?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
staging: rtl8188eu: rename HalSetBrateCfg() - style
[linux/fpc-iii.git] / fs / btrfs / free-space-cache.c
blob0adf38b00fa0477f7a36d2570246213b9442ab23
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2008 Red Hat. All rights reserved.
4 */
5
6 #include <linux/pagemap.h>
7 #include <linux/sched.h>
8 #include <linux/sched/signal.h>
9 #include <linux/slab.h>
10 #include <linux/math64.h>
11 #include <linux/ratelimit.h>
12 #include <linux/error-injection.h>
13 #include "ctree.h"
14 #include "free-space-cache.h"
15 #include "transaction.h"
16 #include "disk-io.h"
17 #include "extent_io.h"
18 #include "inode-map.h"
19 #include "volumes.h"
20
21 #define BITS_PER_BITMAP (PAGE_SIZE * 8UL)
22 #define MAX_CACHE_BYTES_PER_GIG SZ_32K
23
24 struct btrfs_trim_range {
25 u64 start;
26 u64 bytes;
27 struct list_head list;
28 };
29
30 static int link_free_space(struct btrfs_free_space_ctl *ctl,
31 struct btrfs_free_space *info);
32 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
33 struct btrfs_free_space *info);
34 static int btrfs_wait_cache_io_root(struct btrfs_root *root,
35 struct btrfs_trans_handle *trans,
36 struct btrfs_io_ctl *io_ctl,
37 struct btrfs_path *path);
38
39 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
40 struct btrfs_path *path,
41 u64 offset)
42 {
43 struct btrfs_fs_info *fs_info = root->fs_info;
44 struct btrfs_key key;
45 struct btrfs_key location;
46 struct btrfs_disk_key disk_key;
47 struct btrfs_free_space_header *header;
48 struct extent_buffer *leaf;
49 struct inode *inode = NULL;
50 int ret;
51
52 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
53 key.offset = offset;
54 key.type = 0;
55
56 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
57 if (ret < 0)
58 return ERR_PTR(ret);
59 if (ret > 0) {
60 btrfs_release_path(path);
61 return ERR_PTR(-ENOENT);
62 }
63
64 leaf = path->nodes[0];
65 header = btrfs_item_ptr(leaf, path->slots[0],
66 struct btrfs_free_space_header);
67 btrfs_free_space_key(leaf, header, &disk_key);
68 btrfs_disk_key_to_cpu(&location, &disk_key);
69 btrfs_release_path(path);
70
71 inode = btrfs_iget(fs_info->sb, &location, root, NULL);
72 if (IS_ERR(inode))
73 return inode;
74
75 mapping_set_gfp_mask(inode->i_mapping,
76 mapping_gfp_constraint(inode->i_mapping,
77 ~(__GFP_FS | __GFP_HIGHMEM)));
78
79 return inode;
80 }
81
82 struct inode *lookup_free_space_inode(struct btrfs_fs_info *fs_info,
83 struct btrfs_block_group_cache
84 *block_group, struct btrfs_path *path)
85 {
86 struct inode *inode = NULL;
87 u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
88
89 spin_lock(&block_group->lock);
90 if (block_group->inode)
91 inode = igrab(block_group->inode);
92 spin_unlock(&block_group->lock);
93 if (inode)
94 return inode;
95
96 inode = __lookup_free_space_inode(fs_info->tree_root, path,
97 block_group->key.objectid);
98 if (IS_ERR(inode))
99 return inode;
100
101 spin_lock(&block_group->lock);
102 if (!((BTRFS_I(inode)->flags & flags) == flags)) {
103 btrfs_info(fs_info, "Old style space inode found, converting.");
104 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
105 BTRFS_INODE_NODATACOW;
106 block_group->disk_cache_state = BTRFS_DC_CLEAR;
107 }
108
109 if (!block_group->iref) {
110 block_group->inode = igrab(inode);
111 block_group->iref = 1;
112 }
113 spin_unlock(&block_group->lock);
114
115 return inode;
116 }
117
118 static int __create_free_space_inode(struct btrfs_root *root,
119 struct btrfs_trans_handle *trans,
120 struct btrfs_path *path,
121 u64 ino, u64 offset)
122 {
123 struct btrfs_key key;
124 struct btrfs_disk_key disk_key;
125 struct btrfs_free_space_header *header;
126 struct btrfs_inode_item *inode_item;
127 struct extent_buffer *leaf;
128 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
129 int ret;
130
131 ret = btrfs_insert_empty_inode(trans, root, path, ino);
132 if (ret)
133 return ret;
134
135 /* We inline crc's for the free disk space cache */
136 if (ino != BTRFS_FREE_INO_OBJECTID)
137 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
138
139 leaf = path->nodes[0];
140 inode_item = btrfs_item_ptr(leaf, path->slots[0],
141 struct btrfs_inode_item);
142 btrfs_item_key(leaf, &disk_key, path->slots[0]);
143 memzero_extent_buffer(leaf, (unsigned long)inode_item,
144 sizeof(*inode_item));
145 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
146 btrfs_set_inode_size(leaf, inode_item, 0);
147 btrfs_set_inode_nbytes(leaf, inode_item, 0);
148 btrfs_set_inode_uid(leaf, inode_item, 0);
149 btrfs_set_inode_gid(leaf, inode_item, 0);
150 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
151 btrfs_set_inode_flags(leaf, inode_item, flags);
152 btrfs_set_inode_nlink(leaf, inode_item, 1);
153 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
154 btrfs_set_inode_block_group(leaf, inode_item, offset);
155 btrfs_mark_buffer_dirty(leaf);
156 btrfs_release_path(path);
157
158 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
159 key.offset = offset;
160 key.type = 0;
161 ret = btrfs_insert_empty_item(trans, root, path, &key,
162 sizeof(struct btrfs_free_space_header));
163 if (ret < 0) {
164 btrfs_release_path(path);
165 return ret;
166 }
167
168 leaf = path->nodes[0];
169 header = btrfs_item_ptr(leaf, path->slots[0],
170 struct btrfs_free_space_header);
171 memzero_extent_buffer(leaf, (unsigned long)header, sizeof(*header));
172 btrfs_set_free_space_key(leaf, header, &disk_key);
173 btrfs_mark_buffer_dirty(leaf);
174 btrfs_release_path(path);
175
176 return 0;
177 }
178
179 int create_free_space_inode(struct btrfs_fs_info *fs_info,
180 struct btrfs_trans_handle *trans,
181 struct btrfs_block_group_cache *block_group,
182 struct btrfs_path *path)
183 {
184 int ret;
185 u64 ino;
186
187 ret = btrfs_find_free_objectid(fs_info->tree_root, &ino);
188 if (ret < 0)
189 return ret;
190
191 return __create_free_space_inode(fs_info->tree_root, trans, path, ino,
192 block_group->key.objectid);
193 }
194
195 int btrfs_check_trunc_cache_free_space(struct btrfs_fs_info *fs_info,
196 struct btrfs_block_rsv *rsv)
197 {
198 u64 needed_bytes;
199 int ret;
200
201 /* 1 for slack space, 1 for updating the inode */
202 needed_bytes = btrfs_calc_trunc_metadata_size(fs_info, 1) +
203 btrfs_calc_trans_metadata_size(fs_info, 1);
204
205 spin_lock(&rsv->lock);
206 if (rsv->reserved < needed_bytes)
207 ret = -ENOSPC;
208 else
209 ret = 0;
210 spin_unlock(&rsv->lock);
211 return ret;
212 }
213
214 int btrfs_truncate_free_space_cache(struct btrfs_trans_handle *trans,
215 struct btrfs_block_group_cache *block_group,
216 struct inode *inode)
217 {
218 struct btrfs_root *root = BTRFS_I(inode)->root;
219 int ret = 0;
220 bool locked = false;
221
222 if (block_group) {
223 struct btrfs_path *path = btrfs_alloc_path();
224
225 if (!path) {
226 ret = -ENOMEM;
227 goto fail;
228 }
229 locked = true;
230 mutex_lock(&trans->transaction->cache_write_mutex);
231 if (!list_empty(&block_group->io_list)) {
232 list_del_init(&block_group->io_list);
233
234 btrfs_wait_cache_io(trans, block_group, path);
235 btrfs_put_block_group(block_group);
236 }
237
238 /*
239 * now that we've truncated the cache away, its no longer
240 * setup or written
241 */
242 spin_lock(&block_group->lock);
243 block_group->disk_cache_state = BTRFS_DC_CLEAR;
244 spin_unlock(&block_group->lock);
245 btrfs_free_path(path);
246 }
247
248 btrfs_i_size_write(BTRFS_I(inode), 0);
249 truncate_pagecache(inode, 0);
250
251 /*
252 * We skip the throttling logic for free space cache inodes, so we don't
253 * need to check for -EAGAIN.
254 */
255 ret = btrfs_truncate_inode_items(trans, root, inode,
256 0, BTRFS_EXTENT_DATA_KEY);
257 if (ret)
258 goto fail;
259
260 ret = btrfs_update_inode(trans, root, inode);
261
262 fail:
263 if (locked)
264 mutex_unlock(&trans->transaction->cache_write_mutex);
265 if (ret)
266 btrfs_abort_transaction(trans, ret);
267
268 return ret;
269 }
270
271 static void readahead_cache(struct inode *inode)
272 {
273 struct file_ra_state *ra;
274 unsigned long last_index;
275
276 ra = kzalloc(sizeof(*ra), GFP_NOFS);
277 if (!ra)
278 return;
279
280 file_ra_state_init(ra, inode->i_mapping);
281 last_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
282
283 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
284
285 kfree(ra);
286 }
287
288 static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode,
289 int write)
290 {
291 int num_pages;
292 int check_crcs = 0;
293
294 num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
295
296 if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FREE_INO_OBJECTID)
297 check_crcs = 1;
298
299 /* Make sure we can fit our crcs and generation into the first page */
300 if (write && check_crcs &&
301 (num_pages * sizeof(u32) + sizeof(u64)) > PAGE_SIZE)
302 return -ENOSPC;
303
304 memset(io_ctl, 0, sizeof(struct btrfs_io_ctl));
305
306 io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS);
307 if (!io_ctl->pages)
308 return -ENOMEM;
309
310 io_ctl->num_pages = num_pages;
311 io_ctl->fs_info = btrfs_sb(inode->i_sb);
312 io_ctl->check_crcs = check_crcs;
313 io_ctl->inode = inode;
314
315 return 0;
316 }
317 ALLOW_ERROR_INJECTION(io_ctl_init, ERRNO);
318
319 static void io_ctl_free(struct btrfs_io_ctl *io_ctl)
320 {
321 kfree(io_ctl->pages);
322 io_ctl->pages = NULL;
323 }
324
325 static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl)
326 {
327 if (io_ctl->cur) {
328 io_ctl->cur = NULL;
329 io_ctl->orig = NULL;
330 }
331 }
332
333 static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear)
334 {
335 ASSERT(io_ctl->index < io_ctl->num_pages);
336 io_ctl->page = io_ctl->pages[io_ctl->index++];
337 io_ctl->cur = page_address(io_ctl->page);
338 io_ctl->orig = io_ctl->cur;
339 io_ctl->size = PAGE_SIZE;
340 if (clear)
341 clear_page(io_ctl->cur);
342 }
343
344 static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl)
345 {
346 int i;
347
348 io_ctl_unmap_page(io_ctl);
349
350 for (i = 0; i < io_ctl->num_pages; i++) {
351 if (io_ctl->pages[i]) {
352 ClearPageChecked(io_ctl->pages[i]);
353 unlock_page(io_ctl->pages[i]);
354 put_page(io_ctl->pages[i]);
355 }
356 }
357 }
358
359 static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, struct inode *inode,
360 int uptodate)
361 {
362 struct page *page;
363 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
364 int i;
365
366 for (i = 0; i < io_ctl->num_pages; i++) {
367 page = find_or_create_page(inode->i_mapping, i, mask);
368 if (!page) {
369 io_ctl_drop_pages(io_ctl);
370 return -ENOMEM;
371 }
372 io_ctl->pages[i] = page;
373 if (uptodate && !PageUptodate(page)) {
374 btrfs_readpage(NULL, page);
375 lock_page(page);
376 if (!PageUptodate(page)) {
377 btrfs_err(BTRFS_I(inode)->root->fs_info,
378 "error reading free space cache");
379 io_ctl_drop_pages(io_ctl);
380 return -EIO;
381 }
382 }
383 }
384
385 for (i = 0; i < io_ctl->num_pages; i++) {
386 clear_page_dirty_for_io(io_ctl->pages[i]);
387 set_page_extent_mapped(io_ctl->pages[i]);
388 }
389
390 return 0;
391 }
392
393 static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
394 {
395 __le64 *val;
396
397 io_ctl_map_page(io_ctl, 1);
398
399 /*
400 * Skip the csum areas. If we don't check crcs then we just have a
401 * 64bit chunk at the front of the first page.
402 */
403 if (io_ctl->check_crcs) {
404 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
405 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
406 } else {
407 io_ctl->cur += sizeof(u64);
408 io_ctl->size -= sizeof(u64) * 2;
409 }
410
411 val = io_ctl->cur;
412 *val = cpu_to_le64(generation);
413 io_ctl->cur += sizeof(u64);
414 }
415
416 static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
417 {
418 __le64 *gen;
419
420 /*
421 * Skip the crc area. If we don't check crcs then we just have a 64bit
422 * chunk at the front of the first page.
423 */
424 if (io_ctl->check_crcs) {
425 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
426 io_ctl->size -= sizeof(u64) +
427 (sizeof(u32) * io_ctl->num_pages);
428 } else {
429 io_ctl->cur += sizeof(u64);
430 io_ctl->size -= sizeof(u64) * 2;
431 }
432
433 gen = io_ctl->cur;
434 if (le64_to_cpu(*gen) != generation) {
435 btrfs_err_rl(io_ctl->fs_info,
436 "space cache generation (%llu) does not match inode (%llu)",
437 *gen, generation);
438 io_ctl_unmap_page(io_ctl);
439 return -EIO;
440 }
441 io_ctl->cur += sizeof(u64);
442 return 0;
443 }
444
445 static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
446 {
447 u32 *tmp;
448 u32 crc = ~(u32)0;
449 unsigned offset = 0;
450
451 if (!io_ctl->check_crcs) {
452 io_ctl_unmap_page(io_ctl);
453 return;
454 }
455
456 if (index == 0)
457 offset = sizeof(u32) * io_ctl->num_pages;
458
459 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
460 PAGE_SIZE - offset);
461 btrfs_csum_final(crc, (u8 *)&crc);
462 io_ctl_unmap_page(io_ctl);
463 tmp = page_address(io_ctl->pages[0]);
464 tmp += index;
465 *tmp = crc;
466 }
467
468 static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
469 {
470 u32 *tmp, val;
471 u32 crc = ~(u32)0;
472 unsigned offset = 0;
473
474 if (!io_ctl->check_crcs) {
475 io_ctl_map_page(io_ctl, 0);
476 return 0;
477 }
478
479 if (index == 0)
480 offset = sizeof(u32) * io_ctl->num_pages;
481
482 tmp = page_address(io_ctl->pages[0]);
483 tmp += index;
484 val = *tmp;
485
486 io_ctl_map_page(io_ctl, 0);
487 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
488 PAGE_SIZE - offset);
489 btrfs_csum_final(crc, (u8 *)&crc);
490 if (val != crc) {
491 btrfs_err_rl(io_ctl->fs_info,
492 "csum mismatch on free space cache");
493 io_ctl_unmap_page(io_ctl);
494 return -EIO;
495 }
496
497 return 0;
498 }
499
500 static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
501 void *bitmap)
502 {
503 struct btrfs_free_space_entry *entry;
504
505 if (!io_ctl->cur)
506 return -ENOSPC;
507
508 entry = io_ctl->cur;
509 entry->offset = cpu_to_le64(offset);
510 entry->bytes = cpu_to_le64(bytes);
511 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
512 BTRFS_FREE_SPACE_EXTENT;
513 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
514 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
515
516 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
517 return 0;
518
519 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
520
521 /* No more pages to map */
522 if (io_ctl->index >= io_ctl->num_pages)
523 return 0;
524
525 /* map the next page */
526 io_ctl_map_page(io_ctl, 1);
527 return 0;
528 }
529
530 static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
531 {
532 if (!io_ctl->cur)
533 return -ENOSPC;
534
535 /*
536 * If we aren't at the start of the current page, unmap this one and
537 * map the next one if there is any left.
538 */
539 if (io_ctl->cur != io_ctl->orig) {
540 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
541 if (io_ctl->index >= io_ctl->num_pages)
542 return -ENOSPC;
543 io_ctl_map_page(io_ctl, 0);
544 }
545
546 copy_page(io_ctl->cur, bitmap);
547 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
548 if (io_ctl->index < io_ctl->num_pages)
549 io_ctl_map_page(io_ctl, 0);
550 return 0;
551 }
552
553 static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
554 {
555 /*
556 * If we're not on the boundary we know we've modified the page and we
557 * need to crc the page.
558 */
559 if (io_ctl->cur != io_ctl->orig)
560 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
561 else
562 io_ctl_unmap_page(io_ctl);
563
564 while (io_ctl->index < io_ctl->num_pages) {
565 io_ctl_map_page(io_ctl, 1);
566 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
567 }
568 }
569
570 static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
571 struct btrfs_free_space *entry, u8 *type)
572 {
573 struct btrfs_free_space_entry *e;
574 int ret;
575
576 if (!io_ctl->cur) {
577 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
578 if (ret)
579 return ret;
580 }
581
582 e = io_ctl->cur;
583 entry->offset = le64_to_cpu(e->offset);
584 entry->bytes = le64_to_cpu(e->bytes);
585 *type = e->type;
586 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
587 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
588
589 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
590 return 0;
591
592 io_ctl_unmap_page(io_ctl);
593
594 return 0;
595 }
596
597 static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
598 struct btrfs_free_space *entry)
599 {
600 int ret;
601
602 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
603 if (ret)
604 return ret;
605
606 copy_page(entry->bitmap, io_ctl->cur);
607 io_ctl_unmap_page(io_ctl);
608
609 return 0;
610 }
611
612 /*
613 * Since we attach pinned extents after the fact we can have contiguous sections
614 * of free space that are split up in entries. This poses a problem with the
615 * tree logging stuff since it could have allocated across what appears to be 2
616 * entries since we would have merged the entries when adding the pinned extents
617 * back to the free space cache. So run through the space cache that we just
618 * loaded and merge contiguous entries. This will make the log replay stuff not
619 * blow up and it will make for nicer allocator behavior.
620 */
621 static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
622 {
623 struct btrfs_free_space *e, *prev = NULL;
624 struct rb_node *n;
625
626 again:
627 spin_lock(&ctl->tree_lock);
628 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
629 e = rb_entry(n, struct btrfs_free_space, offset_index);
630 if (!prev)
631 goto next;
632 if (e->bitmap || prev->bitmap)
633 goto next;
634 if (prev->offset + prev->bytes == e->offset) {
635 unlink_free_space(ctl, prev);
636 unlink_free_space(ctl, e);
637 prev->bytes += e->bytes;
638 kmem_cache_free(btrfs_free_space_cachep, e);
639 link_free_space(ctl, prev);
640 prev = NULL;
641 spin_unlock(&ctl->tree_lock);
642 goto again;
643 }
644 next:
645 prev = e;
646 }
647 spin_unlock(&ctl->tree_lock);
648 }
649
650 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
651 struct btrfs_free_space_ctl *ctl,
652 struct btrfs_path *path, u64 offset)
653 {
654 struct btrfs_fs_info *fs_info = root->fs_info;
655 struct btrfs_free_space_header *header;
656 struct extent_buffer *leaf;
657 struct btrfs_io_ctl io_ctl;
658 struct btrfs_key key;
659 struct btrfs_free_space *e, *n;
660 LIST_HEAD(bitmaps);
661 u64 num_entries;
662 u64 num_bitmaps;
663 u64 generation;
664 u8 type;
665 int ret = 0;
666
667 /* Nothing in the space cache, goodbye */
668 if (!i_size_read(inode))
669 return 0;
670
671 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
672 key.offset = offset;
673 key.type = 0;
674
675 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
676 if (ret < 0)
677 return 0;
678 else if (ret > 0) {
679 btrfs_release_path(path);
680 return 0;
681 }
682
683 ret = -1;
684
685 leaf = path->nodes[0];
686 header = btrfs_item_ptr(leaf, path->slots[0],
687 struct btrfs_free_space_header);
688 num_entries = btrfs_free_space_entries(leaf, header);
689 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
690 generation = btrfs_free_space_generation(leaf, header);
691 btrfs_release_path(path);
692
693 if (!BTRFS_I(inode)->generation) {
694 btrfs_info(fs_info,
695 "the free space cache file (%llu) is invalid, skip it",
696 offset);
697 return 0;
698 }
699
700 if (BTRFS_I(inode)->generation != generation) {
701 btrfs_err(fs_info,
702 "free space inode generation (%llu) did not match free space cache generation (%llu)",
703 BTRFS_I(inode)->generation, generation);
704 return 0;
705 }
706
707 if (!num_entries)
708 return 0;
709
710 ret = io_ctl_init(&io_ctl, inode, 0);
711 if (ret)
712 return ret;
713
714 readahead_cache(inode);
715
716 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
717 if (ret)
718 goto out;
719
720 ret = io_ctl_check_crc(&io_ctl, 0);
721 if (ret)
722 goto free_cache;
723
724 ret = io_ctl_check_generation(&io_ctl, generation);
725 if (ret)
726 goto free_cache;
727
728 while (num_entries) {
729 e = kmem_cache_zalloc(btrfs_free_space_cachep,
730 GFP_NOFS);
731 if (!e)
732 goto free_cache;
733
734 ret = io_ctl_read_entry(&io_ctl, e, &type);
735 if (ret) {
736 kmem_cache_free(btrfs_free_space_cachep, e);
737 goto free_cache;
738 }
739
740 if (!e->bytes) {
741 kmem_cache_free(btrfs_free_space_cachep, e);
742 goto free_cache;
743 }
744
745 if (type == BTRFS_FREE_SPACE_EXTENT) {
746 spin_lock(&ctl->tree_lock);
747 ret = link_free_space(ctl, e);
748 spin_unlock(&ctl->tree_lock);
749 if (ret) {
750 btrfs_err(fs_info,
751 "Duplicate entries in free space cache, dumping");
752 kmem_cache_free(btrfs_free_space_cachep, e);
753 goto free_cache;
754 }
755 } else {
756 ASSERT(num_bitmaps);
757 num_bitmaps--;
758 e->bitmap = kzalloc(PAGE_SIZE, GFP_NOFS);
759 if (!e->bitmap) {
760 kmem_cache_free(
761 btrfs_free_space_cachep, e);
762 goto free_cache;
763 }
764 spin_lock(&ctl->tree_lock);
765 ret = link_free_space(ctl, e);
766 ctl->total_bitmaps++;
767 ctl->op->recalc_thresholds(ctl);
768 spin_unlock(&ctl->tree_lock);
769 if (ret) {
770 btrfs_err(fs_info,
771 "Duplicate entries in free space cache, dumping");
772 kmem_cache_free(btrfs_free_space_cachep, e);
773 goto free_cache;
774 }
775 list_add_tail(&e->list, &bitmaps);
776 }
777
778 num_entries--;
779 }
780
781 io_ctl_unmap_page(&io_ctl);
782
783 /*
784 * We add the bitmaps at the end of the entries in order that
785 * the bitmap entries are added to the cache.
786 */
787 list_for_each_entry_safe(e, n, &bitmaps, list) {
788 list_del_init(&e->list);
789 ret = io_ctl_read_bitmap(&io_ctl, e);
790 if (ret)
791 goto free_cache;
792 }
793
794 io_ctl_drop_pages(&io_ctl);
795 merge_space_tree(ctl);
796 ret = 1;
797 out:
798 io_ctl_free(&io_ctl);
799 return ret;
800 free_cache:
801 io_ctl_drop_pages(&io_ctl);
802 __btrfs_remove_free_space_cache(ctl);
803 goto out;
804 }
805
806 int load_free_space_cache(struct btrfs_fs_info *fs_info,
807 struct btrfs_block_group_cache *block_group)
808 {
809 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
810 struct inode *inode;
811 struct btrfs_path *path;
812 int ret = 0;
813 bool matched;
814 u64 used = btrfs_block_group_used(&block_group->item);
815
816 /*
817 * If this block group has been marked to be cleared for one reason or
818 * another then we can't trust the on disk cache, so just return.
819 */
820 spin_lock(&block_group->lock);
821 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
822 spin_unlock(&block_group->lock);
823 return 0;
824 }
825 spin_unlock(&block_group->lock);
826
827 path = btrfs_alloc_path();
828 if (!path)
829 return 0;
830 path->search_commit_root = 1;
831 path->skip_locking = 1;
832
833 inode = lookup_free_space_inode(fs_info, block_group, path);
834 if (IS_ERR(inode)) {
835 btrfs_free_path(path);
836 return 0;
837 }
838
839 /* We may have converted the inode and made the cache invalid. */
840 spin_lock(&block_group->lock);
841 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
842 spin_unlock(&block_group->lock);
843 btrfs_free_path(path);
844 goto out;
845 }
846 spin_unlock(&block_group->lock);
847
848 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
849 path, block_group->key.objectid);
850 btrfs_free_path(path);
851 if (ret <= 0)
852 goto out;
853
854 spin_lock(&ctl->tree_lock);
855 matched = (ctl->free_space == (block_group->key.offset - used -
856 block_group->bytes_super));
857 spin_unlock(&ctl->tree_lock);
858
859 if (!matched) {
860 __btrfs_remove_free_space_cache(ctl);
861 btrfs_warn(fs_info,
862 "block group %llu has wrong amount of free space",
863 block_group->key.objectid);
864 ret = -1;
865 }
866 out:
867 if (ret < 0) {
868 /* This cache is bogus, make sure it gets cleared */
869 spin_lock(&block_group->lock);
870 block_group->disk_cache_state = BTRFS_DC_CLEAR;
871 spin_unlock(&block_group->lock);
872 ret = 0;
873
874 btrfs_warn(fs_info,
875 "failed to load free space cache for block group %llu, rebuilding it now",
876 block_group->key.objectid);
877 }
878
879 iput(inode);
880 return ret;
881 }
882
883 static noinline_for_stack
884 int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
885 struct btrfs_free_space_ctl *ctl,
886 struct btrfs_block_group_cache *block_group,
887 int *entries, int *bitmaps,
888 struct list_head *bitmap_list)
889 {
890 int ret;
891 struct btrfs_free_cluster *cluster = NULL;
892 struct btrfs_free_cluster *cluster_locked = NULL;
893 struct rb_node *node = rb_first(&ctl->free_space_offset);
894 struct btrfs_trim_range *trim_entry;
895
896 /* Get the cluster for this block_group if it exists */
897 if (block_group && !list_empty(&block_group->cluster_list)) {
898 cluster = list_entry(block_group->cluster_list.next,
899 struct btrfs_free_cluster,
900 block_group_list);
901 }
902
903 if (!node && cluster) {
904 cluster_locked = cluster;
905 spin_lock(&cluster_locked->lock);
906 node = rb_first(&cluster->root);
907 cluster = NULL;
908 }
909
910 /* Write out the extent entries */
911 while (node) {
912 struct btrfs_free_space *e;
913
914 e = rb_entry(node, struct btrfs_free_space, offset_index);
915 *entries += 1;
916
917 ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
918 e->bitmap);
919 if (ret)
920 goto fail;
921
922 if (e->bitmap) {
923 list_add_tail(&e->list, bitmap_list);
924 *bitmaps += 1;
925 }
926 node = rb_next(node);
927 if (!node && cluster) {
928 node = rb_first(&cluster->root);
929 cluster_locked = cluster;
930 spin_lock(&cluster_locked->lock);
931 cluster = NULL;
932 }
933 }
934 if (cluster_locked) {
935 spin_unlock(&cluster_locked->lock);
936 cluster_locked = NULL;
937 }
938
939 /*
940 * Make sure we don't miss any range that was removed from our rbtree
941 * because trimming is running. Otherwise after a umount+mount (or crash
942 * after committing the transaction) we would leak free space and get
943 * an inconsistent free space cache report from fsck.
944 */
945 list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
946 ret = io_ctl_add_entry(io_ctl, trim_entry->start,
947 trim_entry->bytes, NULL);
948 if (ret)
949 goto fail;
950 *entries += 1;
951 }
952
953 return 0;
954 fail:
955 if (cluster_locked)
956 spin_unlock(&cluster_locked->lock);
957 return -ENOSPC;
958 }
959
960 static noinline_for_stack int
961 update_cache_item(struct btrfs_trans_handle *trans,
962 struct btrfs_root *root,
963 struct inode *inode,
964 struct btrfs_path *path, u64 offset,
965 int entries, int bitmaps)
966 {
967 struct btrfs_key key;
968 struct btrfs_free_space_header *header;
969 struct extent_buffer *leaf;
970 int ret;
971
972 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
973 key.offset = offset;
974 key.type = 0;
975
976 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
977 if (ret < 0) {
978 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
979 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL);
980 goto fail;
981 }
982 leaf = path->nodes[0];
983 if (ret > 0) {
984 struct btrfs_key found_key;
985 ASSERT(path->slots[0]);
986 path->slots[0]--;
987 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
988 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
989 found_key.offset != offset) {
990 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
991 inode->i_size - 1,
992 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
993 NULL);
994 btrfs_release_path(path);
995 goto fail;
996 }
997 }
998
999 BTRFS_I(inode)->generation = trans->transid;
1000 header = btrfs_item_ptr(leaf, path->slots[0],
1001 struct btrfs_free_space_header);
1002 btrfs_set_free_space_entries(leaf, header, entries);
1003 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1004 btrfs_set_free_space_generation(leaf, header, trans->transid);
1005 btrfs_mark_buffer_dirty(leaf);
1006 btrfs_release_path(path);
1007
1008 return 0;
1009
1010 fail:
1011 return -1;
1012 }
1013
1014 static noinline_for_stack int
1015 write_pinned_extent_entries(struct btrfs_fs_info *fs_info,
1016 struct btrfs_block_group_cache *block_group,
1017 struct btrfs_io_ctl *io_ctl,
1018 int *entries)
1019 {
1020 u64 start, extent_start, extent_end, len;
1021 struct extent_io_tree *unpin = NULL;
1022 int ret;
1023
1024 if (!block_group)
1025 return 0;
1026
1027 /*
1028 * We want to add any pinned extents to our free space cache
1029 * so we don't leak the space
1030 *
1031 * We shouldn't have switched the pinned extents yet so this is the
1032 * right one
1033 */
1034 unpin = fs_info->pinned_extents;
1035
1036 start = block_group->key.objectid;
1037
1038 while (start < block_group->key.objectid + block_group->key.offset) {
1039 ret = find_first_extent_bit(unpin, start,
1040 &extent_start, &extent_end,
1041 EXTENT_DIRTY, NULL);
1042 if (ret)
1043 return 0;
1044
1045 /* This pinned extent is out of our range */
1046 if (extent_start >= block_group->key.objectid +
1047 block_group->key.offset)
1048 return 0;
1049
1050 extent_start = max(extent_start, start);
1051 extent_end = min(block_group->key.objectid +
1052 block_group->key.offset, extent_end + 1);
1053 len = extent_end - extent_start;
1054
1055 *entries += 1;
1056 ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1057 if (ret)
1058 return -ENOSPC;
1059
1060 start = extent_end;
1061 }
1062
1063 return 0;
1064 }
1065
1066 static noinline_for_stack int
1067 write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1068 {
1069 struct btrfs_free_space *entry, *next;
1070 int ret;
1071
1072 /* Write out the bitmaps */
1073 list_for_each_entry_safe(entry, next, bitmap_list, list) {
1074 ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1075 if (ret)
1076 return -ENOSPC;
1077 list_del_init(&entry->list);
1078 }
1079
1080 return 0;
1081 }
1082
1083 static int flush_dirty_cache(struct inode *inode)
1084 {
1085 int ret;
1086
1087 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1088 if (ret)
1089 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1090 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL);
1091
1092 return ret;
1093 }
1094
1095 static void noinline_for_stack
1096 cleanup_bitmap_list(struct list_head *bitmap_list)
1097 {
1098 struct btrfs_free_space *entry, *next;
1099
1100 list_for_each_entry_safe(entry, next, bitmap_list, list)
1101 list_del_init(&entry->list);
1102 }
1103
1104 static void noinline_for_stack
1105 cleanup_write_cache_enospc(struct inode *inode,
1106 struct btrfs_io_ctl *io_ctl,
1107 struct extent_state **cached_state)
1108 {
1109 io_ctl_drop_pages(io_ctl);
1110 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1111 i_size_read(inode) - 1, cached_state);
1112 }
1113
1114 static int __btrfs_wait_cache_io(struct btrfs_root *root,
1115 struct btrfs_trans_handle *trans,
1116 struct btrfs_block_group_cache *block_group,
1117 struct btrfs_io_ctl *io_ctl,
1118 struct btrfs_path *path, u64 offset)
1119 {
1120 int ret;
1121 struct inode *inode = io_ctl->inode;
1122
1123 if (!inode)
1124 return 0;
1125
1126 /* Flush the dirty pages in the cache file. */
1127 ret = flush_dirty_cache(inode);
1128 if (ret)
1129 goto out;
1130
1131 /* Update the cache item to tell everyone this cache file is valid. */
1132 ret = update_cache_item(trans, root, inode, path, offset,
1133 io_ctl->entries, io_ctl->bitmaps);
1134 out:
1135 io_ctl_free(io_ctl);
1136 if (ret) {
1137 invalidate_inode_pages2(inode->i_mapping);
1138 BTRFS_I(inode)->generation = 0;
1139 if (block_group) {
1140 #ifdef DEBUG
1141 btrfs_err(root->fs_info,
1142 "failed to write free space cache for block group %llu",
1143 block_group->key.objectid);
1144 #endif
1145 }
1146 }
1147 btrfs_update_inode(trans, root, inode);
1148
1149 if (block_group) {
1150 /* the dirty list is protected by the dirty_bgs_lock */
1151 spin_lock(&trans->transaction->dirty_bgs_lock);
1152
1153 /* the disk_cache_state is protected by the block group lock */
1154 spin_lock(&block_group->lock);
1155
1156 /*
1157 * only mark this as written if we didn't get put back on
1158 * the dirty list while waiting for IO. Otherwise our
1159 * cache state won't be right, and we won't get written again
1160 */
1161 if (!ret && list_empty(&block_group->dirty_list))
1162 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1163 else if (ret)
1164 block_group->disk_cache_state = BTRFS_DC_ERROR;
1165
1166 spin_unlock(&block_group->lock);
1167 spin_unlock(&trans->transaction->dirty_bgs_lock);
1168 io_ctl->inode = NULL;
1169 iput(inode);
1170 }
1171
1172 return ret;
1173
1174 }
1175
1176 static int btrfs_wait_cache_io_root(struct btrfs_root *root,
1177 struct btrfs_trans_handle *trans,
1178 struct btrfs_io_ctl *io_ctl,
1179 struct btrfs_path *path)
1180 {
1181 return __btrfs_wait_cache_io(root, trans, NULL, io_ctl, path, 0);
1182 }
1183
1184 int btrfs_wait_cache_io(struct btrfs_trans_handle *trans,
1185 struct btrfs_block_group_cache *block_group,
1186 struct btrfs_path *path)
1187 {
1188 return __btrfs_wait_cache_io(block_group->fs_info->tree_root, trans,
1189 block_group, &block_group->io_ctl,
1190 path, block_group->key.objectid);
1191 }
1192
1193 /**
1194 * __btrfs_write_out_cache - write out cached info to an inode
1195 * @root - the root the inode belongs to
1196 * @ctl - the free space cache we are going to write out
1197 * @block_group - the block_group for this cache if it belongs to a block_group
1198 * @trans - the trans handle
1199 *
1200 * This function writes out a free space cache struct to disk for quick recovery
1201 * on mount. This will return 0 if it was successful in writing the cache out,
1202 * or an errno if it was not.
1203 */
1204 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
1205 struct btrfs_free_space_ctl *ctl,
1206 struct btrfs_block_group_cache *block_group,
1207 struct btrfs_io_ctl *io_ctl,
1208 struct btrfs_trans_handle *trans)
1209 {
1210 struct btrfs_fs_info *fs_info = root->fs_info;
1211 struct extent_state *cached_state = NULL;
1212 LIST_HEAD(bitmap_list);
1213 int entries = 0;
1214 int bitmaps = 0;
1215 int ret;
1216 int must_iput = 0;
1217
1218 if (!i_size_read(inode))
1219 return -EIO;
1220
1221 WARN_ON(io_ctl->pages);
1222 ret = io_ctl_init(io_ctl, inode, 1);
1223 if (ret)
1224 return ret;
1225
1226 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1227 down_write(&block_group->data_rwsem);
1228 spin_lock(&block_group->lock);
1229 if (block_group->delalloc_bytes) {
1230 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1231 spin_unlock(&block_group->lock);
1232 up_write(&block_group->data_rwsem);
1233 BTRFS_I(inode)->generation = 0;
1234 ret = 0;
1235 must_iput = 1;
1236 goto out;
1237 }
1238 spin_unlock(&block_group->lock);
1239 }
1240
1241 /* Lock all pages first so we can lock the extent safely. */
1242 ret = io_ctl_prepare_pages(io_ctl, inode, 0);
1243 if (ret)
1244 goto out_unlock;
1245
1246 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1247 &cached_state);
1248
1249 io_ctl_set_generation(io_ctl, trans->transid);
1250
1251 mutex_lock(&ctl->cache_writeout_mutex);
1252 /* Write out the extent entries in the free space cache */
1253 spin_lock(&ctl->tree_lock);
1254 ret = write_cache_extent_entries(io_ctl, ctl,
1255 block_group, &entries, &bitmaps,
1256 &bitmap_list);
1257 if (ret)
1258 goto out_nospc_locked;
1259
1260 /*
1261 * Some spaces that are freed in the current transaction are pinned,
1262 * they will be added into free space cache after the transaction is
1263 * committed, we shouldn't lose them.
1264 *
1265 * If this changes while we are working we'll get added back to
1266 * the dirty list and redo it. No locking needed
1267 */
1268 ret = write_pinned_extent_entries(fs_info, block_group,
1269 io_ctl, &entries);
1270 if (ret)
1271 goto out_nospc_locked;
1272
1273 /*
1274 * At last, we write out all the bitmaps and keep cache_writeout_mutex
1275 * locked while doing it because a concurrent trim can be manipulating
1276 * or freeing the bitmap.
1277 */
1278 ret = write_bitmap_entries(io_ctl, &bitmap_list);
1279 spin_unlock(&ctl->tree_lock);
1280 mutex_unlock(&ctl->cache_writeout_mutex);
1281 if (ret)
1282 goto out_nospc;
1283
1284 /* Zero out the rest of the pages just to make sure */
1285 io_ctl_zero_remaining_pages(io_ctl);
1286
1287 /* Everything is written out, now we dirty the pages in the file. */
1288 ret = btrfs_dirty_pages(inode, io_ctl->pages, io_ctl->num_pages, 0,
1289 i_size_read(inode), &cached_state);
1290 if (ret)
1291 goto out_nospc;
1292
1293 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1294 up_write(&block_group->data_rwsem);
1295 /*
1296 * Release the pages and unlock the extent, we will flush
1297 * them out later
1298 */
1299 io_ctl_drop_pages(io_ctl);
1300
1301 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1302 i_size_read(inode) - 1, &cached_state);
1303
1304 /*
1305 * at this point the pages are under IO and we're happy,
1306 * The caller is responsible for waiting on them and updating the
1307 * the cache and the inode
1308 */
1309 io_ctl->entries = entries;
1310 io_ctl->bitmaps = bitmaps;
1311
1312 ret = btrfs_fdatawrite_range(inode, 0, (u64)-1);
1313 if (ret)
1314 goto out;
1315
1316 return 0;
1317
1318 out:
1319 io_ctl->inode = NULL;
1320 io_ctl_free(io_ctl);
1321 if (ret) {
1322 invalidate_inode_pages2(inode->i_mapping);
1323 BTRFS_I(inode)->generation = 0;
1324 }
1325 btrfs_update_inode(trans, root, inode);
1326 if (must_iput)
1327 iput(inode);
1328 return ret;
1329
1330 out_nospc_locked:
1331 cleanup_bitmap_list(&bitmap_list);
1332 spin_unlock(&ctl->tree_lock);
1333 mutex_unlock(&ctl->cache_writeout_mutex);
1334
1335 out_nospc:
1336 cleanup_write_cache_enospc(inode, io_ctl, &cached_state);
1337
1338 out_unlock:
1339 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1340 up_write(&block_group->data_rwsem);
1341
1342 goto out;
1343 }
1344
1345 int btrfs_write_out_cache(struct btrfs_fs_info *fs_info,
1346 struct btrfs_trans_handle *trans,
1347 struct btrfs_block_group_cache *block_group,
1348 struct btrfs_path *path)
1349 {
1350 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1351 struct inode *inode;
1352 int ret = 0;
1353
1354 spin_lock(&block_group->lock);
1355 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1356 spin_unlock(&block_group->lock);
1357 return 0;
1358 }
1359 spin_unlock(&block_group->lock);
1360
1361 inode = lookup_free_space_inode(fs_info, block_group, path);
1362 if (IS_ERR(inode))
1363 return 0;
1364
1365 ret = __btrfs_write_out_cache(fs_info->tree_root, inode, ctl,
1366 block_group, &block_group->io_ctl, trans);
1367 if (ret) {
1368 #ifdef DEBUG
1369 btrfs_err(fs_info,
1370 "failed to write free space cache for block group %llu",
1371 block_group->key.objectid);
1372 #endif
1373 spin_lock(&block_group->lock);
1374 block_group->disk_cache_state = BTRFS_DC_ERROR;
1375 spin_unlock(&block_group->lock);
1376
1377 block_group->io_ctl.inode = NULL;
1378 iput(inode);
1379 }
1380
1381 /*
1382 * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1383 * to wait for IO and put the inode
1384 */
1385
1386 return ret;
1387 }
1388
1389 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1390 u64 offset)
1391 {
1392 ASSERT(offset >= bitmap_start);
1393 offset -= bitmap_start;
1394 return (unsigned long)(div_u64(offset, unit));
1395 }
1396
1397 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1398 {
1399 return (unsigned long)(div_u64(bytes, unit));
1400 }
1401
1402 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1403 u64 offset)
1404 {
1405 u64 bitmap_start;
1406 u64 bytes_per_bitmap;
1407
1408 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1409 bitmap_start = offset - ctl->start;
1410 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1411 bitmap_start *= bytes_per_bitmap;
1412 bitmap_start += ctl->start;
1413
1414 return bitmap_start;
1415 }
1416
1417 static int tree_insert_offset(struct rb_root *root, u64 offset,
1418 struct rb_node *node, int bitmap)
1419 {
1420 struct rb_node **p = &root->rb_node;
1421 struct rb_node *parent = NULL;
1422 struct btrfs_free_space *info;
1423
1424 while (*p) {
1425 parent = *p;
1426 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1427
1428 if (offset < info->offset) {
1429 p = &(*p)->rb_left;
1430 } else if (offset > info->offset) {
1431 p = &(*p)->rb_right;
1432 } else {
1433 /*
1434 * we could have a bitmap entry and an extent entry
1435 * share the same offset. If this is the case, we want
1436 * the extent entry to always be found first if we do a
1437 * linear search through the tree, since we want to have
1438 * the quickest allocation time, and allocating from an
1439 * extent is faster than allocating from a bitmap. So
1440 * if we're inserting a bitmap and we find an entry at
1441 * this offset, we want to go right, or after this entry
1442 * logically. If we are inserting an extent and we've
1443 * found a bitmap, we want to go left, or before
1444 * logically.
1445 */
1446 if (bitmap) {
1447 if (info->bitmap) {
1448 WARN_ON_ONCE(1);
1449 return -EEXIST;
1450 }
1451 p = &(*p)->rb_right;
1452 } else {
1453 if (!info->bitmap) {
1454 WARN_ON_ONCE(1);
1455 return -EEXIST;
1456 }
1457 p = &(*p)->rb_left;
1458 }
1459 }
1460 }
1461
1462 rb_link_node(node, parent, p);
1463 rb_insert_color(node, root);
1464
1465 return 0;
1466 }
1467
1468 /*
1469 * searches the tree for the given offset.
1470 *
1471 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1472 * want a section that has at least bytes size and comes at or after the given
1473 * offset.
1474 */
1475 static struct btrfs_free_space *
1476 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1477 u64 offset, int bitmap_only, int fuzzy)
1478 {
1479 struct rb_node *n = ctl->free_space_offset.rb_node;
1480 struct btrfs_free_space *entry, *prev = NULL;
1481
1482 /* find entry that is closest to the 'offset' */
1483 while (1) {
1484 if (!n) {
1485 entry = NULL;
1486 break;
1487 }
1488
1489 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1490 prev = entry;
1491
1492 if (offset < entry->offset)
1493 n = n->rb_left;
1494 else if (offset > entry->offset)
1495 n = n->rb_right;
1496 else
1497 break;
1498 }
1499
1500 if (bitmap_only) {
1501 if (!entry)
1502 return NULL;
1503 if (entry->bitmap)
1504 return entry;
1505
1506 /*
1507 * bitmap entry and extent entry may share same offset,
1508 * in that case, bitmap entry comes after extent entry.
1509 */
1510 n = rb_next(n);
1511 if (!n)
1512 return NULL;
1513 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1514 if (entry->offset != offset)
1515 return NULL;
1516
1517 WARN_ON(!entry->bitmap);
1518 return entry;
1519 } else if (entry) {
1520 if (entry->bitmap) {
1521 /*
1522 * if previous extent entry covers the offset,
1523 * we should return it instead of the bitmap entry
1524 */
1525 n = rb_prev(&entry->offset_index);
1526 if (n) {
1527 prev = rb_entry(n, struct btrfs_free_space,
1528 offset_index);
1529 if (!prev->bitmap &&
1530 prev->offset + prev->bytes > offset)
1531 entry = prev;
1532 }
1533 }
1534 return entry;
1535 }
1536
1537 if (!prev)
1538 return NULL;
1539
1540 /* find last entry before the 'offset' */
1541 entry = prev;
1542 if (entry->offset > offset) {
1543 n = rb_prev(&entry->offset_index);
1544 if (n) {
1545 entry = rb_entry(n, struct btrfs_free_space,
1546 offset_index);
1547 ASSERT(entry->offset <= offset);
1548 } else {
1549 if (fuzzy)
1550 return entry;
1551 else
1552 return NULL;
1553 }
1554 }
1555
1556 if (entry->bitmap) {
1557 n = rb_prev(&entry->offset_index);
1558 if (n) {
1559 prev = rb_entry(n, struct btrfs_free_space,
1560 offset_index);
1561 if (!prev->bitmap &&
1562 prev->offset + prev->bytes > offset)
1563 return prev;
1564 }
1565 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1566 return entry;
1567 } else if (entry->offset + entry->bytes > offset)
1568 return entry;
1569
1570 if (!fuzzy)
1571 return NULL;
1572
1573 while (1) {
1574 if (entry->bitmap) {
1575 if (entry->offset + BITS_PER_BITMAP *
1576 ctl->unit > offset)
1577 break;
1578 } else {
1579 if (entry->offset + entry->bytes > offset)
1580 break;
1581 }
1582
1583 n = rb_next(&entry->offset_index);
1584 if (!n)
1585 return NULL;
1586 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1587 }
1588 return entry;
1589 }
1590
1591 static inline void
1592 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1593 struct btrfs_free_space *info)
1594 {
1595 rb_erase(&info->offset_index, &ctl->free_space_offset);
1596 ctl->free_extents--;
1597 }
1598
1599 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1600 struct btrfs_free_space *info)
1601 {
1602 __unlink_free_space(ctl, info);
1603 ctl->free_space -= info->bytes;
1604 }
1605
1606 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1607 struct btrfs_free_space *info)
1608 {
1609 int ret = 0;
1610
1611 ASSERT(info->bytes || info->bitmap);
1612 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1613 &info->offset_index, (info->bitmap != NULL));
1614 if (ret)
1615 return ret;
1616
1617 ctl->free_space += info->bytes;
1618 ctl->free_extents++;
1619 return ret;
1620 }
1621
1622 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1623 {
1624 struct btrfs_block_group_cache *block_group = ctl->private;
1625 u64 max_bytes;
1626 u64 bitmap_bytes;
1627 u64 extent_bytes;
1628 u64 size = block_group->key.offset;
1629 u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1630 u64 max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1631
1632 max_bitmaps = max_t(u64, max_bitmaps, 1);
1633
1634 ASSERT(ctl->total_bitmaps <= max_bitmaps);
1635
1636 /*
1637 * The goal is to keep the total amount of memory used per 1gb of space
1638 * at or below 32k, so we need to adjust how much memory we allow to be
1639 * used by extent based free space tracking
1640 */
1641 if (size < SZ_1G)
1642 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1643 else
1644 max_bytes = MAX_CACHE_BYTES_PER_GIG * div_u64(size, SZ_1G);
1645
1646 /*
1647 * we want to account for 1 more bitmap than what we have so we can make
1648 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1649 * we add more bitmaps.
1650 */
1651 bitmap_bytes = (ctl->total_bitmaps + 1) * ctl->unit;
1652
1653 if (bitmap_bytes >= max_bytes) {
1654 ctl->extents_thresh = 0;
1655 return;
1656 }
1657
1658 /*
1659 * we want the extent entry threshold to always be at most 1/2 the max
1660 * bytes we can have, or whatever is less than that.
1661 */
1662 extent_bytes = max_bytes - bitmap_bytes;
1663 extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
1664
1665 ctl->extents_thresh =
1666 div_u64(extent_bytes, sizeof(struct btrfs_free_space));
1667 }
1668
1669 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1670 struct btrfs_free_space *info,
1671 u64 offset, u64 bytes)
1672 {
1673 unsigned long start, count;
1674
1675 start = offset_to_bit(info->offset, ctl->unit, offset);
1676 count = bytes_to_bits(bytes, ctl->unit);
1677 ASSERT(start + count <= BITS_PER_BITMAP);
1678
1679 bitmap_clear(info->bitmap, start, count);
1680
1681 info->bytes -= bytes;
1682 }
1683
1684 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1685 struct btrfs_free_space *info, u64 offset,
1686 u64 bytes)
1687 {
1688 __bitmap_clear_bits(ctl, info, offset, bytes);
1689 ctl->free_space -= bytes;
1690 }
1691
1692 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1693 struct btrfs_free_space *info, u64 offset,
1694 u64 bytes)
1695 {
1696 unsigned long start, count;
1697
1698 start = offset_to_bit(info->offset, ctl->unit, offset);
1699 count = bytes_to_bits(bytes, ctl->unit);
1700 ASSERT(start + count <= BITS_PER_BITMAP);
1701
1702 bitmap_set(info->bitmap, start, count);
1703
1704 info->bytes += bytes;
1705 ctl->free_space += bytes;
1706 }
1707
1708 /*
1709 * If we can not find suitable extent, we will use bytes to record
1710 * the size of the max extent.
1711 */
1712 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1713 struct btrfs_free_space *bitmap_info, u64 *offset,
1714 u64 *bytes, bool for_alloc)
1715 {
1716 unsigned long found_bits = 0;
1717 unsigned long max_bits = 0;
1718 unsigned long bits, i;
1719 unsigned long next_zero;
1720 unsigned long extent_bits;
1721
1722 /*
1723 * Skip searching the bitmap if we don't have a contiguous section that
1724 * is large enough for this allocation.
1725 */
1726 if (for_alloc &&
1727 bitmap_info->max_extent_size &&
1728 bitmap_info->max_extent_size < *bytes) {
1729 *bytes = bitmap_info->max_extent_size;
1730 return -1;
1731 }
1732
1733 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1734 max_t(u64, *offset, bitmap_info->offset));
1735 bits = bytes_to_bits(*bytes, ctl->unit);
1736
1737 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1738 if (for_alloc && bits == 1) {
1739 found_bits = 1;
1740 break;
1741 }
1742 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1743 BITS_PER_BITMAP, i);
1744 extent_bits = next_zero - i;
1745 if (extent_bits >= bits) {
1746 found_bits = extent_bits;
1747 break;
1748 } else if (extent_bits > max_bits) {
1749 max_bits = extent_bits;
1750 }
1751 i = next_zero;
1752 }
1753
1754 if (found_bits) {
1755 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1756 *bytes = (u64)(found_bits) * ctl->unit;
1757 return 0;
1758 }
1759
1760 *bytes = (u64)(max_bits) * ctl->unit;
1761 bitmap_info->max_extent_size = *bytes;
1762 return -1;
1763 }
1764
1765 /* Cache the size of the max extent in bytes */
1766 static struct btrfs_free_space *
1767 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1768 unsigned long align, u64 *max_extent_size)
1769 {
1770 struct btrfs_free_space *entry;
1771 struct rb_node *node;
1772 u64 tmp;
1773 u64 align_off;
1774 int ret;
1775
1776 if (!ctl->free_space_offset.rb_node)
1777 goto out;
1778
1779 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1780 if (!entry)
1781 goto out;
1782
1783 for (node = &entry->offset_index; node; node = rb_next(node)) {
1784 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1785 if (entry->bytes < *bytes) {
1786 if (entry->bytes > *max_extent_size)
1787 *max_extent_size = entry->bytes;
1788 continue;
1789 }
1790
1791 /* make sure the space returned is big enough
1792 * to match our requested alignment
1793 */
1794 if (*bytes >= align) {
1795 tmp = entry->offset - ctl->start + align - 1;
1796 tmp = div64_u64(tmp, align);
1797 tmp = tmp * align + ctl->start;
1798 align_off = tmp - entry->offset;
1799 } else {
1800 align_off = 0;
1801 tmp = entry->offset;
1802 }
1803
1804 if (entry->bytes < *bytes + align_off) {
1805 if (entry->bytes > *max_extent_size)
1806 *max_extent_size = entry->bytes;
1807 continue;
1808 }
1809
1810 if (entry->bitmap) {
1811 u64 size = *bytes;
1812
1813 ret = search_bitmap(ctl, entry, &tmp, &size, true);
1814 if (!ret) {
1815 *offset = tmp;
1816 *bytes = size;
1817 return entry;
1818 } else if (size > *max_extent_size) {
1819 *max_extent_size = size;
1820 }
1821 continue;
1822 }
1823
1824 *offset = tmp;
1825 *bytes = entry->bytes - align_off;
1826 return entry;
1827 }
1828 out:
1829 return NULL;
1830 }
1831
1832 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1833 struct btrfs_free_space *info, u64 offset)
1834 {
1835 info->offset = offset_to_bitmap(ctl, offset);
1836 info->bytes = 0;
1837 INIT_LIST_HEAD(&info->list);
1838 link_free_space(ctl, info);
1839 ctl->total_bitmaps++;
1840
1841 ctl->op->recalc_thresholds(ctl);
1842 }
1843
1844 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1845 struct btrfs_free_space *bitmap_info)
1846 {
1847 unlink_free_space(ctl, bitmap_info);
1848 kfree(bitmap_info->bitmap);
1849 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1850 ctl->total_bitmaps--;
1851 ctl->op->recalc_thresholds(ctl);
1852 }
1853
1854 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1855 struct btrfs_free_space *bitmap_info,
1856 u64 *offset, u64 *bytes)
1857 {
1858 u64 end;
1859 u64 search_start, search_bytes;
1860 int ret;
1861
1862 again:
1863 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1864
1865 /*
1866 * We need to search for bits in this bitmap. We could only cover some
1867 * of the extent in this bitmap thanks to how we add space, so we need
1868 * to search for as much as it as we can and clear that amount, and then
1869 * go searching for the next bit.
1870 */
1871 search_start = *offset;
1872 search_bytes = ctl->unit;
1873 search_bytes = min(search_bytes, end - search_start + 1);
1874 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes,
1875 false);
1876 if (ret < 0 || search_start != *offset)
1877 return -EINVAL;
1878
1879 /* We may have found more bits than what we need */
1880 search_bytes = min(search_bytes, *bytes);
1881
1882 /* Cannot clear past the end of the bitmap */
1883 search_bytes = min(search_bytes, end - search_start + 1);
1884
1885 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1886 *offset += search_bytes;
1887 *bytes -= search_bytes;
1888
1889 if (*bytes) {
1890 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1891 if (!bitmap_info->bytes)
1892 free_bitmap(ctl, bitmap_info);
1893
1894 /*
1895 * no entry after this bitmap, but we still have bytes to
1896 * remove, so something has gone wrong.
1897 */
1898 if (!next)
1899 return -EINVAL;
1900
1901 bitmap_info = rb_entry(next, struct btrfs_free_space,
1902 offset_index);
1903
1904 /*
1905 * if the next entry isn't a bitmap we need to return to let the
1906 * extent stuff do its work.
1907 */
1908 if (!bitmap_info->bitmap)
1909 return -EAGAIN;
1910
1911 /*
1912 * Ok the next item is a bitmap, but it may not actually hold
1913 * the information for the rest of this free space stuff, so
1914 * look for it, and if we don't find it return so we can try
1915 * everything over again.
1916 */
1917 search_start = *offset;
1918 search_bytes = ctl->unit;
1919 ret = search_bitmap(ctl, bitmap_info, &search_start,
1920 &search_bytes, false);
1921 if (ret < 0 || search_start != *offset)
1922 return -EAGAIN;
1923
1924 goto again;
1925 } else if (!bitmap_info->bytes)
1926 free_bitmap(ctl, bitmap_info);
1927
1928 return 0;
1929 }
1930
1931 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1932 struct btrfs_free_space *info, u64 offset,
1933 u64 bytes)
1934 {
1935 u64 bytes_to_set = 0;
1936 u64 end;
1937
1938 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1939
1940 bytes_to_set = min(end - offset, bytes);
1941
1942 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1943
1944 /*
1945 * We set some bytes, we have no idea what the max extent size is
1946 * anymore.
1947 */
1948 info->max_extent_size = 0;
1949
1950 return bytes_to_set;
1951
1952 }
1953
1954 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1955 struct btrfs_free_space *info)
1956 {
1957 struct btrfs_block_group_cache *block_group = ctl->private;
1958 struct btrfs_fs_info *fs_info = block_group->fs_info;
1959 bool forced = false;
1960
1961 #ifdef CONFIG_BTRFS_DEBUG
1962 if (btrfs_should_fragment_free_space(block_group))
1963 forced = true;
1964 #endif
1965
1966 /*
1967 * If we are below the extents threshold then we can add this as an
1968 * extent, and don't have to deal with the bitmap
1969 */
1970 if (!forced && ctl->free_extents < ctl->extents_thresh) {
1971 /*
1972 * If this block group has some small extents we don't want to
1973 * use up all of our free slots in the cache with them, we want
1974 * to reserve them to larger extents, however if we have plenty
1975 * of cache left then go ahead an dadd them, no sense in adding
1976 * the overhead of a bitmap if we don't have to.
1977 */
1978 if (info->bytes <= fs_info->sectorsize * 4) {
1979 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1980 return false;
1981 } else {
1982 return false;
1983 }
1984 }
1985
1986 /*
1987 * The original block groups from mkfs can be really small, like 8
1988 * megabytes, so don't bother with a bitmap for those entries. However
1989 * some block groups can be smaller than what a bitmap would cover but
1990 * are still large enough that they could overflow the 32k memory limit,
1991 * so allow those block groups to still be allowed to have a bitmap
1992 * entry.
1993 */
1994 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
1995 return false;
1996
1997 return true;
1998 }
1999
2000 static const struct btrfs_free_space_op free_space_op = {
2001 .recalc_thresholds = recalculate_thresholds,
2002 .use_bitmap = use_bitmap,
2003 };
2004
2005 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
2006 struct btrfs_free_space *info)
2007 {
2008 struct btrfs_free_space *bitmap_info;
2009 struct btrfs_block_group_cache *block_group = NULL;
2010 int added = 0;
2011 u64 bytes, offset, bytes_added;
2012 int ret;
2013
2014 bytes = info->bytes;
2015 offset = info->offset;
2016
2017 if (!ctl->op->use_bitmap(ctl, info))
2018 return 0;
2019
2020 if (ctl->op == &free_space_op)
2021 block_group = ctl->private;
2022 again:
2023 /*
2024 * Since we link bitmaps right into the cluster we need to see if we
2025 * have a cluster here, and if so and it has our bitmap we need to add
2026 * the free space to that bitmap.
2027 */
2028 if (block_group && !list_empty(&block_group->cluster_list)) {
2029 struct btrfs_free_cluster *cluster;
2030 struct rb_node *node;
2031 struct btrfs_free_space *entry;
2032
2033 cluster = list_entry(block_group->cluster_list.next,
2034 struct btrfs_free_cluster,
2035 block_group_list);
2036 spin_lock(&cluster->lock);
2037 node = rb_first(&cluster->root);
2038 if (!node) {
2039 spin_unlock(&cluster->lock);
2040 goto no_cluster_bitmap;
2041 }
2042
2043 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2044 if (!entry->bitmap) {
2045 spin_unlock(&cluster->lock);
2046 goto no_cluster_bitmap;
2047 }
2048
2049 if (entry->offset == offset_to_bitmap(ctl, offset)) {
2050 bytes_added = add_bytes_to_bitmap(ctl, entry,
2051 offset, bytes);
2052 bytes -= bytes_added;
2053 offset += bytes_added;
2054 }
2055 spin_unlock(&cluster->lock);
2056 if (!bytes) {
2057 ret = 1;
2058 goto out;
2059 }
2060 }
2061
2062 no_cluster_bitmap:
2063 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2064 1, 0);
2065 if (!bitmap_info) {
2066 ASSERT(added == 0);
2067 goto new_bitmap;
2068 }
2069
2070 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
2071 bytes -= bytes_added;
2072 offset += bytes_added;
2073 added = 0;
2074
2075 if (!bytes) {
2076 ret = 1;
2077 goto out;
2078 } else
2079 goto again;
2080
2081 new_bitmap:
2082 if (info && info->bitmap) {
2083 add_new_bitmap(ctl, info, offset);
2084 added = 1;
2085 info = NULL;
2086 goto again;
2087 } else {
2088 spin_unlock(&ctl->tree_lock);
2089
2090 /* no pre-allocated info, allocate a new one */
2091 if (!info) {
2092 info = kmem_cache_zalloc(btrfs_free_space_cachep,
2093 GFP_NOFS);
2094 if (!info) {
2095 spin_lock(&ctl->tree_lock);
2096 ret = -ENOMEM;
2097 goto out;
2098 }
2099 }
2100
2101 /* allocate the bitmap */
2102 info->bitmap = kzalloc(PAGE_SIZE, GFP_NOFS);
2103 spin_lock(&ctl->tree_lock);
2104 if (!info->bitmap) {
2105 ret = -ENOMEM;
2106 goto out;
2107 }
2108 goto again;
2109 }
2110
2111 out:
2112 if (info) {
2113 if (info->bitmap)
2114 kfree(info->bitmap);
2115 kmem_cache_free(btrfs_free_space_cachep, info);
2116 }
2117
2118 return ret;
2119 }
2120
2121 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2122 struct btrfs_free_space *info, bool update_stat)
2123 {
2124 struct btrfs_free_space *left_info;
2125 struct btrfs_free_space *right_info;
2126 bool merged = false;
2127 u64 offset = info->offset;
2128 u64 bytes = info->bytes;
2129
2130 /*
2131 * first we want to see if there is free space adjacent to the range we
2132 * are adding, if there is remove that struct and add a new one to
2133 * cover the entire range
2134 */
2135 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
2136 if (right_info && rb_prev(&right_info->offset_index))
2137 left_info = rb_entry(rb_prev(&right_info->offset_index),
2138 struct btrfs_free_space, offset_index);
2139 else
2140 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
2141
2142 if (right_info && !right_info->bitmap) {
2143 if (update_stat)
2144 unlink_free_space(ctl, right_info);
2145 else
2146 __unlink_free_space(ctl, right_info);
2147 info->bytes += right_info->bytes;
2148 kmem_cache_free(btrfs_free_space_cachep, right_info);
2149 merged = true;
2150 }
2151
2152 if (left_info && !left_info->bitmap &&
2153 left_info->offset + left_info->bytes == offset) {
2154 if (update_stat)
2155 unlink_free_space(ctl, left_info);
2156 else
2157 __unlink_free_space(ctl, left_info);
2158 info->offset = left_info->offset;
2159 info->bytes += left_info->bytes;
2160 kmem_cache_free(btrfs_free_space_cachep, left_info);
2161 merged = true;
2162 }
2163
2164 return merged;
2165 }
2166
2167 static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2168 struct btrfs_free_space *info,
2169 bool update_stat)
2170 {
2171 struct btrfs_free_space *bitmap;
2172 unsigned long i;
2173 unsigned long j;
2174 const u64 end = info->offset + info->bytes;
2175 const u64 bitmap_offset = offset_to_bitmap(ctl, end);
2176 u64 bytes;
2177
2178 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2179 if (!bitmap)
2180 return false;
2181
2182 i = offset_to_bit(bitmap->offset, ctl->unit, end);
2183 j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
2184 if (j == i)
2185 return false;
2186 bytes = (j - i) * ctl->unit;
2187 info->bytes += bytes;
2188
2189 if (update_stat)
2190 bitmap_clear_bits(ctl, bitmap, end, bytes);
2191 else
2192 __bitmap_clear_bits(ctl, bitmap, end, bytes);
2193
2194 if (!bitmap->bytes)
2195 free_bitmap(ctl, bitmap);
2196
2197 return true;
2198 }
2199
2200 static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2201 struct btrfs_free_space *info,
2202 bool update_stat)
2203 {
2204 struct btrfs_free_space *bitmap;
2205 u64 bitmap_offset;
2206 unsigned long i;
2207 unsigned long j;
2208 unsigned long prev_j;
2209 u64 bytes;
2210
2211 bitmap_offset = offset_to_bitmap(ctl, info->offset);
2212 /* If we're on a boundary, try the previous logical bitmap. */
2213 if (bitmap_offset == info->offset) {
2214 if (info->offset == 0)
2215 return false;
2216 bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
2217 }
2218
2219 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2220 if (!bitmap)
2221 return false;
2222
2223 i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
2224 j = 0;
2225 prev_j = (unsigned long)-1;
2226 for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2227 if (j > i)
2228 break;
2229 prev_j = j;
2230 }
2231 if (prev_j == i)
2232 return false;
2233
2234 if (prev_j == (unsigned long)-1)
2235 bytes = (i + 1) * ctl->unit;
2236 else
2237 bytes = (i - prev_j) * ctl->unit;
2238
2239 info->offset -= bytes;
2240 info->bytes += bytes;
2241
2242 if (update_stat)
2243 bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2244 else
2245 __bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2246
2247 if (!bitmap->bytes)
2248 free_bitmap(ctl, bitmap);
2249
2250 return true;
2251 }
2252
2253 /*
2254 * We prefer always to allocate from extent entries, both for clustered and
2255 * non-clustered allocation requests. So when attempting to add a new extent
2256 * entry, try to see if there's adjacent free space in bitmap entries, and if
2257 * there is, migrate that space from the bitmaps to the extent.
2258 * Like this we get better chances of satisfying space allocation requests
2259 * because we attempt to satisfy them based on a single cache entry, and never
2260 * on 2 or more entries - even if the entries represent a contiguous free space
2261 * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2262 * ends).
2263 */
2264 static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2265 struct btrfs_free_space *info,
2266 bool update_stat)
2267 {
2268 /*
2269 * Only work with disconnected entries, as we can change their offset,
2270 * and must be extent entries.
2271 */
2272 ASSERT(!info->bitmap);
2273 ASSERT(RB_EMPTY_NODE(&info->offset_index));
2274
2275 if (ctl->total_bitmaps > 0) {
2276 bool stole_end;
2277 bool stole_front = false;
2278
2279 stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2280 if (ctl->total_bitmaps > 0)
2281 stole_front = steal_from_bitmap_to_front(ctl, info,
2282 update_stat);
2283
2284 if (stole_end || stole_front)
2285 try_merge_free_space(ctl, info, update_stat);
2286 }
2287 }
2288
2289 int __btrfs_add_free_space(struct btrfs_fs_info *fs_info,
2290 struct btrfs_free_space_ctl *ctl,
2291 u64 offset, u64 bytes)
2292 {
2293 struct btrfs_free_space *info;
2294 int ret = 0;
2295
2296 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2297 if (!info)
2298 return -ENOMEM;
2299
2300 info->offset = offset;
2301 info->bytes = bytes;
2302 RB_CLEAR_NODE(&info->offset_index);
2303
2304 spin_lock(&ctl->tree_lock);
2305
2306 if (try_merge_free_space(ctl, info, true))
2307 goto link;
2308
2309 /*
2310 * There was no extent directly to the left or right of this new
2311 * extent then we know we're going to have to allocate a new extent, so
2312 * before we do that see if we need to drop this into a bitmap
2313 */
2314 ret = insert_into_bitmap(ctl, info);
2315 if (ret < 0) {
2316 goto out;
2317 } else if (ret) {
2318 ret = 0;
2319 goto out;
2320 }
2321 link:
2322 /*
2323 * Only steal free space from adjacent bitmaps if we're sure we're not
2324 * going to add the new free space to existing bitmap entries - because
2325 * that would mean unnecessary work that would be reverted. Therefore
2326 * attempt to steal space from bitmaps if we're adding an extent entry.
2327 */
2328 steal_from_bitmap(ctl, info, true);
2329
2330 ret = link_free_space(ctl, info);
2331 if (ret)
2332 kmem_cache_free(btrfs_free_space_cachep, info);
2333 out:
2334 spin_unlock(&ctl->tree_lock);
2335
2336 if (ret) {
2337 btrfs_crit(fs_info, "unable to add free space :%d", ret);
2338 ASSERT(ret != -EEXIST);
2339 }
2340
2341 return ret;
2342 }
2343
2344 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
2345 u64 offset, u64 bytes)
2346 {
2347 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2348 struct btrfs_free_space *info;
2349 int ret;
2350 bool re_search = false;
2351
2352 spin_lock(&ctl->tree_lock);
2353
2354 again:
2355 ret = 0;
2356 if (!bytes)
2357 goto out_lock;
2358
2359 info = tree_search_offset(ctl, offset, 0, 0);
2360 if (!info) {
2361 /*
2362 * oops didn't find an extent that matched the space we wanted
2363 * to remove, look for a bitmap instead
2364 */
2365 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2366 1, 0);
2367 if (!info) {
2368 /*
2369 * If we found a partial bit of our free space in a
2370 * bitmap but then couldn't find the other part this may
2371 * be a problem, so WARN about it.
2372 */
2373 WARN_ON(re_search);
2374 goto out_lock;
2375 }
2376 }
2377
2378 re_search = false;
2379 if (!info->bitmap) {
2380 unlink_free_space(ctl, info);
2381 if (offset == info->offset) {
2382 u64 to_free = min(bytes, info->bytes);
2383
2384 info->bytes -= to_free;
2385 info->offset += to_free;
2386 if (info->bytes) {
2387 ret = link_free_space(ctl, info);
2388 WARN_ON(ret);
2389 } else {
2390 kmem_cache_free(btrfs_free_space_cachep, info);
2391 }
2392
2393 offset += to_free;
2394 bytes -= to_free;
2395 goto again;
2396 } else {
2397 u64 old_end = info->bytes + info->offset;
2398
2399 info->bytes = offset - info->offset;
2400 ret = link_free_space(ctl, info);
2401 WARN_ON(ret);
2402 if (ret)
2403 goto out_lock;
2404
2405 /* Not enough bytes in this entry to satisfy us */
2406 if (old_end < offset + bytes) {
2407 bytes -= old_end - offset;
2408 offset = old_end;
2409 goto again;
2410 } else if (old_end == offset + bytes) {
2411 /* all done */
2412 goto out_lock;
2413 }
2414 spin_unlock(&ctl->tree_lock);
2415
2416 ret = btrfs_add_free_space(block_group, offset + bytes,
2417 old_end - (offset + bytes));
2418 WARN_ON(ret);
2419 goto out;
2420 }
2421 }
2422
2423 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2424 if (ret == -EAGAIN) {
2425 re_search = true;
2426 goto again;
2427 }
2428 out_lock:
2429 spin_unlock(&ctl->tree_lock);
2430 out:
2431 return ret;
2432 }
2433
2434 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
2435 u64 bytes)
2436 {
2437 struct btrfs_fs_info *fs_info = block_group->fs_info;
2438 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2439 struct btrfs_free_space *info;
2440 struct rb_node *n;
2441 int count = 0;
2442
2443 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2444 info = rb_entry(n, struct btrfs_free_space, offset_index);
2445 if (info->bytes >= bytes && !block_group->ro)
2446 count++;
2447 btrfs_crit(fs_info, "entry offset %llu, bytes %llu, bitmap %s",
2448 info->offset, info->bytes,
2449 (info->bitmap) ? "yes" : "no");
2450 }
2451 btrfs_info(fs_info, "block group has cluster?: %s",
2452 list_empty(&block_group->cluster_list) ? "no" : "yes");
2453 btrfs_info(fs_info,
2454 "%d blocks of free space at or bigger than bytes is", count);
2455 }
2456
2457 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2458 {
2459 struct btrfs_fs_info *fs_info = block_group->fs_info;
2460 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2461
2462 spin_lock_init(&ctl->tree_lock);
2463 ctl->unit = fs_info->sectorsize;
2464 ctl->start = block_group->key.objectid;
2465 ctl->private = block_group;
2466 ctl->op = &free_space_op;
2467 INIT_LIST_HEAD(&ctl->trimming_ranges);
2468 mutex_init(&ctl->cache_writeout_mutex);
2469
2470 /*
2471 * we only want to have 32k of ram per block group for keeping
2472 * track of free space, and if we pass 1/2 of that we want to
2473 * start converting things over to using bitmaps
2474 */
2475 ctl->extents_thresh = (SZ_32K / 2) / sizeof(struct btrfs_free_space);
2476 }
2477
2478 /*
2479 * for a given cluster, put all of its extents back into the free
2480 * space cache. If the block group passed doesn't match the block group
2481 * pointed to by the cluster, someone else raced in and freed the
2482 * cluster already. In that case, we just return without changing anything
2483 */
2484 static int
2485 __btrfs_return_cluster_to_free_space(
2486 struct btrfs_block_group_cache *block_group,
2487 struct btrfs_free_cluster *cluster)
2488 {
2489 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2490 struct btrfs_free_space *entry;
2491 struct rb_node *node;
2492
2493 spin_lock(&cluster->lock);
2494 if (cluster->block_group != block_group)
2495 goto out;
2496
2497 cluster->block_group = NULL;
2498 cluster->window_start = 0;
2499 list_del_init(&cluster->block_group_list);
2500
2501 node = rb_first(&cluster->root);
2502 while (node) {
2503 bool bitmap;
2504
2505 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2506 node = rb_next(&entry->offset_index);
2507 rb_erase(&entry->offset_index, &cluster->root);
2508 RB_CLEAR_NODE(&entry->offset_index);
2509
2510 bitmap = (entry->bitmap != NULL);
2511 if (!bitmap) {
2512 try_merge_free_space(ctl, entry, false);
2513 steal_from_bitmap(ctl, entry, false);
2514 }
2515 tree_insert_offset(&ctl->free_space_offset,
2516 entry->offset, &entry->offset_index, bitmap);
2517 }
2518 cluster->root = RB_ROOT;
2519
2520 out:
2521 spin_unlock(&cluster->lock);
2522 btrfs_put_block_group(block_group);
2523 return 0;
2524 }
2525
2526 static void __btrfs_remove_free_space_cache_locked(
2527 struct btrfs_free_space_ctl *ctl)
2528 {
2529 struct btrfs_free_space *info;
2530 struct rb_node *node;
2531
2532 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2533 info = rb_entry(node, struct btrfs_free_space, offset_index);
2534 if (!info->bitmap) {
2535 unlink_free_space(ctl, info);
2536 kmem_cache_free(btrfs_free_space_cachep, info);
2537 } else {
2538 free_bitmap(ctl, info);
2539 }
2540
2541 cond_resched_lock(&ctl->tree_lock);
2542 }
2543 }
2544
2545 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2546 {
2547 spin_lock(&ctl->tree_lock);
2548 __btrfs_remove_free_space_cache_locked(ctl);
2549 spin_unlock(&ctl->tree_lock);
2550 }
2551
2552 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2553 {
2554 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2555 struct btrfs_free_cluster *cluster;
2556 struct list_head *head;
2557
2558 spin_lock(&ctl->tree_lock);
2559 while ((head = block_group->cluster_list.next) !=
2560 &block_group->cluster_list) {
2561 cluster = list_entry(head, struct btrfs_free_cluster,
2562 block_group_list);
2563
2564 WARN_ON(cluster->block_group != block_group);
2565 __btrfs_return_cluster_to_free_space(block_group, cluster);
2566
2567 cond_resched_lock(&ctl->tree_lock);
2568 }
2569 __btrfs_remove_free_space_cache_locked(ctl);
2570 spin_unlock(&ctl->tree_lock);
2571
2572 }
2573
2574 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2575 u64 offset, u64 bytes, u64 empty_size,
2576 u64 *max_extent_size)
2577 {
2578 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2579 struct btrfs_free_space *entry = NULL;
2580 u64 bytes_search = bytes + empty_size;
2581 u64 ret = 0;
2582 u64 align_gap = 0;
2583 u64 align_gap_len = 0;
2584
2585 spin_lock(&ctl->tree_lock);
2586 entry = find_free_space(ctl, &offset, &bytes_search,
2587 block_group->full_stripe_len, max_extent_size);
2588 if (!entry)
2589 goto out;
2590
2591 ret = offset;
2592 if (entry->bitmap) {
2593 bitmap_clear_bits(ctl, entry, offset, bytes);
2594 if (!entry->bytes)
2595 free_bitmap(ctl, entry);
2596 } else {
2597 unlink_free_space(ctl, entry);
2598 align_gap_len = offset - entry->offset;
2599 align_gap = entry->offset;
2600
2601 entry->offset = offset + bytes;
2602 WARN_ON(entry->bytes < bytes + align_gap_len);
2603
2604 entry->bytes -= bytes + align_gap_len;
2605 if (!entry->bytes)
2606 kmem_cache_free(btrfs_free_space_cachep, entry);
2607 else
2608 link_free_space(ctl, entry);
2609 }
2610 out:
2611 spin_unlock(&ctl->tree_lock);
2612
2613 if (align_gap_len)
2614 __btrfs_add_free_space(block_group->fs_info, ctl,
2615 align_gap, align_gap_len);
2616 return ret;
2617 }
2618
2619 /*
2620 * given a cluster, put all of its extents back into the free space
2621 * cache. If a block group is passed, this function will only free
2622 * a cluster that belongs to the passed block group.
2623 *
2624 * Otherwise, it'll get a reference on the block group pointed to by the
2625 * cluster and remove the cluster from it.
2626 */
2627 int btrfs_return_cluster_to_free_space(
2628 struct btrfs_block_group_cache *block_group,
2629 struct btrfs_free_cluster *cluster)
2630 {
2631 struct btrfs_free_space_ctl *ctl;
2632 int ret;
2633
2634 /* first, get a safe pointer to the block group */
2635 spin_lock(&cluster->lock);
2636 if (!block_group) {
2637 block_group = cluster->block_group;
2638 if (!block_group) {
2639 spin_unlock(&cluster->lock);
2640 return 0;
2641 }
2642 } else if (cluster->block_group != block_group) {
2643 /* someone else has already freed it don't redo their work */
2644 spin_unlock(&cluster->lock);
2645 return 0;
2646 }
2647 atomic_inc(&block_group->count);
2648 spin_unlock(&cluster->lock);
2649
2650 ctl = block_group->free_space_ctl;
2651
2652 /* now return any extents the cluster had on it */
2653 spin_lock(&ctl->tree_lock);
2654 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2655 spin_unlock(&ctl->tree_lock);
2656
2657 /* finally drop our ref */
2658 btrfs_put_block_group(block_group);
2659 return ret;
2660 }
2661
2662 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2663 struct btrfs_free_cluster *cluster,
2664 struct btrfs_free_space *entry,
2665 u64 bytes, u64 min_start,
2666 u64 *max_extent_size)
2667 {
2668 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2669 int err;
2670 u64 search_start = cluster->window_start;
2671 u64 search_bytes = bytes;
2672 u64 ret = 0;
2673
2674 search_start = min_start;
2675 search_bytes = bytes;
2676
2677 err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
2678 if (err) {
2679 if (search_bytes > *max_extent_size)
2680 *max_extent_size = search_bytes;
2681 return 0;
2682 }
2683
2684 ret = search_start;
2685 __bitmap_clear_bits(ctl, entry, ret, bytes);
2686
2687 return ret;
2688 }
2689
2690 /*
2691 * given a cluster, try to allocate 'bytes' from it, returns 0
2692 * if it couldn't find anything suitably large, or a logical disk offset
2693 * if things worked out
2694 */
2695 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2696 struct btrfs_free_cluster *cluster, u64 bytes,
2697 u64 min_start, u64 *max_extent_size)
2698 {
2699 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2700 struct btrfs_free_space *entry = NULL;
2701 struct rb_node *node;
2702 u64 ret = 0;
2703
2704 spin_lock(&cluster->lock);
2705 if (bytes > cluster->max_size)
2706 goto out;
2707
2708 if (cluster->block_group != block_group)
2709 goto out;
2710
2711 node = rb_first(&cluster->root);
2712 if (!node)
2713 goto out;
2714
2715 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2716 while (1) {
2717 if (entry->bytes < bytes && entry->bytes > *max_extent_size)
2718 *max_extent_size = entry->bytes;
2719
2720 if (entry->bytes < bytes ||
2721 (!entry->bitmap && entry->offset < min_start)) {
2722 node = rb_next(&entry->offset_index);
2723 if (!node)
2724 break;
2725 entry = rb_entry(node, struct btrfs_free_space,
2726 offset_index);
2727 continue;
2728 }
2729
2730 if (entry->bitmap) {
2731 ret = btrfs_alloc_from_bitmap(block_group,
2732 cluster, entry, bytes,
2733 cluster->window_start,
2734 max_extent_size);
2735 if (ret == 0) {
2736 node = rb_next(&entry->offset_index);
2737 if (!node)
2738 break;
2739 entry = rb_entry(node, struct btrfs_free_space,
2740 offset_index);
2741 continue;
2742 }
2743 cluster->window_start += bytes;
2744 } else {
2745 ret = entry->offset;
2746
2747 entry->offset += bytes;
2748 entry->bytes -= bytes;
2749 }
2750
2751 if (entry->bytes == 0)
2752 rb_erase(&entry->offset_index, &cluster->root);
2753 break;
2754 }
2755 out:
2756 spin_unlock(&cluster->lock);
2757
2758 if (!ret)
2759 return 0;
2760
2761 spin_lock(&ctl->tree_lock);
2762
2763 ctl->free_space -= bytes;
2764 if (entry->bytes == 0) {
2765 ctl->free_extents--;
2766 if (entry->bitmap) {
2767 kfree(entry->bitmap);
2768 ctl->total_bitmaps--;
2769 ctl->op->recalc_thresholds(ctl);
2770 }
2771 kmem_cache_free(btrfs_free_space_cachep, entry);
2772 }
2773
2774 spin_unlock(&ctl->tree_lock);
2775
2776 return ret;
2777 }
2778
2779 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2780 struct btrfs_free_space *entry,
2781 struct btrfs_free_cluster *cluster,
2782 u64 offset, u64 bytes,
2783 u64 cont1_bytes, u64 min_bytes)
2784 {
2785 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2786 unsigned long next_zero;
2787 unsigned long i;
2788 unsigned long want_bits;
2789 unsigned long min_bits;
2790 unsigned long found_bits;
2791 unsigned long max_bits = 0;
2792 unsigned long start = 0;
2793 unsigned long total_found = 0;
2794 int ret;
2795
2796 i = offset_to_bit(entry->offset, ctl->unit,
2797 max_t(u64, offset, entry->offset));
2798 want_bits = bytes_to_bits(bytes, ctl->unit);
2799 min_bits = bytes_to_bits(min_bytes, ctl->unit);
2800
2801 /*
2802 * Don't bother looking for a cluster in this bitmap if it's heavily
2803 * fragmented.
2804 */
2805 if (entry->max_extent_size &&
2806 entry->max_extent_size < cont1_bytes)
2807 return -ENOSPC;
2808 again:
2809 found_bits = 0;
2810 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2811 next_zero = find_next_zero_bit(entry->bitmap,
2812 BITS_PER_BITMAP, i);
2813 if (next_zero - i >= min_bits) {
2814 found_bits = next_zero - i;
2815 if (found_bits > max_bits)
2816 max_bits = found_bits;
2817 break;
2818 }
2819 if (next_zero - i > max_bits)
2820 max_bits = next_zero - i;
2821 i = next_zero;
2822 }
2823
2824 if (!found_bits) {
2825 entry->max_extent_size = (u64)max_bits * ctl->unit;
2826 return -ENOSPC;
2827 }
2828
2829 if (!total_found) {
2830 start = i;
2831 cluster->max_size = 0;
2832 }
2833
2834 total_found += found_bits;
2835
2836 if (cluster->max_size < found_bits * ctl->unit)
2837 cluster->max_size = found_bits * ctl->unit;
2838
2839 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2840 i = next_zero + 1;
2841 goto again;
2842 }
2843
2844 cluster->window_start = start * ctl->unit + entry->offset;
2845 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2846 ret = tree_insert_offset(&cluster->root, entry->offset,
2847 &entry->offset_index, 1);
2848 ASSERT(!ret); /* -EEXIST; Logic error */
2849
2850 trace_btrfs_setup_cluster(block_group, cluster,
2851 total_found * ctl->unit, 1);
2852 return 0;
2853 }
2854
2855 /*
2856 * This searches the block group for just extents to fill the cluster with.
2857 * Try to find a cluster with at least bytes total bytes, at least one
2858 * extent of cont1_bytes, and other clusters of at least min_bytes.
2859 */
2860 static noinline int
2861 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2862 struct btrfs_free_cluster *cluster,
2863 struct list_head *bitmaps, u64 offset, u64 bytes,
2864 u64 cont1_bytes, u64 min_bytes)
2865 {
2866 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2867 struct btrfs_free_space *first = NULL;
2868 struct btrfs_free_space *entry = NULL;
2869 struct btrfs_free_space *last;
2870 struct rb_node *node;
2871 u64 window_free;
2872 u64 max_extent;
2873 u64 total_size = 0;
2874
2875 entry = tree_search_offset(ctl, offset, 0, 1);
2876 if (!entry)
2877 return -ENOSPC;
2878
2879 /*
2880 * We don't want bitmaps, so just move along until we find a normal
2881 * extent entry.
2882 */
2883 while (entry->bitmap || entry->bytes < min_bytes) {
2884 if (entry->bitmap && list_empty(&entry->list))
2885 list_add_tail(&entry->list, bitmaps);
2886 node = rb_next(&entry->offset_index);
2887 if (!node)
2888 return -ENOSPC;
2889 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2890 }
2891
2892 window_free = entry->bytes;
2893 max_extent = entry->bytes;
2894 first = entry;
2895 last = entry;
2896
2897 for (node = rb_next(&entry->offset_index); node;
2898 node = rb_next(&entry->offset_index)) {
2899 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2900
2901 if (entry->bitmap) {
2902 if (list_empty(&entry->list))
2903 list_add_tail(&entry->list, bitmaps);
2904 continue;
2905 }
2906
2907 if (entry->bytes < min_bytes)
2908 continue;
2909
2910 last = entry;
2911 window_free += entry->bytes;
2912 if (entry->bytes > max_extent)
2913 max_extent = entry->bytes;
2914 }
2915
2916 if (window_free < bytes || max_extent < cont1_bytes)
2917 return -ENOSPC;
2918
2919 cluster->window_start = first->offset;
2920
2921 node = &first->offset_index;
2922
2923 /*
2924 * now we've found our entries, pull them out of the free space
2925 * cache and put them into the cluster rbtree
2926 */
2927 do {
2928 int ret;
2929
2930 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2931 node = rb_next(&entry->offset_index);
2932 if (entry->bitmap || entry->bytes < min_bytes)
2933 continue;
2934
2935 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2936 ret = tree_insert_offset(&cluster->root, entry->offset,
2937 &entry->offset_index, 0);
2938 total_size += entry->bytes;
2939 ASSERT(!ret); /* -EEXIST; Logic error */
2940 } while (node && entry != last);
2941
2942 cluster->max_size = max_extent;
2943 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2944 return 0;
2945 }
2946
2947 /*
2948 * This specifically looks for bitmaps that may work in the cluster, we assume
2949 * that we have already failed to find extents that will work.
2950 */
2951 static noinline int
2952 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2953 struct btrfs_free_cluster *cluster,
2954 struct list_head *bitmaps, u64 offset, u64 bytes,
2955 u64 cont1_bytes, u64 min_bytes)
2956 {
2957 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2958 struct btrfs_free_space *entry = NULL;
2959 int ret = -ENOSPC;
2960 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2961
2962 if (ctl->total_bitmaps == 0)
2963 return -ENOSPC;
2964
2965 /*
2966 * The bitmap that covers offset won't be in the list unless offset
2967 * is just its start offset.
2968 */
2969 if (!list_empty(bitmaps))
2970 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2971
2972 if (!entry || entry->offset != bitmap_offset) {
2973 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2974 if (entry && list_empty(&entry->list))
2975 list_add(&entry->list, bitmaps);
2976 }
2977
2978 list_for_each_entry(entry, bitmaps, list) {
2979 if (entry->bytes < bytes)
2980 continue;
2981 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2982 bytes, cont1_bytes, min_bytes);
2983 if (!ret)
2984 return 0;
2985 }
2986
2987 /*
2988 * The bitmaps list has all the bitmaps that record free space
2989 * starting after offset, so no more search is required.
2990 */
2991 return -ENOSPC;
2992 }
2993
2994 /*
2995 * here we try to find a cluster of blocks in a block group. The goal
2996 * is to find at least bytes+empty_size.
2997 * We might not find them all in one contiguous area.
2998 *
2999 * returns zero and sets up cluster if things worked out, otherwise
3000 * it returns -enospc
3001 */
3002 int btrfs_find_space_cluster(struct btrfs_fs_info *fs_info,
3003 struct btrfs_block_group_cache *block_group,
3004 struct btrfs_free_cluster *cluster,
3005 u64 offset, u64 bytes, u64 empty_size)
3006 {
3007 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3008 struct btrfs_free_space *entry, *tmp;
3009 LIST_HEAD(bitmaps);
3010 u64 min_bytes;
3011 u64 cont1_bytes;
3012 int ret;
3013
3014 /*
3015 * Choose the minimum extent size we'll require for this
3016 * cluster. For SSD_SPREAD, don't allow any fragmentation.
3017 * For metadata, allow allocates with smaller extents. For
3018 * data, keep it dense.
3019 */
3020 if (btrfs_test_opt(fs_info, SSD_SPREAD)) {
3021 cont1_bytes = min_bytes = bytes + empty_size;
3022 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3023 cont1_bytes = bytes;
3024 min_bytes = fs_info->sectorsize;
3025 } else {
3026 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3027 min_bytes = fs_info->sectorsize;
3028 }
3029
3030 spin_lock(&ctl->tree_lock);
3031
3032 /*
3033 * If we know we don't have enough space to make a cluster don't even
3034 * bother doing all the work to try and find one.
3035 */
3036 if (ctl->free_space < bytes) {
3037 spin_unlock(&ctl->tree_lock);
3038 return -ENOSPC;
3039 }
3040
3041 spin_lock(&cluster->lock);
3042
3043 /* someone already found a cluster, hooray */
3044 if (cluster->block_group) {
3045 ret = 0;
3046 goto out;
3047 }
3048
3049 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
3050 min_bytes);
3051
3052 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
3053 bytes + empty_size,
3054 cont1_bytes, min_bytes);
3055 if (ret)
3056 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
3057 offset, bytes + empty_size,
3058 cont1_bytes, min_bytes);
3059
3060 /* Clear our temporary list */
3061 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3062 list_del_init(&entry->list);
3063
3064 if (!ret) {
3065 atomic_inc(&block_group->count);
3066 list_add_tail(&cluster->block_group_list,
3067 &block_group->cluster_list);
3068 cluster->block_group = block_group;
3069 } else {
3070 trace_btrfs_failed_cluster_setup(block_group);
3071 }
3072 out:
3073 spin_unlock(&cluster->lock);
3074 spin_unlock(&ctl->tree_lock);
3075
3076 return ret;
3077 }
3078
3079 /*
3080 * simple code to zero out a cluster
3081 */
3082 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3083 {
3084 spin_lock_init(&cluster->lock);
3085 spin_lock_init(&cluster->refill_lock);
3086 cluster->root = RB_ROOT;
3087 cluster->max_size = 0;
3088 cluster->fragmented = false;
3089 INIT_LIST_HEAD(&cluster->block_group_list);
3090 cluster->block_group = NULL;
3091 }
3092
3093 static int do_trimming(struct btrfs_block_group_cache *block_group,
3094 u64 *total_trimmed, u64 start, u64 bytes,
3095 u64 reserved_start, u64 reserved_bytes,
3096 struct btrfs_trim_range *trim_entry)
3097 {
3098 struct btrfs_space_info *space_info = block_group->space_info;
3099 struct btrfs_fs_info *fs_info = block_group->fs_info;
3100 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3101 int ret;
3102 int update = 0;
3103 u64 trimmed = 0;
3104
3105 spin_lock(&space_info->lock);
3106 spin_lock(&block_group->lock);
3107 if (!block_group->ro) {
3108 block_group->reserved += reserved_bytes;
3109 space_info->bytes_reserved += reserved_bytes;
3110 update = 1;
3111 }
3112 spin_unlock(&block_group->lock);
3113 spin_unlock(&space_info->lock);
3114
3115 ret = btrfs_discard_extent(fs_info, start, bytes, &trimmed);
3116 if (!ret)
3117 *total_trimmed += trimmed;
3118
3119 mutex_lock(&ctl->cache_writeout_mutex);
3120 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
3121 list_del(&trim_entry->list);
3122 mutex_unlock(&ctl->cache_writeout_mutex);
3123
3124 if (update) {
3125 spin_lock(&space_info->lock);
3126 spin_lock(&block_group->lock);
3127 if (block_group->ro)
3128 space_info->bytes_readonly += reserved_bytes;
3129 block_group->reserved -= reserved_bytes;
3130 space_info->bytes_reserved -= reserved_bytes;
3131 spin_unlock(&space_info->lock);
3132 spin_unlock(&block_group->lock);
3133 }
3134
3135 return ret;
3136 }
3137
3138 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
3139 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
3140 {
3141 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3142 struct btrfs_free_space *entry;
3143 struct rb_node *node;
3144 int ret = 0;
3145 u64 extent_start;
3146 u64 extent_bytes;
3147 u64 bytes;
3148
3149 while (start < end) {
3150 struct btrfs_trim_range trim_entry;
3151
3152 mutex_lock(&ctl->cache_writeout_mutex);
3153 spin_lock(&ctl->tree_lock);
3154
3155 if (ctl->free_space < minlen) {
3156 spin_unlock(&ctl->tree_lock);
3157 mutex_unlock(&ctl->cache_writeout_mutex);
3158 break;
3159 }
3160
3161 entry = tree_search_offset(ctl, start, 0, 1);
3162 if (!entry) {
3163 spin_unlock(&ctl->tree_lock);
3164 mutex_unlock(&ctl->cache_writeout_mutex);
3165 break;
3166 }
3167
3168 /* skip bitmaps */
3169 while (entry->bitmap) {
3170 node = rb_next(&entry->offset_index);
3171 if (!node) {
3172 spin_unlock(&ctl->tree_lock);
3173 mutex_unlock(&ctl->cache_writeout_mutex);
3174 goto out;
3175 }
3176 entry = rb_entry(node, struct btrfs_free_space,
3177 offset_index);
3178 }
3179
3180 if (entry->offset >= end) {
3181 spin_unlock(&ctl->tree_lock);
3182 mutex_unlock(&ctl->cache_writeout_mutex);
3183 break;
3184 }
3185
3186 extent_start = entry->offset;
3187 extent_bytes = entry->bytes;
3188 start = max(start, extent_start);
3189 bytes = min(extent_start + extent_bytes, end) - start;
3190 if (bytes < minlen) {
3191 spin_unlock(&ctl->tree_lock);
3192 mutex_unlock(&ctl->cache_writeout_mutex);
3193 goto next;
3194 }
3195
3196 unlink_free_space(ctl, entry);
3197 kmem_cache_free(btrfs_free_space_cachep, entry);
3198
3199 spin_unlock(&ctl->tree_lock);
3200 trim_entry.start = extent_start;
3201 trim_entry.bytes = extent_bytes;
3202 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3203 mutex_unlock(&ctl->cache_writeout_mutex);
3204
3205 ret = do_trimming(block_group, total_trimmed, start, bytes,
3206 extent_start, extent_bytes, &trim_entry);
3207 if (ret)
3208 break;
3209 next:
3210 start += bytes;
3211
3212 if (fatal_signal_pending(current)) {
3213 ret = -ERESTARTSYS;
3214 break;
3215 }
3216
3217 cond_resched();
3218 }
3219 out:
3220 return ret;
3221 }
3222
3223 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
3224 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
3225 {
3226 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3227 struct btrfs_free_space *entry;
3228 int ret = 0;
3229 int ret2;
3230 u64 bytes;
3231 u64 offset = offset_to_bitmap(ctl, start);
3232
3233 while (offset < end) {
3234 bool next_bitmap = false;
3235 struct btrfs_trim_range trim_entry;
3236
3237 mutex_lock(&ctl->cache_writeout_mutex);
3238 spin_lock(&ctl->tree_lock);
3239
3240 if (ctl->free_space < minlen) {
3241 spin_unlock(&ctl->tree_lock);
3242 mutex_unlock(&ctl->cache_writeout_mutex);
3243 break;
3244 }
3245
3246 entry = tree_search_offset(ctl, offset, 1, 0);
3247 if (!entry) {
3248 spin_unlock(&ctl->tree_lock);
3249 mutex_unlock(&ctl->cache_writeout_mutex);
3250 next_bitmap = true;
3251 goto next;
3252 }
3253
3254 bytes = minlen;
3255 ret2 = search_bitmap(ctl, entry, &start, &bytes, false);
3256 if (ret2 || start >= end) {
3257 spin_unlock(&ctl->tree_lock);
3258 mutex_unlock(&ctl->cache_writeout_mutex);
3259 next_bitmap = true;
3260 goto next;
3261 }
3262
3263 bytes = min(bytes, end - start);
3264 if (bytes < minlen) {
3265 spin_unlock(&ctl->tree_lock);
3266 mutex_unlock(&ctl->cache_writeout_mutex);
3267 goto next;
3268 }
3269
3270 bitmap_clear_bits(ctl, entry, start, bytes);
3271 if (entry->bytes == 0)
3272 free_bitmap(ctl, entry);
3273
3274 spin_unlock(&ctl->tree_lock);
3275 trim_entry.start = start;
3276 trim_entry.bytes = bytes;
3277 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3278 mutex_unlock(&ctl->cache_writeout_mutex);
3279
3280 ret = do_trimming(block_group, total_trimmed, start, bytes,
3281 start, bytes, &trim_entry);
3282 if (ret)
3283 break;
3284 next:
3285 if (next_bitmap) {
3286 offset += BITS_PER_BITMAP * ctl->unit;
3287 } else {
3288 start += bytes;
3289 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
3290 offset += BITS_PER_BITMAP * ctl->unit;
3291 }
3292
3293 if (fatal_signal_pending(current)) {
3294 ret = -ERESTARTSYS;
3295 break;
3296 }
3297
3298 cond_resched();
3299 }
3300
3301 return ret;
3302 }
3303
3304 void btrfs_get_block_group_trimming(struct btrfs_block_group_cache *cache)
3305 {
3306 atomic_inc(&cache->trimming);
3307 }
3308
3309 void btrfs_put_block_group_trimming(struct btrfs_block_group_cache *block_group)
3310 {
3311 struct btrfs_fs_info *fs_info = block_group->fs_info;
3312 struct extent_map_tree *em_tree;
3313 struct extent_map *em;
3314 bool cleanup;
3315
3316 spin_lock(&block_group->lock);
3317 cleanup = (atomic_dec_and_test(&block_group->trimming) &&
3318 block_group->removed);
3319 spin_unlock(&block_group->lock);
3320
3321 if (cleanup) {
3322 mutex_lock(&fs_info->chunk_mutex);
3323 em_tree = &fs_info->mapping_tree.map_tree;
3324 write_lock(&em_tree->lock);
3325 em = lookup_extent_mapping(em_tree, block_group->key.objectid,
3326 1);
3327 BUG_ON(!em); /* logic error, can't happen */
3328 /*
3329 * remove_extent_mapping() will delete us from the pinned_chunks
3330 * list, which is protected by the chunk mutex.
3331 */
3332 remove_extent_mapping(em_tree, em);
3333 write_unlock(&em_tree->lock);
3334 mutex_unlock(&fs_info->chunk_mutex);
3335
3336 /* once for us and once for the tree */
3337 free_extent_map(em);
3338 free_extent_map(em);
3339
3340 /*
3341 * We've left one free space entry and other tasks trimming
3342 * this block group have left 1 entry each one. Free them.
3343 */
3344 __btrfs_remove_free_space_cache(block_group->free_space_ctl);
3345 }
3346 }
3347
3348 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
3349 u64 *trimmed, u64 start, u64 end, u64 minlen)
3350 {
3351 int ret;
3352
3353 *trimmed = 0;
3354
3355 spin_lock(&block_group->lock);
3356 if (block_group->removed) {
3357 spin_unlock(&block_group->lock);
3358 return 0;
3359 }
3360 btrfs_get_block_group_trimming(block_group);
3361 spin_unlock(&block_group->lock);
3362
3363 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
3364 if (ret)
3365 goto out;
3366
3367 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
3368 out:
3369 btrfs_put_block_group_trimming(block_group);
3370 return ret;
3371 }
3372
3373 /*
3374 * Find the left-most item in the cache tree, and then return the
3375 * smallest inode number in the item.
3376 *
3377 * Note: the returned inode number may not be the smallest one in
3378 * the tree, if the left-most item is a bitmap.
3379 */
3380 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
3381 {
3382 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
3383 struct btrfs_free_space *entry = NULL;
3384 u64 ino = 0;
3385
3386 spin_lock(&ctl->tree_lock);
3387
3388 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
3389 goto out;
3390
3391 entry = rb_entry(rb_first(&ctl->free_space_offset),
3392 struct btrfs_free_space, offset_index);
3393
3394 if (!entry->bitmap) {
3395 ino = entry->offset;
3396
3397 unlink_free_space(ctl, entry);
3398 entry->offset++;
3399 entry->bytes--;
3400 if (!entry->bytes)
3401 kmem_cache_free(btrfs_free_space_cachep, entry);
3402 else
3403 link_free_space(ctl, entry);
3404 } else {
3405 u64 offset = 0;
3406 u64 count = 1;
3407 int ret;
3408
3409 ret = search_bitmap(ctl, entry, &offset, &count, true);
3410 /* Logic error; Should be empty if it can't find anything */
3411 ASSERT(!ret);
3412
3413 ino = offset;
3414 bitmap_clear_bits(ctl, entry, offset, 1);
3415 if (entry->bytes == 0)
3416 free_bitmap(ctl, entry);
3417 }
3418 out:
3419 spin_unlock(&ctl->tree_lock);
3420
3421 return ino;
3422 }
3423
3424 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
3425 struct btrfs_path *path)
3426 {
3427 struct inode *inode = NULL;
3428
3429 spin_lock(&root->ino_cache_lock);
3430 if (root->ino_cache_inode)
3431 inode = igrab(root->ino_cache_inode);
3432 spin_unlock(&root->ino_cache_lock);
3433 if (inode)
3434 return inode;
3435
3436 inode = __lookup_free_space_inode(root, path, 0);
3437 if (IS_ERR(inode))
3438 return inode;
3439
3440 spin_lock(&root->ino_cache_lock);
3441 if (!btrfs_fs_closing(root->fs_info))
3442 root->ino_cache_inode = igrab(inode);
3443 spin_unlock(&root->ino_cache_lock);
3444
3445 return inode;
3446 }
3447
3448 int create_free_ino_inode(struct btrfs_root *root,
3449 struct btrfs_trans_handle *trans,
3450 struct btrfs_path *path)
3451 {
3452 return __create_free_space_inode(root, trans, path,
3453 BTRFS_FREE_INO_OBJECTID, 0);
3454 }
3455
3456 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3457 {
3458 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
3459 struct btrfs_path *path;
3460 struct inode *inode;
3461 int ret = 0;
3462 u64 root_gen = btrfs_root_generation(&root->root_item);
3463
3464 if (!btrfs_test_opt(fs_info, INODE_MAP_CACHE))
3465 return 0;
3466
3467 /*
3468 * If we're unmounting then just return, since this does a search on the
3469 * normal root and not the commit root and we could deadlock.
3470 */
3471 if (btrfs_fs_closing(fs_info))
3472 return 0;
3473
3474 path = btrfs_alloc_path();
3475 if (!path)
3476 return 0;
3477
3478 inode = lookup_free_ino_inode(root, path);
3479 if (IS_ERR(inode))
3480 goto out;
3481
3482 if (root_gen != BTRFS_I(inode)->generation)
3483 goto out_put;
3484
3485 ret = __load_free_space_cache(root, inode, ctl, path, 0);
3486
3487 if (ret < 0)
3488 btrfs_err(fs_info,
3489 "failed to load free ino cache for root %llu",
3490 root->root_key.objectid);
3491 out_put:
3492 iput(inode);
3493 out:
3494 btrfs_free_path(path);
3495 return ret;
3496 }
3497
3498 int btrfs_write_out_ino_cache(struct btrfs_root *root,
3499 struct btrfs_trans_handle *trans,
3500 struct btrfs_path *path,
3501 struct inode *inode)
3502 {
3503 struct btrfs_fs_info *fs_info = root->fs_info;
3504 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
3505 int ret;
3506 struct btrfs_io_ctl io_ctl;
3507 bool release_metadata = true;
3508
3509 if (!btrfs_test_opt(fs_info, INODE_MAP_CACHE))
3510 return 0;
3511
3512 memset(&io_ctl, 0, sizeof(io_ctl));
3513 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, &io_ctl, trans);
3514 if (!ret) {
3515 /*
3516 * At this point writepages() didn't error out, so our metadata
3517 * reservation is released when the writeback finishes, at
3518 * inode.c:btrfs_finish_ordered_io(), regardless of it finishing
3519 * with or without an error.
3520 */
3521 release_metadata = false;
3522 ret = btrfs_wait_cache_io_root(root, trans, &io_ctl, path);
3523 }
3524
3525 if (ret) {
3526 if (release_metadata)
3527 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3528 inode->i_size, true);
3529 #ifdef DEBUG
3530 btrfs_err(fs_info,
3531 "failed to write free ino cache for root %llu",
3532 root->root_key.objectid);
3533 #endif
3534 }
3535
3536 return ret;
3537 }
3538
3539 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3540 /*
3541 * Use this if you need to make a bitmap or extent entry specifically, it
3542 * doesn't do any of the merging that add_free_space does, this acts a lot like
3543 * how the free space cache loading stuff works, so you can get really weird
3544 * configurations.
3545 */
3546 int test_add_free_space_entry(struct btrfs_block_group_cache *cache,
3547 u64 offset, u64 bytes, bool bitmap)
3548 {
3549 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3550 struct btrfs_free_space *info = NULL, *bitmap_info;
3551 void *map = NULL;
3552 u64 bytes_added;
3553 int ret;
3554
3555 again:
3556 if (!info) {
3557 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
3558 if (!info)
3559 return -ENOMEM;
3560 }
3561
3562 if (!bitmap) {
3563 spin_lock(&ctl->tree_lock);
3564 info->offset = offset;
3565 info->bytes = bytes;
3566 info->max_extent_size = 0;
3567 ret = link_free_space(ctl, info);
3568 spin_unlock(&ctl->tree_lock);
3569 if (ret)
3570 kmem_cache_free(btrfs_free_space_cachep, info);
3571 return ret;
3572 }
3573
3574 if (!map) {
3575 map = kzalloc(PAGE_SIZE, GFP_NOFS);
3576 if (!map) {
3577 kmem_cache_free(btrfs_free_space_cachep, info);
3578 return -ENOMEM;
3579 }
3580 }
3581
3582 spin_lock(&ctl->tree_lock);
3583 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3584 1, 0);
3585 if (!bitmap_info) {
3586 info->bitmap = map;
3587 map = NULL;
3588 add_new_bitmap(ctl, info, offset);
3589 bitmap_info = info;
3590 info = NULL;
3591 }
3592
3593 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
3594
3595 bytes -= bytes_added;
3596 offset += bytes_added;
3597 spin_unlock(&ctl->tree_lock);
3598
3599 if (bytes)
3600 goto again;
3601
3602 if (info)
3603 kmem_cache_free(btrfs_free_space_cachep, info);
3604 if (map)
3605 kfree(map);
3606 return 0;
3607 }
3608
3609 /*
3610 * Checks to see if the given range is in the free space cache. This is really
3611 * just used to check the absence of space, so if there is free space in the
3612 * range at all we will return 1.
3613 */
3614 int test_check_exists(struct btrfs_block_group_cache *cache,
3615 u64 offset, u64 bytes)
3616 {
3617 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3618 struct btrfs_free_space *info;
3619 int ret = 0;
3620
3621 spin_lock(&ctl->tree_lock);
3622 info = tree_search_offset(ctl, offset, 0, 0);
3623 if (!info) {
3624 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3625 1, 0);
3626 if (!info)
3627 goto out;
3628 }
3629
3630 have_info:
3631 if (info->bitmap) {
3632 u64 bit_off, bit_bytes;
3633 struct rb_node *n;
3634 struct btrfs_free_space *tmp;
3635
3636 bit_off = offset;
3637 bit_bytes = ctl->unit;
3638 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false);
3639 if (!ret) {
3640 if (bit_off == offset) {
3641 ret = 1;
3642 goto out;
3643 } else if (bit_off > offset &&
3644 offset + bytes > bit_off) {
3645 ret = 1;
3646 goto out;
3647 }
3648 }
3649
3650 n = rb_prev(&info->offset_index);
3651 while (n) {
3652 tmp = rb_entry(n, struct btrfs_free_space,
3653 offset_index);
3654 if (tmp->offset + tmp->bytes < offset)
3655 break;
3656 if (offset + bytes < tmp->offset) {
3657 n = rb_prev(&tmp->offset_index);
3658 continue;
3659 }
3660 info = tmp;
3661 goto have_info;
3662 }
3663
3664 n = rb_next(&info->offset_index);
3665 while (n) {
3666 tmp = rb_entry(n, struct btrfs_free_space,
3667 offset_index);
3668 if (offset + bytes < tmp->offset)
3669 break;
3670 if (tmp->offset + tmp->bytes < offset) {
3671 n = rb_next(&tmp->offset_index);
3672 continue;
3673 }
3674 info = tmp;
3675 goto have_info;
3676 }
3677
3678 ret = 0;
3679 goto out;
3680 }
3681
3682 if (info->offset == offset) {
3683 ret = 1;
3684 goto out;
3685 }
3686
3687 if (offset > info->offset && offset < info->offset + info->bytes)
3688 ret = 1;
3689 out:
3690 spin_unlock(&ctl->tree_lock);
3691 return ret;
3692 }
3693 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */
Linux with added FPC-III support