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 #include "subpage.h"
26 #include "inode-item.h"
27
28 #define BITS_PER_BITMAP (PAGE_SIZE * 8UL)
29 #define MAX_CACHE_BYTES_PER_GIG SZ_64K
30 #define FORCE_EXTENT_THRESHOLD SZ_1M
31
32 struct btrfs_trim_range {
33 u64 start;
34 u64 bytes;
35 struct list_head list;
36 };
37
38 static int link_free_space(struct btrfs_free_space_ctl *ctl,
39 struct btrfs_free_space *info);
40 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
41 struct btrfs_free_space *info, bool update_stat);
42 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
43 struct btrfs_free_space *bitmap_info, u64 *offset,
44 u64 *bytes, bool for_alloc);
45 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
46 struct btrfs_free_space *bitmap_info);
47 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
48 struct btrfs_free_space *info, u64 offset,
49 u64 bytes, bool update_stats);
50
__lookup_free_space_inode(struct btrfs_root * root,struct btrfs_path * path,u64 offset)51 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
52 struct btrfs_path *path,
53 u64 offset)
54 {
55 struct btrfs_fs_info *fs_info = root->fs_info;
56 struct btrfs_key key;
57 struct btrfs_key location;
58 struct btrfs_disk_key disk_key;
59 struct btrfs_free_space_header *header;
60 struct extent_buffer *leaf;
61 struct inode *inode = NULL;
62 unsigned nofs_flag;
63 int ret;
64
65 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
66 key.offset = offset;
67 key.type = 0;
68
69 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
70 if (ret < 0)
71 return ERR_PTR(ret);
72 if (ret > 0) {
73 btrfs_release_path(path);
74 return ERR_PTR(-ENOENT);
75 }
76
77 leaf = path->nodes[0];
78 header = btrfs_item_ptr(leaf, path->slots[0],
79 struct btrfs_free_space_header);
80 btrfs_free_space_key(leaf, header, &disk_key);
81 btrfs_disk_key_to_cpu(&location, &disk_key);
82 btrfs_release_path(path);
83
84 /*
85 * We are often under a trans handle at this point, so we need to make
86 * sure NOFS is set to keep us from deadlocking.
87 */
88 nofs_flag = memalloc_nofs_save();
89 inode = btrfs_iget_path(fs_info->sb, location.objectid, root, path);
90 btrfs_release_path(path);
91 memalloc_nofs_restore(nofs_flag);
92 if (IS_ERR(inode))
93 return inode;
94
95 mapping_set_gfp_mask(inode->i_mapping,
96 mapping_gfp_constraint(inode->i_mapping,
97 ~(__GFP_FS | __GFP_HIGHMEM)));
98
99 return inode;
100 }
101
lookup_free_space_inode(struct btrfs_block_group * block_group,struct btrfs_path * path)102 struct inode *lookup_free_space_inode(struct btrfs_block_group *block_group,
103 struct btrfs_path *path)
104 {
105 struct btrfs_fs_info *fs_info = block_group->fs_info;
106 struct inode *inode = NULL;
107 u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
108
109 spin_lock(&block_group->lock);
110 if (block_group->inode)
111 inode = igrab(block_group->inode);
112 spin_unlock(&block_group->lock);
113 if (inode)
114 return inode;
115
116 inode = __lookup_free_space_inode(fs_info->tree_root, path,
117 block_group->start);
118 if (IS_ERR(inode))
119 return inode;
120
121 spin_lock(&block_group->lock);
122 if (!((BTRFS_I(inode)->flags & flags) == flags)) {
123 btrfs_info(fs_info, "Old style space inode found, converting.");
124 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
125 BTRFS_INODE_NODATACOW;
126 block_group->disk_cache_state = BTRFS_DC_CLEAR;
127 }
128
129 if (!block_group->iref) {
130 block_group->inode = igrab(inode);
131 block_group->iref = 1;
132 }
133 spin_unlock(&block_group->lock);
134
135 return inode;
136 }
137
__create_free_space_inode(struct btrfs_root * root,struct btrfs_trans_handle * trans,struct btrfs_path * path,u64 ino,u64 offset)138 static int __create_free_space_inode(struct btrfs_root *root,
139 struct btrfs_trans_handle *trans,
140 struct btrfs_path *path,
141 u64 ino, u64 offset)
142 {
143 struct btrfs_key key;
144 struct btrfs_disk_key disk_key;
145 struct btrfs_free_space_header *header;
146 struct btrfs_inode_item *inode_item;
147 struct extent_buffer *leaf;
148 /* We inline CRCs for the free disk space cache */
149 const u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC |
150 BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
151 int ret;
152
153 ret = btrfs_insert_empty_inode(trans, root, path, ino);
154 if (ret)
155 return ret;
156
157 leaf = path->nodes[0];
158 inode_item = btrfs_item_ptr(leaf, path->slots[0],
159 struct btrfs_inode_item);
160 btrfs_item_key(leaf, &disk_key, path->slots[0]);
161 memzero_extent_buffer(leaf, (unsigned long)inode_item,
162 sizeof(*inode_item));
163 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
164 btrfs_set_inode_size(leaf, inode_item, 0);
165 btrfs_set_inode_nbytes(leaf, inode_item, 0);
166 btrfs_set_inode_uid(leaf, inode_item, 0);
167 btrfs_set_inode_gid(leaf, inode_item, 0);
168 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
169 btrfs_set_inode_flags(leaf, inode_item, flags);
170 btrfs_set_inode_nlink(leaf, inode_item, 1);
171 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
172 btrfs_set_inode_block_group(leaf, inode_item, offset);
173 btrfs_mark_buffer_dirty(leaf);
174 btrfs_release_path(path);
175
176 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
177 key.offset = offset;
178 key.type = 0;
179 ret = btrfs_insert_empty_item(trans, root, path, &key,
180 sizeof(struct btrfs_free_space_header));
181 if (ret < 0) {
182 btrfs_release_path(path);
183 return ret;
184 }
185
186 leaf = path->nodes[0];
187 header = btrfs_item_ptr(leaf, path->slots[0],
188 struct btrfs_free_space_header);
189 memzero_extent_buffer(leaf, (unsigned long)header, sizeof(*header));
190 btrfs_set_free_space_key(leaf, header, &disk_key);
191 btrfs_mark_buffer_dirty(leaf);
192 btrfs_release_path(path);
193
194 return 0;
195 }
196
create_free_space_inode(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path)197 int create_free_space_inode(struct btrfs_trans_handle *trans,
198 struct btrfs_block_group *block_group,
199 struct btrfs_path *path)
200 {
201 int ret;
202 u64 ino;
203
204 ret = btrfs_get_free_objectid(trans->fs_info->tree_root, &ino);
205 if (ret < 0)
206 return ret;
207
208 return __create_free_space_inode(trans->fs_info->tree_root, trans, path,
209 ino, block_group->start);
210 }
211
212 /*
213 * inode is an optional sink: if it is NULL, btrfs_remove_free_space_inode
214 * handles lookup, otherwise it takes ownership and iputs the inode.
215 * Don't reuse an inode pointer after passing it into this function.
216 */
btrfs_remove_free_space_inode(struct btrfs_trans_handle * trans,struct inode * inode,struct btrfs_block_group * block_group)217 int btrfs_remove_free_space_inode(struct btrfs_trans_handle *trans,
218 struct inode *inode,
219 struct btrfs_block_group *block_group)
220 {
221 struct btrfs_path *path;
222 struct btrfs_key key;
223 int ret = 0;
224
225 path = btrfs_alloc_path();
226 if (!path)
227 return -ENOMEM;
228
229 if (!inode)
230 inode = lookup_free_space_inode(block_group, path);
231 if (IS_ERR(inode)) {
232 if (PTR_ERR(inode) != -ENOENT)
233 ret = PTR_ERR(inode);
234 goto out;
235 }
236 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
237 if (ret) {
238 btrfs_add_delayed_iput(inode);
239 goto out;
240 }
241 clear_nlink(inode);
242 /* One for the block groups ref */
243 spin_lock(&block_group->lock);
244 if (block_group->iref) {
245 block_group->iref = 0;
246 block_group->inode = NULL;
247 spin_unlock(&block_group->lock);
248 iput(inode);
249 } else {
250 spin_unlock(&block_group->lock);
251 }
252 /* One for the lookup ref */
253 btrfs_add_delayed_iput(inode);
254
255 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
256 key.type = 0;
257 key.offset = block_group->start;
258 ret = btrfs_search_slot(trans, trans->fs_info->tree_root, &key, path,
259 -1, 1);
260 if (ret) {
261 if (ret > 0)
262 ret = 0;
263 goto out;
264 }
265 ret = btrfs_del_item(trans, trans->fs_info->tree_root, path);
266 out:
267 btrfs_free_path(path);
268 return ret;
269 }
270
btrfs_check_trunc_cache_free_space(struct btrfs_fs_info * fs_info,struct btrfs_block_rsv * rsv)271 int btrfs_check_trunc_cache_free_space(struct btrfs_fs_info *fs_info,
272 struct btrfs_block_rsv *rsv)
273 {
274 u64 needed_bytes;
275 int ret;
276
277 /* 1 for slack space, 1 for updating the inode */
278 needed_bytes = btrfs_calc_insert_metadata_size(fs_info, 1) +
279 btrfs_calc_metadata_size(fs_info, 1);
280
281 spin_lock(&rsv->lock);
282 if (rsv->reserved < needed_bytes)
283 ret = -ENOSPC;
284 else
285 ret = 0;
286 spin_unlock(&rsv->lock);
287 return ret;
288 }
289
btrfs_truncate_free_space_cache(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct inode * vfs_inode)290 int btrfs_truncate_free_space_cache(struct btrfs_trans_handle *trans,
291 struct btrfs_block_group *block_group,
292 struct inode *vfs_inode)
293 {
294 struct btrfs_truncate_control control = {
295 .inode = BTRFS_I(vfs_inode),
296 .new_size = 0,
297 .ino = btrfs_ino(BTRFS_I(vfs_inode)),
298 .min_type = BTRFS_EXTENT_DATA_KEY,
299 .clear_extent_range = true,
300 };
301 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
302 struct btrfs_root *root = inode->root;
303 struct extent_state *cached_state = NULL;
304 int ret = 0;
305 bool locked = false;
306
307 if (block_group) {
308 struct btrfs_path *path = btrfs_alloc_path();
309
310 if (!path) {
311 ret = -ENOMEM;
312 goto fail;
313 }
314 locked = true;
315 mutex_lock(&trans->transaction->cache_write_mutex);
316 if (!list_empty(&block_group->io_list)) {
317 list_del_init(&block_group->io_list);
318
319 btrfs_wait_cache_io(trans, block_group, path);
320 btrfs_put_block_group(block_group);
321 }
322
323 /*
324 * now that we've truncated the cache away, its no longer
325 * setup or written
326 */
327 spin_lock(&block_group->lock);
328 block_group->disk_cache_state = BTRFS_DC_CLEAR;
329 spin_unlock(&block_group->lock);
330 btrfs_free_path(path);
331 }
332
333 btrfs_i_size_write(inode, 0);
334 truncate_pagecache(vfs_inode, 0);
335
336 lock_extent_bits(&inode->io_tree, 0, (u64)-1, &cached_state);
337 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
338
339 /*
340 * We skip the throttling logic for free space cache inodes, so we don't
341 * need to check for -EAGAIN.
342 */
343 ret = btrfs_truncate_inode_items(trans, root, &control);
344
345 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
346 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
347
348 unlock_extent_cached(&inode->io_tree, 0, (u64)-1, &cached_state);
349 if (ret)
350 goto fail;
351
352 ret = btrfs_update_inode(trans, root, inode);
353
354 fail:
355 if (locked)
356 mutex_unlock(&trans->transaction->cache_write_mutex);
357 if (ret)
358 btrfs_abort_transaction(trans, ret);
359
360 return ret;
361 }
362
readahead_cache(struct inode * inode)363 static void readahead_cache(struct inode *inode)
364 {
365 struct file_ra_state ra;
366 unsigned long last_index;
367
368 file_ra_state_init(&ra, inode->i_mapping);
369 last_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
370
371 page_cache_sync_readahead(inode->i_mapping, &ra, NULL, 0, last_index);
372 }
373
io_ctl_init(struct btrfs_io_ctl * io_ctl,struct inode * inode,int write)374 static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode,
375 int write)
376 {
377 int num_pages;
378
379 num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
380
381 /* Make sure we can fit our crcs and generation into the first page */
382 if (write && (num_pages * sizeof(u32) + sizeof(u64)) > PAGE_SIZE)
383 return -ENOSPC;
384
385 memset(io_ctl, 0, sizeof(struct btrfs_io_ctl));
386
387 io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS);
388 if (!io_ctl->pages)
389 return -ENOMEM;
390
391 io_ctl->num_pages = num_pages;
392 io_ctl->fs_info = btrfs_sb(inode->i_sb);
393 io_ctl->inode = inode;
394
395 return 0;
396 }
397 ALLOW_ERROR_INJECTION(io_ctl_init, ERRNO);
398
io_ctl_free(struct btrfs_io_ctl * io_ctl)399 static void io_ctl_free(struct btrfs_io_ctl *io_ctl)
400 {
401 kfree(io_ctl->pages);
402 io_ctl->pages = NULL;
403 }
404
io_ctl_unmap_page(struct btrfs_io_ctl * io_ctl)405 static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl)
406 {
407 if (io_ctl->cur) {
408 io_ctl->cur = NULL;
409 io_ctl->orig = NULL;
410 }
411 }
412
io_ctl_map_page(struct btrfs_io_ctl * io_ctl,int clear)413 static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear)
414 {
415 ASSERT(io_ctl->index < io_ctl->num_pages);
416 io_ctl->page = io_ctl->pages[io_ctl->index++];
417 io_ctl->cur = page_address(io_ctl->page);
418 io_ctl->orig = io_ctl->cur;
419 io_ctl->size = PAGE_SIZE;
420 if (clear)
421 clear_page(io_ctl->cur);
422 }
423
io_ctl_drop_pages(struct btrfs_io_ctl * io_ctl)424 static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl)
425 {
426 int i;
427
428 io_ctl_unmap_page(io_ctl);
429
430 for (i = 0; i < io_ctl->num_pages; i++) {
431 if (io_ctl->pages[i]) {
432 btrfs_page_clear_checked(io_ctl->fs_info,
433 io_ctl->pages[i],
434 page_offset(io_ctl->pages[i]),
435 PAGE_SIZE);
436 unlock_page(io_ctl->pages[i]);
437 put_page(io_ctl->pages[i]);
438 }
439 }
440 }
441
io_ctl_prepare_pages(struct btrfs_io_ctl * io_ctl,bool uptodate)442 static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, bool uptodate)
443 {
444 struct page *page;
445 struct inode *inode = io_ctl->inode;
446 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
447 int i;
448
449 for (i = 0; i < io_ctl->num_pages; i++) {
450 int ret;
451
452 page = find_or_create_page(inode->i_mapping, i, mask);
453 if (!page) {
454 io_ctl_drop_pages(io_ctl);
455 return -ENOMEM;
456 }
457
458 ret = set_page_extent_mapped(page);
459 if (ret < 0) {
460 unlock_page(page);
461 put_page(page);
462 io_ctl_drop_pages(io_ctl);
463 return ret;
464 }
465
466 io_ctl->pages[i] = page;
467 if (uptodate && !PageUptodate(page)) {
468 btrfs_read_folio(NULL, page_folio(page));
469 lock_page(page);
470 if (page->mapping != inode->i_mapping) {
471 btrfs_err(BTRFS_I(inode)->root->fs_info,
472 "free space cache page truncated");
473 io_ctl_drop_pages(io_ctl);
474 return -EIO;
475 }
476 if (!PageUptodate(page)) {
477 btrfs_err(BTRFS_I(inode)->root->fs_info,
478 "error reading free space cache");
479 io_ctl_drop_pages(io_ctl);
480 return -EIO;
481 }
482 }
483 }
484
485 for (i = 0; i < io_ctl->num_pages; i++)
486 clear_page_dirty_for_io(io_ctl->pages[i]);
487
488 return 0;
489 }
490
io_ctl_set_generation(struct btrfs_io_ctl * io_ctl,u64 generation)491 static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
492 {
493 io_ctl_map_page(io_ctl, 1);
494
495 /*
496 * Skip the csum areas. If we don't check crcs then we just have a
497 * 64bit chunk at the front of the first page.
498 */
499 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
500 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
501
502 put_unaligned_le64(generation, io_ctl->cur);
503 io_ctl->cur += sizeof(u64);
504 }
505
io_ctl_check_generation(struct btrfs_io_ctl * io_ctl,u64 generation)506 static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
507 {
508 u64 cache_gen;
509
510 /*
511 * Skip the crc area. If we don't check crcs then we just have a 64bit
512 * chunk at the front of the first page.
513 */
514 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
515 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
516
517 cache_gen = get_unaligned_le64(io_ctl->cur);
518 if (cache_gen != generation) {
519 btrfs_err_rl(io_ctl->fs_info,
520 "space cache generation (%llu) does not match inode (%llu)",
521 cache_gen, generation);
522 io_ctl_unmap_page(io_ctl);
523 return -EIO;
524 }
525 io_ctl->cur += sizeof(u64);
526 return 0;
527 }
528
io_ctl_set_crc(struct btrfs_io_ctl * io_ctl,int index)529 static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
530 {
531 u32 *tmp;
532 u32 crc = ~(u32)0;
533 unsigned offset = 0;
534
535 if (index == 0)
536 offset = sizeof(u32) * io_ctl->num_pages;
537
538 crc = btrfs_crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
539 btrfs_crc32c_final(crc, (u8 *)&crc);
540 io_ctl_unmap_page(io_ctl);
541 tmp = page_address(io_ctl->pages[0]);
542 tmp += index;
543 *tmp = crc;
544 }
545
io_ctl_check_crc(struct btrfs_io_ctl * io_ctl,int index)546 static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
547 {
548 u32 *tmp, val;
549 u32 crc = ~(u32)0;
550 unsigned offset = 0;
551
552 if (index == 0)
553 offset = sizeof(u32) * io_ctl->num_pages;
554
555 tmp = page_address(io_ctl->pages[0]);
556 tmp += index;
557 val = *tmp;
558
559 io_ctl_map_page(io_ctl, 0);
560 crc = btrfs_crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
561 btrfs_crc32c_final(crc, (u8 *)&crc);
562 if (val != crc) {
563 btrfs_err_rl(io_ctl->fs_info,
564 "csum mismatch on free space cache");
565 io_ctl_unmap_page(io_ctl);
566 return -EIO;
567 }
568
569 return 0;
570 }
571
io_ctl_add_entry(struct btrfs_io_ctl * io_ctl,u64 offset,u64 bytes,void * bitmap)572 static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
573 void *bitmap)
574 {
575 struct btrfs_free_space_entry *entry;
576
577 if (!io_ctl->cur)
578 return -ENOSPC;
579
580 entry = io_ctl->cur;
581 put_unaligned_le64(offset, &entry->offset);
582 put_unaligned_le64(bytes, &entry->bytes);
583 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
584 BTRFS_FREE_SPACE_EXTENT;
585 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
586 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
587
588 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
589 return 0;
590
591 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
592
593 /* No more pages to map */
594 if (io_ctl->index >= io_ctl->num_pages)
595 return 0;
596
597 /* map the next page */
598 io_ctl_map_page(io_ctl, 1);
599 return 0;
600 }
601
io_ctl_add_bitmap(struct btrfs_io_ctl * io_ctl,void * bitmap)602 static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
603 {
604 if (!io_ctl->cur)
605 return -ENOSPC;
606
607 /*
608 * If we aren't at the start of the current page, unmap this one and
609 * map the next one if there is any left.
610 */
611 if (io_ctl->cur != io_ctl->orig) {
612 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
613 if (io_ctl->index >= io_ctl->num_pages)
614 return -ENOSPC;
615 io_ctl_map_page(io_ctl, 0);
616 }
617
618 copy_page(io_ctl->cur, bitmap);
619 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
620 if (io_ctl->index < io_ctl->num_pages)
621 io_ctl_map_page(io_ctl, 0);
622 return 0;
623 }
624
io_ctl_zero_remaining_pages(struct btrfs_io_ctl * io_ctl)625 static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
626 {
627 /*
628 * If we're not on the boundary we know we've modified the page and we
629 * need to crc the page.
630 */
631 if (io_ctl->cur != io_ctl->orig)
632 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
633 else
634 io_ctl_unmap_page(io_ctl);
635
636 while (io_ctl->index < io_ctl->num_pages) {
637 io_ctl_map_page(io_ctl, 1);
638 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
639 }
640 }
641
io_ctl_read_entry(struct btrfs_io_ctl * io_ctl,struct btrfs_free_space * entry,u8 * type)642 static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
643 struct btrfs_free_space *entry, u8 *type)
644 {
645 struct btrfs_free_space_entry *e;
646 int ret;
647
648 if (!io_ctl->cur) {
649 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
650 if (ret)
651 return ret;
652 }
653
654 e = io_ctl->cur;
655 entry->offset = get_unaligned_le64(&e->offset);
656 entry->bytes = get_unaligned_le64(&e->bytes);
657 *type = e->type;
658 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
659 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
660
661 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
662 return 0;
663
664 io_ctl_unmap_page(io_ctl);
665
666 return 0;
667 }
668
io_ctl_read_bitmap(struct btrfs_io_ctl * io_ctl,struct btrfs_free_space * entry)669 static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
670 struct btrfs_free_space *entry)
671 {
672 int ret;
673
674 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
675 if (ret)
676 return ret;
677
678 copy_page(entry->bitmap, io_ctl->cur);
679 io_ctl_unmap_page(io_ctl);
680
681 return 0;
682 }
683
recalculate_thresholds(struct btrfs_free_space_ctl * ctl)684 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
685 {
686 struct btrfs_block_group *block_group = ctl->block_group;
687 u64 max_bytes;
688 u64 bitmap_bytes;
689 u64 extent_bytes;
690 u64 size = block_group->length;
691 u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
692 u64 max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
693
694 max_bitmaps = max_t(u64, max_bitmaps, 1);
695
696 ASSERT(ctl->total_bitmaps <= max_bitmaps);
697
698 /*
699 * We are trying to keep the total amount of memory used per 1GiB of
700 * space to be MAX_CACHE_BYTES_PER_GIG. However, with a reclamation
701 * mechanism of pulling extents >= FORCE_EXTENT_THRESHOLD out of
702 * bitmaps, we may end up using more memory than this.
703 */
704 if (size < SZ_1G)
705 max_bytes = MAX_CACHE_BYTES_PER_GIG;
706 else
707 max_bytes = MAX_CACHE_BYTES_PER_GIG * div_u64(size, SZ_1G);
708
709 bitmap_bytes = ctl->total_bitmaps * ctl->unit;
710
711 /*
712 * we want the extent entry threshold to always be at most 1/2 the max
713 * bytes we can have, or whatever is less than that.
714 */
715 extent_bytes = max_bytes - bitmap_bytes;
716 extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
717
718 ctl->extents_thresh =
719 div_u64(extent_bytes, sizeof(struct btrfs_free_space));
720 }
721
__load_free_space_cache(struct btrfs_root * root,struct inode * inode,struct btrfs_free_space_ctl * ctl,struct btrfs_path * path,u64 offset)722 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
723 struct btrfs_free_space_ctl *ctl,
724 struct btrfs_path *path, u64 offset)
725 {
726 struct btrfs_fs_info *fs_info = root->fs_info;
727 struct btrfs_free_space_header *header;
728 struct extent_buffer *leaf;
729 struct btrfs_io_ctl io_ctl;
730 struct btrfs_key key;
731 struct btrfs_free_space *e, *n;
732 LIST_HEAD(bitmaps);
733 u64 num_entries;
734 u64 num_bitmaps;
735 u64 generation;
736 u8 type;
737 int ret = 0;
738
739 /* Nothing in the space cache, goodbye */
740 if (!i_size_read(inode))
741 return 0;
742
743 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
744 key.offset = offset;
745 key.type = 0;
746
747 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
748 if (ret < 0)
749 return 0;
750 else if (ret > 0) {
751 btrfs_release_path(path);
752 return 0;
753 }
754
755 ret = -1;
756
757 leaf = path->nodes[0];
758 header = btrfs_item_ptr(leaf, path->slots[0],
759 struct btrfs_free_space_header);
760 num_entries = btrfs_free_space_entries(leaf, header);
761 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
762 generation = btrfs_free_space_generation(leaf, header);
763 btrfs_release_path(path);
764
765 if (!BTRFS_I(inode)->generation) {
766 btrfs_info(fs_info,
767 "the free space cache file (%llu) is invalid, skip it",
768 offset);
769 return 0;
770 }
771
772 if (BTRFS_I(inode)->generation != generation) {
773 btrfs_err(fs_info,
774 "free space inode generation (%llu) did not match free space cache generation (%llu)",
775 BTRFS_I(inode)->generation, generation);
776 return 0;
777 }
778
779 if (!num_entries)
780 return 0;
781
782 ret = io_ctl_init(&io_ctl, inode, 0);
783 if (ret)
784 return ret;
785
786 readahead_cache(inode);
787
788 ret = io_ctl_prepare_pages(&io_ctl, true);
789 if (ret)
790 goto out;
791
792 ret = io_ctl_check_crc(&io_ctl, 0);
793 if (ret)
794 goto free_cache;
795
796 ret = io_ctl_check_generation(&io_ctl, generation);
797 if (ret)
798 goto free_cache;
799
800 while (num_entries) {
801 e = kmem_cache_zalloc(btrfs_free_space_cachep,
802 GFP_NOFS);
803 if (!e) {
804 ret = -ENOMEM;
805 goto free_cache;
806 }
807
808 ret = io_ctl_read_entry(&io_ctl, e, &type);
809 if (ret) {
810 kmem_cache_free(btrfs_free_space_cachep, e);
811 goto free_cache;
812 }
813
814 if (!e->bytes) {
815 ret = -1;
816 kmem_cache_free(btrfs_free_space_cachep, e);
817 goto free_cache;
818 }
819
820 if (type == BTRFS_FREE_SPACE_EXTENT) {
821 spin_lock(&ctl->tree_lock);
822 ret = link_free_space(ctl, e);
823 spin_unlock(&ctl->tree_lock);
824 if (ret) {
825 btrfs_err(fs_info,
826 "Duplicate entries in free space cache, dumping");
827 kmem_cache_free(btrfs_free_space_cachep, e);
828 goto free_cache;
829 }
830 } else {
831 ASSERT(num_bitmaps);
832 num_bitmaps--;
833 e->bitmap = kmem_cache_zalloc(
834 btrfs_free_space_bitmap_cachep, GFP_NOFS);
835 if (!e->bitmap) {
836 ret = -ENOMEM;
837 kmem_cache_free(
838 btrfs_free_space_cachep, e);
839 goto free_cache;
840 }
841 spin_lock(&ctl->tree_lock);
842 ret = link_free_space(ctl, e);
843 ctl->total_bitmaps++;
844 recalculate_thresholds(ctl);
845 spin_unlock(&ctl->tree_lock);
846 if (ret) {
847 btrfs_err(fs_info,
848 "Duplicate entries in free space cache, dumping");
849 kmem_cache_free(btrfs_free_space_cachep, e);
850 goto free_cache;
851 }
852 list_add_tail(&e->list, &bitmaps);
853 }
854
855 num_entries--;
856 }
857
858 io_ctl_unmap_page(&io_ctl);
859
860 /*
861 * We add the bitmaps at the end of the entries in order that
862 * the bitmap entries are added to the cache.
863 */
864 list_for_each_entry_safe(e, n, &bitmaps, list) {
865 list_del_init(&e->list);
866 ret = io_ctl_read_bitmap(&io_ctl, e);
867 if (ret)
868 goto free_cache;
869 }
870
871 io_ctl_drop_pages(&io_ctl);
872 ret = 1;
873 out:
874 io_ctl_free(&io_ctl);
875 return ret;
876 free_cache:
877 io_ctl_drop_pages(&io_ctl);
878 __btrfs_remove_free_space_cache(ctl);
879 goto out;
880 }
881
copy_free_space_cache(struct btrfs_block_group * block_group,struct btrfs_free_space_ctl * ctl)882 static int copy_free_space_cache(struct btrfs_block_group *block_group,
883 struct btrfs_free_space_ctl *ctl)
884 {
885 struct btrfs_free_space *info;
886 struct rb_node *n;
887 int ret = 0;
888
889 while (!ret && (n = rb_first(&ctl->free_space_offset)) != NULL) {
890 info = rb_entry(n, struct btrfs_free_space, offset_index);
891 if (!info->bitmap) {
892 unlink_free_space(ctl, info, true);
893 ret = btrfs_add_free_space(block_group, info->offset,
894 info->bytes);
895 kmem_cache_free(btrfs_free_space_cachep, info);
896 } else {
897 u64 offset = info->offset;
898 u64 bytes = ctl->unit;
899
900 while (search_bitmap(ctl, info, &offset, &bytes,
901 false) == 0) {
902 ret = btrfs_add_free_space(block_group, offset,
903 bytes);
904 if (ret)
905 break;
906 bitmap_clear_bits(ctl, info, offset, bytes, true);
907 offset = info->offset;
908 bytes = ctl->unit;
909 }
910 free_bitmap(ctl, info);
911 }
912 cond_resched();
913 }
914 return ret;
915 }
916
load_free_space_cache(struct btrfs_block_group * block_group)917 int load_free_space_cache(struct btrfs_block_group *block_group)
918 {
919 struct btrfs_fs_info *fs_info = block_group->fs_info;
920 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
921 struct btrfs_free_space_ctl tmp_ctl = {};
922 struct inode *inode;
923 struct btrfs_path *path;
924 int ret = 0;
925 bool matched;
926 u64 used = block_group->used;
927
928 /*
929 * Because we could potentially discard our loaded free space, we want
930 * to load everything into a temporary structure first, and then if it's
931 * valid copy it all into the actual free space ctl.
932 */
933 btrfs_init_free_space_ctl(block_group, &tmp_ctl);
934
935 /*
936 * If this block group has been marked to be cleared for one reason or
937 * another then we can't trust the on disk cache, so just return.
938 */
939 spin_lock(&block_group->lock);
940 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
941 spin_unlock(&block_group->lock);
942 return 0;
943 }
944 spin_unlock(&block_group->lock);
945
946 path = btrfs_alloc_path();
947 if (!path)
948 return 0;
949 path->search_commit_root = 1;
950 path->skip_locking = 1;
951
952 /*
953 * We must pass a path with search_commit_root set to btrfs_iget in
954 * order to avoid a deadlock when allocating extents for the tree root.
955 *
956 * When we are COWing an extent buffer from the tree root, when looking
957 * for a free extent, at extent-tree.c:find_free_extent(), we can find
958 * block group without its free space cache loaded. When we find one
959 * we must load its space cache which requires reading its free space
960 * cache's inode item from the root tree. If this inode item is located
961 * in the same leaf that we started COWing before, then we end up in
962 * deadlock on the extent buffer (trying to read lock it when we
963 * previously write locked it).
964 *
965 * It's safe to read the inode item using the commit root because
966 * block groups, once loaded, stay in memory forever (until they are
967 * removed) as well as their space caches once loaded. New block groups
968 * once created get their ->cached field set to BTRFS_CACHE_FINISHED so
969 * we will never try to read their inode item while the fs is mounted.
970 */
971 inode = lookup_free_space_inode(block_group, path);
972 if (IS_ERR(inode)) {
973 btrfs_free_path(path);
974 return 0;
975 }
976
977 /* We may have converted the inode and made the cache invalid. */
978 spin_lock(&block_group->lock);
979 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
980 spin_unlock(&block_group->lock);
981 btrfs_free_path(path);
982 goto out;
983 }
984 spin_unlock(&block_group->lock);
985
986 ret = __load_free_space_cache(fs_info->tree_root, inode, &tmp_ctl,
987 path, block_group->start);
988 btrfs_free_path(path);
989 if (ret <= 0)
990 goto out;
991
992 matched = (tmp_ctl.free_space == (block_group->length - used -
993 block_group->bytes_super));
994
995 if (matched) {
996 ret = copy_free_space_cache(block_group, &tmp_ctl);
997 /*
998 * ret == 1 means we successfully loaded the free space cache,
999 * so we need to re-set it here.
1000 */
1001 if (ret == 0)
1002 ret = 1;
1003 } else {
1004 __btrfs_remove_free_space_cache(&tmp_ctl);
1005 btrfs_warn(fs_info,
1006 "block group %llu has wrong amount of free space",
1007 block_group->start);
1008 ret = -1;
1009 }
1010 out:
1011 if (ret < 0) {
1012 /* This cache is bogus, make sure it gets cleared */
1013 spin_lock(&block_group->lock);
1014 block_group->disk_cache_state = BTRFS_DC_CLEAR;
1015 spin_unlock(&block_group->lock);
1016 ret = 0;
1017
1018 btrfs_warn(fs_info,
1019 "failed to load free space cache for block group %llu, rebuilding it now",
1020 block_group->start);
1021 }
1022
1023 spin_lock(&ctl->tree_lock);
1024 btrfs_discard_update_discardable(block_group);
1025 spin_unlock(&ctl->tree_lock);
1026 iput(inode);
1027 return ret;
1028 }
1029
1030 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)1031 int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
1032 struct btrfs_free_space_ctl *ctl,
1033 struct btrfs_block_group *block_group,
1034 int *entries, int *bitmaps,
1035 struct list_head *bitmap_list)
1036 {
1037 int ret;
1038 struct btrfs_free_cluster *cluster = NULL;
1039 struct btrfs_free_cluster *cluster_locked = NULL;
1040 struct rb_node *node = rb_first(&ctl->free_space_offset);
1041 struct btrfs_trim_range *trim_entry;
1042
1043 /* Get the cluster for this block_group if it exists */
1044 if (block_group && !list_empty(&block_group->cluster_list)) {
1045 cluster = list_entry(block_group->cluster_list.next,
1046 struct btrfs_free_cluster,
1047 block_group_list);
1048 }
1049
1050 if (!node && cluster) {
1051 cluster_locked = cluster;
1052 spin_lock(&cluster_locked->lock);
1053 node = rb_first(&cluster->root);
1054 cluster = NULL;
1055 }
1056
1057 /* Write out the extent entries */
1058 while (node) {
1059 struct btrfs_free_space *e;
1060
1061 e = rb_entry(node, struct btrfs_free_space, offset_index);
1062 *entries += 1;
1063
1064 ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
1065 e->bitmap);
1066 if (ret)
1067 goto fail;
1068
1069 if (e->bitmap) {
1070 list_add_tail(&e->list, bitmap_list);
1071 *bitmaps += 1;
1072 }
1073 node = rb_next(node);
1074 if (!node && cluster) {
1075 node = rb_first(&cluster->root);
1076 cluster_locked = cluster;
1077 spin_lock(&cluster_locked->lock);
1078 cluster = NULL;
1079 }
1080 }
1081 if (cluster_locked) {
1082 spin_unlock(&cluster_locked->lock);
1083 cluster_locked = NULL;
1084 }
1085
1086 /*
1087 * Make sure we don't miss any range that was removed from our rbtree
1088 * because trimming is running. Otherwise after a umount+mount (or crash
1089 * after committing the transaction) we would leak free space and get
1090 * an inconsistent free space cache report from fsck.
1091 */
1092 list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
1093 ret = io_ctl_add_entry(io_ctl, trim_entry->start,
1094 trim_entry->bytes, NULL);
1095 if (ret)
1096 goto fail;
1097 *entries += 1;
1098 }
1099
1100 return 0;
1101 fail:
1102 if (cluster_locked)
1103 spin_unlock(&cluster_locked->lock);
1104 return -ENOSPC;
1105 }
1106
1107 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)1108 update_cache_item(struct btrfs_trans_handle *trans,
1109 struct btrfs_root *root,
1110 struct inode *inode,
1111 struct btrfs_path *path, u64 offset,
1112 int entries, int bitmaps)
1113 {
1114 struct btrfs_key key;
1115 struct btrfs_free_space_header *header;
1116 struct extent_buffer *leaf;
1117 int ret;
1118
1119 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1120 key.offset = offset;
1121 key.type = 0;
1122
1123 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1124 if (ret < 0) {
1125 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1126 EXTENT_DELALLOC, 0, 0, NULL);
1127 goto fail;
1128 }
1129 leaf = path->nodes[0];
1130 if (ret > 0) {
1131 struct btrfs_key found_key;
1132 ASSERT(path->slots[0]);
1133 path->slots[0]--;
1134 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1135 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1136 found_key.offset != offset) {
1137 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1138 inode->i_size - 1, EXTENT_DELALLOC, 0,
1139 0, NULL);
1140 btrfs_release_path(path);
1141 goto fail;
1142 }
1143 }
1144
1145 BTRFS_I(inode)->generation = trans->transid;
1146 header = btrfs_item_ptr(leaf, path->slots[0],
1147 struct btrfs_free_space_header);
1148 btrfs_set_free_space_entries(leaf, header, entries);
1149 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1150 btrfs_set_free_space_generation(leaf, header, trans->transid);
1151 btrfs_mark_buffer_dirty(leaf);
1152 btrfs_release_path(path);
1153
1154 return 0;
1155
1156 fail:
1157 return -1;
1158 }
1159
write_pinned_extent_entries(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_io_ctl * io_ctl,int * entries)1160 static noinline_for_stack int write_pinned_extent_entries(
1161 struct btrfs_trans_handle *trans,
1162 struct btrfs_block_group *block_group,
1163 struct btrfs_io_ctl *io_ctl,
1164 int *entries)
1165 {
1166 u64 start, extent_start, extent_end, len;
1167 struct extent_io_tree *unpin = NULL;
1168 int ret;
1169
1170 if (!block_group)
1171 return 0;
1172
1173 /*
1174 * We want to add any pinned extents to our free space cache
1175 * so we don't leak the space
1176 *
1177 * We shouldn't have switched the pinned extents yet so this is the
1178 * right one
1179 */
1180 unpin = &trans->transaction->pinned_extents;
1181
1182 start = block_group->start;
1183
1184 while (start < block_group->start + block_group->length) {
1185 ret = find_first_extent_bit(unpin, start,
1186 &extent_start, &extent_end,
1187 EXTENT_DIRTY, NULL);
1188 if (ret)
1189 return 0;
1190
1191 /* This pinned extent is out of our range */
1192 if (extent_start >= block_group->start + block_group->length)
1193 return 0;
1194
1195 extent_start = max(extent_start, start);
1196 extent_end = min(block_group->start + block_group->length,
1197 extent_end + 1);
1198 len = extent_end - extent_start;
1199
1200 *entries += 1;
1201 ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1202 if (ret)
1203 return -ENOSPC;
1204
1205 start = extent_end;
1206 }
1207
1208 return 0;
1209 }
1210
1211 static noinline_for_stack int
write_bitmap_entries(struct btrfs_io_ctl * io_ctl,struct list_head * bitmap_list)1212 write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1213 {
1214 struct btrfs_free_space *entry, *next;
1215 int ret;
1216
1217 /* Write out the bitmaps */
1218 list_for_each_entry_safe(entry, next, bitmap_list, list) {
1219 ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1220 if (ret)
1221 return -ENOSPC;
1222 list_del_init(&entry->list);
1223 }
1224
1225 return 0;
1226 }
1227
flush_dirty_cache(struct inode * inode)1228 static int flush_dirty_cache(struct inode *inode)
1229 {
1230 int ret;
1231
1232 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1233 if (ret)
1234 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1235 EXTENT_DELALLOC, 0, 0, NULL);
1236
1237 return ret;
1238 }
1239
1240 static void noinline_for_stack
cleanup_bitmap_list(struct list_head * bitmap_list)1241 cleanup_bitmap_list(struct list_head *bitmap_list)
1242 {
1243 struct btrfs_free_space *entry, *next;
1244
1245 list_for_each_entry_safe(entry, next, bitmap_list, list)
1246 list_del_init(&entry->list);
1247 }
1248
1249 static void noinline_for_stack
cleanup_write_cache_enospc(struct inode * inode,struct btrfs_io_ctl * io_ctl,struct extent_state ** cached_state)1250 cleanup_write_cache_enospc(struct inode *inode,
1251 struct btrfs_io_ctl *io_ctl,
1252 struct extent_state **cached_state)
1253 {
1254 io_ctl_drop_pages(io_ctl);
1255 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1256 i_size_read(inode) - 1, cached_state);
1257 }
1258
__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)1259 static int __btrfs_wait_cache_io(struct btrfs_root *root,
1260 struct btrfs_trans_handle *trans,
1261 struct btrfs_block_group *block_group,
1262 struct btrfs_io_ctl *io_ctl,
1263 struct btrfs_path *path, u64 offset)
1264 {
1265 int ret;
1266 struct inode *inode = io_ctl->inode;
1267
1268 if (!inode)
1269 return 0;
1270
1271 /* Flush the dirty pages in the cache file. */
1272 ret = flush_dirty_cache(inode);
1273 if (ret)
1274 goto out;
1275
1276 /* Update the cache item to tell everyone this cache file is valid. */
1277 ret = update_cache_item(trans, root, inode, path, offset,
1278 io_ctl->entries, io_ctl->bitmaps);
1279 out:
1280 if (ret) {
1281 invalidate_inode_pages2(inode->i_mapping);
1282 BTRFS_I(inode)->generation = 0;
1283 if (block_group)
1284 btrfs_debug(root->fs_info,
1285 "failed to write free space cache for block group %llu error %d",
1286 block_group->start, ret);
1287 }
1288 btrfs_update_inode(trans, root, BTRFS_I(inode));
1289
1290 if (block_group) {
1291 /* the dirty list is protected by the dirty_bgs_lock */
1292 spin_lock(&trans->transaction->dirty_bgs_lock);
1293
1294 /* the disk_cache_state is protected by the block group lock */
1295 spin_lock(&block_group->lock);
1296
1297 /*
1298 * only mark this as written if we didn't get put back on
1299 * the dirty list while waiting for IO. Otherwise our
1300 * cache state won't be right, and we won't get written again
1301 */
1302 if (!ret && list_empty(&block_group->dirty_list))
1303 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1304 else if (ret)
1305 block_group->disk_cache_state = BTRFS_DC_ERROR;
1306
1307 spin_unlock(&block_group->lock);
1308 spin_unlock(&trans->transaction->dirty_bgs_lock);
1309 io_ctl->inode = NULL;
1310 iput(inode);
1311 }
1312
1313 return ret;
1314
1315 }
1316
btrfs_wait_cache_io(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path)1317 int btrfs_wait_cache_io(struct btrfs_trans_handle *trans,
1318 struct btrfs_block_group *block_group,
1319 struct btrfs_path *path)
1320 {
1321 return __btrfs_wait_cache_io(block_group->fs_info->tree_root, trans,
1322 block_group, &block_group->io_ctl,
1323 path, block_group->start);
1324 }
1325
1326 /**
1327 * Write out cached info to an inode
1328 *
1329 * @root: root the inode belongs to
1330 * @inode: freespace inode we are writing out
1331 * @ctl: free space cache we are going to write out
1332 * @block_group: block_group for this cache if it belongs to a block_group
1333 * @io_ctl: holds context for the io
1334 * @trans: the trans handle
1335 *
1336 * This function writes out a free space cache struct to disk for quick recovery
1337 * on mount. This will return 0 if it was successful in writing the cache out,
1338 * or an errno if it was not.
1339 */
__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)1340 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
1341 struct btrfs_free_space_ctl *ctl,
1342 struct btrfs_block_group *block_group,
1343 struct btrfs_io_ctl *io_ctl,
1344 struct btrfs_trans_handle *trans)
1345 {
1346 struct extent_state *cached_state = NULL;
1347 LIST_HEAD(bitmap_list);
1348 int entries = 0;
1349 int bitmaps = 0;
1350 int ret;
1351 int must_iput = 0;
1352
1353 if (!i_size_read(inode))
1354 return -EIO;
1355
1356 WARN_ON(io_ctl->pages);
1357 ret = io_ctl_init(io_ctl, inode, 1);
1358 if (ret)
1359 return ret;
1360
1361 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1362 down_write(&block_group->data_rwsem);
1363 spin_lock(&block_group->lock);
1364 if (block_group->delalloc_bytes) {
1365 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1366 spin_unlock(&block_group->lock);
1367 up_write(&block_group->data_rwsem);
1368 BTRFS_I(inode)->generation = 0;
1369 ret = 0;
1370 must_iput = 1;
1371 goto out;
1372 }
1373 spin_unlock(&block_group->lock);
1374 }
1375
1376 /* Lock all pages first so we can lock the extent safely. */
1377 ret = io_ctl_prepare_pages(io_ctl, false);
1378 if (ret)
1379 goto out_unlock;
1380
1381 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1382 &cached_state);
1383
1384 io_ctl_set_generation(io_ctl, trans->transid);
1385
1386 mutex_lock(&ctl->cache_writeout_mutex);
1387 /* Write out the extent entries in the free space cache */
1388 spin_lock(&ctl->tree_lock);
1389 ret = write_cache_extent_entries(io_ctl, ctl,
1390 block_group, &entries, &bitmaps,
1391 &bitmap_list);
1392 if (ret)
1393 goto out_nospc_locked;
1394
1395 /*
1396 * Some spaces that are freed in the current transaction are pinned,
1397 * they will be added into free space cache after the transaction is
1398 * committed, we shouldn't lose them.
1399 *
1400 * If this changes while we are working we'll get added back to
1401 * the dirty list and redo it. No locking needed
1402 */
1403 ret = write_pinned_extent_entries(trans, block_group, io_ctl, &entries);
1404 if (ret)
1405 goto out_nospc_locked;
1406
1407 /*
1408 * At last, we write out all the bitmaps and keep cache_writeout_mutex
1409 * locked while doing it because a concurrent trim can be manipulating
1410 * or freeing the bitmap.
1411 */
1412 ret = write_bitmap_entries(io_ctl, &bitmap_list);
1413 spin_unlock(&ctl->tree_lock);
1414 mutex_unlock(&ctl->cache_writeout_mutex);
1415 if (ret)
1416 goto out_nospc;
1417
1418 /* Zero out the rest of the pages just to make sure */
1419 io_ctl_zero_remaining_pages(io_ctl);
1420
1421 /* Everything is written out, now we dirty the pages in the file. */
1422 ret = btrfs_dirty_pages(BTRFS_I(inode), io_ctl->pages,
1423 io_ctl->num_pages, 0, i_size_read(inode),
1424 &cached_state, false);
1425 if (ret)
1426 goto out_nospc;
1427
1428 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1429 up_write(&block_group->data_rwsem);
1430 /*
1431 * Release the pages and unlock the extent, we will flush
1432 * them out later
1433 */
1434 io_ctl_drop_pages(io_ctl);
1435 io_ctl_free(io_ctl);
1436
1437 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1438 i_size_read(inode) - 1, &cached_state);
1439
1440 /*
1441 * at this point the pages are under IO and we're happy,
1442 * The caller is responsible for waiting on them and updating
1443 * the cache and the inode
1444 */
1445 io_ctl->entries = entries;
1446 io_ctl->bitmaps = bitmaps;
1447
1448 ret = btrfs_fdatawrite_range(inode, 0, (u64)-1);
1449 if (ret)
1450 goto out;
1451
1452 return 0;
1453
1454 out_nospc_locked:
1455 cleanup_bitmap_list(&bitmap_list);
1456 spin_unlock(&ctl->tree_lock);
1457 mutex_unlock(&ctl->cache_writeout_mutex);
1458
1459 out_nospc:
1460 cleanup_write_cache_enospc(inode, io_ctl, &cached_state);
1461
1462 out_unlock:
1463 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1464 up_write(&block_group->data_rwsem);
1465
1466 out:
1467 io_ctl->inode = NULL;
1468 io_ctl_free(io_ctl);
1469 if (ret) {
1470 invalidate_inode_pages2(inode->i_mapping);
1471 BTRFS_I(inode)->generation = 0;
1472 }
1473 btrfs_update_inode(trans, root, BTRFS_I(inode));
1474 if (must_iput)
1475 iput(inode);
1476 return ret;
1477 }
1478
btrfs_write_out_cache(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path)1479 int btrfs_write_out_cache(struct btrfs_trans_handle *trans,
1480 struct btrfs_block_group *block_group,
1481 struct btrfs_path *path)
1482 {
1483 struct btrfs_fs_info *fs_info = trans->fs_info;
1484 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1485 struct inode *inode;
1486 int ret = 0;
1487
1488 spin_lock(&block_group->lock);
1489 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1490 spin_unlock(&block_group->lock);
1491 return 0;
1492 }
1493 spin_unlock(&block_group->lock);
1494
1495 inode = lookup_free_space_inode(block_group, path);
1496 if (IS_ERR(inode))
1497 return 0;
1498
1499 ret = __btrfs_write_out_cache(fs_info->tree_root, inode, ctl,
1500 block_group, &block_group->io_ctl, trans);
1501 if (ret) {
1502 btrfs_debug(fs_info,
1503 "failed to write free space cache for block group %llu error %d",
1504 block_group->start, ret);
1505 spin_lock(&block_group->lock);
1506 block_group->disk_cache_state = BTRFS_DC_ERROR;
1507 spin_unlock(&block_group->lock);
1508
1509 block_group->io_ctl.inode = NULL;
1510 iput(inode);
1511 }
1512
1513 /*
1514 * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1515 * to wait for IO and put the inode
1516 */
1517
1518 return ret;
1519 }
1520
offset_to_bit(u64 bitmap_start,u32 unit,u64 offset)1521 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1522 u64 offset)
1523 {
1524 ASSERT(offset >= bitmap_start);
1525 offset -= bitmap_start;
1526 return (unsigned long)(div_u64(offset, unit));
1527 }
1528
bytes_to_bits(u64 bytes,u32 unit)1529 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1530 {
1531 return (unsigned long)(div_u64(bytes, unit));
1532 }
1533
offset_to_bitmap(struct btrfs_free_space_ctl * ctl,u64 offset)1534 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1535 u64 offset)
1536 {
1537 u64 bitmap_start;
1538 u64 bytes_per_bitmap;
1539
1540 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1541 bitmap_start = offset - ctl->start;
1542 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1543 bitmap_start *= bytes_per_bitmap;
1544 bitmap_start += ctl->start;
1545
1546 return bitmap_start;
1547 }
1548
tree_insert_offset(struct rb_root * root,u64 offset,struct rb_node * node,int bitmap)1549 static int tree_insert_offset(struct rb_root *root, u64 offset,
1550 struct rb_node *node, int bitmap)
1551 {
1552 struct rb_node **p = &root->rb_node;
1553 struct rb_node *parent = NULL;
1554 struct btrfs_free_space *info;
1555
1556 while (*p) {
1557 parent = *p;
1558 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1559
1560 if (offset < info->offset) {
1561 p = &(*p)->rb_left;
1562 } else if (offset > info->offset) {
1563 p = &(*p)->rb_right;
1564 } else {
1565 /*
1566 * we could have a bitmap entry and an extent entry
1567 * share the same offset. If this is the case, we want
1568 * the extent entry to always be found first if we do a
1569 * linear search through the tree, since we want to have
1570 * the quickest allocation time, and allocating from an
1571 * extent is faster than allocating from a bitmap. So
1572 * if we're inserting a bitmap and we find an entry at
1573 * this offset, we want to go right, or after this entry
1574 * logically. If we are inserting an extent and we've
1575 * found a bitmap, we want to go left, or before
1576 * logically.
1577 */
1578 if (bitmap) {
1579 if (info->bitmap) {
1580 WARN_ON_ONCE(1);
1581 return -EEXIST;
1582 }
1583 p = &(*p)->rb_right;
1584 } else {
1585 if (!info->bitmap) {
1586 WARN_ON_ONCE(1);
1587 return -EEXIST;
1588 }
1589 p = &(*p)->rb_left;
1590 }
1591 }
1592 }
1593
1594 rb_link_node(node, parent, p);
1595 rb_insert_color(node, root);
1596
1597 return 0;
1598 }
1599
1600 /*
1601 * This is a little subtle. We *only* have ->max_extent_size set if we actually
1602 * searched through the bitmap and figured out the largest ->max_extent_size,
1603 * otherwise it's 0. In the case that it's 0 we don't want to tell the
1604 * allocator the wrong thing, we want to use the actual real max_extent_size
1605 * we've found already if it's larger, or we want to use ->bytes.
1606 *
1607 * This matters because find_free_space() will skip entries who's ->bytes is
1608 * less than the required bytes. So if we didn't search down this bitmap, we
1609 * may pick some previous entry that has a smaller ->max_extent_size than we
1610 * have. For example, assume we have two entries, one that has
1611 * ->max_extent_size set to 4K and ->bytes set to 1M. A second entry hasn't set
1612 * ->max_extent_size yet, has ->bytes set to 8K and it's contiguous. We will
1613 * call into find_free_space(), and return with max_extent_size == 4K, because
1614 * that first bitmap entry had ->max_extent_size set, but the second one did
1615 * not. If instead we returned 8K we'd come in searching for 8K, and find the
1616 * 8K contiguous range.
1617 *
1618 * Consider the other case, we have 2 8K chunks in that second entry and still
1619 * don't have ->max_extent_size set. We'll return 16K, and the next time the
1620 * allocator comes in it'll fully search our second bitmap, and this time it'll
1621 * get an uptodate value of 8K as the maximum chunk size. Then we'll get the
1622 * right allocation the next loop through.
1623 */
get_max_extent_size(const struct btrfs_free_space * entry)1624 static inline u64 get_max_extent_size(const struct btrfs_free_space *entry)
1625 {
1626 if (entry->bitmap && entry->max_extent_size)
1627 return entry->max_extent_size;
1628 return entry->bytes;
1629 }
1630
1631 /*
1632 * We want the largest entry to be leftmost, so this is inverted from what you'd
1633 * normally expect.
1634 */
entry_less(struct rb_node * node,const struct rb_node * parent)1635 static bool entry_less(struct rb_node *node, const struct rb_node *parent)
1636 {
1637 const struct btrfs_free_space *entry, *exist;
1638
1639 entry = rb_entry(node, struct btrfs_free_space, bytes_index);
1640 exist = rb_entry(parent, struct btrfs_free_space, bytes_index);
1641 return get_max_extent_size(exist) < get_max_extent_size(entry);
1642 }
1643
1644 /*
1645 * searches the tree for the given offset.
1646 *
1647 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1648 * want a section that has at least bytes size and comes at or after the given
1649 * offset.
1650 */
1651 static struct btrfs_free_space *
tree_search_offset(struct btrfs_free_space_ctl * ctl,u64 offset,int bitmap_only,int fuzzy)1652 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1653 u64 offset, int bitmap_only, int fuzzy)
1654 {
1655 struct rb_node *n = ctl->free_space_offset.rb_node;
1656 struct btrfs_free_space *entry = NULL, *prev = NULL;
1657
1658 /* find entry that is closest to the 'offset' */
1659 while (n) {
1660 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1661 prev = entry;
1662
1663 if (offset < entry->offset)
1664 n = n->rb_left;
1665 else if (offset > entry->offset)
1666 n = n->rb_right;
1667 else
1668 break;
1669
1670 entry = NULL;
1671 }
1672
1673 if (bitmap_only) {
1674 if (!entry)
1675 return NULL;
1676 if (entry->bitmap)
1677 return entry;
1678
1679 /*
1680 * bitmap entry and extent entry may share same offset,
1681 * in that case, bitmap entry comes after extent entry.
1682 */
1683 n = rb_next(n);
1684 if (!n)
1685 return NULL;
1686 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1687 if (entry->offset != offset)
1688 return NULL;
1689
1690 WARN_ON(!entry->bitmap);
1691 return entry;
1692 } else if (entry) {
1693 if (entry->bitmap) {
1694 /*
1695 * if previous extent entry covers the offset,
1696 * we should return it instead of the bitmap entry
1697 */
1698 n = rb_prev(&entry->offset_index);
1699 if (n) {
1700 prev = rb_entry(n, struct btrfs_free_space,
1701 offset_index);
1702 if (!prev->bitmap &&
1703 prev->offset + prev->bytes > offset)
1704 entry = prev;
1705 }
1706 }
1707 return entry;
1708 }
1709
1710 if (!prev)
1711 return NULL;
1712
1713 /* find last entry before the 'offset' */
1714 entry = prev;
1715 if (entry->offset > offset) {
1716 n = rb_prev(&entry->offset_index);
1717 if (n) {
1718 entry = rb_entry(n, struct btrfs_free_space,
1719 offset_index);
1720 ASSERT(entry->offset <= offset);
1721 } else {
1722 if (fuzzy)
1723 return entry;
1724 else
1725 return NULL;
1726 }
1727 }
1728
1729 if (entry->bitmap) {
1730 n = rb_prev(&entry->offset_index);
1731 if (n) {
1732 prev = rb_entry(n, struct btrfs_free_space,
1733 offset_index);
1734 if (!prev->bitmap &&
1735 prev->offset + prev->bytes > offset)
1736 return prev;
1737 }
1738 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1739 return entry;
1740 } else if (entry->offset + entry->bytes > offset)
1741 return entry;
1742
1743 if (!fuzzy)
1744 return NULL;
1745
1746 while (1) {
1747 n = rb_next(&entry->offset_index);
1748 if (!n)
1749 return NULL;
1750 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1751 if (entry->bitmap) {
1752 if (entry->offset + BITS_PER_BITMAP *
1753 ctl->unit > offset)
1754 break;
1755 } else {
1756 if (entry->offset + entry->bytes > offset)
1757 break;
1758 }
1759 }
1760 return entry;
1761 }
1762
unlink_free_space(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)1763 static inline void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1764 struct btrfs_free_space *info,
1765 bool update_stat)
1766 {
1767 rb_erase(&info->offset_index, &ctl->free_space_offset);
1768 rb_erase_cached(&info->bytes_index, &ctl->free_space_bytes);
1769 ctl->free_extents--;
1770
1771 if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1772 ctl->discardable_extents[BTRFS_STAT_CURR]--;
1773 ctl->discardable_bytes[BTRFS_STAT_CURR] -= info->bytes;
1774 }
1775
1776 if (update_stat)
1777 ctl->free_space -= info->bytes;
1778 }
1779
link_free_space(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)1780 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1781 struct btrfs_free_space *info)
1782 {
1783 int ret = 0;
1784
1785 ASSERT(info->bytes || info->bitmap);
1786 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1787 &info->offset_index, (info->bitmap != NULL));
1788 if (ret)
1789 return ret;
1790
1791 rb_add_cached(&info->bytes_index, &ctl->free_space_bytes, entry_less);
1792
1793 if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1794 ctl->discardable_extents[BTRFS_STAT_CURR]++;
1795 ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
1796 }
1797
1798 ctl->free_space += info->bytes;
1799 ctl->free_extents++;
1800 return ret;
1801 }
1802
relink_bitmap_entry(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)1803 static void relink_bitmap_entry(struct btrfs_free_space_ctl *ctl,
1804 struct btrfs_free_space *info)
1805 {
1806 ASSERT(info->bitmap);
1807
1808 /*
1809 * If our entry is empty it's because we're on a cluster and we don't
1810 * want to re-link it into our ctl bytes index.
1811 */
1812 if (RB_EMPTY_NODE(&info->bytes_index))
1813 return;
1814
1815 rb_erase_cached(&info->bytes_index, &ctl->free_space_bytes);
1816 rb_add_cached(&info->bytes_index, &ctl->free_space_bytes, entry_less);
1817 }
1818
bitmap_clear_bits(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset,u64 bytes,bool update_stat)1819 static inline void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1820 struct btrfs_free_space *info,
1821 u64 offset, u64 bytes, bool update_stat)
1822 {
1823 unsigned long start, count, end;
1824 int extent_delta = -1;
1825
1826 start = offset_to_bit(info->offset, ctl->unit, offset);
1827 count = bytes_to_bits(bytes, ctl->unit);
1828 end = start + count;
1829 ASSERT(end <= BITS_PER_BITMAP);
1830
1831 bitmap_clear(info->bitmap, start, count);
1832
1833 info->bytes -= bytes;
1834 if (info->max_extent_size > ctl->unit)
1835 info->max_extent_size = 0;
1836
1837 relink_bitmap_entry(ctl, info);
1838
1839 if (start && test_bit(start - 1, info->bitmap))
1840 extent_delta++;
1841
1842 if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1843 extent_delta++;
1844
1845 info->bitmap_extents += extent_delta;
1846 if (!btrfs_free_space_trimmed(info)) {
1847 ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1848 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
1849 }
1850
1851 if (update_stat)
1852 ctl->free_space -= bytes;
1853 }
1854
bitmap_set_bits(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset,u64 bytes)1855 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1856 struct btrfs_free_space *info, u64 offset,
1857 u64 bytes)
1858 {
1859 unsigned long start, count, end;
1860 int extent_delta = 1;
1861
1862 start = offset_to_bit(info->offset, ctl->unit, offset);
1863 count = bytes_to_bits(bytes, ctl->unit);
1864 end = start + count;
1865 ASSERT(end <= BITS_PER_BITMAP);
1866
1867 bitmap_set(info->bitmap, start, count);
1868
1869 /*
1870 * We set some bytes, we have no idea what the max extent size is
1871 * anymore.
1872 */
1873 info->max_extent_size = 0;
1874 info->bytes += bytes;
1875 ctl->free_space += bytes;
1876
1877 relink_bitmap_entry(ctl, info);
1878
1879 if (start && test_bit(start - 1, info->bitmap))
1880 extent_delta--;
1881
1882 if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1883 extent_delta--;
1884
1885 info->bitmap_extents += extent_delta;
1886 if (!btrfs_free_space_trimmed(info)) {
1887 ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1888 ctl->discardable_bytes[BTRFS_STAT_CURR] += bytes;
1889 }
1890 }
1891
1892 /*
1893 * If we can not find suitable extent, we will use bytes to record
1894 * the size of the max extent.
1895 */
search_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * bitmap_info,u64 * offset,u64 * bytes,bool for_alloc)1896 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1897 struct btrfs_free_space *bitmap_info, u64 *offset,
1898 u64 *bytes, bool for_alloc)
1899 {
1900 unsigned long found_bits = 0;
1901 unsigned long max_bits = 0;
1902 unsigned long bits, i;
1903 unsigned long next_zero;
1904 unsigned long extent_bits;
1905
1906 /*
1907 * Skip searching the bitmap if we don't have a contiguous section that
1908 * is large enough for this allocation.
1909 */
1910 if (for_alloc &&
1911 bitmap_info->max_extent_size &&
1912 bitmap_info->max_extent_size < *bytes) {
1913 *bytes = bitmap_info->max_extent_size;
1914 return -1;
1915 }
1916
1917 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1918 max_t(u64, *offset, bitmap_info->offset));
1919 bits = bytes_to_bits(*bytes, ctl->unit);
1920
1921 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1922 if (for_alloc && bits == 1) {
1923 found_bits = 1;
1924 break;
1925 }
1926 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1927 BITS_PER_BITMAP, i);
1928 extent_bits = next_zero - i;
1929 if (extent_bits >= bits) {
1930 found_bits = extent_bits;
1931 break;
1932 } else if (extent_bits > max_bits) {
1933 max_bits = extent_bits;
1934 }
1935 i = next_zero;
1936 }
1937
1938 if (found_bits) {
1939 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1940 *bytes = (u64)(found_bits) * ctl->unit;
1941 return 0;
1942 }
1943
1944 *bytes = (u64)(max_bits) * ctl->unit;
1945 bitmap_info->max_extent_size = *bytes;
1946 relink_bitmap_entry(ctl, bitmap_info);
1947 return -1;
1948 }
1949
1950 /* Cache the size of the max extent in bytes */
1951 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,bool use_bytes_index)1952 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1953 unsigned long align, u64 *max_extent_size, bool use_bytes_index)
1954 {
1955 struct btrfs_free_space *entry;
1956 struct rb_node *node;
1957 u64 tmp;
1958 u64 align_off;
1959 int ret;
1960
1961 if (!ctl->free_space_offset.rb_node)
1962 goto out;
1963 again:
1964 if (use_bytes_index) {
1965 node = rb_first_cached(&ctl->free_space_bytes);
1966 } else {
1967 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset),
1968 0, 1);
1969 if (!entry)
1970 goto out;
1971 node = &entry->offset_index;
1972 }
1973
1974 for (; node; node = rb_next(node)) {
1975 if (use_bytes_index)
1976 entry = rb_entry(node, struct btrfs_free_space,
1977 bytes_index);
1978 else
1979 entry = rb_entry(node, struct btrfs_free_space,
1980 offset_index);
1981
1982 /*
1983 * If we are using the bytes index then all subsequent entries
1984 * in this tree are going to be < bytes, so simply set the max
1985 * extent size and exit the loop.
1986 *
1987 * If we're using the offset index then we need to keep going
1988 * through the rest of the tree.
1989 */
1990 if (entry->bytes < *bytes) {
1991 *max_extent_size = max(get_max_extent_size(entry),
1992 *max_extent_size);
1993 if (use_bytes_index)
1994 break;
1995 continue;
1996 }
1997
1998 /* make sure the space returned is big enough
1999 * to match our requested alignment
2000 */
2001 if (*bytes >= align) {
2002 tmp = entry->offset - ctl->start + align - 1;
2003 tmp = div64_u64(tmp, align);
2004 tmp = tmp * align + ctl->start;
2005 align_off = tmp - entry->offset;
2006 } else {
2007 align_off = 0;
2008 tmp = entry->offset;
2009 }
2010
2011 /*
2012 * We don't break here if we're using the bytes index because we
2013 * may have another entry that has the correct alignment that is
2014 * the right size, so we don't want to miss that possibility.
2015 * At worst this adds another loop through the logic, but if we
2016 * broke here we could prematurely ENOSPC.
2017 */
2018 if (entry->bytes < *bytes + align_off) {
2019 *max_extent_size = max(get_max_extent_size(entry),
2020 *max_extent_size);
2021 continue;
2022 }
2023
2024 if (entry->bitmap) {
2025 struct rb_node *old_next = rb_next(node);
2026 u64 size = *bytes;
2027
2028 ret = search_bitmap(ctl, entry, &tmp, &size, true);
2029 if (!ret) {
2030 *offset = tmp;
2031 *bytes = size;
2032 return entry;
2033 } else {
2034 *max_extent_size =
2035 max(get_max_extent_size(entry),
2036 *max_extent_size);
2037 }
2038
2039 /*
2040 * The bitmap may have gotten re-arranged in the space
2041 * index here because the max_extent_size may have been
2042 * updated. Start from the beginning again if this
2043 * happened.
2044 */
2045 if (use_bytes_index && old_next != rb_next(node))
2046 goto again;
2047 continue;
2048 }
2049
2050 *offset = tmp;
2051 *bytes = entry->bytes - align_off;
2052 return entry;
2053 }
2054 out:
2055 return NULL;
2056 }
2057
add_new_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset)2058 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
2059 struct btrfs_free_space *info, u64 offset)
2060 {
2061 info->offset = offset_to_bitmap(ctl, offset);
2062 info->bytes = 0;
2063 info->bitmap_extents = 0;
2064 INIT_LIST_HEAD(&info->list);
2065 link_free_space(ctl, info);
2066 ctl->total_bitmaps++;
2067 recalculate_thresholds(ctl);
2068 }
2069
free_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * bitmap_info)2070 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
2071 struct btrfs_free_space *bitmap_info)
2072 {
2073 /*
2074 * Normally when this is called, the bitmap is completely empty. However,
2075 * if we are blowing up the free space cache for one reason or another
2076 * via __btrfs_remove_free_space_cache(), then it may not be freed and
2077 * we may leave stats on the table.
2078 */
2079 if (bitmap_info->bytes && !btrfs_free_space_trimmed(bitmap_info)) {
2080 ctl->discardable_extents[BTRFS_STAT_CURR] -=
2081 bitmap_info->bitmap_extents;
2082 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bitmap_info->bytes;
2083
2084 }
2085 unlink_free_space(ctl, bitmap_info, true);
2086 kmem_cache_free(btrfs_free_space_bitmap_cachep, bitmap_info->bitmap);
2087 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
2088 ctl->total_bitmaps--;
2089 recalculate_thresholds(ctl);
2090 }
2091
remove_from_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * bitmap_info,u64 * offset,u64 * bytes)2092 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
2093 struct btrfs_free_space *bitmap_info,
2094 u64 *offset, u64 *bytes)
2095 {
2096 u64 end;
2097 u64 search_start, search_bytes;
2098 int ret;
2099
2100 again:
2101 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
2102
2103 /*
2104 * We need to search for bits in this bitmap. We could only cover some
2105 * of the extent in this bitmap thanks to how we add space, so we need
2106 * to search for as much as it as we can and clear that amount, and then
2107 * go searching for the next bit.
2108 */
2109 search_start = *offset;
2110 search_bytes = ctl->unit;
2111 search_bytes = min(search_bytes, end - search_start + 1);
2112 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes,
2113 false);
2114 if (ret < 0 || search_start != *offset)
2115 return -EINVAL;
2116
2117 /* We may have found more bits than what we need */
2118 search_bytes = min(search_bytes, *bytes);
2119
2120 /* Cannot clear past the end of the bitmap */
2121 search_bytes = min(search_bytes, end - search_start + 1);
2122
2123 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes, true);
2124 *offset += search_bytes;
2125 *bytes -= search_bytes;
2126
2127 if (*bytes) {
2128 struct rb_node *next = rb_next(&bitmap_info->offset_index);
2129 if (!bitmap_info->bytes)
2130 free_bitmap(ctl, bitmap_info);
2131
2132 /*
2133 * no entry after this bitmap, but we still have bytes to
2134 * remove, so something has gone wrong.
2135 */
2136 if (!next)
2137 return -EINVAL;
2138
2139 bitmap_info = rb_entry(next, struct btrfs_free_space,
2140 offset_index);
2141
2142 /*
2143 * if the next entry isn't a bitmap we need to return to let the
2144 * extent stuff do its work.
2145 */
2146 if (!bitmap_info->bitmap)
2147 return -EAGAIN;
2148
2149 /*
2150 * Ok the next item is a bitmap, but it may not actually hold
2151 * the information for the rest of this free space stuff, so
2152 * look for it, and if we don't find it return so we can try
2153 * everything over again.
2154 */
2155 search_start = *offset;
2156 search_bytes = ctl->unit;
2157 ret = search_bitmap(ctl, bitmap_info, &search_start,
2158 &search_bytes, false);
2159 if (ret < 0 || search_start != *offset)
2160 return -EAGAIN;
2161
2162 goto again;
2163 } else if (!bitmap_info->bytes)
2164 free_bitmap(ctl, bitmap_info);
2165
2166 return 0;
2167 }
2168
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)2169 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
2170 struct btrfs_free_space *info, u64 offset,
2171 u64 bytes, enum btrfs_trim_state trim_state)
2172 {
2173 u64 bytes_to_set = 0;
2174 u64 end;
2175
2176 /*
2177 * This is a tradeoff to make bitmap trim state minimal. We mark the
2178 * whole bitmap untrimmed if at any point we add untrimmed regions.
2179 */
2180 if (trim_state == BTRFS_TRIM_STATE_UNTRIMMED) {
2181 if (btrfs_free_space_trimmed(info)) {
2182 ctl->discardable_extents[BTRFS_STAT_CURR] +=
2183 info->bitmap_extents;
2184 ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
2185 }
2186 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2187 }
2188
2189 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
2190
2191 bytes_to_set = min(end - offset, bytes);
2192
2193 bitmap_set_bits(ctl, info, offset, bytes_to_set);
2194
2195 return bytes_to_set;
2196
2197 }
2198
use_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)2199 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
2200 struct btrfs_free_space *info)
2201 {
2202 struct btrfs_block_group *block_group = ctl->block_group;
2203 struct btrfs_fs_info *fs_info = block_group->fs_info;
2204 bool forced = false;
2205
2206 #ifdef CONFIG_BTRFS_DEBUG
2207 if (btrfs_should_fragment_free_space(block_group))
2208 forced = true;
2209 #endif
2210
2211 /* This is a way to reclaim large regions from the bitmaps. */
2212 if (!forced && info->bytes >= FORCE_EXTENT_THRESHOLD)
2213 return false;
2214
2215 /*
2216 * If we are below the extents threshold then we can add this as an
2217 * extent, and don't have to deal with the bitmap
2218 */
2219 if (!forced && ctl->free_extents < ctl->extents_thresh) {
2220 /*
2221 * If this block group has some small extents we don't want to
2222 * use up all of our free slots in the cache with them, we want
2223 * to reserve them to larger extents, however if we have plenty
2224 * of cache left then go ahead an dadd them, no sense in adding
2225 * the overhead of a bitmap if we don't have to.
2226 */
2227 if (info->bytes <= fs_info->sectorsize * 8) {
2228 if (ctl->free_extents * 3 <= ctl->extents_thresh)
2229 return false;
2230 } else {
2231 return false;
2232 }
2233 }
2234
2235 /*
2236 * The original block groups from mkfs can be really small, like 8
2237 * megabytes, so don't bother with a bitmap for those entries. However
2238 * some block groups can be smaller than what a bitmap would cover but
2239 * are still large enough that they could overflow the 32k memory limit,
2240 * so allow those block groups to still be allowed to have a bitmap
2241 * entry.
2242 */
2243 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->length)
2244 return false;
2245
2246 return true;
2247 }
2248
2249 static const struct btrfs_free_space_op free_space_op = {
2250 .use_bitmap = use_bitmap,
2251 };
2252
insert_into_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)2253 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
2254 struct btrfs_free_space *info)
2255 {
2256 struct btrfs_free_space *bitmap_info;
2257 struct btrfs_block_group *block_group = NULL;
2258 int added = 0;
2259 u64 bytes, offset, bytes_added;
2260 enum btrfs_trim_state trim_state;
2261 int ret;
2262
2263 bytes = info->bytes;
2264 offset = info->offset;
2265 trim_state = info->trim_state;
2266
2267 if (!ctl->op->use_bitmap(ctl, info))
2268 return 0;
2269
2270 if (ctl->op == &free_space_op)
2271 block_group = ctl->block_group;
2272 again:
2273 /*
2274 * Since we link bitmaps right into the cluster we need to see if we
2275 * have a cluster here, and if so and it has our bitmap we need to add
2276 * the free space to that bitmap.
2277 */
2278 if (block_group && !list_empty(&block_group->cluster_list)) {
2279 struct btrfs_free_cluster *cluster;
2280 struct rb_node *node;
2281 struct btrfs_free_space *entry;
2282
2283 cluster = list_entry(block_group->cluster_list.next,
2284 struct btrfs_free_cluster,
2285 block_group_list);
2286 spin_lock(&cluster->lock);
2287 node = rb_first(&cluster->root);
2288 if (!node) {
2289 spin_unlock(&cluster->lock);
2290 goto no_cluster_bitmap;
2291 }
2292
2293 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2294 if (!entry->bitmap) {
2295 spin_unlock(&cluster->lock);
2296 goto no_cluster_bitmap;
2297 }
2298
2299 if (entry->offset == offset_to_bitmap(ctl, offset)) {
2300 bytes_added = add_bytes_to_bitmap(ctl, entry, offset,
2301 bytes, trim_state);
2302 bytes -= bytes_added;
2303 offset += bytes_added;
2304 }
2305 spin_unlock(&cluster->lock);
2306 if (!bytes) {
2307 ret = 1;
2308 goto out;
2309 }
2310 }
2311
2312 no_cluster_bitmap:
2313 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2314 1, 0);
2315 if (!bitmap_info) {
2316 ASSERT(added == 0);
2317 goto new_bitmap;
2318 }
2319
2320 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
2321 trim_state);
2322 bytes -= bytes_added;
2323 offset += bytes_added;
2324 added = 0;
2325
2326 if (!bytes) {
2327 ret = 1;
2328 goto out;
2329 } else
2330 goto again;
2331
2332 new_bitmap:
2333 if (info && info->bitmap) {
2334 add_new_bitmap(ctl, info, offset);
2335 added = 1;
2336 info = NULL;
2337 goto again;
2338 } else {
2339 spin_unlock(&ctl->tree_lock);
2340
2341 /* no pre-allocated info, allocate a new one */
2342 if (!info) {
2343 info = kmem_cache_zalloc(btrfs_free_space_cachep,
2344 GFP_NOFS);
2345 if (!info) {
2346 spin_lock(&ctl->tree_lock);
2347 ret = -ENOMEM;
2348 goto out;
2349 }
2350 }
2351
2352 /* allocate the bitmap */
2353 info->bitmap = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep,
2354 GFP_NOFS);
2355 info->trim_state = BTRFS_TRIM_STATE_TRIMMED;
2356 spin_lock(&ctl->tree_lock);
2357 if (!info->bitmap) {
2358 ret = -ENOMEM;
2359 goto out;
2360 }
2361 goto again;
2362 }
2363
2364 out:
2365 if (info) {
2366 if (info->bitmap)
2367 kmem_cache_free(btrfs_free_space_bitmap_cachep,
2368 info->bitmap);
2369 kmem_cache_free(btrfs_free_space_cachep, info);
2370 }
2371
2372 return ret;
2373 }
2374
2375 /*
2376 * Free space merging rules:
2377 * 1) Merge trimmed areas together
2378 * 2) Let untrimmed areas coalesce with trimmed areas
2379 * 3) Always pull neighboring regions from bitmaps
2380 *
2381 * The above rules are for when we merge free space based on btrfs_trim_state.
2382 * Rules 2 and 3 are subtle because they are suboptimal, but are done for the
2383 * same reason: to promote larger extent regions which makes life easier for
2384 * find_free_extent(). Rule 2 enables coalescing based on the common path
2385 * being returning free space from btrfs_finish_extent_commit(). So when free
2386 * space is trimmed, it will prevent aggregating trimmed new region and
2387 * untrimmed regions in the rb_tree. Rule 3 is purely to obtain larger extents
2388 * and provide find_free_extent() with the largest extents possible hoping for
2389 * the reuse path.
2390 */
try_merge_free_space(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2391 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2392 struct btrfs_free_space *info, bool update_stat)
2393 {
2394 struct btrfs_free_space *left_info = NULL;
2395 struct btrfs_free_space *right_info;
2396 bool merged = false;
2397 u64 offset = info->offset;
2398 u64 bytes = info->bytes;
2399 const bool is_trimmed = btrfs_free_space_trimmed(info);
2400
2401 /*
2402 * first we want to see if there is free space adjacent to the range we
2403 * are adding, if there is remove that struct and add a new one to
2404 * cover the entire range
2405 */
2406 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
2407 if (right_info && rb_prev(&right_info->offset_index))
2408 left_info = rb_entry(rb_prev(&right_info->offset_index),
2409 struct btrfs_free_space, offset_index);
2410 else if (!right_info)
2411 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
2412
2413 /* See try_merge_free_space() comment. */
2414 if (right_info && !right_info->bitmap &&
2415 (!is_trimmed || btrfs_free_space_trimmed(right_info))) {
2416 unlink_free_space(ctl, right_info, update_stat);
2417 info->bytes += right_info->bytes;
2418 kmem_cache_free(btrfs_free_space_cachep, right_info);
2419 merged = true;
2420 }
2421
2422 /* See try_merge_free_space() comment. */
2423 if (left_info && !left_info->bitmap &&
2424 left_info->offset + left_info->bytes == offset &&
2425 (!is_trimmed || btrfs_free_space_trimmed(left_info))) {
2426 unlink_free_space(ctl, left_info, update_stat);
2427 info->offset = left_info->offset;
2428 info->bytes += left_info->bytes;
2429 kmem_cache_free(btrfs_free_space_cachep, left_info);
2430 merged = true;
2431 }
2432
2433 return merged;
2434 }
2435
steal_from_bitmap_to_end(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2436 static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2437 struct btrfs_free_space *info,
2438 bool update_stat)
2439 {
2440 struct btrfs_free_space *bitmap;
2441 unsigned long i;
2442 unsigned long j;
2443 const u64 end = info->offset + info->bytes;
2444 const u64 bitmap_offset = offset_to_bitmap(ctl, end);
2445 u64 bytes;
2446
2447 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2448 if (!bitmap)
2449 return false;
2450
2451 i = offset_to_bit(bitmap->offset, ctl->unit, end);
2452 j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
2453 if (j == i)
2454 return false;
2455 bytes = (j - i) * ctl->unit;
2456 info->bytes += bytes;
2457
2458 /* See try_merge_free_space() comment. */
2459 if (!btrfs_free_space_trimmed(bitmap))
2460 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2461
2462 bitmap_clear_bits(ctl, bitmap, end, bytes, update_stat);
2463
2464 if (!bitmap->bytes)
2465 free_bitmap(ctl, bitmap);
2466
2467 return true;
2468 }
2469
steal_from_bitmap_to_front(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2470 static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2471 struct btrfs_free_space *info,
2472 bool update_stat)
2473 {
2474 struct btrfs_free_space *bitmap;
2475 u64 bitmap_offset;
2476 unsigned long i;
2477 unsigned long j;
2478 unsigned long prev_j;
2479 u64 bytes;
2480
2481 bitmap_offset = offset_to_bitmap(ctl, info->offset);
2482 /* If we're on a boundary, try the previous logical bitmap. */
2483 if (bitmap_offset == info->offset) {
2484 if (info->offset == 0)
2485 return false;
2486 bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
2487 }
2488
2489 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2490 if (!bitmap)
2491 return false;
2492
2493 i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
2494 j = 0;
2495 prev_j = (unsigned long)-1;
2496 for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2497 if (j > i)
2498 break;
2499 prev_j = j;
2500 }
2501 if (prev_j == i)
2502 return false;
2503
2504 if (prev_j == (unsigned long)-1)
2505 bytes = (i + 1) * ctl->unit;
2506 else
2507 bytes = (i - prev_j) * ctl->unit;
2508
2509 info->offset -= bytes;
2510 info->bytes += bytes;
2511
2512 /* See try_merge_free_space() comment. */
2513 if (!btrfs_free_space_trimmed(bitmap))
2514 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2515
2516 bitmap_clear_bits(ctl, bitmap, info->offset, bytes, update_stat);
2517
2518 if (!bitmap->bytes)
2519 free_bitmap(ctl, bitmap);
2520
2521 return true;
2522 }
2523
2524 /*
2525 * We prefer always to allocate from extent entries, both for clustered and
2526 * non-clustered allocation requests. So when attempting to add a new extent
2527 * entry, try to see if there's adjacent free space in bitmap entries, and if
2528 * there is, migrate that space from the bitmaps to the extent.
2529 * Like this we get better chances of satisfying space allocation requests
2530 * because we attempt to satisfy them based on a single cache entry, and never
2531 * on 2 or more entries - even if the entries represent a contiguous free space
2532 * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2533 * ends).
2534 */
steal_from_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2535 static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2536 struct btrfs_free_space *info,
2537 bool update_stat)
2538 {
2539 /*
2540 * Only work with disconnected entries, as we can change their offset,
2541 * and must be extent entries.
2542 */
2543 ASSERT(!info->bitmap);
2544 ASSERT(RB_EMPTY_NODE(&info->offset_index));
2545
2546 if (ctl->total_bitmaps > 0) {
2547 bool stole_end;
2548 bool stole_front = false;
2549
2550 stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2551 if (ctl->total_bitmaps > 0)
2552 stole_front = steal_from_bitmap_to_front(ctl, info,
2553 update_stat);
2554
2555 if (stole_end || stole_front)
2556 try_merge_free_space(ctl, info, update_stat);
2557 }
2558 }
2559
__btrfs_add_free_space(struct btrfs_block_group * block_group,u64 offset,u64 bytes,enum btrfs_trim_state trim_state)2560 int __btrfs_add_free_space(struct btrfs_block_group *block_group,
2561 u64 offset, u64 bytes,
2562 enum btrfs_trim_state trim_state)
2563 {
2564 struct btrfs_fs_info *fs_info = block_group->fs_info;
2565 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2566 struct btrfs_free_space *info;
2567 int ret = 0;
2568 u64 filter_bytes = bytes;
2569
2570 ASSERT(!btrfs_is_zoned(fs_info));
2571
2572 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2573 if (!info)
2574 return -ENOMEM;
2575
2576 info->offset = offset;
2577 info->bytes = bytes;
2578 info->trim_state = trim_state;
2579 RB_CLEAR_NODE(&info->offset_index);
2580 RB_CLEAR_NODE(&info->bytes_index);
2581
2582 spin_lock(&ctl->tree_lock);
2583
2584 if (try_merge_free_space(ctl, info, true))
2585 goto link;
2586
2587 /*
2588 * There was no extent directly to the left or right of this new
2589 * extent then we know we're going to have to allocate a new extent, so
2590 * before we do that see if we need to drop this into a bitmap
2591 */
2592 ret = insert_into_bitmap(ctl, info);
2593 if (ret < 0) {
2594 goto out;
2595 } else if (ret) {
2596 ret = 0;
2597 goto out;
2598 }
2599 link:
2600 /*
2601 * Only steal free space from adjacent bitmaps if we're sure we're not
2602 * going to add the new free space to existing bitmap entries - because
2603 * that would mean unnecessary work that would be reverted. Therefore
2604 * attempt to steal space from bitmaps if we're adding an extent entry.
2605 */
2606 steal_from_bitmap(ctl, info, true);
2607
2608 filter_bytes = max(filter_bytes, info->bytes);
2609
2610 ret = link_free_space(ctl, info);
2611 if (ret)
2612 kmem_cache_free(btrfs_free_space_cachep, info);
2613 out:
2614 btrfs_discard_update_discardable(block_group);
2615 spin_unlock(&ctl->tree_lock);
2616
2617 if (ret) {
2618 btrfs_crit(fs_info, "unable to add free space :%d", ret);
2619 ASSERT(ret != -EEXIST);
2620 }
2621
2622 if (trim_state != BTRFS_TRIM_STATE_TRIMMED) {
2623 btrfs_discard_check_filter(block_group, filter_bytes);
2624 btrfs_discard_queue_work(&fs_info->discard_ctl, block_group);
2625 }
2626
2627 return ret;
2628 }
2629
__btrfs_add_free_space_zoned(struct btrfs_block_group * block_group,u64 bytenr,u64 size,bool used)2630 static int __btrfs_add_free_space_zoned(struct btrfs_block_group *block_group,
2631 u64 bytenr, u64 size, bool used)
2632 {
2633 struct btrfs_space_info *sinfo = block_group->space_info;
2634 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2635 u64 offset = bytenr - block_group->start;
2636 u64 to_free, to_unusable;
2637 int bg_reclaim_threshold = 0;
2638 bool initial = (size == block_group->length);
2639 u64 reclaimable_unusable;
2640
2641 WARN_ON(!initial && offset + size > block_group->zone_capacity);
2642
2643 if (!initial)
2644 bg_reclaim_threshold = READ_ONCE(sinfo->bg_reclaim_threshold);
2645
2646 spin_lock(&ctl->tree_lock);
2647 if (!used)
2648 to_free = size;
2649 else if (initial)
2650 to_free = block_group->zone_capacity;
2651 else if (offset >= block_group->alloc_offset)
2652 to_free = size;
2653 else if (offset + size <= block_group->alloc_offset)
2654 to_free = 0;
2655 else
2656 to_free = offset + size - block_group->alloc_offset;
2657 to_unusable = size - to_free;
2658
2659 ctl->free_space += to_free;
2660 /*
2661 * If the block group is read-only, we should account freed space into
2662 * bytes_readonly.
2663 */
2664 if (!block_group->ro)
2665 block_group->zone_unusable += to_unusable;
2666 spin_unlock(&ctl->tree_lock);
2667 if (!used) {
2668 spin_lock(&block_group->lock);
2669 block_group->alloc_offset -= size;
2670 spin_unlock(&block_group->lock);
2671 }
2672
2673 reclaimable_unusable = block_group->zone_unusable -
2674 (block_group->length - block_group->zone_capacity);
2675 /* All the region is now unusable. Mark it as unused and reclaim */
2676 if (block_group->zone_unusable == block_group->length) {
2677 btrfs_mark_bg_unused(block_group);
2678 } else if (bg_reclaim_threshold &&
2679 reclaimable_unusable >=
2680 div_factor_fine(block_group->zone_capacity,
2681 bg_reclaim_threshold)) {
2682 btrfs_mark_bg_to_reclaim(block_group);
2683 }
2684
2685 return 0;
2686 }
2687
btrfs_add_free_space(struct btrfs_block_group * block_group,u64 bytenr,u64 size)2688 int btrfs_add_free_space(struct btrfs_block_group *block_group,
2689 u64 bytenr, u64 size)
2690 {
2691 enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2692
2693 if (btrfs_is_zoned(block_group->fs_info))
2694 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2695 true);
2696
2697 if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC))
2698 trim_state = BTRFS_TRIM_STATE_TRIMMED;
2699
2700 return __btrfs_add_free_space(block_group, bytenr, size, trim_state);
2701 }
2702
btrfs_add_free_space_unused(struct btrfs_block_group * block_group,u64 bytenr,u64 size)2703 int btrfs_add_free_space_unused(struct btrfs_block_group *block_group,
2704 u64 bytenr, u64 size)
2705 {
2706 if (btrfs_is_zoned(block_group->fs_info))
2707 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2708 false);
2709
2710 return btrfs_add_free_space(block_group, bytenr, size);
2711 }
2712
2713 /*
2714 * This is a subtle distinction because when adding free space back in general,
2715 * we want it to be added as untrimmed for async. But in the case where we add
2716 * it on loading of a block group, we want to consider it trimmed.
2717 */
btrfs_add_free_space_async_trimmed(struct btrfs_block_group * block_group,u64 bytenr,u64 size)2718 int btrfs_add_free_space_async_trimmed(struct btrfs_block_group *block_group,
2719 u64 bytenr, u64 size)
2720 {
2721 enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2722
2723 if (btrfs_is_zoned(block_group->fs_info))
2724 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2725 true);
2726
2727 if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC) ||
2728 btrfs_test_opt(block_group->fs_info, DISCARD_ASYNC))
2729 trim_state = BTRFS_TRIM_STATE_TRIMMED;
2730
2731 return __btrfs_add_free_space(block_group, bytenr, size, trim_state);
2732 }
2733
btrfs_remove_free_space(struct btrfs_block_group * block_group,u64 offset,u64 bytes)2734 int btrfs_remove_free_space(struct btrfs_block_group *block_group,
2735 u64 offset, u64 bytes)
2736 {
2737 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2738 struct btrfs_free_space *info;
2739 int ret;
2740 bool re_search = false;
2741
2742 if (btrfs_is_zoned(block_group->fs_info)) {
2743 /*
2744 * This can happen with conventional zones when replaying log.
2745 * Since the allocation info of tree-log nodes are not recorded
2746 * to the extent-tree, calculate_alloc_pointer() failed to
2747 * advance the allocation pointer after last allocated tree log
2748 * node blocks.
2749 *
2750 * This function is called from
2751 * btrfs_pin_extent_for_log_replay() when replaying the log.
2752 * Advance the pointer not to overwrite the tree-log nodes.
2753 */
2754 if (block_group->start + block_group->alloc_offset <
2755 offset + bytes) {
2756 block_group->alloc_offset =
2757 offset + bytes - block_group->start;
2758 }
2759 return 0;
2760 }
2761
2762 spin_lock(&ctl->tree_lock);
2763
2764 again:
2765 ret = 0;
2766 if (!bytes)
2767 goto out_lock;
2768
2769 info = tree_search_offset(ctl, offset, 0, 0);
2770 if (!info) {
2771 /*
2772 * oops didn't find an extent that matched the space we wanted
2773 * to remove, look for a bitmap instead
2774 */
2775 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2776 1, 0);
2777 if (!info) {
2778 /*
2779 * If we found a partial bit of our free space in a
2780 * bitmap but then couldn't find the other part this may
2781 * be a problem, so WARN about it.
2782 */
2783 WARN_ON(re_search);
2784 goto out_lock;
2785 }
2786 }
2787
2788 re_search = false;
2789 if (!info->bitmap) {
2790 unlink_free_space(ctl, info, true);
2791 if (offset == info->offset) {
2792 u64 to_free = min(bytes, info->bytes);
2793
2794 info->bytes -= to_free;
2795 info->offset += to_free;
2796 if (info->bytes) {
2797 ret = link_free_space(ctl, info);
2798 WARN_ON(ret);
2799 } else {
2800 kmem_cache_free(btrfs_free_space_cachep, info);
2801 }
2802
2803 offset += to_free;
2804 bytes -= to_free;
2805 goto again;
2806 } else {
2807 u64 old_end = info->bytes + info->offset;
2808
2809 info->bytes = offset - info->offset;
2810 ret = link_free_space(ctl, info);
2811 WARN_ON(ret);
2812 if (ret)
2813 goto out_lock;
2814
2815 /* Not enough bytes in this entry to satisfy us */
2816 if (old_end < offset + bytes) {
2817 bytes -= old_end - offset;
2818 offset = old_end;
2819 goto again;
2820 } else if (old_end == offset + bytes) {
2821 /* all done */
2822 goto out_lock;
2823 }
2824 spin_unlock(&ctl->tree_lock);
2825
2826 ret = __btrfs_add_free_space(block_group,
2827 offset + bytes,
2828 old_end - (offset + bytes),
2829 info->trim_state);
2830 WARN_ON(ret);
2831 goto out;
2832 }
2833 }
2834
2835 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2836 if (ret == -EAGAIN) {
2837 re_search = true;
2838 goto again;
2839 }
2840 out_lock:
2841 btrfs_discard_update_discardable(block_group);
2842 spin_unlock(&ctl->tree_lock);
2843 out:
2844 return ret;
2845 }
2846
btrfs_dump_free_space(struct btrfs_block_group * block_group,u64 bytes)2847 void btrfs_dump_free_space(struct btrfs_block_group *block_group,
2848 u64 bytes)
2849 {
2850 struct btrfs_fs_info *fs_info = block_group->fs_info;
2851 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2852 struct btrfs_free_space *info;
2853 struct rb_node *n;
2854 int count = 0;
2855
2856 /*
2857 * Zoned btrfs does not use free space tree and cluster. Just print
2858 * out the free space after the allocation offset.
2859 */
2860 if (btrfs_is_zoned(fs_info)) {
2861 btrfs_info(fs_info, "free space %llu active %d",
2862 block_group->zone_capacity - block_group->alloc_offset,
2863 block_group->zone_is_active);
2864 return;
2865 }
2866
2867 spin_lock(&ctl->tree_lock);
2868 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2869 info = rb_entry(n, struct btrfs_free_space, offset_index);
2870 if (info->bytes >= bytes && !block_group->ro)
2871 count++;
2872 btrfs_crit(fs_info, "entry offset %llu, bytes %llu, bitmap %s",
2873 info->offset, info->bytes,
2874 (info->bitmap) ? "yes" : "no");
2875 }
2876 spin_unlock(&ctl->tree_lock);
2877 btrfs_info(fs_info, "block group has cluster?: %s",
2878 list_empty(&block_group->cluster_list) ? "no" : "yes");
2879 btrfs_info(fs_info,
2880 "%d blocks of free space at or bigger than bytes is", count);
2881 }
2882
btrfs_init_free_space_ctl(struct btrfs_block_group * block_group,struct btrfs_free_space_ctl * ctl)2883 void btrfs_init_free_space_ctl(struct btrfs_block_group *block_group,
2884 struct btrfs_free_space_ctl *ctl)
2885 {
2886 struct btrfs_fs_info *fs_info = block_group->fs_info;
2887
2888 spin_lock_init(&ctl->tree_lock);
2889 ctl->unit = fs_info->sectorsize;
2890 ctl->start = block_group->start;
2891 ctl->block_group = block_group;
2892 ctl->op = &free_space_op;
2893 ctl->free_space_bytes = RB_ROOT_CACHED;
2894 INIT_LIST_HEAD(&ctl->trimming_ranges);
2895 mutex_init(&ctl->cache_writeout_mutex);
2896
2897 /*
2898 * we only want to have 32k of ram per block group for keeping
2899 * track of free space, and if we pass 1/2 of that we want to
2900 * start converting things over to using bitmaps
2901 */
2902 ctl->extents_thresh = (SZ_32K / 2) / sizeof(struct btrfs_free_space);
2903 }
2904
2905 /*
2906 * for a given cluster, put all of its extents back into the free
2907 * space cache. If the block group passed doesn't match the block group
2908 * pointed to by the cluster, someone else raced in and freed the
2909 * cluster already. In that case, we just return without changing anything
2910 */
__btrfs_return_cluster_to_free_space(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster)2911 static void __btrfs_return_cluster_to_free_space(
2912 struct btrfs_block_group *block_group,
2913 struct btrfs_free_cluster *cluster)
2914 {
2915 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2916 struct btrfs_free_space *entry;
2917 struct rb_node *node;
2918
2919 spin_lock(&cluster->lock);
2920 if (cluster->block_group != block_group) {
2921 spin_unlock(&cluster->lock);
2922 return;
2923 }
2924
2925 cluster->block_group = NULL;
2926 cluster->window_start = 0;
2927 list_del_init(&cluster->block_group_list);
2928
2929 node = rb_first(&cluster->root);
2930 while (node) {
2931 bool bitmap;
2932
2933 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2934 node = rb_next(&entry->offset_index);
2935 rb_erase(&entry->offset_index, &cluster->root);
2936 RB_CLEAR_NODE(&entry->offset_index);
2937
2938 bitmap = (entry->bitmap != NULL);
2939 if (!bitmap) {
2940 /* Merging treats extents as if they were new */
2941 if (!btrfs_free_space_trimmed(entry)) {
2942 ctl->discardable_extents[BTRFS_STAT_CURR]--;
2943 ctl->discardable_bytes[BTRFS_STAT_CURR] -=
2944 entry->bytes;
2945 }
2946
2947 try_merge_free_space(ctl, entry, false);
2948 steal_from_bitmap(ctl, entry, false);
2949
2950 /* As we insert directly, update these statistics */
2951 if (!btrfs_free_space_trimmed(entry)) {
2952 ctl->discardable_extents[BTRFS_STAT_CURR]++;
2953 ctl->discardable_bytes[BTRFS_STAT_CURR] +=
2954 entry->bytes;
2955 }
2956 }
2957 tree_insert_offset(&ctl->free_space_offset,
2958 entry->offset, &entry->offset_index, bitmap);
2959 rb_add_cached(&entry->bytes_index, &ctl->free_space_bytes,
2960 entry_less);
2961 }
2962 cluster->root = RB_ROOT;
2963 spin_unlock(&cluster->lock);
2964 btrfs_put_block_group(block_group);
2965 }
2966
__btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl * ctl)2967 static void __btrfs_remove_free_space_cache_locked(
2968 struct btrfs_free_space_ctl *ctl)
2969 {
2970 struct btrfs_free_space *info;
2971 struct rb_node *node;
2972
2973 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2974 info = rb_entry(node, struct btrfs_free_space, offset_index);
2975 if (!info->bitmap) {
2976 unlink_free_space(ctl, info, true);
2977 kmem_cache_free(btrfs_free_space_cachep, info);
2978 } else {
2979 free_bitmap(ctl, info);
2980 }
2981
2982 cond_resched_lock(&ctl->tree_lock);
2983 }
2984 }
2985
__btrfs_remove_free_space_cache(struct btrfs_free_space_ctl * ctl)2986 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2987 {
2988 spin_lock(&ctl->tree_lock);
2989 __btrfs_remove_free_space_cache_locked(ctl);
2990 if (ctl->block_group)
2991 btrfs_discard_update_discardable(ctl->block_group);
2992 spin_unlock(&ctl->tree_lock);
2993 }
2994
btrfs_remove_free_space_cache(struct btrfs_block_group * block_group)2995 void btrfs_remove_free_space_cache(struct btrfs_block_group *block_group)
2996 {
2997 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2998 struct btrfs_free_cluster *cluster;
2999 struct list_head *head;
3000
3001 spin_lock(&ctl->tree_lock);
3002 while ((head = block_group->cluster_list.next) !=
3003 &block_group->cluster_list) {
3004 cluster = list_entry(head, struct btrfs_free_cluster,
3005 block_group_list);
3006
3007 WARN_ON(cluster->block_group != block_group);
3008 __btrfs_return_cluster_to_free_space(block_group, cluster);
3009
3010 cond_resched_lock(&ctl->tree_lock);
3011 }
3012 __btrfs_remove_free_space_cache_locked(ctl);
3013 btrfs_discard_update_discardable(block_group);
3014 spin_unlock(&ctl->tree_lock);
3015
3016 }
3017
3018 /**
3019 * btrfs_is_free_space_trimmed - see if everything is trimmed
3020 * @block_group: block_group of interest
3021 *
3022 * Walk @block_group's free space rb_tree to determine if everything is trimmed.
3023 */
btrfs_is_free_space_trimmed(struct btrfs_block_group * block_group)3024 bool btrfs_is_free_space_trimmed(struct btrfs_block_group *block_group)
3025 {
3026 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3027 struct btrfs_free_space *info;
3028 struct rb_node *node;
3029 bool ret = true;
3030
3031 spin_lock(&ctl->tree_lock);
3032 node = rb_first(&ctl->free_space_offset);
3033
3034 while (node) {
3035 info = rb_entry(node, struct btrfs_free_space, offset_index);
3036
3037 if (!btrfs_free_space_trimmed(info)) {
3038 ret = false;
3039 break;
3040 }
3041
3042 node = rb_next(node);
3043 }
3044
3045 spin_unlock(&ctl->tree_lock);
3046 return ret;
3047 }
3048
btrfs_find_space_for_alloc(struct btrfs_block_group * block_group,u64 offset,u64 bytes,u64 empty_size,u64 * max_extent_size)3049 u64 btrfs_find_space_for_alloc(struct btrfs_block_group *block_group,
3050 u64 offset, u64 bytes, u64 empty_size,
3051 u64 *max_extent_size)
3052 {
3053 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3054 struct btrfs_discard_ctl *discard_ctl =
3055 &block_group->fs_info->discard_ctl;
3056 struct btrfs_free_space *entry = NULL;
3057 u64 bytes_search = bytes + empty_size;
3058 u64 ret = 0;
3059 u64 align_gap = 0;
3060 u64 align_gap_len = 0;
3061 enum btrfs_trim_state align_gap_trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3062 bool use_bytes_index = (offset == block_group->start);
3063
3064 ASSERT(!btrfs_is_zoned(block_group->fs_info));
3065
3066 spin_lock(&ctl->tree_lock);
3067 entry = find_free_space(ctl, &offset, &bytes_search,
3068 block_group->full_stripe_len, max_extent_size,
3069 use_bytes_index);
3070 if (!entry)
3071 goto out;
3072
3073 ret = offset;
3074 if (entry->bitmap) {
3075 bitmap_clear_bits(ctl, entry, offset, bytes, true);
3076
3077 if (!btrfs_free_space_trimmed(entry))
3078 atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3079
3080 if (!entry->bytes)
3081 free_bitmap(ctl, entry);
3082 } else {
3083 unlink_free_space(ctl, entry, true);
3084 align_gap_len = offset - entry->offset;
3085 align_gap = entry->offset;
3086 align_gap_trim_state = entry->trim_state;
3087
3088 if (!btrfs_free_space_trimmed(entry))
3089 atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3090
3091 entry->offset = offset + bytes;
3092 WARN_ON(entry->bytes < bytes + align_gap_len);
3093
3094 entry->bytes -= bytes + align_gap_len;
3095 if (!entry->bytes)
3096 kmem_cache_free(btrfs_free_space_cachep, entry);
3097 else
3098 link_free_space(ctl, entry);
3099 }
3100 out:
3101 btrfs_discard_update_discardable(block_group);
3102 spin_unlock(&ctl->tree_lock);
3103
3104 if (align_gap_len)
3105 __btrfs_add_free_space(block_group, align_gap, align_gap_len,
3106 align_gap_trim_state);
3107 return ret;
3108 }
3109
3110 /*
3111 * given a cluster, put all of its extents back into the free space
3112 * cache. If a block group is passed, this function will only free
3113 * a cluster that belongs to the passed block group.
3114 *
3115 * Otherwise, it'll get a reference on the block group pointed to by the
3116 * cluster and remove the cluster from it.
3117 */
btrfs_return_cluster_to_free_space(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster)3118 void btrfs_return_cluster_to_free_space(
3119 struct btrfs_block_group *block_group,
3120 struct btrfs_free_cluster *cluster)
3121 {
3122 struct btrfs_free_space_ctl *ctl;
3123
3124 /* first, get a safe pointer to the block group */
3125 spin_lock(&cluster->lock);
3126 if (!block_group) {
3127 block_group = cluster->block_group;
3128 if (!block_group) {
3129 spin_unlock(&cluster->lock);
3130 return;
3131 }
3132 } else if (cluster->block_group != block_group) {
3133 /* someone else has already freed it don't redo their work */
3134 spin_unlock(&cluster->lock);
3135 return;
3136 }
3137 btrfs_get_block_group(block_group);
3138 spin_unlock(&cluster->lock);
3139
3140 ctl = block_group->free_space_ctl;
3141
3142 /* now return any extents the cluster had on it */
3143 spin_lock(&ctl->tree_lock);
3144 __btrfs_return_cluster_to_free_space(block_group, cluster);
3145 spin_unlock(&ctl->tree_lock);
3146
3147 btrfs_discard_queue_work(&block_group->fs_info->discard_ctl, block_group);
3148
3149 /* finally drop our ref */
3150 btrfs_put_block_group(block_group);
3151 }
3152
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)3153 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group *block_group,
3154 struct btrfs_free_cluster *cluster,
3155 struct btrfs_free_space *entry,
3156 u64 bytes, u64 min_start,
3157 u64 *max_extent_size)
3158 {
3159 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3160 int err;
3161 u64 search_start = cluster->window_start;
3162 u64 search_bytes = bytes;
3163 u64 ret = 0;
3164
3165 search_start = min_start;
3166 search_bytes = bytes;
3167
3168 err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
3169 if (err) {
3170 *max_extent_size = max(get_max_extent_size(entry),
3171 *max_extent_size);
3172 return 0;
3173 }
3174
3175 ret = search_start;
3176 bitmap_clear_bits(ctl, entry, ret, bytes, false);
3177
3178 return ret;
3179 }
3180
3181 /*
3182 * given a cluster, try to allocate 'bytes' from it, returns 0
3183 * if it couldn't find anything suitably large, or a logical disk offset
3184 * if things worked out
3185 */
btrfs_alloc_from_cluster(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster,u64 bytes,u64 min_start,u64 * max_extent_size)3186 u64 btrfs_alloc_from_cluster(struct btrfs_block_group *block_group,
3187 struct btrfs_free_cluster *cluster, u64 bytes,
3188 u64 min_start, u64 *max_extent_size)
3189 {
3190 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3191 struct btrfs_discard_ctl *discard_ctl =
3192 &block_group->fs_info->discard_ctl;
3193 struct btrfs_free_space *entry = NULL;
3194 struct rb_node *node;
3195 u64 ret = 0;
3196
3197 ASSERT(!btrfs_is_zoned(block_group->fs_info));
3198
3199 spin_lock(&cluster->lock);
3200 if (bytes > cluster->max_size)
3201 goto out;
3202
3203 if (cluster->block_group != block_group)
3204 goto out;
3205
3206 node = rb_first(&cluster->root);
3207 if (!node)
3208 goto out;
3209
3210 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3211 while (1) {
3212 if (entry->bytes < bytes)
3213 *max_extent_size = max(get_max_extent_size(entry),
3214 *max_extent_size);
3215
3216 if (entry->bytes < bytes ||
3217 (!entry->bitmap && entry->offset < min_start)) {
3218 node = rb_next(&entry->offset_index);
3219 if (!node)
3220 break;
3221 entry = rb_entry(node, struct btrfs_free_space,
3222 offset_index);
3223 continue;
3224 }
3225
3226 if (entry->bitmap) {
3227 ret = btrfs_alloc_from_bitmap(block_group,
3228 cluster, entry, bytes,
3229 cluster->window_start,
3230 max_extent_size);
3231 if (ret == 0) {
3232 node = rb_next(&entry->offset_index);
3233 if (!node)
3234 break;
3235 entry = rb_entry(node, struct btrfs_free_space,
3236 offset_index);
3237 continue;
3238 }
3239 cluster->window_start += bytes;
3240 } else {
3241 ret = entry->offset;
3242
3243 entry->offset += bytes;
3244 entry->bytes -= bytes;
3245 }
3246
3247 break;
3248 }
3249 out:
3250 spin_unlock(&cluster->lock);
3251
3252 if (!ret)
3253 return 0;
3254
3255 spin_lock(&ctl->tree_lock);
3256
3257 if (!btrfs_free_space_trimmed(entry))
3258 atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3259
3260 ctl->free_space -= bytes;
3261 if (!entry->bitmap && !btrfs_free_space_trimmed(entry))
3262 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
3263
3264 spin_lock(&cluster->lock);
3265 if (entry->bytes == 0) {
3266 rb_erase(&entry->offset_index, &cluster->root);
3267 ctl->free_extents--;
3268 if (entry->bitmap) {
3269 kmem_cache_free(btrfs_free_space_bitmap_cachep,
3270 entry->bitmap);
3271 ctl->total_bitmaps--;
3272 recalculate_thresholds(ctl);
3273 } else if (!btrfs_free_space_trimmed(entry)) {
3274 ctl->discardable_extents[BTRFS_STAT_CURR]--;
3275 }
3276 kmem_cache_free(btrfs_free_space_cachep, entry);
3277 }
3278
3279 spin_unlock(&cluster->lock);
3280 spin_unlock(&ctl->tree_lock);
3281
3282 return ret;
3283 }
3284
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)3285 static int btrfs_bitmap_cluster(struct btrfs_block_group *block_group,
3286 struct btrfs_free_space *entry,
3287 struct btrfs_free_cluster *cluster,
3288 u64 offset, u64 bytes,
3289 u64 cont1_bytes, u64 min_bytes)
3290 {
3291 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3292 unsigned long next_zero;
3293 unsigned long i;
3294 unsigned long want_bits;
3295 unsigned long min_bits;
3296 unsigned long found_bits;
3297 unsigned long max_bits = 0;
3298 unsigned long start = 0;
3299 unsigned long total_found = 0;
3300 int ret;
3301
3302 i = offset_to_bit(entry->offset, ctl->unit,
3303 max_t(u64, offset, entry->offset));
3304 want_bits = bytes_to_bits(bytes, ctl->unit);
3305 min_bits = bytes_to_bits(min_bytes, ctl->unit);
3306
3307 /*
3308 * Don't bother looking for a cluster in this bitmap if it's heavily
3309 * fragmented.
3310 */
3311 if (entry->max_extent_size &&
3312 entry->max_extent_size < cont1_bytes)
3313 return -ENOSPC;
3314 again:
3315 found_bits = 0;
3316 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
3317 next_zero = find_next_zero_bit(entry->bitmap,
3318 BITS_PER_BITMAP, i);
3319 if (next_zero - i >= min_bits) {
3320 found_bits = next_zero - i;
3321 if (found_bits > max_bits)
3322 max_bits = found_bits;
3323 break;
3324 }
3325 if (next_zero - i > max_bits)
3326 max_bits = next_zero - i;
3327 i = next_zero;
3328 }
3329
3330 if (!found_bits) {
3331 entry->max_extent_size = (u64)max_bits * ctl->unit;
3332 return -ENOSPC;
3333 }
3334
3335 if (!total_found) {
3336 start = i;
3337 cluster->max_size = 0;
3338 }
3339
3340 total_found += found_bits;
3341
3342 if (cluster->max_size < found_bits * ctl->unit)
3343 cluster->max_size = found_bits * ctl->unit;
3344
3345 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
3346 i = next_zero + 1;
3347 goto again;
3348 }
3349
3350 cluster->window_start = start * ctl->unit + entry->offset;
3351 rb_erase(&entry->offset_index, &ctl->free_space_offset);
3352 rb_erase_cached(&entry->bytes_index, &ctl->free_space_bytes);
3353
3354 /*
3355 * We need to know if we're currently on the normal space index when we
3356 * manipulate the bitmap so that we know we need to remove and re-insert
3357 * it into the space_index tree. Clear the bytes_index node here so the
3358 * bitmap manipulation helpers know not to mess with the space_index
3359 * until this bitmap entry is added back into the normal cache.
3360 */
3361 RB_CLEAR_NODE(&entry->bytes_index);
3362
3363 ret = tree_insert_offset(&cluster->root, entry->offset,
3364 &entry->offset_index, 1);
3365 ASSERT(!ret); /* -EEXIST; Logic error */
3366
3367 trace_btrfs_setup_cluster(block_group, cluster,
3368 total_found * ctl->unit, 1);
3369 return 0;
3370 }
3371
3372 /*
3373 * This searches the block group for just extents to fill the cluster with.
3374 * Try to find a cluster with at least bytes total bytes, at least one
3375 * extent of cont1_bytes, and other clusters of at least min_bytes.
3376 */
3377 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)3378 setup_cluster_no_bitmap(struct btrfs_block_group *block_group,
3379 struct btrfs_free_cluster *cluster,
3380 struct list_head *bitmaps, u64 offset, u64 bytes,
3381 u64 cont1_bytes, u64 min_bytes)
3382 {
3383 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3384 struct btrfs_free_space *first = NULL;
3385 struct btrfs_free_space *entry = NULL;
3386 struct btrfs_free_space *last;
3387 struct rb_node *node;
3388 u64 window_free;
3389 u64 max_extent;
3390 u64 total_size = 0;
3391
3392 entry = tree_search_offset(ctl, offset, 0, 1);
3393 if (!entry)
3394 return -ENOSPC;
3395
3396 /*
3397 * We don't want bitmaps, so just move along until we find a normal
3398 * extent entry.
3399 */
3400 while (entry->bitmap || entry->bytes < min_bytes) {
3401 if (entry->bitmap && list_empty(&entry->list))
3402 list_add_tail(&entry->list, bitmaps);
3403 node = rb_next(&entry->offset_index);
3404 if (!node)
3405 return -ENOSPC;
3406 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3407 }
3408
3409 window_free = entry->bytes;
3410 max_extent = entry->bytes;
3411 first = entry;
3412 last = entry;
3413
3414 for (node = rb_next(&entry->offset_index); node;
3415 node = rb_next(&entry->offset_index)) {
3416 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3417
3418 if (entry->bitmap) {
3419 if (list_empty(&entry->list))
3420 list_add_tail(&entry->list, bitmaps);
3421 continue;
3422 }
3423
3424 if (entry->bytes < min_bytes)
3425 continue;
3426
3427 last = entry;
3428 window_free += entry->bytes;
3429 if (entry->bytes > max_extent)
3430 max_extent = entry->bytes;
3431 }
3432
3433 if (window_free < bytes || max_extent < cont1_bytes)
3434 return -ENOSPC;
3435
3436 cluster->window_start = first->offset;
3437
3438 node = &first->offset_index;
3439
3440 /*
3441 * now we've found our entries, pull them out of the free space
3442 * cache and put them into the cluster rbtree
3443 */
3444 do {
3445 int ret;
3446
3447 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3448 node = rb_next(&entry->offset_index);
3449 if (entry->bitmap || entry->bytes < min_bytes)
3450 continue;
3451
3452 rb_erase(&entry->offset_index, &ctl->free_space_offset);
3453 rb_erase_cached(&entry->bytes_index, &ctl->free_space_bytes);
3454 ret = tree_insert_offset(&cluster->root, entry->offset,
3455 &entry->offset_index, 0);
3456 total_size += entry->bytes;
3457 ASSERT(!ret); /* -EEXIST; Logic error */
3458 } while (node && entry != last);
3459
3460 cluster->max_size = max_extent;
3461 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
3462 return 0;
3463 }
3464
3465 /*
3466 * This specifically looks for bitmaps that may work in the cluster, we assume
3467 * that we have already failed to find extents that will work.
3468 */
3469 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)3470 setup_cluster_bitmap(struct btrfs_block_group *block_group,
3471 struct btrfs_free_cluster *cluster,
3472 struct list_head *bitmaps, u64 offset, u64 bytes,
3473 u64 cont1_bytes, u64 min_bytes)
3474 {
3475 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3476 struct btrfs_free_space *entry = NULL;
3477 int ret = -ENOSPC;
3478 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
3479
3480 if (ctl->total_bitmaps == 0)
3481 return -ENOSPC;
3482
3483 /*
3484 * The bitmap that covers offset won't be in the list unless offset
3485 * is just its start offset.
3486 */
3487 if (!list_empty(bitmaps))
3488 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
3489
3490 if (!entry || entry->offset != bitmap_offset) {
3491 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
3492 if (entry && list_empty(&entry->list))
3493 list_add(&entry->list, bitmaps);
3494 }
3495
3496 list_for_each_entry(entry, bitmaps, list) {
3497 if (entry->bytes < bytes)
3498 continue;
3499 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
3500 bytes, cont1_bytes, min_bytes);
3501 if (!ret)
3502 return 0;
3503 }
3504
3505 /*
3506 * The bitmaps list has all the bitmaps that record free space
3507 * starting after offset, so no more search is required.
3508 */
3509 return -ENOSPC;
3510 }
3511
3512 /*
3513 * here we try to find a cluster of blocks in a block group. The goal
3514 * is to find at least bytes+empty_size.
3515 * We might not find them all in one contiguous area.
3516 *
3517 * returns zero and sets up cluster if things worked out, otherwise
3518 * it returns -enospc
3519 */
btrfs_find_space_cluster(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster,u64 offset,u64 bytes,u64 empty_size)3520 int btrfs_find_space_cluster(struct btrfs_block_group *block_group,
3521 struct btrfs_free_cluster *cluster,
3522 u64 offset, u64 bytes, u64 empty_size)
3523 {
3524 struct btrfs_fs_info *fs_info = block_group->fs_info;
3525 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3526 struct btrfs_free_space *entry, *tmp;
3527 LIST_HEAD(bitmaps);
3528 u64 min_bytes;
3529 u64 cont1_bytes;
3530 int ret;
3531
3532 /*
3533 * Choose the minimum extent size we'll require for this
3534 * cluster. For SSD_SPREAD, don't allow any fragmentation.
3535 * For metadata, allow allocates with smaller extents. For
3536 * data, keep it dense.
3537 */
3538 if (btrfs_test_opt(fs_info, SSD_SPREAD)) {
3539 cont1_bytes = min_bytes = bytes + empty_size;
3540 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3541 cont1_bytes = bytes;
3542 min_bytes = fs_info->sectorsize;
3543 } else {
3544 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3545 min_bytes = fs_info->sectorsize;
3546 }
3547
3548 spin_lock(&ctl->tree_lock);
3549
3550 /*
3551 * If we know we don't have enough space to make a cluster don't even
3552 * bother doing all the work to try and find one.
3553 */
3554 if (ctl->free_space < bytes) {
3555 spin_unlock(&ctl->tree_lock);
3556 return -ENOSPC;
3557 }
3558
3559 spin_lock(&cluster->lock);
3560
3561 /* someone already found a cluster, hooray */
3562 if (cluster->block_group) {
3563 ret = 0;
3564 goto out;
3565 }
3566
3567 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
3568 min_bytes);
3569
3570 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
3571 bytes + empty_size,
3572 cont1_bytes, min_bytes);
3573 if (ret)
3574 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
3575 offset, bytes + empty_size,
3576 cont1_bytes, min_bytes);
3577
3578 /* Clear our temporary list */
3579 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3580 list_del_init(&entry->list);
3581
3582 if (!ret) {
3583 btrfs_get_block_group(block_group);
3584 list_add_tail(&cluster->block_group_list,
3585 &block_group->cluster_list);
3586 cluster->block_group = block_group;
3587 } else {
3588 trace_btrfs_failed_cluster_setup(block_group);
3589 }
3590 out:
3591 spin_unlock(&cluster->lock);
3592 spin_unlock(&ctl->tree_lock);
3593
3594 return ret;
3595 }
3596
3597 /*
3598 * simple code to zero out a cluster
3599 */
btrfs_init_free_cluster(struct btrfs_free_cluster * cluster)3600 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3601 {
3602 spin_lock_init(&cluster->lock);
3603 spin_lock_init(&cluster->refill_lock);
3604 cluster->root = RB_ROOT;
3605 cluster->max_size = 0;
3606 cluster->fragmented = false;
3607 INIT_LIST_HEAD(&cluster->block_group_list);
3608 cluster->block_group = NULL;
3609 }
3610
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)3611 static int do_trimming(struct btrfs_block_group *block_group,
3612 u64 *total_trimmed, u64 start, u64 bytes,
3613 u64 reserved_start, u64 reserved_bytes,
3614 enum btrfs_trim_state reserved_trim_state,
3615 struct btrfs_trim_range *trim_entry)
3616 {
3617 struct btrfs_space_info *space_info = block_group->space_info;
3618 struct btrfs_fs_info *fs_info = block_group->fs_info;
3619 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3620 int ret;
3621 int update = 0;
3622 const u64 end = start + bytes;
3623 const u64 reserved_end = reserved_start + reserved_bytes;
3624 enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3625 u64 trimmed = 0;
3626
3627 spin_lock(&space_info->lock);
3628 spin_lock(&block_group->lock);
3629 if (!block_group->ro) {
3630 block_group->reserved += reserved_bytes;
3631 space_info->bytes_reserved += reserved_bytes;
3632 update = 1;
3633 }
3634 spin_unlock(&block_group->lock);
3635 spin_unlock(&space_info->lock);
3636
3637 ret = btrfs_discard_extent(fs_info, start, bytes, &trimmed);
3638 if (!ret) {
3639 *total_trimmed += trimmed;
3640 trim_state = BTRFS_TRIM_STATE_TRIMMED;
3641 }
3642
3643 mutex_lock(&ctl->cache_writeout_mutex);
3644 if (reserved_start < start)
3645 __btrfs_add_free_space(block_group, reserved_start,
3646 start - reserved_start,
3647 reserved_trim_state);
3648 if (start + bytes < reserved_start + reserved_bytes)
3649 __btrfs_add_free_space(block_group, end, reserved_end - end,
3650 reserved_trim_state);
3651 __btrfs_add_free_space(block_group, start, bytes, trim_state);
3652 list_del(&trim_entry->list);
3653 mutex_unlock(&ctl->cache_writeout_mutex);
3654
3655 if (update) {
3656 spin_lock(&space_info->lock);
3657 spin_lock(&block_group->lock);
3658 if (block_group->ro)
3659 space_info->bytes_readonly += reserved_bytes;
3660 block_group->reserved -= reserved_bytes;
3661 space_info->bytes_reserved -= reserved_bytes;
3662 spin_unlock(&block_group->lock);
3663 spin_unlock(&space_info->lock);
3664 }
3665
3666 return ret;
3667 }
3668
3669 /*
3670 * If @async is set, then we will trim 1 region and return.
3671 */
trim_no_bitmap(struct btrfs_block_group * block_group,u64 * total_trimmed,u64 start,u64 end,u64 minlen,bool async)3672 static int trim_no_bitmap(struct btrfs_block_group *block_group,
3673 u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3674 bool async)
3675 {
3676 struct btrfs_discard_ctl *discard_ctl =
3677 &block_group->fs_info->discard_ctl;
3678 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3679 struct btrfs_free_space *entry;
3680 struct rb_node *node;
3681 int ret = 0;
3682 u64 extent_start;
3683 u64 extent_bytes;
3684 enum btrfs_trim_state extent_trim_state;
3685 u64 bytes;
3686 const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3687
3688 while (start < end) {
3689 struct btrfs_trim_range trim_entry;
3690
3691 mutex_lock(&ctl->cache_writeout_mutex);
3692 spin_lock(&ctl->tree_lock);
3693
3694 if (ctl->free_space < minlen)
3695 goto out_unlock;
3696
3697 entry = tree_search_offset(ctl, start, 0, 1);
3698 if (!entry)
3699 goto out_unlock;
3700
3701 /* Skip bitmaps and if async, already trimmed entries */
3702 while (entry->bitmap ||
3703 (async && btrfs_free_space_trimmed(entry))) {
3704 node = rb_next(&entry->offset_index);
3705 if (!node)
3706 goto out_unlock;
3707 entry = rb_entry(node, struct btrfs_free_space,
3708 offset_index);
3709 }
3710
3711 if (entry->offset >= end)
3712 goto out_unlock;
3713
3714 extent_start = entry->offset;
3715 extent_bytes = entry->bytes;
3716 extent_trim_state = entry->trim_state;
3717 if (async) {
3718 start = entry->offset;
3719 bytes = entry->bytes;
3720 if (bytes < minlen) {
3721 spin_unlock(&ctl->tree_lock);
3722 mutex_unlock(&ctl->cache_writeout_mutex);
3723 goto next;
3724 }
3725 unlink_free_space(ctl, entry, true);
3726 /*
3727 * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3728 * If X < BTRFS_ASYNC_DISCARD_MIN_FILTER, we won't trim
3729 * X when we come back around. So trim it now.
3730 */
3731 if (max_discard_size &&
3732 bytes >= (max_discard_size +
3733 BTRFS_ASYNC_DISCARD_MIN_FILTER)) {
3734 bytes = max_discard_size;
3735 extent_bytes = max_discard_size;
3736 entry->offset += max_discard_size;
3737 entry->bytes -= max_discard_size;
3738 link_free_space(ctl, entry);
3739 } else {
3740 kmem_cache_free(btrfs_free_space_cachep, entry);
3741 }
3742 } else {
3743 start = max(start, extent_start);
3744 bytes = min(extent_start + extent_bytes, end) - start;
3745 if (bytes < minlen) {
3746 spin_unlock(&ctl->tree_lock);
3747 mutex_unlock(&ctl->cache_writeout_mutex);
3748 goto next;
3749 }
3750
3751 unlink_free_space(ctl, entry, true);
3752 kmem_cache_free(btrfs_free_space_cachep, entry);
3753 }
3754
3755 spin_unlock(&ctl->tree_lock);
3756 trim_entry.start = extent_start;
3757 trim_entry.bytes = extent_bytes;
3758 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3759 mutex_unlock(&ctl->cache_writeout_mutex);
3760
3761 ret = do_trimming(block_group, total_trimmed, start, bytes,
3762 extent_start, extent_bytes, extent_trim_state,
3763 &trim_entry);
3764 if (ret) {
3765 block_group->discard_cursor = start + bytes;
3766 break;
3767 }
3768 next:
3769 start += bytes;
3770 block_group->discard_cursor = start;
3771 if (async && *total_trimmed)
3772 break;
3773
3774 if (fatal_signal_pending(current)) {
3775 ret = -ERESTARTSYS;
3776 break;
3777 }
3778
3779 cond_resched();
3780 }
3781
3782 return ret;
3783
3784 out_unlock:
3785 block_group->discard_cursor = btrfs_block_group_end(block_group);
3786 spin_unlock(&ctl->tree_lock);
3787 mutex_unlock(&ctl->cache_writeout_mutex);
3788
3789 return ret;
3790 }
3791
3792 /*
3793 * If we break out of trimming a bitmap prematurely, we should reset the
3794 * trimming bit. In a rather contrieved case, it's possible to race here so
3795 * reset the state to BTRFS_TRIM_STATE_UNTRIMMED.
3796 *
3797 * start = start of bitmap
3798 * end = near end of bitmap
3799 *
3800 * Thread 1: Thread 2:
3801 * trim_bitmaps(start)
3802 * trim_bitmaps(end)
3803 * end_trimming_bitmap()
3804 * reset_trimming_bitmap()
3805 */
reset_trimming_bitmap(struct btrfs_free_space_ctl * ctl,u64 offset)3806 static void reset_trimming_bitmap(struct btrfs_free_space_ctl *ctl, u64 offset)
3807 {
3808 struct btrfs_free_space *entry;
3809
3810 spin_lock(&ctl->tree_lock);
3811 entry = tree_search_offset(ctl, offset, 1, 0);
3812 if (entry) {
3813 if (btrfs_free_space_trimmed(entry)) {
3814 ctl->discardable_extents[BTRFS_STAT_CURR] +=
3815 entry->bitmap_extents;
3816 ctl->discardable_bytes[BTRFS_STAT_CURR] += entry->bytes;
3817 }
3818 entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3819 }
3820
3821 spin_unlock(&ctl->tree_lock);
3822 }
3823
end_trimming_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * entry)3824 static void end_trimming_bitmap(struct btrfs_free_space_ctl *ctl,
3825 struct btrfs_free_space *entry)
3826 {
3827 if (btrfs_free_space_trimming_bitmap(entry)) {
3828 entry->trim_state = BTRFS_TRIM_STATE_TRIMMED;
3829 ctl->discardable_extents[BTRFS_STAT_CURR] -=
3830 entry->bitmap_extents;
3831 ctl->discardable_bytes[BTRFS_STAT_CURR] -= entry->bytes;
3832 }
3833 }
3834
3835 /*
3836 * If @async is set, then we will trim 1 region and return.
3837 */
trim_bitmaps(struct btrfs_block_group * block_group,u64 * total_trimmed,u64 start,u64 end,u64 minlen,u64 maxlen,bool async)3838 static int trim_bitmaps(struct btrfs_block_group *block_group,
3839 u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3840 u64 maxlen, bool async)
3841 {
3842 struct btrfs_discard_ctl *discard_ctl =
3843 &block_group->fs_info->discard_ctl;
3844 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3845 struct btrfs_free_space *entry;
3846 int ret = 0;
3847 int ret2;
3848 u64 bytes;
3849 u64 offset = offset_to_bitmap(ctl, start);
3850 const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3851
3852 while (offset < end) {
3853 bool next_bitmap = false;
3854 struct btrfs_trim_range trim_entry;
3855
3856 mutex_lock(&ctl->cache_writeout_mutex);
3857 spin_lock(&ctl->tree_lock);
3858
3859 if (ctl->free_space < minlen) {
3860 block_group->discard_cursor =
3861 btrfs_block_group_end(block_group);
3862 spin_unlock(&ctl->tree_lock);
3863 mutex_unlock(&ctl->cache_writeout_mutex);
3864 break;
3865 }
3866
3867 entry = tree_search_offset(ctl, offset, 1, 0);
3868 /*
3869 * Bitmaps are marked trimmed lossily now to prevent constant
3870 * discarding of the same bitmap (the reason why we are bound
3871 * by the filters). So, retrim the block group bitmaps when we
3872 * are preparing to punt to the unused_bgs list. This uses
3873 * @minlen to determine if we are in BTRFS_DISCARD_INDEX_UNUSED
3874 * which is the only discard index which sets minlen to 0.
3875 */
3876 if (!entry || (async && minlen && start == offset &&
3877 btrfs_free_space_trimmed(entry))) {
3878 spin_unlock(&ctl->tree_lock);
3879 mutex_unlock(&ctl->cache_writeout_mutex);
3880 next_bitmap = true;
3881 goto next;
3882 }
3883
3884 /*
3885 * Async discard bitmap trimming begins at by setting the start
3886 * to be key.objectid and the offset_to_bitmap() aligns to the
3887 * start of the bitmap. This lets us know we are fully
3888 * scanning the bitmap rather than only some portion of it.
3889 */
3890 if (start == offset)
3891 entry->trim_state = BTRFS_TRIM_STATE_TRIMMING;
3892
3893 bytes = minlen;
3894 ret2 = search_bitmap(ctl, entry, &start, &bytes, false);
3895 if (ret2 || start >= end) {
3896 /*
3897 * We lossily consider a bitmap trimmed if we only skip
3898 * over regions <= BTRFS_ASYNC_DISCARD_MIN_FILTER.
3899 */
3900 if (ret2 && minlen <= BTRFS_ASYNC_DISCARD_MIN_FILTER)
3901 end_trimming_bitmap(ctl, entry);
3902 else
3903 entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3904 spin_unlock(&ctl->tree_lock);
3905 mutex_unlock(&ctl->cache_writeout_mutex);
3906 next_bitmap = true;
3907 goto next;
3908 }
3909
3910 /*
3911 * We already trimmed a region, but are using the locking above
3912 * to reset the trim_state.
3913 */
3914 if (async && *total_trimmed) {
3915 spin_unlock(&ctl->tree_lock);
3916 mutex_unlock(&ctl->cache_writeout_mutex);
3917 goto out;
3918 }
3919
3920 bytes = min(bytes, end - start);
3921 if (bytes < minlen || (async && maxlen && bytes > maxlen)) {
3922 spin_unlock(&ctl->tree_lock);
3923 mutex_unlock(&ctl->cache_writeout_mutex);
3924 goto next;
3925 }
3926
3927 /*
3928 * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3929 * If X < @minlen, we won't trim X when we come back around.
3930 * So trim it now. We differ here from trimming extents as we
3931 * don't keep individual state per bit.
3932 */
3933 if (async &&
3934 max_discard_size &&
3935 bytes > (max_discard_size + minlen))
3936 bytes = max_discard_size;
3937
3938 bitmap_clear_bits(ctl, entry, start, bytes, true);
3939 if (entry->bytes == 0)
3940 free_bitmap(ctl, entry);
3941
3942 spin_unlock(&ctl->tree_lock);
3943 trim_entry.start = start;
3944 trim_entry.bytes = bytes;
3945 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3946 mutex_unlock(&ctl->cache_writeout_mutex);
3947
3948 ret = do_trimming(block_group, total_trimmed, start, bytes,
3949 start, bytes, 0, &trim_entry);
3950 if (ret) {
3951 reset_trimming_bitmap(ctl, offset);
3952 block_group->discard_cursor =
3953 btrfs_block_group_end(block_group);
3954 break;
3955 }
3956 next:
3957 if (next_bitmap) {
3958 offset += BITS_PER_BITMAP * ctl->unit;
3959 start = offset;
3960 } else {
3961 start += bytes;
3962 }
3963 block_group->discard_cursor = start;
3964
3965 if (fatal_signal_pending(current)) {
3966 if (start != offset)
3967 reset_trimming_bitmap(ctl, offset);
3968 ret = -ERESTARTSYS;
3969 break;
3970 }
3971
3972 cond_resched();
3973 }
3974
3975 if (offset >= end)
3976 block_group->discard_cursor = end;
3977
3978 out:
3979 return ret;
3980 }
3981
btrfs_trim_block_group(struct btrfs_block_group * block_group,u64 * trimmed,u64 start,u64 end,u64 minlen)3982 int btrfs_trim_block_group(struct btrfs_block_group *block_group,
3983 u64 *trimmed, u64 start, u64 end, u64 minlen)
3984 {
3985 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3986 int ret;
3987 u64 rem = 0;
3988
3989 ASSERT(!btrfs_is_zoned(block_group->fs_info));
3990
3991 *trimmed = 0;
3992
3993 spin_lock(&block_group->lock);
3994 if (block_group->removed) {
3995 spin_unlock(&block_group->lock);
3996 return 0;
3997 }
3998 btrfs_freeze_block_group(block_group);
3999 spin_unlock(&block_group->lock);
4000
4001 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, false);
4002 if (ret)
4003 goto out;
4004
4005 ret = trim_bitmaps(block_group, trimmed, start, end, minlen, 0, false);
4006 div64_u64_rem(end, BITS_PER_BITMAP * ctl->unit, &rem);
4007 /* If we ended in the middle of a bitmap, reset the trimming flag */
4008 if (rem)
4009 reset_trimming_bitmap(ctl, offset_to_bitmap(ctl, end));
4010 out:
4011 btrfs_unfreeze_block_group(block_group);
4012 return ret;
4013 }
4014
btrfs_trim_block_group_extents(struct btrfs_block_group * block_group,u64 * trimmed,u64 start,u64 end,u64 minlen,bool async)4015 int btrfs_trim_block_group_extents(struct btrfs_block_group *block_group,
4016 u64 *trimmed, u64 start, u64 end, u64 minlen,
4017 bool async)
4018 {
4019 int ret;
4020
4021 *trimmed = 0;
4022
4023 spin_lock(&block_group->lock);
4024 if (block_group->removed) {
4025 spin_unlock(&block_group->lock);
4026 return 0;
4027 }
4028 btrfs_freeze_block_group(block_group);
4029 spin_unlock(&block_group->lock);
4030
4031 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, async);
4032 btrfs_unfreeze_block_group(block_group);
4033
4034 return ret;
4035 }
4036
btrfs_trim_block_group_bitmaps(struct btrfs_block_group * block_group,u64 * trimmed,u64 start,u64 end,u64 minlen,u64 maxlen,bool async)4037 int btrfs_trim_block_group_bitmaps(struct btrfs_block_group *block_group,
4038 u64 *trimmed, u64 start, u64 end, u64 minlen,
4039 u64 maxlen, bool async)
4040 {
4041 int ret;
4042
4043 *trimmed = 0;
4044
4045 spin_lock(&block_group->lock);
4046 if (block_group->removed) {
4047 spin_unlock(&block_group->lock);
4048 return 0;
4049 }
4050 btrfs_freeze_block_group(block_group);
4051 spin_unlock(&block_group->lock);
4052
4053 ret = trim_bitmaps(block_group, trimmed, start, end, minlen, maxlen,
4054 async);
4055
4056 btrfs_unfreeze_block_group(block_group);
4057
4058 return ret;
4059 }
4060
btrfs_free_space_cache_v1_active(struct btrfs_fs_info * fs_info)4061 bool btrfs_free_space_cache_v1_active(struct btrfs_fs_info *fs_info)
4062 {
4063 return btrfs_super_cache_generation(fs_info->super_copy);
4064 }
4065
cleanup_free_space_cache_v1(struct btrfs_fs_info * fs_info,struct btrfs_trans_handle * trans)4066 static int cleanup_free_space_cache_v1(struct btrfs_fs_info *fs_info,
4067 struct btrfs_trans_handle *trans)
4068 {
4069 struct btrfs_block_group *block_group;
4070 struct rb_node *node;
4071 int ret = 0;
4072
4073 btrfs_info(fs_info, "cleaning free space cache v1");
4074
4075 node = rb_first_cached(&fs_info->block_group_cache_tree);
4076 while (node) {
4077 block_group = rb_entry(node, struct btrfs_block_group, cache_node);
4078 ret = btrfs_remove_free_space_inode(trans, NULL, block_group);
4079 if (ret)
4080 goto out;
4081 node = rb_next(node);
4082 }
4083 out:
4084 return ret;
4085 }
4086
btrfs_set_free_space_cache_v1_active(struct btrfs_fs_info * fs_info,bool active)4087 int btrfs_set_free_space_cache_v1_active(struct btrfs_fs_info *fs_info, bool active)
4088 {
4089 struct btrfs_trans_handle *trans;
4090 int ret;
4091
4092 /*
4093 * update_super_roots will appropriately set or unset
4094 * super_copy->cache_generation based on SPACE_CACHE and
4095 * BTRFS_FS_CLEANUP_SPACE_CACHE_V1. For this reason, we need a
4096 * transaction commit whether we are enabling space cache v1 and don't
4097 * have any other work to do, or are disabling it and removing free
4098 * space inodes.
4099 */
4100 trans = btrfs_start_transaction(fs_info->tree_root, 0);
4101 if (IS_ERR(trans))
4102 return PTR_ERR(trans);
4103
4104 if (!active) {
4105 set_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);
4106 ret = cleanup_free_space_cache_v1(fs_info, trans);
4107 if (ret) {
4108 btrfs_abort_transaction(trans, ret);
4109 btrfs_end_transaction(trans);
4110 goto out;
4111 }
4112 }
4113
4114 ret = btrfs_commit_transaction(trans);
4115 out:
4116 clear_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);
4117
4118 return ret;
4119 }
4120
4121 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4122 /*
4123 * Use this if you need to make a bitmap or extent entry specifically, it
4124 * doesn't do any of the merging that add_free_space does, this acts a lot like
4125 * how the free space cache loading stuff works, so you can get really weird
4126 * configurations.
4127 */
test_add_free_space_entry(struct btrfs_block_group * cache,u64 offset,u64 bytes,bool bitmap)4128 int test_add_free_space_entry(struct btrfs_block_group *cache,
4129 u64 offset, u64 bytes, bool bitmap)
4130 {
4131 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4132 struct btrfs_free_space *info = NULL, *bitmap_info;
4133 void *map = NULL;
4134 enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_TRIMMED;
4135 u64 bytes_added;
4136 int ret;
4137
4138 again:
4139 if (!info) {
4140 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
4141 if (!info)
4142 return -ENOMEM;
4143 }
4144
4145 if (!bitmap) {
4146 spin_lock(&ctl->tree_lock);
4147 info->offset = offset;
4148 info->bytes = bytes;
4149 info->max_extent_size = 0;
4150 ret = link_free_space(ctl, info);
4151 spin_unlock(&ctl->tree_lock);
4152 if (ret)
4153 kmem_cache_free(btrfs_free_space_cachep, info);
4154 return ret;
4155 }
4156
4157 if (!map) {
4158 map = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep, GFP_NOFS);
4159 if (!map) {
4160 kmem_cache_free(btrfs_free_space_cachep, info);
4161 return -ENOMEM;
4162 }
4163 }
4164
4165 spin_lock(&ctl->tree_lock);
4166 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
4167 1, 0);
4168 if (!bitmap_info) {
4169 info->bitmap = map;
4170 map = NULL;
4171 add_new_bitmap(ctl, info, offset);
4172 bitmap_info = info;
4173 info = NULL;
4174 }
4175
4176 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
4177 trim_state);
4178
4179 bytes -= bytes_added;
4180 offset += bytes_added;
4181 spin_unlock(&ctl->tree_lock);
4182
4183 if (bytes)
4184 goto again;
4185
4186 if (info)
4187 kmem_cache_free(btrfs_free_space_cachep, info);
4188 if (map)
4189 kmem_cache_free(btrfs_free_space_bitmap_cachep, map);
4190 return 0;
4191 }
4192
4193 /*
4194 * Checks to see if the given range is in the free space cache. This is really
4195 * just used to check the absence of space, so if there is free space in the
4196 * range at all we will return 1.
4197 */
test_check_exists(struct btrfs_block_group * cache,u64 offset,u64 bytes)4198 int test_check_exists(struct btrfs_block_group *cache,
4199 u64 offset, u64 bytes)
4200 {
4201 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4202 struct btrfs_free_space *info;
4203 int ret = 0;
4204
4205 spin_lock(&ctl->tree_lock);
4206 info = tree_search_offset(ctl, offset, 0, 0);
4207 if (!info) {
4208 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
4209 1, 0);
4210 if (!info)
4211 goto out;
4212 }
4213
4214 have_info:
4215 if (info->bitmap) {
4216 u64 bit_off, bit_bytes;
4217 struct rb_node *n;
4218 struct btrfs_free_space *tmp;
4219
4220 bit_off = offset;
4221 bit_bytes = ctl->unit;
4222 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false);
4223 if (!ret) {
4224 if (bit_off == offset) {
4225 ret = 1;
4226 goto out;
4227 } else if (bit_off > offset &&
4228 offset + bytes > bit_off) {
4229 ret = 1;
4230 goto out;
4231 }
4232 }
4233
4234 n = rb_prev(&info->offset_index);
4235 while (n) {
4236 tmp = rb_entry(n, struct btrfs_free_space,
4237 offset_index);
4238 if (tmp->offset + tmp->bytes < offset)
4239 break;
4240 if (offset + bytes < tmp->offset) {
4241 n = rb_prev(&tmp->offset_index);
4242 continue;
4243 }
4244 info = tmp;
4245 goto have_info;
4246 }
4247
4248 n = rb_next(&info->offset_index);
4249 while (n) {
4250 tmp = rb_entry(n, struct btrfs_free_space,
4251 offset_index);
4252 if (offset + bytes < tmp->offset)
4253 break;
4254 if (tmp->offset + tmp->bytes < offset) {
4255 n = rb_next(&tmp->offset_index);
4256 continue;
4257 }
4258 info = tmp;
4259 goto have_info;
4260 }
4261
4262 ret = 0;
4263 goto out;
4264 }
4265
4266 if (info->offset == offset) {
4267 ret = 1;
4268 goto out;
4269 }
4270
4271 if (offset > info->offset && offset < info->offset + info->bytes)
4272 ret = 1;
4273 out:
4274 spin_unlock(&ctl->tree_lock);
4275 return ret;
4276 }
4277 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */
4278