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 "ctree.h"
15 #include "fs.h"
16 #include "messages.h"
17 #include "misc.h"
18 #include "free-space-cache.h"
19 #include "transaction.h"
20 #include "disk-io.h"
21 #include "extent_io.h"
22 #include "volumes.h"
23 #include "space-info.h"
24 #include "delalloc-space.h"
25 #include "block-group.h"
26 #include "discard.h"
27 #include "subpage.h"
28 #include "inode-item.h"
29 #include "accessors.h"
30 #include "file-item.h"
31 #include "file.h"
32 #include "super.h"
33 
34 #define BITS_PER_BITMAP		(PAGE_SIZE * 8UL)
35 #define MAX_CACHE_BYTES_PER_GIG	SZ_64K
36 #define FORCE_EXTENT_THRESHOLD	SZ_1M
37 
38 static struct kmem_cache *btrfs_free_space_cachep;
39 static struct kmem_cache *btrfs_free_space_bitmap_cachep;
40 
41 struct btrfs_trim_range {
42 	u64 start;
43 	u64 bytes;
44 	struct list_head list;
45 };
46 
47 static int link_free_space(struct btrfs_free_space_ctl *ctl,
48 			   struct btrfs_free_space *info);
49 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
50 			      struct btrfs_free_space *info, bool update_stat);
51 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
52 			 struct btrfs_free_space *bitmap_info, u64 *offset,
53 			 u64 *bytes, bool for_alloc);
54 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
55 			struct btrfs_free_space *bitmap_info);
56 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
57 			      struct btrfs_free_space *info, u64 offset,
58 			      u64 bytes, bool update_stats);
59 
__btrfs_remove_free_space_cache(struct btrfs_free_space_ctl * ctl)60 static void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
61 {
62 	struct btrfs_free_space *info;
63 	struct rb_node *node;
64 
65 	while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
66 		info = rb_entry(node, struct btrfs_free_space, offset_index);
67 		if (!info->bitmap) {
68 			unlink_free_space(ctl, info, true);
69 			kmem_cache_free(btrfs_free_space_cachep, info);
70 		} else {
71 			free_bitmap(ctl, info);
72 		}
73 
74 		cond_resched_lock(&ctl->tree_lock);
75 	}
76 }
77 
__lookup_free_space_inode(struct btrfs_root * root,struct btrfs_path * path,u64 offset)78 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
79 					       struct btrfs_path *path,
80 					       u64 offset)
81 {
82 	struct btrfs_fs_info *fs_info = root->fs_info;
83 	struct btrfs_key key;
84 	struct btrfs_key location;
85 	struct btrfs_disk_key disk_key;
86 	struct btrfs_free_space_header *header;
87 	struct extent_buffer *leaf;
88 	struct inode *inode = NULL;
89 	unsigned nofs_flag;
90 	int ret;
91 
92 	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
93 	key.offset = offset;
94 	key.type = 0;
95 
96 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
97 	if (ret < 0)
98 		return ERR_PTR(ret);
99 	if (ret > 0) {
100 		btrfs_release_path(path);
101 		return ERR_PTR(-ENOENT);
102 	}
103 
104 	leaf = path->nodes[0];
105 	header = btrfs_item_ptr(leaf, path->slots[0],
106 				struct btrfs_free_space_header);
107 	btrfs_free_space_key(leaf, header, &disk_key);
108 	btrfs_disk_key_to_cpu(&location, &disk_key);
109 	btrfs_release_path(path);
110 
111 	/*
112 	 * We are often under a trans handle at this point, so we need to make
113 	 * sure NOFS is set to keep us from deadlocking.
114 	 */
115 	nofs_flag = memalloc_nofs_save();
116 	inode = btrfs_iget_path(fs_info->sb, location.objectid, root, path);
117 	btrfs_release_path(path);
118 	memalloc_nofs_restore(nofs_flag);
119 	if (IS_ERR(inode))
120 		return inode;
121 
122 	mapping_set_gfp_mask(inode->i_mapping,
123 			mapping_gfp_constraint(inode->i_mapping,
124 			~(__GFP_FS | __GFP_HIGHMEM)));
125 
126 	return inode;
127 }
128 
lookup_free_space_inode(struct btrfs_block_group * block_group,struct btrfs_path * path)129 struct inode *lookup_free_space_inode(struct btrfs_block_group *block_group,
130 		struct btrfs_path *path)
131 {
132 	struct btrfs_fs_info *fs_info = block_group->fs_info;
133 	struct inode *inode = NULL;
134 	u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
135 
136 	spin_lock(&block_group->lock);
137 	if (block_group->inode)
138 		inode = igrab(block_group->inode);
139 	spin_unlock(&block_group->lock);
140 	if (inode)
141 		return inode;
142 
143 	inode = __lookup_free_space_inode(fs_info->tree_root, path,
144 					  block_group->start);
145 	if (IS_ERR(inode))
146 		return inode;
147 
148 	spin_lock(&block_group->lock);
149 	if (!((BTRFS_I(inode)->flags & flags) == flags)) {
150 		btrfs_info(fs_info, "Old style space inode found, converting.");
151 		BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
152 			BTRFS_INODE_NODATACOW;
153 		block_group->disk_cache_state = BTRFS_DC_CLEAR;
154 	}
155 
156 	if (!test_and_set_bit(BLOCK_GROUP_FLAG_IREF, &block_group->runtime_flags))
157 		block_group->inode = igrab(inode);
158 	spin_unlock(&block_group->lock);
159 
160 	return inode;
161 }
162 
__create_free_space_inode(struct btrfs_root * root,struct btrfs_trans_handle * trans,struct btrfs_path * path,u64 ino,u64 offset)163 static int __create_free_space_inode(struct btrfs_root *root,
164 				     struct btrfs_trans_handle *trans,
165 				     struct btrfs_path *path,
166 				     u64 ino, u64 offset)
167 {
168 	struct btrfs_key key;
169 	struct btrfs_disk_key disk_key;
170 	struct btrfs_free_space_header *header;
171 	struct btrfs_inode_item *inode_item;
172 	struct extent_buffer *leaf;
173 	/* We inline CRCs for the free disk space cache */
174 	const u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC |
175 			  BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
176 	int ret;
177 
178 	ret = btrfs_insert_empty_inode(trans, root, path, ino);
179 	if (ret)
180 		return ret;
181 
182 	leaf = path->nodes[0];
183 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
184 				    struct btrfs_inode_item);
185 	btrfs_item_key(leaf, &disk_key, path->slots[0]);
186 	memzero_extent_buffer(leaf, (unsigned long)inode_item,
187 			     sizeof(*inode_item));
188 	btrfs_set_inode_generation(leaf, inode_item, trans->transid);
189 	btrfs_set_inode_size(leaf, inode_item, 0);
190 	btrfs_set_inode_nbytes(leaf, inode_item, 0);
191 	btrfs_set_inode_uid(leaf, inode_item, 0);
192 	btrfs_set_inode_gid(leaf, inode_item, 0);
193 	btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
194 	btrfs_set_inode_flags(leaf, inode_item, flags);
195 	btrfs_set_inode_nlink(leaf, inode_item, 1);
196 	btrfs_set_inode_transid(leaf, inode_item, trans->transid);
197 	btrfs_set_inode_block_group(leaf, inode_item, offset);
198 	btrfs_mark_buffer_dirty(leaf);
199 	btrfs_release_path(path);
200 
201 	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
202 	key.offset = offset;
203 	key.type = 0;
204 	ret = btrfs_insert_empty_item(trans, root, path, &key,
205 				      sizeof(struct btrfs_free_space_header));
206 	if (ret < 0) {
207 		btrfs_release_path(path);
208 		return ret;
209 	}
210 
211 	leaf = path->nodes[0];
212 	header = btrfs_item_ptr(leaf, path->slots[0],
213 				struct btrfs_free_space_header);
214 	memzero_extent_buffer(leaf, (unsigned long)header, sizeof(*header));
215 	btrfs_set_free_space_key(leaf, header, &disk_key);
216 	btrfs_mark_buffer_dirty(leaf);
217 	btrfs_release_path(path);
218 
219 	return 0;
220 }
221 
create_free_space_inode(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path)222 int create_free_space_inode(struct btrfs_trans_handle *trans,
223 			    struct btrfs_block_group *block_group,
224 			    struct btrfs_path *path)
225 {
226 	int ret;
227 	u64 ino;
228 
229 	ret = btrfs_get_free_objectid(trans->fs_info->tree_root, &ino);
230 	if (ret < 0)
231 		return ret;
232 
233 	return __create_free_space_inode(trans->fs_info->tree_root, trans, path,
234 					 ino, block_group->start);
235 }
236 
237 /*
238  * inode is an optional sink: if it is NULL, btrfs_remove_free_space_inode
239  * handles lookup, otherwise it takes ownership and iputs the inode.
240  * Don't reuse an inode pointer after passing it into this function.
241  */
btrfs_remove_free_space_inode(struct btrfs_trans_handle * trans,struct inode * inode,struct btrfs_block_group * block_group)242 int btrfs_remove_free_space_inode(struct btrfs_trans_handle *trans,
243 				  struct inode *inode,
244 				  struct btrfs_block_group *block_group)
245 {
246 	struct btrfs_path *path;
247 	struct btrfs_key key;
248 	int ret = 0;
249 
250 	path = btrfs_alloc_path();
251 	if (!path)
252 		return -ENOMEM;
253 
254 	if (!inode)
255 		inode = lookup_free_space_inode(block_group, path);
256 	if (IS_ERR(inode)) {
257 		if (PTR_ERR(inode) != -ENOENT)
258 			ret = PTR_ERR(inode);
259 		goto out;
260 	}
261 	ret = btrfs_orphan_add(trans, BTRFS_I(inode));
262 	if (ret) {
263 		btrfs_add_delayed_iput(BTRFS_I(inode));
264 		goto out;
265 	}
266 	clear_nlink(inode);
267 	/* One for the block groups ref */
268 	spin_lock(&block_group->lock);
269 	if (test_and_clear_bit(BLOCK_GROUP_FLAG_IREF, &block_group->runtime_flags)) {
270 		block_group->inode = NULL;
271 		spin_unlock(&block_group->lock);
272 		iput(inode);
273 	} else {
274 		spin_unlock(&block_group->lock);
275 	}
276 	/* One for the lookup ref */
277 	btrfs_add_delayed_iput(BTRFS_I(inode));
278 
279 	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
280 	key.type = 0;
281 	key.offset = block_group->start;
282 	ret = btrfs_search_slot(trans, trans->fs_info->tree_root, &key, path,
283 				-1, 1);
284 	if (ret) {
285 		if (ret > 0)
286 			ret = 0;
287 		goto out;
288 	}
289 	ret = btrfs_del_item(trans, trans->fs_info->tree_root, path);
290 out:
291 	btrfs_free_path(path);
292 	return ret;
293 }
294 
btrfs_truncate_free_space_cache(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct inode * vfs_inode)295 int btrfs_truncate_free_space_cache(struct btrfs_trans_handle *trans,
296 				    struct btrfs_block_group *block_group,
297 				    struct inode *vfs_inode)
298 {
299 	struct btrfs_truncate_control control = {
300 		.inode = BTRFS_I(vfs_inode),
301 		.new_size = 0,
302 		.ino = btrfs_ino(BTRFS_I(vfs_inode)),
303 		.min_type = BTRFS_EXTENT_DATA_KEY,
304 		.clear_extent_range = true,
305 	};
306 	struct btrfs_inode *inode = BTRFS_I(vfs_inode);
307 	struct btrfs_root *root = inode->root;
308 	struct extent_state *cached_state = NULL;
309 	int ret = 0;
310 	bool locked = false;
311 
312 	if (block_group) {
313 		struct btrfs_path *path = btrfs_alloc_path();
314 
315 		if (!path) {
316 			ret = -ENOMEM;
317 			goto fail;
318 		}
319 		locked = true;
320 		mutex_lock(&trans->transaction->cache_write_mutex);
321 		if (!list_empty(&block_group->io_list)) {
322 			list_del_init(&block_group->io_list);
323 
324 			btrfs_wait_cache_io(trans, block_group, path);
325 			btrfs_put_block_group(block_group);
326 		}
327 
328 		/*
329 		 * now that we've truncated the cache away, its no longer
330 		 * setup or written
331 		 */
332 		spin_lock(&block_group->lock);
333 		block_group->disk_cache_state = BTRFS_DC_CLEAR;
334 		spin_unlock(&block_group->lock);
335 		btrfs_free_path(path);
336 	}
337 
338 	btrfs_i_size_write(inode, 0);
339 	truncate_pagecache(vfs_inode, 0);
340 
341 	lock_extent(&inode->io_tree, 0, (u64)-1, &cached_state);
342 	btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
343 
344 	/*
345 	 * We skip the throttling logic for free space cache inodes, so we don't
346 	 * need to check for -EAGAIN.
347 	 */
348 	ret = btrfs_truncate_inode_items(trans, root, &control);
349 
350 	inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
351 	btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
352 
353 	unlock_extent(&inode->io_tree, 0, (u64)-1, &cached_state);
354 	if (ret)
355 		goto fail;
356 
357 	ret = btrfs_update_inode(trans, root, inode);
358 
359 fail:
360 	if (locked)
361 		mutex_unlock(&trans->transaction->cache_write_mutex);
362 	if (ret)
363 		btrfs_abort_transaction(trans, ret);
364 
365 	return ret;
366 }
367 
readahead_cache(struct inode * inode)368 static void readahead_cache(struct inode *inode)
369 {
370 	struct file_ra_state ra;
371 	unsigned long last_index;
372 
373 	file_ra_state_init(&ra, inode->i_mapping);
374 	last_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
375 
376 	page_cache_sync_readahead(inode->i_mapping, &ra, NULL, 0, last_index);
377 }
378 
io_ctl_init(struct btrfs_io_ctl * io_ctl,struct inode * inode,int write)379 static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode,
380 		       int write)
381 {
382 	int num_pages;
383 
384 	num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
385 
386 	/* Make sure we can fit our crcs and generation into the first page */
387 	if (write && (num_pages * sizeof(u32) + sizeof(u64)) > PAGE_SIZE)
388 		return -ENOSPC;
389 
390 	memset(io_ctl, 0, sizeof(struct btrfs_io_ctl));
391 
392 	io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS);
393 	if (!io_ctl->pages)
394 		return -ENOMEM;
395 
396 	io_ctl->num_pages = num_pages;
397 	io_ctl->fs_info = btrfs_sb(inode->i_sb);
398 	io_ctl->inode = inode;
399 
400 	return 0;
401 }
402 ALLOW_ERROR_INJECTION(io_ctl_init, ERRNO);
403 
io_ctl_free(struct btrfs_io_ctl * io_ctl)404 static void io_ctl_free(struct btrfs_io_ctl *io_ctl)
405 {
406 	kfree(io_ctl->pages);
407 	io_ctl->pages = NULL;
408 }
409 
io_ctl_unmap_page(struct btrfs_io_ctl * io_ctl)410 static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl)
411 {
412 	if (io_ctl->cur) {
413 		io_ctl->cur = NULL;
414 		io_ctl->orig = NULL;
415 	}
416 }
417 
io_ctl_map_page(struct btrfs_io_ctl * io_ctl,int clear)418 static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear)
419 {
420 	ASSERT(io_ctl->index < io_ctl->num_pages);
421 	io_ctl->page = io_ctl->pages[io_ctl->index++];
422 	io_ctl->cur = page_address(io_ctl->page);
423 	io_ctl->orig = io_ctl->cur;
424 	io_ctl->size = PAGE_SIZE;
425 	if (clear)
426 		clear_page(io_ctl->cur);
427 }
428 
io_ctl_drop_pages(struct btrfs_io_ctl * io_ctl)429 static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl)
430 {
431 	int i;
432 
433 	io_ctl_unmap_page(io_ctl);
434 
435 	for (i = 0; i < io_ctl->num_pages; i++) {
436 		if (io_ctl->pages[i]) {
437 			btrfs_page_clear_checked(io_ctl->fs_info,
438 					io_ctl->pages[i],
439 					page_offset(io_ctl->pages[i]),
440 					PAGE_SIZE);
441 			unlock_page(io_ctl->pages[i]);
442 			put_page(io_ctl->pages[i]);
443 		}
444 	}
445 }
446 
io_ctl_prepare_pages(struct btrfs_io_ctl * io_ctl,bool uptodate)447 static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, bool uptodate)
448 {
449 	struct page *page;
450 	struct inode *inode = io_ctl->inode;
451 	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
452 	int i;
453 
454 	for (i = 0; i < io_ctl->num_pages; i++) {
455 		int ret;
456 
457 		page = find_or_create_page(inode->i_mapping, i, mask);
458 		if (!page) {
459 			io_ctl_drop_pages(io_ctl);
460 			return -ENOMEM;
461 		}
462 
463 		ret = set_page_extent_mapped(page);
464 		if (ret < 0) {
465 			unlock_page(page);
466 			put_page(page);
467 			io_ctl_drop_pages(io_ctl);
468 			return ret;
469 		}
470 
471 		io_ctl->pages[i] = page;
472 		if (uptodate && !PageUptodate(page)) {
473 			btrfs_read_folio(NULL, page_folio(page));
474 			lock_page(page);
475 			if (page->mapping != inode->i_mapping) {
476 				btrfs_err(BTRFS_I(inode)->root->fs_info,
477 					  "free space cache page truncated");
478 				io_ctl_drop_pages(io_ctl);
479 				return -EIO;
480 			}
481 			if (!PageUptodate(page)) {
482 				btrfs_err(BTRFS_I(inode)->root->fs_info,
483 					   "error reading free space cache");
484 				io_ctl_drop_pages(io_ctl);
485 				return -EIO;
486 			}
487 		}
488 	}
489 
490 	for (i = 0; i < io_ctl->num_pages; i++)
491 		clear_page_dirty_for_io(io_ctl->pages[i]);
492 
493 	return 0;
494 }
495 
io_ctl_set_generation(struct btrfs_io_ctl * io_ctl,u64 generation)496 static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
497 {
498 	io_ctl_map_page(io_ctl, 1);
499 
500 	/*
501 	 * Skip the csum areas.  If we don't check crcs then we just have a
502 	 * 64bit chunk at the front of the first page.
503 	 */
504 	io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
505 	io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
506 
507 	put_unaligned_le64(generation, io_ctl->cur);
508 	io_ctl->cur += sizeof(u64);
509 }
510 
io_ctl_check_generation(struct btrfs_io_ctl * io_ctl,u64 generation)511 static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
512 {
513 	u64 cache_gen;
514 
515 	/*
516 	 * Skip the crc area.  If we don't check crcs then we just have a 64bit
517 	 * chunk at the front of the first page.
518 	 */
519 	io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
520 	io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
521 
522 	cache_gen = get_unaligned_le64(io_ctl->cur);
523 	if (cache_gen != generation) {
524 		btrfs_err_rl(io_ctl->fs_info,
525 			"space cache generation (%llu) does not match inode (%llu)",
526 				cache_gen, generation);
527 		io_ctl_unmap_page(io_ctl);
528 		return -EIO;
529 	}
530 	io_ctl->cur += sizeof(u64);
531 	return 0;
532 }
533 
io_ctl_set_crc(struct btrfs_io_ctl * io_ctl,int index)534 static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
535 {
536 	u32 *tmp;
537 	u32 crc = ~(u32)0;
538 	unsigned offset = 0;
539 
540 	if (index == 0)
541 		offset = sizeof(u32) * io_ctl->num_pages;
542 
543 	crc = btrfs_crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
544 	btrfs_crc32c_final(crc, (u8 *)&crc);
545 	io_ctl_unmap_page(io_ctl);
546 	tmp = page_address(io_ctl->pages[0]);
547 	tmp += index;
548 	*tmp = crc;
549 }
550 
io_ctl_check_crc(struct btrfs_io_ctl * io_ctl,int index)551 static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
552 {
553 	u32 *tmp, val;
554 	u32 crc = ~(u32)0;
555 	unsigned offset = 0;
556 
557 	if (index == 0)
558 		offset = sizeof(u32) * io_ctl->num_pages;
559 
560 	tmp = page_address(io_ctl->pages[0]);
561 	tmp += index;
562 	val = *tmp;
563 
564 	io_ctl_map_page(io_ctl, 0);
565 	crc = btrfs_crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
566 	btrfs_crc32c_final(crc, (u8 *)&crc);
567 	if (val != crc) {
568 		btrfs_err_rl(io_ctl->fs_info,
569 			"csum mismatch on free space cache");
570 		io_ctl_unmap_page(io_ctl);
571 		return -EIO;
572 	}
573 
574 	return 0;
575 }
576 
io_ctl_add_entry(struct btrfs_io_ctl * io_ctl,u64 offset,u64 bytes,void * bitmap)577 static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
578 			    void *bitmap)
579 {
580 	struct btrfs_free_space_entry *entry;
581 
582 	if (!io_ctl->cur)
583 		return -ENOSPC;
584 
585 	entry = io_ctl->cur;
586 	put_unaligned_le64(offset, &entry->offset);
587 	put_unaligned_le64(bytes, &entry->bytes);
588 	entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
589 		BTRFS_FREE_SPACE_EXTENT;
590 	io_ctl->cur += sizeof(struct btrfs_free_space_entry);
591 	io_ctl->size -= sizeof(struct btrfs_free_space_entry);
592 
593 	if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
594 		return 0;
595 
596 	io_ctl_set_crc(io_ctl, io_ctl->index - 1);
597 
598 	/* No more pages to map */
599 	if (io_ctl->index >= io_ctl->num_pages)
600 		return 0;
601 
602 	/* map the next page */
603 	io_ctl_map_page(io_ctl, 1);
604 	return 0;
605 }
606 
io_ctl_add_bitmap(struct btrfs_io_ctl * io_ctl,void * bitmap)607 static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
608 {
609 	if (!io_ctl->cur)
610 		return -ENOSPC;
611 
612 	/*
613 	 * If we aren't at the start of the current page, unmap this one and
614 	 * map the next one if there is any left.
615 	 */
616 	if (io_ctl->cur != io_ctl->orig) {
617 		io_ctl_set_crc(io_ctl, io_ctl->index - 1);
618 		if (io_ctl->index >= io_ctl->num_pages)
619 			return -ENOSPC;
620 		io_ctl_map_page(io_ctl, 0);
621 	}
622 
623 	copy_page(io_ctl->cur, bitmap);
624 	io_ctl_set_crc(io_ctl, io_ctl->index - 1);
625 	if (io_ctl->index < io_ctl->num_pages)
626 		io_ctl_map_page(io_ctl, 0);
627 	return 0;
628 }
629 
io_ctl_zero_remaining_pages(struct btrfs_io_ctl * io_ctl)630 static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
631 {
632 	/*
633 	 * If we're not on the boundary we know we've modified the page and we
634 	 * need to crc the page.
635 	 */
636 	if (io_ctl->cur != io_ctl->orig)
637 		io_ctl_set_crc(io_ctl, io_ctl->index - 1);
638 	else
639 		io_ctl_unmap_page(io_ctl);
640 
641 	while (io_ctl->index < io_ctl->num_pages) {
642 		io_ctl_map_page(io_ctl, 1);
643 		io_ctl_set_crc(io_ctl, io_ctl->index - 1);
644 	}
645 }
646 
io_ctl_read_entry(struct btrfs_io_ctl * io_ctl,struct btrfs_free_space * entry,u8 * type)647 static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
648 			    struct btrfs_free_space *entry, u8 *type)
649 {
650 	struct btrfs_free_space_entry *e;
651 	int ret;
652 
653 	if (!io_ctl->cur) {
654 		ret = io_ctl_check_crc(io_ctl, io_ctl->index);
655 		if (ret)
656 			return ret;
657 	}
658 
659 	e = io_ctl->cur;
660 	entry->offset = get_unaligned_le64(&e->offset);
661 	entry->bytes = get_unaligned_le64(&e->bytes);
662 	*type = e->type;
663 	io_ctl->cur += sizeof(struct btrfs_free_space_entry);
664 	io_ctl->size -= sizeof(struct btrfs_free_space_entry);
665 
666 	if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
667 		return 0;
668 
669 	io_ctl_unmap_page(io_ctl);
670 
671 	return 0;
672 }
673 
io_ctl_read_bitmap(struct btrfs_io_ctl * io_ctl,struct btrfs_free_space * entry)674 static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
675 			      struct btrfs_free_space *entry)
676 {
677 	int ret;
678 
679 	ret = io_ctl_check_crc(io_ctl, io_ctl->index);
680 	if (ret)
681 		return ret;
682 
683 	copy_page(entry->bitmap, io_ctl->cur);
684 	io_ctl_unmap_page(io_ctl);
685 
686 	return 0;
687 }
688 
recalculate_thresholds(struct btrfs_free_space_ctl * ctl)689 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
690 {
691 	struct btrfs_block_group *block_group = ctl->block_group;
692 	u64 max_bytes;
693 	u64 bitmap_bytes;
694 	u64 extent_bytes;
695 	u64 size = block_group->length;
696 	u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
697 	u64 max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
698 
699 	max_bitmaps = max_t(u64, max_bitmaps, 1);
700 
701 	if (ctl->total_bitmaps > max_bitmaps)
702 		btrfs_err(block_group->fs_info,
703 "invalid free space control: bg start=%llu len=%llu total_bitmaps=%u unit=%u max_bitmaps=%llu bytes_per_bg=%llu",
704 			  block_group->start, block_group->length,
705 			  ctl->total_bitmaps, ctl->unit, max_bitmaps,
706 			  bytes_per_bg);
707 	ASSERT(ctl->total_bitmaps <= max_bitmaps);
708 
709 	/*
710 	 * We are trying to keep the total amount of memory used per 1GiB of
711 	 * space to be MAX_CACHE_BYTES_PER_GIG.  However, with a reclamation
712 	 * mechanism of pulling extents >= FORCE_EXTENT_THRESHOLD out of
713 	 * bitmaps, we may end up using more memory than this.
714 	 */
715 	if (size < SZ_1G)
716 		max_bytes = MAX_CACHE_BYTES_PER_GIG;
717 	else
718 		max_bytes = MAX_CACHE_BYTES_PER_GIG * div_u64(size, SZ_1G);
719 
720 	bitmap_bytes = ctl->total_bitmaps * ctl->unit;
721 
722 	/*
723 	 * we want the extent entry threshold to always be at most 1/2 the max
724 	 * bytes we can have, or whatever is less than that.
725 	 */
726 	extent_bytes = max_bytes - bitmap_bytes;
727 	extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
728 
729 	ctl->extents_thresh =
730 		div_u64(extent_bytes, sizeof(struct btrfs_free_space));
731 }
732 
__load_free_space_cache(struct btrfs_root * root,struct inode * inode,struct btrfs_free_space_ctl * ctl,struct btrfs_path * path,u64 offset)733 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
734 				   struct btrfs_free_space_ctl *ctl,
735 				   struct btrfs_path *path, u64 offset)
736 {
737 	struct btrfs_fs_info *fs_info = root->fs_info;
738 	struct btrfs_free_space_header *header;
739 	struct extent_buffer *leaf;
740 	struct btrfs_io_ctl io_ctl;
741 	struct btrfs_key key;
742 	struct btrfs_free_space *e, *n;
743 	LIST_HEAD(bitmaps);
744 	u64 num_entries;
745 	u64 num_bitmaps;
746 	u64 generation;
747 	u8 type;
748 	int ret = 0;
749 
750 	/* Nothing in the space cache, goodbye */
751 	if (!i_size_read(inode))
752 		return 0;
753 
754 	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
755 	key.offset = offset;
756 	key.type = 0;
757 
758 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
759 	if (ret < 0)
760 		return 0;
761 	else if (ret > 0) {
762 		btrfs_release_path(path);
763 		return 0;
764 	}
765 
766 	ret = -1;
767 
768 	leaf = path->nodes[0];
769 	header = btrfs_item_ptr(leaf, path->slots[0],
770 				struct btrfs_free_space_header);
771 	num_entries = btrfs_free_space_entries(leaf, header);
772 	num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
773 	generation = btrfs_free_space_generation(leaf, header);
774 	btrfs_release_path(path);
775 
776 	if (!BTRFS_I(inode)->generation) {
777 		btrfs_info(fs_info,
778 			   "the free space cache file (%llu) is invalid, skip it",
779 			   offset);
780 		return 0;
781 	}
782 
783 	if (BTRFS_I(inode)->generation != generation) {
784 		btrfs_err(fs_info,
785 			  "free space inode generation (%llu) did not match free space cache generation (%llu)",
786 			  BTRFS_I(inode)->generation, generation);
787 		return 0;
788 	}
789 
790 	if (!num_entries)
791 		return 0;
792 
793 	ret = io_ctl_init(&io_ctl, inode, 0);
794 	if (ret)
795 		return ret;
796 
797 	readahead_cache(inode);
798 
799 	ret = io_ctl_prepare_pages(&io_ctl, true);
800 	if (ret)
801 		goto out;
802 
803 	ret = io_ctl_check_crc(&io_ctl, 0);
804 	if (ret)
805 		goto free_cache;
806 
807 	ret = io_ctl_check_generation(&io_ctl, generation);
808 	if (ret)
809 		goto free_cache;
810 
811 	while (num_entries) {
812 		e = kmem_cache_zalloc(btrfs_free_space_cachep,
813 				      GFP_NOFS);
814 		if (!e) {
815 			ret = -ENOMEM;
816 			goto free_cache;
817 		}
818 
819 		ret = io_ctl_read_entry(&io_ctl, e, &type);
820 		if (ret) {
821 			kmem_cache_free(btrfs_free_space_cachep, e);
822 			goto free_cache;
823 		}
824 
825 		if (!e->bytes) {
826 			ret = -1;
827 			kmem_cache_free(btrfs_free_space_cachep, e);
828 			goto free_cache;
829 		}
830 
831 		if (type == BTRFS_FREE_SPACE_EXTENT) {
832 			spin_lock(&ctl->tree_lock);
833 			ret = link_free_space(ctl, e);
834 			spin_unlock(&ctl->tree_lock);
835 			if (ret) {
836 				btrfs_err(fs_info,
837 					"Duplicate entries in free space cache, dumping");
838 				kmem_cache_free(btrfs_free_space_cachep, e);
839 				goto free_cache;
840 			}
841 		} else {
842 			ASSERT(num_bitmaps);
843 			num_bitmaps--;
844 			e->bitmap = kmem_cache_zalloc(
845 					btrfs_free_space_bitmap_cachep, GFP_NOFS);
846 			if (!e->bitmap) {
847 				ret = -ENOMEM;
848 				kmem_cache_free(
849 					btrfs_free_space_cachep, e);
850 				goto free_cache;
851 			}
852 			spin_lock(&ctl->tree_lock);
853 			ret = link_free_space(ctl, e);
854 			if (ret) {
855 				spin_unlock(&ctl->tree_lock);
856 				btrfs_err(fs_info,
857 					"Duplicate entries in free space cache, dumping");
858 				kmem_cache_free(btrfs_free_space_cachep, e);
859 				goto free_cache;
860 			}
861 			ctl->total_bitmaps++;
862 			recalculate_thresholds(ctl);
863 			spin_unlock(&ctl->tree_lock);
864 			list_add_tail(&e->list, &bitmaps);
865 		}
866 
867 		num_entries--;
868 	}
869 
870 	io_ctl_unmap_page(&io_ctl);
871 
872 	/*
873 	 * We add the bitmaps at the end of the entries in order that
874 	 * the bitmap entries are added to the cache.
875 	 */
876 	list_for_each_entry_safe(e, n, &bitmaps, list) {
877 		list_del_init(&e->list);
878 		ret = io_ctl_read_bitmap(&io_ctl, e);
879 		if (ret)
880 			goto free_cache;
881 	}
882 
883 	io_ctl_drop_pages(&io_ctl);
884 	ret = 1;
885 out:
886 	io_ctl_free(&io_ctl);
887 	return ret;
888 free_cache:
889 	io_ctl_drop_pages(&io_ctl);
890 
891 	spin_lock(&ctl->tree_lock);
892 	__btrfs_remove_free_space_cache(ctl);
893 	spin_unlock(&ctl->tree_lock);
894 	goto out;
895 }
896 
copy_free_space_cache(struct btrfs_block_group * block_group,struct btrfs_free_space_ctl * ctl)897 static int copy_free_space_cache(struct btrfs_block_group *block_group,
898 				 struct btrfs_free_space_ctl *ctl)
899 {
900 	struct btrfs_free_space *info;
901 	struct rb_node *n;
902 	int ret = 0;
903 
904 	while (!ret && (n = rb_first(&ctl->free_space_offset)) != NULL) {
905 		info = rb_entry(n, struct btrfs_free_space, offset_index);
906 		if (!info->bitmap) {
907 			const u64 offset = info->offset;
908 			const u64 bytes = info->bytes;
909 
910 			unlink_free_space(ctl, info, true);
911 			spin_unlock(&ctl->tree_lock);
912 			kmem_cache_free(btrfs_free_space_cachep, info);
913 			ret = btrfs_add_free_space(block_group, offset, bytes);
914 			spin_lock(&ctl->tree_lock);
915 		} else {
916 			u64 offset = info->offset;
917 			u64 bytes = ctl->unit;
918 
919 			ret = search_bitmap(ctl, info, &offset, &bytes, false);
920 			if (ret == 0) {
921 				bitmap_clear_bits(ctl, info, offset, bytes, true);
922 				spin_unlock(&ctl->tree_lock);
923 				ret = btrfs_add_free_space(block_group, offset,
924 							   bytes);
925 				spin_lock(&ctl->tree_lock);
926 			} else {
927 				free_bitmap(ctl, info);
928 				ret = 0;
929 			}
930 		}
931 		cond_resched_lock(&ctl->tree_lock);
932 	}
933 	return ret;
934 }
935 
936 static struct lock_class_key btrfs_free_space_inode_key;
937 
load_free_space_cache(struct btrfs_block_group * block_group)938 int load_free_space_cache(struct btrfs_block_group *block_group)
939 {
940 	struct btrfs_fs_info *fs_info = block_group->fs_info;
941 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
942 	struct btrfs_free_space_ctl tmp_ctl = {};
943 	struct inode *inode;
944 	struct btrfs_path *path;
945 	int ret = 0;
946 	bool matched;
947 	u64 used = block_group->used;
948 
949 	/*
950 	 * Because we could potentially discard our loaded free space, we want
951 	 * to load everything into a temporary structure first, and then if it's
952 	 * valid copy it all into the actual free space ctl.
953 	 */
954 	btrfs_init_free_space_ctl(block_group, &tmp_ctl);
955 
956 	/*
957 	 * If this block group has been marked to be cleared for one reason or
958 	 * another then we can't trust the on disk cache, so just return.
959 	 */
960 	spin_lock(&block_group->lock);
961 	if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
962 		spin_unlock(&block_group->lock);
963 		return 0;
964 	}
965 	spin_unlock(&block_group->lock);
966 
967 	path = btrfs_alloc_path();
968 	if (!path)
969 		return 0;
970 	path->search_commit_root = 1;
971 	path->skip_locking = 1;
972 
973 	/*
974 	 * We must pass a path with search_commit_root set to btrfs_iget in
975 	 * order to avoid a deadlock when allocating extents for the tree root.
976 	 *
977 	 * When we are COWing an extent buffer from the tree root, when looking
978 	 * for a free extent, at extent-tree.c:find_free_extent(), we can find
979 	 * block group without its free space cache loaded. When we find one
980 	 * we must load its space cache which requires reading its free space
981 	 * cache's inode item from the root tree. If this inode item is located
982 	 * in the same leaf that we started COWing before, then we end up in
983 	 * deadlock on the extent buffer (trying to read lock it when we
984 	 * previously write locked it).
985 	 *
986 	 * It's safe to read the inode item using the commit root because
987 	 * block groups, once loaded, stay in memory forever (until they are
988 	 * removed) as well as their space caches once loaded. New block groups
989 	 * once created get their ->cached field set to BTRFS_CACHE_FINISHED so
990 	 * we will never try to read their inode item while the fs is mounted.
991 	 */
992 	inode = lookup_free_space_inode(block_group, path);
993 	if (IS_ERR(inode)) {
994 		btrfs_free_path(path);
995 		return 0;
996 	}
997 
998 	/* We may have converted the inode and made the cache invalid. */
999 	spin_lock(&block_group->lock);
1000 	if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
1001 		spin_unlock(&block_group->lock);
1002 		btrfs_free_path(path);
1003 		goto out;
1004 	}
1005 	spin_unlock(&block_group->lock);
1006 
1007 	/*
1008 	 * Reinitialize the class of struct inode's mapping->invalidate_lock for
1009 	 * free space inodes to prevent false positives related to locks for normal
1010 	 * inodes.
1011 	 */
1012 	lockdep_set_class(&(&inode->i_data)->invalidate_lock,
1013 			  &btrfs_free_space_inode_key);
1014 
1015 	ret = __load_free_space_cache(fs_info->tree_root, inode, &tmp_ctl,
1016 				      path, block_group->start);
1017 	btrfs_free_path(path);
1018 	if (ret <= 0)
1019 		goto out;
1020 
1021 	matched = (tmp_ctl.free_space == (block_group->length - used -
1022 					  block_group->bytes_super));
1023 
1024 	if (matched) {
1025 		spin_lock(&tmp_ctl.tree_lock);
1026 		ret = copy_free_space_cache(block_group, &tmp_ctl);
1027 		spin_unlock(&tmp_ctl.tree_lock);
1028 		/*
1029 		 * ret == 1 means we successfully loaded the free space cache,
1030 		 * so we need to re-set it here.
1031 		 */
1032 		if (ret == 0)
1033 			ret = 1;
1034 	} else {
1035 		/*
1036 		 * We need to call the _locked variant so we don't try to update
1037 		 * the discard counters.
1038 		 */
1039 		spin_lock(&tmp_ctl.tree_lock);
1040 		__btrfs_remove_free_space_cache(&tmp_ctl);
1041 		spin_unlock(&tmp_ctl.tree_lock);
1042 		btrfs_warn(fs_info,
1043 			   "block group %llu has wrong amount of free space",
1044 			   block_group->start);
1045 		ret = -1;
1046 	}
1047 out:
1048 	if (ret < 0) {
1049 		/* This cache is bogus, make sure it gets cleared */
1050 		spin_lock(&block_group->lock);
1051 		block_group->disk_cache_state = BTRFS_DC_CLEAR;
1052 		spin_unlock(&block_group->lock);
1053 		ret = 0;
1054 
1055 		btrfs_warn(fs_info,
1056 			   "failed to load free space cache for block group %llu, rebuilding it now",
1057 			   block_group->start);
1058 	}
1059 
1060 	spin_lock(&ctl->tree_lock);
1061 	btrfs_discard_update_discardable(block_group);
1062 	spin_unlock(&ctl->tree_lock);
1063 	iput(inode);
1064 	return ret;
1065 }
1066 
1067 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)1068 int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
1069 			      struct btrfs_free_space_ctl *ctl,
1070 			      struct btrfs_block_group *block_group,
1071 			      int *entries, int *bitmaps,
1072 			      struct list_head *bitmap_list)
1073 {
1074 	int ret;
1075 	struct btrfs_free_cluster *cluster = NULL;
1076 	struct btrfs_free_cluster *cluster_locked = NULL;
1077 	struct rb_node *node = rb_first(&ctl->free_space_offset);
1078 	struct btrfs_trim_range *trim_entry;
1079 
1080 	/* Get the cluster for this block_group if it exists */
1081 	if (block_group && !list_empty(&block_group->cluster_list)) {
1082 		cluster = list_entry(block_group->cluster_list.next,
1083 				     struct btrfs_free_cluster,
1084 				     block_group_list);
1085 	}
1086 
1087 	if (!node && cluster) {
1088 		cluster_locked = cluster;
1089 		spin_lock(&cluster_locked->lock);
1090 		node = rb_first(&cluster->root);
1091 		cluster = NULL;
1092 	}
1093 
1094 	/* Write out the extent entries */
1095 	while (node) {
1096 		struct btrfs_free_space *e;
1097 
1098 		e = rb_entry(node, struct btrfs_free_space, offset_index);
1099 		*entries += 1;
1100 
1101 		ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
1102 				       e->bitmap);
1103 		if (ret)
1104 			goto fail;
1105 
1106 		if (e->bitmap) {
1107 			list_add_tail(&e->list, bitmap_list);
1108 			*bitmaps += 1;
1109 		}
1110 		node = rb_next(node);
1111 		if (!node && cluster) {
1112 			node = rb_first(&cluster->root);
1113 			cluster_locked = cluster;
1114 			spin_lock(&cluster_locked->lock);
1115 			cluster = NULL;
1116 		}
1117 	}
1118 	if (cluster_locked) {
1119 		spin_unlock(&cluster_locked->lock);
1120 		cluster_locked = NULL;
1121 	}
1122 
1123 	/*
1124 	 * Make sure we don't miss any range that was removed from our rbtree
1125 	 * because trimming is running. Otherwise after a umount+mount (or crash
1126 	 * after committing the transaction) we would leak free space and get
1127 	 * an inconsistent free space cache report from fsck.
1128 	 */
1129 	list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
1130 		ret = io_ctl_add_entry(io_ctl, trim_entry->start,
1131 				       trim_entry->bytes, NULL);
1132 		if (ret)
1133 			goto fail;
1134 		*entries += 1;
1135 	}
1136 
1137 	return 0;
1138 fail:
1139 	if (cluster_locked)
1140 		spin_unlock(&cluster_locked->lock);
1141 	return -ENOSPC;
1142 }
1143 
1144 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)1145 update_cache_item(struct btrfs_trans_handle *trans,
1146 		  struct btrfs_root *root,
1147 		  struct inode *inode,
1148 		  struct btrfs_path *path, u64 offset,
1149 		  int entries, int bitmaps)
1150 {
1151 	struct btrfs_key key;
1152 	struct btrfs_free_space_header *header;
1153 	struct extent_buffer *leaf;
1154 	int ret;
1155 
1156 	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1157 	key.offset = offset;
1158 	key.type = 0;
1159 
1160 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1161 	if (ret < 0) {
1162 		clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1163 				 EXTENT_DELALLOC, NULL);
1164 		goto fail;
1165 	}
1166 	leaf = path->nodes[0];
1167 	if (ret > 0) {
1168 		struct btrfs_key found_key;
1169 		ASSERT(path->slots[0]);
1170 		path->slots[0]--;
1171 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1172 		if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1173 		    found_key.offset != offset) {
1174 			clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1175 					 inode->i_size - 1, EXTENT_DELALLOC,
1176 					 NULL);
1177 			btrfs_release_path(path);
1178 			goto fail;
1179 		}
1180 	}
1181 
1182 	BTRFS_I(inode)->generation = trans->transid;
1183 	header = btrfs_item_ptr(leaf, path->slots[0],
1184 				struct btrfs_free_space_header);
1185 	btrfs_set_free_space_entries(leaf, header, entries);
1186 	btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1187 	btrfs_set_free_space_generation(leaf, header, trans->transid);
1188 	btrfs_mark_buffer_dirty(leaf);
1189 	btrfs_release_path(path);
1190 
1191 	return 0;
1192 
1193 fail:
1194 	return -1;
1195 }
1196 
write_pinned_extent_entries(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_io_ctl * io_ctl,int * entries)1197 static noinline_for_stack int write_pinned_extent_entries(
1198 			    struct btrfs_trans_handle *trans,
1199 			    struct btrfs_block_group *block_group,
1200 			    struct btrfs_io_ctl *io_ctl,
1201 			    int *entries)
1202 {
1203 	u64 start, extent_start, extent_end, len;
1204 	struct extent_io_tree *unpin = NULL;
1205 	int ret;
1206 
1207 	if (!block_group)
1208 		return 0;
1209 
1210 	/*
1211 	 * We want to add any pinned extents to our free space cache
1212 	 * so we don't leak the space
1213 	 *
1214 	 * We shouldn't have switched the pinned extents yet so this is the
1215 	 * right one
1216 	 */
1217 	unpin = &trans->transaction->pinned_extents;
1218 
1219 	start = block_group->start;
1220 
1221 	while (start < block_group->start + block_group->length) {
1222 		if (!find_first_extent_bit(unpin, start,
1223 					   &extent_start, &extent_end,
1224 					   EXTENT_DIRTY, NULL))
1225 			return 0;
1226 
1227 		/* This pinned extent is out of our range */
1228 		if (extent_start >= block_group->start + block_group->length)
1229 			return 0;
1230 
1231 		extent_start = max(extent_start, start);
1232 		extent_end = min(block_group->start + block_group->length,
1233 				 extent_end + 1);
1234 		len = extent_end - extent_start;
1235 
1236 		*entries += 1;
1237 		ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1238 		if (ret)
1239 			return -ENOSPC;
1240 
1241 		start = extent_end;
1242 	}
1243 
1244 	return 0;
1245 }
1246 
1247 static noinline_for_stack int
write_bitmap_entries(struct btrfs_io_ctl * io_ctl,struct list_head * bitmap_list)1248 write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1249 {
1250 	struct btrfs_free_space *entry, *next;
1251 	int ret;
1252 
1253 	/* Write out the bitmaps */
1254 	list_for_each_entry_safe(entry, next, bitmap_list, list) {
1255 		ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1256 		if (ret)
1257 			return -ENOSPC;
1258 		list_del_init(&entry->list);
1259 	}
1260 
1261 	return 0;
1262 }
1263 
flush_dirty_cache(struct inode * inode)1264 static int flush_dirty_cache(struct inode *inode)
1265 {
1266 	int ret;
1267 
1268 	ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1269 	if (ret)
1270 		clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1271 				 EXTENT_DELALLOC, NULL);
1272 
1273 	return ret;
1274 }
1275 
1276 static void noinline_for_stack
cleanup_bitmap_list(struct list_head * bitmap_list)1277 cleanup_bitmap_list(struct list_head *bitmap_list)
1278 {
1279 	struct btrfs_free_space *entry, *next;
1280 
1281 	list_for_each_entry_safe(entry, next, bitmap_list, list)
1282 		list_del_init(&entry->list);
1283 }
1284 
1285 static void noinline_for_stack
cleanup_write_cache_enospc(struct inode * inode,struct btrfs_io_ctl * io_ctl,struct extent_state ** cached_state)1286 cleanup_write_cache_enospc(struct inode *inode,
1287 			   struct btrfs_io_ctl *io_ctl,
1288 			   struct extent_state **cached_state)
1289 {
1290 	io_ctl_drop_pages(io_ctl);
1291 	unlock_extent(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1292 		      cached_state);
1293 }
1294 
__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)1295 static int __btrfs_wait_cache_io(struct btrfs_root *root,
1296 				 struct btrfs_trans_handle *trans,
1297 				 struct btrfs_block_group *block_group,
1298 				 struct btrfs_io_ctl *io_ctl,
1299 				 struct btrfs_path *path, u64 offset)
1300 {
1301 	int ret;
1302 	struct inode *inode = io_ctl->inode;
1303 
1304 	if (!inode)
1305 		return 0;
1306 
1307 	/* Flush the dirty pages in the cache file. */
1308 	ret = flush_dirty_cache(inode);
1309 	if (ret)
1310 		goto out;
1311 
1312 	/* Update the cache item to tell everyone this cache file is valid. */
1313 	ret = update_cache_item(trans, root, inode, path, offset,
1314 				io_ctl->entries, io_ctl->bitmaps);
1315 out:
1316 	if (ret) {
1317 		invalidate_inode_pages2(inode->i_mapping);
1318 		BTRFS_I(inode)->generation = 0;
1319 		if (block_group)
1320 			btrfs_debug(root->fs_info,
1321 	  "failed to write free space cache for block group %llu error %d",
1322 				  block_group->start, ret);
1323 	}
1324 	btrfs_update_inode(trans, root, BTRFS_I(inode));
1325 
1326 	if (block_group) {
1327 		/* the dirty list is protected by the dirty_bgs_lock */
1328 		spin_lock(&trans->transaction->dirty_bgs_lock);
1329 
1330 		/* the disk_cache_state is protected by the block group lock */
1331 		spin_lock(&block_group->lock);
1332 
1333 		/*
1334 		 * only mark this as written if we didn't get put back on
1335 		 * the dirty list while waiting for IO.   Otherwise our
1336 		 * cache state won't be right, and we won't get written again
1337 		 */
1338 		if (!ret && list_empty(&block_group->dirty_list))
1339 			block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1340 		else if (ret)
1341 			block_group->disk_cache_state = BTRFS_DC_ERROR;
1342 
1343 		spin_unlock(&block_group->lock);
1344 		spin_unlock(&trans->transaction->dirty_bgs_lock);
1345 		io_ctl->inode = NULL;
1346 		iput(inode);
1347 	}
1348 
1349 	return ret;
1350 
1351 }
1352 
btrfs_wait_cache_io(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path)1353 int btrfs_wait_cache_io(struct btrfs_trans_handle *trans,
1354 			struct btrfs_block_group *block_group,
1355 			struct btrfs_path *path)
1356 {
1357 	return __btrfs_wait_cache_io(block_group->fs_info->tree_root, trans,
1358 				     block_group, &block_group->io_ctl,
1359 				     path, block_group->start);
1360 }
1361 
1362 /*
1363  * Write out cached info to an inode.
1364  *
1365  * @root:        root the inode belongs to
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 btrfs_root * root,struct inode * inode,struct btrfs_free_space_ctl * ctl,struct btrfs_block_group * block_group,struct btrfs_io_ctl * io_ctl,struct btrfs_trans_handle * trans)1376 static int __btrfs_write_out_cache(struct btrfs_root *root, 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 
1389 	if (!i_size_read(inode))
1390 		return -EIO;
1391 
1392 	WARN_ON(io_ctl->pages);
1393 	ret = io_ctl_init(io_ctl, inode, 1);
1394 	if (ret)
1395 		return ret;
1396 
1397 	if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1398 		down_write(&block_group->data_rwsem);
1399 		spin_lock(&block_group->lock);
1400 		if (block_group->delalloc_bytes) {
1401 			block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1402 			spin_unlock(&block_group->lock);
1403 			up_write(&block_group->data_rwsem);
1404 			BTRFS_I(inode)->generation = 0;
1405 			ret = 0;
1406 			must_iput = 1;
1407 			goto out;
1408 		}
1409 		spin_unlock(&block_group->lock);
1410 	}
1411 
1412 	/* Lock all pages first so we can lock the extent safely. */
1413 	ret = io_ctl_prepare_pages(io_ctl, false);
1414 	if (ret)
1415 		goto out_unlock;
1416 
1417 	lock_extent(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1418 		    &cached_state);
1419 
1420 	io_ctl_set_generation(io_ctl, trans->transid);
1421 
1422 	mutex_lock(&ctl->cache_writeout_mutex);
1423 	/* Write out the extent entries in the free space cache */
1424 	spin_lock(&ctl->tree_lock);
1425 	ret = write_cache_extent_entries(io_ctl, ctl,
1426 					 block_group, &entries, &bitmaps,
1427 					 &bitmap_list);
1428 	if (ret)
1429 		goto out_nospc_locked;
1430 
1431 	/*
1432 	 * Some spaces that are freed in the current transaction are pinned,
1433 	 * they will be added into free space cache after the transaction is
1434 	 * committed, we shouldn't lose them.
1435 	 *
1436 	 * If this changes while we are working we'll get added back to
1437 	 * the dirty list and redo it.  No locking needed
1438 	 */
1439 	ret = write_pinned_extent_entries(trans, block_group, io_ctl, &entries);
1440 	if (ret)
1441 		goto out_nospc_locked;
1442 
1443 	/*
1444 	 * At last, we write out all the bitmaps and keep cache_writeout_mutex
1445 	 * locked while doing it because a concurrent trim can be manipulating
1446 	 * or freeing the bitmap.
1447 	 */
1448 	ret = write_bitmap_entries(io_ctl, &bitmap_list);
1449 	spin_unlock(&ctl->tree_lock);
1450 	mutex_unlock(&ctl->cache_writeout_mutex);
1451 	if (ret)
1452 		goto out_nospc;
1453 
1454 	/* Zero out the rest of the pages just to make sure */
1455 	io_ctl_zero_remaining_pages(io_ctl);
1456 
1457 	/* Everything is written out, now we dirty the pages in the file. */
1458 	ret = btrfs_dirty_pages(BTRFS_I(inode), io_ctl->pages,
1459 				io_ctl->num_pages, 0, i_size_read(inode),
1460 				&cached_state, false);
1461 	if (ret)
1462 		goto out_nospc;
1463 
1464 	if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1465 		up_write(&block_group->data_rwsem);
1466 	/*
1467 	 * Release the pages and unlock the extent, we will flush
1468 	 * them out later
1469 	 */
1470 	io_ctl_drop_pages(io_ctl);
1471 	io_ctl_free(io_ctl);
1472 
1473 	unlock_extent(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1474 		      &cached_state);
1475 
1476 	/*
1477 	 * at this point the pages are under IO and we're happy,
1478 	 * The caller is responsible for waiting on them and updating
1479 	 * the cache and the inode
1480 	 */
1481 	io_ctl->entries = entries;
1482 	io_ctl->bitmaps = bitmaps;
1483 
1484 	ret = btrfs_fdatawrite_range(inode, 0, (u64)-1);
1485 	if (ret)
1486 		goto out;
1487 
1488 	return 0;
1489 
1490 out_nospc_locked:
1491 	cleanup_bitmap_list(&bitmap_list);
1492 	spin_unlock(&ctl->tree_lock);
1493 	mutex_unlock(&ctl->cache_writeout_mutex);
1494 
1495 out_nospc:
1496 	cleanup_write_cache_enospc(inode, io_ctl, &cached_state);
1497 
1498 out_unlock:
1499 	if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1500 		up_write(&block_group->data_rwsem);
1501 
1502 out:
1503 	io_ctl->inode = NULL;
1504 	io_ctl_free(io_ctl);
1505 	if (ret) {
1506 		invalidate_inode_pages2(inode->i_mapping);
1507 		BTRFS_I(inode)->generation = 0;
1508 	}
1509 	btrfs_update_inode(trans, root, BTRFS_I(inode));
1510 	if (must_iput)
1511 		iput(inode);
1512 	return ret;
1513 }
1514 
btrfs_write_out_cache(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path)1515 int btrfs_write_out_cache(struct btrfs_trans_handle *trans,
1516 			  struct btrfs_block_group *block_group,
1517 			  struct btrfs_path *path)
1518 {
1519 	struct btrfs_fs_info *fs_info = trans->fs_info;
1520 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1521 	struct inode *inode;
1522 	int ret = 0;
1523 
1524 	spin_lock(&block_group->lock);
1525 	if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1526 		spin_unlock(&block_group->lock);
1527 		return 0;
1528 	}
1529 	spin_unlock(&block_group->lock);
1530 
1531 	inode = lookup_free_space_inode(block_group, path);
1532 	if (IS_ERR(inode))
1533 		return 0;
1534 
1535 	ret = __btrfs_write_out_cache(fs_info->tree_root, inode, ctl,
1536 				block_group, &block_group->io_ctl, trans);
1537 	if (ret) {
1538 		btrfs_debug(fs_info,
1539 	  "failed to write free space cache for block group %llu error %d",
1540 			  block_group->start, ret);
1541 		spin_lock(&block_group->lock);
1542 		block_group->disk_cache_state = BTRFS_DC_ERROR;
1543 		spin_unlock(&block_group->lock);
1544 
1545 		block_group->io_ctl.inode = NULL;
1546 		iput(inode);
1547 	}
1548 
1549 	/*
1550 	 * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1551 	 * to wait for IO and put the inode
1552 	 */
1553 
1554 	return ret;
1555 }
1556 
offset_to_bit(u64 bitmap_start,u32 unit,u64 offset)1557 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1558 					  u64 offset)
1559 {
1560 	ASSERT(offset >= bitmap_start);
1561 	offset -= bitmap_start;
1562 	return (unsigned long)(div_u64(offset, unit));
1563 }
1564 
bytes_to_bits(u64 bytes,u32 unit)1565 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1566 {
1567 	return (unsigned long)(div_u64(bytes, unit));
1568 }
1569 
offset_to_bitmap(struct btrfs_free_space_ctl * ctl,u64 offset)1570 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1571 				   u64 offset)
1572 {
1573 	u64 bitmap_start;
1574 	u64 bytes_per_bitmap;
1575 
1576 	bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1577 	bitmap_start = offset - ctl->start;
1578 	bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1579 	bitmap_start *= bytes_per_bitmap;
1580 	bitmap_start += ctl->start;
1581 
1582 	return bitmap_start;
1583 }
1584 
tree_insert_offset(struct btrfs_free_space_ctl * ctl,struct btrfs_free_cluster * cluster,struct btrfs_free_space * new_entry)1585 static int tree_insert_offset(struct btrfs_free_space_ctl *ctl,
1586 			      struct btrfs_free_cluster *cluster,
1587 			      struct btrfs_free_space *new_entry)
1588 {
1589 	struct rb_root *root;
1590 	struct rb_node **p;
1591 	struct rb_node *parent = NULL;
1592 
1593 	lockdep_assert_held(&ctl->tree_lock);
1594 
1595 	if (cluster) {
1596 		lockdep_assert_held(&cluster->lock);
1597 		root = &cluster->root;
1598 	} else {
1599 		root = &ctl->free_space_offset;
1600 	}
1601 
1602 	p = &root->rb_node;
1603 
1604 	while (*p) {
1605 		struct btrfs_free_space *info;
1606 
1607 		parent = *p;
1608 		info = rb_entry(parent, struct btrfs_free_space, offset_index);
1609 
1610 		if (new_entry->offset < info->offset) {
1611 			p = &(*p)->rb_left;
1612 		} else if (new_entry->offset > info->offset) {
1613 			p = &(*p)->rb_right;
1614 		} else {
1615 			/*
1616 			 * we could have a bitmap entry and an extent entry
1617 			 * share the same offset.  If this is the case, we want
1618 			 * the extent entry to always be found first if we do a
1619 			 * linear search through the tree, since we want to have
1620 			 * the quickest allocation time, and allocating from an
1621 			 * extent is faster than allocating from a bitmap.  So
1622 			 * if we're inserting a bitmap and we find an entry at
1623 			 * this offset, we want to go right, or after this entry
1624 			 * logically.  If we are inserting an extent and we've
1625 			 * found a bitmap, we want to go left, or before
1626 			 * logically.
1627 			 */
1628 			if (new_entry->bitmap) {
1629 				if (info->bitmap) {
1630 					WARN_ON_ONCE(1);
1631 					return -EEXIST;
1632 				}
1633 				p = &(*p)->rb_right;
1634 			} else {
1635 				if (!info->bitmap) {
1636 					WARN_ON_ONCE(1);
1637 					return -EEXIST;
1638 				}
1639 				p = &(*p)->rb_left;
1640 			}
1641 		}
1642 	}
1643 
1644 	rb_link_node(&new_entry->offset_index, parent, p);
1645 	rb_insert_color(&new_entry->offset_index, root);
1646 
1647 	return 0;
1648 }
1649 
1650 /*
1651  * This is a little subtle.  We *only* have ->max_extent_size set if we actually
1652  * searched through the bitmap and figured out the largest ->max_extent_size,
1653  * otherwise it's 0.  In the case that it's 0 we don't want to tell the
1654  * allocator the wrong thing, we want to use the actual real max_extent_size
1655  * we've found already if it's larger, or we want to use ->bytes.
1656  *
1657  * This matters because find_free_space() will skip entries who's ->bytes is
1658  * less than the required bytes.  So if we didn't search down this bitmap, we
1659  * may pick some previous entry that has a smaller ->max_extent_size than we
1660  * have.  For example, assume we have two entries, one that has
1661  * ->max_extent_size set to 4K and ->bytes set to 1M.  A second entry hasn't set
1662  * ->max_extent_size yet, has ->bytes set to 8K and it's contiguous.  We will
1663  *  call into find_free_space(), and return with max_extent_size == 4K, because
1664  *  that first bitmap entry had ->max_extent_size set, but the second one did
1665  *  not.  If instead we returned 8K we'd come in searching for 8K, and find the
1666  *  8K contiguous range.
1667  *
1668  *  Consider the other case, we have 2 8K chunks in that second entry and still
1669  *  don't have ->max_extent_size set.  We'll return 16K, and the next time the
1670  *  allocator comes in it'll fully search our second bitmap, and this time it'll
1671  *  get an uptodate value of 8K as the maximum chunk size.  Then we'll get the
1672  *  right allocation the next loop through.
1673  */
get_max_extent_size(const struct btrfs_free_space * entry)1674 static inline u64 get_max_extent_size(const struct btrfs_free_space *entry)
1675 {
1676 	if (entry->bitmap && entry->max_extent_size)
1677 		return entry->max_extent_size;
1678 	return entry->bytes;
1679 }
1680 
1681 /*
1682  * We want the largest entry to be leftmost, so this is inverted from what you'd
1683  * normally expect.
1684  */
entry_less(struct rb_node * node,const struct rb_node * parent)1685 static bool entry_less(struct rb_node *node, const struct rb_node *parent)
1686 {
1687 	const struct btrfs_free_space *entry, *exist;
1688 
1689 	entry = rb_entry(node, struct btrfs_free_space, bytes_index);
1690 	exist = rb_entry(parent, struct btrfs_free_space, bytes_index);
1691 	return get_max_extent_size(exist) < get_max_extent_size(entry);
1692 }
1693 
1694 /*
1695  * searches the tree for the given offset.
1696  *
1697  * fuzzy - If this is set, then we are trying to make an allocation, and we just
1698  * want a section that has at least bytes size and comes at or after the given
1699  * offset.
1700  */
1701 static struct btrfs_free_space *
tree_search_offset(struct btrfs_free_space_ctl * ctl,u64 offset,int bitmap_only,int fuzzy)1702 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1703 		   u64 offset, int bitmap_only, int fuzzy)
1704 {
1705 	struct rb_node *n = ctl->free_space_offset.rb_node;
1706 	struct btrfs_free_space *entry = NULL, *prev = NULL;
1707 
1708 	lockdep_assert_held(&ctl->tree_lock);
1709 
1710 	/* find entry that is closest to the 'offset' */
1711 	while (n) {
1712 		entry = rb_entry(n, struct btrfs_free_space, offset_index);
1713 		prev = entry;
1714 
1715 		if (offset < entry->offset)
1716 			n = n->rb_left;
1717 		else if (offset > entry->offset)
1718 			n = n->rb_right;
1719 		else
1720 			break;
1721 
1722 		entry = NULL;
1723 	}
1724 
1725 	if (bitmap_only) {
1726 		if (!entry)
1727 			return NULL;
1728 		if (entry->bitmap)
1729 			return entry;
1730 
1731 		/*
1732 		 * bitmap entry and extent entry may share same offset,
1733 		 * in that case, bitmap entry comes after extent entry.
1734 		 */
1735 		n = rb_next(n);
1736 		if (!n)
1737 			return NULL;
1738 		entry = rb_entry(n, struct btrfs_free_space, offset_index);
1739 		if (entry->offset != offset)
1740 			return NULL;
1741 
1742 		WARN_ON(!entry->bitmap);
1743 		return entry;
1744 	} else if (entry) {
1745 		if (entry->bitmap) {
1746 			/*
1747 			 * if previous extent entry covers the offset,
1748 			 * we should return it instead of the bitmap entry
1749 			 */
1750 			n = rb_prev(&entry->offset_index);
1751 			if (n) {
1752 				prev = rb_entry(n, struct btrfs_free_space,
1753 						offset_index);
1754 				if (!prev->bitmap &&
1755 				    prev->offset + prev->bytes > offset)
1756 					entry = prev;
1757 			}
1758 		}
1759 		return entry;
1760 	}
1761 
1762 	if (!prev)
1763 		return NULL;
1764 
1765 	/* find last entry before the 'offset' */
1766 	entry = prev;
1767 	if (entry->offset > offset) {
1768 		n = rb_prev(&entry->offset_index);
1769 		if (n) {
1770 			entry = rb_entry(n, struct btrfs_free_space,
1771 					offset_index);
1772 			ASSERT(entry->offset <= offset);
1773 		} else {
1774 			if (fuzzy)
1775 				return entry;
1776 			else
1777 				return NULL;
1778 		}
1779 	}
1780 
1781 	if (entry->bitmap) {
1782 		n = rb_prev(&entry->offset_index);
1783 		if (n) {
1784 			prev = rb_entry(n, struct btrfs_free_space,
1785 					offset_index);
1786 			if (!prev->bitmap &&
1787 			    prev->offset + prev->bytes > offset)
1788 				return prev;
1789 		}
1790 		if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1791 			return entry;
1792 	} else if (entry->offset + entry->bytes > offset)
1793 		return entry;
1794 
1795 	if (!fuzzy)
1796 		return NULL;
1797 
1798 	while (1) {
1799 		n = rb_next(&entry->offset_index);
1800 		if (!n)
1801 			return NULL;
1802 		entry = rb_entry(n, struct btrfs_free_space, offset_index);
1803 		if (entry->bitmap) {
1804 			if (entry->offset + BITS_PER_BITMAP *
1805 			    ctl->unit > offset)
1806 				break;
1807 		} else {
1808 			if (entry->offset + entry->bytes > offset)
1809 				break;
1810 		}
1811 	}
1812 	return entry;
1813 }
1814 
unlink_free_space(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)1815 static inline void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1816 				     struct btrfs_free_space *info,
1817 				     bool update_stat)
1818 {
1819 	lockdep_assert_held(&ctl->tree_lock);
1820 
1821 	rb_erase(&info->offset_index, &ctl->free_space_offset);
1822 	rb_erase_cached(&info->bytes_index, &ctl->free_space_bytes);
1823 	ctl->free_extents--;
1824 
1825 	if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1826 		ctl->discardable_extents[BTRFS_STAT_CURR]--;
1827 		ctl->discardable_bytes[BTRFS_STAT_CURR] -= info->bytes;
1828 	}
1829 
1830 	if (update_stat)
1831 		ctl->free_space -= info->bytes;
1832 }
1833 
link_free_space(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)1834 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1835 			   struct btrfs_free_space *info)
1836 {
1837 	int ret = 0;
1838 
1839 	lockdep_assert_held(&ctl->tree_lock);
1840 
1841 	ASSERT(info->bytes || info->bitmap);
1842 	ret = tree_insert_offset(ctl, NULL, info);
1843 	if (ret)
1844 		return ret;
1845 
1846 	rb_add_cached(&info->bytes_index, &ctl->free_space_bytes, entry_less);
1847 
1848 	if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1849 		ctl->discardable_extents[BTRFS_STAT_CURR]++;
1850 		ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
1851 	}
1852 
1853 	ctl->free_space += info->bytes;
1854 	ctl->free_extents++;
1855 	return ret;
1856 }
1857 
relink_bitmap_entry(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)1858 static void relink_bitmap_entry(struct btrfs_free_space_ctl *ctl,
1859 				struct btrfs_free_space *info)
1860 {
1861 	ASSERT(info->bitmap);
1862 
1863 	/*
1864 	 * If our entry is empty it's because we're on a cluster and we don't
1865 	 * want to re-link it into our ctl bytes index.
1866 	 */
1867 	if (RB_EMPTY_NODE(&info->bytes_index))
1868 		return;
1869 
1870 	lockdep_assert_held(&ctl->tree_lock);
1871 
1872 	rb_erase_cached(&info->bytes_index, &ctl->free_space_bytes);
1873 	rb_add_cached(&info->bytes_index, &ctl->free_space_bytes, entry_less);
1874 }
1875 
bitmap_clear_bits(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset,u64 bytes,bool update_stat)1876 static inline void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1877 				     struct btrfs_free_space *info,
1878 				     u64 offset, u64 bytes, bool update_stat)
1879 {
1880 	unsigned long start, count, end;
1881 	int extent_delta = -1;
1882 
1883 	start = offset_to_bit(info->offset, ctl->unit, offset);
1884 	count = bytes_to_bits(bytes, ctl->unit);
1885 	end = start + count;
1886 	ASSERT(end <= BITS_PER_BITMAP);
1887 
1888 	bitmap_clear(info->bitmap, start, count);
1889 
1890 	info->bytes -= bytes;
1891 	if (info->max_extent_size > ctl->unit)
1892 		info->max_extent_size = 0;
1893 
1894 	relink_bitmap_entry(ctl, info);
1895 
1896 	if (start && test_bit(start - 1, info->bitmap))
1897 		extent_delta++;
1898 
1899 	if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1900 		extent_delta++;
1901 
1902 	info->bitmap_extents += extent_delta;
1903 	if (!btrfs_free_space_trimmed(info)) {
1904 		ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1905 		ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
1906 	}
1907 
1908 	if (update_stat)
1909 		ctl->free_space -= bytes;
1910 }
1911 
bitmap_set_bits(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset,u64 bytes)1912 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1913 			    struct btrfs_free_space *info, u64 offset,
1914 			    u64 bytes)
1915 {
1916 	unsigned long start, count, end;
1917 	int extent_delta = 1;
1918 
1919 	start = offset_to_bit(info->offset, ctl->unit, offset);
1920 	count = bytes_to_bits(bytes, ctl->unit);
1921 	end = start + count;
1922 	ASSERT(end <= BITS_PER_BITMAP);
1923 
1924 	bitmap_set(info->bitmap, start, count);
1925 
1926 	/*
1927 	 * We set some bytes, we have no idea what the max extent size is
1928 	 * anymore.
1929 	 */
1930 	info->max_extent_size = 0;
1931 	info->bytes += bytes;
1932 	ctl->free_space += bytes;
1933 
1934 	relink_bitmap_entry(ctl, info);
1935 
1936 	if (start && test_bit(start - 1, info->bitmap))
1937 		extent_delta--;
1938 
1939 	if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1940 		extent_delta--;
1941 
1942 	info->bitmap_extents += extent_delta;
1943 	if (!btrfs_free_space_trimmed(info)) {
1944 		ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1945 		ctl->discardable_bytes[BTRFS_STAT_CURR] += bytes;
1946 	}
1947 }
1948 
1949 /*
1950  * If we can not find suitable extent, we will use bytes to record
1951  * the size of the max extent.
1952  */
search_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * bitmap_info,u64 * offset,u64 * bytes,bool for_alloc)1953 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1954 			 struct btrfs_free_space *bitmap_info, u64 *offset,
1955 			 u64 *bytes, bool for_alloc)
1956 {
1957 	unsigned long found_bits = 0;
1958 	unsigned long max_bits = 0;
1959 	unsigned long bits, i;
1960 	unsigned long next_zero;
1961 	unsigned long extent_bits;
1962 
1963 	/*
1964 	 * Skip searching the bitmap if we don't have a contiguous section that
1965 	 * is large enough for this allocation.
1966 	 */
1967 	if (for_alloc &&
1968 	    bitmap_info->max_extent_size &&
1969 	    bitmap_info->max_extent_size < *bytes) {
1970 		*bytes = bitmap_info->max_extent_size;
1971 		return -1;
1972 	}
1973 
1974 	i = offset_to_bit(bitmap_info->offset, ctl->unit,
1975 			  max_t(u64, *offset, bitmap_info->offset));
1976 	bits = bytes_to_bits(*bytes, ctl->unit);
1977 
1978 	for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1979 		if (for_alloc && bits == 1) {
1980 			found_bits = 1;
1981 			break;
1982 		}
1983 		next_zero = find_next_zero_bit(bitmap_info->bitmap,
1984 					       BITS_PER_BITMAP, i);
1985 		extent_bits = next_zero - i;
1986 		if (extent_bits >= bits) {
1987 			found_bits = extent_bits;
1988 			break;
1989 		} else if (extent_bits > max_bits) {
1990 			max_bits = extent_bits;
1991 		}
1992 		i = next_zero;
1993 	}
1994 
1995 	if (found_bits) {
1996 		*offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1997 		*bytes = (u64)(found_bits) * ctl->unit;
1998 		return 0;
1999 	}
2000 
2001 	*bytes = (u64)(max_bits) * ctl->unit;
2002 	bitmap_info->max_extent_size = *bytes;
2003 	relink_bitmap_entry(ctl, bitmap_info);
2004 	return -1;
2005 }
2006 
2007 /* Cache the size of the max extent in bytes */
2008 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)2009 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
2010 		unsigned long align, u64 *max_extent_size, bool use_bytes_index)
2011 {
2012 	struct btrfs_free_space *entry;
2013 	struct rb_node *node;
2014 	u64 tmp;
2015 	u64 align_off;
2016 	int ret;
2017 
2018 	if (!ctl->free_space_offset.rb_node)
2019 		goto out;
2020 again:
2021 	if (use_bytes_index) {
2022 		node = rb_first_cached(&ctl->free_space_bytes);
2023 	} else {
2024 		entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset),
2025 					   0, 1);
2026 		if (!entry)
2027 			goto out;
2028 		node = &entry->offset_index;
2029 	}
2030 
2031 	for (; node; node = rb_next(node)) {
2032 		if (use_bytes_index)
2033 			entry = rb_entry(node, struct btrfs_free_space,
2034 					 bytes_index);
2035 		else
2036 			entry = rb_entry(node, struct btrfs_free_space,
2037 					 offset_index);
2038 
2039 		/*
2040 		 * If we are using the bytes index then all subsequent entries
2041 		 * in this tree are going to be < bytes, so simply set the max
2042 		 * extent size and exit the loop.
2043 		 *
2044 		 * If we're using the offset index then we need to keep going
2045 		 * through the rest of the tree.
2046 		 */
2047 		if (entry->bytes < *bytes) {
2048 			*max_extent_size = max(get_max_extent_size(entry),
2049 					       *max_extent_size);
2050 			if (use_bytes_index)
2051 				break;
2052 			continue;
2053 		}
2054 
2055 		/* make sure the space returned is big enough
2056 		 * to match our requested alignment
2057 		 */
2058 		if (*bytes >= align) {
2059 			tmp = entry->offset - ctl->start + align - 1;
2060 			tmp = div64_u64(tmp, align);
2061 			tmp = tmp * align + ctl->start;
2062 			align_off = tmp - entry->offset;
2063 		} else {
2064 			align_off = 0;
2065 			tmp = entry->offset;
2066 		}
2067 
2068 		/*
2069 		 * We don't break here if we're using the bytes index because we
2070 		 * may have another entry that has the correct alignment that is
2071 		 * the right size, so we don't want to miss that possibility.
2072 		 * At worst this adds another loop through the logic, but if we
2073 		 * broke here we could prematurely ENOSPC.
2074 		 */
2075 		if (entry->bytes < *bytes + align_off) {
2076 			*max_extent_size = max(get_max_extent_size(entry),
2077 					       *max_extent_size);
2078 			continue;
2079 		}
2080 
2081 		if (entry->bitmap) {
2082 			struct rb_node *old_next = rb_next(node);
2083 			u64 size = *bytes;
2084 
2085 			ret = search_bitmap(ctl, entry, &tmp, &size, true);
2086 			if (!ret) {
2087 				*offset = tmp;
2088 				*bytes = size;
2089 				return entry;
2090 			} else {
2091 				*max_extent_size =
2092 					max(get_max_extent_size(entry),
2093 					    *max_extent_size);
2094 			}
2095 
2096 			/*
2097 			 * The bitmap may have gotten re-arranged in the space
2098 			 * index here because the max_extent_size may have been
2099 			 * updated.  Start from the beginning again if this
2100 			 * happened.
2101 			 */
2102 			if (use_bytes_index && old_next != rb_next(node))
2103 				goto again;
2104 			continue;
2105 		}
2106 
2107 		*offset = tmp;
2108 		*bytes = entry->bytes - align_off;
2109 		return entry;
2110 	}
2111 out:
2112 	return NULL;
2113 }
2114 
add_new_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset)2115 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
2116 			   struct btrfs_free_space *info, u64 offset)
2117 {
2118 	info->offset = offset_to_bitmap(ctl, offset);
2119 	info->bytes = 0;
2120 	info->bitmap_extents = 0;
2121 	INIT_LIST_HEAD(&info->list);
2122 	link_free_space(ctl, info);
2123 	ctl->total_bitmaps++;
2124 	recalculate_thresholds(ctl);
2125 }
2126 
free_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * bitmap_info)2127 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
2128 			struct btrfs_free_space *bitmap_info)
2129 {
2130 	/*
2131 	 * Normally when this is called, the bitmap is completely empty. However,
2132 	 * if we are blowing up the free space cache for one reason or another
2133 	 * via __btrfs_remove_free_space_cache(), then it may not be freed and
2134 	 * we may leave stats on the table.
2135 	 */
2136 	if (bitmap_info->bytes && !btrfs_free_space_trimmed(bitmap_info)) {
2137 		ctl->discardable_extents[BTRFS_STAT_CURR] -=
2138 			bitmap_info->bitmap_extents;
2139 		ctl->discardable_bytes[BTRFS_STAT_CURR] -= bitmap_info->bytes;
2140 
2141 	}
2142 	unlink_free_space(ctl, bitmap_info, true);
2143 	kmem_cache_free(btrfs_free_space_bitmap_cachep, bitmap_info->bitmap);
2144 	kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
2145 	ctl->total_bitmaps--;
2146 	recalculate_thresholds(ctl);
2147 }
2148 
remove_from_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * bitmap_info,u64 * offset,u64 * bytes)2149 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
2150 			      struct btrfs_free_space *bitmap_info,
2151 			      u64 *offset, u64 *bytes)
2152 {
2153 	u64 end;
2154 	u64 search_start, search_bytes;
2155 	int ret;
2156 
2157 again:
2158 	end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
2159 
2160 	/*
2161 	 * We need to search for bits in this bitmap.  We could only cover some
2162 	 * of the extent in this bitmap thanks to how we add space, so we need
2163 	 * to search for as much as it as we can and clear that amount, and then
2164 	 * go searching for the next bit.
2165 	 */
2166 	search_start = *offset;
2167 	search_bytes = ctl->unit;
2168 	search_bytes = min(search_bytes, end - search_start + 1);
2169 	ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes,
2170 			    false);
2171 	if (ret < 0 || search_start != *offset)
2172 		return -EINVAL;
2173 
2174 	/* We may have found more bits than what we need */
2175 	search_bytes = min(search_bytes, *bytes);
2176 
2177 	/* Cannot clear past the end of the bitmap */
2178 	search_bytes = min(search_bytes, end - search_start + 1);
2179 
2180 	bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes, true);
2181 	*offset += search_bytes;
2182 	*bytes -= search_bytes;
2183 
2184 	if (*bytes) {
2185 		struct rb_node *next = rb_next(&bitmap_info->offset_index);
2186 		if (!bitmap_info->bytes)
2187 			free_bitmap(ctl, bitmap_info);
2188 
2189 		/*
2190 		 * no entry after this bitmap, but we still have bytes to
2191 		 * remove, so something has gone wrong.
2192 		 */
2193 		if (!next)
2194 			return -EINVAL;
2195 
2196 		bitmap_info = rb_entry(next, struct btrfs_free_space,
2197 				       offset_index);
2198 
2199 		/*
2200 		 * if the next entry isn't a bitmap we need to return to let the
2201 		 * extent stuff do its work.
2202 		 */
2203 		if (!bitmap_info->bitmap)
2204 			return -EAGAIN;
2205 
2206 		/*
2207 		 * Ok the next item is a bitmap, but it may not actually hold
2208 		 * the information for the rest of this free space stuff, so
2209 		 * look for it, and if we don't find it return so we can try
2210 		 * everything over again.
2211 		 */
2212 		search_start = *offset;
2213 		search_bytes = ctl->unit;
2214 		ret = search_bitmap(ctl, bitmap_info, &search_start,
2215 				    &search_bytes, false);
2216 		if (ret < 0 || search_start != *offset)
2217 			return -EAGAIN;
2218 
2219 		goto again;
2220 	} else if (!bitmap_info->bytes)
2221 		free_bitmap(ctl, bitmap_info);
2222 
2223 	return 0;
2224 }
2225 
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)2226 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
2227 			       struct btrfs_free_space *info, u64 offset,
2228 			       u64 bytes, enum btrfs_trim_state trim_state)
2229 {
2230 	u64 bytes_to_set = 0;
2231 	u64 end;
2232 
2233 	/*
2234 	 * This is a tradeoff to make bitmap trim state minimal.  We mark the
2235 	 * whole bitmap untrimmed if at any point we add untrimmed regions.
2236 	 */
2237 	if (trim_state == BTRFS_TRIM_STATE_UNTRIMMED) {
2238 		if (btrfs_free_space_trimmed(info)) {
2239 			ctl->discardable_extents[BTRFS_STAT_CURR] +=
2240 				info->bitmap_extents;
2241 			ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
2242 		}
2243 		info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2244 	}
2245 
2246 	end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
2247 
2248 	bytes_to_set = min(end - offset, bytes);
2249 
2250 	bitmap_set_bits(ctl, info, offset, bytes_to_set);
2251 
2252 	return bytes_to_set;
2253 
2254 }
2255 
use_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)2256 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
2257 		      struct btrfs_free_space *info)
2258 {
2259 	struct btrfs_block_group *block_group = ctl->block_group;
2260 	struct btrfs_fs_info *fs_info = block_group->fs_info;
2261 	bool forced = false;
2262 
2263 #ifdef CONFIG_BTRFS_DEBUG
2264 	if (btrfs_should_fragment_free_space(block_group))
2265 		forced = true;
2266 #endif
2267 
2268 	/* This is a way to reclaim large regions from the bitmaps. */
2269 	if (!forced && info->bytes >= FORCE_EXTENT_THRESHOLD)
2270 		return false;
2271 
2272 	/*
2273 	 * If we are below the extents threshold then we can add this as an
2274 	 * extent, and don't have to deal with the bitmap
2275 	 */
2276 	if (!forced && ctl->free_extents < ctl->extents_thresh) {
2277 		/*
2278 		 * If this block group has some small extents we don't want to
2279 		 * use up all of our free slots in the cache with them, we want
2280 		 * to reserve them to larger extents, however if we have plenty
2281 		 * of cache left then go ahead an dadd them, no sense in adding
2282 		 * the overhead of a bitmap if we don't have to.
2283 		 */
2284 		if (info->bytes <= fs_info->sectorsize * 8) {
2285 			if (ctl->free_extents * 3 <= ctl->extents_thresh)
2286 				return false;
2287 		} else {
2288 			return false;
2289 		}
2290 	}
2291 
2292 	/*
2293 	 * The original block groups from mkfs can be really small, like 8
2294 	 * megabytes, so don't bother with a bitmap for those entries.  However
2295 	 * some block groups can be smaller than what a bitmap would cover but
2296 	 * are still large enough that they could overflow the 32k memory limit,
2297 	 * so allow those block groups to still be allowed to have a bitmap
2298 	 * entry.
2299 	 */
2300 	if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->length)
2301 		return false;
2302 
2303 	return true;
2304 }
2305 
2306 static const struct btrfs_free_space_op free_space_op = {
2307 	.use_bitmap		= use_bitmap,
2308 };
2309 
insert_into_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)2310 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
2311 			      struct btrfs_free_space *info)
2312 {
2313 	struct btrfs_free_space *bitmap_info;
2314 	struct btrfs_block_group *block_group = NULL;
2315 	int added = 0;
2316 	u64 bytes, offset, bytes_added;
2317 	enum btrfs_trim_state trim_state;
2318 	int ret;
2319 
2320 	bytes = info->bytes;
2321 	offset = info->offset;
2322 	trim_state = info->trim_state;
2323 
2324 	if (!ctl->op->use_bitmap(ctl, info))
2325 		return 0;
2326 
2327 	if (ctl->op == &free_space_op)
2328 		block_group = ctl->block_group;
2329 again:
2330 	/*
2331 	 * Since we link bitmaps right into the cluster we need to see if we
2332 	 * have a cluster here, and if so and it has our bitmap we need to add
2333 	 * the free space to that bitmap.
2334 	 */
2335 	if (block_group && !list_empty(&block_group->cluster_list)) {
2336 		struct btrfs_free_cluster *cluster;
2337 		struct rb_node *node;
2338 		struct btrfs_free_space *entry;
2339 
2340 		cluster = list_entry(block_group->cluster_list.next,
2341 				     struct btrfs_free_cluster,
2342 				     block_group_list);
2343 		spin_lock(&cluster->lock);
2344 		node = rb_first(&cluster->root);
2345 		if (!node) {
2346 			spin_unlock(&cluster->lock);
2347 			goto no_cluster_bitmap;
2348 		}
2349 
2350 		entry = rb_entry(node, struct btrfs_free_space, offset_index);
2351 		if (!entry->bitmap) {
2352 			spin_unlock(&cluster->lock);
2353 			goto no_cluster_bitmap;
2354 		}
2355 
2356 		if (entry->offset == offset_to_bitmap(ctl, offset)) {
2357 			bytes_added = add_bytes_to_bitmap(ctl, entry, offset,
2358 							  bytes, trim_state);
2359 			bytes -= bytes_added;
2360 			offset += bytes_added;
2361 		}
2362 		spin_unlock(&cluster->lock);
2363 		if (!bytes) {
2364 			ret = 1;
2365 			goto out;
2366 		}
2367 	}
2368 
2369 no_cluster_bitmap:
2370 	bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2371 					 1, 0);
2372 	if (!bitmap_info) {
2373 		ASSERT(added == 0);
2374 		goto new_bitmap;
2375 	}
2376 
2377 	bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
2378 					  trim_state);
2379 	bytes -= bytes_added;
2380 	offset += bytes_added;
2381 	added = 0;
2382 
2383 	if (!bytes) {
2384 		ret = 1;
2385 		goto out;
2386 	} else
2387 		goto again;
2388 
2389 new_bitmap:
2390 	if (info && info->bitmap) {
2391 		add_new_bitmap(ctl, info, offset);
2392 		added = 1;
2393 		info = NULL;
2394 		goto again;
2395 	} else {
2396 		spin_unlock(&ctl->tree_lock);
2397 
2398 		/* no pre-allocated info, allocate a new one */
2399 		if (!info) {
2400 			info = kmem_cache_zalloc(btrfs_free_space_cachep,
2401 						 GFP_NOFS);
2402 			if (!info) {
2403 				spin_lock(&ctl->tree_lock);
2404 				ret = -ENOMEM;
2405 				goto out;
2406 			}
2407 		}
2408 
2409 		/* allocate the bitmap */
2410 		info->bitmap = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep,
2411 						 GFP_NOFS);
2412 		info->trim_state = BTRFS_TRIM_STATE_TRIMMED;
2413 		spin_lock(&ctl->tree_lock);
2414 		if (!info->bitmap) {
2415 			ret = -ENOMEM;
2416 			goto out;
2417 		}
2418 		goto again;
2419 	}
2420 
2421 out:
2422 	if (info) {
2423 		if (info->bitmap)
2424 			kmem_cache_free(btrfs_free_space_bitmap_cachep,
2425 					info->bitmap);
2426 		kmem_cache_free(btrfs_free_space_cachep, info);
2427 	}
2428 
2429 	return ret;
2430 }
2431 
2432 /*
2433  * Free space merging rules:
2434  *  1) Merge trimmed areas together
2435  *  2) Let untrimmed areas coalesce with trimmed areas
2436  *  3) Always pull neighboring regions from bitmaps
2437  *
2438  * The above rules are for when we merge free space based on btrfs_trim_state.
2439  * Rules 2 and 3 are subtle because they are suboptimal, but are done for the
2440  * same reason: to promote larger extent regions which makes life easier for
2441  * find_free_extent().  Rule 2 enables coalescing based on the common path
2442  * being returning free space from btrfs_finish_extent_commit().  So when free
2443  * space is trimmed, it will prevent aggregating trimmed new region and
2444  * untrimmed regions in the rb_tree.  Rule 3 is purely to obtain larger extents
2445  * and provide find_free_extent() with the largest extents possible hoping for
2446  * the reuse path.
2447  */
try_merge_free_space(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2448 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2449 			  struct btrfs_free_space *info, bool update_stat)
2450 {
2451 	struct btrfs_free_space *left_info = NULL;
2452 	struct btrfs_free_space *right_info;
2453 	bool merged = false;
2454 	u64 offset = info->offset;
2455 	u64 bytes = info->bytes;
2456 	const bool is_trimmed = btrfs_free_space_trimmed(info);
2457 	struct rb_node *right_prev = NULL;
2458 
2459 	/*
2460 	 * first we want to see if there is free space adjacent to the range we
2461 	 * are adding, if there is remove that struct and add a new one to
2462 	 * cover the entire range
2463 	 */
2464 	right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
2465 	if (right_info)
2466 		right_prev = rb_prev(&right_info->offset_index);
2467 
2468 	if (right_prev)
2469 		left_info = rb_entry(right_prev, struct btrfs_free_space, offset_index);
2470 	else if (!right_info)
2471 		left_info = tree_search_offset(ctl, offset - 1, 0, 0);
2472 
2473 	/* See try_merge_free_space() comment. */
2474 	if (right_info && !right_info->bitmap &&
2475 	    (!is_trimmed || btrfs_free_space_trimmed(right_info))) {
2476 		unlink_free_space(ctl, right_info, update_stat);
2477 		info->bytes += right_info->bytes;
2478 		kmem_cache_free(btrfs_free_space_cachep, right_info);
2479 		merged = true;
2480 	}
2481 
2482 	/* See try_merge_free_space() comment. */
2483 	if (left_info && !left_info->bitmap &&
2484 	    left_info->offset + left_info->bytes == offset &&
2485 	    (!is_trimmed || btrfs_free_space_trimmed(left_info))) {
2486 		unlink_free_space(ctl, left_info, update_stat);
2487 		info->offset = left_info->offset;
2488 		info->bytes += left_info->bytes;
2489 		kmem_cache_free(btrfs_free_space_cachep, left_info);
2490 		merged = true;
2491 	}
2492 
2493 	return merged;
2494 }
2495 
steal_from_bitmap_to_end(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2496 static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2497 				     struct btrfs_free_space *info,
2498 				     bool update_stat)
2499 {
2500 	struct btrfs_free_space *bitmap;
2501 	unsigned long i;
2502 	unsigned long j;
2503 	const u64 end = info->offset + info->bytes;
2504 	const u64 bitmap_offset = offset_to_bitmap(ctl, end);
2505 	u64 bytes;
2506 
2507 	bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2508 	if (!bitmap)
2509 		return false;
2510 
2511 	i = offset_to_bit(bitmap->offset, ctl->unit, end);
2512 	j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
2513 	if (j == i)
2514 		return false;
2515 	bytes = (j - i) * ctl->unit;
2516 	info->bytes += bytes;
2517 
2518 	/* See try_merge_free_space() comment. */
2519 	if (!btrfs_free_space_trimmed(bitmap))
2520 		info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2521 
2522 	bitmap_clear_bits(ctl, bitmap, end, bytes, update_stat);
2523 
2524 	if (!bitmap->bytes)
2525 		free_bitmap(ctl, bitmap);
2526 
2527 	return true;
2528 }
2529 
steal_from_bitmap_to_front(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2530 static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2531 				       struct btrfs_free_space *info,
2532 				       bool update_stat)
2533 {
2534 	struct btrfs_free_space *bitmap;
2535 	u64 bitmap_offset;
2536 	unsigned long i;
2537 	unsigned long j;
2538 	unsigned long prev_j;
2539 	u64 bytes;
2540 
2541 	bitmap_offset = offset_to_bitmap(ctl, info->offset);
2542 	/* If we're on a boundary, try the previous logical bitmap. */
2543 	if (bitmap_offset == info->offset) {
2544 		if (info->offset == 0)
2545 			return false;
2546 		bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
2547 	}
2548 
2549 	bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2550 	if (!bitmap)
2551 		return false;
2552 
2553 	i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
2554 	j = 0;
2555 	prev_j = (unsigned long)-1;
2556 	for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2557 		if (j > i)
2558 			break;
2559 		prev_j = j;
2560 	}
2561 	if (prev_j == i)
2562 		return false;
2563 
2564 	if (prev_j == (unsigned long)-1)
2565 		bytes = (i + 1) * ctl->unit;
2566 	else
2567 		bytes = (i - prev_j) * ctl->unit;
2568 
2569 	info->offset -= bytes;
2570 	info->bytes += bytes;
2571 
2572 	/* See try_merge_free_space() comment. */
2573 	if (!btrfs_free_space_trimmed(bitmap))
2574 		info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2575 
2576 	bitmap_clear_bits(ctl, bitmap, info->offset, bytes, update_stat);
2577 
2578 	if (!bitmap->bytes)
2579 		free_bitmap(ctl, bitmap);
2580 
2581 	return true;
2582 }
2583 
2584 /*
2585  * We prefer always to allocate from extent entries, both for clustered and
2586  * non-clustered allocation requests. So when attempting to add a new extent
2587  * entry, try to see if there's adjacent free space in bitmap entries, and if
2588  * there is, migrate that space from the bitmaps to the extent.
2589  * Like this we get better chances of satisfying space allocation requests
2590  * because we attempt to satisfy them based on a single cache entry, and never
2591  * on 2 or more entries - even if the entries represent a contiguous free space
2592  * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2593  * ends).
2594  */
steal_from_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2595 static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2596 			      struct btrfs_free_space *info,
2597 			      bool update_stat)
2598 {
2599 	/*
2600 	 * Only work with disconnected entries, as we can change their offset,
2601 	 * and must be extent entries.
2602 	 */
2603 	ASSERT(!info->bitmap);
2604 	ASSERT(RB_EMPTY_NODE(&info->offset_index));
2605 
2606 	if (ctl->total_bitmaps > 0) {
2607 		bool stole_end;
2608 		bool stole_front = false;
2609 
2610 		stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2611 		if (ctl->total_bitmaps > 0)
2612 			stole_front = steal_from_bitmap_to_front(ctl, info,
2613 								 update_stat);
2614 
2615 		if (stole_end || stole_front)
2616 			try_merge_free_space(ctl, info, update_stat);
2617 	}
2618 }
2619 
__btrfs_add_free_space(struct btrfs_block_group * block_group,u64 offset,u64 bytes,enum btrfs_trim_state trim_state)2620 int __btrfs_add_free_space(struct btrfs_block_group *block_group,
2621 			   u64 offset, u64 bytes,
2622 			   enum btrfs_trim_state trim_state)
2623 {
2624 	struct btrfs_fs_info *fs_info = block_group->fs_info;
2625 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2626 	struct btrfs_free_space *info;
2627 	int ret = 0;
2628 	u64 filter_bytes = bytes;
2629 
2630 	ASSERT(!btrfs_is_zoned(fs_info));
2631 
2632 	info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2633 	if (!info)
2634 		return -ENOMEM;
2635 
2636 	info->offset = offset;
2637 	info->bytes = bytes;
2638 	info->trim_state = trim_state;
2639 	RB_CLEAR_NODE(&info->offset_index);
2640 	RB_CLEAR_NODE(&info->bytes_index);
2641 
2642 	spin_lock(&ctl->tree_lock);
2643 
2644 	if (try_merge_free_space(ctl, info, true))
2645 		goto link;
2646 
2647 	/*
2648 	 * There was no extent directly to the left or right of this new
2649 	 * extent then we know we're going to have to allocate a new extent, so
2650 	 * before we do that see if we need to drop this into a bitmap
2651 	 */
2652 	ret = insert_into_bitmap(ctl, info);
2653 	if (ret < 0) {
2654 		goto out;
2655 	} else if (ret) {
2656 		ret = 0;
2657 		goto out;
2658 	}
2659 link:
2660 	/*
2661 	 * Only steal free space from adjacent bitmaps if we're sure we're not
2662 	 * going to add the new free space to existing bitmap entries - because
2663 	 * that would mean unnecessary work that would be reverted. Therefore
2664 	 * attempt to steal space from bitmaps if we're adding an extent entry.
2665 	 */
2666 	steal_from_bitmap(ctl, info, true);
2667 
2668 	filter_bytes = max(filter_bytes, info->bytes);
2669 
2670 	ret = link_free_space(ctl, info);
2671 	if (ret)
2672 		kmem_cache_free(btrfs_free_space_cachep, info);
2673 out:
2674 	btrfs_discard_update_discardable(block_group);
2675 	spin_unlock(&ctl->tree_lock);
2676 
2677 	if (ret) {
2678 		btrfs_crit(fs_info, "unable to add free space :%d", ret);
2679 		ASSERT(ret != -EEXIST);
2680 	}
2681 
2682 	if (trim_state != BTRFS_TRIM_STATE_TRIMMED) {
2683 		btrfs_discard_check_filter(block_group, filter_bytes);
2684 		btrfs_discard_queue_work(&fs_info->discard_ctl, block_group);
2685 	}
2686 
2687 	return ret;
2688 }
2689 
__btrfs_add_free_space_zoned(struct btrfs_block_group * block_group,u64 bytenr,u64 size,bool used)2690 static int __btrfs_add_free_space_zoned(struct btrfs_block_group *block_group,
2691 					u64 bytenr, u64 size, bool used)
2692 {
2693 	struct btrfs_space_info *sinfo = block_group->space_info;
2694 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2695 	u64 offset = bytenr - block_group->start;
2696 	u64 to_free, to_unusable;
2697 	int bg_reclaim_threshold = 0;
2698 	bool initial = (size == block_group->length);
2699 	u64 reclaimable_unusable;
2700 
2701 	WARN_ON(!initial && offset + size > block_group->zone_capacity);
2702 
2703 	if (!initial)
2704 		bg_reclaim_threshold = READ_ONCE(sinfo->bg_reclaim_threshold);
2705 
2706 	spin_lock(&ctl->tree_lock);
2707 	if (!used)
2708 		to_free = size;
2709 	else if (initial)
2710 		to_free = block_group->zone_capacity;
2711 	else if (offset >= block_group->alloc_offset)
2712 		to_free = size;
2713 	else if (offset + size <= block_group->alloc_offset)
2714 		to_free = 0;
2715 	else
2716 		to_free = offset + size - block_group->alloc_offset;
2717 	to_unusable = size - to_free;
2718 
2719 	ctl->free_space += to_free;
2720 	/*
2721 	 * If the block group is read-only, we should account freed space into
2722 	 * bytes_readonly.
2723 	 */
2724 	if (!block_group->ro)
2725 		block_group->zone_unusable += to_unusable;
2726 	spin_unlock(&ctl->tree_lock);
2727 	if (!used) {
2728 		spin_lock(&block_group->lock);
2729 		block_group->alloc_offset -= size;
2730 		spin_unlock(&block_group->lock);
2731 	}
2732 
2733 	reclaimable_unusable = block_group->zone_unusable -
2734 			       (block_group->length - block_group->zone_capacity);
2735 	/* All the region is now unusable. Mark it as unused and reclaim */
2736 	if (block_group->zone_unusable == block_group->length) {
2737 		btrfs_mark_bg_unused(block_group);
2738 	} else if (bg_reclaim_threshold &&
2739 		   reclaimable_unusable >=
2740 		   mult_perc(block_group->zone_capacity, bg_reclaim_threshold)) {
2741 		btrfs_mark_bg_to_reclaim(block_group);
2742 	}
2743 
2744 	return 0;
2745 }
2746 
btrfs_add_free_space(struct btrfs_block_group * block_group,u64 bytenr,u64 size)2747 int btrfs_add_free_space(struct btrfs_block_group *block_group,
2748 			 u64 bytenr, u64 size)
2749 {
2750 	enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2751 
2752 	if (btrfs_is_zoned(block_group->fs_info))
2753 		return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2754 						    true);
2755 
2756 	if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC))
2757 		trim_state = BTRFS_TRIM_STATE_TRIMMED;
2758 
2759 	return __btrfs_add_free_space(block_group, bytenr, size, trim_state);
2760 }
2761 
btrfs_add_free_space_unused(struct btrfs_block_group * block_group,u64 bytenr,u64 size)2762 int btrfs_add_free_space_unused(struct btrfs_block_group *block_group,
2763 				u64 bytenr, u64 size)
2764 {
2765 	if (btrfs_is_zoned(block_group->fs_info))
2766 		return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2767 						    false);
2768 
2769 	return btrfs_add_free_space(block_group, bytenr, size);
2770 }
2771 
2772 /*
2773  * This is a subtle distinction because when adding free space back in general,
2774  * we want it to be added as untrimmed for async. But in the case where we add
2775  * it on loading of a block group, we want to consider it trimmed.
2776  */
btrfs_add_free_space_async_trimmed(struct btrfs_block_group * block_group,u64 bytenr,u64 size)2777 int btrfs_add_free_space_async_trimmed(struct btrfs_block_group *block_group,
2778 				       u64 bytenr, u64 size)
2779 {
2780 	enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2781 
2782 	if (btrfs_is_zoned(block_group->fs_info))
2783 		return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2784 						    true);
2785 
2786 	if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC) ||
2787 	    btrfs_test_opt(block_group->fs_info, DISCARD_ASYNC))
2788 		trim_state = BTRFS_TRIM_STATE_TRIMMED;
2789 
2790 	return __btrfs_add_free_space(block_group, bytenr, size, trim_state);
2791 }
2792 
btrfs_remove_free_space(struct btrfs_block_group * block_group,u64 offset,u64 bytes)2793 int btrfs_remove_free_space(struct btrfs_block_group *block_group,
2794 			    u64 offset, u64 bytes)
2795 {
2796 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2797 	struct btrfs_free_space *info;
2798 	int ret;
2799 	bool re_search = false;
2800 
2801 	if (btrfs_is_zoned(block_group->fs_info)) {
2802 		/*
2803 		 * This can happen with conventional zones when replaying log.
2804 		 * Since the allocation info of tree-log nodes are not recorded
2805 		 * to the extent-tree, calculate_alloc_pointer() failed to
2806 		 * advance the allocation pointer after last allocated tree log
2807 		 * node blocks.
2808 		 *
2809 		 * This function is called from
2810 		 * btrfs_pin_extent_for_log_replay() when replaying the log.
2811 		 * Advance the pointer not to overwrite the tree-log nodes.
2812 		 */
2813 		if (block_group->start + block_group->alloc_offset <
2814 		    offset + bytes) {
2815 			block_group->alloc_offset =
2816 				offset + bytes - block_group->start;
2817 		}
2818 		return 0;
2819 	}
2820 
2821 	spin_lock(&ctl->tree_lock);
2822 
2823 again:
2824 	ret = 0;
2825 	if (!bytes)
2826 		goto out_lock;
2827 
2828 	info = tree_search_offset(ctl, offset, 0, 0);
2829 	if (!info) {
2830 		/*
2831 		 * oops didn't find an extent that matched the space we wanted
2832 		 * to remove, look for a bitmap instead
2833 		 */
2834 		info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2835 					  1, 0);
2836 		if (!info) {
2837 			/*
2838 			 * If we found a partial bit of our free space in a
2839 			 * bitmap but then couldn't find the other part this may
2840 			 * be a problem, so WARN about it.
2841 			 */
2842 			WARN_ON(re_search);
2843 			goto out_lock;
2844 		}
2845 	}
2846 
2847 	re_search = false;
2848 	if (!info->bitmap) {
2849 		unlink_free_space(ctl, info, true);
2850 		if (offset == info->offset) {
2851 			u64 to_free = min(bytes, info->bytes);
2852 
2853 			info->bytes -= to_free;
2854 			info->offset += to_free;
2855 			if (info->bytes) {
2856 				ret = link_free_space(ctl, info);
2857 				WARN_ON(ret);
2858 			} else {
2859 				kmem_cache_free(btrfs_free_space_cachep, info);
2860 			}
2861 
2862 			offset += to_free;
2863 			bytes -= to_free;
2864 			goto again;
2865 		} else {
2866 			u64 old_end = info->bytes + info->offset;
2867 
2868 			info->bytes = offset - info->offset;
2869 			ret = link_free_space(ctl, info);
2870 			WARN_ON(ret);
2871 			if (ret)
2872 				goto out_lock;
2873 
2874 			/* Not enough bytes in this entry to satisfy us */
2875 			if (old_end < offset + bytes) {
2876 				bytes -= old_end - offset;
2877 				offset = old_end;
2878 				goto again;
2879 			} else if (old_end == offset + bytes) {
2880 				/* all done */
2881 				goto out_lock;
2882 			}
2883 			spin_unlock(&ctl->tree_lock);
2884 
2885 			ret = __btrfs_add_free_space(block_group,
2886 						     offset + bytes,
2887 						     old_end - (offset + bytes),
2888 						     info->trim_state);
2889 			WARN_ON(ret);
2890 			goto out;
2891 		}
2892 	}
2893 
2894 	ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2895 	if (ret == -EAGAIN) {
2896 		re_search = true;
2897 		goto again;
2898 	}
2899 out_lock:
2900 	btrfs_discard_update_discardable(block_group);
2901 	spin_unlock(&ctl->tree_lock);
2902 out:
2903 	return ret;
2904 }
2905 
btrfs_dump_free_space(struct btrfs_block_group * block_group,u64 bytes)2906 void btrfs_dump_free_space(struct btrfs_block_group *block_group,
2907 			   u64 bytes)
2908 {
2909 	struct btrfs_fs_info *fs_info = block_group->fs_info;
2910 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2911 	struct btrfs_free_space *info;
2912 	struct rb_node *n;
2913 	int count = 0;
2914 
2915 	/*
2916 	 * Zoned btrfs does not use free space tree and cluster. Just print
2917 	 * out the free space after the allocation offset.
2918 	 */
2919 	if (btrfs_is_zoned(fs_info)) {
2920 		btrfs_info(fs_info, "free space %llu active %d",
2921 			   block_group->zone_capacity - block_group->alloc_offset,
2922 			   test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
2923 				    &block_group->runtime_flags));
2924 		return;
2925 	}
2926 
2927 	spin_lock(&ctl->tree_lock);
2928 	for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2929 		info = rb_entry(n, struct btrfs_free_space, offset_index);
2930 		if (info->bytes >= bytes && !block_group->ro)
2931 			count++;
2932 		btrfs_crit(fs_info, "entry offset %llu, bytes %llu, bitmap %s",
2933 			   info->offset, info->bytes,
2934 		       (info->bitmap) ? "yes" : "no");
2935 	}
2936 	spin_unlock(&ctl->tree_lock);
2937 	btrfs_info(fs_info, "block group has cluster?: %s",
2938 	       list_empty(&block_group->cluster_list) ? "no" : "yes");
2939 	btrfs_info(fs_info,
2940 		   "%d free space entries at or bigger than %llu bytes",
2941 		   count, bytes);
2942 }
2943 
btrfs_init_free_space_ctl(struct btrfs_block_group * block_group,struct btrfs_free_space_ctl * ctl)2944 void btrfs_init_free_space_ctl(struct btrfs_block_group *block_group,
2945 			       struct btrfs_free_space_ctl *ctl)
2946 {
2947 	struct btrfs_fs_info *fs_info = block_group->fs_info;
2948 
2949 	spin_lock_init(&ctl->tree_lock);
2950 	ctl->unit = fs_info->sectorsize;
2951 	ctl->start = block_group->start;
2952 	ctl->block_group = block_group;
2953 	ctl->op = &free_space_op;
2954 	ctl->free_space_bytes = RB_ROOT_CACHED;
2955 	INIT_LIST_HEAD(&ctl->trimming_ranges);
2956 	mutex_init(&ctl->cache_writeout_mutex);
2957 
2958 	/*
2959 	 * we only want to have 32k of ram per block group for keeping
2960 	 * track of free space, and if we pass 1/2 of that we want to
2961 	 * start converting things over to using bitmaps
2962 	 */
2963 	ctl->extents_thresh = (SZ_32K / 2) / sizeof(struct btrfs_free_space);
2964 }
2965 
2966 /*
2967  * for a given cluster, put all of its extents back into the free
2968  * space cache.  If the block group passed doesn't match the block group
2969  * pointed to by the cluster, someone else raced in and freed the
2970  * cluster already.  In that case, we just return without changing anything
2971  */
__btrfs_return_cluster_to_free_space(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster)2972 static void __btrfs_return_cluster_to_free_space(
2973 			     struct btrfs_block_group *block_group,
2974 			     struct btrfs_free_cluster *cluster)
2975 {
2976 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2977 	struct rb_node *node;
2978 
2979 	lockdep_assert_held(&ctl->tree_lock);
2980 
2981 	spin_lock(&cluster->lock);
2982 	if (cluster->block_group != block_group) {
2983 		spin_unlock(&cluster->lock);
2984 		return;
2985 	}
2986 
2987 	cluster->block_group = NULL;
2988 	cluster->window_start = 0;
2989 	list_del_init(&cluster->block_group_list);
2990 
2991 	node = rb_first(&cluster->root);
2992 	while (node) {
2993 		struct btrfs_free_space *entry;
2994 
2995 		entry = rb_entry(node, struct btrfs_free_space, offset_index);
2996 		node = rb_next(&entry->offset_index);
2997 		rb_erase(&entry->offset_index, &cluster->root);
2998 		RB_CLEAR_NODE(&entry->offset_index);
2999 
3000 		if (!entry->bitmap) {
3001 			/* Merging treats extents as if they were new */
3002 			if (!btrfs_free_space_trimmed(entry)) {
3003 				ctl->discardable_extents[BTRFS_STAT_CURR]--;
3004 				ctl->discardable_bytes[BTRFS_STAT_CURR] -=
3005 					entry->bytes;
3006 			}
3007 
3008 			try_merge_free_space(ctl, entry, false);
3009 			steal_from_bitmap(ctl, entry, false);
3010 
3011 			/* As we insert directly, update these statistics */
3012 			if (!btrfs_free_space_trimmed(entry)) {
3013 				ctl->discardable_extents[BTRFS_STAT_CURR]++;
3014 				ctl->discardable_bytes[BTRFS_STAT_CURR] +=
3015 					entry->bytes;
3016 			}
3017 		}
3018 		tree_insert_offset(ctl, NULL, entry);
3019 		rb_add_cached(&entry->bytes_index, &ctl->free_space_bytes,
3020 			      entry_less);
3021 	}
3022 	cluster->root = RB_ROOT;
3023 	spin_unlock(&cluster->lock);
3024 	btrfs_put_block_group(block_group);
3025 }
3026 
btrfs_remove_free_space_cache(struct btrfs_block_group * block_group)3027 void btrfs_remove_free_space_cache(struct btrfs_block_group *block_group)
3028 {
3029 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3030 	struct btrfs_free_cluster *cluster;
3031 	struct list_head *head;
3032 
3033 	spin_lock(&ctl->tree_lock);
3034 	while ((head = block_group->cluster_list.next) !=
3035 	       &block_group->cluster_list) {
3036 		cluster = list_entry(head, struct btrfs_free_cluster,
3037 				     block_group_list);
3038 
3039 		WARN_ON(cluster->block_group != block_group);
3040 		__btrfs_return_cluster_to_free_space(block_group, cluster);
3041 
3042 		cond_resched_lock(&ctl->tree_lock);
3043 	}
3044 	__btrfs_remove_free_space_cache(ctl);
3045 	btrfs_discard_update_discardable(block_group);
3046 	spin_unlock(&ctl->tree_lock);
3047 
3048 }
3049 
3050 /*
3051  * Walk @block_group's free space rb_tree to determine if everything is trimmed.
3052  */
btrfs_is_free_space_trimmed(struct btrfs_block_group * block_group)3053 bool btrfs_is_free_space_trimmed(struct btrfs_block_group *block_group)
3054 {
3055 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3056 	struct btrfs_free_space *info;
3057 	struct rb_node *node;
3058 	bool ret = true;
3059 
3060 	spin_lock(&ctl->tree_lock);
3061 	node = rb_first(&ctl->free_space_offset);
3062 
3063 	while (node) {
3064 		info = rb_entry(node, struct btrfs_free_space, offset_index);
3065 
3066 		if (!btrfs_free_space_trimmed(info)) {
3067 			ret = false;
3068 			break;
3069 		}
3070 
3071 		node = rb_next(node);
3072 	}
3073 
3074 	spin_unlock(&ctl->tree_lock);
3075 	return ret;
3076 }
3077 
btrfs_find_space_for_alloc(struct btrfs_block_group * block_group,u64 offset,u64 bytes,u64 empty_size,u64 * max_extent_size)3078 u64 btrfs_find_space_for_alloc(struct btrfs_block_group *block_group,
3079 			       u64 offset, u64 bytes, u64 empty_size,
3080 			       u64 *max_extent_size)
3081 {
3082 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3083 	struct btrfs_discard_ctl *discard_ctl =
3084 					&block_group->fs_info->discard_ctl;
3085 	struct btrfs_free_space *entry = NULL;
3086 	u64 bytes_search = bytes + empty_size;
3087 	u64 ret = 0;
3088 	u64 align_gap = 0;
3089 	u64 align_gap_len = 0;
3090 	enum btrfs_trim_state align_gap_trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3091 	bool use_bytes_index = (offset == block_group->start);
3092 
3093 	ASSERT(!btrfs_is_zoned(block_group->fs_info));
3094 
3095 	spin_lock(&ctl->tree_lock);
3096 	entry = find_free_space(ctl, &offset, &bytes_search,
3097 				block_group->full_stripe_len, max_extent_size,
3098 				use_bytes_index);
3099 	if (!entry)
3100 		goto out;
3101 
3102 	ret = offset;
3103 	if (entry->bitmap) {
3104 		bitmap_clear_bits(ctl, entry, offset, bytes, true);
3105 
3106 		if (!btrfs_free_space_trimmed(entry))
3107 			atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3108 
3109 		if (!entry->bytes)
3110 			free_bitmap(ctl, entry);
3111 	} else {
3112 		unlink_free_space(ctl, entry, true);
3113 		align_gap_len = offset - entry->offset;
3114 		align_gap = entry->offset;
3115 		align_gap_trim_state = entry->trim_state;
3116 
3117 		if (!btrfs_free_space_trimmed(entry))
3118 			atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3119 
3120 		entry->offset = offset + bytes;
3121 		WARN_ON(entry->bytes < bytes + align_gap_len);
3122 
3123 		entry->bytes -= bytes + align_gap_len;
3124 		if (!entry->bytes)
3125 			kmem_cache_free(btrfs_free_space_cachep, entry);
3126 		else
3127 			link_free_space(ctl, entry);
3128 	}
3129 out:
3130 	btrfs_discard_update_discardable(block_group);
3131 	spin_unlock(&ctl->tree_lock);
3132 
3133 	if (align_gap_len)
3134 		__btrfs_add_free_space(block_group, align_gap, align_gap_len,
3135 				       align_gap_trim_state);
3136 	return ret;
3137 }
3138 
3139 /*
3140  * given a cluster, put all of its extents back into the free space
3141  * cache.  If a block group is passed, this function will only free
3142  * a cluster that belongs to the passed block group.
3143  *
3144  * Otherwise, it'll get a reference on the block group pointed to by the
3145  * cluster and remove the cluster from it.
3146  */
btrfs_return_cluster_to_free_space(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster)3147 void btrfs_return_cluster_to_free_space(
3148 			       struct btrfs_block_group *block_group,
3149 			       struct btrfs_free_cluster *cluster)
3150 {
3151 	struct btrfs_free_space_ctl *ctl;
3152 
3153 	/* first, get a safe pointer to the block group */
3154 	spin_lock(&cluster->lock);
3155 	if (!block_group) {
3156 		block_group = cluster->block_group;
3157 		if (!block_group) {
3158 			spin_unlock(&cluster->lock);
3159 			return;
3160 		}
3161 	} else if (cluster->block_group != block_group) {
3162 		/* someone else has already freed it don't redo their work */
3163 		spin_unlock(&cluster->lock);
3164 		return;
3165 	}
3166 	btrfs_get_block_group(block_group);
3167 	spin_unlock(&cluster->lock);
3168 
3169 	ctl = block_group->free_space_ctl;
3170 
3171 	/* now return any extents the cluster had on it */
3172 	spin_lock(&ctl->tree_lock);
3173 	__btrfs_return_cluster_to_free_space(block_group, cluster);
3174 	spin_unlock(&ctl->tree_lock);
3175 
3176 	btrfs_discard_queue_work(&block_group->fs_info->discard_ctl, block_group);
3177 
3178 	/* finally drop our ref */
3179 	btrfs_put_block_group(block_group);
3180 }
3181 
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)3182 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group *block_group,
3183 				   struct btrfs_free_cluster *cluster,
3184 				   struct btrfs_free_space *entry,
3185 				   u64 bytes, u64 min_start,
3186 				   u64 *max_extent_size)
3187 {
3188 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3189 	int err;
3190 	u64 search_start = cluster->window_start;
3191 	u64 search_bytes = bytes;
3192 	u64 ret = 0;
3193 
3194 	search_start = min_start;
3195 	search_bytes = bytes;
3196 
3197 	err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
3198 	if (err) {
3199 		*max_extent_size = max(get_max_extent_size(entry),
3200 				       *max_extent_size);
3201 		return 0;
3202 	}
3203 
3204 	ret = search_start;
3205 	bitmap_clear_bits(ctl, entry, ret, bytes, false);
3206 
3207 	return ret;
3208 }
3209 
3210 /*
3211  * given a cluster, try to allocate 'bytes' from it, returns 0
3212  * if it couldn't find anything suitably large, or a logical disk offset
3213  * if things worked out
3214  */
btrfs_alloc_from_cluster(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster,u64 bytes,u64 min_start,u64 * max_extent_size)3215 u64 btrfs_alloc_from_cluster(struct btrfs_block_group *block_group,
3216 			     struct btrfs_free_cluster *cluster, u64 bytes,
3217 			     u64 min_start, u64 *max_extent_size)
3218 {
3219 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3220 	struct btrfs_discard_ctl *discard_ctl =
3221 					&block_group->fs_info->discard_ctl;
3222 	struct btrfs_free_space *entry = NULL;
3223 	struct rb_node *node;
3224 	u64 ret = 0;
3225 
3226 	ASSERT(!btrfs_is_zoned(block_group->fs_info));
3227 
3228 	spin_lock(&cluster->lock);
3229 	if (bytes > cluster->max_size)
3230 		goto out;
3231 
3232 	if (cluster->block_group != block_group)
3233 		goto out;
3234 
3235 	node = rb_first(&cluster->root);
3236 	if (!node)
3237 		goto out;
3238 
3239 	entry = rb_entry(node, struct btrfs_free_space, offset_index);
3240 	while (1) {
3241 		if (entry->bytes < bytes)
3242 			*max_extent_size = max(get_max_extent_size(entry),
3243 					       *max_extent_size);
3244 
3245 		if (entry->bytes < bytes ||
3246 		    (!entry->bitmap && entry->offset < min_start)) {
3247 			node = rb_next(&entry->offset_index);
3248 			if (!node)
3249 				break;
3250 			entry = rb_entry(node, struct btrfs_free_space,
3251 					 offset_index);
3252 			continue;
3253 		}
3254 
3255 		if (entry->bitmap) {
3256 			ret = btrfs_alloc_from_bitmap(block_group,
3257 						      cluster, entry, bytes,
3258 						      cluster->window_start,
3259 						      max_extent_size);
3260 			if (ret == 0) {
3261 				node = rb_next(&entry->offset_index);
3262 				if (!node)
3263 					break;
3264 				entry = rb_entry(node, struct btrfs_free_space,
3265 						 offset_index);
3266 				continue;
3267 			}
3268 			cluster->window_start += bytes;
3269 		} else {
3270 			ret = entry->offset;
3271 
3272 			entry->offset += bytes;
3273 			entry->bytes -= bytes;
3274 		}
3275 
3276 		break;
3277 	}
3278 out:
3279 	spin_unlock(&cluster->lock);
3280 
3281 	if (!ret)
3282 		return 0;
3283 
3284 	spin_lock(&ctl->tree_lock);
3285 
3286 	if (!btrfs_free_space_trimmed(entry))
3287 		atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3288 
3289 	ctl->free_space -= bytes;
3290 	if (!entry->bitmap && !btrfs_free_space_trimmed(entry))
3291 		ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
3292 
3293 	spin_lock(&cluster->lock);
3294 	if (entry->bytes == 0) {
3295 		rb_erase(&entry->offset_index, &cluster->root);
3296 		ctl->free_extents--;
3297 		if (entry->bitmap) {
3298 			kmem_cache_free(btrfs_free_space_bitmap_cachep,
3299 					entry->bitmap);
3300 			ctl->total_bitmaps--;
3301 			recalculate_thresholds(ctl);
3302 		} else if (!btrfs_free_space_trimmed(entry)) {
3303 			ctl->discardable_extents[BTRFS_STAT_CURR]--;
3304 		}
3305 		kmem_cache_free(btrfs_free_space_cachep, entry);
3306 	}
3307 
3308 	spin_unlock(&cluster->lock);
3309 	spin_unlock(&ctl->tree_lock);
3310 
3311 	return ret;
3312 }
3313 
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)3314 static int btrfs_bitmap_cluster(struct btrfs_block_group *block_group,
3315 				struct btrfs_free_space *entry,
3316 				struct btrfs_free_cluster *cluster,
3317 				u64 offset, u64 bytes,
3318 				u64 cont1_bytes, u64 min_bytes)
3319 {
3320 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3321 	unsigned long next_zero;
3322 	unsigned long i;
3323 	unsigned long want_bits;
3324 	unsigned long min_bits;
3325 	unsigned long found_bits;
3326 	unsigned long max_bits = 0;
3327 	unsigned long start = 0;
3328 	unsigned long total_found = 0;
3329 	int ret;
3330 
3331 	lockdep_assert_held(&ctl->tree_lock);
3332 
3333 	i = offset_to_bit(entry->offset, ctl->unit,
3334 			  max_t(u64, offset, entry->offset));
3335 	want_bits = bytes_to_bits(bytes, ctl->unit);
3336 	min_bits = bytes_to_bits(min_bytes, ctl->unit);
3337 
3338 	/*
3339 	 * Don't bother looking for a cluster in this bitmap if it's heavily
3340 	 * fragmented.
3341 	 */
3342 	if (entry->max_extent_size &&
3343 	    entry->max_extent_size < cont1_bytes)
3344 		return -ENOSPC;
3345 again:
3346 	found_bits = 0;
3347 	for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
3348 		next_zero = find_next_zero_bit(entry->bitmap,
3349 					       BITS_PER_BITMAP, i);
3350 		if (next_zero - i >= min_bits) {
3351 			found_bits = next_zero - i;
3352 			if (found_bits > max_bits)
3353 				max_bits = found_bits;
3354 			break;
3355 		}
3356 		if (next_zero - i > max_bits)
3357 			max_bits = next_zero - i;
3358 		i = next_zero;
3359 	}
3360 
3361 	if (!found_bits) {
3362 		entry->max_extent_size = (u64)max_bits * ctl->unit;
3363 		return -ENOSPC;
3364 	}
3365 
3366 	if (!total_found) {
3367 		start = i;
3368 		cluster->max_size = 0;
3369 	}
3370 
3371 	total_found += found_bits;
3372 
3373 	if (cluster->max_size < found_bits * ctl->unit)
3374 		cluster->max_size = found_bits * ctl->unit;
3375 
3376 	if (total_found < want_bits || cluster->max_size < cont1_bytes) {
3377 		i = next_zero + 1;
3378 		goto again;
3379 	}
3380 
3381 	cluster->window_start = start * ctl->unit + entry->offset;
3382 	rb_erase(&entry->offset_index, &ctl->free_space_offset);
3383 	rb_erase_cached(&entry->bytes_index, &ctl->free_space_bytes);
3384 
3385 	/*
3386 	 * We need to know if we're currently on the normal space index when we
3387 	 * manipulate the bitmap so that we know we need to remove and re-insert
3388 	 * it into the space_index tree.  Clear the bytes_index node here so the
3389 	 * bitmap manipulation helpers know not to mess with the space_index
3390 	 * until this bitmap entry is added back into the normal cache.
3391 	 */
3392 	RB_CLEAR_NODE(&entry->bytes_index);
3393 
3394 	ret = tree_insert_offset(ctl, cluster, entry);
3395 	ASSERT(!ret); /* -EEXIST; Logic error */
3396 
3397 	trace_btrfs_setup_cluster(block_group, cluster,
3398 				  total_found * ctl->unit, 1);
3399 	return 0;
3400 }
3401 
3402 /*
3403  * This searches the block group for just extents to fill the cluster with.
3404  * Try to find a cluster with at least bytes total bytes, at least one
3405  * extent of cont1_bytes, and other clusters of at least min_bytes.
3406  */
3407 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)3408 setup_cluster_no_bitmap(struct btrfs_block_group *block_group,
3409 			struct btrfs_free_cluster *cluster,
3410 			struct list_head *bitmaps, u64 offset, u64 bytes,
3411 			u64 cont1_bytes, u64 min_bytes)
3412 {
3413 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3414 	struct btrfs_free_space *first = NULL;
3415 	struct btrfs_free_space *entry = NULL;
3416 	struct btrfs_free_space *last;
3417 	struct rb_node *node;
3418 	u64 window_free;
3419 	u64 max_extent;
3420 	u64 total_size = 0;
3421 
3422 	lockdep_assert_held(&ctl->tree_lock);
3423 
3424 	entry = tree_search_offset(ctl, offset, 0, 1);
3425 	if (!entry)
3426 		return -ENOSPC;
3427 
3428 	/*
3429 	 * We don't want bitmaps, so just move along until we find a normal
3430 	 * extent entry.
3431 	 */
3432 	while (entry->bitmap || entry->bytes < min_bytes) {
3433 		if (entry->bitmap && list_empty(&entry->list))
3434 			list_add_tail(&entry->list, bitmaps);
3435 		node = rb_next(&entry->offset_index);
3436 		if (!node)
3437 			return -ENOSPC;
3438 		entry = rb_entry(node, struct btrfs_free_space, offset_index);
3439 	}
3440 
3441 	window_free = entry->bytes;
3442 	max_extent = entry->bytes;
3443 	first = entry;
3444 	last = entry;
3445 
3446 	for (node = rb_next(&entry->offset_index); node;
3447 	     node = rb_next(&entry->offset_index)) {
3448 		entry = rb_entry(node, struct btrfs_free_space, offset_index);
3449 
3450 		if (entry->bitmap) {
3451 			if (list_empty(&entry->list))
3452 				list_add_tail(&entry->list, bitmaps);
3453 			continue;
3454 		}
3455 
3456 		if (entry->bytes < min_bytes)
3457 			continue;
3458 
3459 		last = entry;
3460 		window_free += entry->bytes;
3461 		if (entry->bytes > max_extent)
3462 			max_extent = entry->bytes;
3463 	}
3464 
3465 	if (window_free < bytes || max_extent < cont1_bytes)
3466 		return -ENOSPC;
3467 
3468 	cluster->window_start = first->offset;
3469 
3470 	node = &first->offset_index;
3471 
3472 	/*
3473 	 * now we've found our entries, pull them out of the free space
3474 	 * cache and put them into the cluster rbtree
3475 	 */
3476 	do {
3477 		int ret;
3478 
3479 		entry = rb_entry(node, struct btrfs_free_space, offset_index);
3480 		node = rb_next(&entry->offset_index);
3481 		if (entry->bitmap || entry->bytes < min_bytes)
3482 			continue;
3483 
3484 		rb_erase(&entry->offset_index, &ctl->free_space_offset);
3485 		rb_erase_cached(&entry->bytes_index, &ctl->free_space_bytes);
3486 		ret = tree_insert_offset(ctl, cluster, entry);
3487 		total_size += entry->bytes;
3488 		ASSERT(!ret); /* -EEXIST; Logic error */
3489 	} while (node && entry != last);
3490 
3491 	cluster->max_size = max_extent;
3492 	trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
3493 	return 0;
3494 }
3495 
3496 /*
3497  * This specifically looks for bitmaps that may work in the cluster, we assume
3498  * that we have already failed to find extents that will work.
3499  */
3500 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)3501 setup_cluster_bitmap(struct btrfs_block_group *block_group,
3502 		     struct btrfs_free_cluster *cluster,
3503 		     struct list_head *bitmaps, u64 offset, u64 bytes,
3504 		     u64 cont1_bytes, u64 min_bytes)
3505 {
3506 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3507 	struct btrfs_free_space *entry = NULL;
3508 	int ret = -ENOSPC;
3509 	u64 bitmap_offset = offset_to_bitmap(ctl, offset);
3510 
3511 	if (ctl->total_bitmaps == 0)
3512 		return -ENOSPC;
3513 
3514 	/*
3515 	 * The bitmap that covers offset won't be in the list unless offset
3516 	 * is just its start offset.
3517 	 */
3518 	if (!list_empty(bitmaps))
3519 		entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
3520 
3521 	if (!entry || entry->offset != bitmap_offset) {
3522 		entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
3523 		if (entry && list_empty(&entry->list))
3524 			list_add(&entry->list, bitmaps);
3525 	}
3526 
3527 	list_for_each_entry(entry, bitmaps, list) {
3528 		if (entry->bytes < bytes)
3529 			continue;
3530 		ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
3531 					   bytes, cont1_bytes, min_bytes);
3532 		if (!ret)
3533 			return 0;
3534 	}
3535 
3536 	/*
3537 	 * The bitmaps list has all the bitmaps that record free space
3538 	 * starting after offset, so no more search is required.
3539 	 */
3540 	return -ENOSPC;
3541 }
3542 
3543 /*
3544  * here we try to find a cluster of blocks in a block group.  The goal
3545  * is to find at least bytes+empty_size.
3546  * We might not find them all in one contiguous area.
3547  *
3548  * returns zero and sets up cluster if things worked out, otherwise
3549  * it returns -enospc
3550  */
btrfs_find_space_cluster(struct btrfs_block_group * block_group,struct btrfs_free_cluster * cluster,u64 offset,u64 bytes,u64 empty_size)3551 int btrfs_find_space_cluster(struct btrfs_block_group *block_group,
3552 			     struct btrfs_free_cluster *cluster,
3553 			     u64 offset, u64 bytes, u64 empty_size)
3554 {
3555 	struct btrfs_fs_info *fs_info = block_group->fs_info;
3556 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3557 	struct btrfs_free_space *entry, *tmp;
3558 	LIST_HEAD(bitmaps);
3559 	u64 min_bytes;
3560 	u64 cont1_bytes;
3561 	int ret;
3562 
3563 	/*
3564 	 * Choose the minimum extent size we'll require for this
3565 	 * cluster.  For SSD_SPREAD, don't allow any fragmentation.
3566 	 * For metadata, allow allocates with smaller extents.  For
3567 	 * data, keep it dense.
3568 	 */
3569 	if (btrfs_test_opt(fs_info, SSD_SPREAD)) {
3570 		cont1_bytes = bytes + empty_size;
3571 		min_bytes = cont1_bytes;
3572 	} else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3573 		cont1_bytes = bytes;
3574 		min_bytes = fs_info->sectorsize;
3575 	} else {
3576 		cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3577 		min_bytes = fs_info->sectorsize;
3578 	}
3579 
3580 	spin_lock(&ctl->tree_lock);
3581 
3582 	/*
3583 	 * If we know we don't have enough space to make a cluster don't even
3584 	 * bother doing all the work to try and find one.
3585 	 */
3586 	if (ctl->free_space < bytes) {
3587 		spin_unlock(&ctl->tree_lock);
3588 		return -ENOSPC;
3589 	}
3590 
3591 	spin_lock(&cluster->lock);
3592 
3593 	/* someone already found a cluster, hooray */
3594 	if (cluster->block_group) {
3595 		ret = 0;
3596 		goto out;
3597 	}
3598 
3599 	trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
3600 				 min_bytes);
3601 
3602 	ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
3603 				      bytes + empty_size,
3604 				      cont1_bytes, min_bytes);
3605 	if (ret)
3606 		ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
3607 					   offset, bytes + empty_size,
3608 					   cont1_bytes, min_bytes);
3609 
3610 	/* Clear our temporary list */
3611 	list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3612 		list_del_init(&entry->list);
3613 
3614 	if (!ret) {
3615 		btrfs_get_block_group(block_group);
3616 		list_add_tail(&cluster->block_group_list,
3617 			      &block_group->cluster_list);
3618 		cluster->block_group = block_group;
3619 	} else {
3620 		trace_btrfs_failed_cluster_setup(block_group);
3621 	}
3622 out:
3623 	spin_unlock(&cluster->lock);
3624 	spin_unlock(&ctl->tree_lock);
3625 
3626 	return ret;
3627 }
3628 
3629 /*
3630  * simple code to zero out a cluster
3631  */
btrfs_init_free_cluster(struct btrfs_free_cluster * cluster)3632 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3633 {
3634 	spin_lock_init(&cluster->lock);
3635 	spin_lock_init(&cluster->refill_lock);
3636 	cluster->root = RB_ROOT;
3637 	cluster->max_size = 0;
3638 	cluster->fragmented = false;
3639 	INIT_LIST_HEAD(&cluster->block_group_list);
3640 	cluster->block_group = NULL;
3641 }
3642 
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)3643 static int do_trimming(struct btrfs_block_group *block_group,
3644 		       u64 *total_trimmed, u64 start, u64 bytes,
3645 		       u64 reserved_start, u64 reserved_bytes,
3646 		       enum btrfs_trim_state reserved_trim_state,
3647 		       struct btrfs_trim_range *trim_entry)
3648 {
3649 	struct btrfs_space_info *space_info = block_group->space_info;
3650 	struct btrfs_fs_info *fs_info = block_group->fs_info;
3651 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3652 	int ret;
3653 	int update = 0;
3654 	const u64 end = start + bytes;
3655 	const u64 reserved_end = reserved_start + reserved_bytes;
3656 	enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3657 	u64 trimmed = 0;
3658 
3659 	spin_lock(&space_info->lock);
3660 	spin_lock(&block_group->lock);
3661 	if (!block_group->ro) {
3662 		block_group->reserved += reserved_bytes;
3663 		space_info->bytes_reserved += reserved_bytes;
3664 		update = 1;
3665 	}
3666 	spin_unlock(&block_group->lock);
3667 	spin_unlock(&space_info->lock);
3668 
3669 	ret = btrfs_discard_extent(fs_info, start, bytes, &trimmed);
3670 	if (!ret) {
3671 		*total_trimmed += trimmed;
3672 		trim_state = BTRFS_TRIM_STATE_TRIMMED;
3673 	}
3674 
3675 	mutex_lock(&ctl->cache_writeout_mutex);
3676 	if (reserved_start < start)
3677 		__btrfs_add_free_space(block_group, reserved_start,
3678 				       start - reserved_start,
3679 				       reserved_trim_state);
3680 	if (end < reserved_end)
3681 		__btrfs_add_free_space(block_group, end, reserved_end - end,
3682 				       reserved_trim_state);
3683 	__btrfs_add_free_space(block_group, start, bytes, trim_state);
3684 	list_del(&trim_entry->list);
3685 	mutex_unlock(&ctl->cache_writeout_mutex);
3686 
3687 	if (update) {
3688 		spin_lock(&space_info->lock);
3689 		spin_lock(&block_group->lock);
3690 		if (block_group->ro)
3691 			space_info->bytes_readonly += reserved_bytes;
3692 		block_group->reserved -= reserved_bytes;
3693 		space_info->bytes_reserved -= reserved_bytes;
3694 		spin_unlock(&block_group->lock);
3695 		spin_unlock(&space_info->lock);
3696 	}
3697 
3698 	return ret;
3699 }
3700 
3701 /*
3702  * If @async is set, then we will trim 1 region and return.
3703  */
trim_no_bitmap(struct btrfs_block_group * block_group,u64 * total_trimmed,u64 start,u64 end,u64 minlen,bool async)3704 static int trim_no_bitmap(struct btrfs_block_group *block_group,
3705 			  u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3706 			  bool async)
3707 {
3708 	struct btrfs_discard_ctl *discard_ctl =
3709 					&block_group->fs_info->discard_ctl;
3710 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3711 	struct btrfs_free_space *entry;
3712 	struct rb_node *node;
3713 	int ret = 0;
3714 	u64 extent_start;
3715 	u64 extent_bytes;
3716 	enum btrfs_trim_state extent_trim_state;
3717 	u64 bytes;
3718 	const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3719 
3720 	while (start < end) {
3721 		struct btrfs_trim_range trim_entry;
3722 
3723 		mutex_lock(&ctl->cache_writeout_mutex);
3724 		spin_lock(&ctl->tree_lock);
3725 
3726 		if (ctl->free_space < minlen)
3727 			goto out_unlock;
3728 
3729 		entry = tree_search_offset(ctl, start, 0, 1);
3730 		if (!entry)
3731 			goto out_unlock;
3732 
3733 		/* Skip bitmaps and if async, already trimmed entries */
3734 		while (entry->bitmap ||
3735 		       (async && btrfs_free_space_trimmed(entry))) {
3736 			node = rb_next(&entry->offset_index);
3737 			if (!node)
3738 				goto out_unlock;
3739 			entry = rb_entry(node, struct btrfs_free_space,
3740 					 offset_index);
3741 		}
3742 
3743 		if (entry->offset >= end)
3744 			goto out_unlock;
3745 
3746 		extent_start = entry->offset;
3747 		extent_bytes = entry->bytes;
3748 		extent_trim_state = entry->trim_state;
3749 		if (async) {
3750 			start = entry->offset;
3751 			bytes = entry->bytes;
3752 			if (bytes < minlen) {
3753 				spin_unlock(&ctl->tree_lock);
3754 				mutex_unlock(&ctl->cache_writeout_mutex);
3755 				goto next;
3756 			}
3757 			unlink_free_space(ctl, entry, true);
3758 			/*
3759 			 * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3760 			 * If X < BTRFS_ASYNC_DISCARD_MIN_FILTER, we won't trim
3761 			 * X when we come back around.  So trim it now.
3762 			 */
3763 			if (max_discard_size &&
3764 			    bytes >= (max_discard_size +
3765 				      BTRFS_ASYNC_DISCARD_MIN_FILTER)) {
3766 				bytes = max_discard_size;
3767 				extent_bytes = max_discard_size;
3768 				entry->offset += max_discard_size;
3769 				entry->bytes -= max_discard_size;
3770 				link_free_space(ctl, entry);
3771 			} else {
3772 				kmem_cache_free(btrfs_free_space_cachep, entry);
3773 			}
3774 		} else {
3775 			start = max(start, extent_start);
3776 			bytes = min(extent_start + extent_bytes, end) - start;
3777 			if (bytes < minlen) {
3778 				spin_unlock(&ctl->tree_lock);
3779 				mutex_unlock(&ctl->cache_writeout_mutex);
3780 				goto next;
3781 			}
3782 
3783 			unlink_free_space(ctl, entry, true);
3784 			kmem_cache_free(btrfs_free_space_cachep, entry);
3785 		}
3786 
3787 		spin_unlock(&ctl->tree_lock);
3788 		trim_entry.start = extent_start;
3789 		trim_entry.bytes = extent_bytes;
3790 		list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3791 		mutex_unlock(&ctl->cache_writeout_mutex);
3792 
3793 		ret = do_trimming(block_group, total_trimmed, start, bytes,
3794 				  extent_start, extent_bytes, extent_trim_state,
3795 				  &trim_entry);
3796 		if (ret) {
3797 			block_group->discard_cursor = start + bytes;
3798 			break;
3799 		}
3800 next:
3801 		start += bytes;
3802 		block_group->discard_cursor = start;
3803 		if (async && *total_trimmed)
3804 			break;
3805 
3806 		if (fatal_signal_pending(current)) {
3807 			ret = -ERESTARTSYS;
3808 			break;
3809 		}
3810 
3811 		cond_resched();
3812 	}
3813 
3814 	return ret;
3815 
3816 out_unlock:
3817 	block_group->discard_cursor = btrfs_block_group_end(block_group);
3818 	spin_unlock(&ctl->tree_lock);
3819 	mutex_unlock(&ctl->cache_writeout_mutex);
3820 
3821 	return ret;
3822 }
3823 
3824 /*
3825  * If we break out of trimming a bitmap prematurely, we should reset the
3826  * trimming bit.  In a rather contrieved case, it's possible to race here so
3827  * reset the state to BTRFS_TRIM_STATE_UNTRIMMED.
3828  *
3829  * start = start of bitmap
3830  * end = near end of bitmap
3831  *
3832  * Thread 1:			Thread 2:
3833  * trim_bitmaps(start)
3834  *				trim_bitmaps(end)
3835  *				end_trimming_bitmap()
3836  * reset_trimming_bitmap()
3837  */
reset_trimming_bitmap(struct btrfs_free_space_ctl * ctl,u64 offset)3838 static void reset_trimming_bitmap(struct btrfs_free_space_ctl *ctl, u64 offset)
3839 {
3840 	struct btrfs_free_space *entry;
3841 
3842 	spin_lock(&ctl->tree_lock);
3843 	entry = tree_search_offset(ctl, offset, 1, 0);
3844 	if (entry) {
3845 		if (btrfs_free_space_trimmed(entry)) {
3846 			ctl->discardable_extents[BTRFS_STAT_CURR] +=
3847 				entry->bitmap_extents;
3848 			ctl->discardable_bytes[BTRFS_STAT_CURR] += entry->bytes;
3849 		}
3850 		entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3851 	}
3852 
3853 	spin_unlock(&ctl->tree_lock);
3854 }
3855 
end_trimming_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * entry)3856 static void end_trimming_bitmap(struct btrfs_free_space_ctl *ctl,
3857 				struct btrfs_free_space *entry)
3858 {
3859 	if (btrfs_free_space_trimming_bitmap(entry)) {
3860 		entry->trim_state = BTRFS_TRIM_STATE_TRIMMED;
3861 		ctl->discardable_extents[BTRFS_STAT_CURR] -=
3862 			entry->bitmap_extents;
3863 		ctl->discardable_bytes[BTRFS_STAT_CURR] -= entry->bytes;
3864 	}
3865 }
3866 
3867 /*
3868  * If @async is set, then we will trim 1 region and return.
3869  */
trim_bitmaps(struct btrfs_block_group * block_group,u64 * total_trimmed,u64 start,u64 end,u64 minlen,u64 maxlen,bool async)3870 static int trim_bitmaps(struct btrfs_block_group *block_group,
3871 			u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3872 			u64 maxlen, bool async)
3873 {
3874 	struct btrfs_discard_ctl *discard_ctl =
3875 					&block_group->fs_info->discard_ctl;
3876 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3877 	struct btrfs_free_space *entry;
3878 	int ret = 0;
3879 	int ret2;
3880 	u64 bytes;
3881 	u64 offset = offset_to_bitmap(ctl, start);
3882 	const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3883 
3884 	while (offset < end) {
3885 		bool next_bitmap = false;
3886 		struct btrfs_trim_range trim_entry;
3887 
3888 		mutex_lock(&ctl->cache_writeout_mutex);
3889 		spin_lock(&ctl->tree_lock);
3890 
3891 		if (ctl->free_space < minlen) {
3892 			block_group->discard_cursor =
3893 				btrfs_block_group_end(block_group);
3894 			spin_unlock(&ctl->tree_lock);
3895 			mutex_unlock(&ctl->cache_writeout_mutex);
3896 			break;
3897 		}
3898 
3899 		entry = tree_search_offset(ctl, offset, 1, 0);
3900 		/*
3901 		 * Bitmaps are marked trimmed lossily now to prevent constant
3902 		 * discarding of the same bitmap (the reason why we are bound
3903 		 * by the filters).  So, retrim the block group bitmaps when we
3904 		 * are preparing to punt to the unused_bgs list.  This uses
3905 		 * @minlen to determine if we are in BTRFS_DISCARD_INDEX_UNUSED
3906 		 * which is the only discard index which sets minlen to 0.
3907 		 */
3908 		if (!entry || (async && minlen && start == offset &&
3909 			       btrfs_free_space_trimmed(entry))) {
3910 			spin_unlock(&ctl->tree_lock);
3911 			mutex_unlock(&ctl->cache_writeout_mutex);
3912 			next_bitmap = true;
3913 			goto next;
3914 		}
3915 
3916 		/*
3917 		 * Async discard bitmap trimming begins at by setting the start
3918 		 * to be key.objectid and the offset_to_bitmap() aligns to the
3919 		 * start of the bitmap.  This lets us know we are fully
3920 		 * scanning the bitmap rather than only some portion of it.
3921 		 */
3922 		if (start == offset)
3923 			entry->trim_state = BTRFS_TRIM_STATE_TRIMMING;
3924 
3925 		bytes = minlen;
3926 		ret2 = search_bitmap(ctl, entry, &start, &bytes, false);
3927 		if (ret2 || start >= end) {
3928 			/*
3929 			 * We lossily consider a bitmap trimmed if we only skip
3930 			 * over regions <= BTRFS_ASYNC_DISCARD_MIN_FILTER.
3931 			 */
3932 			if (ret2 && minlen <= BTRFS_ASYNC_DISCARD_MIN_FILTER)
3933 				end_trimming_bitmap(ctl, entry);
3934 			else
3935 				entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3936 			spin_unlock(&ctl->tree_lock);
3937 			mutex_unlock(&ctl->cache_writeout_mutex);
3938 			next_bitmap = true;
3939 			goto next;
3940 		}
3941 
3942 		/*
3943 		 * We already trimmed a region, but are using the locking above
3944 		 * to reset the trim_state.
3945 		 */
3946 		if (async && *total_trimmed) {
3947 			spin_unlock(&ctl->tree_lock);
3948 			mutex_unlock(&ctl->cache_writeout_mutex);
3949 			goto out;
3950 		}
3951 
3952 		bytes = min(bytes, end - start);
3953 		if (bytes < minlen || (async && maxlen && bytes > maxlen)) {
3954 			spin_unlock(&ctl->tree_lock);
3955 			mutex_unlock(&ctl->cache_writeout_mutex);
3956 			goto next;
3957 		}
3958 
3959 		/*
3960 		 * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3961 		 * If X < @minlen, we won't trim X when we come back around.
3962 		 * So trim it now.  We differ here from trimming extents as we
3963 		 * don't keep individual state per bit.
3964 		 */
3965 		if (async &&
3966 		    max_discard_size &&
3967 		    bytes > (max_discard_size + minlen))
3968 			bytes = max_discard_size;
3969 
3970 		bitmap_clear_bits(ctl, entry, start, bytes, true);
3971 		if (entry->bytes == 0)
3972 			free_bitmap(ctl, entry);
3973 
3974 		spin_unlock(&ctl->tree_lock);
3975 		trim_entry.start = start;
3976 		trim_entry.bytes = bytes;
3977 		list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3978 		mutex_unlock(&ctl->cache_writeout_mutex);
3979 
3980 		ret = do_trimming(block_group, total_trimmed, start, bytes,
3981 				  start, bytes, 0, &trim_entry);
3982 		if (ret) {
3983 			reset_trimming_bitmap(ctl, offset);
3984 			block_group->discard_cursor =
3985 				btrfs_block_group_end(block_group);
3986 			break;
3987 		}
3988 next:
3989 		if (next_bitmap) {
3990 			offset += BITS_PER_BITMAP * ctl->unit;
3991 			start = offset;
3992 		} else {
3993 			start += bytes;
3994 		}
3995 		block_group->discard_cursor = start;
3996 
3997 		if (fatal_signal_pending(current)) {
3998 			if (start != offset)
3999 				reset_trimming_bitmap(ctl, offset);
4000 			ret = -ERESTARTSYS;
4001 			break;
4002 		}
4003 
4004 		cond_resched();
4005 	}
4006 
4007 	if (offset >= end)
4008 		block_group->discard_cursor = end;
4009 
4010 out:
4011 	return ret;
4012 }
4013 
btrfs_trim_block_group(struct btrfs_block_group * block_group,u64 * trimmed,u64 start,u64 end,u64 minlen)4014 int btrfs_trim_block_group(struct btrfs_block_group *block_group,
4015 			   u64 *trimmed, u64 start, u64 end, u64 minlen)
4016 {
4017 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
4018 	int ret;
4019 	u64 rem = 0;
4020 
4021 	ASSERT(!btrfs_is_zoned(block_group->fs_info));
4022 
4023 	*trimmed = 0;
4024 
4025 	spin_lock(&block_group->lock);
4026 	if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4027 		spin_unlock(&block_group->lock);
4028 		return 0;
4029 	}
4030 	btrfs_freeze_block_group(block_group);
4031 	spin_unlock(&block_group->lock);
4032 
4033 	ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, false);
4034 	if (ret)
4035 		goto out;
4036 
4037 	ret = trim_bitmaps(block_group, trimmed, start, end, minlen, 0, false);
4038 	div64_u64_rem(end, BITS_PER_BITMAP * ctl->unit, &rem);
4039 	/* If we ended in the middle of a bitmap, reset the trimming flag */
4040 	if (rem)
4041 		reset_trimming_bitmap(ctl, offset_to_bitmap(ctl, end));
4042 out:
4043 	btrfs_unfreeze_block_group(block_group);
4044 	return ret;
4045 }
4046 
btrfs_trim_block_group_extents(struct btrfs_block_group * block_group,u64 * trimmed,u64 start,u64 end,u64 minlen,bool async)4047 int btrfs_trim_block_group_extents(struct btrfs_block_group *block_group,
4048 				   u64 *trimmed, u64 start, u64 end, u64 minlen,
4049 				   bool async)
4050 {
4051 	int ret;
4052 
4053 	*trimmed = 0;
4054 
4055 	spin_lock(&block_group->lock);
4056 	if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4057 		spin_unlock(&block_group->lock);
4058 		return 0;
4059 	}
4060 	btrfs_freeze_block_group(block_group);
4061 	spin_unlock(&block_group->lock);
4062 
4063 	ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, async);
4064 	btrfs_unfreeze_block_group(block_group);
4065 
4066 	return ret;
4067 }
4068 
btrfs_trim_block_group_bitmaps(struct btrfs_block_group * block_group,u64 * trimmed,u64 start,u64 end,u64 minlen,u64 maxlen,bool async)4069 int btrfs_trim_block_group_bitmaps(struct btrfs_block_group *block_group,
4070 				   u64 *trimmed, u64 start, u64 end, u64 minlen,
4071 				   u64 maxlen, bool async)
4072 {
4073 	int ret;
4074 
4075 	*trimmed = 0;
4076 
4077 	spin_lock(&block_group->lock);
4078 	if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4079 		spin_unlock(&block_group->lock);
4080 		return 0;
4081 	}
4082 	btrfs_freeze_block_group(block_group);
4083 	spin_unlock(&block_group->lock);
4084 
4085 	ret = trim_bitmaps(block_group, trimmed, start, end, minlen, maxlen,
4086 			   async);
4087 
4088 	btrfs_unfreeze_block_group(block_group);
4089 
4090 	return ret;
4091 }
4092 
btrfs_free_space_cache_v1_active(struct btrfs_fs_info * fs_info)4093 bool btrfs_free_space_cache_v1_active(struct btrfs_fs_info *fs_info)
4094 {
4095 	return btrfs_super_cache_generation(fs_info->super_copy);
4096 }
4097 
cleanup_free_space_cache_v1(struct btrfs_fs_info * fs_info,struct btrfs_trans_handle * trans)4098 static int cleanup_free_space_cache_v1(struct btrfs_fs_info *fs_info,
4099 				       struct btrfs_trans_handle *trans)
4100 {
4101 	struct btrfs_block_group *block_group;
4102 	struct rb_node *node;
4103 	int ret = 0;
4104 
4105 	btrfs_info(fs_info, "cleaning free space cache v1");
4106 
4107 	node = rb_first_cached(&fs_info->block_group_cache_tree);
4108 	while (node) {
4109 		block_group = rb_entry(node, struct btrfs_block_group, cache_node);
4110 		ret = btrfs_remove_free_space_inode(trans, NULL, block_group);
4111 		if (ret)
4112 			goto out;
4113 		node = rb_next(node);
4114 	}
4115 out:
4116 	return ret;
4117 }
4118 
btrfs_set_free_space_cache_v1_active(struct btrfs_fs_info * fs_info,bool active)4119 int btrfs_set_free_space_cache_v1_active(struct btrfs_fs_info *fs_info, bool active)
4120 {
4121 	struct btrfs_trans_handle *trans;
4122 	int ret;
4123 
4124 	/*
4125 	 * update_super_roots will appropriately set or unset
4126 	 * super_copy->cache_generation based on SPACE_CACHE and
4127 	 * BTRFS_FS_CLEANUP_SPACE_CACHE_V1. For this reason, we need a
4128 	 * transaction commit whether we are enabling space cache v1 and don't
4129 	 * have any other work to do, or are disabling it and removing free
4130 	 * space inodes.
4131 	 */
4132 	trans = btrfs_start_transaction(fs_info->tree_root, 0);
4133 	if (IS_ERR(trans))
4134 		return PTR_ERR(trans);
4135 
4136 	if (!active) {
4137 		set_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);
4138 		ret = cleanup_free_space_cache_v1(fs_info, trans);
4139 		if (ret) {
4140 			btrfs_abort_transaction(trans, ret);
4141 			btrfs_end_transaction(trans);
4142 			goto out;
4143 		}
4144 	}
4145 
4146 	ret = btrfs_commit_transaction(trans);
4147 out:
4148 	clear_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);
4149 
4150 	return ret;
4151 }
4152 
btrfs_free_space_init(void)4153 int __init btrfs_free_space_init(void)
4154 {
4155 	btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
4156 			sizeof(struct btrfs_free_space), 0,
4157 			SLAB_MEM_SPREAD, NULL);
4158 	if (!btrfs_free_space_cachep)
4159 		return -ENOMEM;
4160 
4161 	btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
4162 							PAGE_SIZE, PAGE_SIZE,
4163 							SLAB_MEM_SPREAD, NULL);
4164 	if (!btrfs_free_space_bitmap_cachep) {
4165 		kmem_cache_destroy(btrfs_free_space_cachep);
4166 		return -ENOMEM;
4167 	}
4168 
4169 	return 0;
4170 }
4171 
btrfs_free_space_exit(void)4172 void __cold btrfs_free_space_exit(void)
4173 {
4174 	kmem_cache_destroy(btrfs_free_space_cachep);
4175 	kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
4176 }
4177 
4178 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4179 /*
4180  * Use this if you need to make a bitmap or extent entry specifically, it
4181  * doesn't do any of the merging that add_free_space does, this acts a lot like
4182  * how the free space cache loading stuff works, so you can get really weird
4183  * configurations.
4184  */
test_add_free_space_entry(struct btrfs_block_group * cache,u64 offset,u64 bytes,bool bitmap)4185 int test_add_free_space_entry(struct btrfs_block_group *cache,
4186 			      u64 offset, u64 bytes, bool bitmap)
4187 {
4188 	struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4189 	struct btrfs_free_space *info = NULL, *bitmap_info;
4190 	void *map = NULL;
4191 	enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_TRIMMED;
4192 	u64 bytes_added;
4193 	int ret;
4194 
4195 again:
4196 	if (!info) {
4197 		info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
4198 		if (!info)
4199 			return -ENOMEM;
4200 	}
4201 
4202 	if (!bitmap) {
4203 		spin_lock(&ctl->tree_lock);
4204 		info->offset = offset;
4205 		info->bytes = bytes;
4206 		info->max_extent_size = 0;
4207 		ret = link_free_space(ctl, info);
4208 		spin_unlock(&ctl->tree_lock);
4209 		if (ret)
4210 			kmem_cache_free(btrfs_free_space_cachep, info);
4211 		return ret;
4212 	}
4213 
4214 	if (!map) {
4215 		map = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep, GFP_NOFS);
4216 		if (!map) {
4217 			kmem_cache_free(btrfs_free_space_cachep, info);
4218 			return -ENOMEM;
4219 		}
4220 	}
4221 
4222 	spin_lock(&ctl->tree_lock);
4223 	bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
4224 					 1, 0);
4225 	if (!bitmap_info) {
4226 		info->bitmap = map;
4227 		map = NULL;
4228 		add_new_bitmap(ctl, info, offset);
4229 		bitmap_info = info;
4230 		info = NULL;
4231 	}
4232 
4233 	bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
4234 					  trim_state);
4235 
4236 	bytes -= bytes_added;
4237 	offset += bytes_added;
4238 	spin_unlock(&ctl->tree_lock);
4239 
4240 	if (bytes)
4241 		goto again;
4242 
4243 	if (info)
4244 		kmem_cache_free(btrfs_free_space_cachep, info);
4245 	if (map)
4246 		kmem_cache_free(btrfs_free_space_bitmap_cachep, map);
4247 	return 0;
4248 }
4249 
4250 /*
4251  * Checks to see if the given range is in the free space cache.  This is really
4252  * just used to check the absence of space, so if there is free space in the
4253  * range at all we will return 1.
4254  */
test_check_exists(struct btrfs_block_group * cache,u64 offset,u64 bytes)4255 int test_check_exists(struct btrfs_block_group *cache,
4256 		      u64 offset, u64 bytes)
4257 {
4258 	struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4259 	struct btrfs_free_space *info;
4260 	int ret = 0;
4261 
4262 	spin_lock(&ctl->tree_lock);
4263 	info = tree_search_offset(ctl, offset, 0, 0);
4264 	if (!info) {
4265 		info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
4266 					  1, 0);
4267 		if (!info)
4268 			goto out;
4269 	}
4270 
4271 have_info:
4272 	if (info->bitmap) {
4273 		u64 bit_off, bit_bytes;
4274 		struct rb_node *n;
4275 		struct btrfs_free_space *tmp;
4276 
4277 		bit_off = offset;
4278 		bit_bytes = ctl->unit;
4279 		ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false);
4280 		if (!ret) {
4281 			if (bit_off == offset) {
4282 				ret = 1;
4283 				goto out;
4284 			} else if (bit_off > offset &&
4285 				   offset + bytes > bit_off) {
4286 				ret = 1;
4287 				goto out;
4288 			}
4289 		}
4290 
4291 		n = rb_prev(&info->offset_index);
4292 		while (n) {
4293 			tmp = rb_entry(n, struct btrfs_free_space,
4294 				       offset_index);
4295 			if (tmp->offset + tmp->bytes < offset)
4296 				break;
4297 			if (offset + bytes < tmp->offset) {
4298 				n = rb_prev(&tmp->offset_index);
4299 				continue;
4300 			}
4301 			info = tmp;
4302 			goto have_info;
4303 		}
4304 
4305 		n = rb_next(&info->offset_index);
4306 		while (n) {
4307 			tmp = rb_entry(n, struct btrfs_free_space,
4308 				       offset_index);
4309 			if (offset + bytes < tmp->offset)
4310 				break;
4311 			if (tmp->offset + tmp->bytes < offset) {
4312 				n = rb_next(&tmp->offset_index);
4313 				continue;
4314 			}
4315 			info = tmp;
4316 			goto have_info;
4317 		}
4318 
4319 		ret = 0;
4320 		goto out;
4321 	}
4322 
4323 	if (info->offset == offset) {
4324 		ret = 1;
4325 		goto out;
4326 	}
4327 
4328 	if (offset > info->offset && offset < info->offset + info->bytes)
4329 		ret = 1;
4330 out:
4331 	spin_unlock(&ctl->tree_lock);
4332 	return ret;
4333 }
4334 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */
4335