xref: /linux/fs/btrfs/free-space-cache.c (revision 0eb4aaa230d725fa9b1cd758c0f17abca5597af6)
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 <linux/string_choices.h>
15 #include "extent-tree.h"
16 #include "fs.h"
17 #include "messages.h"
18 #include "misc.h"
19 #include "free-space-cache.h"
20 #include "transaction.h"
21 #include "disk-io.h"
22 #include "extent_io.h"
23 #include "space-info.h"
24 #include "block-group.h"
25 #include "discard.h"
26 #include "subpage.h"
27 #include "inode-item.h"
28 #include "accessors.h"
29 #include "file-item.h"
30 #include "file.h"
31 #include "super.h"
32 
33 #define BITS_PER_BITMAP		(PAGE_SIZE * 8UL)
34 #define MAX_CACHE_BYTES_PER_GIG	SZ_64K
35 #define FORCE_EXTENT_THRESHOLD	SZ_1M
36 
37 static struct kmem_cache *btrfs_free_space_cachep;
38 static struct kmem_cache *btrfs_free_space_bitmap_cachep;
39 
40 struct btrfs_trim_range {
41 	u64 start;
42 	u64 bytes;
43 	struct list_head list;
44 };
45 
46 static int link_free_space(struct btrfs_free_space_ctl *ctl,
47 			   struct btrfs_free_space *info);
48 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
49 			      struct btrfs_free_space *info, bool update_stat);
50 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
51 			 struct btrfs_free_space *bitmap_info, u64 *offset,
52 			 u64 *bytes, bool for_alloc);
53 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
54 			struct btrfs_free_space *bitmap_info);
55 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
56 			      struct btrfs_free_space *info, u64 offset,
57 			      u64 bytes, bool update_stats);
58 
btrfs_crc32c_final(u32 crc,u8 * result)59 static void btrfs_crc32c_final(u32 crc, u8 *result)
60 {
61 	put_unaligned_le32(~crc, result);
62 }
63 
__btrfs_remove_free_space_cache(struct btrfs_free_space_ctl * ctl)64 static void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
65 {
66 	struct btrfs_free_space *info;
67 	struct rb_node *node;
68 
69 	while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
70 		info = rb_entry(node, struct btrfs_free_space, offset_index);
71 		if (!info->bitmap) {
72 			unlink_free_space(ctl, info, true);
73 			kmem_cache_free(btrfs_free_space_cachep, info);
74 		} else {
75 			free_bitmap(ctl, info);
76 		}
77 
78 		cond_resched_lock(&ctl->tree_lock);
79 	}
80 }
81 
__lookup_free_space_inode(struct btrfs_root * root,struct btrfs_path * path,u64 offset)82 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
83 					       struct btrfs_path *path,
84 					       u64 offset)
85 {
86 	struct btrfs_key key;
87 	struct btrfs_key location;
88 	struct btrfs_disk_key disk_key;
89 	struct btrfs_free_space_header *header;
90 	struct extent_buffer *leaf;
91 	struct inode *inode = NULL;
92 	unsigned nofs_flag;
93 	int ret;
94 
95 	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
96 	key.offset = offset;
97 	key.type = 0;
98 
99 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
100 	if (ret < 0)
101 		return ERR_PTR(ret);
102 	if (ret > 0) {
103 		btrfs_release_path(path);
104 		return ERR_PTR(-ENOENT);
105 	}
106 
107 	leaf = path->nodes[0];
108 	header = btrfs_item_ptr(leaf, path->slots[0],
109 				struct btrfs_free_space_header);
110 	btrfs_free_space_key(leaf, header, &disk_key);
111 	btrfs_disk_key_to_cpu(&location, &disk_key);
112 	btrfs_release_path(path);
113 
114 	/*
115 	 * We are often under a trans handle at this point, so we need to make
116 	 * sure NOFS is set to keep us from deadlocking.
117 	 */
118 	nofs_flag = memalloc_nofs_save();
119 	inode = btrfs_iget_path(location.objectid, root, path);
120 	btrfs_release_path(path);
121 	memalloc_nofs_restore(nofs_flag);
122 	if (IS_ERR(inode))
123 		return inode;
124 
125 	mapping_set_gfp_mask(inode->i_mapping,
126 			mapping_gfp_constraint(inode->i_mapping,
127 			~(__GFP_FS | __GFP_HIGHMEM)));
128 
129 	return inode;
130 }
131 
lookup_free_space_inode(struct btrfs_block_group * block_group,struct btrfs_path * path)132 struct inode *lookup_free_space_inode(struct btrfs_block_group *block_group,
133 		struct btrfs_path *path)
134 {
135 	struct btrfs_fs_info *fs_info = block_group->fs_info;
136 	struct inode *inode = NULL;
137 	u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
138 
139 	spin_lock(&block_group->lock);
140 	if (block_group->inode)
141 		inode = igrab(&block_group->inode->vfs_inode);
142 	spin_unlock(&block_group->lock);
143 	if (inode)
144 		return inode;
145 
146 	inode = __lookup_free_space_inode(fs_info->tree_root, path,
147 					  block_group->start);
148 	if (IS_ERR(inode))
149 		return inode;
150 
151 	spin_lock(&block_group->lock);
152 	if (!((BTRFS_I(inode)->flags & flags) == flags)) {
153 		btrfs_info(fs_info, "Old style space inode found, converting.");
154 		BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
155 			BTRFS_INODE_NODATACOW;
156 		block_group->disk_cache_state = BTRFS_DC_CLEAR;
157 	}
158 
159 	if (!test_and_set_bit(BLOCK_GROUP_FLAG_IREF, &block_group->runtime_flags))
160 		block_group->inode = BTRFS_I(igrab(inode));
161 	spin_unlock(&block_group->lock);
162 
163 	return inode;
164 }
165 
__create_free_space_inode(struct btrfs_root * root,struct btrfs_trans_handle * trans,struct btrfs_path * path,u64 ino,u64 offset)166 static int __create_free_space_inode(struct btrfs_root *root,
167 				     struct btrfs_trans_handle *trans,
168 				     struct btrfs_path *path,
169 				     u64 ino, u64 offset)
170 {
171 	struct btrfs_key key;
172 	struct btrfs_disk_key disk_key;
173 	struct btrfs_free_space_header *header;
174 	struct btrfs_inode_item *inode_item;
175 	struct extent_buffer *leaf;
176 	/* We inline CRCs for the free disk space cache */
177 	const u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC |
178 			  BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
179 	int ret;
180 
181 	ret = btrfs_insert_empty_inode(trans, root, path, ino);
182 	if (ret)
183 		return ret;
184 
185 	leaf = path->nodes[0];
186 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
187 				    struct btrfs_inode_item);
188 	btrfs_item_key(leaf, &disk_key, path->slots[0]);
189 	memzero_extent_buffer(leaf, (unsigned long)inode_item,
190 			     sizeof(*inode_item));
191 	btrfs_set_inode_generation(leaf, inode_item, trans->transid);
192 	btrfs_set_inode_size(leaf, inode_item, 0);
193 	btrfs_set_inode_nbytes(leaf, inode_item, 0);
194 	btrfs_set_inode_uid(leaf, inode_item, 0);
195 	btrfs_set_inode_gid(leaf, inode_item, 0);
196 	btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
197 	btrfs_set_inode_flags(leaf, inode_item, flags);
198 	btrfs_set_inode_nlink(leaf, inode_item, 1);
199 	btrfs_set_inode_transid(leaf, inode_item, trans->transid);
200 	btrfs_set_inode_block_group(leaf, inode_item, offset);
201 	btrfs_release_path(path);
202 
203 	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
204 	key.offset = offset;
205 	key.type = 0;
206 	ret = btrfs_insert_empty_item(trans, root, path, &key,
207 				      sizeof(struct btrfs_free_space_header));
208 	if (ret < 0) {
209 		btrfs_release_path(path);
210 		return ret;
211 	}
212 
213 	leaf = path->nodes[0];
214 	header = btrfs_item_ptr(leaf, path->slots[0],
215 				struct btrfs_free_space_header);
216 	memzero_extent_buffer(leaf, (unsigned long)header, sizeof(*header));
217 	btrfs_set_free_space_key(leaf, header, &disk_key);
218 	btrfs_release_path(path);
219 
220 	return 0;
221 }
222 
create_free_space_inode(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path)223 int create_free_space_inode(struct btrfs_trans_handle *trans,
224 			    struct btrfs_block_group *block_group,
225 			    struct btrfs_path *path)
226 {
227 	int ret;
228 	u64 ino;
229 
230 	ret = btrfs_get_free_objectid(trans->fs_info->tree_root, &ino);
231 	if (ret < 0)
232 		return ret;
233 
234 	return __create_free_space_inode(trans->fs_info->tree_root, trans, path,
235 					 ino, block_group->start);
236 }
237 
238 /*
239  * inode is an optional sink: if it is NULL, btrfs_remove_free_space_inode
240  * handles lookup, otherwise it takes ownership and iputs the inode.
241  * Don't reuse an inode pointer after passing it into this function.
242  */
btrfs_remove_free_space_inode(struct btrfs_trans_handle * trans,struct inode * inode,struct btrfs_block_group * block_group)243 int btrfs_remove_free_space_inode(struct btrfs_trans_handle *trans,
244 				  struct inode *inode,
245 				  struct btrfs_block_group *block_group)
246 {
247 	struct btrfs_path *path;
248 	struct btrfs_key key;
249 	int ret = 0;
250 
251 	path = btrfs_alloc_path();
252 	if (!path)
253 		return -ENOMEM;
254 
255 	if (!inode)
256 		inode = lookup_free_space_inode(block_group, path);
257 	if (IS_ERR(inode)) {
258 		if (PTR_ERR(inode) != -ENOENT)
259 			ret = PTR_ERR(inode);
260 		goto out;
261 	}
262 	ret = btrfs_orphan_add(trans, BTRFS_I(inode));
263 	if (ret) {
264 		btrfs_add_delayed_iput(BTRFS_I(inode));
265 		goto out;
266 	}
267 	clear_nlink(inode);
268 	/* One for the block groups ref */
269 	spin_lock(&block_group->lock);
270 	if (test_and_clear_bit(BLOCK_GROUP_FLAG_IREF, &block_group->runtime_flags)) {
271 		block_group->inode = NULL;
272 		spin_unlock(&block_group->lock);
273 		iput(inode);
274 	} else {
275 		spin_unlock(&block_group->lock);
276 	}
277 	/* One for the lookup ref */
278 	btrfs_add_delayed_iput(BTRFS_I(inode));
279 
280 	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
281 	key.type = 0;
282 	key.offset = block_group->start;
283 	ret = btrfs_search_slot(trans, trans->fs_info->tree_root, &key, path,
284 				-1, 1);
285 	if (ret) {
286 		if (ret > 0)
287 			ret = 0;
288 		goto out;
289 	}
290 	ret = btrfs_del_item(trans, trans->fs_info->tree_root, path);
291 out:
292 	btrfs_free_path(path);
293 	return ret;
294 }
295 
btrfs_truncate_free_space_cache(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct inode * vfs_inode)296 int btrfs_truncate_free_space_cache(struct btrfs_trans_handle *trans,
297 				    struct btrfs_block_group *block_group,
298 				    struct inode *vfs_inode)
299 {
300 	struct btrfs_truncate_control control = {
301 		.inode = BTRFS_I(vfs_inode),
302 		.new_size = 0,
303 		.ino = btrfs_ino(BTRFS_I(vfs_inode)),
304 		.min_type = BTRFS_EXTENT_DATA_KEY,
305 		.clear_extent_range = true,
306 	};
307 	struct btrfs_inode *inode = BTRFS_I(vfs_inode);
308 	struct btrfs_root *root = inode->root;
309 	struct extent_state *cached_state = NULL;
310 	int ret = 0;
311 	bool locked = false;
312 
313 	if (block_group) {
314 		struct btrfs_path *path = btrfs_alloc_path();
315 
316 		if (!path) {
317 			ret = -ENOMEM;
318 			goto fail;
319 		}
320 		locked = true;
321 		mutex_lock(&trans->transaction->cache_write_mutex);
322 		if (!list_empty(&block_group->io_list)) {
323 			list_del_init(&block_group->io_list);
324 
325 			btrfs_wait_cache_io(trans, block_group, path);
326 			btrfs_put_block_group(block_group);
327 		}
328 
329 		/*
330 		 * now that we've truncated the cache away, its no longer
331 		 * setup or written
332 		 */
333 		spin_lock(&block_group->lock);
334 		block_group->disk_cache_state = BTRFS_DC_CLEAR;
335 		spin_unlock(&block_group->lock);
336 		btrfs_free_path(path);
337 	}
338 
339 	btrfs_i_size_write(inode, 0);
340 	truncate_pagecache(vfs_inode, 0);
341 
342 	lock_extent(&inode->io_tree, 0, (u64)-1, &cached_state);
343 	btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
344 
345 	/*
346 	 * We skip the throttling logic for free space cache inodes, so we don't
347 	 * need to check for -EAGAIN.
348 	 */
349 	ret = btrfs_truncate_inode_items(trans, root, &control);
350 
351 	inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
352 	btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
353 
354 	unlock_extent(&inode->io_tree, 0, (u64)-1, &cached_state);
355 	if (ret)
356 		goto fail;
357 
358 	ret = btrfs_update_inode(trans, inode);
359 
360 fail:
361 	if (locked)
362 		mutex_unlock(&trans->transaction->cache_write_mutex);
363 	if (ret)
364 		btrfs_abort_transaction(trans, ret);
365 
366 	return ret;
367 }
368 
readahead_cache(struct inode * inode)369 static void readahead_cache(struct inode *inode)
370 {
371 	struct file_ra_state ra;
372 	unsigned long last_index;
373 
374 	file_ra_state_init(&ra, inode->i_mapping);
375 	last_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
376 
377 	page_cache_sync_readahead(inode->i_mapping, &ra, NULL, 0, last_index);
378 }
379 
io_ctl_init(struct btrfs_io_ctl * io_ctl,struct inode * inode,int write)380 static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode,
381 		       int write)
382 {
383 	int num_pages;
384 
385 	num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
386 
387 	/* Make sure we can fit our crcs and generation into the first page */
388 	if (write && (num_pages * sizeof(u32) + sizeof(u64)) > PAGE_SIZE)
389 		return -ENOSPC;
390 
391 	memset(io_ctl, 0, sizeof(struct btrfs_io_ctl));
392 
393 	io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS);
394 	if (!io_ctl->pages)
395 		return -ENOMEM;
396 
397 	io_ctl->num_pages = num_pages;
398 	io_ctl->fs_info = inode_to_fs_info(inode);
399 	io_ctl->inode = inode;
400 
401 	return 0;
402 }
403 ALLOW_ERROR_INJECTION(io_ctl_init, ERRNO);
404 
io_ctl_free(struct btrfs_io_ctl * io_ctl)405 static void io_ctl_free(struct btrfs_io_ctl *io_ctl)
406 {
407 	kfree(io_ctl->pages);
408 	io_ctl->pages = NULL;
409 }
410 
io_ctl_unmap_page(struct btrfs_io_ctl * io_ctl)411 static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl)
412 {
413 	if (io_ctl->cur) {
414 		io_ctl->cur = NULL;
415 		io_ctl->orig = NULL;
416 	}
417 }
418 
io_ctl_map_page(struct btrfs_io_ctl * io_ctl,int clear)419 static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear)
420 {
421 	ASSERT(io_ctl->index < io_ctl->num_pages);
422 	io_ctl->page = io_ctl->pages[io_ctl->index++];
423 	io_ctl->cur = page_address(io_ctl->page);
424 	io_ctl->orig = io_ctl->cur;
425 	io_ctl->size = PAGE_SIZE;
426 	if (clear)
427 		clear_page(io_ctl->cur);
428 }
429 
io_ctl_drop_pages(struct btrfs_io_ctl * io_ctl)430 static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl)
431 {
432 	int i;
433 
434 	io_ctl_unmap_page(io_ctl);
435 
436 	for (i = 0; i < io_ctl->num_pages; i++) {
437 		if (io_ctl->pages[i]) {
438 			btrfs_folio_clear_checked(io_ctl->fs_info,
439 					page_folio(io_ctl->pages[i]),
440 					page_offset(io_ctl->pages[i]),
441 					PAGE_SIZE);
442 			unlock_page(io_ctl->pages[i]);
443 			put_page(io_ctl->pages[i]);
444 		}
445 	}
446 }
447 
io_ctl_prepare_pages(struct btrfs_io_ctl * io_ctl,bool uptodate)448 static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, bool uptodate)
449 {
450 	struct page *page;
451 	struct inode *inode = io_ctl->inode;
452 	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
453 	int i;
454 
455 	for (i = 0; i < io_ctl->num_pages; i++) {
456 		int ret;
457 
458 		page = find_or_create_page(inode->i_mapping, i, mask);
459 		if (!page) {
460 			io_ctl_drop_pages(io_ctl);
461 			return -ENOMEM;
462 		}
463 
464 		ret = set_folio_extent_mapped(page_folio(page));
465 		if (ret < 0) {
466 			unlock_page(page);
467 			put_page(page);
468 			io_ctl_drop_pages(io_ctl);
469 			return ret;
470 		}
471 
472 		io_ctl->pages[i] = page;
473 		if (uptodate && !PageUptodate(page)) {
474 			btrfs_read_folio(NULL, page_folio(page));
475 			lock_page(page);
476 			if (page->mapping != inode->i_mapping) {
477 				btrfs_err(BTRFS_I(inode)->root->fs_info,
478 					  "free space cache page truncated");
479 				io_ctl_drop_pages(io_ctl);
480 				return -EIO;
481 			}
482 			if (!PageUptodate(page)) {
483 				btrfs_err(BTRFS_I(inode)->root->fs_info,
484 					   "error reading free space cache");
485 				io_ctl_drop_pages(io_ctl);
486 				return -EIO;
487 			}
488 		}
489 	}
490 
491 	for (i = 0; i < io_ctl->num_pages; i++)
492 		clear_page_dirty_for_io(io_ctl->pages[i]);
493 
494 	return 0;
495 }
496 
io_ctl_set_generation(struct btrfs_io_ctl * io_ctl,u64 generation)497 static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
498 {
499 	io_ctl_map_page(io_ctl, 1);
500 
501 	/*
502 	 * Skip the csum areas.  If we don't check crcs then we just have a
503 	 * 64bit chunk at the front of the first page.
504 	 */
505 	io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
506 	io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
507 
508 	put_unaligned_le64(generation, io_ctl->cur);
509 	io_ctl->cur += sizeof(u64);
510 }
511 
io_ctl_check_generation(struct btrfs_io_ctl * io_ctl,u64 generation)512 static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
513 {
514 	u64 cache_gen;
515 
516 	/*
517 	 * Skip the crc area.  If we don't check crcs then we just have a 64bit
518 	 * chunk at the front of the first page.
519 	 */
520 	io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
521 	io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
522 
523 	cache_gen = get_unaligned_le64(io_ctl->cur);
524 	if (cache_gen != generation) {
525 		btrfs_err_rl(io_ctl->fs_info,
526 			"space cache generation (%llu) does not match inode (%llu)",
527 				cache_gen, generation);
528 		io_ctl_unmap_page(io_ctl);
529 		return -EIO;
530 	}
531 	io_ctl->cur += sizeof(u64);
532 	return 0;
533 }
534 
io_ctl_set_crc(struct btrfs_io_ctl * io_ctl,int index)535 static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
536 {
537 	u32 *tmp;
538 	u32 crc = ~(u32)0;
539 	unsigned offset = 0;
540 
541 	if (index == 0)
542 		offset = sizeof(u32) * io_ctl->num_pages;
543 
544 	crc = crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
545 	btrfs_crc32c_final(crc, (u8 *)&crc);
546 	io_ctl_unmap_page(io_ctl);
547 	tmp = page_address(io_ctl->pages[0]);
548 	tmp += index;
549 	*tmp = crc;
550 }
551 
io_ctl_check_crc(struct btrfs_io_ctl * io_ctl,int index)552 static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
553 {
554 	u32 *tmp, val;
555 	u32 crc = ~(u32)0;
556 	unsigned offset = 0;
557 
558 	if (index == 0)
559 		offset = sizeof(u32) * io_ctl->num_pages;
560 
561 	tmp = page_address(io_ctl->pages[0]);
562 	tmp += index;
563 	val = *tmp;
564 
565 	io_ctl_map_page(io_ctl, 0);
566 	crc = crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
567 	btrfs_crc32c_final(crc, (u8 *)&crc);
568 	if (val != crc) {
569 		btrfs_err_rl(io_ctl->fs_info,
570 			"csum mismatch on free space cache");
571 		io_ctl_unmap_page(io_ctl);
572 		return -EIO;
573 	}
574 
575 	return 0;
576 }
577 
io_ctl_add_entry(struct btrfs_io_ctl * io_ctl,u64 offset,u64 bytes,void * bitmap)578 static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
579 			    void *bitmap)
580 {
581 	struct btrfs_free_space_entry *entry;
582 
583 	if (!io_ctl->cur)
584 		return -ENOSPC;
585 
586 	entry = io_ctl->cur;
587 	put_unaligned_le64(offset, &entry->offset);
588 	put_unaligned_le64(bytes, &entry->bytes);
589 	entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
590 		BTRFS_FREE_SPACE_EXTENT;
591 	io_ctl->cur += sizeof(struct btrfs_free_space_entry);
592 	io_ctl->size -= sizeof(struct btrfs_free_space_entry);
593 
594 	if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
595 		return 0;
596 
597 	io_ctl_set_crc(io_ctl, io_ctl->index - 1);
598 
599 	/* No more pages to map */
600 	if (io_ctl->index >= io_ctl->num_pages)
601 		return 0;
602 
603 	/* map the next page */
604 	io_ctl_map_page(io_ctl, 1);
605 	return 0;
606 }
607 
io_ctl_add_bitmap(struct btrfs_io_ctl * io_ctl,void * bitmap)608 static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
609 {
610 	if (!io_ctl->cur)
611 		return -ENOSPC;
612 
613 	/*
614 	 * If we aren't at the start of the current page, unmap this one and
615 	 * map the next one if there is any left.
616 	 */
617 	if (io_ctl->cur != io_ctl->orig) {
618 		io_ctl_set_crc(io_ctl, io_ctl->index - 1);
619 		if (io_ctl->index >= io_ctl->num_pages)
620 			return -ENOSPC;
621 		io_ctl_map_page(io_ctl, 0);
622 	}
623 
624 	copy_page(io_ctl->cur, bitmap);
625 	io_ctl_set_crc(io_ctl, io_ctl->index - 1);
626 	if (io_ctl->index < io_ctl->num_pages)
627 		io_ctl_map_page(io_ctl, 0);
628 	return 0;
629 }
630 
io_ctl_zero_remaining_pages(struct btrfs_io_ctl * io_ctl)631 static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
632 {
633 	/*
634 	 * If we're not on the boundary we know we've modified the page and we
635 	 * need to crc the page.
636 	 */
637 	if (io_ctl->cur != io_ctl->orig)
638 		io_ctl_set_crc(io_ctl, io_ctl->index - 1);
639 	else
640 		io_ctl_unmap_page(io_ctl);
641 
642 	while (io_ctl->index < io_ctl->num_pages) {
643 		io_ctl_map_page(io_ctl, 1);
644 		io_ctl_set_crc(io_ctl, io_ctl->index - 1);
645 	}
646 }
647 
io_ctl_read_entry(struct btrfs_io_ctl * io_ctl,struct btrfs_free_space * entry,u8 * type)648 static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
649 			    struct btrfs_free_space *entry, u8 *type)
650 {
651 	struct btrfs_free_space_entry *e;
652 	int ret;
653 
654 	if (!io_ctl->cur) {
655 		ret = io_ctl_check_crc(io_ctl, io_ctl->index);
656 		if (ret)
657 			return ret;
658 	}
659 
660 	e = io_ctl->cur;
661 	entry->offset = get_unaligned_le64(&e->offset);
662 	entry->bytes = get_unaligned_le64(&e->bytes);
663 	*type = e->type;
664 	io_ctl->cur += sizeof(struct btrfs_free_space_entry);
665 	io_ctl->size -= sizeof(struct btrfs_free_space_entry);
666 
667 	if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
668 		return 0;
669 
670 	io_ctl_unmap_page(io_ctl);
671 
672 	return 0;
673 }
674 
io_ctl_read_bitmap(struct btrfs_io_ctl * io_ctl,struct btrfs_free_space * entry)675 static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
676 			      struct btrfs_free_space *entry)
677 {
678 	int ret;
679 
680 	ret = io_ctl_check_crc(io_ctl, io_ctl->index);
681 	if (ret)
682 		return ret;
683 
684 	copy_page(entry->bitmap, io_ctl->cur);
685 	io_ctl_unmap_page(io_ctl);
686 
687 	return 0;
688 }
689 
recalculate_thresholds(struct btrfs_free_space_ctl * ctl)690 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
691 {
692 	struct btrfs_block_group *block_group = ctl->block_group;
693 	u64 max_bytes;
694 	u64 bitmap_bytes;
695 	u64 extent_bytes;
696 	u64 size = block_group->length;
697 	u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
698 	u64 max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
699 
700 	max_bitmaps = max_t(u64, max_bitmaps, 1);
701 
702 	if (ctl->total_bitmaps > max_bitmaps)
703 		btrfs_err(block_group->fs_info,
704 "invalid free space control: bg start=%llu len=%llu total_bitmaps=%u unit=%u max_bitmaps=%llu bytes_per_bg=%llu",
705 			  block_group->start, block_group->length,
706 			  ctl->total_bitmaps, ctl->unit, max_bitmaps,
707 			  bytes_per_bg);
708 	ASSERT(ctl->total_bitmaps <= max_bitmaps);
709 
710 	/*
711 	 * We are trying to keep the total amount of memory used per 1GiB of
712 	 * space to be MAX_CACHE_BYTES_PER_GIG.  However, with a reclamation
713 	 * mechanism of pulling extents >= FORCE_EXTENT_THRESHOLD out of
714 	 * bitmaps, we may end up using more memory than this.
715 	 */
716 	if (size < SZ_1G)
717 		max_bytes = MAX_CACHE_BYTES_PER_GIG;
718 	else
719 		max_bytes = MAX_CACHE_BYTES_PER_GIG * div_u64(size, SZ_1G);
720 
721 	bitmap_bytes = ctl->total_bitmaps * ctl->unit;
722 
723 	/*
724 	 * we want the extent entry threshold to always be at most 1/2 the max
725 	 * bytes we can have, or whatever is less than that.
726 	 */
727 	extent_bytes = max_bytes - bitmap_bytes;
728 	extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
729 
730 	ctl->extents_thresh =
731 		div_u64(extent_bytes, sizeof(struct btrfs_free_space));
732 }
733 
__load_free_space_cache(struct btrfs_root * root,struct inode * inode,struct btrfs_free_space_ctl * ctl,struct btrfs_path * path,u64 offset)734 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
735 				   struct btrfs_free_space_ctl *ctl,
736 				   struct btrfs_path *path, u64 offset)
737 {
738 	struct btrfs_fs_info *fs_info = root->fs_info;
739 	struct btrfs_free_space_header *header;
740 	struct extent_buffer *leaf;
741 	struct btrfs_io_ctl io_ctl;
742 	struct btrfs_key key;
743 	struct btrfs_free_space *e, *n;
744 	LIST_HEAD(bitmaps);
745 	u64 num_entries;
746 	u64 num_bitmaps;
747 	u64 generation;
748 	u8 type;
749 	int ret = 0;
750 
751 	/* Nothing in the space cache, goodbye */
752 	if (!i_size_read(inode))
753 		return 0;
754 
755 	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
756 	key.offset = offset;
757 	key.type = 0;
758 
759 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
760 	if (ret < 0)
761 		return 0;
762 	else if (ret > 0) {
763 		btrfs_release_path(path);
764 		return 0;
765 	}
766 
767 	ret = -1;
768 
769 	leaf = path->nodes[0];
770 	header = btrfs_item_ptr(leaf, path->slots[0],
771 				struct btrfs_free_space_header);
772 	num_entries = btrfs_free_space_entries(leaf, header);
773 	num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
774 	generation = btrfs_free_space_generation(leaf, header);
775 	btrfs_release_path(path);
776 
777 	if (!BTRFS_I(inode)->generation) {
778 		btrfs_info(fs_info,
779 			   "the free space cache file (%llu) is invalid, skip it",
780 			   offset);
781 		return 0;
782 	}
783 
784 	if (BTRFS_I(inode)->generation != generation) {
785 		btrfs_err(fs_info,
786 			  "free space inode generation (%llu) did not match free space cache generation (%llu)",
787 			  BTRFS_I(inode)->generation, generation);
788 		return 0;
789 	}
790 
791 	if (!num_entries)
792 		return 0;
793 
794 	ret = io_ctl_init(&io_ctl, inode, 0);
795 	if (ret)
796 		return ret;
797 
798 	readahead_cache(inode);
799 
800 	ret = io_ctl_prepare_pages(&io_ctl, true);
801 	if (ret)
802 		goto out;
803 
804 	ret = io_ctl_check_crc(&io_ctl, 0);
805 	if (ret)
806 		goto free_cache;
807 
808 	ret = io_ctl_check_generation(&io_ctl, generation);
809 	if (ret)
810 		goto free_cache;
811 
812 	while (num_entries) {
813 		e = kmem_cache_zalloc(btrfs_free_space_cachep,
814 				      GFP_NOFS);
815 		if (!e) {
816 			ret = -ENOMEM;
817 			goto free_cache;
818 		}
819 
820 		ret = io_ctl_read_entry(&io_ctl, e, &type);
821 		if (ret) {
822 			kmem_cache_free(btrfs_free_space_cachep, e);
823 			goto free_cache;
824 		}
825 
826 		if (!e->bytes) {
827 			ret = -1;
828 			kmem_cache_free(btrfs_free_space_cachep, e);
829 			goto free_cache;
830 		}
831 
832 		if (type == BTRFS_FREE_SPACE_EXTENT) {
833 			spin_lock(&ctl->tree_lock);
834 			ret = link_free_space(ctl, e);
835 			spin_unlock(&ctl->tree_lock);
836 			if (ret) {
837 				btrfs_err(fs_info,
838 					"Duplicate entries in free space cache, dumping");
839 				kmem_cache_free(btrfs_free_space_cachep, e);
840 				goto free_cache;
841 			}
842 		} else {
843 			ASSERT(num_bitmaps);
844 			num_bitmaps--;
845 			e->bitmap = kmem_cache_zalloc(
846 					btrfs_free_space_bitmap_cachep, GFP_NOFS);
847 			if (!e->bitmap) {
848 				ret = -ENOMEM;
849 				kmem_cache_free(
850 					btrfs_free_space_cachep, e);
851 				goto free_cache;
852 			}
853 			spin_lock(&ctl->tree_lock);
854 			ret = link_free_space(ctl, e);
855 			if (ret) {
856 				spin_unlock(&ctl->tree_lock);
857 				btrfs_err(fs_info,
858 					"Duplicate entries in free space cache, dumping");
859 				kmem_cache_free(btrfs_free_space_bitmap_cachep, e->bitmap);
860 				kmem_cache_free(btrfs_free_space_cachep, e);
861 				goto free_cache;
862 			}
863 			ctl->total_bitmaps++;
864 			recalculate_thresholds(ctl);
865 			spin_unlock(&ctl->tree_lock);
866 			list_add_tail(&e->list, &bitmaps);
867 		}
868 
869 		num_entries--;
870 	}
871 
872 	io_ctl_unmap_page(&io_ctl);
873 
874 	/*
875 	 * We add the bitmaps at the end of the entries in order that
876 	 * the bitmap entries are added to the cache.
877 	 */
878 	list_for_each_entry_safe(e, n, &bitmaps, list) {
879 		list_del_init(&e->list);
880 		ret = io_ctl_read_bitmap(&io_ctl, e);
881 		if (ret)
882 			goto free_cache;
883 	}
884 
885 	io_ctl_drop_pages(&io_ctl);
886 	ret = 1;
887 out:
888 	io_ctl_free(&io_ctl);
889 	return ret;
890 free_cache:
891 	io_ctl_drop_pages(&io_ctl);
892 
893 	spin_lock(&ctl->tree_lock);
894 	__btrfs_remove_free_space_cache(ctl);
895 	spin_unlock(&ctl->tree_lock);
896 	goto out;
897 }
898 
copy_free_space_cache(struct btrfs_block_group * block_group,struct btrfs_free_space_ctl * ctl)899 static int copy_free_space_cache(struct btrfs_block_group *block_group,
900 				 struct btrfs_free_space_ctl *ctl)
901 {
902 	struct btrfs_free_space *info;
903 	struct rb_node *n;
904 	int ret = 0;
905 
906 	while (!ret && (n = rb_first(&ctl->free_space_offset)) != NULL) {
907 		info = rb_entry(n, struct btrfs_free_space, offset_index);
908 		if (!info->bitmap) {
909 			const u64 offset = info->offset;
910 			const u64 bytes = info->bytes;
911 
912 			unlink_free_space(ctl, info, true);
913 			spin_unlock(&ctl->tree_lock);
914 			kmem_cache_free(btrfs_free_space_cachep, info);
915 			ret = btrfs_add_free_space(block_group, offset, bytes);
916 			spin_lock(&ctl->tree_lock);
917 		} else {
918 			u64 offset = info->offset;
919 			u64 bytes = ctl->unit;
920 
921 			ret = search_bitmap(ctl, info, &offset, &bytes, false);
922 			if (ret == 0) {
923 				bitmap_clear_bits(ctl, info, offset, bytes, true);
924 				spin_unlock(&ctl->tree_lock);
925 				ret = btrfs_add_free_space(block_group, offset,
926 							   bytes);
927 				spin_lock(&ctl->tree_lock);
928 			} else {
929 				free_bitmap(ctl, info);
930 				ret = 0;
931 			}
932 		}
933 		cond_resched_lock(&ctl->tree_lock);
934 	}
935 	return ret;
936 }
937 
938 static struct lock_class_key btrfs_free_space_inode_key;
939 
load_free_space_cache(struct btrfs_block_group * block_group)940 int load_free_space_cache(struct btrfs_block_group *block_group)
941 {
942 	struct btrfs_fs_info *fs_info = block_group->fs_info;
943 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
944 	struct btrfs_free_space_ctl tmp_ctl = {};
945 	struct inode *inode;
946 	struct btrfs_path *path;
947 	int ret = 0;
948 	bool matched;
949 	u64 used = block_group->used;
950 
951 	/*
952 	 * Because we could potentially discard our loaded free space, we want
953 	 * to load everything into a temporary structure first, and then if it's
954 	 * valid copy it all into the actual free space ctl.
955 	 */
956 	btrfs_init_free_space_ctl(block_group, &tmp_ctl);
957 
958 	/*
959 	 * If this block group has been marked to be cleared for one reason or
960 	 * another then we can't trust the on disk cache, so just return.
961 	 */
962 	spin_lock(&block_group->lock);
963 	if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
964 		spin_unlock(&block_group->lock);
965 		return 0;
966 	}
967 	spin_unlock(&block_group->lock);
968 
969 	path = btrfs_alloc_path();
970 	if (!path)
971 		return 0;
972 	path->search_commit_root = 1;
973 	path->skip_locking = 1;
974 
975 	/*
976 	 * We must pass a path with search_commit_root set to btrfs_iget in
977 	 * order to avoid a deadlock when allocating extents for the tree root.
978 	 *
979 	 * When we are COWing an extent buffer from the tree root, when looking
980 	 * for a free extent, at extent-tree.c:find_free_extent(), we can find
981 	 * block group without its free space cache loaded. When we find one
982 	 * we must load its space cache which requires reading its free space
983 	 * cache's inode item from the root tree. If this inode item is located
984 	 * in the same leaf that we started COWing before, then we end up in
985 	 * deadlock on the extent buffer (trying to read lock it when we
986 	 * previously write locked it).
987 	 *
988 	 * It's safe to read the inode item using the commit root because
989 	 * block groups, once loaded, stay in memory forever (until they are
990 	 * removed) as well as their space caches once loaded. New block groups
991 	 * once created get their ->cached field set to BTRFS_CACHE_FINISHED so
992 	 * we will never try to read their inode item while the fs is mounted.
993 	 */
994 	inode = lookup_free_space_inode(block_group, path);
995 	if (IS_ERR(inode)) {
996 		btrfs_free_path(path);
997 		return 0;
998 	}
999 
1000 	/* We may have converted the inode and made the cache invalid. */
1001 	spin_lock(&block_group->lock);
1002 	if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
1003 		spin_unlock(&block_group->lock);
1004 		btrfs_free_path(path);
1005 		goto out;
1006 	}
1007 	spin_unlock(&block_group->lock);
1008 
1009 	/*
1010 	 * Reinitialize the class of struct inode's mapping->invalidate_lock for
1011 	 * free space inodes to prevent false positives related to locks for normal
1012 	 * inodes.
1013 	 */
1014 	lockdep_set_class(&(&inode->i_data)->invalidate_lock,
1015 			  &btrfs_free_space_inode_key);
1016 
1017 	ret = __load_free_space_cache(fs_info->tree_root, inode, &tmp_ctl,
1018 				      path, block_group->start);
1019 	btrfs_free_path(path);
1020 	if (ret <= 0)
1021 		goto out;
1022 
1023 	matched = (tmp_ctl.free_space == (block_group->length - used -
1024 					  block_group->bytes_super));
1025 
1026 	if (matched) {
1027 		spin_lock(&tmp_ctl.tree_lock);
1028 		ret = copy_free_space_cache(block_group, &tmp_ctl);
1029 		spin_unlock(&tmp_ctl.tree_lock);
1030 		/*
1031 		 * ret == 1 means we successfully loaded the free space cache,
1032 		 * so we need to re-set it here.
1033 		 */
1034 		if (ret == 0)
1035 			ret = 1;
1036 	} else {
1037 		/*
1038 		 * We need to call the _locked variant so we don't try to update
1039 		 * the discard counters.
1040 		 */
1041 		spin_lock(&tmp_ctl.tree_lock);
1042 		__btrfs_remove_free_space_cache(&tmp_ctl);
1043 		spin_unlock(&tmp_ctl.tree_lock);
1044 		btrfs_warn(fs_info,
1045 			   "block group %llu has wrong amount of free space",
1046 			   block_group->start);
1047 		ret = -1;
1048 	}
1049 out:
1050 	if (ret < 0) {
1051 		/* This cache is bogus, make sure it gets cleared */
1052 		spin_lock(&block_group->lock);
1053 		block_group->disk_cache_state = BTRFS_DC_CLEAR;
1054 		spin_unlock(&block_group->lock);
1055 		ret = 0;
1056 
1057 		btrfs_warn(fs_info,
1058 			   "failed to load free space cache for block group %llu, rebuilding it now",
1059 			   block_group->start);
1060 	}
1061 
1062 	spin_lock(&ctl->tree_lock);
1063 	btrfs_discard_update_discardable(block_group);
1064 	spin_unlock(&ctl->tree_lock);
1065 	iput(inode);
1066 	return ret;
1067 }
1068 
1069 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)1070 int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
1071 			      struct btrfs_free_space_ctl *ctl,
1072 			      struct btrfs_block_group *block_group,
1073 			      int *entries, int *bitmaps,
1074 			      struct list_head *bitmap_list)
1075 {
1076 	int ret;
1077 	struct btrfs_free_cluster *cluster = NULL;
1078 	struct btrfs_free_cluster *cluster_locked = NULL;
1079 	struct rb_node *node = rb_first(&ctl->free_space_offset);
1080 	struct btrfs_trim_range *trim_entry;
1081 
1082 	/* Get the cluster for this block_group if it exists */
1083 	if (block_group && !list_empty(&block_group->cluster_list)) {
1084 		cluster = list_entry(block_group->cluster_list.next,
1085 				     struct btrfs_free_cluster,
1086 				     block_group_list);
1087 	}
1088 
1089 	if (!node && cluster) {
1090 		cluster_locked = cluster;
1091 		spin_lock(&cluster_locked->lock);
1092 		node = rb_first(&cluster->root);
1093 		cluster = NULL;
1094 	}
1095 
1096 	/* Write out the extent entries */
1097 	while (node) {
1098 		struct btrfs_free_space *e;
1099 
1100 		e = rb_entry(node, struct btrfs_free_space, offset_index);
1101 		*entries += 1;
1102 
1103 		ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
1104 				       e->bitmap);
1105 		if (ret)
1106 			goto fail;
1107 
1108 		if (e->bitmap) {
1109 			list_add_tail(&e->list, bitmap_list);
1110 			*bitmaps += 1;
1111 		}
1112 		node = rb_next(node);
1113 		if (!node && cluster) {
1114 			node = rb_first(&cluster->root);
1115 			cluster_locked = cluster;
1116 			spin_lock(&cluster_locked->lock);
1117 			cluster = NULL;
1118 		}
1119 	}
1120 	if (cluster_locked) {
1121 		spin_unlock(&cluster_locked->lock);
1122 		cluster_locked = NULL;
1123 	}
1124 
1125 	/*
1126 	 * Make sure we don't miss any range that was removed from our rbtree
1127 	 * because trimming is running. Otherwise after a umount+mount (or crash
1128 	 * after committing the transaction) we would leak free space and get
1129 	 * an inconsistent free space cache report from fsck.
1130 	 */
1131 	list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
1132 		ret = io_ctl_add_entry(io_ctl, trim_entry->start,
1133 				       trim_entry->bytes, NULL);
1134 		if (ret)
1135 			goto fail;
1136 		*entries += 1;
1137 	}
1138 
1139 	return 0;
1140 fail:
1141 	if (cluster_locked)
1142 		spin_unlock(&cluster_locked->lock);
1143 	return -ENOSPC;
1144 }
1145 
1146 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)1147 update_cache_item(struct btrfs_trans_handle *trans,
1148 		  struct btrfs_root *root,
1149 		  struct inode *inode,
1150 		  struct btrfs_path *path, u64 offset,
1151 		  int entries, int bitmaps)
1152 {
1153 	struct btrfs_key key;
1154 	struct btrfs_free_space_header *header;
1155 	struct extent_buffer *leaf;
1156 	int ret;
1157 
1158 	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1159 	key.offset = offset;
1160 	key.type = 0;
1161 
1162 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1163 	if (ret < 0) {
1164 		clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1165 				 EXTENT_DELALLOC, NULL);
1166 		goto fail;
1167 	}
1168 	leaf = path->nodes[0];
1169 	if (ret > 0) {
1170 		struct btrfs_key found_key;
1171 		ASSERT(path->slots[0]);
1172 		path->slots[0]--;
1173 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1174 		if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1175 		    found_key.offset != offset) {
1176 			clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1177 					 inode->i_size - 1, EXTENT_DELALLOC,
1178 					 NULL);
1179 			btrfs_release_path(path);
1180 			goto fail;
1181 		}
1182 	}
1183 
1184 	BTRFS_I(inode)->generation = trans->transid;
1185 	header = btrfs_item_ptr(leaf, path->slots[0],
1186 				struct btrfs_free_space_header);
1187 	btrfs_set_free_space_entries(leaf, header, entries);
1188 	btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1189 	btrfs_set_free_space_generation(leaf, header, trans->transid);
1190 	btrfs_release_path(path);
1191 
1192 	return 0;
1193 
1194 fail:
1195 	return -1;
1196 }
1197 
write_pinned_extent_entries(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_io_ctl * io_ctl,int * entries)1198 static noinline_for_stack int write_pinned_extent_entries(
1199 			    struct btrfs_trans_handle *trans,
1200 			    struct btrfs_block_group *block_group,
1201 			    struct btrfs_io_ctl *io_ctl,
1202 			    int *entries)
1203 {
1204 	u64 start, extent_start, extent_end, len;
1205 	struct extent_io_tree *unpin = NULL;
1206 	int ret;
1207 
1208 	if (!block_group)
1209 		return 0;
1210 
1211 	/*
1212 	 * We want to add any pinned extents to our free space cache
1213 	 * so we don't leak the space
1214 	 *
1215 	 * We shouldn't have switched the pinned extents yet so this is the
1216 	 * right one
1217 	 */
1218 	unpin = &trans->transaction->pinned_extents;
1219 
1220 	start = block_group->start;
1221 
1222 	while (start < block_group->start + block_group->length) {
1223 		if (!find_first_extent_bit(unpin, start,
1224 					   &extent_start, &extent_end,
1225 					   EXTENT_DIRTY, NULL))
1226 			return 0;
1227 
1228 		/* This pinned extent is out of our range */
1229 		if (extent_start >= block_group->start + block_group->length)
1230 			return 0;
1231 
1232 		extent_start = max(extent_start, start);
1233 		extent_end = min(block_group->start + block_group->length,
1234 				 extent_end + 1);
1235 		len = extent_end - extent_start;
1236 
1237 		*entries += 1;
1238 		ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1239 		if (ret)
1240 			return -ENOSPC;
1241 
1242 		start = extent_end;
1243 	}
1244 
1245 	return 0;
1246 }
1247 
1248 static noinline_for_stack int
write_bitmap_entries(struct btrfs_io_ctl * io_ctl,struct list_head * bitmap_list)1249 write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1250 {
1251 	struct btrfs_free_space *entry, *next;
1252 	int ret;
1253 
1254 	/* Write out the bitmaps */
1255 	list_for_each_entry_safe(entry, next, bitmap_list, list) {
1256 		ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1257 		if (ret)
1258 			return -ENOSPC;
1259 		list_del_init(&entry->list);
1260 	}
1261 
1262 	return 0;
1263 }
1264 
flush_dirty_cache(struct inode * inode)1265 static int flush_dirty_cache(struct inode *inode)
1266 {
1267 	int ret;
1268 
1269 	ret = btrfs_wait_ordered_range(BTRFS_I(inode), 0, (u64)-1);
1270 	if (ret)
1271 		clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1272 				 EXTENT_DELALLOC, NULL);
1273 
1274 	return ret;
1275 }
1276 
1277 static void noinline_for_stack
cleanup_bitmap_list(struct list_head * bitmap_list)1278 cleanup_bitmap_list(struct list_head *bitmap_list)
1279 {
1280 	struct btrfs_free_space *entry, *next;
1281 
1282 	list_for_each_entry_safe(entry, next, bitmap_list, list)
1283 		list_del_init(&entry->list);
1284 }
1285 
1286 static void noinline_for_stack
cleanup_write_cache_enospc(struct inode * inode,struct btrfs_io_ctl * io_ctl,struct extent_state ** cached_state)1287 cleanup_write_cache_enospc(struct inode *inode,
1288 			   struct btrfs_io_ctl *io_ctl,
1289 			   struct extent_state **cached_state)
1290 {
1291 	io_ctl_drop_pages(io_ctl);
1292 	unlock_extent(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1293 		      cached_state);
1294 }
1295 
__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)1296 static int __btrfs_wait_cache_io(struct btrfs_root *root,
1297 				 struct btrfs_trans_handle *trans,
1298 				 struct btrfs_block_group *block_group,
1299 				 struct btrfs_io_ctl *io_ctl,
1300 				 struct btrfs_path *path, u64 offset)
1301 {
1302 	int ret;
1303 	struct inode *inode = io_ctl->inode;
1304 
1305 	if (!inode)
1306 		return 0;
1307 
1308 	/* Flush the dirty pages in the cache file. */
1309 	ret = flush_dirty_cache(inode);
1310 	if (ret)
1311 		goto out;
1312 
1313 	/* Update the cache item to tell everyone this cache file is valid. */
1314 	ret = update_cache_item(trans, root, inode, path, offset,
1315 				io_ctl->entries, io_ctl->bitmaps);
1316 out:
1317 	if (ret) {
1318 		invalidate_inode_pages2(inode->i_mapping);
1319 		BTRFS_I(inode)->generation = 0;
1320 		if (block_group)
1321 			btrfs_debug(root->fs_info,
1322 	  "failed to write free space cache for block group %llu error %d",
1323 				  block_group->start, ret);
1324 	}
1325 	btrfs_update_inode(trans, BTRFS_I(inode));
1326 
1327 	if (block_group) {
1328 		/* the dirty list is protected by the dirty_bgs_lock */
1329 		spin_lock(&trans->transaction->dirty_bgs_lock);
1330 
1331 		/* the disk_cache_state is protected by the block group lock */
1332 		spin_lock(&block_group->lock);
1333 
1334 		/*
1335 		 * only mark this as written if we didn't get put back on
1336 		 * the dirty list while waiting for IO.   Otherwise our
1337 		 * cache state won't be right, and we won't get written again
1338 		 */
1339 		if (!ret && list_empty(&block_group->dirty_list))
1340 			block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1341 		else if (ret)
1342 			block_group->disk_cache_state = BTRFS_DC_ERROR;
1343 
1344 		spin_unlock(&block_group->lock);
1345 		spin_unlock(&trans->transaction->dirty_bgs_lock);
1346 		io_ctl->inode = NULL;
1347 		iput(inode);
1348 	}
1349 
1350 	return ret;
1351 
1352 }
1353 
btrfs_wait_cache_io(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path)1354 int btrfs_wait_cache_io(struct btrfs_trans_handle *trans,
1355 			struct btrfs_block_group *block_group,
1356 			struct btrfs_path *path)
1357 {
1358 	return __btrfs_wait_cache_io(block_group->fs_info->tree_root, trans,
1359 				     block_group, &block_group->io_ctl,
1360 				     path, block_group->start);
1361 }
1362 
1363 /*
1364  * Write out cached info to an inode.
1365  *
1366  * @inode:       freespace inode we are writing out
1367  * @ctl:         free space cache we are going to write out
1368  * @block_group: block_group for this cache if it belongs to a block_group
1369  * @io_ctl:      holds context for the io
1370  * @trans:       the trans handle
1371  *
1372  * This function writes out a free space cache struct to disk for quick recovery
1373  * on mount.  This will return 0 if it was successful in writing the cache out,
1374  * or an errno if it was not.
1375  */
__btrfs_write_out_cache(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)1376 static int __btrfs_write_out_cache(struct inode *inode,
1377 				   struct btrfs_free_space_ctl *ctl,
1378 				   struct btrfs_block_group *block_group,
1379 				   struct btrfs_io_ctl *io_ctl,
1380 				   struct btrfs_trans_handle *trans)
1381 {
1382 	struct extent_state *cached_state = NULL;
1383 	LIST_HEAD(bitmap_list);
1384 	int entries = 0;
1385 	int bitmaps = 0;
1386 	int ret;
1387 	int must_iput = 0;
1388 	int i_size;
1389 
1390 	if (!i_size_read(inode))
1391 		return -EIO;
1392 
1393 	WARN_ON(io_ctl->pages);
1394 	ret = io_ctl_init(io_ctl, inode, 1);
1395 	if (ret)
1396 		return ret;
1397 
1398 	if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1399 		down_write(&block_group->data_rwsem);
1400 		spin_lock(&block_group->lock);
1401 		if (block_group->delalloc_bytes) {
1402 			block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1403 			spin_unlock(&block_group->lock);
1404 			up_write(&block_group->data_rwsem);
1405 			BTRFS_I(inode)->generation = 0;
1406 			ret = 0;
1407 			must_iput = 1;
1408 			goto out;
1409 		}
1410 		spin_unlock(&block_group->lock);
1411 	}
1412 
1413 	/* Lock all pages first so we can lock the extent safely. */
1414 	ret = io_ctl_prepare_pages(io_ctl, false);
1415 	if (ret)
1416 		goto out_unlock;
1417 
1418 	lock_extent(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1419 		    &cached_state);
1420 
1421 	io_ctl_set_generation(io_ctl, trans->transid);
1422 
1423 	mutex_lock(&ctl->cache_writeout_mutex);
1424 	/* Write out the extent entries in the free space cache */
1425 	spin_lock(&ctl->tree_lock);
1426 	ret = write_cache_extent_entries(io_ctl, ctl,
1427 					 block_group, &entries, &bitmaps,
1428 					 &bitmap_list);
1429 	if (ret)
1430 		goto out_nospc_locked;
1431 
1432 	/*
1433 	 * Some spaces that are freed in the current transaction are pinned,
1434 	 * they will be added into free space cache after the transaction is
1435 	 * committed, we shouldn't lose them.
1436 	 *
1437 	 * If this changes while we are working we'll get added back to
1438 	 * the dirty list and redo it.  No locking needed
1439 	 */
1440 	ret = write_pinned_extent_entries(trans, block_group, io_ctl, &entries);
1441 	if (ret)
1442 		goto out_nospc_locked;
1443 
1444 	/*
1445 	 * At last, we write out all the bitmaps and keep cache_writeout_mutex
1446 	 * locked while doing it because a concurrent trim can be manipulating
1447 	 * or freeing the bitmap.
1448 	 */
1449 	ret = write_bitmap_entries(io_ctl, &bitmap_list);
1450 	spin_unlock(&ctl->tree_lock);
1451 	mutex_unlock(&ctl->cache_writeout_mutex);
1452 	if (ret)
1453 		goto out_nospc;
1454 
1455 	/* Zero out the rest of the pages just to make sure */
1456 	io_ctl_zero_remaining_pages(io_ctl);
1457 
1458 	/* Everything is written out, now we dirty the pages in the file. */
1459 	i_size = i_size_read(inode);
1460 	for (int i = 0; i < round_up(i_size, PAGE_SIZE) / PAGE_SIZE; i++) {
1461 		u64 dirty_start = i * PAGE_SIZE;
1462 		u64 dirty_len = min_t(u64, dirty_start + PAGE_SIZE, i_size) - dirty_start;
1463 
1464 		ret = btrfs_dirty_folio(BTRFS_I(inode), page_folio(io_ctl->pages[i]),
1465 					dirty_start, dirty_len, &cached_state, false);
1466 		if (ret < 0)
1467 			goto out_nospc;
1468 	}
1469 
1470 	if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1471 		up_write(&block_group->data_rwsem);
1472 	/*
1473 	 * Release the pages and unlock the extent, we will flush
1474 	 * them out later
1475 	 */
1476 	io_ctl_drop_pages(io_ctl);
1477 	io_ctl_free(io_ctl);
1478 
1479 	unlock_extent(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1480 		      &cached_state);
1481 
1482 	/*
1483 	 * at this point the pages are under IO and we're happy,
1484 	 * The caller is responsible for waiting on them and updating
1485 	 * the cache and the inode
1486 	 */
1487 	io_ctl->entries = entries;
1488 	io_ctl->bitmaps = bitmaps;
1489 
1490 	ret = btrfs_fdatawrite_range(BTRFS_I(inode), 0, (u64)-1);
1491 	if (ret)
1492 		goto out;
1493 
1494 	return 0;
1495 
1496 out_nospc_locked:
1497 	cleanup_bitmap_list(&bitmap_list);
1498 	spin_unlock(&ctl->tree_lock);
1499 	mutex_unlock(&ctl->cache_writeout_mutex);
1500 
1501 out_nospc:
1502 	cleanup_write_cache_enospc(inode, io_ctl, &cached_state);
1503 
1504 out_unlock:
1505 	if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1506 		up_write(&block_group->data_rwsem);
1507 
1508 out:
1509 	io_ctl->inode = NULL;
1510 	io_ctl_free(io_ctl);
1511 	if (ret) {
1512 		invalidate_inode_pages2(inode->i_mapping);
1513 		BTRFS_I(inode)->generation = 0;
1514 	}
1515 	btrfs_update_inode(trans, BTRFS_I(inode));
1516 	if (must_iput)
1517 		iput(inode);
1518 	return ret;
1519 }
1520 
btrfs_write_out_cache(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path)1521 int btrfs_write_out_cache(struct btrfs_trans_handle *trans,
1522 			  struct btrfs_block_group *block_group,
1523 			  struct btrfs_path *path)
1524 {
1525 	struct btrfs_fs_info *fs_info = trans->fs_info;
1526 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1527 	struct inode *inode;
1528 	int ret = 0;
1529 
1530 	spin_lock(&block_group->lock);
1531 	if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1532 		spin_unlock(&block_group->lock);
1533 		return 0;
1534 	}
1535 	spin_unlock(&block_group->lock);
1536 
1537 	inode = lookup_free_space_inode(block_group, path);
1538 	if (IS_ERR(inode))
1539 		return 0;
1540 
1541 	ret = __btrfs_write_out_cache(inode, ctl, block_group,
1542 				      &block_group->io_ctl, trans);
1543 	if (ret) {
1544 		btrfs_debug(fs_info,
1545 	  "failed to write free space cache for block group %llu error %d",
1546 			  block_group->start, ret);
1547 		spin_lock(&block_group->lock);
1548 		block_group->disk_cache_state = BTRFS_DC_ERROR;
1549 		spin_unlock(&block_group->lock);
1550 
1551 		block_group->io_ctl.inode = NULL;
1552 		iput(inode);
1553 	}
1554 
1555 	/*
1556 	 * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1557 	 * to wait for IO and put the inode
1558 	 */
1559 
1560 	return ret;
1561 }
1562 
offset_to_bit(u64 bitmap_start,u32 unit,u64 offset)1563 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1564 					  u64 offset)
1565 {
1566 	ASSERT(offset >= bitmap_start);
1567 	offset -= bitmap_start;
1568 	return (unsigned long)(div_u64(offset, unit));
1569 }
1570 
bytes_to_bits(u64 bytes,u32 unit)1571 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1572 {
1573 	return (unsigned long)(div_u64(bytes, unit));
1574 }
1575 
offset_to_bitmap(struct btrfs_free_space_ctl * ctl,u64 offset)1576 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1577 				   u64 offset)
1578 {
1579 	u64 bitmap_start;
1580 	u64 bytes_per_bitmap;
1581 
1582 	bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1583 	bitmap_start = offset - ctl->start;
1584 	bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1585 	bitmap_start *= bytes_per_bitmap;
1586 	bitmap_start += ctl->start;
1587 
1588 	return bitmap_start;
1589 }
1590 
tree_insert_offset(struct btrfs_free_space_ctl * ctl,struct btrfs_free_cluster * cluster,struct btrfs_free_space * new_entry)1591 static int tree_insert_offset(struct btrfs_free_space_ctl *ctl,
1592 			      struct btrfs_free_cluster *cluster,
1593 			      struct btrfs_free_space *new_entry)
1594 {
1595 	struct rb_root *root;
1596 	struct rb_node **p;
1597 	struct rb_node *parent = NULL;
1598 
1599 	lockdep_assert_held(&ctl->tree_lock);
1600 
1601 	if (cluster) {
1602 		lockdep_assert_held(&cluster->lock);
1603 		root = &cluster->root;
1604 	} else {
1605 		root = &ctl->free_space_offset;
1606 	}
1607 
1608 	p = &root->rb_node;
1609 
1610 	while (*p) {
1611 		struct btrfs_free_space *info;
1612 
1613 		parent = *p;
1614 		info = rb_entry(parent, struct btrfs_free_space, offset_index);
1615 
1616 		if (new_entry->offset < info->offset) {
1617 			p = &(*p)->rb_left;
1618 		} else if (new_entry->offset > info->offset) {
1619 			p = &(*p)->rb_right;
1620 		} else {
1621 			/*
1622 			 * we could have a bitmap entry and an extent entry
1623 			 * share the same offset.  If this is the case, we want
1624 			 * the extent entry to always be found first if we do a
1625 			 * linear search through the tree, since we want to have
1626 			 * the quickest allocation time, and allocating from an
1627 			 * extent is faster than allocating from a bitmap.  So
1628 			 * if we're inserting a bitmap and we find an entry at
1629 			 * this offset, we want to go right, or after this entry
1630 			 * logically.  If we are inserting an extent and we've
1631 			 * found a bitmap, we want to go left, or before
1632 			 * logically.
1633 			 */
1634 			if (new_entry->bitmap) {
1635 				if (info->bitmap) {
1636 					WARN_ON_ONCE(1);
1637 					return -EEXIST;
1638 				}
1639 				p = &(*p)->rb_right;
1640 			} else {
1641 				if (!info->bitmap) {
1642 					WARN_ON_ONCE(1);
1643 					return -EEXIST;
1644 				}
1645 				p = &(*p)->rb_left;
1646 			}
1647 		}
1648 	}
1649 
1650 	rb_link_node(&new_entry->offset_index, parent, p);
1651 	rb_insert_color(&new_entry->offset_index, root);
1652 
1653 	return 0;
1654 }
1655 
1656 /*
1657  * This is a little subtle.  We *only* have ->max_extent_size set if we actually
1658  * searched through the bitmap and figured out the largest ->max_extent_size,
1659  * otherwise it's 0.  In the case that it's 0 we don't want to tell the
1660  * allocator the wrong thing, we want to use the actual real max_extent_size
1661  * we've found already if it's larger, or we want to use ->bytes.
1662  *
1663  * This matters because find_free_space() will skip entries who's ->bytes is
1664  * less than the required bytes.  So if we didn't search down this bitmap, we
1665  * may pick some previous entry that has a smaller ->max_extent_size than we
1666  * have.  For example, assume we have two entries, one that has
1667  * ->max_extent_size set to 4K and ->bytes set to 1M.  A second entry hasn't set
1668  * ->max_extent_size yet, has ->bytes set to 8K and it's contiguous.  We will
1669  *  call into find_free_space(), and return with max_extent_size == 4K, because
1670  *  that first bitmap entry had ->max_extent_size set, but the second one did
1671  *  not.  If instead we returned 8K we'd come in searching for 8K, and find the
1672  *  8K contiguous range.
1673  *
1674  *  Consider the other case, we have 2 8K chunks in that second entry and still
1675  *  don't have ->max_extent_size set.  We'll return 16K, and the next time the
1676  *  allocator comes in it'll fully search our second bitmap, and this time it'll
1677  *  get an uptodate value of 8K as the maximum chunk size.  Then we'll get the
1678  *  right allocation the next loop through.
1679  */
get_max_extent_size(const struct btrfs_free_space * entry)1680 static inline u64 get_max_extent_size(const struct btrfs_free_space *entry)
1681 {
1682 	if (entry->bitmap && entry->max_extent_size)
1683 		return entry->max_extent_size;
1684 	return entry->bytes;
1685 }
1686 
1687 /*
1688  * We want the largest entry to be leftmost, so this is inverted from what you'd
1689  * normally expect.
1690  */
entry_less(struct rb_node * node,const struct rb_node * parent)1691 static bool entry_less(struct rb_node *node, const struct rb_node *parent)
1692 {
1693 	const struct btrfs_free_space *entry, *exist;
1694 
1695 	entry = rb_entry(node, struct btrfs_free_space, bytes_index);
1696 	exist = rb_entry(parent, struct btrfs_free_space, bytes_index);
1697 	return get_max_extent_size(exist) < get_max_extent_size(entry);
1698 }
1699 
1700 /*
1701  * searches the tree for the given offset.
1702  *
1703  * fuzzy - If this is set, then we are trying to make an allocation, and we just
1704  * want a section that has at least bytes size and comes at or after the given
1705  * offset.
1706  */
1707 static struct btrfs_free_space *
tree_search_offset(struct btrfs_free_space_ctl * ctl,u64 offset,int bitmap_only,int fuzzy)1708 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1709 		   u64 offset, int bitmap_only, int fuzzy)
1710 {
1711 	struct rb_node *n = ctl->free_space_offset.rb_node;
1712 	struct btrfs_free_space *entry = NULL, *prev = NULL;
1713 
1714 	lockdep_assert_held(&ctl->tree_lock);
1715 
1716 	/* find entry that is closest to the 'offset' */
1717 	while (n) {
1718 		entry = rb_entry(n, struct btrfs_free_space, offset_index);
1719 		prev = entry;
1720 
1721 		if (offset < entry->offset)
1722 			n = n->rb_left;
1723 		else if (offset > entry->offset)
1724 			n = n->rb_right;
1725 		else
1726 			break;
1727 
1728 		entry = NULL;
1729 	}
1730 
1731 	if (bitmap_only) {
1732 		if (!entry)
1733 			return NULL;
1734 		if (entry->bitmap)
1735 			return entry;
1736 
1737 		/*
1738 		 * bitmap entry and extent entry may share same offset,
1739 		 * in that case, bitmap entry comes after extent entry.
1740 		 */
1741 		n = rb_next(n);
1742 		if (!n)
1743 			return NULL;
1744 		entry = rb_entry(n, struct btrfs_free_space, offset_index);
1745 		if (entry->offset != offset)
1746 			return NULL;
1747 
1748 		WARN_ON(!entry->bitmap);
1749 		return entry;
1750 	} else if (entry) {
1751 		if (entry->bitmap) {
1752 			/*
1753 			 * if previous extent entry covers the offset,
1754 			 * we should return it instead of the bitmap entry
1755 			 */
1756 			n = rb_prev(&entry->offset_index);
1757 			if (n) {
1758 				prev = rb_entry(n, struct btrfs_free_space,
1759 						offset_index);
1760 				if (!prev->bitmap &&
1761 				    prev->offset + prev->bytes > offset)
1762 					entry = prev;
1763 			}
1764 		}
1765 		return entry;
1766 	}
1767 
1768 	if (!prev)
1769 		return NULL;
1770 
1771 	/* find last entry before the 'offset' */
1772 	entry = prev;
1773 	if (entry->offset > offset) {
1774 		n = rb_prev(&entry->offset_index);
1775 		if (n) {
1776 			entry = rb_entry(n, struct btrfs_free_space,
1777 					offset_index);
1778 			ASSERT(entry->offset <= offset);
1779 		} else {
1780 			if (fuzzy)
1781 				return entry;
1782 			else
1783 				return NULL;
1784 		}
1785 	}
1786 
1787 	if (entry->bitmap) {
1788 		n = rb_prev(&entry->offset_index);
1789 		if (n) {
1790 			prev = rb_entry(n, struct btrfs_free_space,
1791 					offset_index);
1792 			if (!prev->bitmap &&
1793 			    prev->offset + prev->bytes > offset)
1794 				return prev;
1795 		}
1796 		if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1797 			return entry;
1798 	} else if (entry->offset + entry->bytes > offset)
1799 		return entry;
1800 
1801 	if (!fuzzy)
1802 		return NULL;
1803 
1804 	while (1) {
1805 		n = rb_next(&entry->offset_index);
1806 		if (!n)
1807 			return NULL;
1808 		entry = rb_entry(n, struct btrfs_free_space, offset_index);
1809 		if (entry->bitmap) {
1810 			if (entry->offset + BITS_PER_BITMAP *
1811 			    ctl->unit > offset)
1812 				break;
1813 		} else {
1814 			if (entry->offset + entry->bytes > offset)
1815 				break;
1816 		}
1817 	}
1818 	return entry;
1819 }
1820 
unlink_free_space(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)1821 static inline void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1822 				     struct btrfs_free_space *info,
1823 				     bool update_stat)
1824 {
1825 	lockdep_assert_held(&ctl->tree_lock);
1826 
1827 	rb_erase(&info->offset_index, &ctl->free_space_offset);
1828 	rb_erase_cached(&info->bytes_index, &ctl->free_space_bytes);
1829 	ctl->free_extents--;
1830 
1831 	if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1832 		ctl->discardable_extents[BTRFS_STAT_CURR]--;
1833 		ctl->discardable_bytes[BTRFS_STAT_CURR] -= info->bytes;
1834 	}
1835 
1836 	if (update_stat)
1837 		ctl->free_space -= info->bytes;
1838 }
1839 
link_free_space(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)1840 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1841 			   struct btrfs_free_space *info)
1842 {
1843 	int ret = 0;
1844 
1845 	lockdep_assert_held(&ctl->tree_lock);
1846 
1847 	ASSERT(info->bytes || info->bitmap);
1848 	ret = tree_insert_offset(ctl, NULL, info);
1849 	if (ret)
1850 		return ret;
1851 
1852 	rb_add_cached(&info->bytes_index, &ctl->free_space_bytes, entry_less);
1853 
1854 	if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1855 		ctl->discardable_extents[BTRFS_STAT_CURR]++;
1856 		ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
1857 	}
1858 
1859 	ctl->free_space += info->bytes;
1860 	ctl->free_extents++;
1861 	return ret;
1862 }
1863 
relink_bitmap_entry(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)1864 static void relink_bitmap_entry(struct btrfs_free_space_ctl *ctl,
1865 				struct btrfs_free_space *info)
1866 {
1867 	ASSERT(info->bitmap);
1868 
1869 	/*
1870 	 * If our entry is empty it's because we're on a cluster and we don't
1871 	 * want to re-link it into our ctl bytes index.
1872 	 */
1873 	if (RB_EMPTY_NODE(&info->bytes_index))
1874 		return;
1875 
1876 	lockdep_assert_held(&ctl->tree_lock);
1877 
1878 	rb_erase_cached(&info->bytes_index, &ctl->free_space_bytes);
1879 	rb_add_cached(&info->bytes_index, &ctl->free_space_bytes, entry_less);
1880 }
1881 
bitmap_clear_bits(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset,u64 bytes,bool update_stat)1882 static inline void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1883 				     struct btrfs_free_space *info,
1884 				     u64 offset, u64 bytes, bool update_stat)
1885 {
1886 	unsigned long start, count, end;
1887 	int extent_delta = -1;
1888 
1889 	start = offset_to_bit(info->offset, ctl->unit, offset);
1890 	count = bytes_to_bits(bytes, ctl->unit);
1891 	end = start + count;
1892 	ASSERT(end <= BITS_PER_BITMAP);
1893 
1894 	bitmap_clear(info->bitmap, start, count);
1895 
1896 	info->bytes -= bytes;
1897 	if (info->max_extent_size > ctl->unit)
1898 		info->max_extent_size = 0;
1899 
1900 	relink_bitmap_entry(ctl, info);
1901 
1902 	if (start && test_bit(start - 1, info->bitmap))
1903 		extent_delta++;
1904 
1905 	if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1906 		extent_delta++;
1907 
1908 	info->bitmap_extents += extent_delta;
1909 	if (!btrfs_free_space_trimmed(info)) {
1910 		ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1911 		ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
1912 	}
1913 
1914 	if (update_stat)
1915 		ctl->free_space -= bytes;
1916 }
1917 
btrfs_bitmap_set_bits(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset,u64 bytes)1918 static void btrfs_bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1919 				  struct btrfs_free_space *info, u64 offset,
1920 				  u64 bytes)
1921 {
1922 	unsigned long start, count, end;
1923 	int extent_delta = 1;
1924 
1925 	start = offset_to_bit(info->offset, ctl->unit, offset);
1926 	count = bytes_to_bits(bytes, ctl->unit);
1927 	end = start + count;
1928 	ASSERT(end <= BITS_PER_BITMAP);
1929 
1930 	bitmap_set(info->bitmap, start, count);
1931 
1932 	/*
1933 	 * We set some bytes, we have no idea what the max extent size is
1934 	 * anymore.
1935 	 */
1936 	info->max_extent_size = 0;
1937 	info->bytes += bytes;
1938 	ctl->free_space += bytes;
1939 
1940 	relink_bitmap_entry(ctl, info);
1941 
1942 	if (start && test_bit(start - 1, info->bitmap))
1943 		extent_delta--;
1944 
1945 	if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1946 		extent_delta--;
1947 
1948 	info->bitmap_extents += extent_delta;
1949 	if (!btrfs_free_space_trimmed(info)) {
1950 		ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1951 		ctl->discardable_bytes[BTRFS_STAT_CURR] += bytes;
1952 	}
1953 }
1954 
1955 /*
1956  * If we can not find suitable extent, we will use bytes to record
1957  * the size of the max extent.
1958  */
search_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * bitmap_info,u64 * offset,u64 * bytes,bool for_alloc)1959 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1960 			 struct btrfs_free_space *bitmap_info, u64 *offset,
1961 			 u64 *bytes, bool for_alloc)
1962 {
1963 	unsigned long found_bits = 0;
1964 	unsigned long max_bits = 0;
1965 	unsigned long bits, i;
1966 	unsigned long next_zero;
1967 	unsigned long extent_bits;
1968 
1969 	/*
1970 	 * Skip searching the bitmap if we don't have a contiguous section that
1971 	 * is large enough for this allocation.
1972 	 */
1973 	if (for_alloc &&
1974 	    bitmap_info->max_extent_size &&
1975 	    bitmap_info->max_extent_size < *bytes) {
1976 		*bytes = bitmap_info->max_extent_size;
1977 		return -1;
1978 	}
1979 
1980 	i = offset_to_bit(bitmap_info->offset, ctl->unit,
1981 			  max_t(u64, *offset, bitmap_info->offset));
1982 	bits = bytes_to_bits(*bytes, ctl->unit);
1983 
1984 	for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1985 		if (for_alloc && bits == 1) {
1986 			found_bits = 1;
1987 			break;
1988 		}
1989 		next_zero = find_next_zero_bit(bitmap_info->bitmap,
1990 					       BITS_PER_BITMAP, i);
1991 		extent_bits = next_zero - i;
1992 		if (extent_bits >= bits) {
1993 			found_bits = extent_bits;
1994 			break;
1995 		} else if (extent_bits > max_bits) {
1996 			max_bits = extent_bits;
1997 		}
1998 		i = next_zero;
1999 	}
2000 
2001 	if (found_bits) {
2002 		*offset = (u64)(i * ctl->unit) + bitmap_info->offset;
2003 		*bytes = (u64)(found_bits) * ctl->unit;
2004 		return 0;
2005 	}
2006 
2007 	*bytes = (u64)(max_bits) * ctl->unit;
2008 	bitmap_info->max_extent_size = *bytes;
2009 	relink_bitmap_entry(ctl, bitmap_info);
2010 	return -1;
2011 }
2012 
2013 /* Cache the size of the max extent in bytes */
2014 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)2015 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
2016 		unsigned long align, u64 *max_extent_size, bool use_bytes_index)
2017 {
2018 	struct btrfs_free_space *entry;
2019 	struct rb_node *node;
2020 	u64 tmp;
2021 	u64 align_off;
2022 	int ret;
2023 
2024 	if (!ctl->free_space_offset.rb_node)
2025 		goto out;
2026 again:
2027 	if (use_bytes_index) {
2028 		node = rb_first_cached(&ctl->free_space_bytes);
2029 	} else {
2030 		entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset),
2031 					   0, 1);
2032 		if (!entry)
2033 			goto out;
2034 		node = &entry->offset_index;
2035 	}
2036 
2037 	for (; node; node = rb_next(node)) {
2038 		if (use_bytes_index)
2039 			entry = rb_entry(node, struct btrfs_free_space,
2040 					 bytes_index);
2041 		else
2042 			entry = rb_entry(node, struct btrfs_free_space,
2043 					 offset_index);
2044 
2045 		/*
2046 		 * If we are using the bytes index then all subsequent entries
2047 		 * in this tree are going to be < bytes, so simply set the max
2048 		 * extent size and exit the loop.
2049 		 *
2050 		 * If we're using the offset index then we need to keep going
2051 		 * through the rest of the tree.
2052 		 */
2053 		if (entry->bytes < *bytes) {
2054 			*max_extent_size = max(get_max_extent_size(entry),
2055 					       *max_extent_size);
2056 			if (use_bytes_index)
2057 				break;
2058 			continue;
2059 		}
2060 
2061 		/* make sure the space returned is big enough
2062 		 * to match our requested alignment
2063 		 */
2064 		if (*bytes >= align) {
2065 			tmp = entry->offset - ctl->start + align - 1;
2066 			tmp = div64_u64(tmp, align);
2067 			tmp = tmp * align + ctl->start;
2068 			align_off = tmp - entry->offset;
2069 		} else {
2070 			align_off = 0;
2071 			tmp = entry->offset;
2072 		}
2073 
2074 		/*
2075 		 * We don't break here if we're using the bytes index because we
2076 		 * may have another entry that has the correct alignment that is
2077 		 * the right size, so we don't want to miss that possibility.
2078 		 * At worst this adds another loop through the logic, but if we
2079 		 * broke here we could prematurely ENOSPC.
2080 		 */
2081 		if (entry->bytes < *bytes + align_off) {
2082 			*max_extent_size = max(get_max_extent_size(entry),
2083 					       *max_extent_size);
2084 			continue;
2085 		}
2086 
2087 		if (entry->bitmap) {
2088 			struct rb_node *old_next = rb_next(node);
2089 			u64 size = *bytes;
2090 
2091 			ret = search_bitmap(ctl, entry, &tmp, &size, true);
2092 			if (!ret) {
2093 				*offset = tmp;
2094 				*bytes = size;
2095 				return entry;
2096 			} else {
2097 				*max_extent_size =
2098 					max(get_max_extent_size(entry),
2099 					    *max_extent_size);
2100 			}
2101 
2102 			/*
2103 			 * The bitmap may have gotten re-arranged in the space
2104 			 * index here because the max_extent_size may have been
2105 			 * updated.  Start from the beginning again if this
2106 			 * happened.
2107 			 */
2108 			if (use_bytes_index && old_next != rb_next(node))
2109 				goto again;
2110 			continue;
2111 		}
2112 
2113 		*offset = tmp;
2114 		*bytes = entry->bytes - align_off;
2115 		return entry;
2116 	}
2117 out:
2118 	return NULL;
2119 }
2120 
add_new_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset)2121 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
2122 			   struct btrfs_free_space *info, u64 offset)
2123 {
2124 	info->offset = offset_to_bitmap(ctl, offset);
2125 	info->bytes = 0;
2126 	info->bitmap_extents = 0;
2127 	INIT_LIST_HEAD(&info->list);
2128 	link_free_space(ctl, info);
2129 	ctl->total_bitmaps++;
2130 	recalculate_thresholds(ctl);
2131 }
2132 
free_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * bitmap_info)2133 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
2134 			struct btrfs_free_space *bitmap_info)
2135 {
2136 	/*
2137 	 * Normally when this is called, the bitmap is completely empty. However,
2138 	 * if we are blowing up the free space cache for one reason or another
2139 	 * via __btrfs_remove_free_space_cache(), then it may not be freed and
2140 	 * we may leave stats on the table.
2141 	 */
2142 	if (bitmap_info->bytes && !btrfs_free_space_trimmed(bitmap_info)) {
2143 		ctl->discardable_extents[BTRFS_STAT_CURR] -=
2144 			bitmap_info->bitmap_extents;
2145 		ctl->discardable_bytes[BTRFS_STAT_CURR] -= bitmap_info->bytes;
2146 
2147 	}
2148 	unlink_free_space(ctl, bitmap_info, true);
2149 	kmem_cache_free(btrfs_free_space_bitmap_cachep, bitmap_info->bitmap);
2150 	kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
2151 	ctl->total_bitmaps--;
2152 	recalculate_thresholds(ctl);
2153 }
2154 
remove_from_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * bitmap_info,u64 * offset,u64 * bytes)2155 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
2156 			      struct btrfs_free_space *bitmap_info,
2157 			      u64 *offset, u64 *bytes)
2158 {
2159 	u64 end;
2160 	u64 search_start, search_bytes;
2161 	int ret;
2162 
2163 again:
2164 	end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
2165 
2166 	/*
2167 	 * We need to search for bits in this bitmap.  We could only cover some
2168 	 * of the extent in this bitmap thanks to how we add space, so we need
2169 	 * to search for as much as it as we can and clear that amount, and then
2170 	 * go searching for the next bit.
2171 	 */
2172 	search_start = *offset;
2173 	search_bytes = ctl->unit;
2174 	search_bytes = min(search_bytes, end - search_start + 1);
2175 	ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes,
2176 			    false);
2177 	if (ret < 0 || search_start != *offset)
2178 		return -EINVAL;
2179 
2180 	/* We may have found more bits than what we need */
2181 	search_bytes = min(search_bytes, *bytes);
2182 
2183 	/* Cannot clear past the end of the bitmap */
2184 	search_bytes = min(search_bytes, end - search_start + 1);
2185 
2186 	bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes, true);
2187 	*offset += search_bytes;
2188 	*bytes -= search_bytes;
2189 
2190 	if (*bytes) {
2191 		struct rb_node *next = rb_next(&bitmap_info->offset_index);
2192 		if (!bitmap_info->bytes)
2193 			free_bitmap(ctl, bitmap_info);
2194 
2195 		/*
2196 		 * no entry after this bitmap, but we still have bytes to
2197 		 * remove, so something has gone wrong.
2198 		 */
2199 		if (!next)
2200 			return -EINVAL;
2201 
2202 		bitmap_info = rb_entry(next, struct btrfs_free_space,
2203 				       offset_index);
2204 
2205 		/*
2206 		 * if the next entry isn't a bitmap we need to return to let the
2207 		 * extent stuff do its work.
2208 		 */
2209 		if (!bitmap_info->bitmap)
2210 			return -EAGAIN;
2211 
2212 		/*
2213 		 * Ok the next item is a bitmap, but it may not actually hold
2214 		 * the information for the rest of this free space stuff, so
2215 		 * look for it, and if we don't find it return so we can try
2216 		 * everything over again.
2217 		 */
2218 		search_start = *offset;
2219 		search_bytes = ctl->unit;
2220 		ret = search_bitmap(ctl, bitmap_info, &search_start,
2221 				    &search_bytes, false);
2222 		if (ret < 0 || search_start != *offset)
2223 			return -EAGAIN;
2224 
2225 		goto again;
2226 	} else if (!bitmap_info->bytes)
2227 		free_bitmap(ctl, bitmap_info);
2228 
2229 	return 0;
2230 }
2231 
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)2232 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
2233 			       struct btrfs_free_space *info, u64 offset,
2234 			       u64 bytes, enum btrfs_trim_state trim_state)
2235 {
2236 	u64 bytes_to_set = 0;
2237 	u64 end;
2238 
2239 	/*
2240 	 * This is a tradeoff to make bitmap trim state minimal.  We mark the
2241 	 * whole bitmap untrimmed if at any point we add untrimmed regions.
2242 	 */
2243 	if (trim_state == BTRFS_TRIM_STATE_UNTRIMMED) {
2244 		if (btrfs_free_space_trimmed(info)) {
2245 			ctl->discardable_extents[BTRFS_STAT_CURR] +=
2246 				info->bitmap_extents;
2247 			ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
2248 		}
2249 		info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2250 	}
2251 
2252 	end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
2253 
2254 	bytes_to_set = min(end - offset, bytes);
2255 
2256 	btrfs_bitmap_set_bits(ctl, info, offset, bytes_to_set);
2257 
2258 	return bytes_to_set;
2259 
2260 }
2261 
use_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)2262 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
2263 		      struct btrfs_free_space *info)
2264 {
2265 	struct btrfs_block_group *block_group = ctl->block_group;
2266 	struct btrfs_fs_info *fs_info = block_group->fs_info;
2267 	bool forced = false;
2268 
2269 #ifdef CONFIG_BTRFS_DEBUG
2270 	if (btrfs_should_fragment_free_space(block_group))
2271 		forced = true;
2272 #endif
2273 
2274 	/* This is a way to reclaim large regions from the bitmaps. */
2275 	if (!forced && info->bytes >= FORCE_EXTENT_THRESHOLD)
2276 		return false;
2277 
2278 	/*
2279 	 * If we are below the extents threshold then we can add this as an
2280 	 * extent, and don't have to deal with the bitmap
2281 	 */
2282 	if (!forced && ctl->free_extents < ctl->extents_thresh) {
2283 		/*
2284 		 * If this block group has some small extents we don't want to
2285 		 * use up all of our free slots in the cache with them, we want
2286 		 * to reserve them to larger extents, however if we have plenty
2287 		 * of cache left then go ahead an dadd them, no sense in adding
2288 		 * the overhead of a bitmap if we don't have to.
2289 		 */
2290 		if (info->bytes <= fs_info->sectorsize * 8) {
2291 			if (ctl->free_extents * 3 <= ctl->extents_thresh)
2292 				return false;
2293 		} else {
2294 			return false;
2295 		}
2296 	}
2297 
2298 	/*
2299 	 * The original block groups from mkfs can be really small, like 8
2300 	 * megabytes, so don't bother with a bitmap for those entries.  However
2301 	 * some block groups can be smaller than what a bitmap would cover but
2302 	 * are still large enough that they could overflow the 32k memory limit,
2303 	 * so allow those block groups to still be allowed to have a bitmap
2304 	 * entry.
2305 	 */
2306 	if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->length)
2307 		return false;
2308 
2309 	return true;
2310 }
2311 
2312 static const struct btrfs_free_space_op free_space_op = {
2313 	.use_bitmap		= use_bitmap,
2314 };
2315 
insert_into_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)2316 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
2317 			      struct btrfs_free_space *info)
2318 {
2319 	struct btrfs_free_space *bitmap_info;
2320 	struct btrfs_block_group *block_group = NULL;
2321 	int added = 0;
2322 	u64 bytes, offset, bytes_added;
2323 	enum btrfs_trim_state trim_state;
2324 	int ret;
2325 
2326 	bytes = info->bytes;
2327 	offset = info->offset;
2328 	trim_state = info->trim_state;
2329 
2330 	if (!ctl->op->use_bitmap(ctl, info))
2331 		return 0;
2332 
2333 	if (ctl->op == &free_space_op)
2334 		block_group = ctl->block_group;
2335 again:
2336 	/*
2337 	 * Since we link bitmaps right into the cluster we need to see if we
2338 	 * have a cluster here, and if so and it has our bitmap we need to add
2339 	 * the free space to that bitmap.
2340 	 */
2341 	if (block_group && !list_empty(&block_group->cluster_list)) {
2342 		struct btrfs_free_cluster *cluster;
2343 		struct rb_node *node;
2344 		struct btrfs_free_space *entry;
2345 
2346 		cluster = list_entry(block_group->cluster_list.next,
2347 				     struct btrfs_free_cluster,
2348 				     block_group_list);
2349 		spin_lock(&cluster->lock);
2350 		node = rb_first(&cluster->root);
2351 		if (!node) {
2352 			spin_unlock(&cluster->lock);
2353 			goto no_cluster_bitmap;
2354 		}
2355 
2356 		entry = rb_entry(node, struct btrfs_free_space, offset_index);
2357 		if (!entry->bitmap) {
2358 			spin_unlock(&cluster->lock);
2359 			goto no_cluster_bitmap;
2360 		}
2361 
2362 		if (entry->offset == offset_to_bitmap(ctl, offset)) {
2363 			bytes_added = add_bytes_to_bitmap(ctl, entry, offset,
2364 							  bytes, trim_state);
2365 			bytes -= bytes_added;
2366 			offset += bytes_added;
2367 		}
2368 		spin_unlock(&cluster->lock);
2369 		if (!bytes) {
2370 			ret = 1;
2371 			goto out;
2372 		}
2373 	}
2374 
2375 no_cluster_bitmap:
2376 	bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2377 					 1, 0);
2378 	if (!bitmap_info) {
2379 		ASSERT(added == 0);
2380 		goto new_bitmap;
2381 	}
2382 
2383 	bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
2384 					  trim_state);
2385 	bytes -= bytes_added;
2386 	offset += bytes_added;
2387 	added = 0;
2388 
2389 	if (!bytes) {
2390 		ret = 1;
2391 		goto out;
2392 	} else
2393 		goto again;
2394 
2395 new_bitmap:
2396 	if (info && info->bitmap) {
2397 		add_new_bitmap(ctl, info, offset);
2398 		added = 1;
2399 		info = NULL;
2400 		goto again;
2401 	} else {
2402 		spin_unlock(&ctl->tree_lock);
2403 
2404 		/* no pre-allocated info, allocate a new one */
2405 		if (!info) {
2406 			info = kmem_cache_zalloc(btrfs_free_space_cachep,
2407 						 GFP_NOFS);
2408 			if (!info) {
2409 				spin_lock(&ctl->tree_lock);
2410 				ret = -ENOMEM;
2411 				goto out;
2412 			}
2413 		}
2414 
2415 		/* allocate the bitmap */
2416 		info->bitmap = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep,
2417 						 GFP_NOFS);
2418 		info->trim_state = BTRFS_TRIM_STATE_TRIMMED;
2419 		spin_lock(&ctl->tree_lock);
2420 		if (!info->bitmap) {
2421 			ret = -ENOMEM;
2422 			goto out;
2423 		}
2424 		goto again;
2425 	}
2426 
2427 out:
2428 	if (info) {
2429 		if (info->bitmap)
2430 			kmem_cache_free(btrfs_free_space_bitmap_cachep,
2431 					info->bitmap);
2432 		kmem_cache_free(btrfs_free_space_cachep, info);
2433 	}
2434 
2435 	return ret;
2436 }
2437 
2438 /*
2439  * Free space merging rules:
2440  *  1) Merge trimmed areas together
2441  *  2) Let untrimmed areas coalesce with trimmed areas
2442  *  3) Always pull neighboring regions from bitmaps
2443  *
2444  * The above rules are for when we merge free space based on btrfs_trim_state.
2445  * Rules 2 and 3 are subtle because they are suboptimal, but are done for the
2446  * same reason: to promote larger extent regions which makes life easier for
2447  * find_free_extent().  Rule 2 enables coalescing based on the common path
2448  * being returning free space from btrfs_finish_extent_commit().  So when free
2449  * space is trimmed, it will prevent aggregating trimmed new region and
2450  * untrimmed regions in the rb_tree.  Rule 3 is purely to obtain larger extents
2451  * and provide find_free_extent() with the largest extents possible hoping for
2452  * the reuse path.
2453  */
try_merge_free_space(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2454 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2455 			  struct btrfs_free_space *info, bool update_stat)
2456 {
2457 	struct btrfs_free_space *left_info = NULL;
2458 	struct btrfs_free_space *right_info;
2459 	bool merged = false;
2460 	u64 offset = info->offset;
2461 	u64 bytes = info->bytes;
2462 	const bool is_trimmed = btrfs_free_space_trimmed(info);
2463 	struct rb_node *right_prev = NULL;
2464 
2465 	/*
2466 	 * first we want to see if there is free space adjacent to the range we
2467 	 * are adding, if there is remove that struct and add a new one to
2468 	 * cover the entire range
2469 	 */
2470 	right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
2471 	if (right_info)
2472 		right_prev = rb_prev(&right_info->offset_index);
2473 
2474 	if (right_prev)
2475 		left_info = rb_entry(right_prev, struct btrfs_free_space, offset_index);
2476 	else if (!right_info)
2477 		left_info = tree_search_offset(ctl, offset - 1, 0, 0);
2478 
2479 	/* See try_merge_free_space() comment. */
2480 	if (right_info && !right_info->bitmap &&
2481 	    (!is_trimmed || btrfs_free_space_trimmed(right_info))) {
2482 		unlink_free_space(ctl, right_info, update_stat);
2483 		info->bytes += right_info->bytes;
2484 		kmem_cache_free(btrfs_free_space_cachep, right_info);
2485 		merged = true;
2486 	}
2487 
2488 	/* See try_merge_free_space() comment. */
2489 	if (left_info && !left_info->bitmap &&
2490 	    left_info->offset + left_info->bytes == offset &&
2491 	    (!is_trimmed || btrfs_free_space_trimmed(left_info))) {
2492 		unlink_free_space(ctl, left_info, update_stat);
2493 		info->offset = left_info->offset;
2494 		info->bytes += left_info->bytes;
2495 		kmem_cache_free(btrfs_free_space_cachep, left_info);
2496 		merged = true;
2497 	}
2498 
2499 	return merged;
2500 }
2501 
steal_from_bitmap_to_end(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2502 static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2503 				     struct btrfs_free_space *info,
2504 				     bool update_stat)
2505 {
2506 	struct btrfs_free_space *bitmap;
2507 	unsigned long i;
2508 	unsigned long j;
2509 	const u64 end = info->offset + info->bytes;
2510 	const u64 bitmap_offset = offset_to_bitmap(ctl, end);
2511 	u64 bytes;
2512 
2513 	bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2514 	if (!bitmap)
2515 		return false;
2516 
2517 	i = offset_to_bit(bitmap->offset, ctl->unit, end);
2518 	j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
2519 	if (j == i)
2520 		return false;
2521 	bytes = (j - i) * ctl->unit;
2522 	info->bytes += bytes;
2523 
2524 	/* See try_merge_free_space() comment. */
2525 	if (!btrfs_free_space_trimmed(bitmap))
2526 		info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2527 
2528 	bitmap_clear_bits(ctl, bitmap, end, bytes, update_stat);
2529 
2530 	if (!bitmap->bytes)
2531 		free_bitmap(ctl, bitmap);
2532 
2533 	return true;
2534 }
2535 
steal_from_bitmap_to_front(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2536 static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2537 				       struct btrfs_free_space *info,
2538 				       bool update_stat)
2539 {
2540 	struct btrfs_free_space *bitmap;
2541 	u64 bitmap_offset;
2542 	unsigned long i;
2543 	unsigned long j;
2544 	unsigned long prev_j;
2545 	u64 bytes;
2546 
2547 	bitmap_offset = offset_to_bitmap(ctl, info->offset);
2548 	/* If we're on a boundary, try the previous logical bitmap. */
2549 	if (bitmap_offset == info->offset) {
2550 		if (info->offset == 0)
2551 			return false;
2552 		bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
2553 	}
2554 
2555 	bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2556 	if (!bitmap)
2557 		return false;
2558 
2559 	i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
2560 	j = 0;
2561 	prev_j = (unsigned long)-1;
2562 	for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2563 		if (j > i)
2564 			break;
2565 		prev_j = j;
2566 	}
2567 	if (prev_j == i)
2568 		return false;
2569 
2570 	if (prev_j == (unsigned long)-1)
2571 		bytes = (i + 1) * ctl->unit;
2572 	else
2573 		bytes = (i - prev_j) * ctl->unit;
2574 
2575 	info->offset -= bytes;
2576 	info->bytes += bytes;
2577 
2578 	/* See try_merge_free_space() comment. */
2579 	if (!btrfs_free_space_trimmed(bitmap))
2580 		info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2581 
2582 	bitmap_clear_bits(ctl, bitmap, info->offset, bytes, update_stat);
2583 
2584 	if (!bitmap->bytes)
2585 		free_bitmap(ctl, bitmap);
2586 
2587 	return true;
2588 }
2589 
2590 /*
2591  * We prefer always to allocate from extent entries, both for clustered and
2592  * non-clustered allocation requests. So when attempting to add a new extent
2593  * entry, try to see if there's adjacent free space in bitmap entries, and if
2594  * there is, migrate that space from the bitmaps to the extent.
2595  * Like this we get better chances of satisfying space allocation requests
2596  * because we attempt to satisfy them based on a single cache entry, and never
2597  * on 2 or more entries - even if the entries represent a contiguous free space
2598  * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2599  * ends).
2600  */
steal_from_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2601 static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2602 			      struct btrfs_free_space *info,
2603 			      bool update_stat)
2604 {
2605 	/*
2606 	 * Only work with disconnected entries, as we can change their offset,
2607 	 * and must be extent entries.
2608 	 */
2609 	ASSERT(!info->bitmap);
2610 	ASSERT(RB_EMPTY_NODE(&info->offset_index));
2611 
2612 	if (ctl->total_bitmaps > 0) {
2613 		bool stole_end;
2614 		bool stole_front = false;
2615 
2616 		stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2617 		if (ctl->total_bitmaps > 0)
2618 			stole_front = steal_from_bitmap_to_front(ctl, info,
2619 								 update_stat);
2620 
2621 		if (stole_end || stole_front)
2622 			try_merge_free_space(ctl, info, update_stat);
2623 	}
2624 }
2625 
__btrfs_add_free_space(struct btrfs_block_group * block_group,u64 offset,u64 bytes,enum btrfs_trim_state trim_state)2626 static int __btrfs_add_free_space(struct btrfs_block_group *block_group,
2627 			   u64 offset, u64 bytes,
2628 			   enum btrfs_trim_state trim_state)
2629 {
2630 	struct btrfs_fs_info *fs_info = block_group->fs_info;
2631 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2632 	struct btrfs_free_space *info;
2633 	int ret = 0;
2634 	u64 filter_bytes = bytes;
2635 
2636 	ASSERT(!btrfs_is_zoned(fs_info));
2637 
2638 	info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2639 	if (!info)
2640 		return -ENOMEM;
2641 
2642 	info->offset = offset;
2643 	info->bytes = bytes;
2644 	info->trim_state = trim_state;
2645 	RB_CLEAR_NODE(&info->offset_index);
2646 	RB_CLEAR_NODE(&info->bytes_index);
2647 
2648 	spin_lock(&ctl->tree_lock);
2649 
2650 	if (try_merge_free_space(ctl, info, true))
2651 		goto link;
2652 
2653 	/*
2654 	 * There was no extent directly to the left or right of this new
2655 	 * extent then we know we're going to have to allocate a new extent, so
2656 	 * before we do that see if we need to drop this into a bitmap
2657 	 */
2658 	ret = insert_into_bitmap(ctl, info);
2659 	if (ret < 0) {
2660 		goto out;
2661 	} else if (ret) {
2662 		ret = 0;
2663 		goto out;
2664 	}
2665 link:
2666 	/*
2667 	 * Only steal free space from adjacent bitmaps if we're sure we're not
2668 	 * going to add the new free space to existing bitmap entries - because
2669 	 * that would mean unnecessary work that would be reverted. Therefore
2670 	 * attempt to steal space from bitmaps if we're adding an extent entry.
2671 	 */
2672 	steal_from_bitmap(ctl, info, true);
2673 
2674 	filter_bytes = max(filter_bytes, info->bytes);
2675 
2676 	ret = link_free_space(ctl, info);
2677 	if (ret)
2678 		kmem_cache_free(btrfs_free_space_cachep, info);
2679 out:
2680 	btrfs_discard_update_discardable(block_group);
2681 	spin_unlock(&ctl->tree_lock);
2682 
2683 	if (ret) {
2684 		btrfs_crit(fs_info, "unable to add free space :%d", ret);
2685 		ASSERT(ret != -EEXIST);
2686 	}
2687 
2688 	if (trim_state != BTRFS_TRIM_STATE_TRIMMED) {
2689 		btrfs_discard_check_filter(block_group, filter_bytes);
2690 		btrfs_discard_queue_work(&fs_info->discard_ctl, block_group);
2691 	}
2692 
2693 	return ret;
2694 }
2695 
__btrfs_add_free_space_zoned(struct btrfs_block_group * block_group,u64 bytenr,u64 size,bool used)2696 static int __btrfs_add_free_space_zoned(struct btrfs_block_group *block_group,
2697 					u64 bytenr, u64 size, bool used)
2698 {
2699 	struct btrfs_space_info *sinfo = block_group->space_info;
2700 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2701 	u64 offset = bytenr - block_group->start;
2702 	u64 to_free, to_unusable;
2703 	int bg_reclaim_threshold = 0;
2704 	bool initial;
2705 	u64 reclaimable_unusable;
2706 
2707 	spin_lock(&block_group->lock);
2708 
2709 	initial = ((size == block_group->length) && (block_group->alloc_offset == 0));
2710 	WARN_ON(!initial && offset + size > block_group->zone_capacity);
2711 	if (!initial)
2712 		bg_reclaim_threshold = READ_ONCE(sinfo->bg_reclaim_threshold);
2713 
2714 	if (!used)
2715 		to_free = size;
2716 	else if (initial)
2717 		to_free = block_group->zone_capacity;
2718 	else if (offset >= block_group->alloc_offset)
2719 		to_free = size;
2720 	else if (offset + size <= block_group->alloc_offset)
2721 		to_free = 0;
2722 	else
2723 		to_free = offset + size - block_group->alloc_offset;
2724 	to_unusable = size - to_free;
2725 
2726 	spin_lock(&ctl->tree_lock);
2727 	ctl->free_space += to_free;
2728 	spin_unlock(&ctl->tree_lock);
2729 	/*
2730 	 * If the block group is read-only, we should account freed space into
2731 	 * bytes_readonly.
2732 	 */
2733 	if (!block_group->ro) {
2734 		block_group->zone_unusable += to_unusable;
2735 		WARN_ON(block_group->zone_unusable > block_group->length);
2736 	}
2737 	if (!used) {
2738 		block_group->alloc_offset -= size;
2739 	}
2740 
2741 	reclaimable_unusable = block_group->zone_unusable -
2742 			       (block_group->length - block_group->zone_capacity);
2743 	/* All the region is now unusable. Mark it as unused and reclaim */
2744 	if (block_group->zone_unusable == block_group->length) {
2745 		btrfs_mark_bg_unused(block_group);
2746 	} else if (bg_reclaim_threshold &&
2747 		   reclaimable_unusable >=
2748 		   mult_perc(block_group->zone_capacity, bg_reclaim_threshold)) {
2749 		btrfs_mark_bg_to_reclaim(block_group);
2750 	}
2751 
2752 	spin_unlock(&block_group->lock);
2753 
2754 	return 0;
2755 }
2756 
btrfs_add_free_space(struct btrfs_block_group * block_group,u64 bytenr,u64 size)2757 int btrfs_add_free_space(struct btrfs_block_group *block_group,
2758 			 u64 bytenr, u64 size)
2759 {
2760 	enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2761 
2762 	if (btrfs_is_zoned(block_group->fs_info))
2763 		return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2764 						    true);
2765 
2766 	if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC))
2767 		trim_state = BTRFS_TRIM_STATE_TRIMMED;
2768 
2769 	return __btrfs_add_free_space(block_group, bytenr, size, trim_state);
2770 }
2771 
btrfs_add_free_space_unused(struct btrfs_block_group * block_group,u64 bytenr,u64 size)2772 int btrfs_add_free_space_unused(struct btrfs_block_group *block_group,
2773 				u64 bytenr, u64 size)
2774 {
2775 	if (btrfs_is_zoned(block_group->fs_info))
2776 		return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2777 						    false);
2778 
2779 	return btrfs_add_free_space(block_group, bytenr, size);
2780 }
2781 
2782 /*
2783  * This is a subtle distinction because when adding free space back in general,
2784  * we want it to be added as untrimmed for async. But in the case where we add
2785  * it on loading of a block group, we want to consider it trimmed.
2786  */
btrfs_add_free_space_async_trimmed(struct btrfs_block_group * block_group,u64 bytenr,u64 size)2787 int btrfs_add_free_space_async_trimmed(struct btrfs_block_group *block_group,
2788 				       u64 bytenr, u64 size)
2789 {
2790 	enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2791 
2792 	if (btrfs_is_zoned(block_group->fs_info))
2793 		return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2794 						    true);
2795 
2796 	if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC) ||
2797 	    btrfs_test_opt(block_group->fs_info, DISCARD_ASYNC))
2798 		trim_state = BTRFS_TRIM_STATE_TRIMMED;
2799 
2800 	return __btrfs_add_free_space(block_group, bytenr, size, trim_state);
2801 }
2802 
btrfs_remove_free_space(struct btrfs_block_group * block_group,u64 offset,u64 bytes)2803 int btrfs_remove_free_space(struct btrfs_block_group *block_group,
2804 			    u64 offset, u64 bytes)
2805 {
2806 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2807 	struct btrfs_free_space *info;
2808 	int ret;
2809 	bool re_search = false;
2810 
2811 	if (btrfs_is_zoned(block_group->fs_info)) {
2812 		/*
2813 		 * This can happen with conventional zones when replaying log.
2814 		 * Since the allocation info of tree-log nodes are not recorded
2815 		 * to the extent-tree, calculate_alloc_pointer() failed to
2816 		 * advance the allocation pointer after last allocated tree log
2817 		 * node blocks.
2818 		 *
2819 		 * This function is called from
2820 		 * btrfs_pin_extent_for_log_replay() when replaying the log.
2821 		 * Advance the pointer not to overwrite the tree-log nodes.
2822 		 */
2823 		if (block_group->start + block_group->alloc_offset <
2824 		    offset + bytes) {
2825 			block_group->alloc_offset =
2826 				offset + bytes - block_group->start;
2827 		}
2828 		return 0;
2829 	}
2830 
2831 	spin_lock(&ctl->tree_lock);
2832 
2833 again:
2834 	ret = 0;
2835 	if (!bytes)
2836 		goto out_lock;
2837 
2838 	info = tree_search_offset(ctl, offset, 0, 0);
2839 	if (!info) {
2840 		/*
2841 		 * oops didn't find an extent that matched the space we wanted
2842 		 * to remove, look for a bitmap instead
2843 		 */
2844 		info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2845 					  1, 0);
2846 		if (!info) {
2847 			/*
2848 			 * If we found a partial bit of our free space in a
2849 			 * bitmap but then couldn't find the other part this may
2850 			 * be a problem, so WARN about it.
2851 			 */
2852 			WARN_ON(re_search);
2853 			goto out_lock;
2854 		}
2855 	}
2856 
2857 	re_search = false;
2858 	if (!info->bitmap) {
2859 		unlink_free_space(ctl, info, true);
2860 		if (offset == info->offset) {
2861 			u64 to_free = min(bytes, info->bytes);
2862 
2863 			info->bytes -= to_free;
2864 			info->offset += to_free;
2865 			if (info->bytes) {
2866 				ret = link_free_space(ctl, info);
2867 				WARN_ON(ret);
2868 			} else {
2869 				kmem_cache_free(btrfs_free_space_cachep, info);
2870 			}
2871 
2872 			offset += to_free;
2873 			bytes -= to_free;
2874 			goto again;
2875 		} else {
2876 			u64 old_end = info->bytes + info->offset;
2877 
2878 			info->bytes = offset - info->offset;
2879 			ret = link_free_space(ctl, info);
2880 			WARN_ON(ret);
2881 			if (ret)
2882 				goto out_lock;
2883 
2884 			/* Not enough bytes in this entry to satisfy us */
2885 			if (old_end < offset + bytes) {
2886 				bytes -= old_end - offset;
2887 				offset = old_end;
2888 				goto again;
2889 			} else if (old_end == offset + bytes) {
2890 				/* all done */
2891 				goto out_lock;
2892 			}
2893 			spin_unlock(&ctl->tree_lock);
2894 
2895 			ret = __btrfs_add_free_space(block_group,
2896 						     offset + bytes,
2897 						     old_end - (offset + bytes),
2898 						     info->trim_state);
2899 			WARN_ON(ret);
2900 			goto out;
2901 		}
2902 	}
2903 
2904 	ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2905 	if (ret == -EAGAIN) {
2906 		re_search = true;
2907 		goto again;
2908 	}
2909 out_lock:
2910 	btrfs_discard_update_discardable(block_group);
2911 	spin_unlock(&ctl->tree_lock);
2912 out:
2913 	return ret;
2914 }
2915 
btrfs_dump_free_space(struct btrfs_block_group * block_group,u64 bytes)2916 void btrfs_dump_free_space(struct btrfs_block_group *block_group,
2917 			   u64 bytes)
2918 {
2919 	struct btrfs_fs_info *fs_info = block_group->fs_info;
2920 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2921 	struct btrfs_free_space *info;
2922 	struct rb_node *n;
2923 	int count = 0;
2924 
2925 	/*
2926 	 * Zoned btrfs does not use free space tree and cluster. Just print
2927 	 * out the free space after the allocation offset.
2928 	 */
2929 	if (btrfs_is_zoned(fs_info)) {
2930 		btrfs_info(fs_info, "free space %llu active %d",
2931 			   block_group->zone_capacity - block_group->alloc_offset,
2932 			   test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
2933 				    &block_group->runtime_flags));
2934 		return;
2935 	}
2936 
2937 	spin_lock(&ctl->tree_lock);
2938 	for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2939 		info = rb_entry(n, struct btrfs_free_space, offset_index);
2940 		if (info->bytes >= bytes && !block_group->ro)
2941 			count++;
2942 		btrfs_crit(fs_info, "entry offset %llu, bytes %llu, bitmap %s",
2943 			   info->offset, info->bytes, str_yes_no(info->bitmap));
2944 	}
2945 	spin_unlock(&ctl->tree_lock);
2946 	btrfs_info(fs_info, "block group has cluster?: %s",
2947 	       str_no_yes(list_empty(&block_group->cluster_list)));
2948 	btrfs_info(fs_info,
2949 		   "%d free space entries at or bigger than %llu bytes",
2950 		   count, bytes);
2951 }
2952 
btrfs_init_free_space_ctl(struct btrfs_block_group * block_group,struct btrfs_free_space_ctl * ctl)2953 void btrfs_init_free_space_ctl(struct btrfs_block_group *block_group,
2954 			       struct btrfs_free_space_ctl *ctl)
2955 {
2956 	struct btrfs_fs_info *fs_info = block_group->fs_info;
2957 
2958 	spin_lock_init(&ctl->tree_lock);
2959 	ctl->unit = fs_info->sectorsize;
2960 	ctl->start = block_group->start;
2961 	ctl->block_group = block_group;
2962 	ctl->op = &free_space_op;
2963 	ctl->free_space_bytes = RB_ROOT_CACHED;
2964 	INIT_LIST_HEAD(&ctl->trimming_ranges);
2965 	mutex_init(&ctl->cache_writeout_mutex);
2966 
2967 	/*
2968 	 * we only want to have 32k of ram per block group for keeping
2969 	 * track of free space, and if we pass 1/2 of that we want to
2970 	 * start converting things over to using bitmaps
2971 	 */
2972 	ctl->extents_thresh = (SZ_32K / 2) / sizeof(struct btrfs_free_space);
2973 }
2974 
2975 /*
2976  * for a given cluster, put all of its extents back into the free
2977  * space cache.  If the block group passed doesn't match the block group
2978  * pointed to by the cluster, someone else raced in and freed the
2979  * cluster already.  In that case, we just return without changing anything
2980  */
__btrfs_return_cluster_to_free_space(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster)2981 static void __btrfs_return_cluster_to_free_space(
2982 			     struct btrfs_block_group *block_group,
2983 			     struct btrfs_free_cluster *cluster)
2984 {
2985 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2986 	struct rb_node *node;
2987 
2988 	lockdep_assert_held(&ctl->tree_lock);
2989 
2990 	spin_lock(&cluster->lock);
2991 	if (cluster->block_group != block_group) {
2992 		spin_unlock(&cluster->lock);
2993 		return;
2994 	}
2995 
2996 	cluster->block_group = NULL;
2997 	cluster->window_start = 0;
2998 	list_del_init(&cluster->block_group_list);
2999 
3000 	node = rb_first(&cluster->root);
3001 	while (node) {
3002 		struct btrfs_free_space *entry;
3003 
3004 		entry = rb_entry(node, struct btrfs_free_space, offset_index);
3005 		node = rb_next(&entry->offset_index);
3006 		rb_erase(&entry->offset_index, &cluster->root);
3007 		RB_CLEAR_NODE(&entry->offset_index);
3008 
3009 		if (!entry->bitmap) {
3010 			/* Merging treats extents as if they were new */
3011 			if (!btrfs_free_space_trimmed(entry)) {
3012 				ctl->discardable_extents[BTRFS_STAT_CURR]--;
3013 				ctl->discardable_bytes[BTRFS_STAT_CURR] -=
3014 					entry->bytes;
3015 			}
3016 
3017 			try_merge_free_space(ctl, entry, false);
3018 			steal_from_bitmap(ctl, entry, false);
3019 
3020 			/* As we insert directly, update these statistics */
3021 			if (!btrfs_free_space_trimmed(entry)) {
3022 				ctl->discardable_extents[BTRFS_STAT_CURR]++;
3023 				ctl->discardable_bytes[BTRFS_STAT_CURR] +=
3024 					entry->bytes;
3025 			}
3026 		}
3027 		tree_insert_offset(ctl, NULL, entry);
3028 		rb_add_cached(&entry->bytes_index, &ctl->free_space_bytes,
3029 			      entry_less);
3030 	}
3031 	cluster->root = RB_ROOT;
3032 	spin_unlock(&cluster->lock);
3033 	btrfs_put_block_group(block_group);
3034 }
3035 
btrfs_remove_free_space_cache(struct btrfs_block_group * block_group)3036 void btrfs_remove_free_space_cache(struct btrfs_block_group *block_group)
3037 {
3038 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3039 	struct btrfs_free_cluster *cluster;
3040 	struct list_head *head;
3041 
3042 	spin_lock(&ctl->tree_lock);
3043 	while ((head = block_group->cluster_list.next) !=
3044 	       &block_group->cluster_list) {
3045 		cluster = list_entry(head, struct btrfs_free_cluster,
3046 				     block_group_list);
3047 
3048 		WARN_ON(cluster->block_group != block_group);
3049 		__btrfs_return_cluster_to_free_space(block_group, cluster);
3050 
3051 		cond_resched_lock(&ctl->tree_lock);
3052 	}
3053 	__btrfs_remove_free_space_cache(ctl);
3054 	btrfs_discard_update_discardable(block_group);
3055 	spin_unlock(&ctl->tree_lock);
3056 
3057 }
3058 
3059 /*
3060  * Walk @block_group's free space rb_tree to determine if everything is trimmed.
3061  */
btrfs_is_free_space_trimmed(struct btrfs_block_group * block_group)3062 bool btrfs_is_free_space_trimmed(struct btrfs_block_group *block_group)
3063 {
3064 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3065 	struct btrfs_free_space *info;
3066 	struct rb_node *node;
3067 	bool ret = true;
3068 
3069 	spin_lock(&ctl->tree_lock);
3070 	node = rb_first(&ctl->free_space_offset);
3071 
3072 	while (node) {
3073 		info = rb_entry(node, struct btrfs_free_space, offset_index);
3074 
3075 		if (!btrfs_free_space_trimmed(info)) {
3076 			ret = false;
3077 			break;
3078 		}
3079 
3080 		node = rb_next(node);
3081 	}
3082 
3083 	spin_unlock(&ctl->tree_lock);
3084 	return ret;
3085 }
3086 
btrfs_find_space_for_alloc(struct btrfs_block_group * block_group,u64 offset,u64 bytes,u64 empty_size,u64 * max_extent_size)3087 u64 btrfs_find_space_for_alloc(struct btrfs_block_group *block_group,
3088 			       u64 offset, u64 bytes, u64 empty_size,
3089 			       u64 *max_extent_size)
3090 {
3091 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3092 	struct btrfs_discard_ctl *discard_ctl =
3093 					&block_group->fs_info->discard_ctl;
3094 	struct btrfs_free_space *entry = NULL;
3095 	u64 bytes_search = bytes + empty_size;
3096 	u64 ret = 0;
3097 	u64 align_gap = 0;
3098 	u64 align_gap_len = 0;
3099 	enum btrfs_trim_state align_gap_trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3100 	bool use_bytes_index = (offset == block_group->start);
3101 
3102 	ASSERT(!btrfs_is_zoned(block_group->fs_info));
3103 
3104 	spin_lock(&ctl->tree_lock);
3105 	entry = find_free_space(ctl, &offset, &bytes_search,
3106 				block_group->full_stripe_len, max_extent_size,
3107 				use_bytes_index);
3108 	if (!entry)
3109 		goto out;
3110 
3111 	ret = offset;
3112 	if (entry->bitmap) {
3113 		bitmap_clear_bits(ctl, entry, offset, bytes, true);
3114 
3115 		if (!btrfs_free_space_trimmed(entry))
3116 			atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3117 
3118 		if (!entry->bytes)
3119 			free_bitmap(ctl, entry);
3120 	} else {
3121 		unlink_free_space(ctl, entry, true);
3122 		align_gap_len = offset - entry->offset;
3123 		align_gap = entry->offset;
3124 		align_gap_trim_state = entry->trim_state;
3125 
3126 		if (!btrfs_free_space_trimmed(entry))
3127 			atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3128 
3129 		entry->offset = offset + bytes;
3130 		WARN_ON(entry->bytes < bytes + align_gap_len);
3131 
3132 		entry->bytes -= bytes + align_gap_len;
3133 		if (!entry->bytes)
3134 			kmem_cache_free(btrfs_free_space_cachep, entry);
3135 		else
3136 			link_free_space(ctl, entry);
3137 	}
3138 out:
3139 	btrfs_discard_update_discardable(block_group);
3140 	spin_unlock(&ctl->tree_lock);
3141 
3142 	if (align_gap_len)
3143 		__btrfs_add_free_space(block_group, align_gap, align_gap_len,
3144 				       align_gap_trim_state);
3145 	return ret;
3146 }
3147 
3148 /*
3149  * given a cluster, put all of its extents back into the free space
3150  * cache.  If a block group is passed, this function will only free
3151  * a cluster that belongs to the passed block group.
3152  *
3153  * Otherwise, it'll get a reference on the block group pointed to by the
3154  * cluster and remove the cluster from it.
3155  */
btrfs_return_cluster_to_free_space(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster)3156 void btrfs_return_cluster_to_free_space(
3157 			       struct btrfs_block_group *block_group,
3158 			       struct btrfs_free_cluster *cluster)
3159 {
3160 	struct btrfs_free_space_ctl *ctl;
3161 
3162 	/* first, get a safe pointer to the block group */
3163 	spin_lock(&cluster->lock);
3164 	if (!block_group) {
3165 		block_group = cluster->block_group;
3166 		if (!block_group) {
3167 			spin_unlock(&cluster->lock);
3168 			return;
3169 		}
3170 	} else if (cluster->block_group != block_group) {
3171 		/* someone else has already freed it don't redo their work */
3172 		spin_unlock(&cluster->lock);
3173 		return;
3174 	}
3175 	btrfs_get_block_group(block_group);
3176 	spin_unlock(&cluster->lock);
3177 
3178 	ctl = block_group->free_space_ctl;
3179 
3180 	/* now return any extents the cluster had on it */
3181 	spin_lock(&ctl->tree_lock);
3182 	__btrfs_return_cluster_to_free_space(block_group, cluster);
3183 	spin_unlock(&ctl->tree_lock);
3184 
3185 	btrfs_discard_queue_work(&block_group->fs_info->discard_ctl, block_group);
3186 
3187 	/* finally drop our ref */
3188 	btrfs_put_block_group(block_group);
3189 }
3190 
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)3191 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group *block_group,
3192 				   struct btrfs_free_cluster *cluster,
3193 				   struct btrfs_free_space *entry,
3194 				   u64 bytes, u64 min_start,
3195 				   u64 *max_extent_size)
3196 {
3197 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3198 	int err;
3199 	u64 search_start = cluster->window_start;
3200 	u64 search_bytes = bytes;
3201 	u64 ret = 0;
3202 
3203 	search_start = min_start;
3204 	search_bytes = bytes;
3205 
3206 	err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
3207 	if (err) {
3208 		*max_extent_size = max(get_max_extent_size(entry),
3209 				       *max_extent_size);
3210 		return 0;
3211 	}
3212 
3213 	ret = search_start;
3214 	bitmap_clear_bits(ctl, entry, ret, bytes, false);
3215 
3216 	return ret;
3217 }
3218 
3219 /*
3220  * given a cluster, try to allocate 'bytes' from it, returns 0
3221  * if it couldn't find anything suitably large, or a logical disk offset
3222  * if things worked out
3223  */
btrfs_alloc_from_cluster(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster,u64 bytes,u64 min_start,u64 * max_extent_size)3224 u64 btrfs_alloc_from_cluster(struct btrfs_block_group *block_group,
3225 			     struct btrfs_free_cluster *cluster, u64 bytes,
3226 			     u64 min_start, u64 *max_extent_size)
3227 {
3228 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3229 	struct btrfs_discard_ctl *discard_ctl =
3230 					&block_group->fs_info->discard_ctl;
3231 	struct btrfs_free_space *entry = NULL;
3232 	struct rb_node *node;
3233 	u64 ret = 0;
3234 
3235 	ASSERT(!btrfs_is_zoned(block_group->fs_info));
3236 
3237 	spin_lock(&cluster->lock);
3238 	if (bytes > cluster->max_size)
3239 		goto out;
3240 
3241 	if (cluster->block_group != block_group)
3242 		goto out;
3243 
3244 	node = rb_first(&cluster->root);
3245 	if (!node)
3246 		goto out;
3247 
3248 	entry = rb_entry(node, struct btrfs_free_space, offset_index);
3249 	while (1) {
3250 		if (entry->bytes < bytes)
3251 			*max_extent_size = max(get_max_extent_size(entry),
3252 					       *max_extent_size);
3253 
3254 		if (entry->bytes < bytes ||
3255 		    (!entry->bitmap && entry->offset < min_start)) {
3256 			node = rb_next(&entry->offset_index);
3257 			if (!node)
3258 				break;
3259 			entry = rb_entry(node, struct btrfs_free_space,
3260 					 offset_index);
3261 			continue;
3262 		}
3263 
3264 		if (entry->bitmap) {
3265 			ret = btrfs_alloc_from_bitmap(block_group,
3266 						      cluster, entry, bytes,
3267 						      cluster->window_start,
3268 						      max_extent_size);
3269 			if (ret == 0) {
3270 				node = rb_next(&entry->offset_index);
3271 				if (!node)
3272 					break;
3273 				entry = rb_entry(node, struct btrfs_free_space,
3274 						 offset_index);
3275 				continue;
3276 			}
3277 			cluster->window_start += bytes;
3278 		} else {
3279 			ret = entry->offset;
3280 
3281 			entry->offset += bytes;
3282 			entry->bytes -= bytes;
3283 		}
3284 
3285 		break;
3286 	}
3287 out:
3288 	spin_unlock(&cluster->lock);
3289 
3290 	if (!ret)
3291 		return 0;
3292 
3293 	spin_lock(&ctl->tree_lock);
3294 
3295 	if (!btrfs_free_space_trimmed(entry))
3296 		atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3297 
3298 	ctl->free_space -= bytes;
3299 	if (!entry->bitmap && !btrfs_free_space_trimmed(entry))
3300 		ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
3301 
3302 	spin_lock(&cluster->lock);
3303 	if (entry->bytes == 0) {
3304 		rb_erase(&entry->offset_index, &cluster->root);
3305 		ctl->free_extents--;
3306 		if (entry->bitmap) {
3307 			kmem_cache_free(btrfs_free_space_bitmap_cachep,
3308 					entry->bitmap);
3309 			ctl->total_bitmaps--;
3310 			recalculate_thresholds(ctl);
3311 		} else if (!btrfs_free_space_trimmed(entry)) {
3312 			ctl->discardable_extents[BTRFS_STAT_CURR]--;
3313 		}
3314 		kmem_cache_free(btrfs_free_space_cachep, entry);
3315 	}
3316 
3317 	spin_unlock(&cluster->lock);
3318 	spin_unlock(&ctl->tree_lock);
3319 
3320 	return ret;
3321 }
3322 
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)3323 static int btrfs_bitmap_cluster(struct btrfs_block_group *block_group,
3324 				struct btrfs_free_space *entry,
3325 				struct btrfs_free_cluster *cluster,
3326 				u64 offset, u64 bytes,
3327 				u64 cont1_bytes, u64 min_bytes)
3328 {
3329 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3330 	unsigned long next_zero;
3331 	unsigned long i;
3332 	unsigned long want_bits;
3333 	unsigned long min_bits;
3334 	unsigned long found_bits;
3335 	unsigned long max_bits = 0;
3336 	unsigned long start = 0;
3337 	unsigned long total_found = 0;
3338 	int ret;
3339 
3340 	lockdep_assert_held(&ctl->tree_lock);
3341 
3342 	i = offset_to_bit(entry->offset, ctl->unit,
3343 			  max_t(u64, offset, entry->offset));
3344 	want_bits = bytes_to_bits(bytes, ctl->unit);
3345 	min_bits = bytes_to_bits(min_bytes, ctl->unit);
3346 
3347 	/*
3348 	 * Don't bother looking for a cluster in this bitmap if it's heavily
3349 	 * fragmented.
3350 	 */
3351 	if (entry->max_extent_size &&
3352 	    entry->max_extent_size < cont1_bytes)
3353 		return -ENOSPC;
3354 again:
3355 	found_bits = 0;
3356 	for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
3357 		next_zero = find_next_zero_bit(entry->bitmap,
3358 					       BITS_PER_BITMAP, i);
3359 		if (next_zero - i >= min_bits) {
3360 			found_bits = next_zero - i;
3361 			if (found_bits > max_bits)
3362 				max_bits = found_bits;
3363 			break;
3364 		}
3365 		if (next_zero - i > max_bits)
3366 			max_bits = next_zero - i;
3367 		i = next_zero;
3368 	}
3369 
3370 	if (!found_bits) {
3371 		entry->max_extent_size = (u64)max_bits * ctl->unit;
3372 		return -ENOSPC;
3373 	}
3374 
3375 	if (!total_found) {
3376 		start = i;
3377 		cluster->max_size = 0;
3378 	}
3379 
3380 	total_found += found_bits;
3381 
3382 	if (cluster->max_size < found_bits * ctl->unit)
3383 		cluster->max_size = found_bits * ctl->unit;
3384 
3385 	if (total_found < want_bits || cluster->max_size < cont1_bytes) {
3386 		i = next_zero + 1;
3387 		goto again;
3388 	}
3389 
3390 	cluster->window_start = start * ctl->unit + entry->offset;
3391 	rb_erase(&entry->offset_index, &ctl->free_space_offset);
3392 	rb_erase_cached(&entry->bytes_index, &ctl->free_space_bytes);
3393 
3394 	/*
3395 	 * We need to know if we're currently on the normal space index when we
3396 	 * manipulate the bitmap so that we know we need to remove and re-insert
3397 	 * it into the space_index tree.  Clear the bytes_index node here so the
3398 	 * bitmap manipulation helpers know not to mess with the space_index
3399 	 * until this bitmap entry is added back into the normal cache.
3400 	 */
3401 	RB_CLEAR_NODE(&entry->bytes_index);
3402 
3403 	ret = tree_insert_offset(ctl, cluster, entry);
3404 	ASSERT(!ret); /* -EEXIST; Logic error */
3405 
3406 	trace_btrfs_setup_cluster(block_group, cluster,
3407 				  total_found * ctl->unit, 1);
3408 	return 0;
3409 }
3410 
3411 /*
3412  * This searches the block group for just extents to fill the cluster with.
3413  * Try to find a cluster with at least bytes total bytes, at least one
3414  * extent of cont1_bytes, and other clusters of at least min_bytes.
3415  */
3416 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)3417 setup_cluster_no_bitmap(struct btrfs_block_group *block_group,
3418 			struct btrfs_free_cluster *cluster,
3419 			struct list_head *bitmaps, u64 offset, u64 bytes,
3420 			u64 cont1_bytes, u64 min_bytes)
3421 {
3422 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3423 	struct btrfs_free_space *first = NULL;
3424 	struct btrfs_free_space *entry = NULL;
3425 	struct btrfs_free_space *last;
3426 	struct rb_node *node;
3427 	u64 window_free;
3428 	u64 max_extent;
3429 	u64 total_size = 0;
3430 
3431 	lockdep_assert_held(&ctl->tree_lock);
3432 
3433 	entry = tree_search_offset(ctl, offset, 0, 1);
3434 	if (!entry)
3435 		return -ENOSPC;
3436 
3437 	/*
3438 	 * We don't want bitmaps, so just move along until we find a normal
3439 	 * extent entry.
3440 	 */
3441 	while (entry->bitmap || entry->bytes < min_bytes) {
3442 		if (entry->bitmap && list_empty(&entry->list))
3443 			list_add_tail(&entry->list, bitmaps);
3444 		node = rb_next(&entry->offset_index);
3445 		if (!node)
3446 			return -ENOSPC;
3447 		entry = rb_entry(node, struct btrfs_free_space, offset_index);
3448 	}
3449 
3450 	window_free = entry->bytes;
3451 	max_extent = entry->bytes;
3452 	first = entry;
3453 	last = entry;
3454 
3455 	for (node = rb_next(&entry->offset_index); node;
3456 	     node = rb_next(&entry->offset_index)) {
3457 		entry = rb_entry(node, struct btrfs_free_space, offset_index);
3458 
3459 		if (entry->bitmap) {
3460 			if (list_empty(&entry->list))
3461 				list_add_tail(&entry->list, bitmaps);
3462 			continue;
3463 		}
3464 
3465 		if (entry->bytes < min_bytes)
3466 			continue;
3467 
3468 		last = entry;
3469 		window_free += entry->bytes;
3470 		if (entry->bytes > max_extent)
3471 			max_extent = entry->bytes;
3472 	}
3473 
3474 	if (window_free < bytes || max_extent < cont1_bytes)
3475 		return -ENOSPC;
3476 
3477 	cluster->window_start = first->offset;
3478 
3479 	node = &first->offset_index;
3480 
3481 	/*
3482 	 * now we've found our entries, pull them out of the free space
3483 	 * cache and put them into the cluster rbtree
3484 	 */
3485 	do {
3486 		int ret;
3487 
3488 		entry = rb_entry(node, struct btrfs_free_space, offset_index);
3489 		node = rb_next(&entry->offset_index);
3490 		if (entry->bitmap || entry->bytes < min_bytes)
3491 			continue;
3492 
3493 		rb_erase(&entry->offset_index, &ctl->free_space_offset);
3494 		rb_erase_cached(&entry->bytes_index, &ctl->free_space_bytes);
3495 		ret = tree_insert_offset(ctl, cluster, entry);
3496 		total_size += entry->bytes;
3497 		ASSERT(!ret); /* -EEXIST; Logic error */
3498 	} while (node && entry != last);
3499 
3500 	cluster->max_size = max_extent;
3501 	trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
3502 	return 0;
3503 }
3504 
3505 /*
3506  * This specifically looks for bitmaps that may work in the cluster, we assume
3507  * that we have already failed to find extents that will work.
3508  */
3509 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)3510 setup_cluster_bitmap(struct btrfs_block_group *block_group,
3511 		     struct btrfs_free_cluster *cluster,
3512 		     struct list_head *bitmaps, u64 offset, u64 bytes,
3513 		     u64 cont1_bytes, u64 min_bytes)
3514 {
3515 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3516 	struct btrfs_free_space *entry = NULL;
3517 	int ret = -ENOSPC;
3518 	u64 bitmap_offset = offset_to_bitmap(ctl, offset);
3519 
3520 	if (ctl->total_bitmaps == 0)
3521 		return -ENOSPC;
3522 
3523 	/*
3524 	 * The bitmap that covers offset won't be in the list unless offset
3525 	 * is just its start offset.
3526 	 */
3527 	if (!list_empty(bitmaps))
3528 		entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
3529 
3530 	if (!entry || entry->offset != bitmap_offset) {
3531 		entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
3532 		if (entry && list_empty(&entry->list))
3533 			list_add(&entry->list, bitmaps);
3534 	}
3535 
3536 	list_for_each_entry(entry, bitmaps, list) {
3537 		if (entry->bytes < bytes)
3538 			continue;
3539 		ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
3540 					   bytes, cont1_bytes, min_bytes);
3541 		if (!ret)
3542 			return 0;
3543 	}
3544 
3545 	/*
3546 	 * The bitmaps list has all the bitmaps that record free space
3547 	 * starting after offset, so no more search is required.
3548 	 */
3549 	return -ENOSPC;
3550 }
3551 
3552 /*
3553  * here we try to find a cluster of blocks in a block group.  The goal
3554  * is to find at least bytes+empty_size.
3555  * We might not find them all in one contiguous area.
3556  *
3557  * returns zero and sets up cluster if things worked out, otherwise
3558  * it returns -enospc
3559  */
btrfs_find_space_cluster(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster,u64 offset,u64 bytes,u64 empty_size)3560 int btrfs_find_space_cluster(struct btrfs_block_group *block_group,
3561 			     struct btrfs_free_cluster *cluster,
3562 			     u64 offset, u64 bytes, u64 empty_size)
3563 {
3564 	struct btrfs_fs_info *fs_info = block_group->fs_info;
3565 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3566 	struct btrfs_free_space *entry, *tmp;
3567 	LIST_HEAD(bitmaps);
3568 	u64 min_bytes;
3569 	u64 cont1_bytes;
3570 	int ret;
3571 
3572 	/*
3573 	 * Choose the minimum extent size we'll require for this
3574 	 * cluster.  For SSD_SPREAD, don't allow any fragmentation.
3575 	 * For metadata, allow allocates with smaller extents.  For
3576 	 * data, keep it dense.
3577 	 */
3578 	if (btrfs_test_opt(fs_info, SSD_SPREAD)) {
3579 		cont1_bytes = bytes + empty_size;
3580 		min_bytes = cont1_bytes;
3581 	} else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3582 		cont1_bytes = bytes;
3583 		min_bytes = fs_info->sectorsize;
3584 	} else {
3585 		cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3586 		min_bytes = fs_info->sectorsize;
3587 	}
3588 
3589 	spin_lock(&ctl->tree_lock);
3590 
3591 	/*
3592 	 * If we know we don't have enough space to make a cluster don't even
3593 	 * bother doing all the work to try and find one.
3594 	 */
3595 	if (ctl->free_space < bytes) {
3596 		spin_unlock(&ctl->tree_lock);
3597 		return -ENOSPC;
3598 	}
3599 
3600 	spin_lock(&cluster->lock);
3601 
3602 	/* someone already found a cluster, hooray */
3603 	if (cluster->block_group) {
3604 		ret = 0;
3605 		goto out;
3606 	}
3607 
3608 	trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
3609 				 min_bytes);
3610 
3611 	ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
3612 				      bytes + empty_size,
3613 				      cont1_bytes, min_bytes);
3614 	if (ret)
3615 		ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
3616 					   offset, bytes + empty_size,
3617 					   cont1_bytes, min_bytes);
3618 
3619 	/* Clear our temporary list */
3620 	list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3621 		list_del_init(&entry->list);
3622 
3623 	if (!ret) {
3624 		btrfs_get_block_group(block_group);
3625 		list_add_tail(&cluster->block_group_list,
3626 			      &block_group->cluster_list);
3627 		cluster->block_group = block_group;
3628 	} else {
3629 		trace_btrfs_failed_cluster_setup(block_group);
3630 	}
3631 out:
3632 	spin_unlock(&cluster->lock);
3633 	spin_unlock(&ctl->tree_lock);
3634 
3635 	return ret;
3636 }
3637 
3638 /*
3639  * simple code to zero out a cluster
3640  */
btrfs_init_free_cluster(struct btrfs_free_cluster * cluster)3641 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3642 {
3643 	spin_lock_init(&cluster->lock);
3644 	spin_lock_init(&cluster->refill_lock);
3645 	cluster->root = RB_ROOT;
3646 	cluster->max_size = 0;
3647 	cluster->fragmented = false;
3648 	INIT_LIST_HEAD(&cluster->block_group_list);
3649 	cluster->block_group = NULL;
3650 }
3651 
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)3652 static int do_trimming(struct btrfs_block_group *block_group,
3653 		       u64 *total_trimmed, u64 start, u64 bytes,
3654 		       u64 reserved_start, u64 reserved_bytes,
3655 		       enum btrfs_trim_state reserved_trim_state,
3656 		       struct btrfs_trim_range *trim_entry)
3657 {
3658 	struct btrfs_space_info *space_info = block_group->space_info;
3659 	struct btrfs_fs_info *fs_info = block_group->fs_info;
3660 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3661 	int ret;
3662 	int update = 0;
3663 	const u64 end = start + bytes;
3664 	const u64 reserved_end = reserved_start + reserved_bytes;
3665 	enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3666 	u64 trimmed = 0;
3667 
3668 	spin_lock(&space_info->lock);
3669 	spin_lock(&block_group->lock);
3670 	if (!block_group->ro) {
3671 		block_group->reserved += reserved_bytes;
3672 		space_info->bytes_reserved += reserved_bytes;
3673 		update = 1;
3674 	}
3675 	spin_unlock(&block_group->lock);
3676 	spin_unlock(&space_info->lock);
3677 
3678 	ret = btrfs_discard_extent(fs_info, start, bytes, &trimmed);
3679 	if (!ret) {
3680 		*total_trimmed += trimmed;
3681 		trim_state = BTRFS_TRIM_STATE_TRIMMED;
3682 	}
3683 
3684 	mutex_lock(&ctl->cache_writeout_mutex);
3685 	if (reserved_start < start)
3686 		__btrfs_add_free_space(block_group, reserved_start,
3687 				       start - reserved_start,
3688 				       reserved_trim_state);
3689 	if (end < reserved_end)
3690 		__btrfs_add_free_space(block_group, end, reserved_end - end,
3691 				       reserved_trim_state);
3692 	__btrfs_add_free_space(block_group, start, bytes, trim_state);
3693 	list_del(&trim_entry->list);
3694 	mutex_unlock(&ctl->cache_writeout_mutex);
3695 
3696 	if (update) {
3697 		spin_lock(&space_info->lock);
3698 		spin_lock(&block_group->lock);
3699 		if (block_group->ro)
3700 			space_info->bytes_readonly += reserved_bytes;
3701 		block_group->reserved -= reserved_bytes;
3702 		space_info->bytes_reserved -= reserved_bytes;
3703 		spin_unlock(&block_group->lock);
3704 		spin_unlock(&space_info->lock);
3705 	}
3706 
3707 	return ret;
3708 }
3709 
3710 /*
3711  * If @async is set, then we will trim 1 region and return.
3712  */
trim_no_bitmap(struct btrfs_block_group * block_group,u64 * total_trimmed,u64 start,u64 end,u64 minlen,bool async)3713 static int trim_no_bitmap(struct btrfs_block_group *block_group,
3714 			  u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3715 			  bool async)
3716 {
3717 	struct btrfs_discard_ctl *discard_ctl =
3718 					&block_group->fs_info->discard_ctl;
3719 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3720 	struct btrfs_free_space *entry;
3721 	struct rb_node *node;
3722 	int ret = 0;
3723 	u64 extent_start;
3724 	u64 extent_bytes;
3725 	enum btrfs_trim_state extent_trim_state;
3726 	u64 bytes;
3727 	const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3728 
3729 	while (start < end) {
3730 		struct btrfs_trim_range trim_entry;
3731 
3732 		mutex_lock(&ctl->cache_writeout_mutex);
3733 		spin_lock(&ctl->tree_lock);
3734 
3735 		if (ctl->free_space < minlen)
3736 			goto out_unlock;
3737 
3738 		entry = tree_search_offset(ctl, start, 0, 1);
3739 		if (!entry)
3740 			goto out_unlock;
3741 
3742 		/* Skip bitmaps and if async, already trimmed entries */
3743 		while (entry->bitmap ||
3744 		       (async && btrfs_free_space_trimmed(entry))) {
3745 			node = rb_next(&entry->offset_index);
3746 			if (!node)
3747 				goto out_unlock;
3748 			entry = rb_entry(node, struct btrfs_free_space,
3749 					 offset_index);
3750 		}
3751 
3752 		if (entry->offset >= end)
3753 			goto out_unlock;
3754 
3755 		extent_start = entry->offset;
3756 		extent_bytes = entry->bytes;
3757 		extent_trim_state = entry->trim_state;
3758 		if (async) {
3759 			start = entry->offset;
3760 			bytes = entry->bytes;
3761 			if (bytes < minlen) {
3762 				spin_unlock(&ctl->tree_lock);
3763 				mutex_unlock(&ctl->cache_writeout_mutex);
3764 				goto next;
3765 			}
3766 			unlink_free_space(ctl, entry, true);
3767 			/*
3768 			 * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3769 			 * If X < BTRFS_ASYNC_DISCARD_MIN_FILTER, we won't trim
3770 			 * X when we come back around.  So trim it now.
3771 			 */
3772 			if (max_discard_size &&
3773 			    bytes >= (max_discard_size +
3774 				      BTRFS_ASYNC_DISCARD_MIN_FILTER)) {
3775 				bytes = max_discard_size;
3776 				extent_bytes = max_discard_size;
3777 				entry->offset += max_discard_size;
3778 				entry->bytes -= max_discard_size;
3779 				link_free_space(ctl, entry);
3780 			} else {
3781 				kmem_cache_free(btrfs_free_space_cachep, entry);
3782 			}
3783 		} else {
3784 			start = max(start, extent_start);
3785 			bytes = min(extent_start + extent_bytes, end) - start;
3786 			if (bytes < minlen) {
3787 				spin_unlock(&ctl->tree_lock);
3788 				mutex_unlock(&ctl->cache_writeout_mutex);
3789 				goto next;
3790 			}
3791 
3792 			unlink_free_space(ctl, entry, true);
3793 			kmem_cache_free(btrfs_free_space_cachep, entry);
3794 		}
3795 
3796 		spin_unlock(&ctl->tree_lock);
3797 		trim_entry.start = extent_start;
3798 		trim_entry.bytes = extent_bytes;
3799 		list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3800 		mutex_unlock(&ctl->cache_writeout_mutex);
3801 
3802 		ret = do_trimming(block_group, total_trimmed, start, bytes,
3803 				  extent_start, extent_bytes, extent_trim_state,
3804 				  &trim_entry);
3805 		if (ret) {
3806 			block_group->discard_cursor = start + bytes;
3807 			break;
3808 		}
3809 next:
3810 		start += bytes;
3811 		block_group->discard_cursor = start;
3812 		if (async && *total_trimmed)
3813 			break;
3814 
3815 		if (btrfs_trim_interrupted()) {
3816 			ret = -ERESTARTSYS;
3817 			break;
3818 		}
3819 
3820 		cond_resched();
3821 	}
3822 
3823 	return ret;
3824 
3825 out_unlock:
3826 	block_group->discard_cursor = btrfs_block_group_end(block_group);
3827 	spin_unlock(&ctl->tree_lock);
3828 	mutex_unlock(&ctl->cache_writeout_mutex);
3829 
3830 	return ret;
3831 }
3832 
3833 /*
3834  * If we break out of trimming a bitmap prematurely, we should reset the
3835  * trimming bit.  In a rather contrieved case, it's possible to race here so
3836  * reset the state to BTRFS_TRIM_STATE_UNTRIMMED.
3837  *
3838  * start = start of bitmap
3839  * end = near end of bitmap
3840  *
3841  * Thread 1:			Thread 2:
3842  * trim_bitmaps(start)
3843  *				trim_bitmaps(end)
3844  *				end_trimming_bitmap()
3845  * reset_trimming_bitmap()
3846  */
reset_trimming_bitmap(struct btrfs_free_space_ctl * ctl,u64 offset)3847 static void reset_trimming_bitmap(struct btrfs_free_space_ctl *ctl, u64 offset)
3848 {
3849 	struct btrfs_free_space *entry;
3850 
3851 	spin_lock(&ctl->tree_lock);
3852 	entry = tree_search_offset(ctl, offset, 1, 0);
3853 	if (entry) {
3854 		if (btrfs_free_space_trimmed(entry)) {
3855 			ctl->discardable_extents[BTRFS_STAT_CURR] +=
3856 				entry->bitmap_extents;
3857 			ctl->discardable_bytes[BTRFS_STAT_CURR] += entry->bytes;
3858 		}
3859 		entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3860 	}
3861 
3862 	spin_unlock(&ctl->tree_lock);
3863 }
3864 
end_trimming_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * entry)3865 static void end_trimming_bitmap(struct btrfs_free_space_ctl *ctl,
3866 				struct btrfs_free_space *entry)
3867 {
3868 	if (btrfs_free_space_trimming_bitmap(entry)) {
3869 		entry->trim_state = BTRFS_TRIM_STATE_TRIMMED;
3870 		ctl->discardable_extents[BTRFS_STAT_CURR] -=
3871 			entry->bitmap_extents;
3872 		ctl->discardable_bytes[BTRFS_STAT_CURR] -= entry->bytes;
3873 	}
3874 }
3875 
3876 /*
3877  * If @async is set, then we will trim 1 region and return.
3878  */
trim_bitmaps(struct btrfs_block_group * block_group,u64 * total_trimmed,u64 start,u64 end,u64 minlen,u64 maxlen,bool async)3879 static int trim_bitmaps(struct btrfs_block_group *block_group,
3880 			u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3881 			u64 maxlen, bool async)
3882 {
3883 	struct btrfs_discard_ctl *discard_ctl =
3884 					&block_group->fs_info->discard_ctl;
3885 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3886 	struct btrfs_free_space *entry;
3887 	int ret = 0;
3888 	int ret2;
3889 	u64 bytes;
3890 	u64 offset = offset_to_bitmap(ctl, start);
3891 	const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3892 
3893 	while (offset < end) {
3894 		bool next_bitmap = false;
3895 		struct btrfs_trim_range trim_entry;
3896 
3897 		mutex_lock(&ctl->cache_writeout_mutex);
3898 		spin_lock(&ctl->tree_lock);
3899 
3900 		if (ctl->free_space < minlen) {
3901 			block_group->discard_cursor =
3902 				btrfs_block_group_end(block_group);
3903 			spin_unlock(&ctl->tree_lock);
3904 			mutex_unlock(&ctl->cache_writeout_mutex);
3905 			break;
3906 		}
3907 
3908 		entry = tree_search_offset(ctl, offset, 1, 0);
3909 		/*
3910 		 * Bitmaps are marked trimmed lossily now to prevent constant
3911 		 * discarding of the same bitmap (the reason why we are bound
3912 		 * by the filters).  So, retrim the block group bitmaps when we
3913 		 * are preparing to punt to the unused_bgs list.  This uses
3914 		 * @minlen to determine if we are in BTRFS_DISCARD_INDEX_UNUSED
3915 		 * which is the only discard index which sets minlen to 0.
3916 		 */
3917 		if (!entry || (async && minlen && start == offset &&
3918 			       btrfs_free_space_trimmed(entry))) {
3919 			spin_unlock(&ctl->tree_lock);
3920 			mutex_unlock(&ctl->cache_writeout_mutex);
3921 			next_bitmap = true;
3922 			goto next;
3923 		}
3924 
3925 		/*
3926 		 * Async discard bitmap trimming begins at by setting the start
3927 		 * to be key.objectid and the offset_to_bitmap() aligns to the
3928 		 * start of the bitmap.  This lets us know we are fully
3929 		 * scanning the bitmap rather than only some portion of it.
3930 		 */
3931 		if (start == offset)
3932 			entry->trim_state = BTRFS_TRIM_STATE_TRIMMING;
3933 
3934 		bytes = minlen;
3935 		ret2 = search_bitmap(ctl, entry, &start, &bytes, false);
3936 		if (ret2 || start >= end) {
3937 			/*
3938 			 * We lossily consider a bitmap trimmed if we only skip
3939 			 * over regions <= BTRFS_ASYNC_DISCARD_MIN_FILTER.
3940 			 */
3941 			if (ret2 && minlen <= BTRFS_ASYNC_DISCARD_MIN_FILTER)
3942 				end_trimming_bitmap(ctl, entry);
3943 			else
3944 				entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3945 			spin_unlock(&ctl->tree_lock);
3946 			mutex_unlock(&ctl->cache_writeout_mutex);
3947 			next_bitmap = true;
3948 			goto next;
3949 		}
3950 
3951 		/*
3952 		 * We already trimmed a region, but are using the locking above
3953 		 * to reset the trim_state.
3954 		 */
3955 		if (async && *total_trimmed) {
3956 			spin_unlock(&ctl->tree_lock);
3957 			mutex_unlock(&ctl->cache_writeout_mutex);
3958 			goto out;
3959 		}
3960 
3961 		bytes = min(bytes, end - start);
3962 		if (bytes < minlen || (async && maxlen && bytes > maxlen)) {
3963 			spin_unlock(&ctl->tree_lock);
3964 			mutex_unlock(&ctl->cache_writeout_mutex);
3965 			goto next;
3966 		}
3967 
3968 		/*
3969 		 * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3970 		 * If X < @minlen, we won't trim X when we come back around.
3971 		 * So trim it now.  We differ here from trimming extents as we
3972 		 * don't keep individual state per bit.
3973 		 */
3974 		if (async &&
3975 		    max_discard_size &&
3976 		    bytes > (max_discard_size + minlen))
3977 			bytes = max_discard_size;
3978 
3979 		bitmap_clear_bits(ctl, entry, start, bytes, true);
3980 		if (entry->bytes == 0)
3981 			free_bitmap(ctl, entry);
3982 
3983 		spin_unlock(&ctl->tree_lock);
3984 		trim_entry.start = start;
3985 		trim_entry.bytes = bytes;
3986 		list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3987 		mutex_unlock(&ctl->cache_writeout_mutex);
3988 
3989 		ret = do_trimming(block_group, total_trimmed, start, bytes,
3990 				  start, bytes, 0, &trim_entry);
3991 		if (ret) {
3992 			reset_trimming_bitmap(ctl, offset);
3993 			block_group->discard_cursor =
3994 				btrfs_block_group_end(block_group);
3995 			break;
3996 		}
3997 next:
3998 		if (next_bitmap) {
3999 			offset += BITS_PER_BITMAP * ctl->unit;
4000 			start = offset;
4001 		} else {
4002 			start += bytes;
4003 		}
4004 		block_group->discard_cursor = start;
4005 
4006 		if (btrfs_trim_interrupted()) {
4007 			if (start != offset)
4008 				reset_trimming_bitmap(ctl, offset);
4009 			ret = -ERESTARTSYS;
4010 			break;
4011 		}
4012 
4013 		cond_resched();
4014 	}
4015 
4016 	if (offset >= end)
4017 		block_group->discard_cursor = end;
4018 
4019 out:
4020 	return ret;
4021 }
4022 
btrfs_trim_block_group(struct btrfs_block_group * block_group,u64 * trimmed,u64 start,u64 end,u64 minlen)4023 int btrfs_trim_block_group(struct btrfs_block_group *block_group,
4024 			   u64 *trimmed, u64 start, u64 end, u64 minlen)
4025 {
4026 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
4027 	int ret;
4028 	u64 rem = 0;
4029 
4030 	ASSERT(!btrfs_is_zoned(block_group->fs_info));
4031 
4032 	*trimmed = 0;
4033 
4034 	spin_lock(&block_group->lock);
4035 	if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4036 		spin_unlock(&block_group->lock);
4037 		return 0;
4038 	}
4039 	btrfs_freeze_block_group(block_group);
4040 	spin_unlock(&block_group->lock);
4041 
4042 	ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, false);
4043 	if (ret)
4044 		goto out;
4045 
4046 	ret = trim_bitmaps(block_group, trimmed, start, end, minlen, 0, false);
4047 	div64_u64_rem(end, BITS_PER_BITMAP * ctl->unit, &rem);
4048 	/* If we ended in the middle of a bitmap, reset the trimming flag */
4049 	if (rem)
4050 		reset_trimming_bitmap(ctl, offset_to_bitmap(ctl, end));
4051 out:
4052 	btrfs_unfreeze_block_group(block_group);
4053 	return ret;
4054 }
4055 
btrfs_trim_block_group_extents(struct btrfs_block_group * block_group,u64 * trimmed,u64 start,u64 end,u64 minlen,bool async)4056 int btrfs_trim_block_group_extents(struct btrfs_block_group *block_group,
4057 				   u64 *trimmed, u64 start, u64 end, u64 minlen,
4058 				   bool async)
4059 {
4060 	int ret;
4061 
4062 	*trimmed = 0;
4063 
4064 	spin_lock(&block_group->lock);
4065 	if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4066 		spin_unlock(&block_group->lock);
4067 		return 0;
4068 	}
4069 	btrfs_freeze_block_group(block_group);
4070 	spin_unlock(&block_group->lock);
4071 
4072 	ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, async);
4073 	btrfs_unfreeze_block_group(block_group);
4074 
4075 	return ret;
4076 }
4077 
btrfs_trim_block_group_bitmaps(struct btrfs_block_group * block_group,u64 * trimmed,u64 start,u64 end,u64 minlen,u64 maxlen,bool async)4078 int btrfs_trim_block_group_bitmaps(struct btrfs_block_group *block_group,
4079 				   u64 *trimmed, u64 start, u64 end, u64 minlen,
4080 				   u64 maxlen, bool async)
4081 {
4082 	int ret;
4083 
4084 	*trimmed = 0;
4085 
4086 	spin_lock(&block_group->lock);
4087 	if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4088 		spin_unlock(&block_group->lock);
4089 		return 0;
4090 	}
4091 	btrfs_freeze_block_group(block_group);
4092 	spin_unlock(&block_group->lock);
4093 
4094 	ret = trim_bitmaps(block_group, trimmed, start, end, minlen, maxlen,
4095 			   async);
4096 
4097 	btrfs_unfreeze_block_group(block_group);
4098 
4099 	return ret;
4100 }
4101 
btrfs_free_space_cache_v1_active(struct btrfs_fs_info * fs_info)4102 bool btrfs_free_space_cache_v1_active(struct btrfs_fs_info *fs_info)
4103 {
4104 	return btrfs_super_cache_generation(fs_info->super_copy);
4105 }
4106 
cleanup_free_space_cache_v1(struct btrfs_fs_info * fs_info,struct btrfs_trans_handle * trans)4107 static int cleanup_free_space_cache_v1(struct btrfs_fs_info *fs_info,
4108 				       struct btrfs_trans_handle *trans)
4109 {
4110 	struct btrfs_block_group *block_group;
4111 	struct rb_node *node;
4112 	int ret = 0;
4113 
4114 	btrfs_info(fs_info, "cleaning free space cache v1");
4115 
4116 	node = rb_first_cached(&fs_info->block_group_cache_tree);
4117 	while (node) {
4118 		block_group = rb_entry(node, struct btrfs_block_group, cache_node);
4119 		ret = btrfs_remove_free_space_inode(trans, NULL, block_group);
4120 		if (ret)
4121 			goto out;
4122 		node = rb_next(node);
4123 	}
4124 out:
4125 	return ret;
4126 }
4127 
btrfs_set_free_space_cache_v1_active(struct btrfs_fs_info * fs_info,bool active)4128 int btrfs_set_free_space_cache_v1_active(struct btrfs_fs_info *fs_info, bool active)
4129 {
4130 	struct btrfs_trans_handle *trans;
4131 	int ret;
4132 
4133 	/*
4134 	 * update_super_roots will appropriately set or unset
4135 	 * super_copy->cache_generation based on SPACE_CACHE and
4136 	 * BTRFS_FS_CLEANUP_SPACE_CACHE_V1. For this reason, we need a
4137 	 * transaction commit whether we are enabling space cache v1 and don't
4138 	 * have any other work to do, or are disabling it and removing free
4139 	 * space inodes.
4140 	 */
4141 	trans = btrfs_start_transaction(fs_info->tree_root, 0);
4142 	if (IS_ERR(trans))
4143 		return PTR_ERR(trans);
4144 
4145 	if (!active) {
4146 		set_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);
4147 		ret = cleanup_free_space_cache_v1(fs_info, trans);
4148 		if (ret) {
4149 			btrfs_abort_transaction(trans, ret);
4150 			btrfs_end_transaction(trans);
4151 			goto out;
4152 		}
4153 	}
4154 
4155 	ret = btrfs_commit_transaction(trans);
4156 out:
4157 	clear_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);
4158 
4159 	return ret;
4160 }
4161 
btrfs_free_space_init(void)4162 int __init btrfs_free_space_init(void)
4163 {
4164 	btrfs_free_space_cachep = KMEM_CACHE(btrfs_free_space, 0);
4165 	if (!btrfs_free_space_cachep)
4166 		return -ENOMEM;
4167 
4168 	btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
4169 							PAGE_SIZE, PAGE_SIZE,
4170 							0, NULL);
4171 	if (!btrfs_free_space_bitmap_cachep) {
4172 		kmem_cache_destroy(btrfs_free_space_cachep);
4173 		return -ENOMEM;
4174 	}
4175 
4176 	return 0;
4177 }
4178 
btrfs_free_space_exit(void)4179 void __cold btrfs_free_space_exit(void)
4180 {
4181 	kmem_cache_destroy(btrfs_free_space_cachep);
4182 	kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
4183 }
4184 
4185 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4186 /*
4187  * Use this if you need to make a bitmap or extent entry specifically, it
4188  * doesn't do any of the merging that add_free_space does, this acts a lot like
4189  * how the free space cache loading stuff works, so you can get really weird
4190  * configurations.
4191  */
test_add_free_space_entry(struct btrfs_block_group * cache,u64 offset,u64 bytes,bool bitmap)4192 int test_add_free_space_entry(struct btrfs_block_group *cache,
4193 			      u64 offset, u64 bytes, bool bitmap)
4194 {
4195 	struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4196 	struct btrfs_free_space *info = NULL, *bitmap_info;
4197 	void *map = NULL;
4198 	enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_TRIMMED;
4199 	u64 bytes_added;
4200 	int ret;
4201 
4202 again:
4203 	if (!info) {
4204 		info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
4205 		if (!info)
4206 			return -ENOMEM;
4207 	}
4208 
4209 	if (!bitmap) {
4210 		spin_lock(&ctl->tree_lock);
4211 		info->offset = offset;
4212 		info->bytes = bytes;
4213 		info->max_extent_size = 0;
4214 		ret = link_free_space(ctl, info);
4215 		spin_unlock(&ctl->tree_lock);
4216 		if (ret)
4217 			kmem_cache_free(btrfs_free_space_cachep, info);
4218 		return ret;
4219 	}
4220 
4221 	if (!map) {
4222 		map = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep, GFP_NOFS);
4223 		if (!map) {
4224 			kmem_cache_free(btrfs_free_space_cachep, info);
4225 			return -ENOMEM;
4226 		}
4227 	}
4228 
4229 	spin_lock(&ctl->tree_lock);
4230 	bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
4231 					 1, 0);
4232 	if (!bitmap_info) {
4233 		info->bitmap = map;
4234 		map = NULL;
4235 		add_new_bitmap(ctl, info, offset);
4236 		bitmap_info = info;
4237 		info = NULL;
4238 	}
4239 
4240 	bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
4241 					  trim_state);
4242 
4243 	bytes -= bytes_added;
4244 	offset += bytes_added;
4245 	spin_unlock(&ctl->tree_lock);
4246 
4247 	if (bytes)
4248 		goto again;
4249 
4250 	if (info)
4251 		kmem_cache_free(btrfs_free_space_cachep, info);
4252 	if (map)
4253 		kmem_cache_free(btrfs_free_space_bitmap_cachep, map);
4254 	return 0;
4255 }
4256 
4257 /*
4258  * Checks to see if the given range is in the free space cache.  This is really
4259  * just used to check the absence of space, so if there is free space in the
4260  * range at all we will return 1.
4261  */
test_check_exists(struct btrfs_block_group * cache,u64 offset,u64 bytes)4262 int test_check_exists(struct btrfs_block_group *cache,
4263 		      u64 offset, u64 bytes)
4264 {
4265 	struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4266 	struct btrfs_free_space *info;
4267 	int ret = 0;
4268 
4269 	spin_lock(&ctl->tree_lock);
4270 	info = tree_search_offset(ctl, offset, 0, 0);
4271 	if (!info) {
4272 		info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
4273 					  1, 0);
4274 		if (!info)
4275 			goto out;
4276 	}
4277 
4278 have_info:
4279 	if (info->bitmap) {
4280 		u64 bit_off, bit_bytes;
4281 		struct rb_node *n;
4282 		struct btrfs_free_space *tmp;
4283 
4284 		bit_off = offset;
4285 		bit_bytes = ctl->unit;
4286 		ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false);
4287 		if (!ret) {
4288 			if (bit_off == offset) {
4289 				ret = 1;
4290 				goto out;
4291 			} else if (bit_off > offset &&
4292 				   offset + bytes > bit_off) {
4293 				ret = 1;
4294 				goto out;
4295 			}
4296 		}
4297 
4298 		n = rb_prev(&info->offset_index);
4299 		while (n) {
4300 			tmp = rb_entry(n, struct btrfs_free_space,
4301 				       offset_index);
4302 			if (tmp->offset + tmp->bytes < offset)
4303 				break;
4304 			if (offset + bytes < tmp->offset) {
4305 				n = rb_prev(&tmp->offset_index);
4306 				continue;
4307 			}
4308 			info = tmp;
4309 			goto have_info;
4310 		}
4311 
4312 		n = rb_next(&info->offset_index);
4313 		while (n) {
4314 			tmp = rb_entry(n, struct btrfs_free_space,
4315 				       offset_index);
4316 			if (offset + bytes < tmp->offset)
4317 				break;
4318 			if (tmp->offset + tmp->bytes < offset) {
4319 				n = rb_next(&tmp->offset_index);
4320 				continue;
4321 			}
4322 			info = tmp;
4323 			goto have_info;
4324 		}
4325 
4326 		ret = 0;
4327 		goto out;
4328 	}
4329 
4330 	if (info->offset == offset) {
4331 		ret = 1;
4332 		goto out;
4333 	}
4334 
4335 	if (offset > info->offset && offset < info->offset + info->bytes)
4336 		ret = 1;
4337 out:
4338 	spin_unlock(&ctl->tree_lock);
4339 	return ret;
4340 }
4341 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */
4342