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 "free-space-cache.h"
16 #include "transaction.h"
17 #include "disk-io.h"
18 #include "extent_io.h"
19 #include "inode-map.h"
20 #include "volumes.h"
21 #include "space-info.h"
22 #include "delalloc-space.h"
23 #include "block-group.h"
24 
25 #define BITS_PER_BITMAP		(PAGE_SIZE * 8UL)
26 #define MAX_CACHE_BYTES_PER_GIG	SZ_32K
27 
28 struct btrfs_trim_range {
29 	u64 start;
30 	u64 bytes;
31 	struct list_head list;
32 };
33 
34 static int link_free_space(struct btrfs_free_space_ctl *ctl,
35 			   struct btrfs_free_space *info);
36 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
37 			      struct btrfs_free_space *info);
38 static int btrfs_wait_cache_io_root(struct btrfs_root *root,
39 			     struct btrfs_trans_handle *trans,
40 			     struct btrfs_io_ctl *io_ctl,
41 			     struct btrfs_path *path);
42 
__lookup_free_space_inode(struct btrfs_root * root,struct btrfs_path * path,u64 offset)43 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
44 					       struct btrfs_path *path,
45 					       u64 offset)
46 {
47 	struct btrfs_fs_info *fs_info = root->fs_info;
48 	struct btrfs_key key;
49 	struct btrfs_key location;
50 	struct btrfs_disk_key disk_key;
51 	struct btrfs_free_space_header *header;
52 	struct extent_buffer *leaf;
53 	struct inode *inode = NULL;
54 	unsigned nofs_flag;
55 	int ret;
56 
57 	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
58 	key.offset = offset;
59 	key.type = 0;
60 
61 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
62 	if (ret < 0)
63 		return ERR_PTR(ret);
64 	if (ret > 0) {
65 		btrfs_release_path(path);
66 		return ERR_PTR(-ENOENT);
67 	}
68 
69 	leaf = path->nodes[0];
70 	header = btrfs_item_ptr(leaf, path->slots[0],
71 				struct btrfs_free_space_header);
72 	btrfs_free_space_key(leaf, header, &disk_key);
73 	btrfs_disk_key_to_cpu(&location, &disk_key);
74 	btrfs_release_path(path);
75 
76 	/*
77 	 * We are often under a trans handle at this point, so we need to make
78 	 * sure NOFS is set to keep us from deadlocking.
79 	 */
80 	nofs_flag = memalloc_nofs_save();
81 	inode = btrfs_iget_path(fs_info->sb, &location, root, NULL, path);
82 	btrfs_release_path(path);
83 	memalloc_nofs_restore(nofs_flag);
84 	if (IS_ERR(inode))
85 		return inode;
86 
87 	mapping_set_gfp_mask(inode->i_mapping,
88 			mapping_gfp_constraint(inode->i_mapping,
89 			~(__GFP_FS | __GFP_HIGHMEM)));
90 
91 	return inode;
92 }
93 
lookup_free_space_inode(struct btrfs_block_group_cache * block_group,struct btrfs_path * path)94 struct inode *lookup_free_space_inode(
95 		struct btrfs_block_group_cache *block_group,
96 		struct btrfs_path *path)
97 {
98 	struct btrfs_fs_info *fs_info = block_group->fs_info;
99 	struct inode *inode = NULL;
100 	u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
101 
102 	spin_lock(&block_group->lock);
103 	if (block_group->inode)
104 		inode = igrab(block_group->inode);
105 	spin_unlock(&block_group->lock);
106 	if (inode)
107 		return inode;
108 
109 	inode = __lookup_free_space_inode(fs_info->tree_root, path,
110 					  block_group->key.objectid);
111 	if (IS_ERR(inode))
112 		return inode;
113 
114 	spin_lock(&block_group->lock);
115 	if (!((BTRFS_I(inode)->flags & flags) == flags)) {
116 		btrfs_info(fs_info, "Old style space inode found, converting.");
117 		BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
118 			BTRFS_INODE_NODATACOW;
119 		block_group->disk_cache_state = BTRFS_DC_CLEAR;
120 	}
121 
122 	if (!block_group->iref) {
123 		block_group->inode = igrab(inode);
124 		block_group->iref = 1;
125 	}
126 	spin_unlock(&block_group->lock);
127 
128 	return inode;
129 }
130 
__create_free_space_inode(struct btrfs_root * root,struct btrfs_trans_handle * trans,struct btrfs_path * path,u64 ino,u64 offset)131 static int __create_free_space_inode(struct btrfs_root *root,
132 				     struct btrfs_trans_handle *trans,
133 				     struct btrfs_path *path,
134 				     u64 ino, u64 offset)
135 {
136 	struct btrfs_key key;
137 	struct btrfs_disk_key disk_key;
138 	struct btrfs_free_space_header *header;
139 	struct btrfs_inode_item *inode_item;
140 	struct extent_buffer *leaf;
141 	u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
142 	int ret;
143 
144 	ret = btrfs_insert_empty_inode(trans, root, path, ino);
145 	if (ret)
146 		return ret;
147 
148 	/* We inline crc's for the free disk space cache */
149 	if (ino != BTRFS_FREE_INO_OBJECTID)
150 		flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
151 
152 	leaf = path->nodes[0];
153 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
154 				    struct btrfs_inode_item);
155 	btrfs_item_key(leaf, &disk_key, path->slots[0]);
156 	memzero_extent_buffer(leaf, (unsigned long)inode_item,
157 			     sizeof(*inode_item));
158 	btrfs_set_inode_generation(leaf, inode_item, trans->transid);
159 	btrfs_set_inode_size(leaf, inode_item, 0);
160 	btrfs_set_inode_nbytes(leaf, inode_item, 0);
161 	btrfs_set_inode_uid(leaf, inode_item, 0);
162 	btrfs_set_inode_gid(leaf, inode_item, 0);
163 	btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
164 	btrfs_set_inode_flags(leaf, inode_item, flags);
165 	btrfs_set_inode_nlink(leaf, inode_item, 1);
166 	btrfs_set_inode_transid(leaf, inode_item, trans->transid);
167 	btrfs_set_inode_block_group(leaf, inode_item, offset);
168 	btrfs_mark_buffer_dirty(leaf);
169 	btrfs_release_path(path);
170 
171 	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
172 	key.offset = offset;
173 	key.type = 0;
174 	ret = btrfs_insert_empty_item(trans, root, path, &key,
175 				      sizeof(struct btrfs_free_space_header));
176 	if (ret < 0) {
177 		btrfs_release_path(path);
178 		return ret;
179 	}
180 
181 	leaf = path->nodes[0];
182 	header = btrfs_item_ptr(leaf, path->slots[0],
183 				struct btrfs_free_space_header);
184 	memzero_extent_buffer(leaf, (unsigned long)header, sizeof(*header));
185 	btrfs_set_free_space_key(leaf, header, &disk_key);
186 	btrfs_mark_buffer_dirty(leaf);
187 	btrfs_release_path(path);
188 
189 	return 0;
190 }
191 
create_free_space_inode(struct btrfs_trans_handle * trans,struct btrfs_block_group_cache * block_group,struct btrfs_path * path)192 int create_free_space_inode(struct btrfs_trans_handle *trans,
193 			    struct btrfs_block_group_cache *block_group,
194 			    struct btrfs_path *path)
195 {
196 	int ret;
197 	u64 ino;
198 
199 	ret = btrfs_find_free_objectid(trans->fs_info->tree_root, &ino);
200 	if (ret < 0)
201 		return ret;
202 
203 	return __create_free_space_inode(trans->fs_info->tree_root, trans, path,
204 					 ino, block_group->key.objectid);
205 }
206 
btrfs_check_trunc_cache_free_space(struct btrfs_fs_info * fs_info,struct btrfs_block_rsv * rsv)207 int btrfs_check_trunc_cache_free_space(struct btrfs_fs_info *fs_info,
208 				       struct btrfs_block_rsv *rsv)
209 {
210 	u64 needed_bytes;
211 	int ret;
212 
213 	/* 1 for slack space, 1 for updating the inode */
214 	needed_bytes = btrfs_calc_insert_metadata_size(fs_info, 1) +
215 		btrfs_calc_metadata_size(fs_info, 1);
216 
217 	spin_lock(&rsv->lock);
218 	if (rsv->reserved < needed_bytes)
219 		ret = -ENOSPC;
220 	else
221 		ret = 0;
222 	spin_unlock(&rsv->lock);
223 	return ret;
224 }
225 
btrfs_truncate_free_space_cache(struct btrfs_trans_handle * trans,struct btrfs_block_group_cache * block_group,struct inode * inode)226 int btrfs_truncate_free_space_cache(struct btrfs_trans_handle *trans,
227 				    struct btrfs_block_group_cache *block_group,
228 				    struct inode *inode)
229 {
230 	struct btrfs_root *root = BTRFS_I(inode)->root;
231 	int ret = 0;
232 	bool locked = false;
233 
234 	if (block_group) {
235 		struct btrfs_path *path = btrfs_alloc_path();
236 
237 		if (!path) {
238 			ret = -ENOMEM;
239 			goto fail;
240 		}
241 		locked = true;
242 		mutex_lock(&trans->transaction->cache_write_mutex);
243 		if (!list_empty(&block_group->io_list)) {
244 			list_del_init(&block_group->io_list);
245 
246 			btrfs_wait_cache_io(trans, block_group, path);
247 			btrfs_put_block_group(block_group);
248 		}
249 
250 		/*
251 		 * now that we've truncated the cache away, its no longer
252 		 * setup or written
253 		 */
254 		spin_lock(&block_group->lock);
255 		block_group->disk_cache_state = BTRFS_DC_CLEAR;
256 		spin_unlock(&block_group->lock);
257 		btrfs_free_path(path);
258 	}
259 
260 	btrfs_i_size_write(BTRFS_I(inode), 0);
261 	truncate_pagecache(inode, 0);
262 
263 	/*
264 	 * We skip the throttling logic for free space cache inodes, so we don't
265 	 * need to check for -EAGAIN.
266 	 */
267 	ret = btrfs_truncate_inode_items(trans, root, inode,
268 					 0, BTRFS_EXTENT_DATA_KEY);
269 	if (ret)
270 		goto fail;
271 
272 	ret = btrfs_update_inode(trans, root, inode);
273 
274 fail:
275 	if (locked)
276 		mutex_unlock(&trans->transaction->cache_write_mutex);
277 	if (ret)
278 		btrfs_abort_transaction(trans, ret);
279 
280 	return ret;
281 }
282 
readahead_cache(struct inode * inode)283 static void readahead_cache(struct inode *inode)
284 {
285 	struct file_ra_state *ra;
286 	unsigned long last_index;
287 
288 	ra = kzalloc(sizeof(*ra), GFP_NOFS);
289 	if (!ra)
290 		return;
291 
292 	file_ra_state_init(ra, inode->i_mapping);
293 	last_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
294 
295 	page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
296 
297 	kfree(ra);
298 }
299 
io_ctl_init(struct btrfs_io_ctl * io_ctl,struct inode * inode,int write)300 static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode,
301 		       int write)
302 {
303 	int num_pages;
304 	int check_crcs = 0;
305 
306 	num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
307 
308 	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FREE_INO_OBJECTID)
309 		check_crcs = 1;
310 
311 	/* Make sure we can fit our crcs and generation into the first page */
312 	if (write && check_crcs &&
313 	    (num_pages * sizeof(u32) + sizeof(u64)) > PAGE_SIZE)
314 		return -ENOSPC;
315 
316 	memset(io_ctl, 0, sizeof(struct btrfs_io_ctl));
317 
318 	io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS);
319 	if (!io_ctl->pages)
320 		return -ENOMEM;
321 
322 	io_ctl->num_pages = num_pages;
323 	io_ctl->fs_info = btrfs_sb(inode->i_sb);
324 	io_ctl->check_crcs = check_crcs;
325 	io_ctl->inode = inode;
326 
327 	return 0;
328 }
329 ALLOW_ERROR_INJECTION(io_ctl_init, ERRNO);
330 
io_ctl_free(struct btrfs_io_ctl * io_ctl)331 static void io_ctl_free(struct btrfs_io_ctl *io_ctl)
332 {
333 	kfree(io_ctl->pages);
334 	io_ctl->pages = NULL;
335 }
336 
io_ctl_unmap_page(struct btrfs_io_ctl * io_ctl)337 static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl)
338 {
339 	if (io_ctl->cur) {
340 		io_ctl->cur = NULL;
341 		io_ctl->orig = NULL;
342 	}
343 }
344 
io_ctl_map_page(struct btrfs_io_ctl * io_ctl,int clear)345 static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear)
346 {
347 	ASSERT(io_ctl->index < io_ctl->num_pages);
348 	io_ctl->page = io_ctl->pages[io_ctl->index++];
349 	io_ctl->cur = page_address(io_ctl->page);
350 	io_ctl->orig = io_ctl->cur;
351 	io_ctl->size = PAGE_SIZE;
352 	if (clear)
353 		clear_page(io_ctl->cur);
354 }
355 
io_ctl_drop_pages(struct btrfs_io_ctl * io_ctl)356 static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl)
357 {
358 	int i;
359 
360 	io_ctl_unmap_page(io_ctl);
361 
362 	for (i = 0; i < io_ctl->num_pages; i++) {
363 		if (io_ctl->pages[i]) {
364 			ClearPageChecked(io_ctl->pages[i]);
365 			unlock_page(io_ctl->pages[i]);
366 			put_page(io_ctl->pages[i]);
367 		}
368 	}
369 }
370 
io_ctl_prepare_pages(struct btrfs_io_ctl * io_ctl,struct inode * inode,int uptodate)371 static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, struct inode *inode,
372 				int uptodate)
373 {
374 	struct page *page;
375 	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
376 	int i;
377 
378 	for (i = 0; i < io_ctl->num_pages; i++) {
379 		page = find_or_create_page(inode->i_mapping, i, mask);
380 		if (!page) {
381 			io_ctl_drop_pages(io_ctl);
382 			return -ENOMEM;
383 		}
384 		io_ctl->pages[i] = page;
385 		if (uptodate && !PageUptodate(page)) {
386 			btrfs_readpage(NULL, page);
387 			lock_page(page);
388 			if (!PageUptodate(page)) {
389 				btrfs_err(BTRFS_I(inode)->root->fs_info,
390 					   "error reading free space cache");
391 				io_ctl_drop_pages(io_ctl);
392 				return -EIO;
393 			}
394 		}
395 	}
396 
397 	for (i = 0; i < io_ctl->num_pages; i++) {
398 		clear_page_dirty_for_io(io_ctl->pages[i]);
399 		set_page_extent_mapped(io_ctl->pages[i]);
400 	}
401 
402 	return 0;
403 }
404 
io_ctl_set_generation(struct btrfs_io_ctl * io_ctl,u64 generation)405 static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
406 {
407 	__le64 *val;
408 
409 	io_ctl_map_page(io_ctl, 1);
410 
411 	/*
412 	 * Skip the csum areas.  If we don't check crcs then we just have a
413 	 * 64bit chunk at the front of the first page.
414 	 */
415 	if (io_ctl->check_crcs) {
416 		io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
417 		io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
418 	} else {
419 		io_ctl->cur += sizeof(u64);
420 		io_ctl->size -= sizeof(u64) * 2;
421 	}
422 
423 	val = io_ctl->cur;
424 	*val = cpu_to_le64(generation);
425 	io_ctl->cur += sizeof(u64);
426 }
427 
io_ctl_check_generation(struct btrfs_io_ctl * io_ctl,u64 generation)428 static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
429 {
430 	__le64 *gen;
431 
432 	/*
433 	 * Skip the crc area.  If we don't check crcs then we just have a 64bit
434 	 * chunk at the front of the first page.
435 	 */
436 	if (io_ctl->check_crcs) {
437 		io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
438 		io_ctl->size -= sizeof(u64) +
439 			(sizeof(u32) * io_ctl->num_pages);
440 	} else {
441 		io_ctl->cur += sizeof(u64);
442 		io_ctl->size -= sizeof(u64) * 2;
443 	}
444 
445 	gen = io_ctl->cur;
446 	if (le64_to_cpu(*gen) != generation) {
447 		btrfs_err_rl(io_ctl->fs_info,
448 			"space cache generation (%llu) does not match inode (%llu)",
449 				*gen, generation);
450 		io_ctl_unmap_page(io_ctl);
451 		return -EIO;
452 	}
453 	io_ctl->cur += sizeof(u64);
454 	return 0;
455 }
456 
io_ctl_set_crc(struct btrfs_io_ctl * io_ctl,int index)457 static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
458 {
459 	u32 *tmp;
460 	u32 crc = ~(u32)0;
461 	unsigned offset = 0;
462 
463 	if (!io_ctl->check_crcs) {
464 		io_ctl_unmap_page(io_ctl);
465 		return;
466 	}
467 
468 	if (index == 0)
469 		offset = sizeof(u32) * io_ctl->num_pages;
470 
471 	crc = btrfs_crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
472 	btrfs_crc32c_final(crc, (u8 *)&crc);
473 	io_ctl_unmap_page(io_ctl);
474 	tmp = page_address(io_ctl->pages[0]);
475 	tmp += index;
476 	*tmp = crc;
477 }
478 
io_ctl_check_crc(struct btrfs_io_ctl * io_ctl,int index)479 static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
480 {
481 	u32 *tmp, val;
482 	u32 crc = ~(u32)0;
483 	unsigned offset = 0;
484 
485 	if (!io_ctl->check_crcs) {
486 		io_ctl_map_page(io_ctl, 0);
487 		return 0;
488 	}
489 
490 	if (index == 0)
491 		offset = sizeof(u32) * io_ctl->num_pages;
492 
493 	tmp = page_address(io_ctl->pages[0]);
494 	tmp += index;
495 	val = *tmp;
496 
497 	io_ctl_map_page(io_ctl, 0);
498 	crc = btrfs_crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
499 	btrfs_crc32c_final(crc, (u8 *)&crc);
500 	if (val != crc) {
501 		btrfs_err_rl(io_ctl->fs_info,
502 			"csum mismatch on free space cache");
503 		io_ctl_unmap_page(io_ctl);
504 		return -EIO;
505 	}
506 
507 	return 0;
508 }
509 
io_ctl_add_entry(struct btrfs_io_ctl * io_ctl,u64 offset,u64 bytes,void * bitmap)510 static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
511 			    void *bitmap)
512 {
513 	struct btrfs_free_space_entry *entry;
514 
515 	if (!io_ctl->cur)
516 		return -ENOSPC;
517 
518 	entry = io_ctl->cur;
519 	entry->offset = cpu_to_le64(offset);
520 	entry->bytes = cpu_to_le64(bytes);
521 	entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
522 		BTRFS_FREE_SPACE_EXTENT;
523 	io_ctl->cur += sizeof(struct btrfs_free_space_entry);
524 	io_ctl->size -= sizeof(struct btrfs_free_space_entry);
525 
526 	if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
527 		return 0;
528 
529 	io_ctl_set_crc(io_ctl, io_ctl->index - 1);
530 
531 	/* No more pages to map */
532 	if (io_ctl->index >= io_ctl->num_pages)
533 		return 0;
534 
535 	/* map the next page */
536 	io_ctl_map_page(io_ctl, 1);
537 	return 0;
538 }
539 
io_ctl_add_bitmap(struct btrfs_io_ctl * io_ctl,void * bitmap)540 static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
541 {
542 	if (!io_ctl->cur)
543 		return -ENOSPC;
544 
545 	/*
546 	 * If we aren't at the start of the current page, unmap this one and
547 	 * map the next one if there is any left.
548 	 */
549 	if (io_ctl->cur != io_ctl->orig) {
550 		io_ctl_set_crc(io_ctl, io_ctl->index - 1);
551 		if (io_ctl->index >= io_ctl->num_pages)
552 			return -ENOSPC;
553 		io_ctl_map_page(io_ctl, 0);
554 	}
555 
556 	copy_page(io_ctl->cur, bitmap);
557 	io_ctl_set_crc(io_ctl, io_ctl->index - 1);
558 	if (io_ctl->index < io_ctl->num_pages)
559 		io_ctl_map_page(io_ctl, 0);
560 	return 0;
561 }
562 
io_ctl_zero_remaining_pages(struct btrfs_io_ctl * io_ctl)563 static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
564 {
565 	/*
566 	 * If we're not on the boundary we know we've modified the page and we
567 	 * need to crc the page.
568 	 */
569 	if (io_ctl->cur != io_ctl->orig)
570 		io_ctl_set_crc(io_ctl, io_ctl->index - 1);
571 	else
572 		io_ctl_unmap_page(io_ctl);
573 
574 	while (io_ctl->index < io_ctl->num_pages) {
575 		io_ctl_map_page(io_ctl, 1);
576 		io_ctl_set_crc(io_ctl, io_ctl->index - 1);
577 	}
578 }
579 
io_ctl_read_entry(struct btrfs_io_ctl * io_ctl,struct btrfs_free_space * entry,u8 * type)580 static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
581 			    struct btrfs_free_space *entry, u8 *type)
582 {
583 	struct btrfs_free_space_entry *e;
584 	int ret;
585 
586 	if (!io_ctl->cur) {
587 		ret = io_ctl_check_crc(io_ctl, io_ctl->index);
588 		if (ret)
589 			return ret;
590 	}
591 
592 	e = io_ctl->cur;
593 	entry->offset = le64_to_cpu(e->offset);
594 	entry->bytes = le64_to_cpu(e->bytes);
595 	*type = e->type;
596 	io_ctl->cur += sizeof(struct btrfs_free_space_entry);
597 	io_ctl->size -= sizeof(struct btrfs_free_space_entry);
598 
599 	if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
600 		return 0;
601 
602 	io_ctl_unmap_page(io_ctl);
603 
604 	return 0;
605 }
606 
io_ctl_read_bitmap(struct btrfs_io_ctl * io_ctl,struct btrfs_free_space * entry)607 static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
608 			      struct btrfs_free_space *entry)
609 {
610 	int ret;
611 
612 	ret = io_ctl_check_crc(io_ctl, io_ctl->index);
613 	if (ret)
614 		return ret;
615 
616 	copy_page(entry->bitmap, io_ctl->cur);
617 	io_ctl_unmap_page(io_ctl);
618 
619 	return 0;
620 }
621 
622 /*
623  * Since we attach pinned extents after the fact we can have contiguous sections
624  * of free space that are split up in entries.  This poses a problem with the
625  * tree logging stuff since it could have allocated across what appears to be 2
626  * entries since we would have merged the entries when adding the pinned extents
627  * back to the free space cache.  So run through the space cache that we just
628  * loaded and merge contiguous entries.  This will make the log replay stuff not
629  * blow up and it will make for nicer allocator behavior.
630  */
merge_space_tree(struct btrfs_free_space_ctl * ctl)631 static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
632 {
633 	struct btrfs_free_space *e, *prev = NULL;
634 	struct rb_node *n;
635 
636 again:
637 	spin_lock(&ctl->tree_lock);
638 	for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
639 		e = rb_entry(n, struct btrfs_free_space, offset_index);
640 		if (!prev)
641 			goto next;
642 		if (e->bitmap || prev->bitmap)
643 			goto next;
644 		if (prev->offset + prev->bytes == e->offset) {
645 			unlink_free_space(ctl, prev);
646 			unlink_free_space(ctl, e);
647 			prev->bytes += e->bytes;
648 			kmem_cache_free(btrfs_free_space_cachep, e);
649 			link_free_space(ctl, prev);
650 			prev = NULL;
651 			spin_unlock(&ctl->tree_lock);
652 			goto again;
653 		}
654 next:
655 		prev = e;
656 	}
657 	spin_unlock(&ctl->tree_lock);
658 }
659 
__load_free_space_cache(struct btrfs_root * root,struct inode * inode,struct btrfs_free_space_ctl * ctl,struct btrfs_path * path,u64 offset)660 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
661 				   struct btrfs_free_space_ctl *ctl,
662 				   struct btrfs_path *path, u64 offset)
663 {
664 	struct btrfs_fs_info *fs_info = root->fs_info;
665 	struct btrfs_free_space_header *header;
666 	struct extent_buffer *leaf;
667 	struct btrfs_io_ctl io_ctl;
668 	struct btrfs_key key;
669 	struct btrfs_free_space *e, *n;
670 	LIST_HEAD(bitmaps);
671 	u64 num_entries;
672 	u64 num_bitmaps;
673 	u64 generation;
674 	u8 type;
675 	int ret = 0;
676 
677 	/* Nothing in the space cache, goodbye */
678 	if (!i_size_read(inode))
679 		return 0;
680 
681 	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
682 	key.offset = offset;
683 	key.type = 0;
684 
685 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
686 	if (ret < 0)
687 		return 0;
688 	else if (ret > 0) {
689 		btrfs_release_path(path);
690 		return 0;
691 	}
692 
693 	ret = -1;
694 
695 	leaf = path->nodes[0];
696 	header = btrfs_item_ptr(leaf, path->slots[0],
697 				struct btrfs_free_space_header);
698 	num_entries = btrfs_free_space_entries(leaf, header);
699 	num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
700 	generation = btrfs_free_space_generation(leaf, header);
701 	btrfs_release_path(path);
702 
703 	if (!BTRFS_I(inode)->generation) {
704 		btrfs_info(fs_info,
705 			   "the free space cache file (%llu) is invalid, skip it",
706 			   offset);
707 		return 0;
708 	}
709 
710 	if (BTRFS_I(inode)->generation != generation) {
711 		btrfs_err(fs_info,
712 			  "free space inode generation (%llu) did not match free space cache generation (%llu)",
713 			  BTRFS_I(inode)->generation, generation);
714 		return 0;
715 	}
716 
717 	if (!num_entries)
718 		return 0;
719 
720 	ret = io_ctl_init(&io_ctl, inode, 0);
721 	if (ret)
722 		return ret;
723 
724 	readahead_cache(inode);
725 
726 	ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
727 	if (ret)
728 		goto out;
729 
730 	ret = io_ctl_check_crc(&io_ctl, 0);
731 	if (ret)
732 		goto free_cache;
733 
734 	ret = io_ctl_check_generation(&io_ctl, generation);
735 	if (ret)
736 		goto free_cache;
737 
738 	while (num_entries) {
739 		e = kmem_cache_zalloc(btrfs_free_space_cachep,
740 				      GFP_NOFS);
741 		if (!e)
742 			goto free_cache;
743 
744 		ret = io_ctl_read_entry(&io_ctl, e, &type);
745 		if (ret) {
746 			kmem_cache_free(btrfs_free_space_cachep, e);
747 			goto free_cache;
748 		}
749 
750 		if (!e->bytes) {
751 			kmem_cache_free(btrfs_free_space_cachep, e);
752 			goto free_cache;
753 		}
754 
755 		if (type == BTRFS_FREE_SPACE_EXTENT) {
756 			spin_lock(&ctl->tree_lock);
757 			ret = link_free_space(ctl, e);
758 			spin_unlock(&ctl->tree_lock);
759 			if (ret) {
760 				btrfs_err(fs_info,
761 					"Duplicate entries in free space cache, dumping");
762 				kmem_cache_free(btrfs_free_space_cachep, e);
763 				goto free_cache;
764 			}
765 		} else {
766 			ASSERT(num_bitmaps);
767 			num_bitmaps--;
768 			e->bitmap = kmem_cache_zalloc(
769 					btrfs_free_space_bitmap_cachep, GFP_NOFS);
770 			if (!e->bitmap) {
771 				kmem_cache_free(
772 					btrfs_free_space_cachep, e);
773 				goto free_cache;
774 			}
775 			spin_lock(&ctl->tree_lock);
776 			ret = link_free_space(ctl, e);
777 			ctl->total_bitmaps++;
778 			ctl->op->recalc_thresholds(ctl);
779 			spin_unlock(&ctl->tree_lock);
780 			if (ret) {
781 				btrfs_err(fs_info,
782 					"Duplicate entries in free space cache, dumping");
783 				kmem_cache_free(btrfs_free_space_cachep, e);
784 				goto free_cache;
785 			}
786 			list_add_tail(&e->list, &bitmaps);
787 		}
788 
789 		num_entries--;
790 	}
791 
792 	io_ctl_unmap_page(&io_ctl);
793 
794 	/*
795 	 * We add the bitmaps at the end of the entries in order that
796 	 * the bitmap entries are added to the cache.
797 	 */
798 	list_for_each_entry_safe(e, n, &bitmaps, list) {
799 		list_del_init(&e->list);
800 		ret = io_ctl_read_bitmap(&io_ctl, e);
801 		if (ret)
802 			goto free_cache;
803 	}
804 
805 	io_ctl_drop_pages(&io_ctl);
806 	merge_space_tree(ctl);
807 	ret = 1;
808 out:
809 	io_ctl_free(&io_ctl);
810 	return ret;
811 free_cache:
812 	io_ctl_drop_pages(&io_ctl);
813 	__btrfs_remove_free_space_cache(ctl);
814 	goto out;
815 }
816 
load_free_space_cache(struct btrfs_block_group_cache * block_group)817 int load_free_space_cache(struct btrfs_block_group_cache *block_group)
818 {
819 	struct btrfs_fs_info *fs_info = block_group->fs_info;
820 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
821 	struct inode *inode;
822 	struct btrfs_path *path;
823 	int ret = 0;
824 	bool matched;
825 	u64 used = btrfs_block_group_used(&block_group->item);
826 
827 	/*
828 	 * If this block group has been marked to be cleared for one reason or
829 	 * another then we can't trust the on disk cache, so just return.
830 	 */
831 	spin_lock(&block_group->lock);
832 	if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
833 		spin_unlock(&block_group->lock);
834 		return 0;
835 	}
836 	spin_unlock(&block_group->lock);
837 
838 	path = btrfs_alloc_path();
839 	if (!path)
840 		return 0;
841 	path->search_commit_root = 1;
842 	path->skip_locking = 1;
843 
844 	/*
845 	 * We must pass a path with search_commit_root set to btrfs_iget in
846 	 * order to avoid a deadlock when allocating extents for the tree root.
847 	 *
848 	 * When we are COWing an extent buffer from the tree root, when looking
849 	 * for a free extent, at extent-tree.c:find_free_extent(), we can find
850 	 * block group without its free space cache loaded. When we find one
851 	 * we must load its space cache which requires reading its free space
852 	 * cache's inode item from the root tree. If this inode item is located
853 	 * in the same leaf that we started COWing before, then we end up in
854 	 * deadlock on the extent buffer (trying to read lock it when we
855 	 * previously write locked it).
856 	 *
857 	 * It's safe to read the inode item using the commit root because
858 	 * block groups, once loaded, stay in memory forever (until they are
859 	 * removed) as well as their space caches once loaded. New block groups
860 	 * once created get their ->cached field set to BTRFS_CACHE_FINISHED so
861 	 * we will never try to read their inode item while the fs is mounted.
862 	 */
863 	inode = lookup_free_space_inode(block_group, path);
864 	if (IS_ERR(inode)) {
865 		btrfs_free_path(path);
866 		return 0;
867 	}
868 
869 	/* We may have converted the inode and made the cache invalid. */
870 	spin_lock(&block_group->lock);
871 	if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
872 		spin_unlock(&block_group->lock);
873 		btrfs_free_path(path);
874 		goto out;
875 	}
876 	spin_unlock(&block_group->lock);
877 
878 	ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
879 				      path, block_group->key.objectid);
880 	btrfs_free_path(path);
881 	if (ret <= 0)
882 		goto out;
883 
884 	spin_lock(&ctl->tree_lock);
885 	matched = (ctl->free_space == (block_group->key.offset - used -
886 				       block_group->bytes_super));
887 	spin_unlock(&ctl->tree_lock);
888 
889 	if (!matched) {
890 		__btrfs_remove_free_space_cache(ctl);
891 		btrfs_warn(fs_info,
892 			   "block group %llu has wrong amount of free space",
893 			   block_group->key.objectid);
894 		ret = -1;
895 	}
896 out:
897 	if (ret < 0) {
898 		/* This cache is bogus, make sure it gets cleared */
899 		spin_lock(&block_group->lock);
900 		block_group->disk_cache_state = BTRFS_DC_CLEAR;
901 		spin_unlock(&block_group->lock);
902 		ret = 0;
903 
904 		btrfs_warn(fs_info,
905 			   "failed to load free space cache for block group %llu, rebuilding it now",
906 			   block_group->key.objectid);
907 	}
908 
909 	iput(inode);
910 	return ret;
911 }
912 
913 static noinline_for_stack
write_cache_extent_entries(struct btrfs_io_ctl * io_ctl,struct btrfs_free_space_ctl * ctl,struct btrfs_block_group_cache * block_group,int * entries,int * bitmaps,struct list_head * bitmap_list)914 int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
915 			      struct btrfs_free_space_ctl *ctl,
916 			      struct btrfs_block_group_cache *block_group,
917 			      int *entries, int *bitmaps,
918 			      struct list_head *bitmap_list)
919 {
920 	int ret;
921 	struct btrfs_free_cluster *cluster = NULL;
922 	struct btrfs_free_cluster *cluster_locked = NULL;
923 	struct rb_node *node = rb_first(&ctl->free_space_offset);
924 	struct btrfs_trim_range *trim_entry;
925 
926 	/* Get the cluster for this block_group if it exists */
927 	if (block_group && !list_empty(&block_group->cluster_list)) {
928 		cluster = list_entry(block_group->cluster_list.next,
929 				     struct btrfs_free_cluster,
930 				     block_group_list);
931 	}
932 
933 	if (!node && cluster) {
934 		cluster_locked = cluster;
935 		spin_lock(&cluster_locked->lock);
936 		node = rb_first(&cluster->root);
937 		cluster = NULL;
938 	}
939 
940 	/* Write out the extent entries */
941 	while (node) {
942 		struct btrfs_free_space *e;
943 
944 		e = rb_entry(node, struct btrfs_free_space, offset_index);
945 		*entries += 1;
946 
947 		ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
948 				       e->bitmap);
949 		if (ret)
950 			goto fail;
951 
952 		if (e->bitmap) {
953 			list_add_tail(&e->list, bitmap_list);
954 			*bitmaps += 1;
955 		}
956 		node = rb_next(node);
957 		if (!node && cluster) {
958 			node = rb_first(&cluster->root);
959 			cluster_locked = cluster;
960 			spin_lock(&cluster_locked->lock);
961 			cluster = NULL;
962 		}
963 	}
964 	if (cluster_locked) {
965 		spin_unlock(&cluster_locked->lock);
966 		cluster_locked = NULL;
967 	}
968 
969 	/*
970 	 * Make sure we don't miss any range that was removed from our rbtree
971 	 * because trimming is running. Otherwise after a umount+mount (or crash
972 	 * after committing the transaction) we would leak free space and get
973 	 * an inconsistent free space cache report from fsck.
974 	 */
975 	list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
976 		ret = io_ctl_add_entry(io_ctl, trim_entry->start,
977 				       trim_entry->bytes, NULL);
978 		if (ret)
979 			goto fail;
980 		*entries += 1;
981 	}
982 
983 	return 0;
984 fail:
985 	if (cluster_locked)
986 		spin_unlock(&cluster_locked->lock);
987 	return -ENOSPC;
988 }
989 
990 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)991 update_cache_item(struct btrfs_trans_handle *trans,
992 		  struct btrfs_root *root,
993 		  struct inode *inode,
994 		  struct btrfs_path *path, u64 offset,
995 		  int entries, int bitmaps)
996 {
997 	struct btrfs_key key;
998 	struct btrfs_free_space_header *header;
999 	struct extent_buffer *leaf;
1000 	int ret;
1001 
1002 	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1003 	key.offset = offset;
1004 	key.type = 0;
1005 
1006 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1007 	if (ret < 0) {
1008 		clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1009 				 EXTENT_DELALLOC, 0, 0, NULL);
1010 		goto fail;
1011 	}
1012 	leaf = path->nodes[0];
1013 	if (ret > 0) {
1014 		struct btrfs_key found_key;
1015 		ASSERT(path->slots[0]);
1016 		path->slots[0]--;
1017 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1018 		if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1019 		    found_key.offset != offset) {
1020 			clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1021 					 inode->i_size - 1, EXTENT_DELALLOC, 0,
1022 					 0, NULL);
1023 			btrfs_release_path(path);
1024 			goto fail;
1025 		}
1026 	}
1027 
1028 	BTRFS_I(inode)->generation = trans->transid;
1029 	header = btrfs_item_ptr(leaf, path->slots[0],
1030 				struct btrfs_free_space_header);
1031 	btrfs_set_free_space_entries(leaf, header, entries);
1032 	btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1033 	btrfs_set_free_space_generation(leaf, header, trans->transid);
1034 	btrfs_mark_buffer_dirty(leaf);
1035 	btrfs_release_path(path);
1036 
1037 	return 0;
1038 
1039 fail:
1040 	return -1;
1041 }
1042 
write_pinned_extent_entries(struct btrfs_block_group_cache * block_group,struct btrfs_io_ctl * io_ctl,int * entries)1043 static noinline_for_stack int write_pinned_extent_entries(
1044 			    struct btrfs_block_group_cache *block_group,
1045 			    struct btrfs_io_ctl *io_ctl,
1046 			    int *entries)
1047 {
1048 	u64 start, extent_start, extent_end, len;
1049 	struct extent_io_tree *unpin = NULL;
1050 	int ret;
1051 
1052 	if (!block_group)
1053 		return 0;
1054 
1055 	/*
1056 	 * We want to add any pinned extents to our free space cache
1057 	 * so we don't leak the space
1058 	 *
1059 	 * We shouldn't have switched the pinned extents yet so this is the
1060 	 * right one
1061 	 */
1062 	unpin = block_group->fs_info->pinned_extents;
1063 
1064 	start = block_group->key.objectid;
1065 
1066 	while (start < block_group->key.objectid + block_group->key.offset) {
1067 		ret = find_first_extent_bit(unpin, start,
1068 					    &extent_start, &extent_end,
1069 					    EXTENT_DIRTY, NULL);
1070 		if (ret)
1071 			return 0;
1072 
1073 		/* This pinned extent is out of our range */
1074 		if (extent_start >= block_group->key.objectid +
1075 		    block_group->key.offset)
1076 			return 0;
1077 
1078 		extent_start = max(extent_start, start);
1079 		extent_end = min(block_group->key.objectid +
1080 				 block_group->key.offset, extent_end + 1);
1081 		len = extent_end - extent_start;
1082 
1083 		*entries += 1;
1084 		ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1085 		if (ret)
1086 			return -ENOSPC;
1087 
1088 		start = extent_end;
1089 	}
1090 
1091 	return 0;
1092 }
1093 
1094 static noinline_for_stack int
write_bitmap_entries(struct btrfs_io_ctl * io_ctl,struct list_head * bitmap_list)1095 write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1096 {
1097 	struct btrfs_free_space *entry, *next;
1098 	int ret;
1099 
1100 	/* Write out the bitmaps */
1101 	list_for_each_entry_safe(entry, next, bitmap_list, list) {
1102 		ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1103 		if (ret)
1104 			return -ENOSPC;
1105 		list_del_init(&entry->list);
1106 	}
1107 
1108 	return 0;
1109 }
1110 
flush_dirty_cache(struct inode * inode)1111 static int flush_dirty_cache(struct inode *inode)
1112 {
1113 	int ret;
1114 
1115 	ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1116 	if (ret)
1117 		clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1118 				 EXTENT_DELALLOC, 0, 0, NULL);
1119 
1120 	return ret;
1121 }
1122 
1123 static void noinline_for_stack
cleanup_bitmap_list(struct list_head * bitmap_list)1124 cleanup_bitmap_list(struct list_head *bitmap_list)
1125 {
1126 	struct btrfs_free_space *entry, *next;
1127 
1128 	list_for_each_entry_safe(entry, next, bitmap_list, list)
1129 		list_del_init(&entry->list);
1130 }
1131 
1132 static void noinline_for_stack
cleanup_write_cache_enospc(struct inode * inode,struct btrfs_io_ctl * io_ctl,struct extent_state ** cached_state)1133 cleanup_write_cache_enospc(struct inode *inode,
1134 			   struct btrfs_io_ctl *io_ctl,
1135 			   struct extent_state **cached_state)
1136 {
1137 	io_ctl_drop_pages(io_ctl);
1138 	unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1139 			     i_size_read(inode) - 1, cached_state);
1140 }
1141 
__btrfs_wait_cache_io(struct btrfs_root * root,struct btrfs_trans_handle * trans,struct btrfs_block_group_cache * block_group,struct btrfs_io_ctl * io_ctl,struct btrfs_path * path,u64 offset)1142 static int __btrfs_wait_cache_io(struct btrfs_root *root,
1143 				 struct btrfs_trans_handle *trans,
1144 				 struct btrfs_block_group_cache *block_group,
1145 				 struct btrfs_io_ctl *io_ctl,
1146 				 struct btrfs_path *path, u64 offset)
1147 {
1148 	int ret;
1149 	struct inode *inode = io_ctl->inode;
1150 
1151 	if (!inode)
1152 		return 0;
1153 
1154 	/* Flush the dirty pages in the cache file. */
1155 	ret = flush_dirty_cache(inode);
1156 	if (ret)
1157 		goto out;
1158 
1159 	/* Update the cache item to tell everyone this cache file is valid. */
1160 	ret = update_cache_item(trans, root, inode, path, offset,
1161 				io_ctl->entries, io_ctl->bitmaps);
1162 out:
1163 	io_ctl_free(io_ctl);
1164 	if (ret) {
1165 		invalidate_inode_pages2(inode->i_mapping);
1166 		BTRFS_I(inode)->generation = 0;
1167 		if (block_group) {
1168 #ifdef DEBUG
1169 			btrfs_err(root->fs_info,
1170 				  "failed to write free space cache for block group %llu",
1171 				  block_group->key.objectid);
1172 #endif
1173 		}
1174 	}
1175 	btrfs_update_inode(trans, root, inode);
1176 
1177 	if (block_group) {
1178 		/* the dirty list is protected by the dirty_bgs_lock */
1179 		spin_lock(&trans->transaction->dirty_bgs_lock);
1180 
1181 		/* the disk_cache_state is protected by the block group lock */
1182 		spin_lock(&block_group->lock);
1183 
1184 		/*
1185 		 * only mark this as written if we didn't get put back on
1186 		 * the dirty list while waiting for IO.   Otherwise our
1187 		 * cache state won't be right, and we won't get written again
1188 		 */
1189 		if (!ret && list_empty(&block_group->dirty_list))
1190 			block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1191 		else if (ret)
1192 			block_group->disk_cache_state = BTRFS_DC_ERROR;
1193 
1194 		spin_unlock(&block_group->lock);
1195 		spin_unlock(&trans->transaction->dirty_bgs_lock);
1196 		io_ctl->inode = NULL;
1197 		iput(inode);
1198 	}
1199 
1200 	return ret;
1201 
1202 }
1203 
btrfs_wait_cache_io_root(struct btrfs_root * root,struct btrfs_trans_handle * trans,struct btrfs_io_ctl * io_ctl,struct btrfs_path * path)1204 static int btrfs_wait_cache_io_root(struct btrfs_root *root,
1205 				    struct btrfs_trans_handle *trans,
1206 				    struct btrfs_io_ctl *io_ctl,
1207 				    struct btrfs_path *path)
1208 {
1209 	return __btrfs_wait_cache_io(root, trans, NULL, io_ctl, path, 0);
1210 }
1211 
btrfs_wait_cache_io(struct btrfs_trans_handle * trans,struct btrfs_block_group_cache * block_group,struct btrfs_path * path)1212 int btrfs_wait_cache_io(struct btrfs_trans_handle *trans,
1213 			struct btrfs_block_group_cache *block_group,
1214 			struct btrfs_path *path)
1215 {
1216 	return __btrfs_wait_cache_io(block_group->fs_info->tree_root, trans,
1217 				     block_group, &block_group->io_ctl,
1218 				     path, block_group->key.objectid);
1219 }
1220 
1221 /**
1222  * __btrfs_write_out_cache - write out cached info to an inode
1223  * @root - the root the inode belongs to
1224  * @ctl - the free space cache we are going to write out
1225  * @block_group - the block_group for this cache if it belongs to a block_group
1226  * @trans - the trans handle
1227  *
1228  * This function writes out a free space cache struct to disk for quick recovery
1229  * on mount.  This will return 0 if it was successful in writing the cache out,
1230  * or an errno if it was not.
1231  */
__btrfs_write_out_cache(struct btrfs_root * root,struct inode * inode,struct btrfs_free_space_ctl * ctl,struct btrfs_block_group_cache * block_group,struct btrfs_io_ctl * io_ctl,struct btrfs_trans_handle * trans)1232 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
1233 				   struct btrfs_free_space_ctl *ctl,
1234 				   struct btrfs_block_group_cache *block_group,
1235 				   struct btrfs_io_ctl *io_ctl,
1236 				   struct btrfs_trans_handle *trans)
1237 {
1238 	struct extent_state *cached_state = NULL;
1239 	LIST_HEAD(bitmap_list);
1240 	int entries = 0;
1241 	int bitmaps = 0;
1242 	int ret;
1243 	int must_iput = 0;
1244 
1245 	if (!i_size_read(inode))
1246 		return -EIO;
1247 
1248 	WARN_ON(io_ctl->pages);
1249 	ret = io_ctl_init(io_ctl, inode, 1);
1250 	if (ret)
1251 		return ret;
1252 
1253 	if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1254 		down_write(&block_group->data_rwsem);
1255 		spin_lock(&block_group->lock);
1256 		if (block_group->delalloc_bytes) {
1257 			block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1258 			spin_unlock(&block_group->lock);
1259 			up_write(&block_group->data_rwsem);
1260 			BTRFS_I(inode)->generation = 0;
1261 			ret = 0;
1262 			must_iput = 1;
1263 			goto out;
1264 		}
1265 		spin_unlock(&block_group->lock);
1266 	}
1267 
1268 	/* Lock all pages first so we can lock the extent safely. */
1269 	ret = io_ctl_prepare_pages(io_ctl, inode, 0);
1270 	if (ret)
1271 		goto out_unlock;
1272 
1273 	lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1274 			 &cached_state);
1275 
1276 	io_ctl_set_generation(io_ctl, trans->transid);
1277 
1278 	mutex_lock(&ctl->cache_writeout_mutex);
1279 	/* Write out the extent entries in the free space cache */
1280 	spin_lock(&ctl->tree_lock);
1281 	ret = write_cache_extent_entries(io_ctl, ctl,
1282 					 block_group, &entries, &bitmaps,
1283 					 &bitmap_list);
1284 	if (ret)
1285 		goto out_nospc_locked;
1286 
1287 	/*
1288 	 * Some spaces that are freed in the current transaction are pinned,
1289 	 * they will be added into free space cache after the transaction is
1290 	 * committed, we shouldn't lose them.
1291 	 *
1292 	 * If this changes while we are working we'll get added back to
1293 	 * the dirty list and redo it.  No locking needed
1294 	 */
1295 	ret = write_pinned_extent_entries(block_group, io_ctl, &entries);
1296 	if (ret)
1297 		goto out_nospc_locked;
1298 
1299 	/*
1300 	 * At last, we write out all the bitmaps and keep cache_writeout_mutex
1301 	 * locked while doing it because a concurrent trim can be manipulating
1302 	 * or freeing the bitmap.
1303 	 */
1304 	ret = write_bitmap_entries(io_ctl, &bitmap_list);
1305 	spin_unlock(&ctl->tree_lock);
1306 	mutex_unlock(&ctl->cache_writeout_mutex);
1307 	if (ret)
1308 		goto out_nospc;
1309 
1310 	/* Zero out the rest of the pages just to make sure */
1311 	io_ctl_zero_remaining_pages(io_ctl);
1312 
1313 	/* Everything is written out, now we dirty the pages in the file. */
1314 	ret = btrfs_dirty_pages(inode, io_ctl->pages, io_ctl->num_pages, 0,
1315 				i_size_read(inode), &cached_state);
1316 	if (ret)
1317 		goto out_nospc;
1318 
1319 	if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1320 		up_write(&block_group->data_rwsem);
1321 	/*
1322 	 * Release the pages and unlock the extent, we will flush
1323 	 * them out later
1324 	 */
1325 	io_ctl_drop_pages(io_ctl);
1326 
1327 	unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1328 			     i_size_read(inode) - 1, &cached_state);
1329 
1330 	/*
1331 	 * at this point the pages are under IO and we're happy,
1332 	 * The caller is responsible for waiting on them and updating the
1333 	 * the cache and the inode
1334 	 */
1335 	io_ctl->entries = entries;
1336 	io_ctl->bitmaps = bitmaps;
1337 
1338 	ret = btrfs_fdatawrite_range(inode, 0, (u64)-1);
1339 	if (ret)
1340 		goto out;
1341 
1342 	return 0;
1343 
1344 out:
1345 	io_ctl->inode = NULL;
1346 	io_ctl_free(io_ctl);
1347 	if (ret) {
1348 		invalidate_inode_pages2(inode->i_mapping);
1349 		BTRFS_I(inode)->generation = 0;
1350 	}
1351 	btrfs_update_inode(trans, root, inode);
1352 	if (must_iput)
1353 		iput(inode);
1354 	return ret;
1355 
1356 out_nospc_locked:
1357 	cleanup_bitmap_list(&bitmap_list);
1358 	spin_unlock(&ctl->tree_lock);
1359 	mutex_unlock(&ctl->cache_writeout_mutex);
1360 
1361 out_nospc:
1362 	cleanup_write_cache_enospc(inode, io_ctl, &cached_state);
1363 
1364 out_unlock:
1365 	if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1366 		up_write(&block_group->data_rwsem);
1367 
1368 	goto out;
1369 }
1370 
btrfs_write_out_cache(struct btrfs_trans_handle * trans,struct btrfs_block_group_cache * block_group,struct btrfs_path * path)1371 int btrfs_write_out_cache(struct btrfs_trans_handle *trans,
1372 			  struct btrfs_block_group_cache *block_group,
1373 			  struct btrfs_path *path)
1374 {
1375 	struct btrfs_fs_info *fs_info = trans->fs_info;
1376 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1377 	struct inode *inode;
1378 	int ret = 0;
1379 
1380 	spin_lock(&block_group->lock);
1381 	if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1382 		spin_unlock(&block_group->lock);
1383 		return 0;
1384 	}
1385 	spin_unlock(&block_group->lock);
1386 
1387 	inode = lookup_free_space_inode(block_group, path);
1388 	if (IS_ERR(inode))
1389 		return 0;
1390 
1391 	ret = __btrfs_write_out_cache(fs_info->tree_root, inode, ctl,
1392 				block_group, &block_group->io_ctl, trans);
1393 	if (ret) {
1394 #ifdef DEBUG
1395 		btrfs_err(fs_info,
1396 			  "failed to write free space cache for block group %llu",
1397 			  block_group->key.objectid);
1398 #endif
1399 		spin_lock(&block_group->lock);
1400 		block_group->disk_cache_state = BTRFS_DC_ERROR;
1401 		spin_unlock(&block_group->lock);
1402 
1403 		block_group->io_ctl.inode = NULL;
1404 		iput(inode);
1405 	}
1406 
1407 	/*
1408 	 * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1409 	 * to wait for IO and put the inode
1410 	 */
1411 
1412 	return ret;
1413 }
1414 
offset_to_bit(u64 bitmap_start,u32 unit,u64 offset)1415 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1416 					  u64 offset)
1417 {
1418 	ASSERT(offset >= bitmap_start);
1419 	offset -= bitmap_start;
1420 	return (unsigned long)(div_u64(offset, unit));
1421 }
1422 
bytes_to_bits(u64 bytes,u32 unit)1423 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1424 {
1425 	return (unsigned long)(div_u64(bytes, unit));
1426 }
1427 
offset_to_bitmap(struct btrfs_free_space_ctl * ctl,u64 offset)1428 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1429 				   u64 offset)
1430 {
1431 	u64 bitmap_start;
1432 	u64 bytes_per_bitmap;
1433 
1434 	bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1435 	bitmap_start = offset - ctl->start;
1436 	bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1437 	bitmap_start *= bytes_per_bitmap;
1438 	bitmap_start += ctl->start;
1439 
1440 	return bitmap_start;
1441 }
1442 
tree_insert_offset(struct rb_root * root,u64 offset,struct rb_node * node,int bitmap)1443 static int tree_insert_offset(struct rb_root *root, u64 offset,
1444 			      struct rb_node *node, int bitmap)
1445 {
1446 	struct rb_node **p = &root->rb_node;
1447 	struct rb_node *parent = NULL;
1448 	struct btrfs_free_space *info;
1449 
1450 	while (*p) {
1451 		parent = *p;
1452 		info = rb_entry(parent, struct btrfs_free_space, offset_index);
1453 
1454 		if (offset < info->offset) {
1455 			p = &(*p)->rb_left;
1456 		} else if (offset > info->offset) {
1457 			p = &(*p)->rb_right;
1458 		} else {
1459 			/*
1460 			 * we could have a bitmap entry and an extent entry
1461 			 * share the same offset.  If this is the case, we want
1462 			 * the extent entry to always be found first if we do a
1463 			 * linear search through the tree, since we want to have
1464 			 * the quickest allocation time, and allocating from an
1465 			 * extent is faster than allocating from a bitmap.  So
1466 			 * if we're inserting a bitmap and we find an entry at
1467 			 * this offset, we want to go right, or after this entry
1468 			 * logically.  If we are inserting an extent and we've
1469 			 * found a bitmap, we want to go left, or before
1470 			 * logically.
1471 			 */
1472 			if (bitmap) {
1473 				if (info->bitmap) {
1474 					WARN_ON_ONCE(1);
1475 					return -EEXIST;
1476 				}
1477 				p = &(*p)->rb_right;
1478 			} else {
1479 				if (!info->bitmap) {
1480 					WARN_ON_ONCE(1);
1481 					return -EEXIST;
1482 				}
1483 				p = &(*p)->rb_left;
1484 			}
1485 		}
1486 	}
1487 
1488 	rb_link_node(node, parent, p);
1489 	rb_insert_color(node, root);
1490 
1491 	return 0;
1492 }
1493 
1494 /*
1495  * searches the tree for the given offset.
1496  *
1497  * fuzzy - If this is set, then we are trying to make an allocation, and we just
1498  * want a section that has at least bytes size and comes at or after the given
1499  * offset.
1500  */
1501 static struct btrfs_free_space *
tree_search_offset(struct btrfs_free_space_ctl * ctl,u64 offset,int bitmap_only,int fuzzy)1502 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1503 		   u64 offset, int bitmap_only, int fuzzy)
1504 {
1505 	struct rb_node *n = ctl->free_space_offset.rb_node;
1506 	struct btrfs_free_space *entry, *prev = NULL;
1507 
1508 	/* find entry that is closest to the 'offset' */
1509 	while (1) {
1510 		if (!n) {
1511 			entry = NULL;
1512 			break;
1513 		}
1514 
1515 		entry = rb_entry(n, struct btrfs_free_space, offset_index);
1516 		prev = entry;
1517 
1518 		if (offset < entry->offset)
1519 			n = n->rb_left;
1520 		else if (offset > entry->offset)
1521 			n = n->rb_right;
1522 		else
1523 			break;
1524 	}
1525 
1526 	if (bitmap_only) {
1527 		if (!entry)
1528 			return NULL;
1529 		if (entry->bitmap)
1530 			return entry;
1531 
1532 		/*
1533 		 * bitmap entry and extent entry may share same offset,
1534 		 * in that case, bitmap entry comes after extent entry.
1535 		 */
1536 		n = rb_next(n);
1537 		if (!n)
1538 			return NULL;
1539 		entry = rb_entry(n, struct btrfs_free_space, offset_index);
1540 		if (entry->offset != offset)
1541 			return NULL;
1542 
1543 		WARN_ON(!entry->bitmap);
1544 		return entry;
1545 	} else if (entry) {
1546 		if (entry->bitmap) {
1547 			/*
1548 			 * if previous extent entry covers the offset,
1549 			 * we should return it instead of the bitmap entry
1550 			 */
1551 			n = rb_prev(&entry->offset_index);
1552 			if (n) {
1553 				prev = rb_entry(n, struct btrfs_free_space,
1554 						offset_index);
1555 				if (!prev->bitmap &&
1556 				    prev->offset + prev->bytes > offset)
1557 					entry = prev;
1558 			}
1559 		}
1560 		return entry;
1561 	}
1562 
1563 	if (!prev)
1564 		return NULL;
1565 
1566 	/* find last entry before the 'offset' */
1567 	entry = prev;
1568 	if (entry->offset > offset) {
1569 		n = rb_prev(&entry->offset_index);
1570 		if (n) {
1571 			entry = rb_entry(n, struct btrfs_free_space,
1572 					offset_index);
1573 			ASSERT(entry->offset <= offset);
1574 		} else {
1575 			if (fuzzy)
1576 				return entry;
1577 			else
1578 				return NULL;
1579 		}
1580 	}
1581 
1582 	if (entry->bitmap) {
1583 		n = rb_prev(&entry->offset_index);
1584 		if (n) {
1585 			prev = rb_entry(n, struct btrfs_free_space,
1586 					offset_index);
1587 			if (!prev->bitmap &&
1588 			    prev->offset + prev->bytes > offset)
1589 				return prev;
1590 		}
1591 		if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1592 			return entry;
1593 	} else if (entry->offset + entry->bytes > offset)
1594 		return entry;
1595 
1596 	if (!fuzzy)
1597 		return NULL;
1598 
1599 	while (1) {
1600 		if (entry->bitmap) {
1601 			if (entry->offset + BITS_PER_BITMAP *
1602 			    ctl->unit > offset)
1603 				break;
1604 		} else {
1605 			if (entry->offset + entry->bytes > offset)
1606 				break;
1607 		}
1608 
1609 		n = rb_next(&entry->offset_index);
1610 		if (!n)
1611 			return NULL;
1612 		entry = rb_entry(n, struct btrfs_free_space, offset_index);
1613 	}
1614 	return entry;
1615 }
1616 
1617 static inline void
__unlink_free_space(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)1618 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1619 		    struct btrfs_free_space *info)
1620 {
1621 	rb_erase(&info->offset_index, &ctl->free_space_offset);
1622 	ctl->free_extents--;
1623 }
1624 
unlink_free_space(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)1625 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1626 			      struct btrfs_free_space *info)
1627 {
1628 	__unlink_free_space(ctl, info);
1629 	ctl->free_space -= info->bytes;
1630 }
1631 
link_free_space(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)1632 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1633 			   struct btrfs_free_space *info)
1634 {
1635 	int ret = 0;
1636 
1637 	ASSERT(info->bytes || info->bitmap);
1638 	ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1639 				 &info->offset_index, (info->bitmap != NULL));
1640 	if (ret)
1641 		return ret;
1642 
1643 	ctl->free_space += info->bytes;
1644 	ctl->free_extents++;
1645 	return ret;
1646 }
1647 
recalculate_thresholds(struct btrfs_free_space_ctl * ctl)1648 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1649 {
1650 	struct btrfs_block_group_cache *block_group = ctl->private;
1651 	u64 max_bytes;
1652 	u64 bitmap_bytes;
1653 	u64 extent_bytes;
1654 	u64 size = block_group->key.offset;
1655 	u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1656 	u64 max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1657 
1658 	max_bitmaps = max_t(u64, max_bitmaps, 1);
1659 
1660 	ASSERT(ctl->total_bitmaps <= max_bitmaps);
1661 
1662 	/*
1663 	 * The goal is to keep the total amount of memory used per 1gb of space
1664 	 * at or below 32k, so we need to adjust how much memory we allow to be
1665 	 * used by extent based free space tracking
1666 	 */
1667 	if (size < SZ_1G)
1668 		max_bytes = MAX_CACHE_BYTES_PER_GIG;
1669 	else
1670 		max_bytes = MAX_CACHE_BYTES_PER_GIG * div_u64(size, SZ_1G);
1671 
1672 	/*
1673 	 * we want to account for 1 more bitmap than what we have so we can make
1674 	 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1675 	 * we add more bitmaps.
1676 	 */
1677 	bitmap_bytes = (ctl->total_bitmaps + 1) * ctl->unit;
1678 
1679 	if (bitmap_bytes >= max_bytes) {
1680 		ctl->extents_thresh = 0;
1681 		return;
1682 	}
1683 
1684 	/*
1685 	 * we want the extent entry threshold to always be at most 1/2 the max
1686 	 * bytes we can have, or whatever is less than that.
1687 	 */
1688 	extent_bytes = max_bytes - bitmap_bytes;
1689 	extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
1690 
1691 	ctl->extents_thresh =
1692 		div_u64(extent_bytes, sizeof(struct btrfs_free_space));
1693 }
1694 
__bitmap_clear_bits(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset,u64 bytes)1695 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1696 				       struct btrfs_free_space *info,
1697 				       u64 offset, u64 bytes)
1698 {
1699 	unsigned long start, count;
1700 
1701 	start = offset_to_bit(info->offset, ctl->unit, offset);
1702 	count = bytes_to_bits(bytes, ctl->unit);
1703 	ASSERT(start + count <= BITS_PER_BITMAP);
1704 
1705 	bitmap_clear(info->bitmap, start, count);
1706 
1707 	info->bytes -= bytes;
1708 	if (info->max_extent_size > ctl->unit)
1709 		info->max_extent_size = 0;
1710 }
1711 
bitmap_clear_bits(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset,u64 bytes)1712 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1713 			      struct btrfs_free_space *info, u64 offset,
1714 			      u64 bytes)
1715 {
1716 	__bitmap_clear_bits(ctl, info, offset, bytes);
1717 	ctl->free_space -= bytes;
1718 }
1719 
bitmap_set_bits(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset,u64 bytes)1720 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1721 			    struct btrfs_free_space *info, u64 offset,
1722 			    u64 bytes)
1723 {
1724 	unsigned long start, count;
1725 
1726 	start = offset_to_bit(info->offset, ctl->unit, offset);
1727 	count = bytes_to_bits(bytes, ctl->unit);
1728 	ASSERT(start + count <= BITS_PER_BITMAP);
1729 
1730 	bitmap_set(info->bitmap, start, count);
1731 
1732 	info->bytes += bytes;
1733 	ctl->free_space += bytes;
1734 }
1735 
1736 /*
1737  * If we can not find suitable extent, we will use bytes to record
1738  * the size of the max extent.
1739  */
search_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * bitmap_info,u64 * offset,u64 * bytes,bool for_alloc)1740 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1741 			 struct btrfs_free_space *bitmap_info, u64 *offset,
1742 			 u64 *bytes, bool for_alloc)
1743 {
1744 	unsigned long found_bits = 0;
1745 	unsigned long max_bits = 0;
1746 	unsigned long bits, i;
1747 	unsigned long next_zero;
1748 	unsigned long extent_bits;
1749 
1750 	/*
1751 	 * Skip searching the bitmap if we don't have a contiguous section that
1752 	 * is large enough for this allocation.
1753 	 */
1754 	if (for_alloc &&
1755 	    bitmap_info->max_extent_size &&
1756 	    bitmap_info->max_extent_size < *bytes) {
1757 		*bytes = bitmap_info->max_extent_size;
1758 		return -1;
1759 	}
1760 
1761 	i = offset_to_bit(bitmap_info->offset, ctl->unit,
1762 			  max_t(u64, *offset, bitmap_info->offset));
1763 	bits = bytes_to_bits(*bytes, ctl->unit);
1764 
1765 	for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1766 		if (for_alloc && bits == 1) {
1767 			found_bits = 1;
1768 			break;
1769 		}
1770 		next_zero = find_next_zero_bit(bitmap_info->bitmap,
1771 					       BITS_PER_BITMAP, i);
1772 		extent_bits = next_zero - i;
1773 		if (extent_bits >= bits) {
1774 			found_bits = extent_bits;
1775 			break;
1776 		} else if (extent_bits > max_bits) {
1777 			max_bits = extent_bits;
1778 		}
1779 		i = next_zero;
1780 	}
1781 
1782 	if (found_bits) {
1783 		*offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1784 		*bytes = (u64)(found_bits) * ctl->unit;
1785 		return 0;
1786 	}
1787 
1788 	*bytes = (u64)(max_bits) * ctl->unit;
1789 	bitmap_info->max_extent_size = *bytes;
1790 	return -1;
1791 }
1792 
get_max_extent_size(struct btrfs_free_space * entry)1793 static inline u64 get_max_extent_size(struct btrfs_free_space *entry)
1794 {
1795 	if (entry->bitmap)
1796 		return entry->max_extent_size;
1797 	return entry->bytes;
1798 }
1799 
1800 /* Cache the size of the max extent in bytes */
1801 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)1802 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1803 		unsigned long align, u64 *max_extent_size)
1804 {
1805 	struct btrfs_free_space *entry;
1806 	struct rb_node *node;
1807 	u64 tmp;
1808 	u64 align_off;
1809 	int ret;
1810 
1811 	if (!ctl->free_space_offset.rb_node)
1812 		goto out;
1813 
1814 	entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1815 	if (!entry)
1816 		goto out;
1817 
1818 	for (node = &entry->offset_index; node; node = rb_next(node)) {
1819 		entry = rb_entry(node, struct btrfs_free_space, offset_index);
1820 		if (entry->bytes < *bytes) {
1821 			*max_extent_size = max(get_max_extent_size(entry),
1822 					       *max_extent_size);
1823 			continue;
1824 		}
1825 
1826 		/* make sure the space returned is big enough
1827 		 * to match our requested alignment
1828 		 */
1829 		if (*bytes >= align) {
1830 			tmp = entry->offset - ctl->start + align - 1;
1831 			tmp = div64_u64(tmp, align);
1832 			tmp = tmp * align + ctl->start;
1833 			align_off = tmp - entry->offset;
1834 		} else {
1835 			align_off = 0;
1836 			tmp = entry->offset;
1837 		}
1838 
1839 		if (entry->bytes < *bytes + align_off) {
1840 			*max_extent_size = max(get_max_extent_size(entry),
1841 					       *max_extent_size);
1842 			continue;
1843 		}
1844 
1845 		if (entry->bitmap) {
1846 			u64 size = *bytes;
1847 
1848 			ret = search_bitmap(ctl, entry, &tmp, &size, true);
1849 			if (!ret) {
1850 				*offset = tmp;
1851 				*bytes = size;
1852 				return entry;
1853 			} else {
1854 				*max_extent_size =
1855 					max(get_max_extent_size(entry),
1856 					    *max_extent_size);
1857 			}
1858 			continue;
1859 		}
1860 
1861 		*offset = tmp;
1862 		*bytes = entry->bytes - align_off;
1863 		return entry;
1864 	}
1865 out:
1866 	return NULL;
1867 }
1868 
add_new_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset)1869 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1870 			   struct btrfs_free_space *info, u64 offset)
1871 {
1872 	info->offset = offset_to_bitmap(ctl, offset);
1873 	info->bytes = 0;
1874 	INIT_LIST_HEAD(&info->list);
1875 	link_free_space(ctl, info);
1876 	ctl->total_bitmaps++;
1877 
1878 	ctl->op->recalc_thresholds(ctl);
1879 }
1880 
free_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * bitmap_info)1881 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1882 			struct btrfs_free_space *bitmap_info)
1883 {
1884 	unlink_free_space(ctl, bitmap_info);
1885 	kmem_cache_free(btrfs_free_space_bitmap_cachep, bitmap_info->bitmap);
1886 	kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1887 	ctl->total_bitmaps--;
1888 	ctl->op->recalc_thresholds(ctl);
1889 }
1890 
remove_from_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * bitmap_info,u64 * offset,u64 * bytes)1891 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1892 			      struct btrfs_free_space *bitmap_info,
1893 			      u64 *offset, u64 *bytes)
1894 {
1895 	u64 end;
1896 	u64 search_start, search_bytes;
1897 	int ret;
1898 
1899 again:
1900 	end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1901 
1902 	/*
1903 	 * We need to search for bits in this bitmap.  We could only cover some
1904 	 * of the extent in this bitmap thanks to how we add space, so we need
1905 	 * to search for as much as it as we can and clear that amount, and then
1906 	 * go searching for the next bit.
1907 	 */
1908 	search_start = *offset;
1909 	search_bytes = ctl->unit;
1910 	search_bytes = min(search_bytes, end - search_start + 1);
1911 	ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes,
1912 			    false);
1913 	if (ret < 0 || search_start != *offset)
1914 		return -EINVAL;
1915 
1916 	/* We may have found more bits than what we need */
1917 	search_bytes = min(search_bytes, *bytes);
1918 
1919 	/* Cannot clear past the end of the bitmap */
1920 	search_bytes = min(search_bytes, end - search_start + 1);
1921 
1922 	bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1923 	*offset += search_bytes;
1924 	*bytes -= search_bytes;
1925 
1926 	if (*bytes) {
1927 		struct rb_node *next = rb_next(&bitmap_info->offset_index);
1928 		if (!bitmap_info->bytes)
1929 			free_bitmap(ctl, bitmap_info);
1930 
1931 		/*
1932 		 * no entry after this bitmap, but we still have bytes to
1933 		 * remove, so something has gone wrong.
1934 		 */
1935 		if (!next)
1936 			return -EINVAL;
1937 
1938 		bitmap_info = rb_entry(next, struct btrfs_free_space,
1939 				       offset_index);
1940 
1941 		/*
1942 		 * if the next entry isn't a bitmap we need to return to let the
1943 		 * extent stuff do its work.
1944 		 */
1945 		if (!bitmap_info->bitmap)
1946 			return -EAGAIN;
1947 
1948 		/*
1949 		 * Ok the next item is a bitmap, but it may not actually hold
1950 		 * the information for the rest of this free space stuff, so
1951 		 * look for it, and if we don't find it return so we can try
1952 		 * everything over again.
1953 		 */
1954 		search_start = *offset;
1955 		search_bytes = ctl->unit;
1956 		ret = search_bitmap(ctl, bitmap_info, &search_start,
1957 				    &search_bytes, false);
1958 		if (ret < 0 || search_start != *offset)
1959 			return -EAGAIN;
1960 
1961 		goto again;
1962 	} else if (!bitmap_info->bytes)
1963 		free_bitmap(ctl, bitmap_info);
1964 
1965 	return 0;
1966 }
1967 
add_bytes_to_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,u64 offset,u64 bytes)1968 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1969 			       struct btrfs_free_space *info, u64 offset,
1970 			       u64 bytes)
1971 {
1972 	u64 bytes_to_set = 0;
1973 	u64 end;
1974 
1975 	end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1976 
1977 	bytes_to_set = min(end - offset, bytes);
1978 
1979 	bitmap_set_bits(ctl, info, offset, bytes_to_set);
1980 
1981 	/*
1982 	 * We set some bytes, we have no idea what the max extent size is
1983 	 * anymore.
1984 	 */
1985 	info->max_extent_size = 0;
1986 
1987 	return bytes_to_set;
1988 
1989 }
1990 
use_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)1991 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1992 		      struct btrfs_free_space *info)
1993 {
1994 	struct btrfs_block_group_cache *block_group = ctl->private;
1995 	struct btrfs_fs_info *fs_info = block_group->fs_info;
1996 	bool forced = false;
1997 
1998 #ifdef CONFIG_BTRFS_DEBUG
1999 	if (btrfs_should_fragment_free_space(block_group))
2000 		forced = true;
2001 #endif
2002 
2003 	/*
2004 	 * If we are below the extents threshold then we can add this as an
2005 	 * extent, and don't have to deal with the bitmap
2006 	 */
2007 	if (!forced && ctl->free_extents < ctl->extents_thresh) {
2008 		/*
2009 		 * If this block group has some small extents we don't want to
2010 		 * use up all of our free slots in the cache with them, we want
2011 		 * to reserve them to larger extents, however if we have plenty
2012 		 * of cache left then go ahead an dadd them, no sense in adding
2013 		 * the overhead of a bitmap if we don't have to.
2014 		 */
2015 		if (info->bytes <= fs_info->sectorsize * 4) {
2016 			if (ctl->free_extents * 2 <= ctl->extents_thresh)
2017 				return false;
2018 		} else {
2019 			return false;
2020 		}
2021 	}
2022 
2023 	/*
2024 	 * The original block groups from mkfs can be really small, like 8
2025 	 * megabytes, so don't bother with a bitmap for those entries.  However
2026 	 * some block groups can be smaller than what a bitmap would cover but
2027 	 * are still large enough that they could overflow the 32k memory limit,
2028 	 * so allow those block groups to still be allowed to have a bitmap
2029 	 * entry.
2030 	 */
2031 	if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
2032 		return false;
2033 
2034 	return true;
2035 }
2036 
2037 static const struct btrfs_free_space_op free_space_op = {
2038 	.recalc_thresholds	= recalculate_thresholds,
2039 	.use_bitmap		= use_bitmap,
2040 };
2041 
insert_into_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)2042 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
2043 			      struct btrfs_free_space *info)
2044 {
2045 	struct btrfs_free_space *bitmap_info;
2046 	struct btrfs_block_group_cache *block_group = NULL;
2047 	int added = 0;
2048 	u64 bytes, offset, bytes_added;
2049 	int ret;
2050 
2051 	bytes = info->bytes;
2052 	offset = info->offset;
2053 
2054 	if (!ctl->op->use_bitmap(ctl, info))
2055 		return 0;
2056 
2057 	if (ctl->op == &free_space_op)
2058 		block_group = ctl->private;
2059 again:
2060 	/*
2061 	 * Since we link bitmaps right into the cluster we need to see if we
2062 	 * have a cluster here, and if so and it has our bitmap we need to add
2063 	 * the free space to that bitmap.
2064 	 */
2065 	if (block_group && !list_empty(&block_group->cluster_list)) {
2066 		struct btrfs_free_cluster *cluster;
2067 		struct rb_node *node;
2068 		struct btrfs_free_space *entry;
2069 
2070 		cluster = list_entry(block_group->cluster_list.next,
2071 				     struct btrfs_free_cluster,
2072 				     block_group_list);
2073 		spin_lock(&cluster->lock);
2074 		node = rb_first(&cluster->root);
2075 		if (!node) {
2076 			spin_unlock(&cluster->lock);
2077 			goto no_cluster_bitmap;
2078 		}
2079 
2080 		entry = rb_entry(node, struct btrfs_free_space, offset_index);
2081 		if (!entry->bitmap) {
2082 			spin_unlock(&cluster->lock);
2083 			goto no_cluster_bitmap;
2084 		}
2085 
2086 		if (entry->offset == offset_to_bitmap(ctl, offset)) {
2087 			bytes_added = add_bytes_to_bitmap(ctl, entry,
2088 							  offset, bytes);
2089 			bytes -= bytes_added;
2090 			offset += bytes_added;
2091 		}
2092 		spin_unlock(&cluster->lock);
2093 		if (!bytes) {
2094 			ret = 1;
2095 			goto out;
2096 		}
2097 	}
2098 
2099 no_cluster_bitmap:
2100 	bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2101 					 1, 0);
2102 	if (!bitmap_info) {
2103 		ASSERT(added == 0);
2104 		goto new_bitmap;
2105 	}
2106 
2107 	bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
2108 	bytes -= bytes_added;
2109 	offset += bytes_added;
2110 	added = 0;
2111 
2112 	if (!bytes) {
2113 		ret = 1;
2114 		goto out;
2115 	} else
2116 		goto again;
2117 
2118 new_bitmap:
2119 	if (info && info->bitmap) {
2120 		add_new_bitmap(ctl, info, offset);
2121 		added = 1;
2122 		info = NULL;
2123 		goto again;
2124 	} else {
2125 		spin_unlock(&ctl->tree_lock);
2126 
2127 		/* no pre-allocated info, allocate a new one */
2128 		if (!info) {
2129 			info = kmem_cache_zalloc(btrfs_free_space_cachep,
2130 						 GFP_NOFS);
2131 			if (!info) {
2132 				spin_lock(&ctl->tree_lock);
2133 				ret = -ENOMEM;
2134 				goto out;
2135 			}
2136 		}
2137 
2138 		/* allocate the bitmap */
2139 		info->bitmap = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep,
2140 						 GFP_NOFS);
2141 		spin_lock(&ctl->tree_lock);
2142 		if (!info->bitmap) {
2143 			ret = -ENOMEM;
2144 			goto out;
2145 		}
2146 		goto again;
2147 	}
2148 
2149 out:
2150 	if (info) {
2151 		if (info->bitmap)
2152 			kmem_cache_free(btrfs_free_space_bitmap_cachep,
2153 					info->bitmap);
2154 		kmem_cache_free(btrfs_free_space_cachep, info);
2155 	}
2156 
2157 	return ret;
2158 }
2159 
try_merge_free_space(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2160 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2161 			  struct btrfs_free_space *info, bool update_stat)
2162 {
2163 	struct btrfs_free_space *left_info;
2164 	struct btrfs_free_space *right_info;
2165 	bool merged = false;
2166 	u64 offset = info->offset;
2167 	u64 bytes = info->bytes;
2168 
2169 	/*
2170 	 * first we want to see if there is free space adjacent to the range we
2171 	 * are adding, if there is remove that struct and add a new one to
2172 	 * cover the entire range
2173 	 */
2174 	right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
2175 	if (right_info && rb_prev(&right_info->offset_index))
2176 		left_info = rb_entry(rb_prev(&right_info->offset_index),
2177 				     struct btrfs_free_space, offset_index);
2178 	else
2179 		left_info = tree_search_offset(ctl, offset - 1, 0, 0);
2180 
2181 	if (right_info && !right_info->bitmap) {
2182 		if (update_stat)
2183 			unlink_free_space(ctl, right_info);
2184 		else
2185 			__unlink_free_space(ctl, right_info);
2186 		info->bytes += right_info->bytes;
2187 		kmem_cache_free(btrfs_free_space_cachep, right_info);
2188 		merged = true;
2189 	}
2190 
2191 	if (left_info && !left_info->bitmap &&
2192 	    left_info->offset + left_info->bytes == offset) {
2193 		if (update_stat)
2194 			unlink_free_space(ctl, left_info);
2195 		else
2196 			__unlink_free_space(ctl, left_info);
2197 		info->offset = left_info->offset;
2198 		info->bytes += left_info->bytes;
2199 		kmem_cache_free(btrfs_free_space_cachep, left_info);
2200 		merged = true;
2201 	}
2202 
2203 	return merged;
2204 }
2205 
steal_from_bitmap_to_end(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2206 static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2207 				     struct btrfs_free_space *info,
2208 				     bool update_stat)
2209 {
2210 	struct btrfs_free_space *bitmap;
2211 	unsigned long i;
2212 	unsigned long j;
2213 	const u64 end = info->offset + info->bytes;
2214 	const u64 bitmap_offset = offset_to_bitmap(ctl, end);
2215 	u64 bytes;
2216 
2217 	bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2218 	if (!bitmap)
2219 		return false;
2220 
2221 	i = offset_to_bit(bitmap->offset, ctl->unit, end);
2222 	j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
2223 	if (j == i)
2224 		return false;
2225 	bytes = (j - i) * ctl->unit;
2226 	info->bytes += bytes;
2227 
2228 	if (update_stat)
2229 		bitmap_clear_bits(ctl, bitmap, end, bytes);
2230 	else
2231 		__bitmap_clear_bits(ctl, bitmap, end, bytes);
2232 
2233 	if (!bitmap->bytes)
2234 		free_bitmap(ctl, bitmap);
2235 
2236 	return true;
2237 }
2238 
steal_from_bitmap_to_front(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2239 static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2240 				       struct btrfs_free_space *info,
2241 				       bool update_stat)
2242 {
2243 	struct btrfs_free_space *bitmap;
2244 	u64 bitmap_offset;
2245 	unsigned long i;
2246 	unsigned long j;
2247 	unsigned long prev_j;
2248 	u64 bytes;
2249 
2250 	bitmap_offset = offset_to_bitmap(ctl, info->offset);
2251 	/* If we're on a boundary, try the previous logical bitmap. */
2252 	if (bitmap_offset == info->offset) {
2253 		if (info->offset == 0)
2254 			return false;
2255 		bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
2256 	}
2257 
2258 	bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2259 	if (!bitmap)
2260 		return false;
2261 
2262 	i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
2263 	j = 0;
2264 	prev_j = (unsigned long)-1;
2265 	for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2266 		if (j > i)
2267 			break;
2268 		prev_j = j;
2269 	}
2270 	if (prev_j == i)
2271 		return false;
2272 
2273 	if (prev_j == (unsigned long)-1)
2274 		bytes = (i + 1) * ctl->unit;
2275 	else
2276 		bytes = (i - prev_j) * ctl->unit;
2277 
2278 	info->offset -= bytes;
2279 	info->bytes += bytes;
2280 
2281 	if (update_stat)
2282 		bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2283 	else
2284 		__bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2285 
2286 	if (!bitmap->bytes)
2287 		free_bitmap(ctl, bitmap);
2288 
2289 	return true;
2290 }
2291 
2292 /*
2293  * We prefer always to allocate from extent entries, both for clustered and
2294  * non-clustered allocation requests. So when attempting to add a new extent
2295  * entry, try to see if there's adjacent free space in bitmap entries, and if
2296  * there is, migrate that space from the bitmaps to the extent.
2297  * Like this we get better chances of satisfying space allocation requests
2298  * because we attempt to satisfy them based on a single cache entry, and never
2299  * on 2 or more entries - even if the entries represent a contiguous free space
2300  * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2301  * ends).
2302  */
steal_from_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info,bool update_stat)2303 static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2304 			      struct btrfs_free_space *info,
2305 			      bool update_stat)
2306 {
2307 	/*
2308 	 * Only work with disconnected entries, as we can change their offset,
2309 	 * and must be extent entries.
2310 	 */
2311 	ASSERT(!info->bitmap);
2312 	ASSERT(RB_EMPTY_NODE(&info->offset_index));
2313 
2314 	if (ctl->total_bitmaps > 0) {
2315 		bool stole_end;
2316 		bool stole_front = false;
2317 
2318 		stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2319 		if (ctl->total_bitmaps > 0)
2320 			stole_front = steal_from_bitmap_to_front(ctl, info,
2321 								 update_stat);
2322 
2323 		if (stole_end || stole_front)
2324 			try_merge_free_space(ctl, info, update_stat);
2325 	}
2326 }
2327 
__btrfs_add_free_space(struct btrfs_fs_info * fs_info,struct btrfs_free_space_ctl * ctl,u64 offset,u64 bytes)2328 int __btrfs_add_free_space(struct btrfs_fs_info *fs_info,
2329 			   struct btrfs_free_space_ctl *ctl,
2330 			   u64 offset, u64 bytes)
2331 {
2332 	struct btrfs_free_space *info;
2333 	int ret = 0;
2334 
2335 	info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2336 	if (!info)
2337 		return -ENOMEM;
2338 
2339 	info->offset = offset;
2340 	info->bytes = bytes;
2341 	RB_CLEAR_NODE(&info->offset_index);
2342 
2343 	spin_lock(&ctl->tree_lock);
2344 
2345 	if (try_merge_free_space(ctl, info, true))
2346 		goto link;
2347 
2348 	/*
2349 	 * There was no extent directly to the left or right of this new
2350 	 * extent then we know we're going to have to allocate a new extent, so
2351 	 * before we do that see if we need to drop this into a bitmap
2352 	 */
2353 	ret = insert_into_bitmap(ctl, info);
2354 	if (ret < 0) {
2355 		goto out;
2356 	} else if (ret) {
2357 		ret = 0;
2358 		goto out;
2359 	}
2360 link:
2361 	/*
2362 	 * Only steal free space from adjacent bitmaps if we're sure we're not
2363 	 * going to add the new free space to existing bitmap entries - because
2364 	 * that would mean unnecessary work that would be reverted. Therefore
2365 	 * attempt to steal space from bitmaps if we're adding an extent entry.
2366 	 */
2367 	steal_from_bitmap(ctl, info, true);
2368 
2369 	ret = link_free_space(ctl, info);
2370 	if (ret)
2371 		kmem_cache_free(btrfs_free_space_cachep, info);
2372 out:
2373 	spin_unlock(&ctl->tree_lock);
2374 
2375 	if (ret) {
2376 		btrfs_crit(fs_info, "unable to add free space :%d", ret);
2377 		ASSERT(ret != -EEXIST);
2378 	}
2379 
2380 	return ret;
2381 }
2382 
btrfs_add_free_space(struct btrfs_block_group_cache * block_group,u64 bytenr,u64 size)2383 int btrfs_add_free_space(struct btrfs_block_group_cache *block_group,
2384 			 u64 bytenr, u64 size)
2385 {
2386 	return __btrfs_add_free_space(block_group->fs_info,
2387 				      block_group->free_space_ctl,
2388 				      bytenr, size);
2389 }
2390 
btrfs_remove_free_space(struct btrfs_block_group_cache * block_group,u64 offset,u64 bytes)2391 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
2392 			    u64 offset, u64 bytes)
2393 {
2394 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2395 	struct btrfs_free_space *info;
2396 	int ret;
2397 	bool re_search = false;
2398 
2399 	spin_lock(&ctl->tree_lock);
2400 
2401 again:
2402 	ret = 0;
2403 	if (!bytes)
2404 		goto out_lock;
2405 
2406 	info = tree_search_offset(ctl, offset, 0, 0);
2407 	if (!info) {
2408 		/*
2409 		 * oops didn't find an extent that matched the space we wanted
2410 		 * to remove, look for a bitmap instead
2411 		 */
2412 		info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2413 					  1, 0);
2414 		if (!info) {
2415 			/*
2416 			 * If we found a partial bit of our free space in a
2417 			 * bitmap but then couldn't find the other part this may
2418 			 * be a problem, so WARN about it.
2419 			 */
2420 			WARN_ON(re_search);
2421 			goto out_lock;
2422 		}
2423 	}
2424 
2425 	re_search = false;
2426 	if (!info->bitmap) {
2427 		unlink_free_space(ctl, info);
2428 		if (offset == info->offset) {
2429 			u64 to_free = min(bytes, info->bytes);
2430 
2431 			info->bytes -= to_free;
2432 			info->offset += to_free;
2433 			if (info->bytes) {
2434 				ret = link_free_space(ctl, info);
2435 				WARN_ON(ret);
2436 			} else {
2437 				kmem_cache_free(btrfs_free_space_cachep, info);
2438 			}
2439 
2440 			offset += to_free;
2441 			bytes -= to_free;
2442 			goto again;
2443 		} else {
2444 			u64 old_end = info->bytes + info->offset;
2445 
2446 			info->bytes = offset - info->offset;
2447 			ret = link_free_space(ctl, info);
2448 			WARN_ON(ret);
2449 			if (ret)
2450 				goto out_lock;
2451 
2452 			/* Not enough bytes in this entry to satisfy us */
2453 			if (old_end < offset + bytes) {
2454 				bytes -= old_end - offset;
2455 				offset = old_end;
2456 				goto again;
2457 			} else if (old_end == offset + bytes) {
2458 				/* all done */
2459 				goto out_lock;
2460 			}
2461 			spin_unlock(&ctl->tree_lock);
2462 
2463 			ret = btrfs_add_free_space(block_group, offset + bytes,
2464 						   old_end - (offset + bytes));
2465 			WARN_ON(ret);
2466 			goto out;
2467 		}
2468 	}
2469 
2470 	ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2471 	if (ret == -EAGAIN) {
2472 		re_search = true;
2473 		goto again;
2474 	}
2475 out_lock:
2476 	spin_unlock(&ctl->tree_lock);
2477 out:
2478 	return ret;
2479 }
2480 
btrfs_dump_free_space(struct btrfs_block_group_cache * block_group,u64 bytes)2481 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
2482 			   u64 bytes)
2483 {
2484 	struct btrfs_fs_info *fs_info = block_group->fs_info;
2485 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2486 	struct btrfs_free_space *info;
2487 	struct rb_node *n;
2488 	int count = 0;
2489 
2490 	spin_lock(&ctl->tree_lock);
2491 	for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2492 		info = rb_entry(n, struct btrfs_free_space, offset_index);
2493 		if (info->bytes >= bytes && !block_group->ro)
2494 			count++;
2495 		btrfs_crit(fs_info, "entry offset %llu, bytes %llu, bitmap %s",
2496 			   info->offset, info->bytes,
2497 		       (info->bitmap) ? "yes" : "no");
2498 	}
2499 	spin_unlock(&ctl->tree_lock);
2500 	btrfs_info(fs_info, "block group has cluster?: %s",
2501 	       list_empty(&block_group->cluster_list) ? "no" : "yes");
2502 	btrfs_info(fs_info,
2503 		   "%d blocks of free space at or bigger than bytes is", count);
2504 }
2505 
btrfs_init_free_space_ctl(struct btrfs_block_group_cache * block_group)2506 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2507 {
2508 	struct btrfs_fs_info *fs_info = block_group->fs_info;
2509 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2510 
2511 	spin_lock_init(&ctl->tree_lock);
2512 	ctl->unit = fs_info->sectorsize;
2513 	ctl->start = block_group->key.objectid;
2514 	ctl->private = block_group;
2515 	ctl->op = &free_space_op;
2516 	INIT_LIST_HEAD(&ctl->trimming_ranges);
2517 	mutex_init(&ctl->cache_writeout_mutex);
2518 
2519 	/*
2520 	 * we only want to have 32k of ram per block group for keeping
2521 	 * track of free space, and if we pass 1/2 of that we want to
2522 	 * start converting things over to using bitmaps
2523 	 */
2524 	ctl->extents_thresh = (SZ_32K / 2) / sizeof(struct btrfs_free_space);
2525 }
2526 
2527 /*
2528  * for a given cluster, put all of its extents back into the free
2529  * space cache.  If the block group passed doesn't match the block group
2530  * pointed to by the cluster, someone else raced in and freed the
2531  * cluster already.  In that case, we just return without changing anything
2532  */
2533 static int
__btrfs_return_cluster_to_free_space(struct btrfs_block_group_cache * block_group,struct btrfs_free_cluster * cluster)2534 __btrfs_return_cluster_to_free_space(
2535 			     struct btrfs_block_group_cache *block_group,
2536 			     struct btrfs_free_cluster *cluster)
2537 {
2538 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2539 	struct btrfs_free_space *entry;
2540 	struct rb_node *node;
2541 
2542 	spin_lock(&cluster->lock);
2543 	if (cluster->block_group != block_group)
2544 		goto out;
2545 
2546 	cluster->block_group = NULL;
2547 	cluster->window_start = 0;
2548 	list_del_init(&cluster->block_group_list);
2549 
2550 	node = rb_first(&cluster->root);
2551 	while (node) {
2552 		bool bitmap;
2553 
2554 		entry = rb_entry(node, struct btrfs_free_space, offset_index);
2555 		node = rb_next(&entry->offset_index);
2556 		rb_erase(&entry->offset_index, &cluster->root);
2557 		RB_CLEAR_NODE(&entry->offset_index);
2558 
2559 		bitmap = (entry->bitmap != NULL);
2560 		if (!bitmap) {
2561 			try_merge_free_space(ctl, entry, false);
2562 			steal_from_bitmap(ctl, entry, false);
2563 		}
2564 		tree_insert_offset(&ctl->free_space_offset,
2565 				   entry->offset, &entry->offset_index, bitmap);
2566 	}
2567 	cluster->root = RB_ROOT;
2568 
2569 out:
2570 	spin_unlock(&cluster->lock);
2571 	btrfs_put_block_group(block_group);
2572 	return 0;
2573 }
2574 
__btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl * ctl)2575 static void __btrfs_remove_free_space_cache_locked(
2576 				struct btrfs_free_space_ctl *ctl)
2577 {
2578 	struct btrfs_free_space *info;
2579 	struct rb_node *node;
2580 
2581 	while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2582 		info = rb_entry(node, struct btrfs_free_space, offset_index);
2583 		if (!info->bitmap) {
2584 			unlink_free_space(ctl, info);
2585 			kmem_cache_free(btrfs_free_space_cachep, info);
2586 		} else {
2587 			free_bitmap(ctl, info);
2588 		}
2589 
2590 		cond_resched_lock(&ctl->tree_lock);
2591 	}
2592 }
2593 
__btrfs_remove_free_space_cache(struct btrfs_free_space_ctl * ctl)2594 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2595 {
2596 	spin_lock(&ctl->tree_lock);
2597 	__btrfs_remove_free_space_cache_locked(ctl);
2598 	spin_unlock(&ctl->tree_lock);
2599 }
2600 
btrfs_remove_free_space_cache(struct btrfs_block_group_cache * block_group)2601 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2602 {
2603 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2604 	struct btrfs_free_cluster *cluster;
2605 	struct list_head *head;
2606 
2607 	spin_lock(&ctl->tree_lock);
2608 	while ((head = block_group->cluster_list.next) !=
2609 	       &block_group->cluster_list) {
2610 		cluster = list_entry(head, struct btrfs_free_cluster,
2611 				     block_group_list);
2612 
2613 		WARN_ON(cluster->block_group != block_group);
2614 		__btrfs_return_cluster_to_free_space(block_group, cluster);
2615 
2616 		cond_resched_lock(&ctl->tree_lock);
2617 	}
2618 	__btrfs_remove_free_space_cache_locked(ctl);
2619 	spin_unlock(&ctl->tree_lock);
2620 
2621 }
2622 
btrfs_find_space_for_alloc(struct btrfs_block_group_cache * block_group,u64 offset,u64 bytes,u64 empty_size,u64 * max_extent_size)2623 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2624 			       u64 offset, u64 bytes, u64 empty_size,
2625 			       u64 *max_extent_size)
2626 {
2627 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2628 	struct btrfs_free_space *entry = NULL;
2629 	u64 bytes_search = bytes + empty_size;
2630 	u64 ret = 0;
2631 	u64 align_gap = 0;
2632 	u64 align_gap_len = 0;
2633 
2634 	spin_lock(&ctl->tree_lock);
2635 	entry = find_free_space(ctl, &offset, &bytes_search,
2636 				block_group->full_stripe_len, max_extent_size);
2637 	if (!entry)
2638 		goto out;
2639 
2640 	ret = offset;
2641 	if (entry->bitmap) {
2642 		bitmap_clear_bits(ctl, entry, offset, bytes);
2643 		if (!entry->bytes)
2644 			free_bitmap(ctl, entry);
2645 	} else {
2646 		unlink_free_space(ctl, entry);
2647 		align_gap_len = offset - entry->offset;
2648 		align_gap = entry->offset;
2649 
2650 		entry->offset = offset + bytes;
2651 		WARN_ON(entry->bytes < bytes + align_gap_len);
2652 
2653 		entry->bytes -= bytes + align_gap_len;
2654 		if (!entry->bytes)
2655 			kmem_cache_free(btrfs_free_space_cachep, entry);
2656 		else
2657 			link_free_space(ctl, entry);
2658 	}
2659 out:
2660 	spin_unlock(&ctl->tree_lock);
2661 
2662 	if (align_gap_len)
2663 		__btrfs_add_free_space(block_group->fs_info, ctl,
2664 				       align_gap, align_gap_len);
2665 	return ret;
2666 }
2667 
2668 /*
2669  * given a cluster, put all of its extents back into the free space
2670  * cache.  If a block group is passed, this function will only free
2671  * a cluster that belongs to the passed block group.
2672  *
2673  * Otherwise, it'll get a reference on the block group pointed to by the
2674  * cluster and remove the cluster from it.
2675  */
btrfs_return_cluster_to_free_space(struct btrfs_block_group_cache * block_group,struct btrfs_free_cluster * cluster)2676 int btrfs_return_cluster_to_free_space(
2677 			       struct btrfs_block_group_cache *block_group,
2678 			       struct btrfs_free_cluster *cluster)
2679 {
2680 	struct btrfs_free_space_ctl *ctl;
2681 	int ret;
2682 
2683 	/* first, get a safe pointer to the block group */
2684 	spin_lock(&cluster->lock);
2685 	if (!block_group) {
2686 		block_group = cluster->block_group;
2687 		if (!block_group) {
2688 			spin_unlock(&cluster->lock);
2689 			return 0;
2690 		}
2691 	} else if (cluster->block_group != block_group) {
2692 		/* someone else has already freed it don't redo their work */
2693 		spin_unlock(&cluster->lock);
2694 		return 0;
2695 	}
2696 	atomic_inc(&block_group->count);
2697 	spin_unlock(&cluster->lock);
2698 
2699 	ctl = block_group->free_space_ctl;
2700 
2701 	/* now return any extents the cluster had on it */
2702 	spin_lock(&ctl->tree_lock);
2703 	ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2704 	spin_unlock(&ctl->tree_lock);
2705 
2706 	/* finally drop our ref */
2707 	btrfs_put_block_group(block_group);
2708 	return ret;
2709 }
2710 
btrfs_alloc_from_bitmap(struct btrfs_block_group_cache * block_group,struct btrfs_free_cluster * cluster,struct btrfs_free_space * entry,u64 bytes,u64 min_start,u64 * max_extent_size)2711 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2712 				   struct btrfs_free_cluster *cluster,
2713 				   struct btrfs_free_space *entry,
2714 				   u64 bytes, u64 min_start,
2715 				   u64 *max_extent_size)
2716 {
2717 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2718 	int err;
2719 	u64 search_start = cluster->window_start;
2720 	u64 search_bytes = bytes;
2721 	u64 ret = 0;
2722 
2723 	search_start = min_start;
2724 	search_bytes = bytes;
2725 
2726 	err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
2727 	if (err) {
2728 		*max_extent_size = max(get_max_extent_size(entry),
2729 				       *max_extent_size);
2730 		return 0;
2731 	}
2732 
2733 	ret = search_start;
2734 	__bitmap_clear_bits(ctl, entry, ret, bytes);
2735 
2736 	return ret;
2737 }
2738 
2739 /*
2740  * given a cluster, try to allocate 'bytes' from it, returns 0
2741  * if it couldn't find anything suitably large, or a logical disk offset
2742  * if things worked out
2743  */
btrfs_alloc_from_cluster(struct btrfs_block_group_cache * block_group,struct btrfs_free_cluster * cluster,u64 bytes,u64 min_start,u64 * max_extent_size)2744 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2745 			     struct btrfs_free_cluster *cluster, u64 bytes,
2746 			     u64 min_start, u64 *max_extent_size)
2747 {
2748 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2749 	struct btrfs_free_space *entry = NULL;
2750 	struct rb_node *node;
2751 	u64 ret = 0;
2752 
2753 	spin_lock(&cluster->lock);
2754 	if (bytes > cluster->max_size)
2755 		goto out;
2756 
2757 	if (cluster->block_group != block_group)
2758 		goto out;
2759 
2760 	node = rb_first(&cluster->root);
2761 	if (!node)
2762 		goto out;
2763 
2764 	entry = rb_entry(node, struct btrfs_free_space, offset_index);
2765 	while (1) {
2766 		if (entry->bytes < bytes)
2767 			*max_extent_size = max(get_max_extent_size(entry),
2768 					       *max_extent_size);
2769 
2770 		if (entry->bytes < bytes ||
2771 		    (!entry->bitmap && entry->offset < min_start)) {
2772 			node = rb_next(&entry->offset_index);
2773 			if (!node)
2774 				break;
2775 			entry = rb_entry(node, struct btrfs_free_space,
2776 					 offset_index);
2777 			continue;
2778 		}
2779 
2780 		if (entry->bitmap) {
2781 			ret = btrfs_alloc_from_bitmap(block_group,
2782 						      cluster, entry, bytes,
2783 						      cluster->window_start,
2784 						      max_extent_size);
2785 			if (ret == 0) {
2786 				node = rb_next(&entry->offset_index);
2787 				if (!node)
2788 					break;
2789 				entry = rb_entry(node, struct btrfs_free_space,
2790 						 offset_index);
2791 				continue;
2792 			}
2793 			cluster->window_start += bytes;
2794 		} else {
2795 			ret = entry->offset;
2796 
2797 			entry->offset += bytes;
2798 			entry->bytes -= bytes;
2799 		}
2800 
2801 		if (entry->bytes == 0)
2802 			rb_erase(&entry->offset_index, &cluster->root);
2803 		break;
2804 	}
2805 out:
2806 	spin_unlock(&cluster->lock);
2807 
2808 	if (!ret)
2809 		return 0;
2810 
2811 	spin_lock(&ctl->tree_lock);
2812 
2813 	ctl->free_space -= bytes;
2814 	if (entry->bytes == 0) {
2815 		ctl->free_extents--;
2816 		if (entry->bitmap) {
2817 			kmem_cache_free(btrfs_free_space_bitmap_cachep,
2818 					entry->bitmap);
2819 			ctl->total_bitmaps--;
2820 			ctl->op->recalc_thresholds(ctl);
2821 		}
2822 		kmem_cache_free(btrfs_free_space_cachep, entry);
2823 	}
2824 
2825 	spin_unlock(&ctl->tree_lock);
2826 
2827 	return ret;
2828 }
2829 
btrfs_bitmap_cluster(struct btrfs_block_group_cache * block_group,struct btrfs_free_space * entry,struct btrfs_free_cluster * cluster,u64 offset,u64 bytes,u64 cont1_bytes,u64 min_bytes)2830 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2831 				struct btrfs_free_space *entry,
2832 				struct btrfs_free_cluster *cluster,
2833 				u64 offset, u64 bytes,
2834 				u64 cont1_bytes, u64 min_bytes)
2835 {
2836 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2837 	unsigned long next_zero;
2838 	unsigned long i;
2839 	unsigned long want_bits;
2840 	unsigned long min_bits;
2841 	unsigned long found_bits;
2842 	unsigned long max_bits = 0;
2843 	unsigned long start = 0;
2844 	unsigned long total_found = 0;
2845 	int ret;
2846 
2847 	i = offset_to_bit(entry->offset, ctl->unit,
2848 			  max_t(u64, offset, entry->offset));
2849 	want_bits = bytes_to_bits(bytes, ctl->unit);
2850 	min_bits = bytes_to_bits(min_bytes, ctl->unit);
2851 
2852 	/*
2853 	 * Don't bother looking for a cluster in this bitmap if it's heavily
2854 	 * fragmented.
2855 	 */
2856 	if (entry->max_extent_size &&
2857 	    entry->max_extent_size < cont1_bytes)
2858 		return -ENOSPC;
2859 again:
2860 	found_bits = 0;
2861 	for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2862 		next_zero = find_next_zero_bit(entry->bitmap,
2863 					       BITS_PER_BITMAP, i);
2864 		if (next_zero - i >= min_bits) {
2865 			found_bits = next_zero - i;
2866 			if (found_bits > max_bits)
2867 				max_bits = found_bits;
2868 			break;
2869 		}
2870 		if (next_zero - i > max_bits)
2871 			max_bits = next_zero - i;
2872 		i = next_zero;
2873 	}
2874 
2875 	if (!found_bits) {
2876 		entry->max_extent_size = (u64)max_bits * ctl->unit;
2877 		return -ENOSPC;
2878 	}
2879 
2880 	if (!total_found) {
2881 		start = i;
2882 		cluster->max_size = 0;
2883 	}
2884 
2885 	total_found += found_bits;
2886 
2887 	if (cluster->max_size < found_bits * ctl->unit)
2888 		cluster->max_size = found_bits * ctl->unit;
2889 
2890 	if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2891 		i = next_zero + 1;
2892 		goto again;
2893 	}
2894 
2895 	cluster->window_start = start * ctl->unit + entry->offset;
2896 	rb_erase(&entry->offset_index, &ctl->free_space_offset);
2897 	ret = tree_insert_offset(&cluster->root, entry->offset,
2898 				 &entry->offset_index, 1);
2899 	ASSERT(!ret); /* -EEXIST; Logic error */
2900 
2901 	trace_btrfs_setup_cluster(block_group, cluster,
2902 				  total_found * ctl->unit, 1);
2903 	return 0;
2904 }
2905 
2906 /*
2907  * This searches the block group for just extents to fill the cluster with.
2908  * Try to find a cluster with at least bytes total bytes, at least one
2909  * extent of cont1_bytes, and other clusters of at least min_bytes.
2910  */
2911 static noinline int
setup_cluster_no_bitmap(struct btrfs_block_group_cache * block_group,struct btrfs_free_cluster * cluster,struct list_head * bitmaps,u64 offset,u64 bytes,u64 cont1_bytes,u64 min_bytes)2912 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2913 			struct btrfs_free_cluster *cluster,
2914 			struct list_head *bitmaps, u64 offset, u64 bytes,
2915 			u64 cont1_bytes, u64 min_bytes)
2916 {
2917 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2918 	struct btrfs_free_space *first = NULL;
2919 	struct btrfs_free_space *entry = NULL;
2920 	struct btrfs_free_space *last;
2921 	struct rb_node *node;
2922 	u64 window_free;
2923 	u64 max_extent;
2924 	u64 total_size = 0;
2925 
2926 	entry = tree_search_offset(ctl, offset, 0, 1);
2927 	if (!entry)
2928 		return -ENOSPC;
2929 
2930 	/*
2931 	 * We don't want bitmaps, so just move along until we find a normal
2932 	 * extent entry.
2933 	 */
2934 	while (entry->bitmap || entry->bytes < min_bytes) {
2935 		if (entry->bitmap && list_empty(&entry->list))
2936 			list_add_tail(&entry->list, bitmaps);
2937 		node = rb_next(&entry->offset_index);
2938 		if (!node)
2939 			return -ENOSPC;
2940 		entry = rb_entry(node, struct btrfs_free_space, offset_index);
2941 	}
2942 
2943 	window_free = entry->bytes;
2944 	max_extent = entry->bytes;
2945 	first = entry;
2946 	last = entry;
2947 
2948 	for (node = rb_next(&entry->offset_index); node;
2949 	     node = rb_next(&entry->offset_index)) {
2950 		entry = rb_entry(node, struct btrfs_free_space, offset_index);
2951 
2952 		if (entry->bitmap) {
2953 			if (list_empty(&entry->list))
2954 				list_add_tail(&entry->list, bitmaps);
2955 			continue;
2956 		}
2957 
2958 		if (entry->bytes < min_bytes)
2959 			continue;
2960 
2961 		last = entry;
2962 		window_free += entry->bytes;
2963 		if (entry->bytes > max_extent)
2964 			max_extent = entry->bytes;
2965 	}
2966 
2967 	if (window_free < bytes || max_extent < cont1_bytes)
2968 		return -ENOSPC;
2969 
2970 	cluster->window_start = first->offset;
2971 
2972 	node = &first->offset_index;
2973 
2974 	/*
2975 	 * now we've found our entries, pull them out of the free space
2976 	 * cache and put them into the cluster rbtree
2977 	 */
2978 	do {
2979 		int ret;
2980 
2981 		entry = rb_entry(node, struct btrfs_free_space, offset_index);
2982 		node = rb_next(&entry->offset_index);
2983 		if (entry->bitmap || entry->bytes < min_bytes)
2984 			continue;
2985 
2986 		rb_erase(&entry->offset_index, &ctl->free_space_offset);
2987 		ret = tree_insert_offset(&cluster->root, entry->offset,
2988 					 &entry->offset_index, 0);
2989 		total_size += entry->bytes;
2990 		ASSERT(!ret); /* -EEXIST; Logic error */
2991 	} while (node && entry != last);
2992 
2993 	cluster->max_size = max_extent;
2994 	trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2995 	return 0;
2996 }
2997 
2998 /*
2999  * This specifically looks for bitmaps that may work in the cluster, we assume
3000  * that we have already failed to find extents that will work.
3001  */
3002 static noinline int
setup_cluster_bitmap(struct btrfs_block_group_cache * block_group,struct btrfs_free_cluster * cluster,struct list_head * bitmaps,u64 offset,u64 bytes,u64 cont1_bytes,u64 min_bytes)3003 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
3004 		     struct btrfs_free_cluster *cluster,
3005 		     struct list_head *bitmaps, u64 offset, u64 bytes,
3006 		     u64 cont1_bytes, u64 min_bytes)
3007 {
3008 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3009 	struct btrfs_free_space *entry = NULL;
3010 	int ret = -ENOSPC;
3011 	u64 bitmap_offset = offset_to_bitmap(ctl, offset);
3012 
3013 	if (ctl->total_bitmaps == 0)
3014 		return -ENOSPC;
3015 
3016 	/*
3017 	 * The bitmap that covers offset won't be in the list unless offset
3018 	 * is just its start offset.
3019 	 */
3020 	if (!list_empty(bitmaps))
3021 		entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
3022 
3023 	if (!entry || entry->offset != bitmap_offset) {
3024 		entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
3025 		if (entry && list_empty(&entry->list))
3026 			list_add(&entry->list, bitmaps);
3027 	}
3028 
3029 	list_for_each_entry(entry, bitmaps, list) {
3030 		if (entry->bytes < bytes)
3031 			continue;
3032 		ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
3033 					   bytes, cont1_bytes, min_bytes);
3034 		if (!ret)
3035 			return 0;
3036 	}
3037 
3038 	/*
3039 	 * The bitmaps list has all the bitmaps that record free space
3040 	 * starting after offset, so no more search is required.
3041 	 */
3042 	return -ENOSPC;
3043 }
3044 
3045 /*
3046  * here we try to find a cluster of blocks in a block group.  The goal
3047  * is to find at least bytes+empty_size.
3048  * We might not find them all in one contiguous area.
3049  *
3050  * returns zero and sets up cluster if things worked out, otherwise
3051  * it returns -enospc
3052  */
btrfs_find_space_cluster(struct btrfs_block_group_cache * block_group,struct btrfs_free_cluster * cluster,u64 offset,u64 bytes,u64 empty_size)3053 int btrfs_find_space_cluster(struct btrfs_block_group_cache *block_group,
3054 			     struct btrfs_free_cluster *cluster,
3055 			     u64 offset, u64 bytes, u64 empty_size)
3056 {
3057 	struct btrfs_fs_info *fs_info = block_group->fs_info;
3058 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3059 	struct btrfs_free_space *entry, *tmp;
3060 	LIST_HEAD(bitmaps);
3061 	u64 min_bytes;
3062 	u64 cont1_bytes;
3063 	int ret;
3064 
3065 	/*
3066 	 * Choose the minimum extent size we'll require for this
3067 	 * cluster.  For SSD_SPREAD, don't allow any fragmentation.
3068 	 * For metadata, allow allocates with smaller extents.  For
3069 	 * data, keep it dense.
3070 	 */
3071 	if (btrfs_test_opt(fs_info, SSD_SPREAD)) {
3072 		cont1_bytes = min_bytes = bytes + empty_size;
3073 	} else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3074 		cont1_bytes = bytes;
3075 		min_bytes = fs_info->sectorsize;
3076 	} else {
3077 		cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3078 		min_bytes = fs_info->sectorsize;
3079 	}
3080 
3081 	spin_lock(&ctl->tree_lock);
3082 
3083 	/*
3084 	 * If we know we don't have enough space to make a cluster don't even
3085 	 * bother doing all the work to try and find one.
3086 	 */
3087 	if (ctl->free_space < bytes) {
3088 		spin_unlock(&ctl->tree_lock);
3089 		return -ENOSPC;
3090 	}
3091 
3092 	spin_lock(&cluster->lock);
3093 
3094 	/* someone already found a cluster, hooray */
3095 	if (cluster->block_group) {
3096 		ret = 0;
3097 		goto out;
3098 	}
3099 
3100 	trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
3101 				 min_bytes);
3102 
3103 	ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
3104 				      bytes + empty_size,
3105 				      cont1_bytes, min_bytes);
3106 	if (ret)
3107 		ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
3108 					   offset, bytes + empty_size,
3109 					   cont1_bytes, min_bytes);
3110 
3111 	/* Clear our temporary list */
3112 	list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3113 		list_del_init(&entry->list);
3114 
3115 	if (!ret) {
3116 		atomic_inc(&block_group->count);
3117 		list_add_tail(&cluster->block_group_list,
3118 			      &block_group->cluster_list);
3119 		cluster->block_group = block_group;
3120 	} else {
3121 		trace_btrfs_failed_cluster_setup(block_group);
3122 	}
3123 out:
3124 	spin_unlock(&cluster->lock);
3125 	spin_unlock(&ctl->tree_lock);
3126 
3127 	return ret;
3128 }
3129 
3130 /*
3131  * simple code to zero out a cluster
3132  */
btrfs_init_free_cluster(struct btrfs_free_cluster * cluster)3133 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3134 {
3135 	spin_lock_init(&cluster->lock);
3136 	spin_lock_init(&cluster->refill_lock);
3137 	cluster->root = RB_ROOT;
3138 	cluster->max_size = 0;
3139 	cluster->fragmented = false;
3140 	INIT_LIST_HEAD(&cluster->block_group_list);
3141 	cluster->block_group = NULL;
3142 }
3143 
do_trimming(struct btrfs_block_group_cache * block_group,u64 * total_trimmed,u64 start,u64 bytes,u64 reserved_start,u64 reserved_bytes,struct btrfs_trim_range * trim_entry)3144 static int do_trimming(struct btrfs_block_group_cache *block_group,
3145 		       u64 *total_trimmed, u64 start, u64 bytes,
3146 		       u64 reserved_start, u64 reserved_bytes,
3147 		       struct btrfs_trim_range *trim_entry)
3148 {
3149 	struct btrfs_space_info *space_info = block_group->space_info;
3150 	struct btrfs_fs_info *fs_info = block_group->fs_info;
3151 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3152 	int ret;
3153 	int update = 0;
3154 	u64 trimmed = 0;
3155 
3156 	spin_lock(&space_info->lock);
3157 	spin_lock(&block_group->lock);
3158 	if (!block_group->ro) {
3159 		block_group->reserved += reserved_bytes;
3160 		space_info->bytes_reserved += reserved_bytes;
3161 		update = 1;
3162 	}
3163 	spin_unlock(&block_group->lock);
3164 	spin_unlock(&space_info->lock);
3165 
3166 	ret = btrfs_discard_extent(fs_info, start, bytes, &trimmed);
3167 	if (!ret)
3168 		*total_trimmed += trimmed;
3169 
3170 	mutex_lock(&ctl->cache_writeout_mutex);
3171 	btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
3172 	list_del(&trim_entry->list);
3173 	mutex_unlock(&ctl->cache_writeout_mutex);
3174 
3175 	if (update) {
3176 		spin_lock(&space_info->lock);
3177 		spin_lock(&block_group->lock);
3178 		if (block_group->ro)
3179 			space_info->bytes_readonly += reserved_bytes;
3180 		block_group->reserved -= reserved_bytes;
3181 		space_info->bytes_reserved -= reserved_bytes;
3182 		spin_unlock(&block_group->lock);
3183 		spin_unlock(&space_info->lock);
3184 	}
3185 
3186 	return ret;
3187 }
3188 
trim_no_bitmap(struct btrfs_block_group_cache * block_group,u64 * total_trimmed,u64 start,u64 end,u64 minlen)3189 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
3190 			  u64 *total_trimmed, u64 start, u64 end, u64 minlen)
3191 {
3192 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3193 	struct btrfs_free_space *entry;
3194 	struct rb_node *node;
3195 	int ret = 0;
3196 	u64 extent_start;
3197 	u64 extent_bytes;
3198 	u64 bytes;
3199 
3200 	while (start < end) {
3201 		struct btrfs_trim_range trim_entry;
3202 
3203 		mutex_lock(&ctl->cache_writeout_mutex);
3204 		spin_lock(&ctl->tree_lock);
3205 
3206 		if (ctl->free_space < minlen) {
3207 			spin_unlock(&ctl->tree_lock);
3208 			mutex_unlock(&ctl->cache_writeout_mutex);
3209 			break;
3210 		}
3211 
3212 		entry = tree_search_offset(ctl, start, 0, 1);
3213 		if (!entry) {
3214 			spin_unlock(&ctl->tree_lock);
3215 			mutex_unlock(&ctl->cache_writeout_mutex);
3216 			break;
3217 		}
3218 
3219 		/* skip bitmaps */
3220 		while (entry->bitmap) {
3221 			node = rb_next(&entry->offset_index);
3222 			if (!node) {
3223 				spin_unlock(&ctl->tree_lock);
3224 				mutex_unlock(&ctl->cache_writeout_mutex);
3225 				goto out;
3226 			}
3227 			entry = rb_entry(node, struct btrfs_free_space,
3228 					 offset_index);
3229 		}
3230 
3231 		if (entry->offset >= end) {
3232 			spin_unlock(&ctl->tree_lock);
3233 			mutex_unlock(&ctl->cache_writeout_mutex);
3234 			break;
3235 		}
3236 
3237 		extent_start = entry->offset;
3238 		extent_bytes = entry->bytes;
3239 		start = max(start, extent_start);
3240 		bytes = min(extent_start + extent_bytes, end) - start;
3241 		if (bytes < minlen) {
3242 			spin_unlock(&ctl->tree_lock);
3243 			mutex_unlock(&ctl->cache_writeout_mutex);
3244 			goto next;
3245 		}
3246 
3247 		unlink_free_space(ctl, entry);
3248 		kmem_cache_free(btrfs_free_space_cachep, entry);
3249 
3250 		spin_unlock(&ctl->tree_lock);
3251 		trim_entry.start = extent_start;
3252 		trim_entry.bytes = extent_bytes;
3253 		list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3254 		mutex_unlock(&ctl->cache_writeout_mutex);
3255 
3256 		ret = do_trimming(block_group, total_trimmed, start, bytes,
3257 				  extent_start, extent_bytes, &trim_entry);
3258 		if (ret)
3259 			break;
3260 next:
3261 		start += bytes;
3262 
3263 		if (fatal_signal_pending(current)) {
3264 			ret = -ERESTARTSYS;
3265 			break;
3266 		}
3267 
3268 		cond_resched();
3269 	}
3270 out:
3271 	return ret;
3272 }
3273 
trim_bitmaps(struct btrfs_block_group_cache * block_group,u64 * total_trimmed,u64 start,u64 end,u64 minlen)3274 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
3275 			u64 *total_trimmed, u64 start, u64 end, u64 minlen)
3276 {
3277 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3278 	struct btrfs_free_space *entry;
3279 	int ret = 0;
3280 	int ret2;
3281 	u64 bytes;
3282 	u64 offset = offset_to_bitmap(ctl, start);
3283 
3284 	while (offset < end) {
3285 		bool next_bitmap = false;
3286 		struct btrfs_trim_range trim_entry;
3287 
3288 		mutex_lock(&ctl->cache_writeout_mutex);
3289 		spin_lock(&ctl->tree_lock);
3290 
3291 		if (ctl->free_space < minlen) {
3292 			spin_unlock(&ctl->tree_lock);
3293 			mutex_unlock(&ctl->cache_writeout_mutex);
3294 			break;
3295 		}
3296 
3297 		entry = tree_search_offset(ctl, offset, 1, 0);
3298 		if (!entry) {
3299 			spin_unlock(&ctl->tree_lock);
3300 			mutex_unlock(&ctl->cache_writeout_mutex);
3301 			next_bitmap = true;
3302 			goto next;
3303 		}
3304 
3305 		bytes = minlen;
3306 		ret2 = search_bitmap(ctl, entry, &start, &bytes, false);
3307 		if (ret2 || start >= end) {
3308 			spin_unlock(&ctl->tree_lock);
3309 			mutex_unlock(&ctl->cache_writeout_mutex);
3310 			next_bitmap = true;
3311 			goto next;
3312 		}
3313 
3314 		bytes = min(bytes, end - start);
3315 		if (bytes < minlen) {
3316 			spin_unlock(&ctl->tree_lock);
3317 			mutex_unlock(&ctl->cache_writeout_mutex);
3318 			goto next;
3319 		}
3320 
3321 		bitmap_clear_bits(ctl, entry, start, bytes);
3322 		if (entry->bytes == 0)
3323 			free_bitmap(ctl, entry);
3324 
3325 		spin_unlock(&ctl->tree_lock);
3326 		trim_entry.start = start;
3327 		trim_entry.bytes = bytes;
3328 		list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3329 		mutex_unlock(&ctl->cache_writeout_mutex);
3330 
3331 		ret = do_trimming(block_group, total_trimmed, start, bytes,
3332 				  start, bytes, &trim_entry);
3333 		if (ret)
3334 			break;
3335 next:
3336 		if (next_bitmap) {
3337 			offset += BITS_PER_BITMAP * ctl->unit;
3338 		} else {
3339 			start += bytes;
3340 			if (start >= offset + BITS_PER_BITMAP * ctl->unit)
3341 				offset += BITS_PER_BITMAP * ctl->unit;
3342 		}
3343 
3344 		if (fatal_signal_pending(current)) {
3345 			ret = -ERESTARTSYS;
3346 			break;
3347 		}
3348 
3349 		cond_resched();
3350 	}
3351 
3352 	return ret;
3353 }
3354 
btrfs_get_block_group_trimming(struct btrfs_block_group_cache * cache)3355 void btrfs_get_block_group_trimming(struct btrfs_block_group_cache *cache)
3356 {
3357 	atomic_inc(&cache->trimming);
3358 }
3359 
btrfs_put_block_group_trimming(struct btrfs_block_group_cache * block_group)3360 void btrfs_put_block_group_trimming(struct btrfs_block_group_cache *block_group)
3361 {
3362 	struct btrfs_fs_info *fs_info = block_group->fs_info;
3363 	struct extent_map_tree *em_tree;
3364 	struct extent_map *em;
3365 	bool cleanup;
3366 
3367 	spin_lock(&block_group->lock);
3368 	cleanup = (atomic_dec_and_test(&block_group->trimming) &&
3369 		   block_group->removed);
3370 	spin_unlock(&block_group->lock);
3371 
3372 	if (cleanup) {
3373 		mutex_lock(&fs_info->chunk_mutex);
3374 		em_tree = &fs_info->mapping_tree;
3375 		write_lock(&em_tree->lock);
3376 		em = lookup_extent_mapping(em_tree, block_group->key.objectid,
3377 					   1);
3378 		BUG_ON(!em); /* logic error, can't happen */
3379 		remove_extent_mapping(em_tree, em);
3380 		write_unlock(&em_tree->lock);
3381 		mutex_unlock(&fs_info->chunk_mutex);
3382 
3383 		/* once for us and once for the tree */
3384 		free_extent_map(em);
3385 		free_extent_map(em);
3386 
3387 		/*
3388 		 * We've left one free space entry and other tasks trimming
3389 		 * this block group have left 1 entry each one. Free them.
3390 		 */
3391 		__btrfs_remove_free_space_cache(block_group->free_space_ctl);
3392 	}
3393 }
3394 
btrfs_trim_block_group(struct btrfs_block_group_cache * block_group,u64 * trimmed,u64 start,u64 end,u64 minlen)3395 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
3396 			   u64 *trimmed, u64 start, u64 end, u64 minlen)
3397 {
3398 	int ret;
3399 
3400 	*trimmed = 0;
3401 
3402 	spin_lock(&block_group->lock);
3403 	if (block_group->removed) {
3404 		spin_unlock(&block_group->lock);
3405 		return 0;
3406 	}
3407 	btrfs_get_block_group_trimming(block_group);
3408 	spin_unlock(&block_group->lock);
3409 
3410 	ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
3411 	if (ret)
3412 		goto out;
3413 
3414 	ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
3415 out:
3416 	btrfs_put_block_group_trimming(block_group);
3417 	return ret;
3418 }
3419 
3420 /*
3421  * Find the left-most item in the cache tree, and then return the
3422  * smallest inode number in the item.
3423  *
3424  * Note: the returned inode number may not be the smallest one in
3425  * the tree, if the left-most item is a bitmap.
3426  */
btrfs_find_ino_for_alloc(struct btrfs_root * fs_root)3427 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
3428 {
3429 	struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
3430 	struct btrfs_free_space *entry = NULL;
3431 	u64 ino = 0;
3432 
3433 	spin_lock(&ctl->tree_lock);
3434 
3435 	if (RB_EMPTY_ROOT(&ctl->free_space_offset))
3436 		goto out;
3437 
3438 	entry = rb_entry(rb_first(&ctl->free_space_offset),
3439 			 struct btrfs_free_space, offset_index);
3440 
3441 	if (!entry->bitmap) {
3442 		ino = entry->offset;
3443 
3444 		unlink_free_space(ctl, entry);
3445 		entry->offset++;
3446 		entry->bytes--;
3447 		if (!entry->bytes)
3448 			kmem_cache_free(btrfs_free_space_cachep, entry);
3449 		else
3450 			link_free_space(ctl, entry);
3451 	} else {
3452 		u64 offset = 0;
3453 		u64 count = 1;
3454 		int ret;
3455 
3456 		ret = search_bitmap(ctl, entry, &offset, &count, true);
3457 		/* Logic error; Should be empty if it can't find anything */
3458 		ASSERT(!ret);
3459 
3460 		ino = offset;
3461 		bitmap_clear_bits(ctl, entry, offset, 1);
3462 		if (entry->bytes == 0)
3463 			free_bitmap(ctl, entry);
3464 	}
3465 out:
3466 	spin_unlock(&ctl->tree_lock);
3467 
3468 	return ino;
3469 }
3470 
lookup_free_ino_inode(struct btrfs_root * root,struct btrfs_path * path)3471 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
3472 				    struct btrfs_path *path)
3473 {
3474 	struct inode *inode = NULL;
3475 
3476 	spin_lock(&root->ino_cache_lock);
3477 	if (root->ino_cache_inode)
3478 		inode = igrab(root->ino_cache_inode);
3479 	spin_unlock(&root->ino_cache_lock);
3480 	if (inode)
3481 		return inode;
3482 
3483 	inode = __lookup_free_space_inode(root, path, 0);
3484 	if (IS_ERR(inode))
3485 		return inode;
3486 
3487 	spin_lock(&root->ino_cache_lock);
3488 	if (!btrfs_fs_closing(root->fs_info))
3489 		root->ino_cache_inode = igrab(inode);
3490 	spin_unlock(&root->ino_cache_lock);
3491 
3492 	return inode;
3493 }
3494 
create_free_ino_inode(struct btrfs_root * root,struct btrfs_trans_handle * trans,struct btrfs_path * path)3495 int create_free_ino_inode(struct btrfs_root *root,
3496 			  struct btrfs_trans_handle *trans,
3497 			  struct btrfs_path *path)
3498 {
3499 	return __create_free_space_inode(root, trans, path,
3500 					 BTRFS_FREE_INO_OBJECTID, 0);
3501 }
3502 
load_free_ino_cache(struct btrfs_fs_info * fs_info,struct btrfs_root * root)3503 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3504 {
3505 	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
3506 	struct btrfs_path *path;
3507 	struct inode *inode;
3508 	int ret = 0;
3509 	u64 root_gen = btrfs_root_generation(&root->root_item);
3510 
3511 	if (!btrfs_test_opt(fs_info, INODE_MAP_CACHE))
3512 		return 0;
3513 
3514 	/*
3515 	 * If we're unmounting then just return, since this does a search on the
3516 	 * normal root and not the commit root and we could deadlock.
3517 	 */
3518 	if (btrfs_fs_closing(fs_info))
3519 		return 0;
3520 
3521 	path = btrfs_alloc_path();
3522 	if (!path)
3523 		return 0;
3524 
3525 	inode = lookup_free_ino_inode(root, path);
3526 	if (IS_ERR(inode))
3527 		goto out;
3528 
3529 	if (root_gen != BTRFS_I(inode)->generation)
3530 		goto out_put;
3531 
3532 	ret = __load_free_space_cache(root, inode, ctl, path, 0);
3533 
3534 	if (ret < 0)
3535 		btrfs_err(fs_info,
3536 			"failed to load free ino cache for root %llu",
3537 			root->root_key.objectid);
3538 out_put:
3539 	iput(inode);
3540 out:
3541 	btrfs_free_path(path);
3542 	return ret;
3543 }
3544 
btrfs_write_out_ino_cache(struct btrfs_root * root,struct btrfs_trans_handle * trans,struct btrfs_path * path,struct inode * inode)3545 int btrfs_write_out_ino_cache(struct btrfs_root *root,
3546 			      struct btrfs_trans_handle *trans,
3547 			      struct btrfs_path *path,
3548 			      struct inode *inode)
3549 {
3550 	struct btrfs_fs_info *fs_info = root->fs_info;
3551 	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
3552 	int ret;
3553 	struct btrfs_io_ctl io_ctl;
3554 	bool release_metadata = true;
3555 
3556 	if (!btrfs_test_opt(fs_info, INODE_MAP_CACHE))
3557 		return 0;
3558 
3559 	memset(&io_ctl, 0, sizeof(io_ctl));
3560 	ret = __btrfs_write_out_cache(root, inode, ctl, NULL, &io_ctl, trans);
3561 	if (!ret) {
3562 		/*
3563 		 * At this point writepages() didn't error out, so our metadata
3564 		 * reservation is released when the writeback finishes, at
3565 		 * inode.c:btrfs_finish_ordered_io(), regardless of it finishing
3566 		 * with or without an error.
3567 		 */
3568 		release_metadata = false;
3569 		ret = btrfs_wait_cache_io_root(root, trans, &io_ctl, path);
3570 	}
3571 
3572 	if (ret) {
3573 		if (release_metadata)
3574 			btrfs_delalloc_release_metadata(BTRFS_I(inode),
3575 					inode->i_size, true);
3576 #ifdef DEBUG
3577 		btrfs_err(fs_info,
3578 			  "failed to write free ino cache for root %llu",
3579 			  root->root_key.objectid);
3580 #endif
3581 	}
3582 
3583 	return ret;
3584 }
3585 
3586 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3587 /*
3588  * Use this if you need to make a bitmap or extent entry specifically, it
3589  * doesn't do any of the merging that add_free_space does, this acts a lot like
3590  * how the free space cache loading stuff works, so you can get really weird
3591  * configurations.
3592  */
test_add_free_space_entry(struct btrfs_block_group_cache * cache,u64 offset,u64 bytes,bool bitmap)3593 int test_add_free_space_entry(struct btrfs_block_group_cache *cache,
3594 			      u64 offset, u64 bytes, bool bitmap)
3595 {
3596 	struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3597 	struct btrfs_free_space *info = NULL, *bitmap_info;
3598 	void *map = NULL;
3599 	u64 bytes_added;
3600 	int ret;
3601 
3602 again:
3603 	if (!info) {
3604 		info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
3605 		if (!info)
3606 			return -ENOMEM;
3607 	}
3608 
3609 	if (!bitmap) {
3610 		spin_lock(&ctl->tree_lock);
3611 		info->offset = offset;
3612 		info->bytes = bytes;
3613 		info->max_extent_size = 0;
3614 		ret = link_free_space(ctl, info);
3615 		spin_unlock(&ctl->tree_lock);
3616 		if (ret)
3617 			kmem_cache_free(btrfs_free_space_cachep, info);
3618 		return ret;
3619 	}
3620 
3621 	if (!map) {
3622 		map = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep, GFP_NOFS);
3623 		if (!map) {
3624 			kmem_cache_free(btrfs_free_space_cachep, info);
3625 			return -ENOMEM;
3626 		}
3627 	}
3628 
3629 	spin_lock(&ctl->tree_lock);
3630 	bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3631 					 1, 0);
3632 	if (!bitmap_info) {
3633 		info->bitmap = map;
3634 		map = NULL;
3635 		add_new_bitmap(ctl, info, offset);
3636 		bitmap_info = info;
3637 		info = NULL;
3638 	}
3639 
3640 	bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
3641 
3642 	bytes -= bytes_added;
3643 	offset += bytes_added;
3644 	spin_unlock(&ctl->tree_lock);
3645 
3646 	if (bytes)
3647 		goto again;
3648 
3649 	if (info)
3650 		kmem_cache_free(btrfs_free_space_cachep, info);
3651 	if (map)
3652 		kmem_cache_free(btrfs_free_space_bitmap_cachep, map);
3653 	return 0;
3654 }
3655 
3656 /*
3657  * Checks to see if the given range is in the free space cache.  This is really
3658  * just used to check the absence of space, so if there is free space in the
3659  * range at all we will return 1.
3660  */
test_check_exists(struct btrfs_block_group_cache * cache,u64 offset,u64 bytes)3661 int test_check_exists(struct btrfs_block_group_cache *cache,
3662 		      u64 offset, u64 bytes)
3663 {
3664 	struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3665 	struct btrfs_free_space *info;
3666 	int ret = 0;
3667 
3668 	spin_lock(&ctl->tree_lock);
3669 	info = tree_search_offset(ctl, offset, 0, 0);
3670 	if (!info) {
3671 		info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3672 					  1, 0);
3673 		if (!info)
3674 			goto out;
3675 	}
3676 
3677 have_info:
3678 	if (info->bitmap) {
3679 		u64 bit_off, bit_bytes;
3680 		struct rb_node *n;
3681 		struct btrfs_free_space *tmp;
3682 
3683 		bit_off = offset;
3684 		bit_bytes = ctl->unit;
3685 		ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false);
3686 		if (!ret) {
3687 			if (bit_off == offset) {
3688 				ret = 1;
3689 				goto out;
3690 			} else if (bit_off > offset &&
3691 				   offset + bytes > bit_off) {
3692 				ret = 1;
3693 				goto out;
3694 			}
3695 		}
3696 
3697 		n = rb_prev(&info->offset_index);
3698 		while (n) {
3699 			tmp = rb_entry(n, struct btrfs_free_space,
3700 				       offset_index);
3701 			if (tmp->offset + tmp->bytes < offset)
3702 				break;
3703 			if (offset + bytes < tmp->offset) {
3704 				n = rb_prev(&tmp->offset_index);
3705 				continue;
3706 			}
3707 			info = tmp;
3708 			goto have_info;
3709 		}
3710 
3711 		n = rb_next(&info->offset_index);
3712 		while (n) {
3713 			tmp = rb_entry(n, struct btrfs_free_space,
3714 				       offset_index);
3715 			if (offset + bytes < tmp->offset)
3716 				break;
3717 			if (tmp->offset + tmp->bytes < offset) {
3718 				n = rb_next(&tmp->offset_index);
3719 				continue;
3720 			}
3721 			info = tmp;
3722 			goto have_info;
3723 		}
3724 
3725 		ret = 0;
3726 		goto out;
3727 	}
3728 
3729 	if (info->offset == offset) {
3730 		ret = 1;
3731 		goto out;
3732 	}
3733 
3734 	if (offset > info->offset && offset < info->offset + info->bytes)
3735 		ret = 1;
3736 out:
3737 	spin_unlock(&ctl->tree_lock);
3738 	return ret;
3739 }
3740 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */
3741