size_read(inode) - 1, cached_state,
1146 GFP_NOFS);
1147 }
1148
1149 int btrfs_wait_cache_io(struct btrfs_root *root,
1150 struct btrfs_trans_handle *trans,
1151 struct btrfs_block_group_cache *block_group,
1152 struct btrfs_io_ctl *io_ctl,
1153 struct btrfs_path *path, u64 offset)
1154 {
1155 int ret;
1156 struct inode *inode = io_ctl->inode;
1157
1158 if (!inode)
1159 return 0;
1160
1161 if (block_group)
1162 root = root->fs_info->tree_root;
1163
1164 /* Flush the dirty pages in the cache file. */
1165 ret = flush_dirty_cache(inode);
1166 if (ret)
1167 goto out;
1168
1169 /* Update the cache item to tell everyone this cache file is valid. */
1170 ret = update_cache_item(trans, root, inode, path, offset,
1171 io_ctl->entries, io_ctl->bitmaps);
1172 out:
1173 io_ctl_free(io_ctl);
1174 if (ret) {
1175 invalidate_inode_pages2(inode->i_mapping);
1176 BTRFS_I(inode)->generation = 0;
1177 if (block_group) {
1178 #ifdef DEBUG
1179 btrfs_err(root->fs_info,
1180 "failed to write free space cache for block group %llu",
1181 block_group->key.objectid);
1182 #endif
1183 }
1184 }
1185 btrfs_update_inode(trans, root, inode);
1186
1187 if (block_group) {
1188 /* the dirty list is protected by the dirty_bgs_lock */
1189 spin_lock(&trans->transaction->dirty_bgs_lock);
1190
1191 /* the disk_cache_state is protected by the block group lock */
1192 spin_lock(&block_group->lock);
1193
1194 /*
1195 * only mark this as written if we didn't get put back on
1196 * the dirty list while waiting for IO. Otherwise our
1197 * cache state won't be right, and we won't get written again
1198 */
1199 if (!ret && list_empty(&block_group->dirty_list))
1200 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1201 else if (ret)
1202 block_group->disk_cache_state = BTRFS_DC_ERROR;
1203
1204 spin_unlock(&block_group->lock);
1205 spin_unlock(&trans->transaction->dirty_bgs_lock);
1206 io_ctl->inode = NULL;
1207 iput(inode);
1208 }
1209
1210 return ret;
1211
1212 }
1213
1214 /**
1215 * __btrfs_write_out_cache - write out cached info to an inode
1216 * @root - the root the inode belongs to
1217 * @ctl - the free space cache we are going to write out
1218 * @block_group - the block_group for this cache if it belongs to a block_group
1219 * @trans - the trans handle
1220 * @path - the path to use
1221 * @offset - the offset for the key we'll insert
1222 *
1223 * This function writes out a free space cache struct to disk for quick recovery
1224 * on mount. This will return 0 if it was successful in writing the cache out,
1225 * or an errno if it was not.
1226 */
1227 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
1228 struct btrfs_free_space_ctl *ctl,
1229 struct btrfs_block_group_cache *block_group,
1230 struct btrfs_io_ctl *io_ctl,
1231 struct btrfs_trans_handle *trans,
1232 struct btrfs_path *path, u64 offset)
1233 {
1234 struct extent_state *cached_state = NULL;
1235 LIST_HEAD(bitmap_list);
1236 int entries = 0;
1237 int bitmaps = 0;
1238 int ret;
1239 int must_iput = 0;
1240
1241 if (!i_size_read(inode))
1242 return -EIO;
1243
1244 WARN_ON(io_ctl->pages);
1245 ret = io_ctl_init(io_ctl, inode, root, 1);
1246 if (ret)
1247 return ret;
1248
1249 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1250 down_write(&block_group->data_rwsem);
1251 spin_lock(&block_group->lock);
1252 if (block_group->delalloc_bytes) {
1253 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1254 spin_unlock(&block_group->lock);
1255 up_write(&block_group->data_rwsem);
1256 BTRFS_I(inode)->generation = 0;
1257 ret = 0;
1258 must_iput = 1;
1259 goto out;
1260 }
1261 spin_unlock(&block_group->lock);
1262 }
1263
1264 /* Lock all pages first so we can lock the extent safely. */
1265 ret = io_ctl_prepare_pages(io_ctl, inode, 0);
1266 if (ret)
1267 goto out_unlock;
1268
1269 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1270 0, &cached_state);
1271
1272 io_ctl_set_generation(io_ctl, trans->transid);
1273
1274 mutex_lock(&ctl->cache_writeout_mutex);
1275 /* Write out the extent entries in the free space cache */
1276 spin_lock(&ctl->tree_lock);
1277 ret = write_cache_extent_entries(io_ctl, ctl,
1278 block_group, &entries, &bitmaps,
1279 &bitmap_list);
1280 if (ret)
1281 goto out_nospc_locked;
1282
1283 /*
1284 * Some spaces that are freed in the current transaction are pinned,
1285 * they will be added into free space cache after the transaction is
1286 * committed, we shouldn't lose them.
1287 *
1288 * If this changes while we are working we'll get added back to
1289 * the dirty list and redo it. No locking needed
1290 */
1291 ret = write_pinned_extent_entries(root, block_group, io_ctl, &entries);
1292 if (ret)
1293 goto out_nospc_locked;
1294
1295 /*
1296 * At last, we write out all the bitmaps and keep cache_writeout_mutex
1297 * locked while doing it because a concurrent trim can be manipulating
1298 * or freeing the bitmap.
1299 */
1300 ret = write_bitmap_entries(io_ctl, &bitmap_list);
1301 spin_unlock(&ctl->tree_lock);
1302 mutex_unlock(&ctl->cache_writeout_mutex);
1303 if (ret)
1304 goto out_nospc;
1305
1306 /* Zero out the rest of the pages just to make sure */
1307 io_ctl_zero_remaining_pages(io_ctl);
1308
1309 /* Everything is written out, now we dirty the pages in the file. */
1310 ret = btrfs_dirty_pages(root, inode, io_ctl->pages, io_ctl->num_pages,
1311 0, i_size_read(inode), &cached_state);
1312 if (ret)
1313 goto out_nospc;
1314
1315 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1316 up_write(&block_group->data_rwsem);
1317 /*
1318 * Release the pages and unlock the extent, we will flush
1319 * them out later
1320 */
1321 io_ctl_drop_pages(io_ctl);
1322
1323 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1324 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1325
1326 /*
1327 * at this point the pages are under IO and we're happy,
1328 * The caller is responsible for waiting on them and updating the
1329 * the cache and the inode
1330 */
1331 io_ctl->entries = entries;
1332 io_ctl->bitmaps = bitmaps;
1333
1334 ret = btrfs_fdatawrite_range(inode, 0, (u64)-1);
1335 if (ret)
1336 goto out;
1337
1338 return 0;
1339
1340 out:
1341 io_ctl->inode = NULL;
1342 io_ctl_free(io_ctl);
1343 if (ret) {
1344 invalidate_inode_pages2(inode->i_mapping);
1345 BTRFS_I(inode)->generation = 0;
1346 }
1347 btrfs_update_inode(trans, root, inode);
1348 if (must_iput)
1349 iput(inode);
1350 return ret;
1351
1352 out_nospc_locked:
1353 cleanup_bitmap_list(&bitmap_list);
1354 spin_unlock(&ctl->tree_lock);
1355 mutex_unlock(&ctl->cache_writeout_mutex);
1356
1357 out_nospc:
1358 cleanup_write_cache_enospc(inode, io_ctl, &cached_state, &bitmap_list);
1359
1360 out_unlock:
1361 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1362 up_write(&block_group->data_rwsem);
1363
1364 goto out;
1365 }
1366
1367 int btrfs_write_out_cache(struct btrfs_root *root,
1368 struct btrfs_trans_handle *trans,
1369 struct btrfs_block_group_cache *block_group,
1370 struct btrfs_path *path)
1371 {
1372 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1373 struct inode *inode;
1374 int ret = 0;
1375
1376 root = root->fs_info->tree_root;
1377
1378 spin_lock(&block_group->lock);
1379 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1380 spin_unlock(&block_group->lock);
1381 return 0;
1382 }
1383 spin_unlock(&block_group->lock);
1384
1385 inode = lookup_free_space_inode(root, block_group, path);
1386 if (IS_ERR(inode))
1387 return 0;
1388
1389 ret = __btrfs_write_out_cache(root, inode, ctl, block_group,
1390 &block_group->io_ctl, trans,
1391 path, block_group->key.objectid);
1392 if (ret) {
1393 #ifdef DEBUG
1394 btrfs_err(root->fs_info,
1395 "failed to write free space cache for block group %llu",
1396 block_group->key.objectid);
1397 #endif
1398 spin_lock(&block_group->lock);
1399 block_group->disk_cache_state = BTRFS_DC_ERROR;
1400 spin_unlock(&block_group->lock);
1401
1402 block_group->io_ctl.inode = NULL;
1403 iput(inode);
1404 }
1405
1406 /*
1407 * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1408 * to wait for IO and put the inode
1409 */
1410
1411 return ret;
1412 }
1413
1414 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1415 u64 offset)
1416 {
1417 ASSERT(offset >= bitmap_start);
1418 offset -= bitmap_start;
1419 return (unsigned long)(div_u64(offset, unit));
1420 }
1421
1422 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1423 {
1424 return (unsigned long)(div_u64(bytes, unit));
1425 }
1426
1427 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1428 u64 offset)
1429 {
1430 u64 bitmap_start;
1431 u32 bytes_per_bitmap;
1432
1433 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1434 bitmap_start = offset - ctl->start;
1435 bitmap_start = div_u64(bitmap_start, bytes_per_bitmap);
1436 bitmap_start *= bytes_per_bitmap;
1437 bitmap_start += ctl->start;
1438
1439 return bitmap_start;
1440 }
1441
1442 static int tree_insert_offset(struct rb_root *root, u64 offset,
1443 struct rb_node *node, int bitmap)
1444 {
1445 struct rb_node **p = &root->rb_node;
1446 struct rb_node *parent = NULL;
1447 struct btrfs_free_space *info;
1448
1449 while (*p) {
1450 parent = *p;
1451 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1452
1453 if (offset < info->offset) {
1454 p = &(*p)->rb_left;
1455 } else if (offset > info->offset) {
1456 p = &(*p)->rb_right;
1457 } else {
1458 /*
1459 * we could have a bitmap entry and an extent entry
1460 * share the same offset. If this is the case, we want
1461 * the extent entry to always be found first if we do a
1462 * linear search through the tree, since we want to have
1463 * the quickest allocation time, and allocating from an
1464 * extent is faster than allocating from a bitmap. So
1465 * if we're inserting a bitmap and we find an entry at
1466 * this offset, we want to go right, or after this entry
1467 * logically. If we are inserting an extent and we've
1468 * found a bitmap, we want to go left, or before
1469 * logically.
1470 */
1471 if (bitmap) {
1472 if (info->bitmap) {
1473 WARN_ON_ONCE(1);
1474 return -EEXIST;
1475 }
1476 p = &(*p)->rb_right;
1477 } else {
1478 if (!info->bitmap) {
1479 WARN_ON_ONCE(1);
1480 return -EEXIST;
1481 }
1482 p = &(*p)->rb_left;
1483 }
1484 }
1485 }
1486
1487 rb_link_node(node, parent, p);
1488 rb_insert_color(node, root);
1489
1490 return 0;
1491 }
1492
1493 /*
1494 * searches the tree for the given offset.
1495 *
1496 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1497 * want a section that has at least bytes size and comes at or after the given
1498 * offset.
1499 */
1500 static struct btrfs_free_space *
1501 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1502 u64 offset, int bitmap_only, int fuzzy)
1503 {
1504 struct rb_node *n = ctl->free_space_offset.rb_node;
1505 struct btrfs_free_space *entry, *prev = NULL;
1506
1507 /* find entry that is closest to the 'offset' */
1508 while (1) {
1509 if (!n) {
1510 entry = NULL;
1511 break;
1512 }
1513
1514 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1515 prev = entry;
1516
1517 if (offset < entry->offset)
1518 n = n->rb_left;
1519 else if (offset > entry->offset)
1520 n = n->rb_right;
1521 else
1522 break;
1523 }
1524
1525 if (bitmap_only) {
1526 if (!entry)
1527 return NULL;
1528 if (entry->bitmap)
1529 return entry;
1530
1531 /*
1532 * bitmap entry and extent entry may share same offset,
1533 * in that case, bitmap entry comes after extent entry.
1534 */
1535 n = rb_next(n);
1536 if (!n)
1537 return NULL;
1538 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1539 if (entry->offset != offset)
1540 return NULL;
1541
1542 WARN_ON(!entry->bitmap);
1543 return entry;
1544 } else if (entry) {
1545 if (entry->bitmap) {
1546 /*
1547 * if previous extent entry covers the offset,
1548 * we should return it instead of the bitmap entry
1549 */
1550 n = rb_prev(&entry->offset_index);
1551 if (n) {
1552 prev = rb_entry(n, struct btrfs_free_space,
1553 offset_index);
1554 if (!prev->bitmap &&
1555 prev->offset + prev->bytes > offset)
1556 entry = prev;
1557 }
1558 }
1559 return entry;
1560 }
1561
1562 if (!prev)
1563 return NULL;
1564
1565 /* find last entry before the 'offset' */
1566 entry = prev;
1567 if (entry->offset > offset) {
1568 n = rb_prev(&entry->offset_index);
1569 if (n) {
1570 entry = rb_entry(n, struct btrfs_free_space,
1571 offset_index);
1572 ASSERT(entry->offset <= offset);
1573 } else {
1574 if (fuzzy)
1575 return entry;
1576 else
1577 return NULL;
1578 }
1579 }
1580
1581 if (entry->bitmap) {
1582 n = rb_prev(&entry->offset_index);
1583 if (n) {
1584 prev = rb_entry(n, struct btrfs_free_space,
1585 offset_index);
1586 if (!prev->bitmap &&
1587 prev->offset + prev->bytes > offset)
1588 return prev;
1589 }
1590 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1591 return entry;
1592 } else if (entry->offset + entry->bytes > offset)
1593 return entry;
1594
1595 if (!fuzzy)
1596 return NULL;
1597
1598 while (1) {
1599 if (entry->bitmap) {
1600 if (entry->offset + BITS_PER_BITMAP *
1601 ctl->unit > offset)
1602 break;
1603 } else {
1604 if (entry->offset + entry->bytes > offset)
1605 break;
1606 }
1607
1608 n = rb_next(&entry->offset_index);
1609 if (!n)
1610 return NULL;
1611 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1612 }
1613 return entry;
1614 }
1615
1616 static inline void
1617 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1618 struct btrfs_free_space *info)
1619 {
1620 rb_erase(&info->offset_index, &ctl->free_space_offset);
1621 ctl->free_extents--;
1622 }
1623
1624 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1625 struct btrfs_free_space *info)
1626 {
1627 __unlink_free_space(ctl, info);
1628 ctl->free_space -= info->bytes;
1629 }
1630
1631 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1632 struct btrfs_free_space *info)
1633 {
1634 int ret = 0;
1635
1636 ASSERT(info->bytes || info->bitmap);
1637 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1638 &info->offset_index, (info->bitmap != NULL));
1639 if (ret)
1640 return ret;
1641
1642 ctl->free_space += info->bytes;
1643 ctl->free_extents++;
1644 return ret;
1645 }
1646
1647 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1648 {
1649 struct btrfs_block_group_cache *block_group = ctl->private;
1650 u64 max_bytes;
1651 u64 bitmap_bytes;
1652 u64 extent_bytes;
1653 u64 size = block_group->key.offset;
1654 u32 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1655 u32 max_bitmaps = div_u64(size + bytes_per_bg - 1, bytes_per_bg);
1656
1657 max_bitmaps = max_t(u32, max_bitmaps, 1);
1658
1659 ASSERT(ctl->total_bitmaps <= max_bitmaps);
1660
1661 /*
1662 * The goal is to keep the total amount of memory used per 1gb of space
1663 * at or below 32k, so we need to adjust how much memory we allow to be
1664 * used by extent based free space tracking
1665 */
1666 if (size < 1024 * 1024 * 1024)
1667 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1668 else
1669 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1670 div_u64(size, 1024 * 1024 * 1024);
1671
1672 /*
1673 * we want to account for 1 more bitmap than what we have so we can make
1674 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1675 * we add more bitmaps.
1676 */
1677 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1678
1679 if (bitmap_bytes >= max_bytes) {
1680 ctl->extents_thresh = 0;
1681 return;
1682 }
1683
1684 /*
1685 * we want the extent entry threshold to always be at most 1/2 the max
1686 * bytes we can have, or whatever is less than that.
1687 */
1688 extent_bytes = max_bytes - bitmap_bytes;
1689 extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
1690
1691 ctl->extents_thresh =
1692 div_u64(extent_bytes, sizeof(struct btrfs_free_space));
1693 }
1694
1695 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1696 struct btrfs_free_space *info,
1697 u64 offset, u64 bytes)
1698 {
1699 unsigned long start, count;
1700
1701 start = offset_to_bit(info->offset, ctl->unit, offset);
1702 count = bytes_to_bits(bytes, ctl->unit);
1703 ASSERT(start + count <= BITS_PER_BITMAP);
1704
1705 bitmap_clear(info->bitmap, start, count);
1706
1707 info->bytes -= bytes;
1708 if (info->max_extent_size > ctl->unit)
1709 info->max_extent_size = 0;
1710 }
1711
1712 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1713 struct btrfs_free_space *info, u64 offset,
1714 u64 bytes)
1715 {
1716 __bitmap_clear_bits(ctl, info, offset, bytes);
1717 ctl->free_space -= bytes;
1718 }
1719
1720 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1721 struct btrfs_free_space *info, u64 offset,
1722 u64 bytes)
1723 {
1724 unsigned long start, count;
1725
1726 start = offset_to_bit(info->offset, ctl->unit, offset);
1727 count = bytes_to_bits(bytes, ctl->unit);
1728 ASSERT(start + count <= BITS_PER_BITMAP);
1729
1730 bitmap_set(info->bitmap, start, count);
1731
1732 info->bytes += bytes;
1733 ctl->free_space += bytes;
1734 }
1735
1736 /*
1737 * If we can not find suitable extent, we will use bytes to record
1738 * the size of the max extent.
1739 */
1740 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1741 struct btrfs_free_space *bitmap_info, u64 *offset,
1742 u64 *bytes, bool for_alloc)
1743 {
1744 unsigned long found_bits = 0;
1745 unsigned long max_bits = 0;
1746 unsigned long bits, i;
1747 unsigned long next_zero;
1748 unsigned long extent_bits;
1749
1750 /*
1751 * Skip searching the bitmap if we don't have a contiguous section that
1752 * is large enough for this allocation.
1753 */
1754 if (for_alloc &&
1755 bitmap_info->max_extent_size &&
1756 bitmap_info->max_extent_size < *bytes) {
1757 *bytes = bitmap_info->max_extent_size;
1758 return -1;
1759 }
1760
1761 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1762 max_t(u64, *offset, bitmap_info->offset));
1763 bits = bytes_to_bits(*bytes, ctl->unit);
1764
1765 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1766 if (for_alloc && bits == 1) {
1767 found_bits = 1;
1768 break;
1769 }
1770 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1771 BITS_PER_BITMAP, i);
1772 extent_bits = next_zero - i;
1773 if (extent_bits >= bits) {
1774 found_bits = extent_bits;
1775 break;
1776 } else if (extent_bits > max_bits) {
1777 max_bits = extent_bits;
1778 }
1779 i = next_zero;
1780 }
1781
1782 if (found_bits) {
1783 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1784 *bytes = (u64)(found_bits) * ctl->unit;
1785 return 0;
1786 }
1787
1788 *bytes = (u64)(max_bits) * ctl->unit;
1789 bitmap_info->max_extent_size = *bytes;
1790 return -1;
1791 }
1792
1793 static inline u64 get_max_extent_size(struct btrfs_free_space *entry)
1794 {
1795 if (entry->bitmap)
1796 return entry->max_extent_size;
1797 return entry->bytes;
1798 }
1799
1800 /* Cache the size of the max extent in bytes */
1801 static struct btrfs_free_space *
1802 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1803 unsigned long align, u64 *max_extent_size)
1804 {
1805 struct btrfs_free_space *entry;
1806 struct rb_node *node;
1807 u64 tmp;
1808 u64 align_off;
1809 int ret;
1810
1811 if (!ctl->free_space_offset.rb_node)
1812 goto out;
1813
1814 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1815 if (!entry)
1816 goto out;
1817
1818 for (node = &entry->offset_index; node; node = rb_next(node)) {
1819 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1820 if (entry->bytes < *bytes) {
1821 *max_extent_size = max(get_max_extent_size(entry),
1822 *max_extent_size);
1823 continue;
1824 }
1825
1826 /* make sure the space returned is big enough
1827 * to match our requested alignment
1828 */
1829 if (*bytes >= align) {
1830 tmp = entry->offset - ctl->start + align - 1;
1831 tmp = div64_u64(tmp, align);
1832 tmp = tmp * align + ctl->start;
1833 align_off = tmp - entry->offset;
1834 } else {
1835 align_off = 0;
1836 tmp = entry->offset;
1837 }
1838
1839 if (entry->bytes < *bytes + align_off) {
1840 *max_extent_size = max(get_max_extent_size(entry),
1841 *max_extent_size);
1842 continue;
1843 }
1844
1845 if (entry->bitmap) {
1846 u64 size = *bytes;
1847
1848 ret = search_bitmap(ctl, entry, &tmp, &size, true);
1849 if (!ret) {
1850 *offset = tmp;
1851 *bytes = size;
1852 return entry;
1853 } else {
1854 *max_extent_size =
1855 max(get_max_extent_size(entry),
1856 *max_extent_size);
1857 }
1858 continue;
1859 }
1860
1861 *offset = tmp;
1862 *bytes = entry->bytes - align_off;
1863 return entry;
1864 }
1865 out:
1866 return NULL;
1867 }
1868
1869 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1870 struct btrfs_free_space *info, u64 offset)
1871 {
1872 info->offset = offset_to_bitmap(ctl, offset);
1873 info->bytes = 0;
1874 INIT_LIST_HEAD(&info->list);
1875 link_free_space(ctl, info);
1876 ctl->total_bitmaps++;
1877
1878 ctl->op->recalc_thresholds(ctl);
1879 }
1880
1881 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1882 struct btrfs_free_space *bitmap_info)
1883 {
1884 unlink_free_space(ctl, bitmap_info);
1885 kfree(bitmap_info->bitmap);
1886 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1887 ctl->total_bitmaps--;
1888 ctl->op->recalc_thresholds(ctl);
1889 }
1890
1891 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1892 struct btrfs_free_space *bitmap_info,
1893 u64 *offset, u64 *bytes)
1894 {
1895 u64 end;
1896 u64 search_start, search_bytes;
1897 int ret;
1898
1899 again:
1900 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1901
1902 /*
1903 * We need to search for bits in this bitmap. We could only cover some
1904 * of the extent in this bitmap thanks to how we add space, so we need
1905 * to search for as much as it as we can and clear that amount, and then
1906 * go searching for the next bit.
1907 */
1908 search_start = *offset;
1909 search_bytes = ctl->unit;
1910 search_bytes = min(search_bytes, end - search_start + 1);
1911 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes,
1912 false);
1913 if (ret < 0 || search_start != *offset)
1914 return -EINVAL;
1915
1916 /* We may have found more bits than what we need */
1917 search_bytes = min(search_bytes, *bytes);
1918
1919 /* Cannot clear past the end of the bitmap */
1920 search_bytes = min(search_bytes, end - search_start + 1);
1921
1922 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1923 *offset += search_bytes;
1924 *bytes -= search_bytes;
1925
1926 if (*bytes) {
1927 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1928 if (!bitmap_info->bytes)
1929 free_bitmap(ctl, bitmap_info);
1930
1931 /*
1932 * no entry after this bitmap, but we still have bytes to
1933 * remove, so something has gone wrong.
1934 */
1935 if (!next)
1936 return -EINVAL;
1937
1938 bitmap_info = rb_entry(next, struct btrfs_free_space,
1939 offset_index);
1940
1941 /*
1942 * if the next entry isn't a bitmap we need to return to let the
1943 * extent stuff do its work.
1944 */
1945 if (!bitmap_info->bitmap)
1946 return -EAGAIN;
1947
1948 /*
1949 * Ok the next item is a bitmap, but it may not actually hold
1950 * the information for the rest of this free space stuff, so
1951 * look for it, and if we don't find it return so we can try
1952 * everything over again.
1953 */
1954 search_start = *offset;
1955 search_bytes = ctl->unit;
1956 ret = search_bitmap(ctl, bitmap_info, &search_start,
1957 &search_bytes, false);
1958 if (ret < 0 || search_start != *offset)
1959 return -EAGAIN;
1960
1961 goto again;
1962 } else if (!bitmap_info->bytes)
1963 free_bitmap(ctl, bitmap_info);
1964
1965 return 0;
1966 }
1967
1968 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1969 struct btrfs_free_space *info, u64 offset,
1970 u64 bytes)
1971 {
1972 u64 bytes_to_set = 0;
1973 u64 end;
1974
1975 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1976
1977 bytes_to_set = min(end - offset, bytes);
1978
1979 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1980
1981 /*
1982 * We set some bytes, we have no idea what the max extent size is
1983 * anymore.
1984 */
1985 info->max_extent_size = 0;
1986
1987 return bytes_to_set;
1988
1989 }
1990
1991 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1992 struct btrfs_free_space *info)
1993 {
1994 struct btrfs_block_group_cache *block_group = ctl->private;
1995 bool forced = false;
1996
1997 #ifdef CONFIG_BTRFS_DEBUG
1998 if (btrfs_should_fragment_free_space(block_group->fs_info->extent_root,
1999 block_group))
2000 forced = true;
2001 #endif
2002
2003 /*
2004 * If we are below the extents threshold then we can add this as an
2005 * extent, and don't have to deal with the bitmap
2006 */
2007 if (!forced && ctl->free_extents < ctl->extents_thresh) {
2008 /*
2009 * If this block group has some small extents we don't want to
2010 * use up all of our free slots in the cache with them, we want
2011 * to reserve them to larger extents, however if we have plent
2012 * of cache left then go ahead an dadd them, no sense in adding
2013 * the overhead of a bitmap if we don't have to.
2014 */
2015 if (info->bytes <= block_group->sectorsize * 4) {
2016 if (ctl->free_extents * 2 <= ctl->extents_thresh)
2017 return false;
2018 } else {
2019 return false;
2020 }
2021 }
2022
2023 /*
2024 * The original block groups from mkfs can be really small, like 8
2025 * megabytes, so don't bother with a bitmap for those entries. However
2026 * some block groups can be smaller than what a bitmap would cover but
2027 * are still large enough that they could overflow the 32k memory limit,
2028 * so allow those block groups to still be allowed to have a bitmap
2029 * entry.
2030 */
2031 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
2032 return false;
2033
2034 return true;
2035 }
2036
2037 static struct btrfs_free_space_op free_space_op = {
2038 .recalc_thresholds = recalculate_thresholds,
2039 .use_bitmap = use_bitmap,
2040 };
2041
2042 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
2043 struct btrfs_free_space *info)
2044 {
2045 struct btrfs_free_space *bitmap_info;
2046 struct btrfs_block_group_cache *block_group = NULL;
2047 int added = 0;
2048 u64 bytes, offset, bytes_added;
2049 int ret;
2050
2051 bytes = info->bytes;
2052 offset = info->offset;
2053
2054 if (!ctl->op->use_bitmap(ctl, info))
2055 return 0;
2056
2057 if (ctl->op == &free_space_op)
2058 block_group = ctl->private;
2059 again:
2060 /*
2061 * Since we link bitmaps right into the cluster we need to see if we
2062 * have a cluster here, and if so and it has our bitmap we need to add
2063 * the free space to that bitmap.
2064 */
2065 if (block_group && !list_empty(&block_group->cluster_list)) {
2066 struct btrfs_free_cluster *cluster;
2067 struct rb_node *node;
2068 struct btrfs_free_space *entry;
2069
2070 cluster = list_entry(block_group->cluster_list.next,
2071 struct btrfs_free_cluster,
2072 block_group_list);
2073 spin_lock(&cluster->lock);
2074 node = rb_first(&cluster->root);
2075 if (!node) {
2076 spin_unlock(&cluster->lock);
2077 goto no_cluster_bitmap;
2078 }
2079
2080 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2081 if (!entry->bitmap) {
2082 spin_unlock(&cluster->lock);
2083 goto no_cluster_bitmap;
2084 }
2085
2086 if (entry->offset == offset_to_bitmap(ctl, offset)) {
2087 bytes_added = add_bytes_to_bitmap(ctl, entry,
2088 offset, bytes);
2089 bytes -= bytes_added;
2090 offset += bytes_added;
2091 }
2092 spin_unlock(&cluster->lock);
2093 if (!bytes) {
2094 ret = 1;
2095 goto out;
2096 }
2097 }
2098
2099 no_cluster_bitmap:
2100 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2101 1, 0);
2102 if (!bitmap_info) {
2103 ASSERT(added == 0);
2104 goto new_bitmap;
2105 }
2106
2107 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
2108 bytes -= bytes_added;
2109 offset += bytes_added;
2110 added = 0;
2111
2112 if (!bytes) {
2113 ret = 1;
2114 goto out;
2115 } else
2116 goto again;
2117
2118 new_bitmap:
2119 if (info && info->bitmap) {
2120 add_new_bitmap(ctl, info, offset);
2121 added = 1;
2122 info = NULL;
2123 goto again;
2124 } else {
2125 spin_unlock(&ctl->tree_lock);
2126
2127 /* no pre-allocated info, allocate a new one */
2128 if (!info) {
2129 info = kmem_cache_zalloc(btrfs_free_space_cachep,
2130 GFP_NOFS);
2131 if (!info) {
2132 spin_lock(&ctl->tree_lock);
2133 ret = -ENOMEM;
2134 goto out;
2135 }
2136 }
2137
2138 /* allocate the bitmap */
2139 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
2140 spin_lock(&ctl->tree_lock);
2141 if (!info->bitmap) {
2142 ret = -ENOMEM;
2143 goto out;
2144 }
2145 goto again;
2146 }
2147
2148 out:
2149 if (info) {
2150 if (info->bitmap)
2151 kfree(info->bitmap);
2152 kmem_cache_free(btrfs_free_space_cachep, info);
2153 }
2154
2155 return ret;
2156 }
2157
2158 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2159 struct btrfs_free_space *info, bool update_stat)
2160 {
2161 struct btrfs_free_space *left_info = NULL;
2162 struct btrfs_free_space *right_info;
2163 bool merged = false;
2164 u64 offset = info->offset;
2165 u64 bytes = info->bytes;
2166
2167 /*
2168 * first we want to see if there is free space adjacent to the range we
2169 * are adding, if there is remove that struct and add a new one to
2170 * cover the entire range
2171 */
2172 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
2173 if (right_info && rb_prev(&right_info->offset_index))
2174 left_info = rb_entry(rb_prev(&right_info->offset_index),
2175 struct btrfs_free_space, offset_index);
2176 else if (!right_info)
2177 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
2178
2179 if (right_info && !right_info->bitmap) {
2180 if (update_stat)
2181 unlink_free_space(ctl, right_info);
2182 else
2183 __unlink_free_space(ctl, right_info);
2184 info->bytes += right_info->bytes;
2185 kmem_cache_free(btrfs_free_space_cachep, right_info);
2186 merged = true;
2187 }
2188
2189 if (left_info && !left_info->bitmap &&
2190 left_info->offset + left_info->bytes == offset) {
2191 if (update_stat)
2192 unlink_free_space(ctl, left_info);
2193 else
2194 __unlink_free_space(ctl, left_info);
2195 info->offset = left_info->offset;
2196 info->bytes += left_info->bytes;
2197 kmem_cache_free(btrfs_free_space_cachep, left_info);
2198 merged = true;
2199 }
2200
2201 return merged;
2202 }
2203
2204 static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2205 struct btrfs_free_space *info,
2206 bool update_stat)
2207 {
2208 struct btrfs_free_space *bitmap;
2209 unsigned long i;
2210 unsigned long j;
2211 const u64 end = info->offset + info->bytes;
2212 const u64 bitmap_offset = offset_to_bitmap(ctl, end);
2213 u64 bytes;
2214
2215 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2216 if (!bitmap)
2217 return false;
2218
2219 i = offset_to_bit(bitmap->offset, ctl->unit, end);
2220 j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
2221 if (j == i)
2222 return false;
2223 bytes = (j - i) * ctl->unit;
2224 info->bytes += bytes;
2225
2226 if (update_stat)
2227 bitmap_clear_bits(ctl, bitmap, end, bytes);
2228 else
2229 __bitmap_clear_bits(ctl, bitmap, end, bytes);
2230
2231 if (!bitmap->bytes)
2232 free_bitmap(ctl, bitmap);
2233
2234 return true;
2235 }
2236
2237 static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2238 struct btrfs_free_space *info,
2239 bool update_stat)
2240 {
2241 struct btrfs_free_space *bitmap;
2242 u64 bitmap_offset;
2243 unsigned long i;
2244 unsigned long j;
2245 unsigned long prev_j;
2246 u64 bytes;
2247
2248 bitmap_offset = offset_to_bitmap(ctl, info->offset);
2249 /* If we're on a boundary, try the previous logical bitmap. */
2250 if (bitmap_offset == info->offset) {
2251 if (info->offset == 0)
2252 return false;
2253 bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
2254 }
2255
2256 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2257 if (!bitmap)
2258 return false;
2259
2260 i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
2261 j = 0;
2262 prev_j = (unsigned long)-1;
2263 for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2264 if (j > i)
2265 break;
2266 prev_j = j;
2267 }
2268 if (prev_j == i)
2269 return false;
2270
2271 if (prev_j == (unsigned long)-1)
2272 bytes = (i + 1) * ctl->unit;
2273 else
2274 bytes = (i - prev_j) * ctl->unit;
2275
2276 info->offset -= bytes;
2277 info->bytes += bytes;
2278
2279 if (update_stat)
2280 bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2281 else
2282 __bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2283
2284 if (!bitmap->bytes)
2285 free_bitmap(ctl, bitmap);
2286
2287 return true;
2288 }
2289
2290 /*
2291 * We prefer always to allocate from extent entries, both for clustered and
2292 * non-clustered allocation requests. So when attempting to add a new extent
2293 * entry, try to see if there's adjacent free space in bitmap entries, and if
2294 * there is, migrate that space from the bitmaps to the extent.
2295 * Like this we get better chances of satisfying space allocation requests
2296 * because we attempt to satisfy them based on a single cache entry, and never
2297 * on 2 or more entries - even if the entries represent a contiguous free space
2298 * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2299 * ends).
2300 */
2301 static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2302 struct btrfs_free_space *info,
2303 bool update_stat)
2304 {
2305 /*
2306 * Only work with disconnected entries, as we can change their offset,
2307 * and must be extent entries.
2308 */
2309 ASSERT(!info->bitmap);
2310 ASSERT(RB_EMPTY_NODE(&info->offset_index));
2311
2312 if (ctl->total_bitmaps > 0) {
2313 bool stole_end;
2314 bool stole_front = false;
2315
2316 stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2317 if (ctl->total_bitmaps > 0)
2318 stole_front = steal_from_bitmap_to_front(ctl, info,
2319 update_stat);
2320
2321 if (stole_end || stole_front)
2322 try_merge_free_space(ctl, info, update_stat);
2323 }
2324 }
2325
2326 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
2327 u64 offset, u64 bytes)
2328 {
2329 struct btrfs_free_space *info;
2330 int ret = 0;
2331
2332 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2333 if (!info)
2334 return -ENOMEM;
2335
2336 info->offset = offset;
2337 info->bytes = bytes;
2338 RB_CLEAR_NODE(&info->offset_index);
2339
2340 spin_lock(&ctl->tree_lock);
2341
2342 if (try_merge_free_space(ctl, info, true))
2343 goto link;
2344
2345 /*
2346 * There was no extent directly to the left or right of this new
2347 * extent then we know we're going to have to allocate a new extent, so
2348 * before we do that see if we need to drop this into a bitmap
2349 */
2350 ret = insert_into_bitmap(ctl, info);
2351 if (ret < 0) {
2352 goto out;
2353 } else if (ret) {
2354 ret = 0;
2355 goto out;
2356 }
2357 link:
2358 /*
2359 * Only steal free space from adjacent bitmaps if we're sure we're not
2360 * going to add the new free space to existing bitmap entries - because
2361 * that would mean unnecessary work that would be reverted. Therefore
2362 * attempt to steal space from bitmaps if we're adding an extent entry.
2363 */
2364 steal_from_bitmap(ctl, info, true);
2365
2366 ret = link_free_space(ctl, info);
2367 if (ret)
2368 kmem_cache_free(btrfs_free_space_cachep, info);
2369 out:
2370 spin_unlock(&ctl->tree_lock);
2371
2372 if (ret) {
2373 printk(KERN_CRIT "BTRFS: unable to add free space :%d\n", ret);
2374 ASSERT(ret != -EEXIST);
2375 }
2376
2377 return ret;
2378 }
2379
2380 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
2381 u64 offset, u64 bytes)
2382 {
2383 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2384 struct btrfs_free_space *info;
2385 int ret;
2386 bool re_search = false;
2387
2388 spin_lock(&ctl->tree_lock);
2389
2390 again:
2391 ret = 0;
2392 if (!bytes)
2393 goto out_lock;
2394
2395 info = tree_search_offset(ctl, offset, 0, 0);
2396 if (!info) {
2397 /*
2398 * oops didn't find an extent that matched the space we wanted
2399 * to remove, look for a bitmap instead
2400 */
2401 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2402 1, 0);
2403 if (!info) {
2404 /*
2405 * If we found a partial bit of our free space in a
2406 * bitmap but then couldn't find the other part this may
2407 * be a problem, so WARN about it.
2408 */
2409 WARN_ON(re_search);
2410 goto out_lock;
2411 }
2412 }
2413
2414 re_search = false;
2415 if (!info->bitmap) {
2416 unlink_free_space(ctl, info);
2417 if (offset == info->offset) {
2418 u64 to_free = min(bytes, info->bytes);
2419
2420 info->bytes -= to_free;
2421 info->offset += to_free;
2422 if (info->bytes) {
2423 ret = link_free_space(ctl, info);
2424 WARN_ON(ret);
2425 } else {
2426 kmem_cache_free(btrfs_free_space_cachep, info);
2427 }
2428
2429 offset += to_free;
2430 bytes -= to_free;
2431 goto again;
2432 } else {
2433 u64 old_end = info->bytes + info->offset;
2434
2435 info->bytes = offset - info->offset;
2436 ret = link_free_space(ctl, info);
2437 WARN_ON(ret);
2438 if (ret)
2439 goto out_lock;
2440
2441 /* Not enough bytes in this entry to satisfy us */
2442 if (old_end < offset + bytes) {
2443 bytes -= old_end - offset;
2444 offset = old_end;
2445 goto again;
2446 } else if (old_end == offset + bytes) {
2447 /* all done */
2448 goto out_lock;
2449 }
2450 spin_unlock(&ctl->tree_lock);
2451
2452 ret = btrfs_add_free_space(block_group, offset + bytes,
2453 old_end - (offset + bytes));
2454 WARN_ON(ret);
2455 goto out;
2456 }
2457 }
2458
2459 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2460 if (ret == -EAGAIN) {
2461 re_search = true;
2462 goto again;
2463 }
2464 out_lock:
2465 spin_unlock(&ctl->tree_lock);
2466 out:
2467 return ret;
2468 }
2469
2470 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
2471 u64 bytes)
2472 {
2473 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2474 struct btrfs_free_space *info;
2475 struct rb_node *n;
2476 int count = 0;
2477
2478 spin_lock(&ctl->tree_lock);
2479 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2480 info = rb_entry(n, struct btrfs_free_space, offset_index);
2481 if (info->bytes >= bytes && !block_group->ro)
2482 count++;
2483 btrfs_crit(block_group->fs_info,
2484 "entry offset %llu, bytes %llu, bitmap %s",
2485 info->offset, info->bytes,
2486 (info->bitmap) ? "yes" : "no");
2487 }
2488 spin_unlock(&ctl->tree_lock);
2489 btrfs_info(block_group->fs_info, "block group has cluster?: %s",
2490 list_empty(&block_group->cluster_list) ? "no" : "yes");
2491 btrfs_info(block_group->fs_info,
2492 "%d blocks of free space at or bigger than bytes is", count);
2493 }
2494
2495 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2496 {
2497 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2498
2499 spin_lock_init(&ctl->tree_lock);
2500 ctl->unit = block_group->sectorsize;
2501 ctl->start = block_group->key.objectid;
2502 ctl->private = block_group;
2503 ctl->op = &free_space_op;
2504 INIT_LIST_HEAD(&ctl->trimming_ranges);
2505 mutex_init(&ctl->cache_writeout_mutex);
2506
2507 /*
2508 * we only want to have 32k of ram per block group for keeping
2509 * track of free space, and if we pass 1/2 of that we want to
2510 * start converting things over to using bitmaps
2511 */
2512 ctl->extents_thresh = ((1024 * 32) / 2) /
2513 sizeof(struct btrfs_free_space);
2514 }
2515
2516 /*
2517 * for a given cluster, put all of its extents back into the free
2518 * space cache. If the block group passed doesn't match the block group
2519 * pointed to by the cluster, someone else raced in and freed the
2520 * cluster already. In that case, we just return without changing anything
2521 */
2522 static int
2523 __btrfs_return_cluster_to_free_space(
2524 struct btrfs_block_group_cache *block_group,
2525 struct btrfs_free_cluster *cluster)
2526 {
2527 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2528 struct btrfs_free_space *entry;
2529 struct rb_node *node;
2530
2531 spin_lock(&cluster->lock);
2532 if (cluster->block_group != block_group)
2533 goto out;
2534
2535 cluster->block_group = NULL;
2536 cluster->window_start = 0;
2537 list_del_init(&cluster->block_group_list);
2538
2539 node = rb_first(&cluster->root);
2540 while (node) {
2541 bool bitmap;
2542
2543 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2544 node = rb_next(&entry->offset_index);
2545 rb_erase(&entry->offset_index, &cluster->root);
2546 RB_CLEAR_NODE(&entry->offset_index);
2547
2548 bitmap = (entry->bitmap != NULL);
2549 if (!bitmap) {
2550 try_merge_free_space(ctl, entry, false);
2551 steal_from_bitmap(ctl, entry, false);
2552 }
2553 tree_insert_offset(&ctl->free_space_offset,
2554 entry->offset, &entry->offset_index, bitmap);
2555 }
2556 cluster->root = RB_ROOT;
2557
2558 out:
2559 spin_unlock(&cluster->lock);
2560 btrfs_put_block_group(block_group);
2561 return 0;
2562 }
2563
2564 static void __btrfs_remove_free_space_cache_locked(
2565 struct btrfs_free_space_ctl *ctl)
2566 {
2567 struct btrfs_free_space *info;
2568 struct rb_node *node;
2569
2570 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2571 info = rb_entry(node, struct btrfs_free_space, offset_index);
2572 if (!info->bitmap) {
2573 unlink_free_space(ctl, info);
2574 kmem_cache_free(btrfs_free_space_cachep, info);
2575 } else {
2576 free_bitmap(ctl, info);
2577 }
2578
2579 cond_resched_lock(&ctl->tree_lock);
2580 }
2581 }
2582
2583 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2584 {
2585 spin_lock(&ctl->tree_lock);
2586 __btrfs_remove_free_space_cache_locked(ctl);
2587 spin_unlock(&ctl->tree_lock);
2588 }
2589
2590 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2591 {
2592 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2593 struct btrfs_free_cluster *cluster;
2594 struct list_head *head;
2595
2596 spin_lock(&ctl->tree_lock);
2597 while ((head = block_group->cluster_list.next) !=
2598 &block_group->cluster_list) {
2599 cluster = list_entry(head, struct btrfs_free_cluster,
2600 block_group_list);
2601
2602 WARN_ON(cluster->block_group != block_group);
2603 __btrfs_return_cluster_to_free_space(block_group, cluster);
2604
2605 cond_resched_lock(&ctl->tree_lock);
2606 }
2607 __btrfs_remove_free_space_cache_locked(ctl);
2608 spin_unlock(&ctl->tree_lock);
2609
2610 }
2611
2612 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2613 u64 offset, u64 bytes, u64 empty_size,
2614 u64 *max_extent_size)
2615 {
2616 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2617 struct btrfs_free_space *entry = NULL;
2618 u64 bytes_search = bytes + empty_size;
2619 u64 ret = 0;
2620 u64 align_gap = 0;
2621 u64 align_gap_len = 0;
2622
2623 spin_lock(&ctl->tree_lock);
2624 entry = find_free_space(ctl, &offset, &bytes_search,
2625 block_group->full_stripe_len, max_extent_size);
2626 if (!entry)
2627 goto out;
2628
2629 ret = offset;
2630 if (entry->bitmap) {
2631 bitmap_clear_bits(ctl, entry, offset, bytes);
2632 if (!entry->bytes)
2633 free_bitmap(ctl, entry);
2634 } else {
2635 unlink_free_space(ctl, entry);
2636 align_gap_len = offset - entry->offset;
2637 align_gap = entry->offset;
2638
2639 entry->offset = offset + bytes;
2640 WARN_ON(entry->bytes < bytes + align_gap_len);
2641
2642 entry->bytes -= bytes + align_gap_len;
2643 if (!entry->bytes)
2644 kmem_cache_free(btrfs_free_space_cachep, entry);
2645 else
2646 link_free_space(ctl, entry);
2647 }
2648 out:
2649 spin_unlock(&ctl->tree_lock);
2650
2651 if (align_gap_len)
2652 __btrfs_add_free_space(ctl, align_gap, align_gap_len);
2653 return ret;
2654 }
2655
2656 /*
2657 * given a cluster, put all of its extents back into the free space
2658 * cache. If a block group is passed, this function will only free
2659 * a cluster that belongs to the passed block group.
2660 *
2661 * Otherwise, it'll get a reference on the block group pointed to by the
2662 * cluster and remove the cluster from it.
2663 */
2664 int btrfs_return_cluster_to_free_space(
2665 struct btrfs_block_group_cache *block_group,
2666 struct btrfs_free_cluster *cluster)
2667 {
2668 struct btrfs_free_space_ctl *ctl;
2669 int ret;
2670
2671 /* first, get a safe pointer to the block group */
2672 spin_lock(&cluster->lock);
2673 if (!block_group) {
2674 block_group = cluster->block_group;
2675 if (!block_group) {
2676 spin_unlock(&cluster->lock);
2677 return 0;
2678 }
2679 } else if (cluster->block_group != block_group) {
2680 /* someone else has already freed it don't redo their work */
2681 spin_unlock(&cluster->lock);
2682 return 0;
2683 }
2684 atomic_inc(&block_group->count);
2685 spin_unlock(&cluster->lock);
2686
2687 ctl = block_group->free_space_ctl;
2688
2689 /* now return any extents the cluster had on it */
2690 spin_lock(&ctl->tree_lock);
2691 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2692 spin_unlock(&ctl->tree_lock);
2693
2694 /* finally drop our ref */
2695 btrfs_put_block_group(block_group);
2696 return ret;
2697 }
2698
2699 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2700 struct btrfs_free_cluster *cluster,
2701 struct btrfs_free_space *entry,
2702 u64 bytes, u64 min_start,
2703 u64 *max_extent_size)
2704 {
2705 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2706 int err;
2707 u64 search_start = cluster->window_start;
2708 u64 search_bytes = bytes;
2709 u64 ret = 0;
2710
2711 search_start = min_start;
2712 search_bytes = bytes;
2713
2714 err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
2715 if (err) {
2716 *max_extent_size = max(get_max_extent_size(entry),
2717 *max_extent_size);
2718 return 0;
2719 }
2720
2721 ret = search_start;
2722 __bitmap_clear_bits(ctl, entry, ret, bytes);
2723
2724 return ret;
2725 }
2726
2727 /*
2728 * given a cluster, try to allocate 'bytes' from it, returns 0
2729 * if it couldn't find anything suitably large, or a logical disk offset
2730 * if things worked out
2731 */
2732 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2733 struct btrfs_free_cluster *cluster, u64 bytes,
2734 u64 min_start, u64 *max_extent_size)
2735 {
2736 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2737 struct btrfs_free_space *entry = NULL;
2738 struct rb_node *node;
2739 u64 ret = 0;
2740
2741 spin_lock(&cluster->lock);
2742 if (bytes > cluster->max_size)
2743 goto out;
2744
2745 if (cluster->block_group != block_group)
2746 goto out;
2747
2748 node = rb_first(&cluster->root);
2749 if (!node)
2750 goto out;
2751
2752 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2753 while (1) {
2754 if (entry->bytes < bytes)
2755 *max_extent_size = max(get_max_extent_size(entry),
2756 *max_extent_size);
2757
2758 if (entry->bytes < bytes ||
2759 (!entry->bitmap && entry->offset < min_start)) {
2760 node = rb_next(&entry->offset_index);
2761 if (!node)
2762 break;
2763 entry = rb_entry(node, struct btrfs_free_space,
2764 offset_index);
2765 continue;
2766 }
2767
2768 if (entry->bitmap) {
2769 ret = btrfs_alloc_from_bitmap(block_group,
2770 cluster, entry, bytes,
2771 cluster->window_start,
2772 max_extent_size);
2773 if (ret == 0) {
2774 node = rb_next(&entry->offset_index);
2775 if (!node)
2776 break;
2777 entry = rb_entry(node, struct btrfs_free_space,
2778 offset_index);
2779 continue;
2780 }
2781 cluster->window_start += bytes;
2782 } else {
2783 ret = entry->offset;
2784
2785 entry->offset += bytes;
2786 entry->bytes -= bytes;
2787 }
2788
2789 if (entry->bytes == 0)
2790 rb_erase(&entry->offset_index, &cluster->root);
2791 break;
2792 }
2793 out:
2794 spin_unlock(&cluster->lock);
2795
2796 if (!ret)
2797 return 0;
2798
2799 spin_lock(&ctl->tree_lock);
2800
2801 ctl->free_space -= bytes;
2802 if (entry->bytes == 0) {
2803 ctl->free_extents--;
2804 if (entry->bitmap) {
2805 kfree(entry->bitmap);
2806 ctl->total_bitmaps--;
2807 ctl->op->recalc_thresholds(ctl);
2808 }
2809 kmem_cache_free(btrfs_free_space_cachep, entry);
2810 }
2811
2812 spin_unlock(&ctl->tree_lock);
2813
2814 return ret;
2815 }
2816
2817 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2818 struct btrfs_free_space *entry,
2819 struct btrfs_free_cluster *cluster,
2820 u64 offset, u64 bytes,
2821 u64 cont1_bytes, u64 min_bytes)
2822 {
2823 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2824 unsigned long next_zero;
2825 unsigned long i;
2826 unsigned long want_bits;
2827 unsigned long min_bits;
2828 unsigned long found_bits;
2829 unsigned long max_bits = 0;
2830 unsigned long start = 0;
2831 unsigned long total_found = 0;
2832 int ret;
2833
2834 i = offset_to_bit(entry->offset, ctl->unit,
2835 max_t(u64, offset, entry->offset));
2836 want_bits = bytes_to_bits(bytes, ctl->unit);
2837 min_bits = bytes_to_bits(min_bytes, ctl->unit);
2838
2839 /*
2840 * Don't bother looking for a cluster in this bitmap if it's heavily
2841 * fragmented.
2842 */
2843 if (entry->max_extent_size &&
2844 entry->max_extent_size < cont1_bytes)
2845 return -ENOSPC;
2846 again:
2847 found_bits = 0;
2848 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2849 next_zero = find_next_zero_bit(entry->bitmap,
2850 BITS_PER_BITMAP, i);
2851 if (next_zero - i >= min_bits) {
2852 found_bits = next_zero - i;
2853 if (found_bits > max_bits)
2854 max_bits = found_bits;
2855 break;
2856 }
2857 if (next_zero - i > max_bits)
2858 max_bits = next_zero - i;
2859 i = next_zero;
2860 }
2861
2862 if (!found_bits) {
2863 entry->max_extent_size = (u64)max_bits * ctl->unit;
2864 return -ENOSPC;
2865 }
2866
2867 if (!total_found) {
2868 start = i;
2869 cluster->max_size = 0;
2870 }
2871
2872 total_found += found_bits;
2873
2874 if (cluster->max_size < found_bits * ctl->unit)
2875 cluster->max_size = found_bits * ctl->unit;
2876
2877 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2878 i = next_zero + 1;
2879 goto again;
2880 }
2881
2882 cluster->window_start = start * ctl->unit + entry->offset;
2883 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2884 ret = tree_insert_offset(&cluster->root, entry->offset,
2885 &entry->offset_index, 1);
2886 ASSERT(!ret); /* -EEXIST; Logic error */
2887
2888 trace_btrfs_setup_cluster(block_group, cluster,
2889 total_found * ctl->unit, 1);
2890 return 0;
2891 }
2892
2893 /*
2894 * This searches the block group for just extents to fill the cluster with.
2895 * Try to find a cluster with at least bytes total bytes, at least one
2896 * extent of cont1_bytes, and other clusters of at least min_bytes.
2897 */
2898 static noinline int
2899 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2900 struct btrfs_free_cluster *cluster,
2901 struct list_head *bitmaps, u64 offset, u64 bytes,
2902 u64 cont1_bytes, u64 min_bytes)
2903 {
2904 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2905 struct btrfs_free_space *first = NULL;
2906 struct btrfs_free_space *entry = NULL;
2907 struct btrfs_free_space *last;
2908 struct rb_node *node;
2909 u64 window_free;
2910 u64 max_extent;
2911 u64 total_size = 0;
2912
2913 entry = tree_search_offset(ctl, offset, 0, 1);
2914 if (!entry)
2915 return -ENOSPC;
2916
2917 /*
2918 * We don't want bitmaps, so just move along until we find a normal
2919 * extent entry.
2920 */
2921 while (entry->bitmap || entry->bytes < min_bytes) {
2922 if (entry->bitmap && list_empty(&entry->list))
2923 list_add_tail(&entry->list, bitmaps);
2924 node = rb_next(&entry->offset_index);
2925 if (!node)
2926 return -ENOSPC;
2927 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2928 }
2929
2930 window_free = entry->bytes;
2931 max_extent = entry->bytes;
2932 first = entry;
2933 last = entry;
2934
2935 for (node = rb_next(&entry->offset_index); node;
2936 node = rb_next(&entry->offset_index)) {
2937 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2938
2939 if (entry->bitmap) {
2940 if (list_empty(&entry->list))
2941 list_add_tail(&entry->list, bitmaps);
2942 continue;
2943 }
2944
2945 if (entry->bytes < min_bytes)
2946 continue;
2947
2948 last = entry;
2949 window_free += entry->bytes;
2950 if (entry->bytes > max_extent)
2951 max_extent = entry->bytes;
2952 }
2953
2954 if (window_free < bytes || max_extent < cont1_bytes)
2955 return -ENOSPC;
2956
2957 cluster->window_start = first->offset;
2958
2959 node = &first->offset_index;
2960
2961 /*
2962 * now we've found our entries, pull them out of the free space
2963 * cache and put them into the cluster rbtree
2964 */
2965 do {
2966 int ret;
2967
2968 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2969 node = rb_next(&entry->offset_index);
2970 if (entry->bitmap || entry->bytes < min_bytes)
2971 continue;
2972
2973 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2974 ret = tree_insert_offset(&cluster->root, entry->offset,
2975 &entry->offset_index, 0);
2976 total_size += entry->bytes;
2977 ASSERT(!ret); /* -EEXIST; Logic error */
2978 } while (node && entry != last);
2979
2980 cluster->max_size = max_extent;
2981 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2982 return 0;
2983 }
2984
2985 /*
2986 * This specifically looks for bitmaps that may work in the cluster, we assume
2987 * that we have already failed to find extents that will work.
2988 */
2989 static noinline int
2990 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2991 struct btrfs_free_cluster *cluster,
2992 struct list_head *bitmaps, u64 offset, u64 bytes,
2993 u64 cont1_bytes, u64 min_bytes)
2994 {
2995 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2996 struct btrfs_free_space *entry = NULL;
2997 int ret = -ENOSPC;
2998 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2999
3000 if (ctl->total_bitmaps == 0)
3001 return -ENOSPC;
3002
3003 /*
3004 * The bitmap that covers offset won't be in the list unless offset
3005 * is just its start offset.
3006 */
3007 if (!list_empty(bitmaps))
3008 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
3009
3010 if (!entry || entry->offset != bitmap_offset) {
3011 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
3012 if (entry && list_empty(&entry->list))
3013 list_add(&entry->list, bitmaps);
3014 }
3015
3016 list_for_each_entry(entry, bitmaps, list) {
3017 if (entry->bytes < bytes)
3018 continue;
3019 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
3020 bytes, cont1_bytes, min_bytes);
3021 if (!ret)
3022 return 0;
3023 }
3024
3025 /*
3026 * The bitmaps list has all the bitmaps that record free space
3027 * starting after offset, so no more search is required.
3028 */
3029 return -ENOSPC;
3030 }
3031
3032 /*
3033 * here we try to find a cluster of blocks in a block group. The goal
3034 * is to find at least bytes+empty_size.
3035 * We might not find them all in one contiguous area.
3036 *
3037 * returns zero and sets up cluster if things worked out, otherwise
3038 * it returns -enospc
3039 */
3040 int btrfs_find_space_cluster(struct btrfs_root *root,
3041 struct btrfs_block_group_cache *block_group,
3042 struct btrfs_free_cluster *cluster,
3043 u64 offset, u64 bytes, u64 empty_size)
3044 {
3045 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3046 struct btrfs_free_space *entry, *tmp;
3047 LIST_HEAD(bitmaps);
3048 u64 min_bytes;
3049 u64 cont1_bytes;
3050 int ret;
3051
3052 /*
3053 * Choose the minimum extent size we'll require for this
3054 * cluster. For SSD_SPREAD, don't allow any fragmentation.
3055 * For metadata, allow allocates with smaller extents. For
3056 * data, keep it dense.
3057 */
3058 if (btrfs_test_opt(root, SSD_SPREAD)) {
3059 cont1_bytes = min_bytes = bytes + empty_size;
3060 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3061 cont1_bytes = bytes;
3062 min_bytes = block_group->sectorsize;
3063 } else {
3064 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3065 min_bytes = block_group->sectorsize;
3066 }
3067
3068 spin_lock(&ctl->tree_lock);
3069
3070 /*
3071 * If we know we don't have enough space to make a cluster don't even
3072 * bother doing all the work to try and find one.
3073 */
3074 if (ctl->free_space < bytes) {
3075 spin_unlock(&ctl->tree_lock);
3076 return -ENOSPC;
3077 }
3078
3079 spin_lock(&cluster->lock);
3080
3081 /* someone already found a cluster, hooray */
3082 if (cluster->block_group) {
3083 ret = 0;
3084 goto out;
3085 }
3086
3087 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
3088 min_bytes);
3089
3090 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
3091 bytes + empty_size,
3092 cont1_bytes, min_bytes);
3093 if (ret)
3094 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
3095 offset, bytes + empty_size,
3096 cont1_bytes, min_bytes);
3097
3098 /* Clear our temporary list */
3099 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3100 list_del_init(&entry->list);
3101
3102 if (!ret) {
3103 atomic_inc(&block_group->count);
3104 list_add_tail(&cluster->block_group_list,
3105 &block_group->cluster_list);
3106 cluster->block_group = block_group;
3107 } else {
3108 trace_btrfs_failed_cluster_setup(block_group);
3109 }
3110 out:
3111 spin_unlock(&cluster->lock);
3112 spin_unlock(&ctl->tree_lock);
3113
3114 return ret;
3115 }
3116
3117 /*
3118 * simple code to zero out a cluster
3119 */
3120 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3121 {
3122 spin_lock_init(&cluster->lock);
3123 spin_lock_init(&cluster->refill_lock);
3124 cluster->root = RB_ROOT;
3125 cluster->max_size = 0;
3126 cluster->fragmented = false;
3127 INIT_LIST_HEAD(&cluster->block_group_list);
3128 cluster->block_group = NULL;
3129 }
3130
3131 static int do_trimming(struct btrfs_block_group_cache *block_group,
3132 u64 *total_trimmed, u64 start, u64 bytes,
3133 u64 reserved_start, u64 reserved_bytes,
3134 struct btrfs_trim_range *trim_entry)
3135 {
3136 struct btrfs_space_info *space_info = block_group->space_info;
3137 struct btrfs_fs_info *fs_info = block_group->fs_info;
3138 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3139 int ret;
3140 int update = 0;
3141 u64 trimmed = 0;
3142
3143 spin_lock(&space_info->lock);
3144 spin_lock(&block_group->lock);
3145 if (!block_group->ro) {
3146 block_group->reserved += reserved_bytes;
3147 space_info->bytes_reserved += reserved_bytes;
3148 update = 1;
3149 }
3150 spin_unlock(&block_group->lock);
3151 spin_unlock(&space_info->lock);
3152
3153 ret = btrfs_discard_extent(fs_info->extent_root,
3154 start, bytes, &trimmed);
3155 if (!ret)
3156 *total_trimmed += trimmed;
3157
3158 mutex_lock(&ctl->cache_writeout_mutex);
3159 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
3160 list_del(&trim_entry->list);
3161 mutex_unlock(&ctl->cache_writeout_mutex);
3162
3163 if (update) {
3164 spin_lock(&space_info->lock);
3165 spin_lock(&block_group->lock);
3166 if (block_group->ro)
3167 space_info->bytes_readonly += reserved_bytes;
3168 block_group->reserved -= reserved_bytes;
3169 space_info->bytes_reserved -= reserved_bytes;
3170 spin_unlock(&space_info->lock);
3171 spin_unlock(&block_group->lock);
3172 }
3173
3174 return ret;
3175 }
3176
3177 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
3178 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
3179 {
3180 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3181 struct btrfs_free_space *entry;
3182 struct rb_node *node;
3183 int ret = 0;
3184 u64 extent_start;
3185 u64 extent_bytes;
3186 u64 bytes;
3187
3188 while (start < end) {
3189 struct btrfs_trim_range trim_entry;
3190
3191 mutex_lock(&ctl->cache_writeout_mutex);
3192 spin_lock(&ctl->tree_lock);
3193
3194 if (ctl->free_space < minlen) {
3195 spin_unlock(&ctl->tree_lock);
3196 mutex_unlock(&ctl->cache_writeout_mutex);
3197 break;
3198 }
3199
3200 entry = tree_search_offset(ctl, start, 0, 1);
3201 if (!entry) {
3202 spin_unlock(&ctl->tree_lock);
3203 mutex_unlock(&ctl->cache_writeout_mutex);
3204 break;
3205 }
3206
3207 /* skip bitmaps */
3208 while (entry->bitmap) {
3209 node = rb_next(&entry->offset_index);
3210 if (!node) {
3211 spin_unlock(&ctl->tree_lock);
3212 mutex_unlock(&ctl->cache_writeout_mutex);
3213 goto out;
3214 }
3215 entry = rb_entry(node, struct btrfs_free_space,
3216 offset_index);
3217 }
3218
3219 if (entry->offset >= end) {
3220 spin_unlock(&ctl->tree_lock);
3221 mutex_unlock(&ctl->cache_writeout_mutex);
3222 break;
3223 }
3224
3225 extent_start = entry->offset;
3226 extent_bytes = entry->bytes;
3227 start = max(start, extent_start);
3228 bytes = min(extent_start + extent_bytes, end) - start;
3229 if (bytes < minlen) {
3230 spin_unlock(&ctl->tree_lock);
3231 mutex_unlock(&ctl->cache_writeout_mutex);
3232 goto next;
3233 }
3234
3235 unlink_free_space(ctl, entry);
3236 kmem_cache_free(btrfs_free_space_cachep, entry);
3237
3238 spin_unlock(&ctl->tree_lock);
3239 trim_entry.start = extent_start;
3240 trim_entry.bytes = extent_bytes;
3241 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3242 mutex_unlock(&ctl->cache_writeout_mutex);
3243
3244 ret = do_trimming(block_group, total_trimmed, start, bytes,
3245 extent_start, extent_bytes, &trim_entry);
3246 if (ret)
3247 break;
3248 next:
3249 start += bytes;
3250
3251 if (fatal_signal_pending(current)) {
3252 ret = -ERESTARTSYS;
3253 break;
3254 }
3255
3256 cond_resched();
3257 }
3258 out:
3259 return ret;
3260 }
3261
3262 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
3263 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
3264 {
3265 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3266 struct btrfs_free_space *entry;
3267 int ret = 0;
3268 int ret2;
3269 u64 bytes;
3270 u64 offset = offset_to_bitmap(ctl, start);
3271
3272 while (offset < end) {
3273 bool next_bitmap = false;
3274 struct btrfs_trim_range trim_entry;
3275
3276 mutex_lock(&ctl->cache_writeout_mutex);
3277 spin_lock(&ctl->tree_lock);
3278
3279 if (ctl->free_space < minlen) {
3280 spin_unlock(&ctl->tree_lock);
3281 mutex_unlock(&ctl->cache_writeout_mutex);
3282 break;
3283 }
3284
3285 entry = tree_search_offset(ctl, offset, 1, 0);
3286 if (!entry) {
3287 spin_unlock(&ctl->tree_lock);
3288 mutex_unlock(&ctl->cache_writeout_mutex);
3289 next_bitmap = true;
3290 goto next;
3291 }
3292
3293 bytes = minlen;
3294 ret2 = search_bitmap(ctl, entry, &start, &bytes, false);
3295 if (ret2 || start >= end) {
3296 spin_unlock(&ctl->tree_lock);
3297 mutex_unlock(&ctl->cache_writeout_mutex);
3298 next_bitmap = true;
3299 goto next;
3300 }
3301
3302 bytes = min(bytes, end - start);
3303 if (bytes < minlen) {
3304 spin_unlock(&ctl->tree_lock);
3305 mutex_unlock(&ctl->cache_writeout_mutex);
3306 goto next;
3307 }
3308
3309 bitmap_clear_bits(ctl, entry, start, bytes);
3310 if (entry->bytes == 0)
3311 free_bitmap(ctl, entry);
3312
3313 spin_unlock(&ctl->tree_lock);
3314 trim_entry.start = start;
3315 trim_entry.bytes = bytes;
3316 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3317 mutex_unlock(&ctl->cache_writeout_mutex);
3318
3319 ret = do_trimming(block_group, total_trimmed, start, bytes,
3320 start, bytes, &trim_entry);
3321 if (ret)
3322 break;
3323 next:
3324 if (next_bitmap) {
3325 offset += BITS_PER_BITMAP * ctl->unit;
3326 } else {
3327 start += bytes;
3328 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
3329 offset += BITS_PER_BITMAP * ctl->unit;
3330 }
3331
3332 if (fatal_signal_pending(current)) {
3333 ret = -ERESTARTSYS;
3334 break;
3335 }
3336
3337 cond_resched();
3338 }
3339
3340 return ret;
3341 }
3342
3343 void btrfs_get_block_group_trimming(struct btrfs_block_group_cache *cache)
3344 {
3345 atomic_inc(&cache->trimming);
3346 }
3347
3348 void btrfs_put_block_group_trimming(struct btrfs_block_group_cache *block_group)
3349 {
3350 struct extent_map_tree *em_tree;
3351 struct extent_map *em;
3352 bool cleanup;
3353
3354 spin_lock(&block_group->lock);
3355 cleanup = (atomic_dec_and_test(&block_group->trimming) &&
3356 block_group->removed);
3357 spin_unlock(&block_group->lock);
3358
3359 if (cleanup) {
3360 lock_chunks(block_group->fs_info->chunk_root);
3361 em_tree = &block_group->fs_info->mapping_tree.map_tree;
3362 write_lock(&em_tree->lock);
3363 em = lookup_extent_mapping(em_tree, block_group->key.objectid,
3364 1);
3365 BUG_ON(!em); /* logic error, can't happen */
3366 /*
3367 * remove_extent_mapping() will delete us from the pinned_chunks
3368 * list, which is protected by the chunk mutex.
3369 */
3370 remove_extent_mapping(em_tree, em);
3371 write_unlock(&em_tree->lock);
3372 unlock_chunks(block_group->fs_info->chunk_root);
3373
3374 /* once for us and once for the tree */
3375 free_extent_map(em);
3376 free_extent_map(em);
3377
3378 /*
3379 * We've left one free space entry and other tasks trimming
3380 * this block group have left 1 entry each one. Free them.
3381 */
3382 __btrfs_remove_free_space_cache(block_group->free_space_ctl);
3383 }
3384 }
3385
3386 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
3387 u64 *trimmed, u64 start, u64 end, u64 minlen)
3388 {
3389 int ret;
3390
3391 *trimmed = 0;
3392
3393 spin_lock(&block_group->lock);
3394 if (block_group->removed) {
3395 spin_unlock(&block_group->lock);
3396 return 0;
3397 }
3398 btrfs_get_block_group_trimming(block_group);
3399 spin_unlock(&block_group->lock);
3400
3401 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
3402 if (ret)
3403 goto out;
3404
3405 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
3406 out:
3407 btrfs_put_block_group_trimming(block_group);
3408 return ret;
3409 }
3410
3411 /*
3412 * Find the left-most item in the cache tree, and then return the
3413 * smallest inode number in the item.
3414 *
3415 * Note: the returned inode number may not be the smallest one in
3416 * the tree, if the left-most item is a bitmap.
3417 */
3418 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
3419 {
3420 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
3421 struct btrfs_free_space *entry = NULL;
3422 u64 ino = 0;
3423
3424 spin_lock(&ctl->tree_lock);
3425
3426 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
3427 goto out;
3428
3429 entry = rb_entry(rb_first(&ctl->free_space_offset),
3430 struct btrfs_free_space, offset_index);
3431
3432 if (!entry->bitmap) {
3433 ino = entry->offset;
3434
3435 unlink_free_space(ctl, entry);
3436 entry->offset++;
3437 entry->bytes--;
3438 if (!entry->bytes)
3439 kmem_cache_free(btrfs_free_space_cachep, entry);
3440 else
3441 link_free_space(ctl, entry);
3442 } else {
3443 u64 offset = 0;
3444 u64 count = 1;
3445 int ret;
3446
3447 ret = search_bitmap(ctl, entry, &offset, &count, true);
3448 /* Logic error; Should be empty if it can't find anything */
3449 ASSERT(!ret);
3450
3451 ino = offset;
3452 bitmap_clear_bits(ctl, entry, offset, 1);
3453 if (entry->bytes == 0)
3454 free_bitmap(ctl, entry);
3455 }
3456 out:
3457 spin_unlock(&ctl->tree_lock);
3458
3459 return ino;
3460 }
3461
3462 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
3463 struct btrfs_path *path)
3464 {
3465 struct inode *inode = NULL;
3466
3467 spin_lock(&root->ino_cache_lock);
3468 if (root->ino_cache_inode)
3469 inode = igrab(root->ino_cache_inode);
3470 spin_unlock(&root->ino_cache_lock);
3471 if (inode)
3472 return inode;
3473
3474 inode = __lookup_free_space_inode(root, path, 0);
3475 if (IS_ERR(inode))
3476 return inode;
3477
3478 spin_lock(&root->ino_cache_lock);
3479 if (!btrfs_fs_closing(root->fs_info))
3480 root->ino_cache_inode = igrab(inode);
3481 spin_unlock(&root->ino_cache_lock);
3482
3483 return inode;
3484 }
3485
3486 int create_free_ino_inode(struct btrfs_root *root,
3487 struct btrfs_trans_handle *trans,
3488 struct btrfs_path *path)
3489 {
3490 return __create_free_space_inode(root, trans, path,
3491 BTRFS_FREE_INO_OBJECTID, 0);
3492 }
3493
3494 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3495 {
3496 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
3497 struct btrfs_path *path;
3498 struct inode *inode;
3499 int ret = 0;
3500 u64 root_gen = btrfs_root_generation(&root->root_item);
3501
3502 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
3503 return 0;
3504
3505 /*
3506 * If we're unmounting then just return, since this does a search on the
3507 * normal root and not the commit root and we could deadlock.
3508 */
3509 if (btrfs_fs_closing(fs_info))
3510 return 0;
3511
3512 path = btrfs_alloc_path();
3513 if (!path)
3514 return 0;
3515
3516 inode = lookup_free_ino_inode(root, path);
3517 if (IS_ERR(inode))
3518 goto out;
3519
3520 if (root_gen != BTRFS_I(inode)->generation)
3521 goto out_put;
3522
3523 ret = __load_free_space_cache(root, inode, ctl, path, 0);
3524
3525 if (ret < 0)
3526 btrfs_err(fs_info,
3527 "failed to load free ino cache for root %llu",
3528 root->root_key.objectid);
3529 out_put:
3530 iput(inode);
3531 out:
3532 btrfs_free_path(path);
3533 return ret;
3534 }
3535
3536 int btrfs_write_out_ino_cache(struct btrfs_root *root,
3537 struct btrfs_trans_handle *trans,
3538 struct btrfs_path *path,
3539 struct inode *inode)
3540 {
3541 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
3542 int ret;
3543 struct btrfs_io_ctl io_ctl;
3544 bool release_metadata = true;
3545
3546 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
3547 return 0;
3548
3549 memset(&io_ctl, 0, sizeof(io_ctl));
3550 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, &io_ctl,
3551 trans, path, 0);
3552 if (!ret) {
3553 /*
3554 * At this point writepages() didn't error out, so our metadata
3555 * reservation is released when the writeback finishes, at
3556 * inode.c:btrfs_finish_ordered_io(), regardless of it finishing
3557 * with or without an error.
3558 */
3559 release_metadata = false;
3560 ret = btrfs_wait_cache_io(root, trans, NULL, &io_ctl, path, 0);
3561 }
3562
3563 if (ret) {
3564 if (release_metadata)
3565 btrfs_delalloc_release_metadata(inode, inode->i_size);
3566 #ifdef DEBUG
3567 btrfs_err(root->fs_info,
3568 "failed to write free ino cache for root %llu",
3569 root->root_key.objectid);
3570 #endif
3571 }
3572
3573 return ret;
3574 }
3575
3576 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3577 /*
3578 * Use this if you need to make a bitmap or extent entry specifically, it
3579 * doesn't do any of the merging that add_free_space does, this acts a lot like
3580 * how the free space cache loading stuff works, so you can get really weird
3581 * configurations.
3582 */
3583 int test_add_free_space_entry(struct btrfs_block_group_cache *cache,
3584 u64 offset, u64 bytes, bool bitmap)
3585 {
3586 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3587 struct btrfs_free_space *info = NULL, *bitmap_info;
3588 void *map = NULL;
3589 u64 bytes_added;
3590 int ret;
3591
3592 again:
3593 if (!info) {
3594 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
3595 if (!info)
3596 return -ENOMEM;
3597 }
3598
3599 if (!bitmap) {
3600 spin_lock(&ctl->tree_lock);
3601 info->offset = offset;
3602 info->bytes = bytes;
3603 info->max_extent_size = 0;
3604 ret = link_free_space(ctl, info);
3605 spin_unlock(&ctl->tree_lock);
3606 if (ret)
3607 kmem_cache_free(btrfs_free_space_cachep, info);
3608 return ret;
3609 }
3610
3611 if (!map) {
3612 map = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
3613 if (!map) {
3614 kmem_cache_free(btrfs_free_space_cachep, info);
3615 return -ENOMEM;
3616 }
3617 }
3618
3619 spin_lock(&ctl->tree_lock);
3620 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3621 1, 0);
3622 if (!bitmap_info) {
3623 info->bitmap = map;
3624 map = NULL;
3625 add_new_bitmap(ctl, info, offset);
3626 bitmap_info = info;
3627 info = NULL;
3628 }
3629
3630 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
3631
3632 bytes -= bytes_added;
3633 offset += bytes_added;
3634 spin_unlock(&ctl->tree_lock);
3635
3636 if (bytes)
3637 goto again;
3638
3639 if (info)
3640 kmem_cache_free(btrfs_free_space_cachep, info);
3641 if (map)
3642 kfree(map);
3643 return 0;
3644 }
3645
3646 /*
3647 * Checks to see if the given range is in the free space cache. This is really
3648 * just used to check the absence of space, so if there is free space in the
3649 * range at all we will return 1.
3650 */
3651 int test_check_exists(struct btrfs_block_group_cache *cache,
3652 u64 offset, u64 bytes)
3653 {
3654 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3655 struct btrfs_free_space *info;
3656 int ret = 0;
3657
3658 spin_lock(&ctl->tree_lock);
3659 info = tree_search_offset(ctl, offset, 0, 0);
3660 if (!info) {
3661 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3662 1, 0);
3663 if (!info)
3664 goto out;
3665 }
3666
3667 have_info:
3668 if (info->bitmap) {
3669 u64 bit_off, bit_bytes;
3670 struct rb_node *n;
3671 struct btrfs_free_space *tmp;
3672
3673 bit_off = offset;
3674 bit_bytes = ctl->unit;
3675 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false);
3676 if (!ret) {
3677 if (bit_off == offset) {
3678 ret = 1;
3679 goto out;
3680 } else if (bit_off > offset &&
3681 offset + bytes > bit_off) {
3682 ret = 1;
3683 goto out;
3684 }
3685 }
3686
3687 n = rb_prev(&info->offset_index);
3688 while (n) {
3689 tmp = rb_entry(n, struct btrfs_free_space,
3690 offset_index);
3691 if (tmp->offset + tmp->bytes < offset)
3692 break;
3693 if (offset + bytes < tmp->offset) {
3694 n = rb_prev(&info->offset_index);
3695 continue;
3696 }
3697 info = tmp;
3698 goto have_info;
3699 }
3700
3701 n = rb_next(&info->offset_index);
3702 while (n) {
3703 tmp = rb_entry(n, struct btrfs_free_space,
3704 offset_index);
3705 if (offset + bytes < tmp->offset)
3706 break;
3707 if (tmp->offset + tmp->bytes < offset) {
3708 n = rb_next(&info->offset_index);
3709 continue;
3710 }
3711 info = tmp;
3712 goto have_info;
3713 }
3714
3715 ret = 0;
3716 goto out;
3717 }
3718
3719 if (info->offset == offset) {
3720 ret = 1;
3721 goto out;
3722 }
3723
3724 if (offset > info->offset && offset < info->offset + info->bytes)
3725 ret = 1;
3726 out:
3727 spin_unlock(&ctl->tree_lock);
3728 return ret;
3729 }
3730 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */
Linux with added FPC-III support