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 <linux/sched/mm.h>
14 #include "misc.h"
15 #include "ctree.h"
16 #include "free-space-cache.h"
17 #include "transaction.h"
18 #include "disk-io.h"
19 #include "extent_io.h"
20 #include "volumes.h"
21 #include "space-info.h"
22 #include "delalloc-space.h"
23 #include "block-group.h"
24 #include "discard.h"
25
26 #define BITS_PER_BITMAP (PAGE_SIZE * 8UL)
27 #define MAX_CACHE_BYTES_PER_GIG SZ_64K
28 #define FORCE_EXTENT_THRESHOLD SZ_1M
29
30 struct btrfs_trim_range {
31 u64 start;
32 u64 bytes;
33 struct list_head list;
34 };
35
36 static int link_free_space(struct btrfs_free_space_ctl *ctl,
37 struct btrfs_free_space *info);
38 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
39 struct btrfs_free_space *info);
40 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
41 struct btrfs_free_space *bitmap_info, u64 *offset,
42 u64 *bytes, bool for_alloc);
43 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
44 struct btrfs_free_space *bitmap_info);
45 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
46 struct btrfs_free_space *info, u64 offset,
47 u64 bytes);
48
__lookup_free_space_inode(struct btrfs_root * root,struct btrfs_path * path,u64 offset)49 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
50 struct btrfs_path *path,
51 u64 offset)
52 {
53 struct btrfs_fs_info *fs_info = root->fs_info;
54 struct btrfs_key key;
55 struct btrfs_key location;
56 struct btrfs_disk_key disk_key;
57 struct btrfs_free_space_header *header;
58 struct extent_buffer *leaf;
59 struct inode *inode = NULL;
60 unsigned nofs_flag;
61 int ret;
62
63 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
64 key.offset = offset;
65 key.type = 0;
66
67 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
68 if (ret < 0)
69 return ERR_PTR(ret);
70 if (ret > 0) {
71 btrfs_release_path(path);
72 return ERR_PTR(-ENOENT);
73 }
74
75 leaf = path->nodes[0];
76 header = btrfs_item_ptr(leaf, path->slots[0],
77 struct btrfs_free_space_header);
78 btrfs_free_space_key(leaf, header, &disk_key);
79 btrfs_disk_key_to_cpu(&location, &disk_key);
80 btrfs_release_path(path);
81
82 /*
83 * We are often under a trans handle at this point, so we need to make
84 * sure NOFS is set to keep us from deadlocking.
85 */
86 nofs_flag = memalloc_nofs_save();
87 inode = btrfs_iget_path(fs_info->sb, location.objectid, root, path);
88 btrfs_release_path(path);
89 memalloc_nofs_restore(nofs_flag);
90 if (IS_ERR(inode))
91 return inode;
92
93 mapping_set_gfp_mask(inode->i_mapping,
94 mapping_gfp_constraint(inode->i_mapping,
95 ~(__GFP_FS | __GFP_HIGHMEM)));
96
97 return inode;
98 }
99
lookup_free_space_inode(struct btrfs_block_group * block_group,struct btrfs_path * path)100 struct inode *lookup_free_space_inode(struct btrfs_block_group *block_group,
101 struct btrfs_path *path)
102 {
103 struct btrfs_fs_info *fs_info = block_group->fs_info;
104 struct inode *inode = NULL;
105 u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
106
107 spin_lock(&block_group->lock);
108 if (block_group->inode)
109 inode = igrab(block_group->inode);
110 spin_unlock(&block_group->lock);
111 if (inode)
112 return inode;
113
114 inode = __lookup_free_space_inode(fs_info->tree_root, path,
115 block_group->start);
116 if (IS_ERR(inode))
117 return inode;
118
119 spin_lock(&block_group->lock);
120 if (!((BTRFS_I(inode)->flags & flags) == flags)) {
121 btrfs_info(fs_info, "Old style space inode found, converting.");
122 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
123 BTRFS_INODE_NODATACOW;
124 block_group->disk_cache_state = BTRFS_DC_CLEAR;
125 }
126
127 if (!block_group->iref) {
128 block_group->inode = igrab(inode);
129 block_group->iref = 1;
130 }
131 spin_unlock(&block_group->lock);
132
133 return inode;
134 }
135
__create_free_space_inode(struct btrfs_root * root,struct btrfs_trans_handle * trans,struct btrfs_path * path,u64 ino,u64 offset)136 static int __create_free_space_inode(struct btrfs_root *root,
137 struct btrfs_trans_handle *trans,
138 struct btrfs_path *path,
139 u64 ino, u64 offset)
140 {
141 struct btrfs_key key;
142 struct btrfs_disk_key disk_key;
143 struct btrfs_free_space_header *header;
144 struct btrfs_inode_item *inode_item;
145 struct extent_buffer *leaf;
146 /* We inline CRCs for the free disk space cache */
147 const u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC |
148 BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
149 int ret;
150
151 ret = btrfs_insert_empty_inode(trans, root, path, ino);
152 if (ret)
153 return ret;
154
155 leaf = path->nodes[0];
156 inode_item = btrfs_item_ptr(leaf, path->slots[0],
157 struct btrfs_inode_item);
158 btrfs_item_key(leaf, &disk_key, path->slots[0]);
159 memzero_extent_buffer(leaf, (unsigned long)inode_item,
160 sizeof(*inode_item));
161 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
162 btrfs_set_inode_size(leaf, inode_item, 0);
163 btrfs_set_inode_nbytes(leaf, inode_item, 0);
164 btrfs_set_inode_uid(leaf, inode_item, 0);
165 btrfs_set_inode_gid(leaf, inode_item, 0);
166 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
167 btrfs_set_inode_flags(leaf, inode_item, flags);
168 btrfs_set_inode_nlink(leaf, inode_item, 1);
169 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
170 btrfs_set_inode_block_group(leaf, inode_item, offset);
171 btrfs_mark_buffer_dirty(leaf);
172 btrfs_release_path(path);
173
174 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
175 key.offset = offset;
176 key.type = 0;
177 ret = btrfs_insert_empty_item(trans, root, path, &key,
178 sizeof(struct btrfs_free_space_header));
179 if (ret < 0) {
180 btrfs_release_path(path);
181 return ret;
182 }
183
184 leaf = path->nodes[0];
185 header = btrfs_item_ptr(leaf, path->slots[0],
186 struct btrfs_free_space_header);
187 memzero_extent_buffer(leaf, (unsigned long)header, sizeof(*header));
188 btrfs_set_free_space_key(leaf, header, &disk_key);
189 btrfs_mark_buffer_dirty(leaf);
190 btrfs_release_path(path);
191
192 return 0;
193 }
194
create_free_space_inode(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path)195 int create_free_space_inode(struct btrfs_trans_handle *trans,
196 struct btrfs_block_group *block_group,
197 struct btrfs_path *path)
198 {
199 int ret;
200 u64 ino;
201
202 ret = btrfs_get_free_objectid(trans->fs_info->tree_root, &ino);
203 if (ret < 0)
204 return ret;
205
206 return __create_free_space_inode(trans->fs_info->tree_root, trans, path,
207 ino, block_group->start);
208 }
209
210 /*
211 * inode is an optional sink: if it is NULL, btrfs_remove_free_space_inode
212 * handles lookup, otherwise it takes ownership and iputs the inode.
213 * Don't reuse an inode pointer after passing it into this function.
214 */
btrfs_remove_free_space_inode(struct btrfs_trans_handle * trans,struct inode * inode,struct btrfs_block_group * block_group)215 int btrfs_remove_free_space_inode(struct btrfs_trans_handle *trans,
216 struct inode *inode,
217 struct btrfs_block_group *block_group)
218 {
219 struct btrfs_path *path;
220 struct btrfs_key key;
221 int ret = 0;
222
223 path = btrfs_alloc_path();
224 if (!path)
225 return -ENOMEM;
226
227 if (!inode)
228 inode = lookup_free_space_inode(block_group, path);
229 if (IS_ERR(inode)) {
230 if (PTR_ERR(inode) != -ENOENT)
231 ret = PTR_ERR(inode);
232 goto out;
233 }
234 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
235 if (ret) {
236 btrfs_add_delayed_iput(inode);
237 goto out;
238 }
239 clear_nlink(inode);
240 /* One for the block groups ref */
241 spin_lock(&block_group->lock);
242 if (block_group->iref) {
243 block_group->iref = 0;
244 block_group->inode = NULL;
245 spin_unlock(&block_group->lock);
246 iput(inode);
247 } else {
248 spin_unlock(&block_group->lock);
249 }
250 /* One for the lookup ref */
251 btrfs_add_delayed_iput(inode);
252
253 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
254 key.type = 0;
255 key.offset = block_group->start;
256 ret = btrfs_search_slot(trans, trans->fs_info->tree_root, &key, path,
257 -1, 1);
258 if (ret) {
259 if (ret > 0)
260 ret = 0;
261 goto out;
262 }
263 ret = btrfs_del_item(trans, trans->fs_info->tree_root, path);
264 out:
265 btrfs_free_path(path);
266 return ret;
267 }
268
btrfs_check_trunc_cache_free_space(struct btrfs_fs_info * fs_info,struct btrfs_block_rsv * rsv)269 int btrfs_check_trunc_cache_free_space(struct btrfs_fs_info *fs_info,
270 struct btrfs_block_rsv *rsv)
271 {
272 u64 needed_bytes;
273 int ret;
274
275 /* 1 for slack space, 1 for updating the inode */
276 needed_bytes = btrfs_calc_insert_metadata_size(fs_info, 1) +
277 btrfs_calc_metadata_size(fs_info, 1);
278
279 spin_lock(&rsv->lock);
280 if (rsv->reserved < needed_bytes)
281 ret = -ENOSPC;
282 else
283 ret = 0;
284 spin_unlock(&rsv->lock);
285 return ret;
286 }
287
btrfs_truncate_free_space_cache(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct inode * inode)288 int btrfs_truncate_free_space_cache(struct btrfs_trans_handle *trans,
289 struct btrfs_block_group *block_group,
290 struct inode *inode)
291 {
292 struct btrfs_root *root = BTRFS_I(inode)->root;
293 int ret = 0;
294 bool locked = false;
295
296 if (block_group) {
297 struct btrfs_path *path = btrfs_alloc_path();
298
299 if (!path) {
300 ret = -ENOMEM;
301 goto fail;
302 }
303 locked = true;
304 mutex_lock(&trans->transaction->cache_write_mutex);
305 if (!list_empty(&block_group->io_list)) {
306 list_del_init(&block_group->io_list);
307
308 btrfs_wait_cache_io(trans, block_group, path);
309 btrfs_put_block_group(block_group);
310 }
311
312 /*
313 * now that we've truncated the cache away, its no longer
314 * setup or written
315 */
316 spin_lock(&block_group->lock);
317 block_group->disk_cache_state = BTRFS_DC_CLEAR;
318 spin_unlock(&block_group->lock);
319 btrfs_free_path(path);
320 }
321
322 btrfs_i_size_write(BTRFS_I(inode), 0);
323 truncate_pagecache(inode, 0);
324
325 /*
326 * We skip the throttling logic for free space cache inodes, so we don't
327 * need to check for -EAGAIN.
328 */
329 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
330 0, BTRFS_EXTENT_DATA_KEY, NULL);
331 if (ret)
332 goto fail;
333
334 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
335
336 fail:
337 if (locked)
338 mutex_unlock(&trans->transaction->cache_write_mutex);
339 if (ret)
340 btrfs_abort_transaction(trans, ret);
341
342 return ret;
343 }
344
readahead_cache(struct inode * inode)345 static void readahead_cache(struct inode *inode)
346 {
347 struct file_ra_state ra;
348 unsigned long last_index;
349
350 file_ra_state_init(&ra, inode->i_mapping);
351 last_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
352
353 page_cache_sync_readahead(inode->i_mapping, &ra, NULL, 0, last_index);
354 }
355
io_ctl_init(struct btrfs_io_ctl * io_ctl,struct inode * inode,int write)356 static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode,
357 int write)
358 {
359 int num_pages;
360
361 num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
362
363 /* Make sure we can fit our crcs and generation into the first page */
364 if (write && (num_pages * sizeof(u32) + sizeof(u64)) > PAGE_SIZE)
365 return -ENOSPC;
366
367 memset(io_ctl, 0, sizeof(struct btrfs_io_ctl));
368
369 io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS);
370 if (!io_ctl->pages)
371 return -ENOMEM;
372
373 io_ctl->num_pages = num_pages;
374 io_ctl->fs_info = btrfs_sb(inode->i_sb);
375 io_ctl->inode = inode;
376
377 return 0;
378 }
379 ALLOW_ERROR_INJECTION(io_ctl_init, ERRNO);
380
io_ctl_free(struct btrfs_io_ctl * io_ctl)381 static void io_ctl_free(struct btrfs_io_ctl *io_ctl)
382 {
383 kfree(io_ctl->pages);
384 io_ctl->pages = NULL;
385 }
386
io_ctl_unmap_page(struct btrfs_io_ctl * io_ctl)387 static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl)
388 {
389 if (io_ctl->cur) {
390 io_ctl->cur = NULL;
391 io_ctl->orig = NULL;
392 }
393 }
394
io_ctl_map_page(struct btrfs_io_ctl * io_ctl,int clear)395 static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear)
396 {
397 ASSERT(io_ctl->index < io_ctl->num_pages);
398 io_ctl->page = io_ctl->pages[io_ctl->index++];
399 io_ctl->cur = page_address(io_ctl->page);
400 io_ctl->orig = io_ctl->cur;
401 io_ctl->size = PAGE_SIZE;
402 if (clear)
403 clear_page(io_ctl->cur);
404 }
405
io_ctl_drop_pages(struct btrfs_io_ctl * io_ctl)406 static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl)
407 {
408 int i;
409
410 io_ctl_unmap_page(io_ctl);
411
412 for (i = 0; i < io_ctl->num_pages; i++) {
413 if (io_ctl->pages[i]) {
414 ClearPageChecked(io_ctl->pages[i]);
415 unlock_page(io_ctl->pages[i]);
416 put_page(io_ctl->pages[i]);
417 }
418 }
419 }
420
io_ctl_prepare_pages(struct btrfs_io_ctl * io_ctl,bool uptodate)421 static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, bool uptodate)
422 {
423 struct page *page;
424 struct inode *inode = io_ctl->inode;
425 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
426 int i;
427
428 for (i = 0; i < io_ctl->num_pages; i++) {
429 int ret;
430
431 page = find_or_create_page(inode->i_mapping, i, mask);
432 if (!page) {
433 io_ctl_drop_pages(io_ctl);
434 return -ENOMEM;
435 }
436
437 ret = set_page_extent_mapped(page);
438 if (ret < 0) {
439 unlock_page(page);
440 put_page(page);
441 io_ctl_drop_pages(io_ctl);
442 return ret;
443 }
444
445 io_ctl->pages[i] = page;
446 if (uptodate && !PageUptodate(page)) {
447 btrfs_readpage(NULL, page);
448 lock_page(page);
449 if (page->mapping != inode->i_mapping) {
450 btrfs_err(BTRFS_I(inode)->root->fs_info,
451 "free space cache page truncated");
452 io_ctl_drop_pages(io_ctl);
453 return -EIO;
454 }
455 if (!PageUptodate(page)) {
456 btrfs_err(BTRFS_I(inode)->root->fs_info,
457 "error reading free space cache");
458 io_ctl_drop_pages(io_ctl);
459 return -EIO;
460 }
461 }
462 }
463
464 for (i = 0; i < io_ctl->num_pages; i++)
465 clear_page_dirty_for_io(io_ctl->pages[i]);
466
467 return 0;
468 }
469
io_ctl_set_generation(struct btrfs_io_ctl * io_ctl,u64 generation)470 static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
471 {
472 io_ctl_map_page(io_ctl, 1);
473
474 /*
475 * Skip the csum areas. If we don't check crcs then we just have a
476 * 64bit chunk at the front of the first page.
477 */
478 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
479 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
480
481 put_unaligned_le64(generation, io_ctl->cur);
482 io_ctl->cur += sizeof(u64);
483 }
484
io_ctl_check_generation(struct btrfs_io_ctl * io_ctl,u64 generation)485 static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
486 {
487 u64 cache_gen;
488
489 /*
490 * Skip the crc area. If we don't check crcs then we just have a 64bit
491 * chunk at the front of the first page.
492 */
493 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
494 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
495
496 cache_gen = get_unaligned_le64(io_ctl->cur);
497 if (cache_gen != generation) {
498 btrfs_err_rl(io_ctl->fs_info,
499 "space cache generation (%llu) does not match inode (%llu)",
500 cache_gen, generation);
501 io_ctl_unmap_page(io_ctl);
502 return -EIO;
503 }
504 io_ctl->cur += sizeof(u64);
505 return 0;
506 }
507
io_ctl_set_crc(struct btrfs_io_ctl * io_ctl,int index)508 static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
509 {
510 u32 *tmp;
511 u32 crc = ~(u32)0;
512 unsigned offset = 0;
513
514 if (index == 0)
515 offset = sizeof(u32) * io_ctl->num_pages;
516
517 crc = btrfs_crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
518 btrfs_crc32c_final(crc, (u8 *)&crc);
519 io_ctl_unmap_page(io_ctl);
520 tmp = page_address(io_ctl->pages[0]);
521 tmp += index;
522 *tmp = crc;
523 }
524
io_ctl_check_crc(struct btrfs_io_ctl * io_ctl,int index)525 static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
526 {
527 u32 *tmp, val;
528 u32 crc = ~(u32)0;
529 unsigned offset = 0;
530
531 if (index == 0)
532 offset = sizeof(u32) * io_ctl->num_pages;
533
534 tmp = page_address(io_ctl->pages[0]);
535 tmp += index;
536 val = *tmp;
537
538 io_ctl_map_page(io_ctl, 0);
539 crc = btrfs_crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
540 btrfs_crc32c_final(crc, (u8 *)&crc);
541 if (val != crc) {
542 btrfs_err_rl(io_ctl->fs_info,
543 "csum mismatch on free space cache");
544 io_ctl_unmap_page(io_ctl);
545 return -EIO;
546 }
547
548 return 0;
549 }
550
io_ctl_add_entry(struct btrfs_io_ctl * io_ctl,u64 offset,u64 bytes,void * bitmap)551 static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
552 void *bitmap)
553 {
554 struct btrfs_free_space_entry *entry;
555
556 if (!io_ctl->cur)
557 return -ENOSPC;
558
559 entry = io_ctl->cur;
560 put_unaligned_le64(offset, &entry->offset);
561 put_unaligned_le64(bytes, &entry->bytes);
562 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
563 BTRFS_FREE_SPACE_EXTENT;
564 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
565 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
566
567 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
568 return 0;
569
570 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
571
572 /* No more pages to map */
573 if (io_ctl->index >= io_ctl->num_pages)
574 return 0;
575
576 /* map the next page */
577 io_ctl_map_page(io_ctl, 1);
578 return 0;
579 }
580
io_ctl_add_bitmap(struct btrfs_io_ctl * io_ctl,void * bitmap)581 static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
582 {
583 if (!io_ctl->cur)
584 return -ENOSPC;
585
586 /*
587 * If we aren't at the start of the current page, unmap this one and
588 * map the next one if there is any left.
589 */
590 if (io_ctl->cur != io_ctl->orig) {
591 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
592 if (io_ctl->index >= io_ctl->num_pages)
593 return -ENOSPC;
594 io_ctl_map_page(io_ctl, 0);
595 }
596
597 copy_page(io_ctl->cur, bitmap);
598 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
599 if (io_ctl->index < io_ctl->num_pages)
600 io_ctl_map_page(io_ctl, 0);
601 return 0;
602 }
603
io_ctl_zero_remaining_pages(struct btrfs_io_ctl * io_ctl)604 static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
605 {
606 /*
607 * If we're not on the boundary we know we've modified the page and we
608 * need to crc the page.
609 */
610 if (io_ctl->cur != io_ctl->orig)
611 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
612 else
613 io_ctl_unmap_page(io_ctl);
614
615 while (io_ctl->index < io_ctl->num_pages) {
616 io_ctl_map_page(io_ctl, 1);
617 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
618 }
619 }
620
io_ctl_read_entry(struct btrfs_io_ctl * io_ctl,struct btrfs_free_space * entry,u8 * type)621 static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
622 struct btrfs_free_space *entry, u8 *type)
623 {
624 struct btrfs_free_space_entry *e;
625 int ret;
626
627 if (!io_ctl->cur) {
628 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
629 if (ret)
630 return ret;
631 }
632
633 e = io_ctl->cur;
634 entry->offset = get_unaligned_le64(&e->offset);
635 entry->bytes = get_unaligned_le64(&e->bytes);
636 *type = e->type;
637 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
638 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
639
640 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
641 return 0;
642
643 io_ctl_unmap_page(io_ctl);
644
645 return 0;
646 }
647
io_ctl_read_bitmap(struct btrfs_io_ctl * io_ctl,struct btrfs_free_space * entry)648 static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
649 struct btrfs_free_space *entry)
650 {
651 int ret;
652
653 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
654 if (ret)
655 return ret;
656
657 copy_page(entry->bitmap, io_ctl->cur);
658 io_ctl_unmap_page(io_ctl);
659
660 return 0;
661 }
662
recalculate_thresholds(struct btrfs_free_space_ctl * ctl)663 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
664 {
665 struct btrfs_block_group *block_group = ctl->private;
666 u64 max_bytes;
667 u64 bitmap_bytes;
668 u64 extent_bytes;
669 u64 size = block_group->length;
670 u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
671 u64 max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
672
673 max_bitmaps = max_t(u64, max_bitmaps, 1);
674
675 ASSERT(ctl->total_bitmaps <= max_bitmaps);
676
677 /*
678 * We are trying to keep the total amount of memory used per 1GiB of
679 * space to be MAX_CACHE_BYTES_PER_GIG. However, with a reclamation
680 * mechanism of pulling extents >= FORCE_EXTENT_THRESHOLD out of
681 * bitmaps, we may end up using more memory than this.
682 */
683 if (size < SZ_1G)
684 max_bytes = MAX_CACHE_BYTES_PER_GIG;
685 else
686 max_bytes = MAX_CACHE_BYTES_PER_GIG * div_u64(size, SZ_1G);
687
688 bitmap_bytes = ctl->total_bitmaps * ctl->unit;
689
690 /*
691 * we want the extent entry threshold to always be at most 1/2 the max
692 * bytes we can have, or whatever is less than that.
693 */
694 extent_bytes = max_bytes - bitmap_bytes;
695 extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
696
697 ctl->extents_thresh =
698 div_u64(extent_bytes, sizeof(struct btrfs_free_space));
699 }
700
__load_free_space_cache(struct btrfs_root * root,struct inode * inode,struct btrfs_free_space_ctl * ctl,struct btrfs_path * path,u64 offset)701 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
702 struct btrfs_free_space_ctl *ctl,
703 struct btrfs_path *path, u64 offset)
704 {
705 struct btrfs_fs_info *fs_info = root->fs_info;
706 struct btrfs_free_space_header *header;
707 struct extent_buffer *leaf;
708 struct btrfs_io_ctl io_ctl;
709 struct btrfs_key key;
710 struct btrfs_free_space *e, *n;
711 LIST_HEAD(bitmaps);
712 u64 num_entries;
713 u64 num_bitmaps;
714 u64 generation;
715 u8 type;
716 int ret = 0;
717
718 /* Nothing in the space cache, goodbye */
719 if (!i_size_read(inode))
720 return 0;
721
722 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
723 key.offset = offset;
724 key.type = 0;
725
726 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
727 if (ret < 0)
728 return 0;
729 else if (ret > 0) {
730 btrfs_release_path(path);
731 return 0;
732 }
733
734 ret = -1;
735
736 leaf = path->nodes[0];
737 header = btrfs_item_ptr(leaf, path->slots[0],
738 struct btrfs_free_space_header);
739 num_entries = btrfs_free_space_entries(leaf, header);
740 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
741 generation = btrfs_free_space_generation(leaf, header);
742 btrfs_release_path(path);
743
744 if (!BTRFS_I(inode)->generation) {
745 btrfs_info(fs_info,
746 "the free space cache file (%llu) is invalid, skip it",
747 offset);
748 return 0;
749 }
750
751 if (BTRFS_I(inode)->generation != generation) {
752 btrfs_err(fs_info,
753 "free space inode generation (%llu) did not match free space cache generation (%llu)",
754 BTRFS_I(inode)->generation, generation);
755 return 0;
756 }
757
758 if (!num_entries)
759 return 0;
760
761 ret = io_ctl_init(&io_ctl, inode, 0);
762 if (ret)
763 return ret;
764
765 readahead_cache(inode);
766
767 ret = io_ctl_prepare_pages(&io_ctl, true);
768 if (ret)
769 goto out;
770
771 ret = io_ctl_check_crc(&io_ctl, 0);
772 if (ret)
773 goto free_cache;
774
775 ret = io_ctl_check_generation(&io_ctl, generation);
776 if (ret)
777 goto free_cache;
778
779 while (num_entries) {
780 e = kmem_cache_zalloc(btrfs_free_space_cachep,
781 GFP_NOFS);
782 if (!e) {
783 ret = -ENOMEM;
784 goto free_cache;
785 }
786
787 ret = io_ctl_read_entry(&io_ctl, e, &type);
788 if (ret) {
789 kmem_cache_free(btrfs_free_space_cachep, e);
790 goto free_cache;
791 }
792
793 if (!e->bytes) {
794 ret = -1;
795 kmem_cache_free(btrfs_free_space_cachep, e);
796 goto free_cache;
797 }
798
799 if (type == BTRFS_FREE_SPACE_EXTENT) {
800 spin_lock(&ctl->tree_lock);
801 ret = link_free_space(ctl, e);
802 spin_unlock(&ctl->tree_lock);
803 if (ret) {
804 btrfs_err(fs_info,
805 "Duplicate entries in free space cache, dumping");
806 kmem_cache_free(btrfs_free_space_cachep, e);
807 goto free_cache;
808 }
809 } else {
810 ASSERT(num_bitmaps);
811 num_bitmaps--;
812 e->bitmap = kmem_cache_zalloc(
813 btrfs_free_space_bitmap_cachep, GFP_NOFS);
814 if (!e->bitmap) {
815 ret = -ENOMEM;
816 kmem_cache_free(
817 btrfs_free_space_cachep, e);
818 goto free_cache;
819 }
820 spin_lock(&ctl->tree_lock);
821 ret = link_free_space(ctl, e);
822 ctl->total_bitmaps++;
823 recalculate_thresholds(ctl);
824 spin_unlock(&ctl->tree_lock);
825 if (ret) {
826 btrfs_err(fs_info,
827 "Duplicate entries in free space cache, dumping");
828 kmem_cache_free(btrfs_free_space_cachep, e);
829 goto free_cache;
830 }
831 list_add_tail(&e->list, &bitmaps);
832 }
833
834 num_entries--;
835 }
836
837 io_ctl_unmap_page(&io_ctl);
838
839 /*
840 * We add the bitmaps at the end of the entries in order that
841 * the bitmap entries are added to the cache.
842 */
843 list_for_each_entry_safe(e, n, &bitmaps, list) {
844 list_del_init(&e->list);
845 ret = io_ctl_read_bitmap(&io_ctl, e);
846 if (ret)
847 goto free_cache;
848 }
849
850 io_ctl_drop_pages(&io_ctl);
851 ret = 1;
852 out:
853 io_ctl_free(&io_ctl);
854 return ret;
855 free_cache:
856 io_ctl_drop_pages(&io_ctl);
857 __btrfs_remove_free_space_cache(ctl);
858 goto out;
859 }
860
copy_free_space_cache(struct btrfs_block_group * block_group,struct btrfs_free_space_ctl * ctl)861 static int copy_free_space_cache(struct btrfs_block_group *block_group,
862 struct btrfs_free_space_ctl *ctl)
863 {
864 struct btrfs_free_space *info;
865 struct rb_node *n;
866 int ret = 0;
867
868 while (!ret && (n = rb_first(&ctl->free_space_offset)) != NULL) {
869 info = rb_entry(n, struct btrfs_free_space, offset_index);
870 if (!info->bitmap) {
871 unlink_free_space(ctl, info);
872 ret = btrfs_add_free_space(block_group, info->offset,
873 info->bytes);
874 kmem_cache_free(btrfs_free_space_cachep, info);
875 } else {
876 u64 offset = info->offset;
877 u64 bytes = ctl->unit;
878
879 while (search_bitmap(ctl, info, &offset, &bytes,
880 false) == 0) {
881 ret = btrfs_add_free_space(block_group, offset,
882 bytes);
883 if (ret)
884 break;
885 bitmap_clear_bits(ctl, info, offset, bytes);
886 offset = info->offset;
887 bytes = ctl->unit;
888 }
889 free_bitmap(ctl, info);
890 }
891 cond_resched();
892 }
893 return ret;
894 }
895
load_free_space_cache(struct btrfs_block_group * block_group)896 int load_free_space_cache(struct btrfs_block_group *block_group)
897 {
898 struct btrfs_fs_info *fs_info = block_group->fs_info;
899 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
900 struct btrfs_free_space_ctl tmp_ctl = {};
901 struct inode *inode;
902 struct btrfs_path *path;
903 int ret = 0;
904 bool matched;
905 u64 used = block_group->used;
906
907 /*
908 * Because we could potentially discard our loaded free space, we want
909 * to load everything into a temporary structure first, and then if it's
910 * valid copy it all into the actual free space ctl.
911 */
912 btrfs_init_free_space_ctl(block_group, &tmp_ctl);
913
914 /*
915 * If this block group has been marked to be cleared for one reason or
916 * another then we can't trust the on disk cache, so just return.
917 */
918 spin_lock(&block_group->lock);
919 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
920 spin_unlock(&block_group->lock);
921 return 0;
922 }
923 spin_unlock(&block_group->lock);
924
925 path = btrfs_alloc_path();
926 if (!path)
927 return 0;
928 path->search_commit_root = 1;
929 path->skip_locking = 1;
930
931 /*
932 * We must pass a path with search_commit_root set to btrfs_iget in
933 * order to avoid a deadlock when allocating extents for the tree root.
934 *
935 * When we are COWing an extent buffer from the tree root, when looking
936 * for a free extent, at extent-tree.c:find_free_extent(), we can find
937 * block group without its free space cache loaded. When we find one
938 * we must load its space cache which requires reading its free space
939 * cache's inode item from the root tree. If this inode item is located
940 * in the same leaf that we started COWing before, then we end up in
941 * deadlock on the extent buffer (trying to read lock it when we
942 * previously write locked it).
943 *
944 * It's safe to read the inode item using the commit root because
945 * block groups, once loaded, stay in memory forever (until they are
946 * removed) as well as their space caches once loaded. New block groups
947 * once created get their ->cached field set to BTRFS_CACHE_FINISHED so
948 * we will never try to read their inode item while the fs is mounted.
949 */
950 inode = lookup_free_space_inode(block_group, path);
951 if (IS_ERR(inode)) {
952 btrfs_free_path(path);
953 return 0;
954 }
955
956 /* We may have converted the inode and made the cache invalid. */
957 spin_lock(&block_group->lock);
958 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
959 spin_unlock(&block_group->lock);
960 btrfs_free_path(path);
961 goto out;
962 }
963 spin_unlock(&block_group->lock);
964
965 ret = __load_free_space_cache(fs_info->tree_root, inode, &tmp_ctl,
966 path, block_group->start);
967 btrfs_free_path(path);
968 if (ret <= 0)
969 goto out;
970
971 matched = (tmp_ctl.free_space == (block_group->length - used -
972 block_group->bytes_super));
973
974 if (matched) {
975 ret = copy_free_space_cache(block_group, &tmp_ctl);
976 /*
977 * ret == 1 means we successfully loaded the free space cache,
978 * so we need to re-set it here.
979 */
980 if (ret == 0)
981 ret = 1;
982 } else {
983 __btrfs_remove_free_space_cache(&tmp_ctl);
984 btrfs_warn(fs_info,
985 "block group %llu has wrong amount of free space",
986 block_group->start);
987 ret = -1;
988 }
989 out:
990 if (ret < 0) {
991 /* This cache is bogus, make sure it gets cleared */
992 spin_lock(&block_group->lock);
993 block_group->disk_cache_state = BTRFS_DC_CLEAR;
994 spin_unlock(&block_group->lock);
995 ret = 0;
996
997 btrfs_warn(fs_info,
998 "failed to load free space cache for block group %llu, rebuilding it now",
999 block_group->start);
1000 }
1001
1002 spin_lock(&ctl->tree_lock);
1003 btrfs_discard_update_discardable(block_group);
1004 spin_unlock(&ctl->tree_lock);
1005 iput(inode);
1006 return ret;
1007 }
1008
1009 static noinline_for_stack
write_cache_extent_entries(struct btrfs_io_ctl * io_ctl,struct btrfs_free_space_ctl * ctl,struct btrfs_block_group * block_group,int * entries,int * bitmaps,struct list_head * bitmap_list)1010 int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
1011 struct btrfs_free_space_ctl *ctl,
1012 struct btrfs_block_group *block_group,
1013 int *entries, int *bitmaps,
1014 struct list_head *bitmap_list)
1015 {
1016 int ret;
1017 struct btrfs_free_cluster *cluster = NULL;
1018 struct btrfs_free_cluster *cluster_locked = NULL;
1019 struct rb_node *node = rb_first(&ctl->free_space_offset);
1020 struct btrfs_trim_range *trim_entry;
1021
1022 /* Get the cluster for this block_group if it exists */
1023 if (block_group && !list_empty(&block_group->cluster_list)) {
1024 cluster = list_entry(block_group->cluster_list.next,
1025 struct btrfs_free_cluster,
1026 block_group_list);
1027 }
1028
1029 if (!node && cluster) {
1030 cluster_locked = cluster;
1031 spin_lock(&cluster_locked->lock);
1032 node = rb_first(&cluster->root);
1033 cluster = NULL;
1034 }
1035
1036 /* Write out the extent entries */
1037 while (node) {
1038 struct btrfs_free_space *e;
1039
1040 e = rb_entry(node, struct btrfs_free_space, offset_index);
1041 *entries += 1;
1042
1043 ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
1044 e->bitmap);
1045 if (ret)
1046 goto fail;
1047
1048 if (e->bitmap) {
1049 list_add_tail(&e->list, bitmap_list);
1050 *bitmaps += 1;
1051 }
1052 node = rb_next(node);
1053 if (!node && cluster) {
1054 node = rb_first(&cluster->root);
1055 cluster_locked = cluster;
1056 spin_lock(&cluster_locked->lock);
1057 cluster = NULL;
1058 }
1059 }
1060 if (cluster_locked) {
1061 spin_unlock(&cluster_locked->lock);
1062 cluster_locked = NULL;
1063 }
1064
1065 /*
1066 * Make sure we don't miss any range that was removed from our rbtree
1067 * because trimming is running. Otherwise after a umount+mount (or crash
1068 * after committing the transaction) we would leak free space and get
1069 * an inconsistent free space cache report from fsck.
1070 */
1071 list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
1072 ret = io_ctl_add_entry(io_ctl, trim_entry->start,
1073 trim_entry->bytes, NULL);
1074 if (ret)
1075 goto fail;
1076 *entries += 1;
1077 }
1078
1079 return 0;
1080 fail:
1081 if (cluster_locked)
1082 spin_unlock(&cluster_locked->lock);
1083 return -ENOSPC;
1084 }
1085
1086 static noinline_for_stack int
update_cache_item(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct inode * inode,struct btrfs_path * path,u64 offset,int entries,int bitmaps)1087 update_cache_item(struct btrfs_trans_handle *trans,
1088 struct btrfs_root *root,
1089 struct inode *inode,
1090 struct btrfs_path *path, u64 offset,
1091 int entries, int bitmaps)
1092 {
1093 struct btrfs_key key;
1094 struct btrfs_free_space_header *header;
1095 struct extent_buffer *leaf;
1096 int ret;
1097
1098 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1099 key.offset = offset;
1100 key.type = 0;
1101
1102 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1103 if (ret < 0) {
1104 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1105 EXTENT_DELALLOC, 0, 0, NULL);
1106 goto fail;
1107 }
1108 leaf = path->nodes[0];
1109 if (ret > 0) {
1110 struct btrfs_key found_key;
1111 ASSERT(path->slots[0]);
1112 path->slots[0]--;
1113 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1114 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1115 found_key.offset != offset) {
1116 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1117 inode->i_size - 1, EXTENT_DELALLOC, 0,
1118 0, NULL);
1119 btrfs_release_path(path);
1120 goto fail;
1121 }
1122 }
1123
1124 BTRFS_I(inode)->generation = trans->transid;
1125 header = btrfs_item_ptr(leaf, path->slots[0],
1126 struct btrfs_free_space_header);
1127 btrfs_set_free_space_entries(leaf, header, entries);
1128 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1129 btrfs_set_free_space_generation(leaf, header, trans->transid);
1130 btrfs_mark_buffer_dirty(leaf);
1131 btrfs_release_path(path);
1132
1133 return 0;
1134
1135 fail:
1136 return -1;
1137 }
1138
write_pinned_extent_entries(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_io_ctl * io_ctl,int * entries)1139 static noinline_for_stack int write_pinned_extent_entries(
1140 struct btrfs_trans_handle *trans,
1141 struct btrfs_block_group *block_group,
1142 struct btrfs_io_ctl *io_ctl,
1143 int *entries)
1144 {
1145 u64 start, extent_start, extent_end, len;
1146 struct extent_io_tree *unpin = NULL;
1147 int ret;
1148
1149 if (!block_group)
1150 return 0;
1151
1152 /*
1153 * We want to add any pinned extents to our free space cache
1154 * so we don't leak the space
1155 *
1156 * We shouldn't have switched the pinned extents yet so this is the
1157 * right one
1158 */
1159 unpin = &trans->transaction->pinned_extents;
1160
1161 start = block_group->start;
1162
1163 while (start < block_group->start + block_group->length) {
1164 ret = find_first_extent_bit(unpin, start,
1165 &extent_start, &extent_end,
1166 EXTENT_DIRTY, NULL);
1167 if (ret)
1168 return 0;
1169
1170 /* This pinned extent is out of our range */
1171 if (extent_start >= block_group->start + block_group->length)
1172 return 0;
1173
1174 extent_start = max(extent_start, start);
1175 extent_end = min(block_group->start + block_group->length,
1176 extent_end + 1);
1177 len = extent_end - extent_start;
1178
1179 *entries += 1;
1180 ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1181 if (ret)
1182 return -ENOSPC;
1183
1184 start = extent_end;
1185 }
1186
1187 return 0;
1188 }
1189
1190 static noinline_for_stack int
write_bitmap_entries(struct btrfs_io_ctl * io_ctl,struct list_head * bitmap_list)1191 write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1192 {
1193 struct btrfs_free_space *entry, *next;
1194 int ret;
1195
1196 /* Write out the bitmaps */
1197 list_for_each_entry_safe(entry, next, bitmap_list, list) {
1198 ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1199 if (ret)
1200 return -ENOSPC;
1201 list_del_init(&entry->list);
1202 }
1203
1204 return 0;
1205 }
1206
flush_dirty_cache(struct inode * inode)1207 static int flush_dirty_cache(struct inode *inode)
1208 {
1209 int ret;
1210
1211 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1212 if (ret)
1213 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1214 EXTENT_DELALLOC, 0, 0, NULL);
1215
1216 return ret;
1217 }
1218
1219 static void noinline_for_stack
cleanup_bitmap_list(struct list_head * bitmap_list)1220 cleanup_bitmap_list(struct list_head *bitmap_list)
1221 {
1222 struct btrfs_free_space *entry, *next;
1223
1224 list_for_each_entry_safe(entry, next, bitmap_list, list)
1225 list_del_init(&entry->list);
1226 }
1227
1228 static void noinline_for_stack
cleanup_write_cache_enospc(struct inode * inode,struct btrfs_io_ctl * io_ctl,struct extent_state ** cached_state)1229 cleanup_write_cache_enospc(struct inode *inode,
1230 struct btrfs_io_ctl *io_ctl,
1231 struct extent_state **cached_state)
1232 {
1233 io_ctl_drop_pages(io_ctl);
1234 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1235 i_size_read(inode) - 1, cached_state);
1236 }
1237
__btrfs_wait_cache_io(struct btrfs_root * root,struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_io_ctl * io_ctl,struct btrfs_path * path,u64 offset)1238 static int __btrfs_wait_cache_io(struct btrfs_root *root,
1239 struct btrfs_trans_handle *trans,
1240 struct btrfs_block_group *block_group,
1241 struct btrfs_io_ctl *io_ctl,
1242 struct btrfs_path *path, u64 offset)
1243 {
1244 int ret;
1245 struct inode *inode = io_ctl->inode;
1246
1247 if (!inode)
1248 return 0;
1249
1250 /* Flush the dirty pages in the cache file. */
1251 ret = flush_dirty_cache(inode);
1252 if (ret)
1253 goto out;
1254
1255 /* Update the cache item to tell everyone this cache file is valid. */
1256 ret = update_cache_item(trans, root, inode, path, offset,
1257 io_ctl->entries, io_ctl->bitmaps);
1258 out:
1259 if (ret) {
1260 invalidate_inode_pages2(inode->i_mapping);
1261 BTRFS_I(inode)->generation = 0;
1262 if (block_group)
1263 btrfs_debug(root->fs_info,
1264 "failed to write free space cache for block group %llu error %d",
1265 block_group->start, ret);
1266 }
1267 btrfs_update_inode(trans, root, BTRFS_I(inode));
1268
1269 if (block_group) {
1270 /* the dirty list is protected by the dirty_bgs_lock */
1271 spin_lock(&trans->transaction->dirty_bgs_lock);
1272
1273 /* the disk_cache_state is protected by the block group lock */
1274 spin_lock(&block_group->lock);
1275
1276 /*
1277 * only mark this as written if we didn't get put back on
1278 * the dirty list while waiting for IO. Otherwise our
1279 * cache state won't be right, and we won't get written again
1280 */
1281 if (!ret && list_empty(&block_group->dirty_list))
1282 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1283 else if (ret)
1284 block_group->disk_cache_state = BTRFS_DC_ERROR;
1285
1286 spin_unlock(&block_group->lock);
1287 spin_unlock(&trans->transaction->dirty_bgs_lock);
1288 io_ctl->inode = NULL;
1289 iput(inode);
1290 }
1291
1292 return ret;
1293
1294 }
1295
btrfs_wait_cache_io(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path)1296 int btrfs_wait_cache_io(struct btrfs_trans_handle *trans,
1297 struct btrfs_block_group *block_group,
1298 struct btrfs_path *path)
1299 {
1300 return __btrfs_wait_cache_io(block_group->fs_info->tree_root, trans,
1301 block_group, &block_group->io_ctl,
1302 path, block_group->start);
1303 }
1304
1305 /**
1306 * Write out cached info to an inode
1307 *
1308 * @root: root the inode belongs to
1309 * @inode: freespace inode we are writing out
1310 * @ctl: free space cache we are going to write out
1311 * @block_group: block_group for this cache if it belongs to a block_group
1312 * @io_ctl: holds context for the io
1313 * @trans: the trans handle
1314 *
1315 * This function writes out a free space cache struct to disk for quick recovery
1316 * on mount. This will return 0 if it was successful in writing the cache out,
1317 * or an errno if it was not.
1318 */
__btrfs_write_out_cache(struct btrfs_root * root,struct inode * inode,struct btrfs_free_space_ctl * ctl,struct btrfs_block_group * block_group,struct btrfs_io_ctl * io_ctl,struct btrfs_trans_handle * trans)1319 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
1320 struct btrfs_free_space_ctl *ctl,
1321 struct btrfs_block_group *block_group,
1322 struct btrfs_io_ctl *io_ctl,
1323 struct btrfs_trans_handle *trans)
1324 {
1325 struct extent_state *cached_state = NULL;
1326 LIST_HEAD(bitmap_list);
1327 int entries = 0;
1328 int bitmaps = 0;
1329 int ret;
1330 int must_iput = 0;
1331
1332 if (!i_size_read(inode))
1333 return -EIO;
1334
1335 WARN_ON(io_ctl->pages);
1336 ret = io_ctl_init(io_ctl, inode, 1);
1337 if (ret)
1338 return ret;
1339
1340 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1341 down_write(&block_group->data_rwsem);
1342 spin_lock(&block_group->lock);
1343 if (block_group->delalloc_bytes) {
1344 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1345 spin_unlock(&block_group->lock);
1346 up_write(&block_group->data_rwsem);
1347 BTRFS_I(inode)->generation = 0;
1348 ret = 0;
1349 must_iput = 1;
1350 goto out;
1351 }
1352 spin_unlock(&block_group->lock);
1353 }
1354
1355 /* Lock all pages first so we can lock the extent safely. */
1356 ret = io_ctl_prepare_pages(io_ctl, false);
1357 if (ret)
1358 goto out_unlock;
1359
1360 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1361 &cached_state);
1362
1363 io_ctl_set_generation(io_ctl, trans->transid);
1364
1365 mutex_lock(&ctl->cache_writeout_mutex);
1366 /* Write out the extent entries in the free space cache */
1367 spin_lock(&ctl->tree_lock);
1368 ret = write_cache_extent_entries(io_ctl, ctl,
1369 block_group, &entries, &bitmaps,
1370 &bitmap_list);
1371 if (ret)
1372 goto out_nospc_locked;
1373
1374 /*
1375 * Some spaces that are freed in the current transaction are pinned,
1376 * they will be added into free space cache after the transaction is
1377 * committed, we shouldn't lose them.
1378 *
1379 * If this changes while we are working we'll get added back to
1380 * the dirty list and redo it. No locking needed
1381 */
1382 ret = write_pinned_extent_entries(trans, block_group, io_ctl, &entries);
1383 if (ret)
1384 goto out_nospc_locked;
1385
1386 /*
1387 * At last, we write out all the bitmaps and keep cache_writeout_mutex
1388 * locked while doing it because a concurrent trim can be manipulating
1389 * or freeing the bitmap.
1390 */
1391 ret = write_bitmap_entries(io_ctl, &bitmap_list);
1392 spin_unlock(&ctl->tree_lock);
1393 mutex_unlock(&ctl->cache_writeout_mutex);
1394 if (ret)
1395 goto out_nospc;
1396
1397 /* Zero out the rest of the pages just to make sure */
1398 io_ctl_zero_remaining_pages(io_ctl);
1399
1400 /* Everything is written out, now we dirty the pages in the file. */
1401 ret = btrfs_dirty_pages(BTRFS_I(inode), io_ctl->pages,
1402 io_ctl->num_pages, 0, i_size_read(inode),
1403 &cached_state, false);
1404 if (ret)
1405 goto out_nospc;
1406
1407 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1408 up_write(&block_group->data_rwsem);
1409 /*
1410 * Release the pages and unlock the extent, we will flush
1411 * them out later
1412 */
1413 io_ctl_drop_pages(io_ctl);
1414 io_ctl_free(io_ctl);
1415
1416 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1417 i_size_read(inode) - 1, &cached_state);
1418
1419 /*
1420 * at this point the pages are under IO and we're happy,
1421 * The caller is responsible for waiting on them and updating
1422 * the cache and the inode
1423 */
1424 io_ctl->entries = entries;
1425 io_ctl->bitmaps = bitmaps;
1426
1427 ret = btrfs_fdatawrite_range(inode, 0, (u64)-1);
1428 if (ret)
1429 goto out;
1430
1431 return 0;
1432
1433 out_nospc_locked:
1434 cleanup_bitmap_list(&bitmap_list);
1435 spin_unlock(&ctl->tree_lock);
1436 mutex_unlock(&ctl->cache_writeout_mutex);
1437
1438 out_nospc:
1439 cleanup_write_cache_enospc(inode, io_ctl, &cached_state);
1440
1441 out_unlock:
1442 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1443 up_write(&block_group->data_rwsem);
1444
1445 out:
1446 io_ctl->inode = NULL;
1447 io_ctl_free(io_ctl);
1448 if (ret) {
1449 invalidate_inode_pages2(inode->i_mapping);
1450 BTRFS_I(inode)->generation = 0;
1451 }
1452 btrfs_update_inode(trans, root, BTRFS_I(inode));
1453 if (must_iput)
1454 iput(inode);
1455 return ret;
1456 }
1457
btrfs_write_out_cache(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path)1458 int btrfs_write_out_cache(struct btrfs_trans_handle *trans,
1459 struct btrfs_block_group *block_group,
1460 struct btrfs_path *path)
1461 {
1462 struct btrfs_fs_info *fs_info = trans->fs_info;
1463 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1464 struct inode *inode;
1465 int ret = 0;
1466
1467 spin_lock(&block_group->lock);
1468 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1469 spin_unlock(&block_group->lock);
1470 return 0;
1471 }
1472 spin_unlock(&block_group->lock);
1473
1474 inode = lookup_free_space_inode(block_group, path);
1475 if (IS_ERR(inode))
1476 return 0;
1477
1478 ret = __btrfs_write_out_cache(fs_info->tree_root, inode, ctl,
1479 block_group, &block_group->io_ctl, trans);
1480 if (ret) {
1481 btrfs_debug(fs_info,
1482 "failed to write free space cache for block group %llu error %d",
1483 block_group->start, ret);
1484 spin_lock(&block_group->lock);
1485 block_group->disk_cache_state = BTRFS_DC_ERROR;
1486 spin_unlock(&block_group->lock);
1487
1488 block_group->io_ctl.inode = NULL;
1489 iput(inode);
1490 }
1491
1492 /*
1493 * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1494 * to wait for IO and put the inode
1495 */
1496
1497 return ret;
1498 }
1499
offset_to_bit(u64 bitmap_start,u32 unit,u64 offset)1500 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1501 u64 offset)
1502 {
1503 ASSERT(offset >= bitmap_start);
1504 offset -= bitmap_start;
1505 return (unsigned long)(div_u64(offset, unit));
1506 }
1507
bytes_to_bits(u64 bytes,u32 unit)1508 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1509 {
1510 return (unsigned long)(div_u64(bytes, unit));
1511 }
1512
offset_to_bitmap(struct btrfs_free_space_ctl * ctl,u64 offset)1513 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1514 u64 offset)
1515 {
1516 u64 bitmap_start;
1517 u64 bytes_per_bitmap;
1518
1519 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1520 bitmap_start = offset - ctl->start;
1521 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1522 bitmap_start *= bytes_per_bitmap;
1523 bitmap_start += ctl->start;
1524
1525 return bitmap_start;
1526 }
1527
tree_insert_offset(struct rb_root * root,u64 offset,struct rb_node * node,int bitmap)1528 static int tree_insert_offset(struct rb_root *root, u64 offset,
1529 struct rb_node *node, int bitmap)
1530 {
1531 struct rb_node **p = &root->rb_node;
1532 struct rb_node *parent = NULL;
1533 struct btrfs_free_space *info;
1534
1535 while (*p) {
1536 parent = *p;
1537 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1538
1539 if (offset < info->offset) {
1540 p = &(*p)->rb_left;
1541 } else if (offset > info->offset) {
1542 p = &(*p)->rb_right;
1543 } else {
1544 /*
1545 * we could have a bitmap entry and an extent entry
1546 * share the same offset. If this is the case, we want
1547 * the extent entry to always be found first if we do a
1548 * linear search through the tree, since we want to have
1549 * the quickest allocation time, and allocating from an
1550 * extent is faster than allocating from a bitmap. So
1551 * if we're inserting a bitmap and we find an entry at
1552 * this offset, we want to go right, or after this entry
1553 * logically. If we are inserting an extent and we've
1554 * found a bitmap, we want to go left, or before
1555 * logically.
1556 */
1557 if (bitmap) {
1558 if (info->bitmap) {
1559 WARN_ON_ONCE(1);
1560 return -EEXIST;
1561 }
1562 p = &(*p)->rb_right;
1563 } else {
1564 if (!info->bitmap) {
1565 WARN_ON_ONCE(1);
1566 return -EEXIST;
1567 }
1568 p = &(*p)->rb_left;
1569 }
1570 }
1571 }
1572
1573 rb_link_node(node, parent, p);
1574 rb_insert_color(node, root);
1575
1576 return 0;
1577 }
1578
1579 /*
1580 * searches the tree for the given offset.
1581 *
1582 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1583 * want a section that has at least bytes size and comes at or after the given
1584 * offset.
1585 */
1586 static struct btrfs_free_space *
tree_search_offset(struct btrfs_free_space_ctl * ctl,u64 offset,int bitmap_only,int fuzzy)1587 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1588 u64 offset, int bitmap_only, int fuzzy)
1589 {
1590 struct rb_node *n = ctl->free_space_offset.rb_node;
1591 struct btrfs_free_space *entry, *prev = NULL;
1592
1593 /* find entry that is closest to the 'offset' */
1594 while (1) {
1595 if (!n) {
1596 entry = NULL;
1597 break;
1598 }
1599
1600 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1601 prev = entry;
1602
1603 if (offset < entry->offset)
1604 n = n->rb_left;
1605 else if (offset > entry->offset)
1606 n = n->rb_right;
1607 else
1608 break;
1609 }
1610
1611 if (bitmap_only) {
1612 if (!entry)
1613 return NULL;
1614 if (entry->bitmap)
1615 return entry;
1616
1617 /*
1618 * bitmap entry and extent entry may share same offset,
1619 * in that case, bitmap entry comes after extent entry.
1620 */
1621 n = rb_next(n);
1622 if (!n)
1623 return NULL;
1624 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1625 if (entry->offset != offset)
1626 return NULL;
1627
1628 WARN_ON(!entry->bitmap);
1629 return entry;
1630 } else if (entry) {
1631 if (entry->bitmap) {
1632 /*
1633 * if previous extent entry covers the offset,
1634 * we should return it instead of the bitmap entry
1635 */
1636 n = rb_prev(&entry->offset_index);
1637 if (n) {
1638 prev = rb_entry(n, struct btrfs_free_space,
1639 offset_index);
1640 if (!prev->bitmap &&
1641 prev->offset + prev->bytes > offset)
1642 entry = prev;
1643 }
1644 }
1645 return entry;
1646 }
1647
1648 if (!prev)
1649 return NULL;
1650
1651 /* find last entry before the 'offset' */
1652 entry = prev;
1653 if (entry->offset > offset) {
1654 n = rb_prev(&entry->offset_index);
1655 if (n) {
1656 entry = rb_entry(n, struct btrfs_free_space,
1657 offset_index);
1658 ASSERT(entry->offset <= offset);
1659 } else {
1660 if (fuzzy)
1661 return entry;
1662 else
1663 return NULL;
1664 }
1665 }
1666
1667 if (entry->bitmap) {
1668 n = rb_prev(&entry->offset_index);
1669 if (n) {
1670 prev = rb_entry(n, struct btrfs_free_space,
1671 offset_index);
1672 if (!prev->bitmap &&
1673 prev->offset + prev->bytes > offset)
1674 return prev;
1675 }
1676 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1677 return entry;
1678 } else if (entry->offset + entry->bytes > offset)
1679 return entry;
1680
1681 if (!fuzzy)
1682 return NULL;
1683
1684 while (1) {
1685 if (entry->bitmap) {
1686 if (entry->offset + BITS_PER_BITMAP *
1687 ctl->unit > offset)
1688 break;
1689 } else {
1690 if (entry->offset + entry->bytes > offset)
1691 break;
1692 }
1693
1694 n = rb_next(&entry->offset_index);
1695 if (!n)
1696 return NULL;
1697 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1698 }
1699 return entry;
1700 }
1701
1702 static inline void
__unlink_free_space(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)1703 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1704 struct btrfs_free_space *info)
1705 {
1706 rb_erase(&info->offset_index, &ctl->free_space_offset);
1707 ctl->free_extents--;
1708
1709 if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1710 ctl->discardable_extents[BTRFS_STAT_CURR]--;
1711 ctl->discardable_bytes[BTRFS_STAT_CURR] -= info->bytes;
1712 }
1713 }
1714
unlink_free_space(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)1715 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1716 struct btrfs_free_space *info)
1717 {
1718 __unlink_free_space(ctl, info);
1719 ctl->free_space -= info->bytes;
1720 }
1721
link_free_space(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)1722 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1723 struct btrfs_free_space *info)
1724 {
1725 int ret = 0;
1726
1727 ASSERT(info->bytes || info->bitmap);
1728 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1729 &info->offset_index, (info->bitmap != NULL));
1730 if (ret)
1731 return ret;
1732
1733 if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1734 ctl->discardable_extents[BTRFS_STAT_CURR]++;
1735 ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
1736 }
1737
1738 ctl->free_space += info->bytes;
1739 ctl->free_extents++;
1740 return ret;
1741 }
1742
__bitmap_clear_bits(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset,u64 bytes)1743 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1744 struct btrfs_free_space *info,
1745 u64 offset, u64 bytes)
1746 {
1747 unsigned long start, count, end;
1748 int extent_delta = -1;
1749
1750 start = offset_to_bit(info->offset, ctl->unit, offset);
1751 count = bytes_to_bits(bytes, ctl->unit);
1752 end = start + count;
1753 ASSERT(end <= BITS_PER_BITMAP);
1754
1755 bitmap_clear(info->bitmap, start, count);
1756
1757 info->bytes -= bytes;
1758 if (info->max_extent_size > ctl->unit)
1759 info->max_extent_size = 0;
1760
1761 if (start && test_bit(start - 1, info->bitmap))
1762 extent_delta++;
1763
1764 if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1765 extent_delta++;
1766
1767 info->bitmap_extents += extent_delta;
1768 if (!btrfs_free_space_trimmed(info)) {
1769 ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1770 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
1771 }
1772 }
1773
bitmap_clear_bits(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset,u64 bytes)1774 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1775 struct btrfs_free_space *info, u64 offset,
1776 u64 bytes)
1777 {
1778 __bitmap_clear_bits(ctl, info, offset, bytes);
1779 ctl->free_space -= bytes;
1780 }
1781
bitmap_set_bits(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset,u64 bytes)1782 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1783 struct btrfs_free_space *info, u64 offset,
1784 u64 bytes)
1785 {
1786 unsigned long start, count, end;
1787 int extent_delta = 1;
1788
1789 start = offset_to_bit(info->offset, ctl->unit, offset);
1790 count = bytes_to_bits(bytes, ctl->unit);
1791 end = start + count;
1792 ASSERT(end <= BITS_PER_BITMAP);
1793
1794 bitmap_set(info->bitmap, start, count);
1795
1796 info->bytes += bytes;
1797 ctl->free_space += bytes;
1798
1799 if (start && test_bit(start - 1, info->bitmap))
1800 extent_delta--;
1801
1802 if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1803 extent_delta--;
1804
1805 info->bitmap_extents += extent_delta;
1806 if (!btrfs_free_space_trimmed(info)) {
1807 ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1808 ctl->discardable_bytes[BTRFS_STAT_CURR] += bytes;
1809 }
1810 }
1811
1812 /*
1813 * If we can not find suitable extent, we will use bytes to record
1814 * the size of the max extent.
1815 */
search_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * bitmap_info,u64 * offset,u64 * bytes,bool for_alloc)1816 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1817 struct btrfs_free_space *bitmap_info, u64 *offset,
1818 u64 *bytes, bool for_alloc)
1819 {
1820 unsigned long found_bits = 0;
1821 unsigned long max_bits = 0;
1822 unsigned long bits, i;
1823 unsigned long next_zero;
1824 unsigned long extent_bits;
1825
1826 /*
1827 * Skip searching the bitmap if we don't have a contiguous section that
1828 * is large enough for this allocation.
1829 */
1830 if (for_alloc &&
1831 bitmap_info->max_extent_size &&
1832 bitmap_info->max_extent_size < *bytes) {
1833 *bytes = bitmap_info->max_extent_size;
1834 return -1;
1835 }
1836
1837 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1838 max_t(u64, *offset, bitmap_info->offset));
1839 bits = bytes_to_bits(*bytes, ctl->unit);
1840
1841 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1842 if (for_alloc && bits == 1) {
1843 found_bits = 1;
1844 break;
1845 }
1846 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1847 BITS_PER_BITMAP, i);
1848 extent_bits = next_zero - i;
1849 if (extent_bits >= bits) {
1850 found_bits = extent_bits;
1851 break;
1852 } else if (extent_bits > max_bits) {
1853 max_bits = extent_bits;
1854 }
1855 i = next_zero;
1856 }
1857
1858 if (found_bits) {
1859 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1860 *bytes = (u64)(found_bits) * ctl->unit;
1861 return 0;
1862 }
1863
1864 *bytes = (u64)(max_bits) * ctl->unit;
1865 bitmap_info->max_extent_size = *bytes;
1866 return -1;
1867 }
1868
get_max_extent_size(struct btrfs_free_space * entry)1869 static inline u64 get_max_extent_size(struct btrfs_free_space *entry)
1870 {
1871 if (entry->bitmap)
1872 return entry->max_extent_size;
1873 return entry->bytes;
1874 }
1875
1876 /* Cache the size of the max extent in bytes */
1877 static struct btrfs_free_space *
find_free_space(struct btrfs_free_space_ctl * ctl,u64 * offset,u64 * bytes,unsigned long align,u64 * max_extent_size)1878 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1879 unsigned long align, u64 *max_extent_size)
1880 {
1881 struct btrfs_free_space *entry;
1882 struct rb_node *node;
1883 u64 tmp;
1884 u64 align_off;
1885 int ret;
1886
1887 if (!ctl->free_space_offset.rb_node)
1888 goto out;
1889
1890 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1891 if (!entry)
1892 goto out;
1893
1894 for (node = &entry->offset_index; node; node = rb_next(node)) {
1895 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1896 if (entry->bytes < *bytes) {
1897 *max_extent_size = max(get_max_extent_size(entry),
1898 *max_extent_size);
1899 continue;
1900 }
1901
1902 /* make sure the space returned is big enough
1903 * to match our requested alignment
1904 */
1905 if (*bytes >= align) {
1906 tmp = entry->offset - ctl->start + align - 1;
1907 tmp = div64_u64(tmp, align);
1908 tmp = tmp * align + ctl->start;
1909 align_off = tmp - entry->offset;
1910 } else {
1911 align_off = 0;
1912 tmp = entry->offset;
1913 }
1914
1915 if (entry->bytes < *bytes + align_off) {
1916 *max_extent_size = max(get_max_extent_size(entry),
1917 *max_extent_size);
1918 continue;
1919 }
1920
1921 if (entry->bitmap) {
1922 u64 size = *bytes;
1923
1924 ret = search_bitmap(ctl, entry, &tmp, &size, true);
1925 if (!ret) {
1926 *offset = tmp;
1927 *bytes = size;
1928 return entry;
1929 } else {
1930 *max_extent_size =
1931 max(get_max_extent_size(entry),
1932 *max_extent_size);
1933 }
1934 continue;
1935 }
1936
1937 *offset = tmp;
1938 *bytes = entry->bytes - align_off;
1939 return entry;
1940 }
1941 out:
1942 return NULL;
1943 }
1944
add_new_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset)1945 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1946 struct btrfs_free_space *info, u64 offset)
1947 {
1948 info->offset = offset_to_bitmap(ctl, offset);
1949 info->bytes = 0;
1950 info->bitmap_extents = 0;
1951 INIT_LIST_HEAD(&info->list);
1952 link_free_space(ctl, info);
1953 ctl->total_bitmaps++;
1954 recalculate_thresholds(ctl);
1955 }
1956
free_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * bitmap_info)1957 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1958 struct btrfs_free_space *bitmap_info)
1959 {
1960 /*
1961 * Normally when this is called, the bitmap is completely empty. However,
1962 * if we are blowing up the free space cache for one reason or another
1963 * via __btrfs_remove_free_space_cache(), then it may not be freed and
1964 * we may leave stats on the table.
1965 */
1966 if (bitmap_info->bytes && !btrfs_free_space_trimmed(bitmap_info)) {
1967 ctl->discardable_extents[BTRFS_STAT_CURR] -=
1968 bitmap_info->bitmap_extents;
1969 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bitmap_info->bytes;
1970
1971 }
1972 unlink_free_space(ctl, bitmap_info);
1973 kmem_cache_free(btrfs_free_space_bitmap_cachep, bitmap_info->bitmap);
1974 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1975 ctl->total_bitmaps--;
1976 recalculate_thresholds(ctl);
1977 }
1978
remove_from_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * bitmap_info,u64 * offset,u64 * bytes)1979 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1980 struct btrfs_free_space *bitmap_info,
1981 u64 *offset, u64 *bytes)
1982 {
1983 u64 end;
1984 u64 search_start, search_bytes;
1985 int ret;
1986
1987 again:
1988 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1989
1990 /*
1991 * We need to search for bits in this bitmap. We could only cover some
1992 * of the extent in this bitmap thanks to how we add space, so we need
1993 * to search for as much as it as we can and clear that amount, and then
1994 * go searching for the next bit.
1995 */
1996 search_start = *offset;
1997 search_bytes = ctl->unit;
1998 search_bytes = min(search_bytes, end - search_start + 1);
1999 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes,
2000 false);
2001 if (ret < 0 || search_start != *offset)
2002 return -EINVAL;
2003
2004 /* We may have found more bits than what we need */
2005 search_bytes = min(search_bytes, *bytes);
2006
2007 /* Cannot clear past the end of the bitmap */
2008 search_bytes = min(search_bytes, end - search_start + 1);
2009
2010 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
2011 *offset += search_bytes;
2012 *bytes -= search_bytes;
2013
2014 if (*bytes) {
2015 struct rb_node *next = rb_next(&bitmap_info->offset_index);
2016 if (!bitmap_info->bytes)
2017 free_bitmap(ctl, bitmap_info);
2018
2019 /*
2020 * no entry after this bitmap, but we still have bytes to
2021 * remove, so something has gone wrong.
2022 */
2023 if (!next)
2024 return -EINVAL;
2025
2026 bitmap_info = rb_entry(next, struct btrfs_free_space,
2027 offset_index);
2028
2029 /*
2030 * if the next entry isn't a bitmap we need to return to let the
2031 * extent stuff do its work.
2032 */
2033 if (!bitmap_info->bitmap)
2034 return -EAGAIN;
2035
2036 /*
2037 * Ok the next item is a bitmap, but it may not actually hold
2038 * the information for the rest of this free space stuff, so
2039 * look for it, and if we don't find it return so we can try
2040 * everything over again.
2041 */
2042 search_start = *offset;
2043 search_bytes = ctl->unit;
2044 ret = search_bitmap(ctl, bitmap_info, &search_start,
2045 &search_bytes, false);
2046 if (ret < 0 || search_start != *offset)
2047 return -EAGAIN;
2048
2049 goto again;
2050 } else if (!bitmap_info->bytes)
2051 free_bitmap(ctl, bitmap_info);
2052
2053 return 0;
2054 }
2055
add_bytes_to_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset,u64 bytes,enum btrfs_trim_state trim_state)2056 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
2057 struct btrfs_free_space *info, u64 offset,
2058 u64 bytes, enum btrfs_trim_state trim_state)
2059 {
2060 u64 bytes_to_set = 0;
2061 u64 end;
2062
2063 /*
2064 * This is a tradeoff to make bitmap trim state minimal. We mark the
2065 * whole bitmap untrimmed if at any point we add untrimmed regions.
2066 */
2067 if (trim_state == BTRFS_TRIM_STATE_UNTRIMMED) {
2068 if (btrfs_free_space_trimmed(info)) {
2069 ctl->discardable_extents[BTRFS_STAT_CURR] +=
2070 info->bitmap_extents;
2071 ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
2072 }
2073 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2074 }
2075
2076 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
2077
2078 bytes_to_set = min(end - offset, bytes);
2079
2080 bitmap_set_bits(ctl, info, offset, bytes_to_set);
2081
2082 /*
2083 * We set some bytes, we have no idea what the max extent size is
2084 * anymore.
2085 */
2086 info->max_extent_size = 0;
2087
2088 return bytes_to_set;
2089
2090 }
2091
use_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)2092 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
2093 struct btrfs_free_space *info)
2094 {
2095 struct btrfs_block_group *block_group = ctl->private;
2096 struct btrfs_fs_info *fs_info = block_group->fs_info;
2097 bool forced = false;
2098
2099 #ifdef CONFIG_BTRFS_DEBUG
2100 if (btrfs_should_fragment_free_space(block_group))
2101 forced = true;
2102 #endif
2103
2104 /* This is a way to reclaim large regions from the bitmaps. */
2105 if (!forced && info->bytes >= FORCE_EXTENT_THRESHOLD)
2106 return false;
2107
2108 /*
2109 * If we are below the extents threshold then we can add this as an
2110 * extent, and don't have to deal with the bitmap
2111 */
2112 if (!forced && ctl->free_extents < ctl->extents_thresh) {
2113 /*
2114 * If this block group has some small extents we don't want to
2115 * use up all of our free slots in the cache with them, we want
2116 * to reserve them to larger extents, however if we have plenty
2117 * of cache left then go ahead an dadd them, no sense in adding
2118 * the overhead of a bitmap if we don't have to.
2119 */
2120 if (info->bytes <= fs_info->sectorsize * 8) {
2121 if (ctl->free_extents * 3 <= ctl->extents_thresh)
2122 return false;
2123 } else {
2124 return false;
2125 }
2126 }
2127
2128 /*
2129 * The original block groups from mkfs can be really small, like 8
2130 * megabytes, so don't bother with a bitmap for those entries. However
2131 * some block groups can be smaller than what a bitmap would cover but
2132 * are still large enough that they could overflow the 32k memory limit,
2133 * so allow those block groups to still be allowed to have a bitmap
2134 * entry.
2135 */
2136 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->length)
2137 return false;
2138
2139 return true;
2140 }
2141
2142 static const struct btrfs_free_space_op free_space_op = {
2143 .use_bitmap = use_bitmap,
2144 };
2145
insert_into_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)2146 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
2147 struct btrfs_free_space *info)
2148 {
2149 struct btrfs_free_space *bitmap_info;
2150 struct btrfs_block_group *block_group = NULL;
2151 int added = 0;
2152 u64 bytes, offset, bytes_added;
2153 enum btrfs_trim_state trim_state;
2154 int ret;
2155
2156 bytes = info->bytes;
2157 offset = info->offset;
2158 trim_state = info->trim_state;
2159
2160 if (!ctl->op->use_bitmap(ctl, info))
2161 return 0;
2162
2163 if (ctl->op == &free_space_op)
2164 block_group = ctl->private;
2165 again:
2166 /*
2167 * Since we link bitmaps right into the cluster we need to see if we
2168 * have a cluster here, and if so and it has our bitmap we need to add
2169 * the free space to that bitmap.
2170 */
2171 if (block_group && !list_empty(&block_group->cluster_list)) {
2172 struct btrfs_free_cluster *cluster;
2173 struct rb_node *node;
2174 struct btrfs_free_space *entry;
2175
2176 cluster = list_entry(block_group->cluster_list.next,
2177 struct btrfs_free_cluster,
2178 block_group_list);
2179 spin_lock(&cluster->lock);
2180 node = rb_first(&cluster->root);
2181 if (!node) {
2182 spin_unlock(&cluster->lock);
2183 goto no_cluster_bitmap;
2184 }
2185
2186 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2187 if (!entry->bitmap) {
2188 spin_unlock(&cluster->lock);
2189 goto no_cluster_bitmap;
2190 }
2191
2192 if (entry->offset == offset_to_bitmap(ctl, offset)) {
2193 bytes_added = add_bytes_to_bitmap(ctl, entry, offset,
2194 bytes, trim_state);
2195 bytes -= bytes_added;
2196 offset += bytes_added;
2197 }
2198 spin_unlock(&cluster->lock);
2199 if (!bytes) {
2200 ret = 1;
2201 goto out;
2202 }
2203 }
2204
2205 no_cluster_bitmap:
2206 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2207 1, 0);
2208 if (!bitmap_info) {
2209 ASSERT(added == 0);
2210 goto new_bitmap;
2211 }
2212
2213 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
2214 trim_state);
2215 bytes -= bytes_added;
2216 offset += bytes_added;
2217 added = 0;
2218
2219 if (!bytes) {
2220 ret = 1;
2221 goto out;
2222 } else
2223 goto again;
2224
2225 new_bitmap:
2226 if (info && info->bitmap) {
2227 add_new_bitmap(ctl, info, offset);
2228 added = 1;
2229 info = NULL;
2230 goto again;
2231 } else {
2232 spin_unlock(&ctl->tree_lock);
2233
2234 /* no pre-allocated info, allocate a new one */
2235 if (!info) {
2236 info = kmem_cache_zalloc(btrfs_free_space_cachep,
2237 GFP_NOFS);
2238 if (!info) {
2239 spin_lock(&ctl->tree_lock);
2240 ret = -ENOMEM;
2241 goto out;
2242 }
2243 }
2244
2245 /* allocate the bitmap */
2246 info->bitmap = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep,
2247 GFP_NOFS);
2248 info->trim_state = BTRFS_TRIM_STATE_TRIMMED;
2249 spin_lock(&ctl->tree_lock);
2250 if (!info->bitmap) {
2251 ret = -ENOMEM;
2252 goto out;
2253 }
2254 goto again;
2255 }
2256
2257 out:
2258 if (info) {
2259 if (info->bitmap)
2260 kmem_cache_free(btrfs_free_space_bitmap_cachep,
2261 info->bitmap);
2262 kmem_cache_free(btrfs_free_space_cachep, info);
2263 }
2264
2265 return ret;
2266 }
2267
2268 /*
2269 * Free space merging rules:
2270 * 1) Merge trimmed areas together
2271 * 2) Let untrimmed areas coalesce with trimmed areas
2272 * 3) Always pull neighboring regions from bitmaps
2273 *
2274 * The above rules are for when we merge free space based on btrfs_trim_state.
2275 * Rules 2 and 3 are subtle because they are suboptimal, but are done for the
2276 * same reason: to promote larger extent regions which makes life easier for
2277 * find_free_extent(). Rule 2 enables coalescing based on the common path
2278 * being returning free space from btrfs_finish_extent_commit(). So when free
2279 * space is trimmed, it will prevent aggregating trimmed new region and
2280 * untrimmed regions in the rb_tree. Rule 3 is purely to obtain larger extents
2281 * and provide find_free_extent() with the largest extents possible hoping for
2282 * the reuse path.
2283 */
try_merge_free_space(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2284 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2285 struct btrfs_free_space *info, bool update_stat)
2286 {
2287 struct btrfs_free_space *left_info = NULL;
2288 struct btrfs_free_space *right_info;
2289 bool merged = false;
2290 u64 offset = info->offset;
2291 u64 bytes = info->bytes;
2292 const bool is_trimmed = btrfs_free_space_trimmed(info);
2293
2294 /*
2295 * first we want to see if there is free space adjacent to the range we
2296 * are adding, if there is remove that struct and add a new one to
2297 * cover the entire range
2298 */
2299 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
2300 if (right_info && rb_prev(&right_info->offset_index))
2301 left_info = rb_entry(rb_prev(&right_info->offset_index),
2302 struct btrfs_free_space, offset_index);
2303 else if (!right_info)
2304 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
2305
2306 /* See try_merge_free_space() comment. */
2307 if (right_info && !right_info->bitmap &&
2308 (!is_trimmed || btrfs_free_space_trimmed(right_info))) {
2309 if (update_stat)
2310 unlink_free_space(ctl, right_info);
2311 else
2312 __unlink_free_space(ctl, right_info);
2313 info->bytes += right_info->bytes;
2314 kmem_cache_free(btrfs_free_space_cachep, right_info);
2315 merged = true;
2316 }
2317
2318 /* See try_merge_free_space() comment. */
2319 if (left_info && !left_info->bitmap &&
2320 left_info->offset + left_info->bytes == offset &&
2321 (!is_trimmed || btrfs_free_space_trimmed(left_info))) {
2322 if (update_stat)
2323 unlink_free_space(ctl, left_info);
2324 else
2325 __unlink_free_space(ctl, left_info);
2326 info->offset = left_info->offset;
2327 info->bytes += left_info->bytes;
2328 kmem_cache_free(btrfs_free_space_cachep, left_info);
2329 merged = true;
2330 }
2331
2332 return merged;
2333 }
2334
steal_from_bitmap_to_end(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2335 static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2336 struct btrfs_free_space *info,
2337 bool update_stat)
2338 {
2339 struct btrfs_free_space *bitmap;
2340 unsigned long i;
2341 unsigned long j;
2342 const u64 end = info->offset + info->bytes;
2343 const u64 bitmap_offset = offset_to_bitmap(ctl, end);
2344 u64 bytes;
2345
2346 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2347 if (!bitmap)
2348 return false;
2349
2350 i = offset_to_bit(bitmap->offset, ctl->unit, end);
2351 j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
2352 if (j == i)
2353 return false;
2354 bytes = (j - i) * ctl->unit;
2355 info->bytes += bytes;
2356
2357 /* See try_merge_free_space() comment. */
2358 if (!btrfs_free_space_trimmed(bitmap))
2359 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2360
2361 if (update_stat)
2362 bitmap_clear_bits(ctl, bitmap, end, bytes);
2363 else
2364 __bitmap_clear_bits(ctl, bitmap, end, bytes);
2365
2366 if (!bitmap->bytes)
2367 free_bitmap(ctl, bitmap);
2368
2369 return true;
2370 }
2371
steal_from_bitmap_to_front(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2372 static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2373 struct btrfs_free_space *info,
2374 bool update_stat)
2375 {
2376 struct btrfs_free_space *bitmap;
2377 u64 bitmap_offset;
2378 unsigned long i;
2379 unsigned long j;
2380 unsigned long prev_j;
2381 u64 bytes;
2382
2383 bitmap_offset = offset_to_bitmap(ctl, info->offset);
2384 /* If we're on a boundary, try the previous logical bitmap. */
2385 if (bitmap_offset == info->offset) {
2386 if (info->offset == 0)
2387 return false;
2388 bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
2389 }
2390
2391 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2392 if (!bitmap)
2393 return false;
2394
2395 i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
2396 j = 0;
2397 prev_j = (unsigned long)-1;
2398 for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2399 if (j > i)
2400 break;
2401 prev_j = j;
2402 }
2403 if (prev_j == i)
2404 return false;
2405
2406 if (prev_j == (unsigned long)-1)
2407 bytes = (i + 1) * ctl->unit;
2408 else
2409 bytes = (i - prev_j) * ctl->unit;
2410
2411 info->offset -= bytes;
2412 info->bytes += bytes;
2413
2414 /* See try_merge_free_space() comment. */
2415 if (!btrfs_free_space_trimmed(bitmap))
2416 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2417
2418 if (update_stat)
2419 bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2420 else
2421 __bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2422
2423 if (!bitmap->bytes)
2424 free_bitmap(ctl, bitmap);
2425
2426 return true;
2427 }
2428
2429 /*
2430 * We prefer always to allocate from extent entries, both for clustered and
2431 * non-clustered allocation requests. So when attempting to add a new extent
2432 * entry, try to see if there's adjacent free space in bitmap entries, and if
2433 * there is, migrate that space from the bitmaps to the extent.
2434 * Like this we get better chances of satisfying space allocation requests
2435 * because we attempt to satisfy them based on a single cache entry, and never
2436 * on 2 or more entries - even if the entries represent a contiguous free space
2437 * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2438 * ends).
2439 */
steal_from_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2440 static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2441 struct btrfs_free_space *info,
2442 bool update_stat)
2443 {
2444 /*
2445 * Only work with disconnected entries, as we can change their offset,
2446 * and must be extent entries.
2447 */
2448 ASSERT(!info->bitmap);
2449 ASSERT(RB_EMPTY_NODE(&info->offset_index));
2450
2451 if (ctl->total_bitmaps > 0) {
2452 bool stole_end;
2453 bool stole_front = false;
2454
2455 stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2456 if (ctl->total_bitmaps > 0)
2457 stole_front = steal_from_bitmap_to_front(ctl, info,
2458 update_stat);
2459
2460 if (stole_end || stole_front)
2461 try_merge_free_space(ctl, info, update_stat);
2462 }
2463 }
2464
__btrfs_add_free_space(struct btrfs_fs_info * fs_info,struct btrfs_free_space_ctl * ctl,u64 offset,u64 bytes,enum btrfs_trim_state trim_state)2465 int __btrfs_add_free_space(struct btrfs_fs_info *fs_info,
2466 struct btrfs_free_space_ctl *ctl,
2467 u64 offset, u64 bytes,
2468 enum btrfs_trim_state trim_state)
2469 {
2470 struct btrfs_block_group *block_group = ctl->private;
2471 struct btrfs_free_space *info;
2472 int ret = 0;
2473 u64 filter_bytes = bytes;
2474
2475 ASSERT(!btrfs_is_zoned(fs_info));
2476
2477 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2478 if (!info)
2479 return -ENOMEM;
2480
2481 info->offset = offset;
2482 info->bytes = bytes;
2483 info->trim_state = trim_state;
2484 RB_CLEAR_NODE(&info->offset_index);
2485
2486 spin_lock(&ctl->tree_lock);
2487
2488 if (try_merge_free_space(ctl, info, true))
2489 goto link;
2490
2491 /*
2492 * There was no extent directly to the left or right of this new
2493 * extent then we know we're going to have to allocate a new extent, so
2494 * before we do that see if we need to drop this into a bitmap
2495 */
2496 ret = insert_into_bitmap(ctl, info);
2497 if (ret < 0) {
2498 goto out;
2499 } else if (ret) {
2500 ret = 0;
2501 goto out;
2502 }
2503 link:
2504 /*
2505 * Only steal free space from adjacent bitmaps if we're sure we're not
2506 * going to add the new free space to existing bitmap entries - because
2507 * that would mean unnecessary work that would be reverted. Therefore
2508 * attempt to steal space from bitmaps if we're adding an extent entry.
2509 */
2510 steal_from_bitmap(ctl, info, true);
2511
2512 filter_bytes = max(filter_bytes, info->bytes);
2513
2514 ret = link_free_space(ctl, info);
2515 if (ret)
2516 kmem_cache_free(btrfs_free_space_cachep, info);
2517 out:
2518 btrfs_discard_update_discardable(block_group);
2519 spin_unlock(&ctl->tree_lock);
2520
2521 if (ret) {
2522 btrfs_crit(fs_info, "unable to add free space :%d", ret);
2523 ASSERT(ret != -EEXIST);
2524 }
2525
2526 if (trim_state != BTRFS_TRIM_STATE_TRIMMED) {
2527 btrfs_discard_check_filter(block_group, filter_bytes);
2528 btrfs_discard_queue_work(&fs_info->discard_ctl, block_group);
2529 }
2530
2531 return ret;
2532 }
2533
__btrfs_add_free_space_zoned(struct btrfs_block_group * block_group,u64 bytenr,u64 size,bool used)2534 static int __btrfs_add_free_space_zoned(struct btrfs_block_group *block_group,
2535 u64 bytenr, u64 size, bool used)
2536 {
2537 struct btrfs_fs_info *fs_info = block_group->fs_info;
2538 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2539 u64 offset = bytenr - block_group->start;
2540 u64 to_free, to_unusable;
2541 const int bg_reclaim_threshold = READ_ONCE(fs_info->bg_reclaim_threshold);
2542
2543 spin_lock(&ctl->tree_lock);
2544 if (!used)
2545 to_free = size;
2546 else if (offset >= block_group->alloc_offset)
2547 to_free = size;
2548 else if (offset + size <= block_group->alloc_offset)
2549 to_free = 0;
2550 else
2551 to_free = offset + size - block_group->alloc_offset;
2552 to_unusable = size - to_free;
2553
2554 ctl->free_space += to_free;
2555 /*
2556 * If the block group is read-only, we should account freed space into
2557 * bytes_readonly.
2558 */
2559 if (!block_group->ro)
2560 block_group->zone_unusable += to_unusable;
2561 spin_unlock(&ctl->tree_lock);
2562 if (!used) {
2563 spin_lock(&block_group->lock);
2564 block_group->alloc_offset -= size;
2565 spin_unlock(&block_group->lock);
2566 }
2567
2568 /* All the region is now unusable. Mark it as unused and reclaim */
2569 if (block_group->zone_unusable == block_group->length) {
2570 btrfs_mark_bg_unused(block_group);
2571 } else if (bg_reclaim_threshold &&
2572 block_group->zone_unusable >=
2573 div_factor_fine(block_group->length, bg_reclaim_threshold)) {
2574 btrfs_mark_bg_to_reclaim(block_group);
2575 }
2576
2577 return 0;
2578 }
2579
btrfs_add_free_space(struct btrfs_block_group * block_group,u64 bytenr,u64 size)2580 int btrfs_add_free_space(struct btrfs_block_group *block_group,
2581 u64 bytenr, u64 size)
2582 {
2583 enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2584
2585 if (btrfs_is_zoned(block_group->fs_info))
2586 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2587 true);
2588
2589 if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC))
2590 trim_state = BTRFS_TRIM_STATE_TRIMMED;
2591
2592 return __btrfs_add_free_space(block_group->fs_info,
2593 block_group->free_space_ctl,
2594 bytenr, size, trim_state);
2595 }
2596
btrfs_add_free_space_unused(struct btrfs_block_group * block_group,u64 bytenr,u64 size)2597 int btrfs_add_free_space_unused(struct btrfs_block_group *block_group,
2598 u64 bytenr, u64 size)
2599 {
2600 if (btrfs_is_zoned(block_group->fs_info))
2601 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2602 false);
2603
2604 return btrfs_add_free_space(block_group, bytenr, size);
2605 }
2606
2607 /*
2608 * This is a subtle distinction because when adding free space back in general,
2609 * we want it to be added as untrimmed for async. But in the case where we add
2610 * it on loading of a block group, we want to consider it trimmed.
2611 */
btrfs_add_free_space_async_trimmed(struct btrfs_block_group * block_group,u64 bytenr,u64 size)2612 int btrfs_add_free_space_async_trimmed(struct btrfs_block_group *block_group,
2613 u64 bytenr, u64 size)
2614 {
2615 enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2616
2617 if (btrfs_is_zoned(block_group->fs_info))
2618 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2619 true);
2620
2621 if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC) ||
2622 btrfs_test_opt(block_group->fs_info, DISCARD_ASYNC))
2623 trim_state = BTRFS_TRIM_STATE_TRIMMED;
2624
2625 return __btrfs_add_free_space(block_group->fs_info,
2626 block_group->free_space_ctl,
2627 bytenr, size, trim_state);
2628 }
2629
btrfs_remove_free_space(struct btrfs_block_group * block_group,u64 offset,u64 bytes)2630 int btrfs_remove_free_space(struct btrfs_block_group *block_group,
2631 u64 offset, u64 bytes)
2632 {
2633 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2634 struct btrfs_free_space *info;
2635 int ret;
2636 bool re_search = false;
2637
2638 if (btrfs_is_zoned(block_group->fs_info)) {
2639 /*
2640 * This can happen with conventional zones when replaying log.
2641 * Since the allocation info of tree-log nodes are not recorded
2642 * to the extent-tree, calculate_alloc_pointer() failed to
2643 * advance the allocation pointer after last allocated tree log
2644 * node blocks.
2645 *
2646 * This function is called from
2647 * btrfs_pin_extent_for_log_replay() when replaying the log.
2648 * Advance the pointer not to overwrite the tree-log nodes.
2649 */
2650 if (block_group->start + block_group->alloc_offset <
2651 offset + bytes) {
2652 block_group->alloc_offset =
2653 offset + bytes - block_group->start;
2654 }
2655 return 0;
2656 }
2657
2658 spin_lock(&ctl->tree_lock);
2659
2660 again:
2661 ret = 0;
2662 if (!bytes)
2663 goto out_lock;
2664
2665 info = tree_search_offset(ctl, offset, 0, 0);
2666 if (!info) {
2667 /*
2668 * oops didn't find an extent that matched the space we wanted
2669 * to remove, look for a bitmap instead
2670 */
2671 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2672 1, 0);
2673 if (!info) {
2674 /*
2675 * If we found a partial bit of our free space in a
2676 * bitmap but then couldn't find the other part this may
2677 * be a problem, so WARN about it.
2678 */
2679 WARN_ON(re_search);
2680 goto out_lock;
2681 }
2682 }
2683
2684 re_search = false;
2685 if (!info->bitmap) {
2686 unlink_free_space(ctl, info);
2687 if (offset == info->offset) {
2688 u64 to_free = min(bytes, info->bytes);
2689
2690 info->bytes -= to_free;
2691 info->offset += to_free;
2692 if (info->bytes) {
2693 ret = link_free_space(ctl, info);
2694 WARN_ON(ret);
2695 } else {
2696 kmem_cache_free(btrfs_free_space_cachep, info);
2697 }
2698
2699 offset += to_free;
2700 bytes -= to_free;
2701 goto again;
2702 } else {
2703 u64 old_end = info->bytes + info->offset;
2704
2705 info->bytes = offset - info->offset;
2706 ret = link_free_space(ctl, info);
2707 WARN_ON(ret);
2708 if (ret)
2709 goto out_lock;
2710
2711 /* Not enough bytes in this entry to satisfy us */
2712 if (old_end < offset + bytes) {
2713 bytes -= old_end - offset;
2714 offset = old_end;
2715 goto again;
2716 } else if (old_end == offset + bytes) {
2717 /* all done */
2718 goto out_lock;
2719 }
2720 spin_unlock(&ctl->tree_lock);
2721
2722 ret = __btrfs_add_free_space(block_group->fs_info, ctl,
2723 offset + bytes,
2724 old_end - (offset + bytes),
2725 info->trim_state);
2726 WARN_ON(ret);
2727 goto out;
2728 }
2729 }
2730
2731 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2732 if (ret == -EAGAIN) {
2733 re_search = true;
2734 goto again;
2735 }
2736 out_lock:
2737 btrfs_discard_update_discardable(block_group);
2738 spin_unlock(&ctl->tree_lock);
2739 out:
2740 return ret;
2741 }
2742
btrfs_dump_free_space(struct btrfs_block_group * block_group,u64 bytes)2743 void btrfs_dump_free_space(struct btrfs_block_group *block_group,
2744 u64 bytes)
2745 {
2746 struct btrfs_fs_info *fs_info = block_group->fs_info;
2747 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2748 struct btrfs_free_space *info;
2749 struct rb_node *n;
2750 int count = 0;
2751
2752 /*
2753 * Zoned btrfs does not use free space tree and cluster. Just print
2754 * out the free space after the allocation offset.
2755 */
2756 if (btrfs_is_zoned(fs_info)) {
2757 btrfs_info(fs_info, "free space %llu",
2758 block_group->length - block_group->alloc_offset);
2759 return;
2760 }
2761
2762 spin_lock(&ctl->tree_lock);
2763 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2764 info = rb_entry(n, struct btrfs_free_space, offset_index);
2765 if (info->bytes >= bytes && !block_group->ro)
2766 count++;
2767 btrfs_crit(fs_info, "entry offset %llu, bytes %llu, bitmap %s",
2768 info->offset, info->bytes,
2769 (info->bitmap) ? "yes" : "no");
2770 }
2771 spin_unlock(&ctl->tree_lock);
2772 btrfs_info(fs_info, "block group has cluster?: %s",
2773 list_empty(&block_group->cluster_list) ? "no" : "yes");
2774 btrfs_info(fs_info,
2775 "%d blocks of free space at or bigger than bytes is", count);
2776 }
2777
btrfs_init_free_space_ctl(struct btrfs_block_group * block_group,struct btrfs_free_space_ctl * ctl)2778 void btrfs_init_free_space_ctl(struct btrfs_block_group *block_group,
2779 struct btrfs_free_space_ctl *ctl)
2780 {
2781 struct btrfs_fs_info *fs_info = block_group->fs_info;
2782
2783 spin_lock_init(&ctl->tree_lock);
2784 ctl->unit = fs_info->sectorsize;
2785 ctl->start = block_group->start;
2786 ctl->private = block_group;
2787 ctl->op = &free_space_op;
2788 INIT_LIST_HEAD(&ctl->trimming_ranges);
2789 mutex_init(&ctl->cache_writeout_mutex);
2790
2791 /*
2792 * we only want to have 32k of ram per block group for keeping
2793 * track of free space, and if we pass 1/2 of that we want to
2794 * start converting things over to using bitmaps
2795 */
2796 ctl->extents_thresh = (SZ_32K / 2) / sizeof(struct btrfs_free_space);
2797 }
2798
2799 /*
2800 * for a given cluster, put all of its extents back into the free
2801 * space cache. If the block group passed doesn't match the block group
2802 * pointed to by the cluster, someone else raced in and freed the
2803 * cluster already. In that case, we just return without changing anything
2804 */
__btrfs_return_cluster_to_free_space(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster)2805 static void __btrfs_return_cluster_to_free_space(
2806 struct btrfs_block_group *block_group,
2807 struct btrfs_free_cluster *cluster)
2808 {
2809 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2810 struct btrfs_free_space *entry;
2811 struct rb_node *node;
2812
2813 spin_lock(&cluster->lock);
2814 if (cluster->block_group != block_group) {
2815 spin_unlock(&cluster->lock);
2816 return;
2817 }
2818
2819 cluster->block_group = NULL;
2820 cluster->window_start = 0;
2821 list_del_init(&cluster->block_group_list);
2822
2823 node = rb_first(&cluster->root);
2824 while (node) {
2825 bool bitmap;
2826
2827 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2828 node = rb_next(&entry->offset_index);
2829 rb_erase(&entry->offset_index, &cluster->root);
2830 RB_CLEAR_NODE(&entry->offset_index);
2831
2832 bitmap = (entry->bitmap != NULL);
2833 if (!bitmap) {
2834 /* Merging treats extents as if they were new */
2835 if (!btrfs_free_space_trimmed(entry)) {
2836 ctl->discardable_extents[BTRFS_STAT_CURR]--;
2837 ctl->discardable_bytes[BTRFS_STAT_CURR] -=
2838 entry->bytes;
2839 }
2840
2841 try_merge_free_space(ctl, entry, false);
2842 steal_from_bitmap(ctl, entry, false);
2843
2844 /* As we insert directly, update these statistics */
2845 if (!btrfs_free_space_trimmed(entry)) {
2846 ctl->discardable_extents[BTRFS_STAT_CURR]++;
2847 ctl->discardable_bytes[BTRFS_STAT_CURR] +=
2848 entry->bytes;
2849 }
2850 }
2851 tree_insert_offset(&ctl->free_space_offset,
2852 entry->offset, &entry->offset_index, bitmap);
2853 }
2854 cluster->root = RB_ROOT;
2855 spin_unlock(&cluster->lock);
2856 btrfs_put_block_group(block_group);
2857 }
2858
__btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl * ctl)2859 static void __btrfs_remove_free_space_cache_locked(
2860 struct btrfs_free_space_ctl *ctl)
2861 {
2862 struct btrfs_free_space *info;
2863 struct rb_node *node;
2864
2865 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2866 info = rb_entry(node, struct btrfs_free_space, offset_index);
2867 if (!info->bitmap) {
2868 unlink_free_space(ctl, info);
2869 kmem_cache_free(btrfs_free_space_cachep, info);
2870 } else {
2871 free_bitmap(ctl, info);
2872 }
2873
2874 cond_resched_lock(&ctl->tree_lock);
2875 }
2876 }
2877
__btrfs_remove_free_space_cache(struct btrfs_free_space_ctl * ctl)2878 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2879 {
2880 spin_lock(&ctl->tree_lock);
2881 __btrfs_remove_free_space_cache_locked(ctl);
2882 if (ctl->private)
2883 btrfs_discard_update_discardable(ctl->private);
2884 spin_unlock(&ctl->tree_lock);
2885 }
2886
btrfs_remove_free_space_cache(struct btrfs_block_group * block_group)2887 void btrfs_remove_free_space_cache(struct btrfs_block_group *block_group)
2888 {
2889 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2890 struct btrfs_free_cluster *cluster;
2891 struct list_head *head;
2892
2893 spin_lock(&ctl->tree_lock);
2894 while ((head = block_group->cluster_list.next) !=
2895 &block_group->cluster_list) {
2896 cluster = list_entry(head, struct btrfs_free_cluster,
2897 block_group_list);
2898
2899 WARN_ON(cluster->block_group != block_group);
2900 __btrfs_return_cluster_to_free_space(block_group, cluster);
2901
2902 cond_resched_lock(&ctl->tree_lock);
2903 }
2904 __btrfs_remove_free_space_cache_locked(ctl);
2905 btrfs_discard_update_discardable(block_group);
2906 spin_unlock(&ctl->tree_lock);
2907
2908 }
2909
2910 /**
2911 * btrfs_is_free_space_trimmed - see if everything is trimmed
2912 * @block_group: block_group of interest
2913 *
2914 * Walk @block_group's free space rb_tree to determine if everything is trimmed.
2915 */
btrfs_is_free_space_trimmed(struct btrfs_block_group * block_group)2916 bool btrfs_is_free_space_trimmed(struct btrfs_block_group *block_group)
2917 {
2918 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2919 struct btrfs_free_space *info;
2920 struct rb_node *node;
2921 bool ret = true;
2922
2923 spin_lock(&ctl->tree_lock);
2924 node = rb_first(&ctl->free_space_offset);
2925
2926 while (node) {
2927 info = rb_entry(node, struct btrfs_free_space, offset_index);
2928
2929 if (!btrfs_free_space_trimmed(info)) {
2930 ret = false;
2931 break;
2932 }
2933
2934 node = rb_next(node);
2935 }
2936
2937 spin_unlock(&ctl->tree_lock);
2938 return ret;
2939 }
2940
btrfs_find_space_for_alloc(struct btrfs_block_group * block_group,u64 offset,u64 bytes,u64 empty_size,u64 * max_extent_size)2941 u64 btrfs_find_space_for_alloc(struct btrfs_block_group *block_group,
2942 u64 offset, u64 bytes, u64 empty_size,
2943 u64 *max_extent_size)
2944 {
2945 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2946 struct btrfs_discard_ctl *discard_ctl =
2947 &block_group->fs_info->discard_ctl;
2948 struct btrfs_free_space *entry = NULL;
2949 u64 bytes_search = bytes + empty_size;
2950 u64 ret = 0;
2951 u64 align_gap = 0;
2952 u64 align_gap_len = 0;
2953 enum btrfs_trim_state align_gap_trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2954
2955 ASSERT(!btrfs_is_zoned(block_group->fs_info));
2956
2957 spin_lock(&ctl->tree_lock);
2958 entry = find_free_space(ctl, &offset, &bytes_search,
2959 block_group->full_stripe_len, max_extent_size);
2960 if (!entry)
2961 goto out;
2962
2963 ret = offset;
2964 if (entry->bitmap) {
2965 bitmap_clear_bits(ctl, entry, offset, bytes);
2966
2967 if (!btrfs_free_space_trimmed(entry))
2968 atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
2969
2970 if (!entry->bytes)
2971 free_bitmap(ctl, entry);
2972 } else {
2973 unlink_free_space(ctl, entry);
2974 align_gap_len = offset - entry->offset;
2975 align_gap = entry->offset;
2976 align_gap_trim_state = entry->trim_state;
2977
2978 if (!btrfs_free_space_trimmed(entry))
2979 atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
2980
2981 entry->offset = offset + bytes;
2982 WARN_ON(entry->bytes < bytes + align_gap_len);
2983
2984 entry->bytes -= bytes + align_gap_len;
2985 if (!entry->bytes)
2986 kmem_cache_free(btrfs_free_space_cachep, entry);
2987 else
2988 link_free_space(ctl, entry);
2989 }
2990 out:
2991 btrfs_discard_update_discardable(block_group);
2992 spin_unlock(&ctl->tree_lock);
2993
2994 if (align_gap_len)
2995 __btrfs_add_free_space(block_group->fs_info, ctl,
2996 align_gap, align_gap_len,
2997 align_gap_trim_state);
2998 return ret;
2999 }
3000
3001 /*
3002 * given a cluster, put all of its extents back into the free space
3003 * cache. If a block group is passed, this function will only free
3004 * a cluster that belongs to the passed block group.
3005 *
3006 * Otherwise, it'll get a reference on the block group pointed to by the
3007 * cluster and remove the cluster from it.
3008 */
btrfs_return_cluster_to_free_space(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster)3009 void btrfs_return_cluster_to_free_space(
3010 struct btrfs_block_group *block_group,
3011 struct btrfs_free_cluster *cluster)
3012 {
3013 struct btrfs_free_space_ctl *ctl;
3014
3015 /* first, get a safe pointer to the block group */
3016 spin_lock(&cluster->lock);
3017 if (!block_group) {
3018 block_group = cluster->block_group;
3019 if (!block_group) {
3020 spin_unlock(&cluster->lock);
3021 return;
3022 }
3023 } else if (cluster->block_group != block_group) {
3024 /* someone else has already freed it don't redo their work */
3025 spin_unlock(&cluster->lock);
3026 return;
3027 }
3028 btrfs_get_block_group(block_group);
3029 spin_unlock(&cluster->lock);
3030
3031 ctl = block_group->free_space_ctl;
3032
3033 /* now return any extents the cluster had on it */
3034 spin_lock(&ctl->tree_lock);
3035 __btrfs_return_cluster_to_free_space(block_group, cluster);
3036 spin_unlock(&ctl->tree_lock);
3037
3038 btrfs_discard_queue_work(&block_group->fs_info->discard_ctl, block_group);
3039
3040 /* finally drop our ref */
3041 btrfs_put_block_group(block_group);
3042 }
3043
btrfs_alloc_from_bitmap(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster,struct btrfs_free_space * entry,u64 bytes,u64 min_start,u64 * max_extent_size)3044 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group *block_group,
3045 struct btrfs_free_cluster *cluster,
3046 struct btrfs_free_space *entry,
3047 u64 bytes, u64 min_start,
3048 u64 *max_extent_size)
3049 {
3050 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3051 int err;
3052 u64 search_start = cluster->window_start;
3053 u64 search_bytes = bytes;
3054 u64 ret = 0;
3055
3056 search_start = min_start;
3057 search_bytes = bytes;
3058
3059 err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
3060 if (err) {
3061 *max_extent_size = max(get_max_extent_size(entry),
3062 *max_extent_size);
3063 return 0;
3064 }
3065
3066 ret = search_start;
3067 __bitmap_clear_bits(ctl, entry, ret, bytes);
3068
3069 return ret;
3070 }
3071
3072 /*
3073 * given a cluster, try to allocate 'bytes' from it, returns 0
3074 * if it couldn't find anything suitably large, or a logical disk offset
3075 * if things worked out
3076 */
btrfs_alloc_from_cluster(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster,u64 bytes,u64 min_start,u64 * max_extent_size)3077 u64 btrfs_alloc_from_cluster(struct btrfs_block_group *block_group,
3078 struct btrfs_free_cluster *cluster, u64 bytes,
3079 u64 min_start, u64 *max_extent_size)
3080 {
3081 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3082 struct btrfs_discard_ctl *discard_ctl =
3083 &block_group->fs_info->discard_ctl;
3084 struct btrfs_free_space *entry = NULL;
3085 struct rb_node *node;
3086 u64 ret = 0;
3087
3088 ASSERT(!btrfs_is_zoned(block_group->fs_info));
3089
3090 spin_lock(&cluster->lock);
3091 if (bytes > cluster->max_size)
3092 goto out;
3093
3094 if (cluster->block_group != block_group)
3095 goto out;
3096
3097 node = rb_first(&cluster->root);
3098 if (!node)
3099 goto out;
3100
3101 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3102 while (1) {
3103 if (entry->bytes < bytes)
3104 *max_extent_size = max(get_max_extent_size(entry),
3105 *max_extent_size);
3106
3107 if (entry->bytes < bytes ||
3108 (!entry->bitmap && entry->offset < min_start)) {
3109 node = rb_next(&entry->offset_index);
3110 if (!node)
3111 break;
3112 entry = rb_entry(node, struct btrfs_free_space,
3113 offset_index);
3114 continue;
3115 }
3116
3117 if (entry->bitmap) {
3118 ret = btrfs_alloc_from_bitmap(block_group,
3119 cluster, entry, bytes,
3120 cluster->window_start,
3121 max_extent_size);
3122 if (ret == 0) {
3123 node = rb_next(&entry->offset_index);
3124 if (!node)
3125 break;
3126 entry = rb_entry(node, struct btrfs_free_space,
3127 offset_index);
3128 continue;
3129 }
3130 cluster->window_start += bytes;
3131 } else {
3132 ret = entry->offset;
3133
3134 entry->offset += bytes;
3135 entry->bytes -= bytes;
3136 }
3137
3138 break;
3139 }
3140 out:
3141 spin_unlock(&cluster->lock);
3142
3143 if (!ret)
3144 return 0;
3145
3146 spin_lock(&ctl->tree_lock);
3147
3148 if (!btrfs_free_space_trimmed(entry))
3149 atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3150
3151 ctl->free_space -= bytes;
3152 if (!entry->bitmap && !btrfs_free_space_trimmed(entry))
3153 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
3154
3155 spin_lock(&cluster->lock);
3156 if (entry->bytes == 0) {
3157 rb_erase(&entry->offset_index, &cluster->root);
3158 ctl->free_extents--;
3159 if (entry->bitmap) {
3160 kmem_cache_free(btrfs_free_space_bitmap_cachep,
3161 entry->bitmap);
3162 ctl->total_bitmaps--;
3163 recalculate_thresholds(ctl);
3164 } else if (!btrfs_free_space_trimmed(entry)) {
3165 ctl->discardable_extents[BTRFS_STAT_CURR]--;
3166 }
3167 kmem_cache_free(btrfs_free_space_cachep, entry);
3168 }
3169
3170 spin_unlock(&cluster->lock);
3171 spin_unlock(&ctl->tree_lock);
3172
3173 return ret;
3174 }
3175
btrfs_bitmap_cluster(struct btrfs_block_group * block_group,struct btrfs_free_space * entry,struct btrfs_free_cluster * cluster,u64 offset,u64 bytes,u64 cont1_bytes,u64 min_bytes)3176 static int btrfs_bitmap_cluster(struct btrfs_block_group *block_group,
3177 struct btrfs_free_space *entry,
3178 struct btrfs_free_cluster *cluster,
3179 u64 offset, u64 bytes,
3180 u64 cont1_bytes, u64 min_bytes)
3181 {
3182 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3183 unsigned long next_zero;
3184 unsigned long i;
3185 unsigned long want_bits;
3186 unsigned long min_bits;
3187 unsigned long found_bits;
3188 unsigned long max_bits = 0;
3189 unsigned long start = 0;
3190 unsigned long total_found = 0;
3191 int ret;
3192
3193 i = offset_to_bit(entry->offset, ctl->unit,
3194 max_t(u64, offset, entry->offset));
3195 want_bits = bytes_to_bits(bytes, ctl->unit);
3196 min_bits = bytes_to_bits(min_bytes, ctl->unit);
3197
3198 /*
3199 * Don't bother looking for a cluster in this bitmap if it's heavily
3200 * fragmented.
3201 */
3202 if (entry->max_extent_size &&
3203 entry->max_extent_size < cont1_bytes)
3204 return -ENOSPC;
3205 again:
3206 found_bits = 0;
3207 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
3208 next_zero = find_next_zero_bit(entry->bitmap,
3209 BITS_PER_BITMAP, i);
3210 if (next_zero - i >= min_bits) {
3211 found_bits = next_zero - i;
3212 if (found_bits > max_bits)
3213 max_bits = found_bits;
3214 break;
3215 }
3216 if (next_zero - i > max_bits)
3217 max_bits = next_zero - i;
3218 i = next_zero;
3219 }
3220
3221 if (!found_bits) {
3222 entry->max_extent_size = (u64)max_bits * ctl->unit;
3223 return -ENOSPC;
3224 }
3225
3226 if (!total_found) {
3227 start = i;
3228 cluster->max_size = 0;
3229 }
3230
3231 total_found += found_bits;
3232
3233 if (cluster->max_size < found_bits * ctl->unit)
3234 cluster->max_size = found_bits * ctl->unit;
3235
3236 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
3237 i = next_zero + 1;
3238 goto again;
3239 }
3240
3241 cluster->window_start = start * ctl->unit + entry->offset;
3242 rb_erase(&entry->offset_index, &ctl->free_space_offset);
3243 ret = tree_insert_offset(&cluster->root, entry->offset,
3244 &entry->offset_index, 1);
3245 ASSERT(!ret); /* -EEXIST; Logic error */
3246
3247 trace_btrfs_setup_cluster(block_group, cluster,
3248 total_found * ctl->unit, 1);
3249 return 0;
3250 }
3251
3252 /*
3253 * This searches the block group for just extents to fill the cluster with.
3254 * Try to find a cluster with at least bytes total bytes, at least one
3255 * extent of cont1_bytes, and other clusters of at least min_bytes.
3256 */
3257 static noinline int
setup_cluster_no_bitmap(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster,struct list_head * bitmaps,u64 offset,u64 bytes,u64 cont1_bytes,u64 min_bytes)3258 setup_cluster_no_bitmap(struct btrfs_block_group *block_group,
3259 struct btrfs_free_cluster *cluster,
3260 struct list_head *bitmaps, u64 offset, u64 bytes,
3261 u64 cont1_bytes, u64 min_bytes)
3262 {
3263 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3264 struct btrfs_free_space *first = NULL;
3265 struct btrfs_free_space *entry = NULL;
3266 struct btrfs_free_space *last;
3267 struct rb_node *node;
3268 u64 window_free;
3269 u64 max_extent;
3270 u64 total_size = 0;
3271
3272 entry = tree_search_offset(ctl, offset, 0, 1);
3273 if (!entry)
3274 return -ENOSPC;
3275
3276 /*
3277 * We don't want bitmaps, so just move along until we find a normal
3278 * extent entry.
3279 */
3280 while (entry->bitmap || entry->bytes < min_bytes) {
3281 if (entry->bitmap && list_empty(&entry->list))
3282 list_add_tail(&entry->list, bitmaps);
3283 node = rb_next(&entry->offset_index);
3284 if (!node)
3285 return -ENOSPC;
3286 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3287 }
3288
3289 window_free = entry->bytes;
3290 max_extent = entry->bytes;
3291 first = entry;
3292 last = entry;
3293
3294 for (node = rb_next(&entry->offset_index); node;
3295 node = rb_next(&entry->offset_index)) {
3296 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3297
3298 if (entry->bitmap) {
3299 if (list_empty(&entry->list))
3300 list_add_tail(&entry->list, bitmaps);
3301 continue;
3302 }
3303
3304 if (entry->bytes < min_bytes)
3305 continue;
3306
3307 last = entry;
3308 window_free += entry->bytes;
3309 if (entry->bytes > max_extent)
3310 max_extent = entry->bytes;
3311 }
3312
3313 if (window_free < bytes || max_extent < cont1_bytes)
3314 return -ENOSPC;
3315
3316 cluster->window_start = first->offset;
3317
3318 node = &first->offset_index;
3319
3320 /*
3321 * now we've found our entries, pull them out of the free space
3322 * cache and put them into the cluster rbtree
3323 */
3324 do {
3325 int ret;
3326
3327 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3328 node = rb_next(&entry->offset_index);
3329 if (entry->bitmap || entry->bytes < min_bytes)
3330 continue;
3331
3332 rb_erase(&entry->offset_index, &ctl->free_space_offset);
3333 ret = tree_insert_offset(&cluster->root, entry->offset,
3334 &entry->offset_index, 0);
3335 total_size += entry->bytes;
3336 ASSERT(!ret); /* -EEXIST; Logic error */
3337 } while (node && entry != last);
3338
3339 cluster->max_size = max_extent;
3340 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
3341 return 0;
3342 }
3343
3344 /*
3345 * This specifically looks for bitmaps that may work in the cluster, we assume
3346 * that we have already failed to find extents that will work.
3347 */
3348 static noinline int
setup_cluster_bitmap(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster,struct list_head * bitmaps,u64 offset,u64 bytes,u64 cont1_bytes,u64 min_bytes)3349 setup_cluster_bitmap(struct btrfs_block_group *block_group,
3350 struct btrfs_free_cluster *cluster,
3351 struct list_head *bitmaps, u64 offset, u64 bytes,
3352 u64 cont1_bytes, u64 min_bytes)
3353 {
3354 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3355 struct btrfs_free_space *entry = NULL;
3356 int ret = -ENOSPC;
3357 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
3358
3359 if (ctl->total_bitmaps == 0)
3360 return -ENOSPC;
3361
3362 /*
3363 * The bitmap that covers offset won't be in the list unless offset
3364 * is just its start offset.
3365 */
3366 if (!list_empty(bitmaps))
3367 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
3368
3369 if (!entry || entry->offset != bitmap_offset) {
3370 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
3371 if (entry && list_empty(&entry->list))
3372 list_add(&entry->list, bitmaps);
3373 }
3374
3375 list_for_each_entry(entry, bitmaps, list) {
3376 if (entry->bytes < bytes)
3377 continue;
3378 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
3379 bytes, cont1_bytes, min_bytes);
3380 if (!ret)
3381 return 0;
3382 }
3383
3384 /*
3385 * The bitmaps list has all the bitmaps that record free space
3386 * starting after offset, so no more search is required.
3387 */
3388 return -ENOSPC;
3389 }
3390
3391 /*
3392 * here we try to find a cluster of blocks in a block group. The goal
3393 * is to find at least bytes+empty_size.
3394 * We might not find them all in one contiguous area.
3395 *
3396 * returns zero and sets up cluster if things worked out, otherwise
3397 * it returns -enospc
3398 */
btrfs_find_space_cluster(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster,u64 offset,u64 bytes,u64 empty_size)3399 int btrfs_find_space_cluster(struct btrfs_block_group *block_group,
3400 struct btrfs_free_cluster *cluster,
3401 u64 offset, u64 bytes, u64 empty_size)
3402 {
3403 struct btrfs_fs_info *fs_info = block_group->fs_info;
3404 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3405 struct btrfs_free_space *entry, *tmp;
3406 LIST_HEAD(bitmaps);
3407 u64 min_bytes;
3408 u64 cont1_bytes;
3409 int ret;
3410
3411 /*
3412 * Choose the minimum extent size we'll require for this
3413 * cluster. For SSD_SPREAD, don't allow any fragmentation.
3414 * For metadata, allow allocates with smaller extents. For
3415 * data, keep it dense.
3416 */
3417 if (btrfs_test_opt(fs_info, SSD_SPREAD)) {
3418 cont1_bytes = min_bytes = bytes + empty_size;
3419 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3420 cont1_bytes = bytes;
3421 min_bytes = fs_info->sectorsize;
3422 } else {
3423 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3424 min_bytes = fs_info->sectorsize;
3425 }
3426
3427 spin_lock(&ctl->tree_lock);
3428
3429 /*
3430 * If we know we don't have enough space to make a cluster don't even
3431 * bother doing all the work to try and find one.
3432 */
3433 if (ctl->free_space < bytes) {
3434 spin_unlock(&ctl->tree_lock);
3435 return -ENOSPC;
3436 }
3437
3438 spin_lock(&cluster->lock);
3439
3440 /* someone already found a cluster, hooray */
3441 if (cluster->block_group) {
3442 ret = 0;
3443 goto out;
3444 }
3445
3446 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
3447 min_bytes);
3448
3449 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
3450 bytes + empty_size,
3451 cont1_bytes, min_bytes);
3452 if (ret)
3453 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
3454 offset, bytes + empty_size,
3455 cont1_bytes, min_bytes);
3456
3457 /* Clear our temporary list */
3458 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3459 list_del_init(&entry->list);
3460
3461 if (!ret) {
3462 btrfs_get_block_group(block_group);
3463 list_add_tail(&cluster->block_group_list,
3464 &block_group->cluster_list);
3465 cluster->block_group = block_group;
3466 } else {
3467 trace_btrfs_failed_cluster_setup(block_group);
3468 }
3469 out:
3470 spin_unlock(&cluster->lock);
3471 spin_unlock(&ctl->tree_lock);
3472
3473 return ret;
3474 }
3475
3476 /*
3477 * simple code to zero out a cluster
3478 */
btrfs_init_free_cluster(struct btrfs_free_cluster * cluster)3479 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3480 {
3481 spin_lock_init(&cluster->lock);
3482 spin_lock_init(&cluster->refill_lock);
3483 cluster->root = RB_ROOT;
3484 cluster->max_size = 0;
3485 cluster->fragmented = false;
3486 INIT_LIST_HEAD(&cluster->block_group_list);
3487 cluster->block_group = NULL;
3488 }
3489
do_trimming(struct btrfs_block_group * block_group,u64 * total_trimmed,u64 start,u64 bytes,u64 reserved_start,u64 reserved_bytes,enum btrfs_trim_state reserved_trim_state,struct btrfs_trim_range * trim_entry)3490 static int do_trimming(struct btrfs_block_group *block_group,
3491 u64 *total_trimmed, u64 start, u64 bytes,
3492 u64 reserved_start, u64 reserved_bytes,
3493 enum btrfs_trim_state reserved_trim_state,
3494 struct btrfs_trim_range *trim_entry)
3495 {
3496 struct btrfs_space_info *space_info = block_group->space_info;
3497 struct btrfs_fs_info *fs_info = block_group->fs_info;
3498 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3499 int ret;
3500 int update = 0;
3501 const u64 end = start + bytes;
3502 const u64 reserved_end = reserved_start + reserved_bytes;
3503 enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3504 u64 trimmed = 0;
3505
3506 spin_lock(&space_info->lock);
3507 spin_lock(&block_group->lock);
3508 if (!block_group->ro) {
3509 block_group->reserved += reserved_bytes;
3510 space_info->bytes_reserved += reserved_bytes;
3511 update = 1;
3512 }
3513 spin_unlock(&block_group->lock);
3514 spin_unlock(&space_info->lock);
3515
3516 ret = btrfs_discard_extent(fs_info, start, bytes, &trimmed);
3517 if (!ret) {
3518 *total_trimmed += trimmed;
3519 trim_state = BTRFS_TRIM_STATE_TRIMMED;
3520 }
3521
3522 mutex_lock(&ctl->cache_writeout_mutex);
3523 if (reserved_start < start)
3524 __btrfs_add_free_space(fs_info, ctl, reserved_start,
3525 start - reserved_start,
3526 reserved_trim_state);
3527 if (start + bytes < reserved_start + reserved_bytes)
3528 __btrfs_add_free_space(fs_info, ctl, end, reserved_end - end,
3529 reserved_trim_state);
3530 __btrfs_add_free_space(fs_info, ctl, start, bytes, trim_state);
3531 list_del(&trim_entry->list);
3532 mutex_unlock(&ctl->cache_writeout_mutex);
3533
3534 if (update) {
3535 spin_lock(&space_info->lock);
3536 spin_lock(&block_group->lock);
3537 if (block_group->ro)
3538 space_info->bytes_readonly += reserved_bytes;
3539 block_group->reserved -= reserved_bytes;
3540 space_info->bytes_reserved -= reserved_bytes;
3541 spin_unlock(&block_group->lock);
3542 spin_unlock(&space_info->lock);
3543 }
3544
3545 return ret;
3546 }
3547
3548 /*
3549 * If @async is set, then we will trim 1 region and return.
3550 */
trim_no_bitmap(struct btrfs_block_group * block_group,u64 * total_trimmed,u64 start,u64 end,u64 minlen,bool async)3551 static int trim_no_bitmap(struct btrfs_block_group *block_group,
3552 u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3553 bool async)
3554 {
3555 struct btrfs_discard_ctl *discard_ctl =
3556 &block_group->fs_info->discard_ctl;
3557 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3558 struct btrfs_free_space *entry;
3559 struct rb_node *node;
3560 int ret = 0;
3561 u64 extent_start;
3562 u64 extent_bytes;
3563 enum btrfs_trim_state extent_trim_state;
3564 u64 bytes;
3565 const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3566
3567 while (start < end) {
3568 struct btrfs_trim_range trim_entry;
3569
3570 mutex_lock(&ctl->cache_writeout_mutex);
3571 spin_lock(&ctl->tree_lock);
3572
3573 if (ctl->free_space < minlen)
3574 goto out_unlock;
3575
3576 entry = tree_search_offset(ctl, start, 0, 1);
3577 if (!entry)
3578 goto out_unlock;
3579
3580 /* Skip bitmaps and if async, already trimmed entries */
3581 while (entry->bitmap ||
3582 (async && btrfs_free_space_trimmed(entry))) {
3583 node = rb_next(&entry->offset_index);
3584 if (!node)
3585 goto out_unlock;
3586 entry = rb_entry(node, struct btrfs_free_space,
3587 offset_index);
3588 }
3589
3590 if (entry->offset >= end)
3591 goto out_unlock;
3592
3593 extent_start = entry->offset;
3594 extent_bytes = entry->bytes;
3595 extent_trim_state = entry->trim_state;
3596 if (async) {
3597 start = entry->offset;
3598 bytes = entry->bytes;
3599 if (bytes < minlen) {
3600 spin_unlock(&ctl->tree_lock);
3601 mutex_unlock(&ctl->cache_writeout_mutex);
3602 goto next;
3603 }
3604 unlink_free_space(ctl, entry);
3605 /*
3606 * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3607 * If X < BTRFS_ASYNC_DISCARD_MIN_FILTER, we won't trim
3608 * X when we come back around. So trim it now.
3609 */
3610 if (max_discard_size &&
3611 bytes >= (max_discard_size +
3612 BTRFS_ASYNC_DISCARD_MIN_FILTER)) {
3613 bytes = max_discard_size;
3614 extent_bytes = max_discard_size;
3615 entry->offset += max_discard_size;
3616 entry->bytes -= max_discard_size;
3617 link_free_space(ctl, entry);
3618 } else {
3619 kmem_cache_free(btrfs_free_space_cachep, entry);
3620 }
3621 } else {
3622 start = max(start, extent_start);
3623 bytes = min(extent_start + extent_bytes, end) - start;
3624 if (bytes < minlen) {
3625 spin_unlock(&ctl->tree_lock);
3626 mutex_unlock(&ctl->cache_writeout_mutex);
3627 goto next;
3628 }
3629
3630 unlink_free_space(ctl, entry);
3631 kmem_cache_free(btrfs_free_space_cachep, entry);
3632 }
3633
3634 spin_unlock(&ctl->tree_lock);
3635 trim_entry.start = extent_start;
3636 trim_entry.bytes = extent_bytes;
3637 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3638 mutex_unlock(&ctl->cache_writeout_mutex);
3639
3640 ret = do_trimming(block_group, total_trimmed, start, bytes,
3641 extent_start, extent_bytes, extent_trim_state,
3642 &trim_entry);
3643 if (ret) {
3644 block_group->discard_cursor = start + bytes;
3645 break;
3646 }
3647 next:
3648 start += bytes;
3649 block_group->discard_cursor = start;
3650 if (async && *total_trimmed)
3651 break;
3652
3653 if (fatal_signal_pending(current)) {
3654 ret = -ERESTARTSYS;
3655 break;
3656 }
3657
3658 cond_resched();
3659 }
3660
3661 return ret;
3662
3663 out_unlock:
3664 block_group->discard_cursor = btrfs_block_group_end(block_group);
3665 spin_unlock(&ctl->tree_lock);
3666 mutex_unlock(&ctl->cache_writeout_mutex);
3667
3668 return ret;
3669 }
3670
3671 /*
3672 * If we break out of trimming a bitmap prematurely, we should reset the
3673 * trimming bit. In a rather contrieved case, it's possible to race here so
3674 * reset the state to BTRFS_TRIM_STATE_UNTRIMMED.
3675 *
3676 * start = start of bitmap
3677 * end = near end of bitmap
3678 *
3679 * Thread 1: Thread 2:
3680 * trim_bitmaps(start)
3681 * trim_bitmaps(end)
3682 * end_trimming_bitmap()
3683 * reset_trimming_bitmap()
3684 */
reset_trimming_bitmap(struct btrfs_free_space_ctl * ctl,u64 offset)3685 static void reset_trimming_bitmap(struct btrfs_free_space_ctl *ctl, u64 offset)
3686 {
3687 struct btrfs_free_space *entry;
3688
3689 spin_lock(&ctl->tree_lock);
3690 entry = tree_search_offset(ctl, offset, 1, 0);
3691 if (entry) {
3692 if (btrfs_free_space_trimmed(entry)) {
3693 ctl->discardable_extents[BTRFS_STAT_CURR] +=
3694 entry->bitmap_extents;
3695 ctl->discardable_bytes[BTRFS_STAT_CURR] += entry->bytes;
3696 }
3697 entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3698 }
3699
3700 spin_unlock(&ctl->tree_lock);
3701 }
3702
end_trimming_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * entry)3703 static void end_trimming_bitmap(struct btrfs_free_space_ctl *ctl,
3704 struct btrfs_free_space *entry)
3705 {
3706 if (btrfs_free_space_trimming_bitmap(entry)) {
3707 entry->trim_state = BTRFS_TRIM_STATE_TRIMMED;
3708 ctl->discardable_extents[BTRFS_STAT_CURR] -=
3709 entry->bitmap_extents;
3710 ctl->discardable_bytes[BTRFS_STAT_CURR] -= entry->bytes;
3711 }
3712 }
3713
3714 /*
3715 * If @async is set, then we will trim 1 region and return.
3716 */
trim_bitmaps(struct btrfs_block_group * block_group,u64 * total_trimmed,u64 start,u64 end,u64 minlen,u64 maxlen,bool async)3717 static int trim_bitmaps(struct btrfs_block_group *block_group,
3718 u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3719 u64 maxlen, bool async)
3720 {
3721 struct btrfs_discard_ctl *discard_ctl =
3722 &block_group->fs_info->discard_ctl;
3723 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3724 struct btrfs_free_space *entry;
3725 int ret = 0;
3726 int ret2;
3727 u64 bytes;
3728 u64 offset = offset_to_bitmap(ctl, start);
3729 const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3730
3731 while (offset < end) {
3732 bool next_bitmap = false;
3733 struct btrfs_trim_range trim_entry;
3734
3735 mutex_lock(&ctl->cache_writeout_mutex);
3736 spin_lock(&ctl->tree_lock);
3737
3738 if (ctl->free_space < minlen) {
3739 block_group->discard_cursor =
3740 btrfs_block_group_end(block_group);
3741 spin_unlock(&ctl->tree_lock);
3742 mutex_unlock(&ctl->cache_writeout_mutex);
3743 break;
3744 }
3745
3746 entry = tree_search_offset(ctl, offset, 1, 0);
3747 /*
3748 * Bitmaps are marked trimmed lossily now to prevent constant
3749 * discarding of the same bitmap (the reason why we are bound
3750 * by the filters). So, retrim the block group bitmaps when we
3751 * are preparing to punt to the unused_bgs list. This uses
3752 * @minlen to determine if we are in BTRFS_DISCARD_INDEX_UNUSED
3753 * which is the only discard index which sets minlen to 0.
3754 */
3755 if (!entry || (async && minlen && start == offset &&
3756 btrfs_free_space_trimmed(entry))) {
3757 spin_unlock(&ctl->tree_lock);
3758 mutex_unlock(&ctl->cache_writeout_mutex);
3759 next_bitmap = true;
3760 goto next;
3761 }
3762
3763 /*
3764 * Async discard bitmap trimming begins at by setting the start
3765 * to be key.objectid and the offset_to_bitmap() aligns to the
3766 * start of the bitmap. This lets us know we are fully
3767 * scanning the bitmap rather than only some portion of it.
3768 */
3769 if (start == offset)
3770 entry->trim_state = BTRFS_TRIM_STATE_TRIMMING;
3771
3772 bytes = minlen;
3773 ret2 = search_bitmap(ctl, entry, &start, &bytes, false);
3774 if (ret2 || start >= end) {
3775 /*
3776 * We lossily consider a bitmap trimmed if we only skip
3777 * over regions <= BTRFS_ASYNC_DISCARD_MIN_FILTER.
3778 */
3779 if (ret2 && minlen <= BTRFS_ASYNC_DISCARD_MIN_FILTER)
3780 end_trimming_bitmap(ctl, entry);
3781 else
3782 entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3783 spin_unlock(&ctl->tree_lock);
3784 mutex_unlock(&ctl->cache_writeout_mutex);
3785 next_bitmap = true;
3786 goto next;
3787 }
3788
3789 /*
3790 * We already trimmed a region, but are using the locking above
3791 * to reset the trim_state.
3792 */
3793 if (async && *total_trimmed) {
3794 spin_unlock(&ctl->tree_lock);
3795 mutex_unlock(&ctl->cache_writeout_mutex);
3796 goto out;
3797 }
3798
3799 bytes = min(bytes, end - start);
3800 if (bytes < minlen || (async && maxlen && bytes > maxlen)) {
3801 spin_unlock(&ctl->tree_lock);
3802 mutex_unlock(&ctl->cache_writeout_mutex);
3803 goto next;
3804 }
3805
3806 /*
3807 * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3808 * If X < @minlen, we won't trim X when we come back around.
3809 * So trim it now. We differ here from trimming extents as we
3810 * don't keep individual state per bit.
3811 */
3812 if (async &&
3813 max_discard_size &&
3814 bytes > (max_discard_size + minlen))
3815 bytes = max_discard_size;
3816
3817 bitmap_clear_bits(ctl, entry, start, bytes);
3818 if (entry->bytes == 0)
3819 free_bitmap(ctl, entry);
3820
3821 spin_unlock(&ctl->tree_lock);
3822 trim_entry.start = start;
3823 trim_entry.bytes = bytes;
3824 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3825 mutex_unlock(&ctl->cache_writeout_mutex);
3826
3827 ret = do_trimming(block_group, total_trimmed, start, bytes,
3828 start, bytes, 0, &trim_entry);
3829 if (ret) {
3830 reset_trimming_bitmap(ctl, offset);
3831 block_group->discard_cursor =
3832 btrfs_block_group_end(block_group);
3833 break;
3834 }
3835 next:
3836 if (next_bitmap) {
3837 offset += BITS_PER_BITMAP * ctl->unit;
3838 start = offset;
3839 } else {
3840 start += bytes;
3841 }
3842 block_group->discard_cursor = start;
3843
3844 if (fatal_signal_pending(current)) {
3845 if (start != offset)
3846 reset_trimming_bitmap(ctl, offset);
3847 ret = -ERESTARTSYS;
3848 break;
3849 }
3850
3851 cond_resched();
3852 }
3853
3854 if (offset >= end)
3855 block_group->discard_cursor = end;
3856
3857 out:
3858 return ret;
3859 }
3860
btrfs_trim_block_group(struct btrfs_block_group * block_group,u64 * trimmed,u64 start,u64 end,u64 minlen)3861 int btrfs_trim_block_group(struct btrfs_block_group *block_group,
3862 u64 *trimmed, u64 start, u64 end, u64 minlen)
3863 {
3864 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3865 int ret;
3866 u64 rem = 0;
3867
3868 ASSERT(!btrfs_is_zoned(block_group->fs_info));
3869
3870 *trimmed = 0;
3871
3872 spin_lock(&block_group->lock);
3873 if (block_group->removed) {
3874 spin_unlock(&block_group->lock);
3875 return 0;
3876 }
3877 btrfs_freeze_block_group(block_group);
3878 spin_unlock(&block_group->lock);
3879
3880 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, false);
3881 if (ret)
3882 goto out;
3883
3884 ret = trim_bitmaps(block_group, trimmed, start, end, minlen, 0, false);
3885 div64_u64_rem(end, BITS_PER_BITMAP * ctl->unit, &rem);
3886 /* If we ended in the middle of a bitmap, reset the trimming flag */
3887 if (rem)
3888 reset_trimming_bitmap(ctl, offset_to_bitmap(ctl, end));
3889 out:
3890 btrfs_unfreeze_block_group(block_group);
3891 return ret;
3892 }
3893
btrfs_trim_block_group_extents(struct btrfs_block_group * block_group,u64 * trimmed,u64 start,u64 end,u64 minlen,bool async)3894 int btrfs_trim_block_group_extents(struct btrfs_block_group *block_group,
3895 u64 *trimmed, u64 start, u64 end, u64 minlen,
3896 bool async)
3897 {
3898 int ret;
3899
3900 *trimmed = 0;
3901
3902 spin_lock(&block_group->lock);
3903 if (block_group->removed) {
3904 spin_unlock(&block_group->lock);
3905 return 0;
3906 }
3907 btrfs_freeze_block_group(block_group);
3908 spin_unlock(&block_group->lock);
3909
3910 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, async);
3911 btrfs_unfreeze_block_group(block_group);
3912
3913 return ret;
3914 }
3915
btrfs_trim_block_group_bitmaps(struct btrfs_block_group * block_group,u64 * trimmed,u64 start,u64 end,u64 minlen,u64 maxlen,bool async)3916 int btrfs_trim_block_group_bitmaps(struct btrfs_block_group *block_group,
3917 u64 *trimmed, u64 start, u64 end, u64 minlen,
3918 u64 maxlen, bool async)
3919 {
3920 int ret;
3921
3922 *trimmed = 0;
3923
3924 spin_lock(&block_group->lock);
3925 if (block_group->removed) {
3926 spin_unlock(&block_group->lock);
3927 return 0;
3928 }
3929 btrfs_freeze_block_group(block_group);
3930 spin_unlock(&block_group->lock);
3931
3932 ret = trim_bitmaps(block_group, trimmed, start, end, minlen, maxlen,
3933 async);
3934
3935 btrfs_unfreeze_block_group(block_group);
3936
3937 return ret;
3938 }
3939
btrfs_free_space_cache_v1_active(struct btrfs_fs_info * fs_info)3940 bool btrfs_free_space_cache_v1_active(struct btrfs_fs_info *fs_info)
3941 {
3942 return btrfs_super_cache_generation(fs_info->super_copy);
3943 }
3944
cleanup_free_space_cache_v1(struct btrfs_fs_info * fs_info,struct btrfs_trans_handle * trans)3945 static int cleanup_free_space_cache_v1(struct btrfs_fs_info *fs_info,
3946 struct btrfs_trans_handle *trans)
3947 {
3948 struct btrfs_block_group *block_group;
3949 struct rb_node *node;
3950 int ret = 0;
3951
3952 btrfs_info(fs_info, "cleaning free space cache v1");
3953
3954 node = rb_first(&fs_info->block_group_cache_tree);
3955 while (node) {
3956 block_group = rb_entry(node, struct btrfs_block_group, cache_node);
3957 ret = btrfs_remove_free_space_inode(trans, NULL, block_group);
3958 if (ret)
3959 goto out;
3960 node = rb_next(node);
3961 }
3962 out:
3963 return ret;
3964 }
3965
btrfs_set_free_space_cache_v1_active(struct btrfs_fs_info * fs_info,bool active)3966 int btrfs_set_free_space_cache_v1_active(struct btrfs_fs_info *fs_info, bool active)
3967 {
3968 struct btrfs_trans_handle *trans;
3969 int ret;
3970
3971 /*
3972 * update_super_roots will appropriately set or unset
3973 * super_copy->cache_generation based on SPACE_CACHE and
3974 * BTRFS_FS_CLEANUP_SPACE_CACHE_V1. For this reason, we need a
3975 * transaction commit whether we are enabling space cache v1 and don't
3976 * have any other work to do, or are disabling it and removing free
3977 * space inodes.
3978 */
3979 trans = btrfs_start_transaction(fs_info->tree_root, 0);
3980 if (IS_ERR(trans))
3981 return PTR_ERR(trans);
3982
3983 if (!active) {
3984 set_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);
3985 ret = cleanup_free_space_cache_v1(fs_info, trans);
3986 if (ret) {
3987 btrfs_abort_transaction(trans, ret);
3988 btrfs_end_transaction(trans);
3989 goto out;
3990 }
3991 }
3992
3993 ret = btrfs_commit_transaction(trans);
3994 out:
3995 clear_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);
3996
3997 return ret;
3998 }
3999
4000 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4001 /*
4002 * Use this if you need to make a bitmap or extent entry specifically, it
4003 * doesn't do any of the merging that add_free_space does, this acts a lot like
4004 * how the free space cache loading stuff works, so you can get really weird
4005 * configurations.
4006 */
test_add_free_space_entry(struct btrfs_block_group * cache,u64 offset,u64 bytes,bool bitmap)4007 int test_add_free_space_entry(struct btrfs_block_group *cache,
4008 u64 offset, u64 bytes, bool bitmap)
4009 {
4010 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4011 struct btrfs_free_space *info = NULL, *bitmap_info;
4012 void *map = NULL;
4013 enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_TRIMMED;
4014 u64 bytes_added;
4015 int ret;
4016
4017 again:
4018 if (!info) {
4019 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
4020 if (!info)
4021 return -ENOMEM;
4022 }
4023
4024 if (!bitmap) {
4025 spin_lock(&ctl->tree_lock);
4026 info->offset = offset;
4027 info->bytes = bytes;
4028 info->max_extent_size = 0;
4029 ret = link_free_space(ctl, info);
4030 spin_unlock(&ctl->tree_lock);
4031 if (ret)
4032 kmem_cache_free(btrfs_free_space_cachep, info);
4033 return ret;
4034 }
4035
4036 if (!map) {
4037 map = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep, GFP_NOFS);
4038 if (!map) {
4039 kmem_cache_free(btrfs_free_space_cachep, info);
4040 return -ENOMEM;
4041 }
4042 }
4043
4044 spin_lock(&ctl->tree_lock);
4045 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
4046 1, 0);
4047 if (!bitmap_info) {
4048 info->bitmap = map;
4049 map = NULL;
4050 add_new_bitmap(ctl, info, offset);
4051 bitmap_info = info;
4052 info = NULL;
4053 }
4054
4055 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
4056 trim_state);
4057
4058 bytes -= bytes_added;
4059 offset += bytes_added;
4060 spin_unlock(&ctl->tree_lock);
4061
4062 if (bytes)
4063 goto again;
4064
4065 if (info)
4066 kmem_cache_free(btrfs_free_space_cachep, info);
4067 if (map)
4068 kmem_cache_free(btrfs_free_space_bitmap_cachep, map);
4069 return 0;
4070 }
4071
4072 /*
4073 * Checks to see if the given range is in the free space cache. This is really
4074 * just used to check the absence of space, so if there is free space in the
4075 * range at all we will return 1.
4076 */
test_check_exists(struct btrfs_block_group * cache,u64 offset,u64 bytes)4077 int test_check_exists(struct btrfs_block_group *cache,
4078 u64 offset, u64 bytes)
4079 {
4080 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4081 struct btrfs_free_space *info;
4082 int ret = 0;
4083
4084 spin_lock(&ctl->tree_lock);
4085 info = tree_search_offset(ctl, offset, 0, 0);
4086 if (!info) {
4087 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
4088 1, 0);
4089 if (!info)
4090 goto out;
4091 }
4092
4093 have_info:
4094 if (info->bitmap) {
4095 u64 bit_off, bit_bytes;
4096 struct rb_node *n;
4097 struct btrfs_free_space *tmp;
4098
4099 bit_off = offset;
4100 bit_bytes = ctl->unit;
4101 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false);
4102 if (!ret) {
4103 if (bit_off == offset) {
4104 ret = 1;
4105 goto out;
4106 } else if (bit_off > offset &&
4107 offset + bytes > bit_off) {
4108 ret = 1;
4109 goto out;
4110 }
4111 }
4112
4113 n = rb_prev(&info->offset_index);
4114 while (n) {
4115 tmp = rb_entry(n, struct btrfs_free_space,
4116 offset_index);
4117 if (tmp->offset + tmp->bytes < offset)
4118 break;
4119 if (offset + bytes < tmp->offset) {
4120 n = rb_prev(&tmp->offset_index);
4121 continue;
4122 }
4123 info = tmp;
4124 goto have_info;
4125 }
4126
4127 n = rb_next(&info->offset_index);
4128 while (n) {
4129 tmp = rb_entry(n, struct btrfs_free_space,
4130 offset_index);
4131 if (offset + bytes < tmp->offset)
4132 break;
4133 if (tmp->offset + tmp->bytes < offset) {
4134 n = rb_next(&tmp->offset_index);
4135 continue;
4136 }
4137 info = tmp;
4138 goto have_info;
4139 }
4140
4141 ret = 0;
4142 goto out;
4143 }
4144
4145 if (info->offset == offset) {
4146 ret = 1;
4147 goto out;
4148 }
4149
4150 if (offset > info->offset && offset < info->offset + info->bytes)
4151 ret = 1;
4152 out:
4153 spin_unlock(&ctl->tree_lock);
4154 return ret;
4155 }
4156 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */
4157