1	// SPDX-License-Identifier: GPL-2.0
2	/*
3	* Copyright (C) 2008 Red Hat. All rights reserved.
4	*/
5
6	#include <linux/pagemap.h>
7	#include <linux/sched.h>
8	#include <linux/sched/signal.h>
9	#include <linux/slab.h>
10	#include <linux/math64.h>
11	#include <linux/ratelimit.h>
12	#include <linux/error-injection.h>
13	#include <linux/sched/mm.h>
14	#include "ctree.h"
15	#include "fs.h"
16	#include "messages.h"
17	#include "misc.h"
18	#include "free-space-cache.h"
19	#include "transaction.h"
20	#include "disk-io.h"
21	#include "extent_io.h"
22	#include "space-info.h"
23	#include "block-group.h"
24	#include "discard.h"
25	#include "subpage.h"
26	#include "inode-item.h"
27	#include "accessors.h"
28	#include "file-item.h"
29	#include "file.h"
30	#include "super.h"
31
32	#define BITS_PER_BITMAP (PAGE_SIZE * 8UL)
33	#define MAX_CACHE_BYTES_PER_GIG SZ_64K
34	#define FORCE_EXTENT_THRESHOLD SZ_1M
35
36	static struct kmem_cache *btrfs_free_space_cachep;
37	static struct kmem_cache *btrfs_free_space_bitmap_cachep;
38
39	struct btrfs_trim_range {
40	u64 start;
41	u64 bytes;
42	struct list_head list;
43	};
44
45	static int link_free_space(struct btrfs_free_space_ctl *ctl,
46	struct btrfs_free_space *info);
47	static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
48	struct btrfs_free_space *info, bool update_stat);
49	static int search_bitmap(struct btrfs_free_space_ctl *ctl,
50	struct btrfs_free_space *bitmap_info, u64 *offset,
51	u64 *bytes, bool for_alloc);
52	static void free_bitmap(struct btrfs_free_space_ctl *ctl,
53	struct btrfs_free_space *bitmap_info);
54	static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
55	struct btrfs_free_space *info, u64 offset,
56	u64 bytes, bool update_stats);
57
58	static void btrfs_crc32c_final(u32 crc, u8 *result)
59	{
60	put_unaligned_le32(val: ~crc, p: result);
61	}
62
63	static void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
64	{
65	struct btrfs_free_space *info;
66	struct rb_node *node;
67
68	while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
69	info = rb_entry(node, struct btrfs_free_space, offset_index);
70	if (!info->bitmap) {
71	unlink_free_space(ctl, info, update_stat: true);
72	kmem_cache_free(s: btrfs_free_space_cachep, objp: info);
73	} else {
74	free_bitmap(ctl, bitmap_info: info);
75	}
76
77	cond_resched_lock(&ctl->tree_lock);
78	}
79	}
80
81	static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
82	struct btrfs_path *path,
83	u64 offset)
84	{
85	struct btrfs_fs_info *fs_info = root->fs_info;
86	struct btrfs_key key;
87	struct btrfs_key location;
88	struct btrfs_disk_key disk_key;
89	struct btrfs_free_space_header *header;
90	struct extent_buffer *leaf;
91	struct inode *inode = NULL;
92	unsigned nofs_flag;
93	int ret;
94
95	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
96	key.offset = offset;
97	key.type = 0;
98
99	ret = btrfs_search_slot(NULL, root, key: &key, p: path, ins_len: 0, cow: 0);
100	if (ret < 0)
101	return ERR_PTR(error: ret);
102	if (ret > 0) {
103	btrfs_release_path(p: path);
104	return ERR_PTR(error: -ENOENT);
105	}
106
107	leaf = path->nodes[0];
108	header = btrfs_item_ptr(leaf, path->slots[0],
109	struct btrfs_free_space_header);
110	btrfs_free_space_key(eb: leaf, h: header, key: &disk_key);
111	btrfs_disk_key_to_cpu(cpu_key: &location, disk_key: &disk_key);
112	btrfs_release_path(p: path);
113
114	/*
115	* We are often under a trans handle at this point, so we need to make
116	* sure NOFS is set to keep us from deadlocking.
117	*/
118	nofs_flag = memalloc_nofs_save();
119	inode = btrfs_iget_path(s: fs_info->sb, ino: location.objectid, root, path);
120	btrfs_release_path(p: path);
121	memalloc_nofs_restore(flags: nofs_flag);
122	if (IS_ERR(ptr: inode))
123	return inode;
124
125	mapping_set_gfp_mask(m: inode->i_mapping,
126	mask: mapping_gfp_constraint(mapping: inode->i_mapping,
127	gfp_mask: ~(__GFP_FS | __GFP_HIGHMEM)));
128
129	return inode;
130	}
131
132	struct inode *lookup_free_space_inode(struct btrfs_block_group *block_group,
133	struct btrfs_path *path)
134	{
135	struct btrfs_fs_info *fs_info = block_group->fs_info;
136	struct inode *inode = NULL;
137	u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
138
139	spin_lock(lock: &block_group->lock);
140	if (block_group->inode)
141	inode = igrab(block_group->inode);
142	spin_unlock(lock: &block_group->lock);
143	if (inode)
144	return inode;
145
146	inode = __lookup_free_space_inode(root: fs_info->tree_root, path,
147	offset: block_group->start);
148	if (IS_ERR(ptr: inode))
149	return inode;
150
151	spin_lock(lock: &block_group->lock);
152	if (!((BTRFS_I(inode)->flags & flags) == flags)) {
153	btrfs_info(fs_info, "Old style space inode found, converting.");
154	BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
155	BTRFS_INODE_NODATACOW;
156	block_group->disk_cache_state = BTRFS_DC_CLEAR;
157	}
158
159	if (!test_and_set_bit(nr: BLOCK_GROUP_FLAG_IREF, addr: &block_group->runtime_flags))
160	block_group->inode = igrab(inode);
161	spin_unlock(lock: &block_group->lock);
162
163	return inode;
164	}
165
166	static int __create_free_space_inode(struct btrfs_root *root,
167	struct btrfs_trans_handle *trans,
168	struct btrfs_path *path,
169	u64 ino, u64 offset)
170	{
171	struct btrfs_key key;
172	struct btrfs_disk_key disk_key;
173	struct btrfs_free_space_header *header;
174	struct btrfs_inode_item *inode_item;
175	struct extent_buffer *leaf;
176	/* We inline CRCs for the free disk space cache */
177	const u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC |
178	BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
179	int ret;
180
181	ret = btrfs_insert_empty_inode(trans, root, path, objectid: ino);
182	if (ret)
183	return ret;
184
185	leaf = path->nodes[0];
186	inode_item = btrfs_item_ptr(leaf, path->slots[0],
187	struct btrfs_inode_item);
188	btrfs_item_key(eb: leaf, disk_key: &disk_key, nr: path->slots[0]);
189	memzero_extent_buffer(eb: leaf, start: (unsigned long)inode_item,
190	len: sizeof(*inode_item));
191	btrfs_set_inode_generation(eb: leaf, s: inode_item, val: trans->transid);
192	btrfs_set_inode_size(eb: leaf, s: inode_item, val: 0);
193	btrfs_set_inode_nbytes(eb: leaf, s: inode_item, val: 0);
194	btrfs_set_inode_uid(eb: leaf, s: inode_item, val: 0);
195	btrfs_set_inode_gid(eb: leaf, s: inode_item, val: 0);
196	btrfs_set_inode_mode(eb: leaf, s: inode_item, S_IFREG | 0600);
197	btrfs_set_inode_flags(eb: leaf, s: inode_item, val: flags);
198	btrfs_set_inode_nlink(eb: leaf, s: inode_item, val: 1);
199	btrfs_set_inode_transid(eb: leaf, s: inode_item, val: trans->transid);
200	btrfs_set_inode_block_group(eb: leaf, s: inode_item, val: offset);
201	btrfs_mark_buffer_dirty(trans, buf: leaf);
202	btrfs_release_path(p: path);
203
204	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
205	key.offset = offset;
206	key.type = 0;
207	ret = btrfs_insert_empty_item(trans, root, path, key: &key,
208	data_size: sizeof(struct btrfs_free_space_header));
209	if (ret < 0) {
210	btrfs_release_path(p: path);
211	return ret;
212	}
213
214	leaf = path->nodes[0];
215	header = btrfs_item_ptr(leaf, path->slots[0],
216	struct btrfs_free_space_header);
217	memzero_extent_buffer(eb: leaf, start: (unsigned long)header, len: sizeof(*header));
218	btrfs_set_free_space_key(eb: leaf, h: header, key: &disk_key);
219	btrfs_mark_buffer_dirty(trans, buf: leaf);
220	btrfs_release_path(p: path);
221
222	return 0;
223	}
224
225	int create_free_space_inode(struct btrfs_trans_handle *trans,
226	struct btrfs_block_group *block_group,
227	struct btrfs_path *path)
228	{
229	int ret;
230	u64 ino;
231
232	ret = btrfs_get_free_objectid(root: trans->fs_info->tree_root, objectid: &ino);
233	if (ret < 0)
234	return ret;
235
236	return __create_free_space_inode(root: trans->fs_info->tree_root, trans, path,
237	ino, offset: block_group->start);
238	}
239
240	/*
241	* inode is an optional sink: if it is NULL, btrfs_remove_free_space_inode
242	* handles lookup, otherwise it takes ownership and iputs the inode.
243	* Don't reuse an inode pointer after passing it into this function.
244	*/
245	int btrfs_remove_free_space_inode(struct btrfs_trans_handle *trans,
246	struct inode *inode,
247	struct btrfs_block_group *block_group)
248	{
249	struct btrfs_path *path;
250	struct btrfs_key key;
251	int ret = 0;
252
253	path = btrfs_alloc_path();
254	if (!path)
255	return -ENOMEM;
256
257	if (!inode)
258	inode = lookup_free_space_inode(block_group, path);
259	if (IS_ERR(ptr: inode)) {
260	if (PTR_ERR(ptr: inode) != -ENOENT)
261	ret = PTR_ERR(ptr: inode);
262	goto out;
263	}
264	ret = btrfs_orphan_add(trans, inode: BTRFS_I(inode));
265	if (ret) {
266	btrfs_add_delayed_iput(inode: BTRFS_I(inode));
267	goto out;
268	}
269	clear_nlink(inode);
270	/* One for the block groups ref */
271	spin_lock(lock: &block_group->lock);
272	if (test_and_clear_bit(nr: BLOCK_GROUP_FLAG_IREF, addr: &block_group->runtime_flags)) {
273	block_group->inode = NULL;
274	spin_unlock(lock: &block_group->lock);
275	iput(inode);
276	} else {
277	spin_unlock(lock: &block_group->lock);
278	}
279	/* One for the lookup ref */
280	btrfs_add_delayed_iput(inode: BTRFS_I(inode));
281
282	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
283	key.type = 0;
284	key.offset = block_group->start;
285	ret = btrfs_search_slot(trans, root: trans->fs_info->tree_root, key: &key, p: path,
286	ins_len: -1, cow: 1);
287	if (ret) {
288	if (ret > 0)
289	ret = 0;
290	goto out;
291	}
292	ret = btrfs_del_item(trans, root: trans->fs_info->tree_root, path);
293	out:
294	btrfs_free_path(p: path);
295	return ret;
296	}
297
298	int btrfs_truncate_free_space_cache(struct btrfs_trans_handle *trans,
299	struct btrfs_block_group *block_group,
300	struct inode *vfs_inode)
301	{
302	struct btrfs_truncate_control control = {
303	.inode = BTRFS_I(inode: vfs_inode),
304	.new_size = 0,
305	.ino = btrfs_ino(inode: BTRFS_I(inode: vfs_inode)),
306	.min_type = BTRFS_EXTENT_DATA_KEY,
307	.clear_extent_range = true,
308	};
309	struct btrfs_inode *inode = BTRFS_I(inode: vfs_inode);
310	struct btrfs_root *root = inode->root;
311	struct extent_state *cached_state = NULL;
312	int ret = 0;
313	bool locked = false;
314
315	if (block_group) {
316	struct btrfs_path *path = btrfs_alloc_path();
317
318	if (!path) {
319	ret = -ENOMEM;
320	goto fail;
321	}
322	locked = true;
323	mutex_lock(&trans->transaction->cache_write_mutex);
324	if (!list_empty(head: &block_group->io_list)) {
325	list_del_init(entry: &block_group->io_list);
326
327	btrfs_wait_cache_io(trans, block_group, path);
328	btrfs_put_block_group(cache: block_group);
329	}
330
331	/*
332	* now that we've truncated the cache away, its no longer
333	* setup or written
334	*/
335	spin_lock(lock: &block_group->lock);
336	block_group->disk_cache_state = BTRFS_DC_CLEAR;
337	spin_unlock(lock: &block_group->lock);
338	btrfs_free_path(p: path);
339	}
340
341	btrfs_i_size_write(inode, size: 0);
342	truncate_pagecache(inode: vfs_inode, new: 0);
343
344	lock_extent(tree: &inode->io_tree, start: 0, end: (u64)-1, cached: &cached_state);
345	btrfs_drop_extent_map_range(inode, start: 0, end: (u64)-1, skip_pinned: false);
346
347	/*
348	* We skip the throttling logic for free space cache inodes, so we don't
349	* need to check for -EAGAIN.
350	*/
351	ret = btrfs_truncate_inode_items(trans, root, control: &control);
352
353	inode_sub_bytes(inode: &inode->vfs_inode, bytes: control.sub_bytes);
354	btrfs_inode_safe_disk_i_size_write(inode, new_i_size: control.last_size);
355
356	unlock_extent(tree: &inode->io_tree, start: 0, end: (u64)-1, cached: &cached_state);
357	if (ret)
358	goto fail;
359
360	ret = btrfs_update_inode(trans, inode);
361
362	fail:
363	if (locked)
364	mutex_unlock(lock: &trans->transaction->cache_write_mutex);
365	if (ret)
366	btrfs_abort_transaction(trans, ret);
367
368	return ret;
369	}
370
371	static void readahead_cache(struct inode *inode)
372	{
373	struct file_ra_state ra;
374	unsigned long last_index;
375
376	file_ra_state_init(ra: &ra, mapping: inode->i_mapping);
377	last_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
378
379	page_cache_sync_readahead(mapping: inode->i_mapping, ra: &ra, NULL, index: 0, req_count: last_index);
380	}
381
382	static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode,
383	int write)
384	{
385	int num_pages;
386
387	num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
388
389	/* Make sure we can fit our crcs and generation into the first page */
390	if (write && (num_pages * sizeof(u32) + sizeof(u64)) > PAGE_SIZE)
391	return -ENOSPC;
392
393	memset(io_ctl, 0, sizeof(struct btrfs_io_ctl));
394
395	io_ctl->pages = kcalloc(n: num_pages, size: sizeof(struct page *), GFP_NOFS);
396	if (!io_ctl->pages)
397	return -ENOMEM;
398
399	io_ctl->num_pages = num_pages;
400	io_ctl->fs_info = inode_to_fs_info(inode);
401	io_ctl->inode = inode;
402
403	return 0;
404	}
405	ALLOW_ERROR_INJECTION(io_ctl_init, ERRNO);
406
407	static void io_ctl_free(struct btrfs_io_ctl *io_ctl)
408	{
409	kfree(objp: io_ctl->pages);
410	io_ctl->pages = NULL;
411	}
412
413	static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl)
414	{
415	if (io_ctl->cur) {
416	io_ctl->cur = NULL;
417	io_ctl->orig = NULL;
418	}
419	}
420
421	static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear)
422	{
423	ASSERT(io_ctl->index < io_ctl->num_pages);
424	io_ctl->page = io_ctl->pages[io_ctl->index++];
425	io_ctl->cur = page_address(io_ctl->page);
426	io_ctl->orig = io_ctl->cur;
427	io_ctl->size = PAGE_SIZE;
428	if (clear)
429	clear_page(page: io_ctl->cur);
430	}
431
432	static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl)
433	{
434	int i;
435
436	io_ctl_unmap_page(io_ctl);
437
438	for (i = 0; i < io_ctl->num_pages; i++) {
439	if (io_ctl->pages[i]) {
440	btrfs_folio_clear_checked(fs_info: io_ctl->fs_info,
441	page_folio(io_ctl->pages[i]),
442	start: page_offset(page: io_ctl->pages[i]),
443	PAGE_SIZE);
444	unlock_page(page: io_ctl->pages[i]);
445	put_page(page: io_ctl->pages[i]);
446	}
447	}
448	}
449
450	static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, bool uptodate)
451	{
452	struct page *page;
453	struct inode *inode = io_ctl->inode;
454	gfp_t mask = btrfs_alloc_write_mask(mapping: inode->i_mapping);
455	int i;
456
457	for (i = 0; i < io_ctl->num_pages; i++) {
458	int ret;
459
460	page = find_or_create_page(mapping: inode->i_mapping, index: i, gfp_mask: mask);
461	if (!page) {
462	io_ctl_drop_pages(io_ctl);
463	return -ENOMEM;
464	}
465
466	ret = set_page_extent_mapped(page);
467	if (ret < 0) {
468	unlock_page(page);
469	put_page(page);
470	io_ctl_drop_pages(io_ctl);
471	return ret;
472	}
473
474	io_ctl->pages[i] = page;
475	if (uptodate && !PageUptodate(page)) {
476	btrfs_read_folio(NULL, page_folio(page));
477	lock_page(page);
478	if (page->mapping != inode->i_mapping) {
479	btrfs_err(BTRFS_I(inode)->root->fs_info,
480	"free space cache page truncated");
481	io_ctl_drop_pages(io_ctl);
482	return -EIO;
483	}
484	if (!PageUptodate(page)) {
485	btrfs_err(BTRFS_I(inode)->root->fs_info,
486	"error reading free space cache");
487	io_ctl_drop_pages(io_ctl);
488	return -EIO;
489	}
490	}
491	}
492
493	for (i = 0; i < io_ctl->num_pages; i++)
494	clear_page_dirty_for_io(page: io_ctl->pages[i]);
495
496	return 0;
497	}
498
499	static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
500	{
501	io_ctl_map_page(io_ctl, clear: 1);
502
503	/*
504	* Skip the csum areas. If we don't check crcs then we just have a
505	* 64bit chunk at the front of the first page.
506	*/
507	io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
508	io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
509
510	put_unaligned_le64(val: generation, p: io_ctl->cur);
511	io_ctl->cur += sizeof(u64);
512	}
513
514	static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
515	{
516	u64 cache_gen;
517
518	/*
519	* Skip the crc area. If we don't check crcs then we just have a 64bit
520	* chunk at the front of the first page.
521	*/
522	io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
523	io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
524
525	cache_gen = get_unaligned_le64(p: io_ctl->cur);
526	if (cache_gen != generation) {
527	btrfs_err_rl(io_ctl->fs_info,
528	"space cache generation (%llu) does not match inode (%llu)",
529	cache_gen, generation);
530	io_ctl_unmap_page(io_ctl);
531	return -EIO;
532	}
533	io_ctl->cur += sizeof(u64);
534	return 0;
535	}
536
537	static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
538	{
539	u32 *tmp;
540	u32 crc = ~(u32)0;
541	unsigned offset = 0;
542
543	if (index == 0)
544	offset = sizeof(u32) * io_ctl->num_pages;
545
546	crc = crc32c(crc, address: io_ctl->orig + offset, PAGE_SIZE - offset);
547	btrfs_crc32c_final(crc, result: (u8 *)&crc);
548	io_ctl_unmap_page(io_ctl);
549	tmp = page_address(io_ctl->pages[0]);
550	tmp += index;
551	*tmp = crc;
552	}
553
554	static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
555	{
556	u32 *tmp, val;
557	u32 crc = ~(u32)0;
558	unsigned offset = 0;
559
560	if (index == 0)
561	offset = sizeof(u32) * io_ctl->num_pages;
562
563	tmp = page_address(io_ctl->pages[0]);
564	tmp += index;
565	val = *tmp;
566
567	io_ctl_map_page(io_ctl, clear: 0);
568	crc = crc32c(crc, address: io_ctl->orig + offset, PAGE_SIZE - offset);
569	btrfs_crc32c_final(crc, result: (u8 *)&crc);
570	if (val != crc) {
571	btrfs_err_rl(io_ctl->fs_info,
572	"csum mismatch on free space cache");
573	io_ctl_unmap_page(io_ctl);
574	return -EIO;
575	}
576
577	return 0;
578	}
579
580	static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
581	void *bitmap)
582	{
583	struct btrfs_free_space_entry *entry;
584
585	if (!io_ctl->cur)
586	return -ENOSPC;
587
588	entry = io_ctl->cur;
589	put_unaligned_le64(val: offset, p: &entry->offset);
590	put_unaligned_le64(val: bytes, p: &entry->bytes);
591	entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
592	BTRFS_FREE_SPACE_EXTENT;
593	io_ctl->cur += sizeof(struct btrfs_free_space_entry);
594	io_ctl->size -= sizeof(struct btrfs_free_space_entry);
595
596	if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
597	return 0;
598
599	io_ctl_set_crc(io_ctl, index: io_ctl->index - 1);
600
601	/* No more pages to map */
602	if (io_ctl->index >= io_ctl->num_pages)
603	return 0;
604
605	/* map the next page */
606	io_ctl_map_page(io_ctl, clear: 1);
607	return 0;
608	}
609
610	static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
611	{
612	if (!io_ctl->cur)
613	return -ENOSPC;
614
615	/*
616	* If we aren't at the start of the current page, unmap this one and
617	* map the next one if there is any left.
618	*/
619	if (io_ctl->cur != io_ctl->orig) {
620	io_ctl_set_crc(io_ctl, index: io_ctl->index - 1);
621	if (io_ctl->index >= io_ctl->num_pages)
622	return -ENOSPC;
623	io_ctl_map_page(io_ctl, clear: 0);
624	}
625
626	copy_page(to: io_ctl->cur, from: bitmap);
627	io_ctl_set_crc(io_ctl, index: io_ctl->index - 1);
628	if (io_ctl->index < io_ctl->num_pages)
629	io_ctl_map_page(io_ctl, clear: 0);
630	return 0;
631	}
632
633	static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
634	{
635	/*
636	* If we're not on the boundary we know we've modified the page and we
637	* need to crc the page.
638	*/
639	if (io_ctl->cur != io_ctl->orig)
640	io_ctl_set_crc(io_ctl, index: io_ctl->index - 1);
641	else
642	io_ctl_unmap_page(io_ctl);
643
644	while (io_ctl->index < io_ctl->num_pages) {
645	io_ctl_map_page(io_ctl, clear: 1);
646	io_ctl_set_crc(io_ctl, index: io_ctl->index - 1);
647	}
648	}
649
650	static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
651	struct btrfs_free_space *entry, u8 *type)
652	{
653	struct btrfs_free_space_entry *e;
654	int ret;
655
656	if (!io_ctl->cur) {
657	ret = io_ctl_check_crc(io_ctl, index: io_ctl->index);
658	if (ret)
659	return ret;
660	}
661
662	e = io_ctl->cur;
663	entry->offset = get_unaligned_le64(p: &e->offset);
664	entry->bytes = get_unaligned_le64(p: &e->bytes);
665	*type = e->type;
666	io_ctl->cur += sizeof(struct btrfs_free_space_entry);
667	io_ctl->size -= sizeof(struct btrfs_free_space_entry);
668
669	if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
670	return 0;
671
672	io_ctl_unmap_page(io_ctl);
673
674	return 0;
675	}
676
677	static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
678	struct btrfs_free_space *entry)
679	{
680	int ret;
681
682	ret = io_ctl_check_crc(io_ctl, index: io_ctl->index);
683	if (ret)
684	return ret;
685
686	copy_page(to: entry->bitmap, from: io_ctl->cur);
687	io_ctl_unmap_page(io_ctl);
688
689	return 0;
690	}
691
692	static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
693	{
694	struct btrfs_block_group *block_group = ctl->block_group;
695	u64 max_bytes;
696	u64 bitmap_bytes;
697	u64 extent_bytes;
698	u64 size = block_group->length;
699	u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
700	u64 max_bitmaps = div64_u64(dividend: size + bytes_per_bg - 1, divisor: bytes_per_bg);
701
702	max_bitmaps = max_t(u64, max_bitmaps, 1);
703
704	if (ctl->total_bitmaps > max_bitmaps)
705	btrfs_err(block_group->fs_info,
706	"invalid free space control: bg start=%llu len=%llu total_bitmaps=%u unit=%u max_bitmaps=%llu bytes_per_bg=%llu",
707	block_group->start, block_group->length,
708	ctl->total_bitmaps, ctl->unit, max_bitmaps,
709	bytes_per_bg);
710	ASSERT(ctl->total_bitmaps <= max_bitmaps);
711
712	/*
713	* We are trying to keep the total amount of memory used per 1GiB of
714	* space to be MAX_CACHE_BYTES_PER_GIG. However, with a reclamation
715	* mechanism of pulling extents >= FORCE_EXTENT_THRESHOLD out of
716	* bitmaps, we may end up using more memory than this.
717	*/
718	if (size < SZ_1G)
719	max_bytes = MAX_CACHE_BYTES_PER_GIG;
720	else
721	max_bytes = MAX_CACHE_BYTES_PER_GIG * div_u64(dividend: size, SZ_1G);
722
723	bitmap_bytes = ctl->total_bitmaps * ctl->unit;
724
725	/*
726	* we want the extent entry threshold to always be at most 1/2 the max
727	* bytes we can have, or whatever is less than that.
728	*/
729	extent_bytes = max_bytes - bitmap_bytes;
730	extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
731
732	ctl->extents_thresh =
733	div_u64(dividend: extent_bytes, divisor: sizeof(struct btrfs_free_space));
734	}
735
736	static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
737	struct btrfs_free_space_ctl *ctl,
738	struct btrfs_path *path, u64 offset)
739	{
740	struct btrfs_fs_info *fs_info = root->fs_info;
741	struct btrfs_free_space_header *header;
742	struct extent_buffer *leaf;
743	struct btrfs_io_ctl io_ctl;
744	struct btrfs_key key;
745	struct btrfs_free_space *e, *n;
746	LIST_HEAD(bitmaps);
747	u64 num_entries;
748	u64 num_bitmaps;
749	u64 generation;
750	u8 type;
751	int ret = 0;
752
753	/* Nothing in the space cache, goodbye */
754	if (!i_size_read(inode))
755	return 0;
756
757	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
758	key.offset = offset;
759	key.type = 0;
760
761	ret = btrfs_search_slot(NULL, root, key: &key, p: path, ins_len: 0, cow: 0);
762	if (ret < 0)
763	return 0;
764	else if (ret > 0) {
765	btrfs_release_path(p: path);
766	return 0;
767	}
768
769	ret = -1;
770
771	leaf = path->nodes[0];
772	header = btrfs_item_ptr(leaf, path->slots[0],
773	struct btrfs_free_space_header);
774	num_entries = btrfs_free_space_entries(eb: leaf, s: header);
775	num_bitmaps = btrfs_free_space_bitmaps(eb: leaf, s: header);
776	generation = btrfs_free_space_generation(eb: leaf, s: header);
777	btrfs_release_path(p: path);
778
779	if (!BTRFS_I(inode)->generation) {
780	btrfs_info(fs_info,
781	"the free space cache file (%llu) is invalid, skip it",
782	offset);
783	return 0;
784	}
785
786	if (BTRFS_I(inode)->generation != generation) {
787	btrfs_err(fs_info,
788	"free space inode generation (%llu) did not match free space cache generation (%llu)",
789	BTRFS_I(inode)->generation, generation);
790	return 0;
791	}
792
793	if (!num_entries)
794	return 0;
795
796	ret = io_ctl_init(io_ctl: &io_ctl, inode, write: 0);
797	if (ret)
798	return ret;
799
800	readahead_cache(inode);
801
802	ret = io_ctl_prepare_pages(io_ctl: &io_ctl, uptodate: true);
803	if (ret)
804	goto out;
805
806	ret = io_ctl_check_crc(io_ctl: &io_ctl, index: 0);
807	if (ret)
808	goto free_cache;
809
810	ret = io_ctl_check_generation(io_ctl: &io_ctl, generation);
811	if (ret)
812	goto free_cache;
813
814	while (num_entries) {
815	e = kmem_cache_zalloc(k: btrfs_free_space_cachep,
816	GFP_NOFS);
817	if (!e) {
818	ret = -ENOMEM;
819	goto free_cache;
820	}
821
822	ret = io_ctl_read_entry(io_ctl: &io_ctl, entry: e, type: &type);
823	if (ret) {
824	kmem_cache_free(s: btrfs_free_space_cachep, objp: e);
825	goto free_cache;
826	}
827
828	if (!e->bytes) {
829	ret = -1;
830	kmem_cache_free(s: btrfs_free_space_cachep, objp: e);
831	goto free_cache;
832	}
833
834	if (type == BTRFS_FREE_SPACE_EXTENT) {
835	spin_lock(lock: &ctl->tree_lock);
836	ret = link_free_space(ctl, info: e);
837	spin_unlock(lock: &ctl->tree_lock);
838	if (ret) {
839	btrfs_err(fs_info,
840	"Duplicate entries in free space cache, dumping");
841	kmem_cache_free(s: btrfs_free_space_cachep, objp: e);
842	goto free_cache;
843	}
844	} else {
845	ASSERT(num_bitmaps);
846	num_bitmaps--;
847	e->bitmap = kmem_cache_zalloc(
848	k: btrfs_free_space_bitmap_cachep, GFP_NOFS);
849	if (!e->bitmap) {
850	ret = -ENOMEM;
851	kmem_cache_free(
852	s: btrfs_free_space_cachep, objp: e);
853	goto free_cache;
854	}
855	spin_lock(lock: &ctl->tree_lock);
856	ret = link_free_space(ctl, info: e);
857	if (ret) {
858	spin_unlock(lock: &ctl->tree_lock);
859	btrfs_err(fs_info,
860	"Duplicate entries in free space cache, dumping");
861	kmem_cache_free(s: btrfs_free_space_cachep, objp: e);
862	goto free_cache;
863	}
864	ctl->total_bitmaps++;
865	recalculate_thresholds(ctl);
866	spin_unlock(lock: &ctl->tree_lock);
867	list_add_tail(new: &e->list, head: &bitmaps);
868	}
869
870	num_entries--;
871	}
872
873	io_ctl_unmap_page(io_ctl: &io_ctl);
874
875	/*
876	* We add the bitmaps at the end of the entries in order that
877	* the bitmap entries are added to the cache.
878	*/
879	list_for_each_entry_safe(e, n, &bitmaps, list) {
880	list_del_init(entry: &e->list);
881	ret = io_ctl_read_bitmap(io_ctl: &io_ctl, entry: e);
882	if (ret)
883	goto free_cache;
884	}
885
886	io_ctl_drop_pages(io_ctl: &io_ctl);
887	ret = 1;
888	out:
889	io_ctl_free(io_ctl: &io_ctl);
890	return ret;
891	free_cache:
892	io_ctl_drop_pages(io_ctl: &io_ctl);
893
894	spin_lock(lock: &ctl->tree_lock);
895	__btrfs_remove_free_space_cache(ctl);
896	spin_unlock(lock: &ctl->tree_lock);
897	goto out;
898	}
899
900	static int copy_free_space_cache(struct btrfs_block_group *block_group,
901	struct btrfs_free_space_ctl *ctl)
902	{
903	struct btrfs_free_space *info;
904	struct rb_node *n;
905	int ret = 0;
906
907	while (!ret && (n = rb_first(&ctl->free_space_offset)) != NULL) {
908	info = rb_entry(n, struct btrfs_free_space, offset_index);
909	if (!info->bitmap) {
910	const u64 offset = info->offset;
911	const u64 bytes = info->bytes;
912
913	unlink_free_space(ctl, info, update_stat: true);
914	spin_unlock(lock: &ctl->tree_lock);
915	kmem_cache_free(s: btrfs_free_space_cachep, objp: info);
916	ret = btrfs_add_free_space(block_group, bytenr: offset, size: bytes);
917	spin_lock(lock: &ctl->tree_lock);
918	} else {
919	u64 offset = info->offset;
920	u64 bytes = ctl->unit;
921
922	ret = search_bitmap(ctl, bitmap_info: info, offset: &offset, bytes: &bytes, for_alloc: false);
923	if (ret == 0) {
924	bitmap_clear_bits(ctl, info, offset, bytes, update_stats: true);
925	spin_unlock(lock: &ctl->tree_lock);
926	ret = btrfs_add_free_space(block_group, bytenr: offset,
927	size: bytes);
928	spin_lock(lock: &ctl->tree_lock);
929	} else {
930	free_bitmap(ctl, bitmap_info: info);
931	ret = 0;
932	}
933	}
934	cond_resched_lock(&ctl->tree_lock);
935	}
936	return ret;
937	}
938
939	static struct lock_class_key btrfs_free_space_inode_key;
940
941	int load_free_space_cache(struct btrfs_block_group *block_group)
942	{
943	struct btrfs_fs_info *fs_info = block_group->fs_info;
944	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
945	struct btrfs_free_space_ctl tmp_ctl = {};
946	struct inode *inode;
947	struct btrfs_path *path;
948	int ret = 0;
949	bool matched;
950	u64 used = block_group->used;
951
952	/*
953	* Because we could potentially discard our loaded free space, we want
954	* to load everything into a temporary structure first, and then if it's
955	* valid copy it all into the actual free space ctl.
956	*/
957	btrfs_init_free_space_ctl(block_group, ctl: &tmp_ctl);
958
959	/*
960	* If this block group has been marked to be cleared for one reason or
961	* another then we can't trust the on disk cache, so just return.
962	*/
963	spin_lock(lock: &block_group->lock);
964	if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
965	spin_unlock(lock: &block_group->lock);
966	return 0;
967	}
968	spin_unlock(lock: &block_group->lock);
969
970	path = btrfs_alloc_path();
971	if (!path)
972	return 0;
973	path->search_commit_root = 1;
974	path->skip_locking = 1;
975
976	/*
977	* We must pass a path with search_commit_root set to btrfs_iget in
978	* order to avoid a deadlock when allocating extents for the tree root.
979	*
980	* When we are COWing an extent buffer from the tree root, when looking
981	* for a free extent, at extent-tree.c:find_free_extent(), we can find
982	* block group without its free space cache loaded. When we find one
983	* we must load its space cache which requires reading its free space
984	* cache's inode item from the root tree. If this inode item is located
985	* in the same leaf that we started COWing before, then we end up in
986	* deadlock on the extent buffer (trying to read lock it when we
987	* previously write locked it).
988	*
989	* It's safe to read the inode item using the commit root because
990	* block groups, once loaded, stay in memory forever (until they are
991	* removed) as well as their space caches once loaded. New block groups
992	* once created get their ->cached field set to BTRFS_CACHE_FINISHED so
993	* we will never try to read their inode item while the fs is mounted.
994	*/
995	inode = lookup_free_space_inode(block_group, path);
996	if (IS_ERR(ptr: inode)) {
997	btrfs_free_path(p: path);
998	return 0;
999	}
1000
1001	/* We may have converted the inode and made the cache invalid. */
1002	spin_lock(lock: &block_group->lock);
1003	if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
1004	spin_unlock(lock: &block_group->lock);
1005	btrfs_free_path(p: path);
1006	goto out;
1007	}
1008	spin_unlock(lock: &block_group->lock);
1009
1010	/*
1011	* Reinitialize the class of struct inode's mapping->invalidate_lock for
1012	* free space inodes to prevent false positives related to locks for normal
1013	* inodes.
1014	*/
1015	lockdep_set_class(&(&inode->i_data)->invalidate_lock,
1016	&btrfs_free_space_inode_key);
1017
1018	ret = __load_free_space_cache(root: fs_info->tree_root, inode, ctl: &tmp_ctl,
1019	path, offset: block_group->start);
1020	btrfs_free_path(p: path);
1021	if (ret <= 0)
1022	goto out;
1023
1024	matched = (tmp_ctl.free_space == (block_group->length - used -
1025	block_group->bytes_super));
1026
1027	if (matched) {
1028	spin_lock(lock: &tmp_ctl.tree_lock);
1029	ret = copy_free_space_cache(block_group, ctl: &tmp_ctl);
1030	spin_unlock(lock: &tmp_ctl.tree_lock);
1031	/*
1032	* ret == 1 means we successfully loaded the free space cache,
1033	* so we need to re-set it here.
1034	*/
1035	if (ret == 0)
1036	ret = 1;
1037	} else {
1038	/*
1039	* We need to call the _locked variant so we don't try to update
1040	* the discard counters.
1041	*/
1042	spin_lock(lock: &tmp_ctl.tree_lock);
1043	__btrfs_remove_free_space_cache(ctl: &tmp_ctl);
1044	spin_unlock(lock: &tmp_ctl.tree_lock);
1045	btrfs_warn(fs_info,
1046	"block group %llu has wrong amount of free space",
1047	block_group->start);
1048	ret = -1;
1049	}
1050	out:
1051	if (ret < 0) {
1052	/* This cache is bogus, make sure it gets cleared */
1053	spin_lock(lock: &block_group->lock);
1054	block_group->disk_cache_state = BTRFS_DC_CLEAR;
1055	spin_unlock(lock: &block_group->lock);
1056	ret = 0;
1057
1058	btrfs_warn(fs_info,
1059	"failed to load free space cache for block group %llu, rebuilding it now",
1060	block_group->start);
1061	}
1062
1063	spin_lock(lock: &ctl->tree_lock);
1064	btrfs_discard_update_discardable(block_group);
1065	spin_unlock(lock: &ctl->tree_lock);
1066	iput(inode);
1067	return ret;
1068	}
1069
1070	static noinline_for_stack
1071	int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
1072	struct btrfs_free_space_ctl *ctl,
1073	struct btrfs_block_group *block_group,
1074	int *entries, int *bitmaps,
1075	struct list_head *bitmap_list)
1076	{
1077	int ret;
1078	struct btrfs_free_cluster *cluster = NULL;
1079	struct btrfs_free_cluster *cluster_locked = NULL;
1080	struct rb_node *node = rb_first(&ctl->free_space_offset);
1081	struct btrfs_trim_range *trim_entry;
1082
1083	/* Get the cluster for this block_group if it exists */
1084	if (block_group && !list_empty(head: &block_group->cluster_list)) {
1085	cluster = list_entry(block_group->cluster_list.next,
1086	struct btrfs_free_cluster,
1087	block_group_list);
1088	}
1089
1090	if (!node && cluster) {
1091	cluster_locked = cluster;
1092	spin_lock(lock: &cluster_locked->lock);
1093	node = rb_first(&cluster->root);
1094	cluster = NULL;
1095	}
1096
1097	/* Write out the extent entries */
1098	while (node) {
1099	struct btrfs_free_space *e;
1100
1101	e = rb_entry(node, struct btrfs_free_space, offset_index);
1102	*entries += 1;
1103
1104	ret = io_ctl_add_entry(io_ctl, offset: e->offset, bytes: e->bytes,
1105	bitmap: e->bitmap);
1106	if (ret)
1107	goto fail;
1108
1109	if (e->bitmap) {
1110	list_add_tail(new: &e->list, head: bitmap_list);
1111	*bitmaps += 1;
1112	}
1113	node = rb_next(node);
1114	if (!node && cluster) {
1115	node = rb_first(&cluster->root);
1116	cluster_locked = cluster;
1117	spin_lock(lock: &cluster_locked->lock);
1118	cluster = NULL;
1119	}
1120	}
1121	if (cluster_locked) {
1122	spin_unlock(lock: &cluster_locked->lock);
1123	cluster_locked = NULL;
1124	}
1125
1126	/*
1127	* Make sure we don't miss any range that was removed from our rbtree
1128	* because trimming is running. Otherwise after a umount+mount (or crash
1129	* after committing the transaction) we would leak free space and get
1130	* an inconsistent free space cache report from fsck.
1131	*/
1132	list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
1133	ret = io_ctl_add_entry(io_ctl, offset: trim_entry->start,
1134	bytes: trim_entry->bytes, NULL);
1135	if (ret)
1136	goto fail;
1137	*entries += 1;
1138	}
1139
1140	return 0;
1141	fail:
1142	if (cluster_locked)
1143	spin_unlock(lock: &cluster_locked->lock);
1144	return -ENOSPC;
1145	}
1146
1147	static noinline_for_stack int
1148	update_cache_item(struct btrfs_trans_handle *trans,
1149	struct btrfs_root *root,
1150	struct inode *inode,
1151	struct btrfs_path *path, u64 offset,
1152	int entries, int bitmaps)
1153	{
1154	struct btrfs_key key;
1155	struct btrfs_free_space_header *header;
1156	struct extent_buffer *leaf;
1157	int ret;
1158
1159	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1160	key.offset = offset;
1161	key.type = 0;
1162
1163	ret = btrfs_search_slot(trans, root, key: &key, p: path, ins_len: 0, cow: 1);
1164	if (ret < 0) {
1165	clear_extent_bit(tree: &BTRFS_I(inode)->io_tree, start: 0, end: inode->i_size - 1,
1166	bits: EXTENT_DELALLOC, NULL);
1167	goto fail;
1168	}
1169	leaf = path->nodes[0];
1170	if (ret > 0) {
1171	struct btrfs_key found_key;
1172	ASSERT(path->slots[0]);
1173	path->slots[0]--;
1174	btrfs_item_key_to_cpu(eb: leaf, cpu_key: &found_key, nr: path->slots[0]);
1175	if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1176	found_key.offset != offset) {
1177	clear_extent_bit(tree: &BTRFS_I(inode)->io_tree, start: 0,
1178	end: inode->i_size - 1, bits: EXTENT_DELALLOC,
1179	NULL);
1180	btrfs_release_path(p: path);
1181	goto fail;
1182	}
1183	}
1184
1185	BTRFS_I(inode)->generation = trans->transid;
1186	header = btrfs_item_ptr(leaf, path->slots[0],
1187	struct btrfs_free_space_header);
1188	btrfs_set_free_space_entries(eb: leaf, s: header, val: entries);
1189	btrfs_set_free_space_bitmaps(eb: leaf, s: header, val: bitmaps);
1190	btrfs_set_free_space_generation(eb: leaf, s: header, val: trans->transid);
1191	btrfs_mark_buffer_dirty(trans, buf: leaf);
1192	btrfs_release_path(p: path);
1193
1194	return 0;
1195
1196	fail:
1197	return -1;
1198	}
1199
1200	static noinline_for_stack int write_pinned_extent_entries(
1201	struct btrfs_trans_handle *trans,
1202	struct btrfs_block_group *block_group,
1203	struct btrfs_io_ctl *io_ctl,
1204	int *entries)
1205	{
1206	u64 start, extent_start, extent_end, len;
1207	struct extent_io_tree *unpin = NULL;
1208	int ret;
1209
1210	if (!block_group)
1211	return 0;
1212
1213	/*
1214	* We want to add any pinned extents to our free space cache
1215	* so we don't leak the space
1216	*
1217	* We shouldn't have switched the pinned extents yet so this is the
1218	* right one
1219	*/
1220	unpin = &trans->transaction->pinned_extents;
1221
1222	start = block_group->start;
1223
1224	while (start < block_group->start + block_group->length) {
1225	if (!find_first_extent_bit(tree: unpin, start,
1226	start_ret: &extent_start, end_ret: &extent_end,
1227	bits: EXTENT_DIRTY, NULL))
1228	return 0;
1229
1230	/* This pinned extent is out of our range */
1231	if (extent_start >= block_group->start + block_group->length)
1232	return 0;
1233
1234	extent_start = max(extent_start, start);
1235	extent_end = min(block_group->start + block_group->length,
1236	extent_end + 1);
1237	len = extent_end - extent_start;
1238
1239	*entries += 1;
1240	ret = io_ctl_add_entry(io_ctl, offset: extent_start, bytes: len, NULL);
1241	if (ret)
1242	return -ENOSPC;
1243
1244	start = extent_end;
1245	}
1246
1247	return 0;
1248	}
1249
1250	static noinline_for_stack int
1251	write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1252	{
1253	struct btrfs_free_space *entry, *next;
1254	int ret;
1255
1256	/* Write out the bitmaps */
1257	list_for_each_entry_safe(entry, next, bitmap_list, list) {
1258	ret = io_ctl_add_bitmap(io_ctl, bitmap: entry->bitmap);
1259	if (ret)
1260	return -ENOSPC;
1261	list_del_init(entry: &entry->list);
1262	}
1263
1264	return 0;
1265	}
1266
1267	static int flush_dirty_cache(struct inode *inode)
1268	{
1269	int ret;
1270
1271	ret = btrfs_wait_ordered_range(inode, start: 0, len: (u64)-1);
1272	if (ret)
1273	clear_extent_bit(tree: &BTRFS_I(inode)->io_tree, start: 0, end: inode->i_size - 1,
1274	bits: EXTENT_DELALLOC, NULL);
1275
1276	return ret;
1277	}
1278
1279	static void noinline_for_stack
1280	cleanup_bitmap_list(struct list_head *bitmap_list)
1281	{
1282	struct btrfs_free_space *entry, *next;
1283
1284	list_for_each_entry_safe(entry, next, bitmap_list, list)
1285	list_del_init(entry: &entry->list);
1286	}
1287
1288	static void noinline_for_stack
1289	cleanup_write_cache_enospc(struct inode *inode,
1290	struct btrfs_io_ctl *io_ctl,
1291	struct extent_state **cached_state)
1292	{
1293	io_ctl_drop_pages(io_ctl);
1294	unlock_extent(tree: &BTRFS_I(inode)->io_tree, start: 0, end: i_size_read(inode) - 1,
1295	cached: cached_state);
1296	}
1297
1298	static int __btrfs_wait_cache_io(struct btrfs_root *root,
1299	struct btrfs_trans_handle *trans,
1300	struct btrfs_block_group *block_group,
1301	struct btrfs_io_ctl *io_ctl,
1302	struct btrfs_path *path, u64 offset)
1303	{
1304	int ret;
1305	struct inode *inode = io_ctl->inode;
1306
1307	if (!inode)
1308	return 0;
1309
1310	/* Flush the dirty pages in the cache file. */
1311	ret = flush_dirty_cache(inode);
1312	if (ret)
1313	goto out;
1314
1315	/* Update the cache item to tell everyone this cache file is valid. */
1316	ret = update_cache_item(trans, root, inode, path, offset,
1317	entries: io_ctl->entries, bitmaps: io_ctl->bitmaps);
1318	out:
1319	if (ret) {
1320	invalidate_inode_pages2(mapping: inode->i_mapping);
1321	BTRFS_I(inode)->generation = 0;
1322	if (block_group)
1323	btrfs_debug(root->fs_info,
1324	"failed to write free space cache for block group %llu error %d",
1325	block_group->start, ret);
1326	}
1327	btrfs_update_inode(trans, inode: BTRFS_I(inode));
1328
1329	if (block_group) {
1330	/* the dirty list is protected by the dirty_bgs_lock */
1331	spin_lock(lock: &trans->transaction->dirty_bgs_lock);
1332
1333	/* the disk_cache_state is protected by the block group lock */
1334	spin_lock(lock: &block_group->lock);
1335
1336	/*
1337	* only mark this as written if we didn't get put back on
1338	* the dirty list while waiting for IO. Otherwise our
1339	* cache state won't be right, and we won't get written again
1340	*/
1341	if (!ret && list_empty(head: &block_group->dirty_list))
1342	block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1343	else if (ret)
1344	block_group->disk_cache_state = BTRFS_DC_ERROR;
1345
1346	spin_unlock(lock: &block_group->lock);
1347	spin_unlock(lock: &trans->transaction->dirty_bgs_lock);
1348	io_ctl->inode = NULL;
1349	iput(inode);
1350	}
1351
1352	return ret;
1353
1354	}
1355
1356	int btrfs_wait_cache_io(struct btrfs_trans_handle *trans,
1357	struct btrfs_block_group *block_group,
1358	struct btrfs_path *path)
1359	{
1360	return __btrfs_wait_cache_io(root: block_group->fs_info->tree_root, trans,
1361	block_group, io_ctl: &block_group->io_ctl,
1362	path, offset: block_group->start);
1363	}
1364
1365	/*
1366	* Write out cached info to an inode.
1367	*
1368	* @inode: freespace inode we are writing out
1369	* @ctl: free space cache we are going to write out
1370	* @block_group: block_group for this cache if it belongs to a block_group
1371	* @io_ctl: holds context for the io
1372	* @trans: the trans handle
1373	*
1374	* This function writes out a free space cache struct to disk for quick recovery
1375	* on mount. This will return 0 if it was successful in writing the cache out,
1376	* or an errno if it was not.
1377	*/
1378	static int __btrfs_write_out_cache(struct inode *inode,
1379	struct btrfs_free_space_ctl *ctl,
1380	struct btrfs_block_group *block_group,
1381	struct btrfs_io_ctl *io_ctl,
1382	struct btrfs_trans_handle *trans)
1383	{
1384	struct extent_state *cached_state = NULL;
1385	LIST_HEAD(bitmap_list);
1386	int entries = 0;
1387	int bitmaps = 0;
1388	int ret;
1389	int must_iput = 0;
1390
1391	if (!i_size_read(inode))
1392	return -EIO;
1393
1394	WARN_ON(io_ctl->pages);
1395	ret = io_ctl_init(io_ctl, inode, write: 1);
1396	if (ret)
1397	return ret;
1398
1399	if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1400	down_write(sem: &block_group->data_rwsem);
1401	spin_lock(lock: &block_group->lock);
1402	if (block_group->delalloc_bytes) {
1403	block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1404	spin_unlock(lock: &block_group->lock);
1405	up_write(sem: &block_group->data_rwsem);
1406	BTRFS_I(inode)->generation = 0;
1407	ret = 0;
1408	must_iput = 1;
1409	goto out;
1410	}
1411	spin_unlock(lock: &block_group->lock);
1412	}
1413
1414	/* Lock all pages first so we can lock the extent safely. */
1415	ret = io_ctl_prepare_pages(io_ctl, uptodate: false);
1416	if (ret)
1417	goto out_unlock;
1418
1419	lock_extent(tree: &BTRFS_I(inode)->io_tree, start: 0, end: i_size_read(inode) - 1,
1420	cached: &cached_state);
1421
1422	io_ctl_set_generation(io_ctl, generation: trans->transid);
1423
1424	mutex_lock(&ctl->cache_writeout_mutex);
1425	/* Write out the extent entries in the free space cache */
1426	spin_lock(lock: &ctl->tree_lock);
1427	ret = write_cache_extent_entries(io_ctl, ctl,
1428	block_group, entries: &entries, bitmaps: &bitmaps,
1429	bitmap_list: &bitmap_list);
1430	if (ret)
1431	goto out_nospc_locked;
1432
1433	/*
1434	* Some spaces that are freed in the current transaction are pinned,
1435	* they will be added into free space cache after the transaction is
1436	* committed, we shouldn't lose them.
1437	*
1438	* If this changes while we are working we'll get added back to
1439	* the dirty list and redo it. No locking needed
1440	*/
1441	ret = write_pinned_extent_entries(trans, block_group, io_ctl, entries: &entries);
1442	if (ret)
1443	goto out_nospc_locked;
1444
1445	/*
1446	* At last, we write out all the bitmaps and keep cache_writeout_mutex
1447	* locked while doing it because a concurrent trim can be manipulating
1448	* or freeing the bitmap.
1449	*/
1450	ret = write_bitmap_entries(io_ctl, bitmap_list: &bitmap_list);
1451	spin_unlock(lock: &ctl->tree_lock);
1452	mutex_unlock(lock: &ctl->cache_writeout_mutex);
1453	if (ret)
1454	goto out_nospc;
1455
1456	/* Zero out the rest of the pages just to make sure */
1457	io_ctl_zero_remaining_pages(io_ctl);
1458
1459	/* Everything is written out, now we dirty the pages in the file. */
1460	ret = btrfs_dirty_pages(inode: BTRFS_I(inode), pages: io_ctl->pages,
1461	num_pages: io_ctl->num_pages, pos: 0, write_bytes: i_size_read(inode),
1462	cached: &cached_state, noreserve: false);
1463	if (ret)
1464	goto out_nospc;
1465
1466	if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1467	up_write(sem: &block_group->data_rwsem);
1468	/*
1469	* Release the pages and unlock the extent, we will flush
1470	* them out later
1471	*/
1472	io_ctl_drop_pages(io_ctl);
1473	io_ctl_free(io_ctl);
1474
1475	unlock_extent(tree: &BTRFS_I(inode)->io_tree, start: 0, end: i_size_read(inode) - 1,
1476	cached: &cached_state);
1477
1478	/*
1479	* at this point the pages are under IO and we're happy,
1480	* The caller is responsible for waiting on them and updating
1481	* the cache and the inode
1482	*/
1483	io_ctl->entries = entries;
1484	io_ctl->bitmaps = bitmaps;
1485
1486	ret = btrfs_fdatawrite_range(inode, start: 0, end: (u64)-1);
1487	if (ret)
1488	goto out;
1489
1490	return 0;
1491
1492	out_nospc_locked:
1493	cleanup_bitmap_list(bitmap_list: &bitmap_list);
1494	spin_unlock(lock: &ctl->tree_lock);
1495	mutex_unlock(lock: &ctl->cache_writeout_mutex);
1496
1497	out_nospc:
1498	cleanup_write_cache_enospc(inode, io_ctl, cached_state: &cached_state);
1499
1500	out_unlock:
1501	if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1502	up_write(sem: &block_group->data_rwsem);
1503
1504	out:
1505	io_ctl->inode = NULL;
1506	io_ctl_free(io_ctl);
1507	if (ret) {
1508	invalidate_inode_pages2(mapping: inode->i_mapping);
1509	BTRFS_I(inode)->generation = 0;
1510	}
1511	btrfs_update_inode(trans, inode: BTRFS_I(inode));
1512	if (must_iput)
1513	iput(inode);
1514	return ret;
1515	}
1516
1517	int btrfs_write_out_cache(struct btrfs_trans_handle *trans,
1518	struct btrfs_block_group *block_group,
1519	struct btrfs_path *path)
1520	{
1521	struct btrfs_fs_info *fs_info = trans->fs_info;
1522	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1523	struct inode *inode;
1524	int ret = 0;
1525
1526	spin_lock(lock: &block_group->lock);
1527	if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1528	spin_unlock(lock: &block_group->lock);
1529	return 0;
1530	}
1531	spin_unlock(lock: &block_group->lock);
1532
1533	inode = lookup_free_space_inode(block_group, path);
1534	if (IS_ERR(ptr: inode))
1535	return 0;
1536
1537	ret = __btrfs_write_out_cache(inode, ctl, block_group,
1538	io_ctl: &block_group->io_ctl, trans);
1539	if (ret) {
1540	btrfs_debug(fs_info,
1541	"failed to write free space cache for block group %llu error %d",
1542	block_group->start, ret);
1543	spin_lock(lock: &block_group->lock);
1544	block_group->disk_cache_state = BTRFS_DC_ERROR;
1545	spin_unlock(lock: &block_group->lock);
1546
1547	block_group->io_ctl.inode = NULL;
1548	iput(inode);
1549	}
1550
1551	/*
1552	* if ret == 0 the caller is expected to call btrfs_wait_cache_io
1553	* to wait for IO and put the inode
1554	*/
1555
1556	return ret;
1557	}
1558
1559	static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1560	u64 offset)
1561	{
1562	ASSERT(offset >= bitmap_start);
1563	offset -= bitmap_start;
1564	return (unsigned long)(div_u64(dividend: offset, divisor: unit));
1565	}
1566
1567	static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1568	{
1569	return (unsigned long)(div_u64(dividend: bytes, divisor: unit));
1570	}
1571
1572	static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1573	u64 offset)
1574	{
1575	u64 bitmap_start;
1576	u64 bytes_per_bitmap;
1577
1578	bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1579	bitmap_start = offset - ctl->start;
1580	bitmap_start = div64_u64(dividend: bitmap_start, divisor: bytes_per_bitmap);
1581	bitmap_start *= bytes_per_bitmap;
1582	bitmap_start += ctl->start;
1583
1584	return bitmap_start;
1585	}
1586
1587	static int tree_insert_offset(struct btrfs_free_space_ctl *ctl,
1588	struct btrfs_free_cluster *cluster,
1589	struct btrfs_free_space *new_entry)
1590	{
1591	struct rb_root *root;
1592	struct rb_node **p;
1593	struct rb_node *parent = NULL;
1594
1595	lockdep_assert_held(&ctl->tree_lock);
1596
1597	if (cluster) {
1598	lockdep_assert_held(&cluster->lock);
1599	root = &cluster->root;
1600	} else {
1601	root = &ctl->free_space_offset;
1602	}
1603
1604	p = &root->rb_node;
1605
1606	while (*p) {
1607	struct btrfs_free_space *info;
1608
1609	parent = *p;
1610	info = rb_entry(parent, struct btrfs_free_space, offset_index);
1611
1612	if (new_entry->offset < info->offset) {
1613	p = &(*p)->rb_left;
1614	} else if (new_entry->offset > info->offset) {
1615	p = &(*p)->rb_right;
1616	} else {
1617	/*
1618	* we could have a bitmap entry and an extent entry
1619	* share the same offset. If this is the case, we want
1620	* the extent entry to always be found first if we do a
1621	* linear search through the tree, since we want to have
1622	* the quickest allocation time, and allocating from an
1623	* extent is faster than allocating from a bitmap. So
1624	* if we're inserting a bitmap and we find an entry at
1625	* this offset, we want to go right, or after this entry
1626	* logically. If we are inserting an extent and we've
1627	* found a bitmap, we want to go left, or before
1628	* logically.
1629	*/
1630	if (new_entry->bitmap) {
1631	if (info->bitmap) {
1632	WARN_ON_ONCE(1);
1633	return -EEXIST;
1634	}
1635	p = &(*p)->rb_right;
1636	} else {
1637	if (!info->bitmap) {
1638	WARN_ON_ONCE(1);
1639	return -EEXIST;
1640	}
1641	p = &(*p)->rb_left;
1642	}
1643	}
1644	}
1645
1646	rb_link_node(node: &new_entry->offset_index, parent, rb_link: p);
1647	rb_insert_color(&new_entry->offset_index, root);
1648
1649	return 0;
1650	}
1651
1652	/*
1653	* This is a little subtle. We *only* have ->max_extent_size set if we actually
1654	* searched through the bitmap and figured out the largest ->max_extent_size,
1655	* otherwise it's 0. In the case that it's 0 we don't want to tell the
1656	* allocator the wrong thing, we want to use the actual real max_extent_size
1657	* we've found already if it's larger, or we want to use ->bytes.
1658	*
1659	* This matters because find_free_space() will skip entries who's ->bytes is
1660	* less than the required bytes. So if we didn't search down this bitmap, we
1661	* may pick some previous entry that has a smaller ->max_extent_size than we
1662	* have. For example, assume we have two entries, one that has
1663	* ->max_extent_size set to 4K and ->bytes set to 1M. A second entry hasn't set
1664	* ->max_extent_size yet, has ->bytes set to 8K and it's contiguous. We will
1665	* call into find_free_space(), and return with max_extent_size == 4K, because
1666	* that first bitmap entry had ->max_extent_size set, but the second one did
1667	* not. If instead we returned 8K we'd come in searching for 8K, and find the
1668	* 8K contiguous range.
1669	*
1670	* Consider the other case, we have 2 8K chunks in that second entry and still
1671	* don't have ->max_extent_size set. We'll return 16K, and the next time the
1672	* allocator comes in it'll fully search our second bitmap, and this time it'll
1673	* get an uptodate value of 8K as the maximum chunk size. Then we'll get the
1674	* right allocation the next loop through.
1675	*/
1676	static inline u64 get_max_extent_size(const struct btrfs_free_space *entry)
1677	{
1678	if (entry->bitmap && entry->max_extent_size)
1679	return entry->max_extent_size;
1680	return entry->bytes;
1681	}
1682
1683	/*
1684	* We want the largest entry to be leftmost, so this is inverted from what you'd
1685	* normally expect.
1686	*/
1687	static bool entry_less(struct rb_node *node, const struct rb_node *parent)
1688	{
1689	const struct btrfs_free_space *entry, *exist;
1690
1691	entry = rb_entry(node, struct btrfs_free_space, bytes_index);
1692	exist = rb_entry(parent, struct btrfs_free_space, bytes_index);
1693	return get_max_extent_size(entry: exist) < get_max_extent_size(entry);
1694	}
1695
1696	/*
1697	* searches the tree for the given offset.
1698	*
1699	* fuzzy - If this is set, then we are trying to make an allocation, and we just
1700	* want a section that has at least bytes size and comes at or after the given
1701	* offset.
1702	*/
1703	static struct btrfs_free_space *
1704	tree_search_offset(struct btrfs_free_space_ctl *ctl,
1705	u64 offset, int bitmap_only, int fuzzy)
1706	{
1707	struct rb_node *n = ctl->free_space_offset.rb_node;
1708	struct btrfs_free_space *entry = NULL, *prev = NULL;
1709
1710	lockdep_assert_held(&ctl->tree_lock);
1711
1712	/* find entry that is closest to the 'offset' */
1713	while (n) {
1714	entry = rb_entry(n, struct btrfs_free_space, offset_index);
1715	prev = entry;
1716
1717	if (offset < entry->offset)
1718	n = n->rb_left;
1719	else if (offset > entry->offset)
1720	n = n->rb_right;
1721	else
1722	break;
1723
1724	entry = NULL;
1725	}
1726
1727	if (bitmap_only) {
1728	if (!entry)
1729	return NULL;
1730	if (entry->bitmap)
1731	return entry;
1732
1733	/*
1734	* bitmap entry and extent entry may share same offset,
1735	* in that case, bitmap entry comes after extent entry.
1736	*/
1737	n = rb_next(n);
1738	if (!n)
1739	return NULL;
1740	entry = rb_entry(n, struct btrfs_free_space, offset_index);
1741	if (entry->offset != offset)
1742	return NULL;
1743
1744	WARN_ON(!entry->bitmap);
1745	return entry;
1746	} else if (entry) {
1747	if (entry->bitmap) {
1748	/*
1749	* if previous extent entry covers the offset,
1750	* we should return it instead of the bitmap entry
1751	*/
1752	n = rb_prev(&entry->offset_index);
1753	if (n) {
1754	prev = rb_entry(n, struct btrfs_free_space,
1755	offset_index);
1756	if (!prev->bitmap &&
1757	prev->offset + prev->bytes > offset)
1758	entry = prev;
1759	}
1760	}
1761	return entry;
1762	}
1763
1764	if (!prev)
1765	return NULL;
1766
1767	/* find last entry before the 'offset' */
1768	entry = prev;
1769	if (entry->offset > offset) {
1770	n = rb_prev(&entry->offset_index);
1771	if (n) {
1772	entry = rb_entry(n, struct btrfs_free_space,
1773	offset_index);
1774	ASSERT(entry->offset <= offset);
1775	} else {
1776	if (fuzzy)
1777	return entry;
1778	else
1779	return NULL;
1780	}
1781	}
1782
1783	if (entry->bitmap) {
1784	n = rb_prev(&entry->offset_index);
1785	if (n) {
1786	prev = rb_entry(n, struct btrfs_free_space,
1787	offset_index);
1788	if (!prev->bitmap &&
1789	prev->offset + prev->bytes > offset)
1790	return prev;
1791	}
1792	if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1793	return entry;
1794	} else if (entry->offset + entry->bytes > offset)
1795	return entry;
1796
1797	if (!fuzzy)
1798	return NULL;
1799
1800	while (1) {
1801	n = rb_next(&entry->offset_index);
1802	if (!n)
1803	return NULL;
1804	entry = rb_entry(n, struct btrfs_free_space, offset_index);
1805	if (entry->bitmap) {
1806	if (entry->offset + BITS_PER_BITMAP *
1807	ctl->unit > offset)
1808	break;
1809	} else {
1810	if (entry->offset + entry->bytes > offset)
1811	break;
1812	}
1813	}
1814	return entry;
1815	}
1816
1817	static inline void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1818	struct btrfs_free_space *info,
1819	bool update_stat)
1820	{
1821	lockdep_assert_held(&ctl->tree_lock);
1822
1823	rb_erase(&info->offset_index, &ctl->free_space_offset);
1824	rb_erase_cached(node: &info->bytes_index, root: &ctl->free_space_bytes);
1825	ctl->free_extents--;
1826
1827	if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1828	ctl->discardable_extents[BTRFS_STAT_CURR]--;
1829	ctl->discardable_bytes[BTRFS_STAT_CURR] -= info->bytes;
1830	}
1831
1832	if (update_stat)
1833	ctl->free_space -= info->bytes;
1834	}
1835
1836	static int link_free_space(struct btrfs_free_space_ctl *ctl,
1837	struct btrfs_free_space *info)
1838	{
1839	int ret = 0;
1840
1841	lockdep_assert_held(&ctl->tree_lock);
1842
1843	ASSERT(info->bytes || info->bitmap);
1844	ret = tree_insert_offset(ctl, NULL, new_entry: info);
1845	if (ret)
1846	return ret;
1847
1848	rb_add_cached(node: &info->bytes_index, tree: &ctl->free_space_bytes, less: entry_less);
1849
1850	if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1851	ctl->discardable_extents[BTRFS_STAT_CURR]++;
1852	ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
1853	}
1854
1855	ctl->free_space += info->bytes;
1856	ctl->free_extents++;
1857	return ret;
1858	}
1859
1860	static void relink_bitmap_entry(struct btrfs_free_space_ctl *ctl,
1861	struct btrfs_free_space *info)
1862	{
1863	ASSERT(info->bitmap);
1864
1865	/*
1866	* If our entry is empty it's because we're on a cluster and we don't
1867	* want to re-link it into our ctl bytes index.
1868	*/
1869	if (RB_EMPTY_NODE(&info->bytes_index))
1870	return;
1871
1872	lockdep_assert_held(&ctl->tree_lock);
1873
1874	rb_erase_cached(node: &info->bytes_index, root: &ctl->free_space_bytes);
1875	rb_add_cached(node: &info->bytes_index, tree: &ctl->free_space_bytes, less: entry_less);
1876	}
1877
1878	static inline void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1879	struct btrfs_free_space *info,
1880	u64 offset, u64 bytes, bool update_stat)
1881	{
1882	unsigned long start, count, end;
1883	int extent_delta = -1;
1884
1885	start = offset_to_bit(bitmap_start: info->offset, unit: ctl->unit, offset);
1886	count = bytes_to_bits(bytes, unit: ctl->unit);
1887	end = start + count;
1888	ASSERT(end <= BITS_PER_BITMAP);
1889
1890	bitmap_clear(map: info->bitmap, start, nbits: count);
1891
1892	info->bytes -= bytes;
1893	if (info->max_extent_size > ctl->unit)
1894	info->max_extent_size = 0;
1895
1896	relink_bitmap_entry(ctl, info);
1897
1898	if (start && test_bit(start - 1, info->bitmap))
1899	extent_delta++;
1900
1901	if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1902	extent_delta++;
1903
1904	info->bitmap_extents += extent_delta;
1905	if (!btrfs_free_space_trimmed(info)) {
1906	ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1907	ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
1908	}
1909
1910	if (update_stat)
1911	ctl->free_space -= bytes;
1912	}
1913
1914	static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1915	struct btrfs_free_space *info, u64 offset,
1916	u64 bytes)
1917	{
1918	unsigned long start, count, end;
1919	int extent_delta = 1;
1920
1921	start = offset_to_bit(bitmap_start: info->offset, unit: ctl->unit, offset);
1922	count = bytes_to_bits(bytes, unit: ctl->unit);
1923	end = start + count;
1924	ASSERT(end <= BITS_PER_BITMAP);
1925
1926	bitmap_set(map: info->bitmap, start, nbits: count);
1927
1928	/*
1929	* We set some bytes, we have no idea what the max extent size is
1930	* anymore.
1931	*/
1932	info->max_extent_size = 0;
1933	info->bytes += bytes;
1934	ctl->free_space += bytes;
1935
1936	relink_bitmap_entry(ctl, info);
1937
1938	if (start && test_bit(start - 1, info->bitmap))
1939	extent_delta--;
1940
1941	if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1942	extent_delta--;
1943
1944	info->bitmap_extents += extent_delta;
1945	if (!btrfs_free_space_trimmed(info)) {
1946	ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1947	ctl->discardable_bytes[BTRFS_STAT_CURR] += bytes;
1948	}
1949	}
1950
1951	/*
1952	* If we can not find suitable extent, we will use bytes to record
1953	* the size of the max extent.
1954	*/
1955	static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1956	struct btrfs_free_space *bitmap_info, u64 *offset,
1957	u64 *bytes, bool for_alloc)
1958	{
1959	unsigned long found_bits = 0;
1960	unsigned long max_bits = 0;
1961	unsigned long bits, i;
1962	unsigned long next_zero;
1963	unsigned long extent_bits;
1964
1965	/*
1966	* Skip searching the bitmap if we don't have a contiguous section that
1967	* is large enough for this allocation.
1968	*/
1969	if (for_alloc &&
1970	bitmap_info->max_extent_size &&
1971	bitmap_info->max_extent_size < *bytes) {
1972	*bytes = bitmap_info->max_extent_size;
1973	return -1;
1974	}
1975
1976	i = offset_to_bit(bitmap_start: bitmap_info->offset, unit: ctl->unit,
1977	max_t(u64, *offset, bitmap_info->offset));
1978	bits = bytes_to_bits(bytes: *bytes, unit: ctl->unit);
1979
1980	for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1981	if (for_alloc && bits == 1) {
1982	found_bits = 1;
1983	break;
1984	}
1985	next_zero = find_next_zero_bit(addr: bitmap_info->bitmap,
1986	BITS_PER_BITMAP, offset: i);
1987	extent_bits = next_zero - i;
1988	if (extent_bits >= bits) {
1989	found_bits = extent_bits;
1990	break;
1991	} else if (extent_bits > max_bits) {
1992	max_bits = extent_bits;
1993	}
1994	i = next_zero;
1995	}
1996
1997	if (found_bits) {
1998	*offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1999	*bytes = (u64)(found_bits) * ctl->unit;
2000	return 0;
2001	}
2002
2003	*bytes = (u64)(max_bits) * ctl->unit;
2004	bitmap_info->max_extent_size = *bytes;
2005	relink_bitmap_entry(ctl, info: bitmap_info);
2006	return -1;
2007	}
2008
2009	/* Cache the size of the max extent in bytes */
2010	static struct btrfs_free_space *
2011	find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
2012	unsigned long align, u64 *max_extent_size, bool use_bytes_index)
2013	{
2014	struct btrfs_free_space *entry;
2015	struct rb_node *node;
2016	u64 tmp;
2017	u64 align_off;
2018	int ret;
2019
2020	if (!ctl->free_space_offset.rb_node)
2021	goto out;
2022	again:
2023	if (use_bytes_index) {
2024	node = rb_first_cached(&ctl->free_space_bytes);
2025	} else {
2026	entry = tree_search_offset(ctl, offset: offset_to_bitmap(ctl, offset: *offset),
2027	bitmap_only: 0, fuzzy: 1);
2028	if (!entry)
2029	goto out;
2030	node = &entry->offset_index;
2031	}
2032
2033	for (; node; node = rb_next(node)) {
2034	if (use_bytes_index)
2035	entry = rb_entry(node, struct btrfs_free_space,
2036	bytes_index);
2037	else
2038	entry = rb_entry(node, struct btrfs_free_space,
2039	offset_index);
2040
2041	/*
2042	* If we are using the bytes index then all subsequent entries
2043	* in this tree are going to be < bytes, so simply set the max
2044	* extent size and exit the loop.
2045	*
2046	* If we're using the offset index then we need to keep going
2047	* through the rest of the tree.
2048	*/
2049	if (entry->bytes < *bytes) {
2050	*max_extent_size = max(get_max_extent_size(entry),
2051	*max_extent_size);
2052	if (use_bytes_index)
2053	break;
2054	continue;
2055	}
2056
2057	/* make sure the space returned is big enough
2058	* to match our requested alignment
2059	*/
2060	if (*bytes >= align) {
2061	tmp = entry->offset - ctl->start + align - 1;
2062	tmp = div64_u64(dividend: tmp, divisor: align);
2063	tmp = tmp * align + ctl->start;
2064	align_off = tmp - entry->offset;
2065	} else {
2066	align_off = 0;
2067	tmp = entry->offset;
2068	}
2069
2070	/*
2071	* We don't break here if we're using the bytes index because we
2072	* may have another entry that has the correct alignment that is
2073	* the right size, so we don't want to miss that possibility.
2074	* At worst this adds another loop through the logic, but if we
2075	* broke here we could prematurely ENOSPC.
2076	*/
2077	if (entry->bytes < *bytes + align_off) {
2078	*max_extent_size = max(get_max_extent_size(entry),
2079	*max_extent_size);
2080	continue;
2081	}
2082
2083	if (entry->bitmap) {
2084	struct rb_node *old_next = rb_next(node);
2085	u64 size = *bytes;
2086
2087	ret = search_bitmap(ctl, bitmap_info: entry, offset: &tmp, bytes: &size, for_alloc: true);
2088	if (!ret) {
2089	*offset = tmp;
2090	*bytes = size;
2091	return entry;
2092	} else {
2093	*max_extent_size =
2094	max(get_max_extent_size(entry),
2095	*max_extent_size);
2096	}
2097
2098	/*
2099	* The bitmap may have gotten re-arranged in the space
2100	* index here because the max_extent_size may have been
2101	* updated. Start from the beginning again if this
2102	* happened.
2103	*/
2104	if (use_bytes_index && old_next != rb_next(node))
2105	goto again;
2106	continue;
2107	}
2108
2109	*offset = tmp;
2110	*bytes = entry->bytes - align_off;
2111	return entry;
2112	}
2113	out:
2114	return NULL;
2115	}
2116
2117	static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
2118	struct btrfs_free_space *info, u64 offset)
2119	{
2120	info->offset = offset_to_bitmap(ctl, offset);
2121	info->bytes = 0;
2122	info->bitmap_extents = 0;
2123	INIT_LIST_HEAD(list: &info->list);
2124	link_free_space(ctl, info);
2125	ctl->total_bitmaps++;
2126	recalculate_thresholds(ctl);
2127	}
2128
2129	static void free_bitmap(struct btrfs_free_space_ctl *ctl,
2130	struct btrfs_free_space *bitmap_info)
2131	{
2132	/*
2133	* Normally when this is called, the bitmap is completely empty. However,
2134	* if we are blowing up the free space cache for one reason or another
2135	* via __btrfs_remove_free_space_cache(), then it may not be freed and
2136	* we may leave stats on the table.
2137	*/
2138	if (bitmap_info->bytes && !btrfs_free_space_trimmed(info: bitmap_info)) {
2139	ctl->discardable_extents[BTRFS_STAT_CURR] -=
2140	bitmap_info->bitmap_extents;
2141	ctl->discardable_bytes[BTRFS_STAT_CURR] -= bitmap_info->bytes;
2142
2143	}
2144	unlink_free_space(ctl, info: bitmap_info, update_stat: true);
2145	kmem_cache_free(s: btrfs_free_space_bitmap_cachep, objp: bitmap_info->bitmap);
2146	kmem_cache_free(s: btrfs_free_space_cachep, objp: bitmap_info);
2147	ctl->total_bitmaps--;
2148	recalculate_thresholds(ctl);
2149	}
2150
2151	static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
2152	struct btrfs_free_space *bitmap_info,
2153	u64 *offset, u64 *bytes)
2154	{
2155	u64 end;
2156	u64 search_start, search_bytes;
2157	int ret;
2158
2159	again:
2160	end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
2161
2162	/*
2163	* We need to search for bits in this bitmap. We could only cover some
2164	* of the extent in this bitmap thanks to how we add space, so we need
2165	* to search for as much as it as we can and clear that amount, and then
2166	* go searching for the next bit.
2167	*/
2168	search_start = *offset;
2169	search_bytes = ctl->unit;
2170	search_bytes = min(search_bytes, end - search_start + 1);
2171	ret = search_bitmap(ctl, bitmap_info, offset: &search_start, bytes: &search_bytes,
2172	for_alloc: false);
2173	if (ret < 0 || search_start != *offset)
2174	return -EINVAL;
2175
2176	/* We may have found more bits than what we need */
2177	search_bytes = min(search_bytes, *bytes);
2178
2179	/* Cannot clear past the end of the bitmap */
2180	search_bytes = min(search_bytes, end - search_start + 1);
2181
2182	bitmap_clear_bits(ctl, info: bitmap_info, offset: search_start, bytes: search_bytes, update_stat: true);
2183	*offset += search_bytes;
2184	*bytes -= search_bytes;
2185
2186	if (*bytes) {
2187	struct rb_node *next = rb_next(&bitmap_info->offset_index);
2188	if (!bitmap_info->bytes)
2189	free_bitmap(ctl, bitmap_info);
2190
2191	/*
2192	* no entry after this bitmap, but we still have bytes to
2193	* remove, so something has gone wrong.
2194	*/
2195	if (!next)
2196	return -EINVAL;
2197
2198	bitmap_info = rb_entry(next, struct btrfs_free_space,
2199	offset_index);
2200
2201	/*
2202	* if the next entry isn't a bitmap we need to return to let the
2203	* extent stuff do its work.
2204	*/
2205	if (!bitmap_info->bitmap)
2206	return -EAGAIN;
2207
2208	/*
2209	* Ok the next item is a bitmap, but it may not actually hold
2210	* the information for the rest of this free space stuff, so
2211	* look for it, and if we don't find it return so we can try
2212	* everything over again.
2213	*/
2214	search_start = *offset;
2215	search_bytes = ctl->unit;
2216	ret = search_bitmap(ctl, bitmap_info, offset: &search_start,
2217	bytes: &search_bytes, for_alloc: false);
2218	if (ret < 0 || search_start != *offset)
2219	return -EAGAIN;
2220
2221	goto again;
2222	} else if (!bitmap_info->bytes)
2223	free_bitmap(ctl, bitmap_info);
2224
2225	return 0;
2226	}
2227
2228	static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
2229	struct btrfs_free_space *info, u64 offset,
2230	u64 bytes, enum btrfs_trim_state trim_state)
2231	{
2232	u64 bytes_to_set = 0;
2233	u64 end;
2234
2235	/*
2236	* This is a tradeoff to make bitmap trim state minimal. We mark the
2237	* whole bitmap untrimmed if at any point we add untrimmed regions.
2238	*/
2239	if (trim_state == BTRFS_TRIM_STATE_UNTRIMMED) {
2240	if (btrfs_free_space_trimmed(info)) {
2241	ctl->discardable_extents[BTRFS_STAT_CURR] +=
2242	info->bitmap_extents;
2243	ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
2244	}
2245	info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2246	}
2247
2248	end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
2249
2250	bytes_to_set = min(end - offset, bytes);
2251
2252	bitmap_set_bits(ctl, info, offset, bytes: bytes_to_set);
2253
2254	return bytes_to_set;
2255
2256	}
2257
2258	static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
2259	struct btrfs_free_space *info)
2260	{
2261	struct btrfs_block_group *block_group = ctl->block_group;
2262	struct btrfs_fs_info *fs_info = block_group->fs_info;
2263	bool forced = false;
2264
2265	#ifdef CONFIG_BTRFS_DEBUG
2266	if (btrfs_should_fragment_free_space(block_group))
2267	forced = true;
2268	#endif
2269
2270	/* This is a way to reclaim large regions from the bitmaps. */
2271	if (!forced && info->bytes >= FORCE_EXTENT_THRESHOLD)
2272	return false;
2273
2274	/*
2275	* If we are below the extents threshold then we can add this as an
2276	* extent, and don't have to deal with the bitmap
2277	*/
2278	if (!forced && ctl->free_extents < ctl->extents_thresh) {
2279	/*
2280	* If this block group has some small extents we don't want to
2281	* use up all of our free slots in the cache with them, we want
2282	* to reserve them to larger extents, however if we have plenty
2283	* of cache left then go ahead an dadd them, no sense in adding
2284	* the overhead of a bitmap if we don't have to.
2285	*/
2286	if (info->bytes <= fs_info->sectorsize * 8) {
2287	if (ctl->free_extents * 3 <= ctl->extents_thresh)
2288	return false;
2289	} else {
2290	return false;
2291	}
2292	}
2293
2294	/*
2295	* The original block groups from mkfs can be really small, like 8
2296	* megabytes, so don't bother with a bitmap for those entries. However
2297	* some block groups can be smaller than what a bitmap would cover but
2298	* are still large enough that they could overflow the 32k memory limit,
2299	* so allow those block groups to still be allowed to have a bitmap
2300	* entry.
2301	*/
2302	if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->length)
2303	return false;
2304
2305	return true;
2306	}
2307
2308	static const struct btrfs_free_space_op free_space_op = {
2309	.use_bitmap = use_bitmap,
2310	};
2311
2312	static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
2313	struct btrfs_free_space *info)
2314	{
2315	struct btrfs_free_space *bitmap_info;
2316	struct btrfs_block_group *block_group = NULL;
2317	int added = 0;
2318	u64 bytes, offset, bytes_added;
2319	enum btrfs_trim_state trim_state;
2320	int ret;
2321
2322	bytes = info->bytes;
2323	offset = info->offset;
2324	trim_state = info->trim_state;
2325
2326	if (!ctl->op->use_bitmap(ctl, info))
2327	return 0;
2328
2329	if (ctl->op == &free_space_op)
2330	block_group = ctl->block_group;
2331	again:
2332	/*
2333	* Since we link bitmaps right into the cluster we need to see if we
2334	* have a cluster here, and if so and it has our bitmap we need to add
2335	* the free space to that bitmap.
2336	*/
2337	if (block_group && !list_empty(head: &block_group->cluster_list)) {
2338	struct btrfs_free_cluster *cluster;
2339	struct rb_node *node;
2340	struct btrfs_free_space *entry;
2341
2342	cluster = list_entry(block_group->cluster_list.next,
2343	struct btrfs_free_cluster,
2344	block_group_list);
2345	spin_lock(lock: &cluster->lock);
2346	node = rb_first(&cluster->root);
2347	if (!node) {
2348	spin_unlock(lock: &cluster->lock);
2349	goto no_cluster_bitmap;
2350	}
2351
2352	entry = rb_entry(node, struct btrfs_free_space, offset_index);
2353	if (!entry->bitmap) {
2354	spin_unlock(lock: &cluster->lock);
2355	goto no_cluster_bitmap;
2356	}
2357
2358	if (entry->offset == offset_to_bitmap(ctl, offset)) {
2359	bytes_added = add_bytes_to_bitmap(ctl, info: entry, offset,
2360	bytes, trim_state);
2361	bytes -= bytes_added;
2362	offset += bytes_added;
2363	}
2364	spin_unlock(lock: &cluster->lock);
2365	if (!bytes) {
2366	ret = 1;
2367	goto out;
2368	}
2369	}
2370
2371	no_cluster_bitmap:
2372	bitmap_info = tree_search_offset(ctl, offset: offset_to_bitmap(ctl, offset),
2373	bitmap_only: 1, fuzzy: 0);
2374	if (!bitmap_info) {
2375	ASSERT(added == 0);
2376	goto new_bitmap;
2377	}
2378
2379	bytes_added = add_bytes_to_bitmap(ctl, info: bitmap_info, offset, bytes,
2380	trim_state);
2381	bytes -= bytes_added;
2382	offset += bytes_added;
2383	added = 0;
2384
2385	if (!bytes) {
2386	ret = 1;
2387	goto out;
2388	} else
2389	goto again;
2390
2391	new_bitmap:
2392	if (info && info->bitmap) {
2393	add_new_bitmap(ctl, info, offset);
2394	added = 1;
2395	info = NULL;
2396	goto again;
2397	} else {
2398	spin_unlock(lock: &ctl->tree_lock);
2399
2400	/* no pre-allocated info, allocate a new one */
2401	if (!info) {
2402	info = kmem_cache_zalloc(k: btrfs_free_space_cachep,
2403	GFP_NOFS);
2404	if (!info) {
2405	spin_lock(lock: &ctl->tree_lock);
2406	ret = -ENOMEM;
2407	goto out;
2408	}
2409	}
2410
2411	/* allocate the bitmap */
2412	info->bitmap = kmem_cache_zalloc(k: btrfs_free_space_bitmap_cachep,
2413	GFP_NOFS);
2414	info->trim_state = BTRFS_TRIM_STATE_TRIMMED;
2415	spin_lock(lock: &ctl->tree_lock);
2416	if (!info->bitmap) {
2417	ret = -ENOMEM;
2418	goto out;
2419	}
2420	goto again;
2421	}
2422
2423	out:
2424	if (info) {
2425	if (info->bitmap)
2426	kmem_cache_free(s: btrfs_free_space_bitmap_cachep,
2427	objp: info->bitmap);
2428	kmem_cache_free(s: btrfs_free_space_cachep, objp: info);
2429	}
2430
2431	return ret;
2432	}
2433
2434	/*
2435	* Free space merging rules:
2436	* 1) Merge trimmed areas together
2437	* 2) Let untrimmed areas coalesce with trimmed areas
2438	* 3) Always pull neighboring regions from bitmaps
2439	*
2440	* The above rules are for when we merge free space based on btrfs_trim_state.
2441	* Rules 2 and 3 are subtle because they are suboptimal, but are done for the
2442	* same reason: to promote larger extent regions which makes life easier for
2443	* find_free_extent(). Rule 2 enables coalescing based on the common path
2444	* being returning free space from btrfs_finish_extent_commit(). So when free
2445	* space is trimmed, it will prevent aggregating trimmed new region and
2446	* untrimmed regions in the rb_tree. Rule 3 is purely to obtain larger extents
2447	* and provide find_free_extent() with the largest extents possible hoping for
2448	* the reuse path.
2449	*/
2450	static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2451	struct btrfs_free_space *info, bool update_stat)
2452	{
2453	struct btrfs_free_space *left_info = NULL;
2454	struct btrfs_free_space *right_info;
2455	bool merged = false;
2456	u64 offset = info->offset;
2457	u64 bytes = info->bytes;
2458	const bool is_trimmed = btrfs_free_space_trimmed(info);
2459	struct rb_node *right_prev = NULL;
2460
2461	/*
2462	* first we want to see if there is free space adjacent to the range we
2463	* are adding, if there is remove that struct and add a new one to
2464	* cover the entire range
2465	*/
2466	right_info = tree_search_offset(ctl, offset: offset + bytes, bitmap_only: 0, fuzzy: 0);
2467	if (right_info)
2468	right_prev = rb_prev(&right_info->offset_index);
2469
2470	if (right_prev)
2471	left_info = rb_entry(right_prev, struct btrfs_free_space, offset_index);
2472	else if (!right_info)
2473	left_info = tree_search_offset(ctl, offset: offset - 1, bitmap_only: 0, fuzzy: 0);
2474
2475	/* See try_merge_free_space() comment. */
2476	if (right_info && !right_info->bitmap &&
2477	(!is_trimmed || btrfs_free_space_trimmed(info: right_info))) {
2478	unlink_free_space(ctl, info: right_info, update_stat);
2479	info->bytes += right_info->bytes;
2480	kmem_cache_free(s: btrfs_free_space_cachep, objp: right_info);
2481	merged = true;
2482	}
2483
2484	/* See try_merge_free_space() comment. */
2485	if (left_info && !left_info->bitmap &&
2486	left_info->offset + left_info->bytes == offset &&
2487	(!is_trimmed || btrfs_free_space_trimmed(info: left_info))) {
2488	unlink_free_space(ctl, info: left_info, update_stat);
2489	info->offset = left_info->offset;
2490	info->bytes += left_info->bytes;
2491	kmem_cache_free(s: btrfs_free_space_cachep, objp: left_info);
2492	merged = true;
2493	}
2494
2495	return merged;
2496	}
2497
2498	static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2499	struct btrfs_free_space *info,
2500	bool update_stat)
2501	{
2502	struct btrfs_free_space *bitmap;
2503	unsigned long i;
2504	unsigned long j;
2505	const u64 end = info->offset + info->bytes;
2506	const u64 bitmap_offset = offset_to_bitmap(ctl, offset: end);
2507	u64 bytes;
2508
2509	bitmap = tree_search_offset(ctl, offset: bitmap_offset, bitmap_only: 1, fuzzy: 0);
2510	if (!bitmap)
2511	return false;
2512
2513	i = offset_to_bit(bitmap_start: bitmap->offset, unit: ctl->unit, offset: end);
2514	j = find_next_zero_bit(addr: bitmap->bitmap, BITS_PER_BITMAP, offset: i);
2515	if (j == i)
2516	return false;
2517	bytes = (j - i) * ctl->unit;
2518	info->bytes += bytes;
2519
2520	/* See try_merge_free_space() comment. */
2521	if (!btrfs_free_space_trimmed(info: bitmap))
2522	info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2523
2524	bitmap_clear_bits(ctl, info: bitmap, offset: end, bytes, update_stat);
2525
2526	if (!bitmap->bytes)
2527	free_bitmap(ctl, bitmap_info: bitmap);
2528
2529	return true;
2530	}
2531
2532	static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2533	struct btrfs_free_space *info,
2534	bool update_stat)
2535	{
2536	struct btrfs_free_space *bitmap;
2537	u64 bitmap_offset;
2538	unsigned long i;
2539	unsigned long j;
2540	unsigned long prev_j;
2541	u64 bytes;
2542
2543	bitmap_offset = offset_to_bitmap(ctl, offset: info->offset);
2544	/* If we're on a boundary, try the previous logical bitmap. */
2545	if (bitmap_offset == info->offset) {
2546	if (info->offset == 0)
2547	return false;
2548	bitmap_offset = offset_to_bitmap(ctl, offset: info->offset - 1);
2549	}
2550
2551	bitmap = tree_search_offset(ctl, offset: bitmap_offset, bitmap_only: 1, fuzzy: 0);
2552	if (!bitmap)
2553	return false;
2554
2555	i = offset_to_bit(bitmap_start: bitmap->offset, unit: ctl->unit, offset: info->offset) - 1;
2556	j = 0;
2557	prev_j = (unsigned long)-1;
2558	for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2559	if (j > i)
2560	break;
2561	prev_j = j;
2562	}
2563	if (prev_j == i)
2564	return false;
2565
2566	if (prev_j == (unsigned long)-1)
2567	bytes = (i + 1) * ctl->unit;
2568	else
2569	bytes = (i - prev_j) * ctl->unit;
2570
2571	info->offset -= bytes;
2572	info->bytes += bytes;
2573
2574	/* See try_merge_free_space() comment. */
2575	if (!btrfs_free_space_trimmed(info: bitmap))
2576	info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2577
2578	bitmap_clear_bits(ctl, info: bitmap, offset: info->offset, bytes, update_stat);
2579
2580	if (!bitmap->bytes)
2581	free_bitmap(ctl, bitmap_info: bitmap);
2582
2583	return true;
2584	}
2585
2586	/*
2587	* We prefer always to allocate from extent entries, both for clustered and
2588	* non-clustered allocation requests. So when attempting to add a new extent
2589	* entry, try to see if there's adjacent free space in bitmap entries, and if
2590	* there is, migrate that space from the bitmaps to the extent.
2591	* Like this we get better chances of satisfying space allocation requests
2592	* because we attempt to satisfy them based on a single cache entry, and never
2593	* on 2 or more entries - even if the entries represent a contiguous free space
2594	* region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2595	* ends).
2596	*/
2597	static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2598	struct btrfs_free_space *info,
2599	bool update_stat)
2600	{
2601	/*
2602	* Only work with disconnected entries, as we can change their offset,
2603	* and must be extent entries.
2604	*/
2605	ASSERT(!info->bitmap);
2606	ASSERT(RB_EMPTY_NODE(&info->offset_index));
2607
2608	if (ctl->total_bitmaps > 0) {
2609	bool stole_end;
2610	bool stole_front = false;
2611
2612	stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2613	if (ctl->total_bitmaps > 0)
2614	stole_front = steal_from_bitmap_to_front(ctl, info,
2615	update_stat);
2616
2617	if (stole_end || stole_front)
2618	try_merge_free_space(ctl, info, update_stat);
2619	}
2620	}
2621
2622	static int __btrfs_add_free_space(struct btrfs_block_group *block_group,
2623	u64 offset, u64 bytes,
2624	enum btrfs_trim_state trim_state)
2625	{
2626	struct btrfs_fs_info *fs_info = block_group->fs_info;
2627	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2628	struct btrfs_free_space *info;
2629	int ret = 0;
2630	u64 filter_bytes = bytes;
2631
2632	ASSERT(!btrfs_is_zoned(fs_info));
2633
2634	info = kmem_cache_zalloc(k: btrfs_free_space_cachep, GFP_NOFS);
2635	if (!info)
2636	return -ENOMEM;
2637
2638	info->offset = offset;
2639	info->bytes = bytes;
2640	info->trim_state = trim_state;
2641	RB_CLEAR_NODE(&info->offset_index);
2642	RB_CLEAR_NODE(&info->bytes_index);
2643
2644	spin_lock(lock: &ctl->tree_lock);
2645
2646	if (try_merge_free_space(ctl, info, update_stat: true))
2647	goto link;
2648
2649	/*
2650	* There was no extent directly to the left or right of this new
2651	* extent then we know we're going to have to allocate a new extent, so
2652	* before we do that see if we need to drop this into a bitmap
2653	*/
2654	ret = insert_into_bitmap(ctl, info);
2655	if (ret < 0) {
2656	goto out;
2657	} else if (ret) {
2658	ret = 0;
2659	goto out;
2660	}
2661	link:
2662	/*
2663	* Only steal free space from adjacent bitmaps if we're sure we're not
2664	* going to add the new free space to existing bitmap entries - because
2665	* that would mean unnecessary work that would be reverted. Therefore
2666	* attempt to steal space from bitmaps if we're adding an extent entry.
2667	*/
2668	steal_from_bitmap(ctl, info, update_stat: true);
2669
2670	filter_bytes = max(filter_bytes, info->bytes);
2671
2672	ret = link_free_space(ctl, info);
2673	if (ret)
2674	kmem_cache_free(s: btrfs_free_space_cachep, objp: info);
2675	out:
2676	btrfs_discard_update_discardable(block_group);
2677	spin_unlock(lock: &ctl->tree_lock);
2678
2679	if (ret) {
2680	btrfs_crit(fs_info, "unable to add free space :%d", ret);
2681	ASSERT(ret != -EEXIST);
2682	}
2683
2684	if (trim_state != BTRFS_TRIM_STATE_TRIMMED) {
2685	btrfs_discard_check_filter(block_group, bytes: filter_bytes);
2686	btrfs_discard_queue_work(discard_ctl: &fs_info->discard_ctl, block_group);
2687	}
2688
2689	return ret;
2690	}
2691
2692	static int __btrfs_add_free_space_zoned(struct btrfs_block_group *block_group,
2693	u64 bytenr, u64 size, bool used)
2694	{
2695	struct btrfs_space_info *sinfo = block_group->space_info;
2696	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2697	u64 offset = bytenr - block_group->start;
2698	u64 to_free, to_unusable;
2699	int bg_reclaim_threshold = 0;
2700	bool initial = (size == block_group->length);
2701	u64 reclaimable_unusable;
2702
2703	WARN_ON(!initial && offset + size > block_group->zone_capacity);
2704
2705	if (!initial)
2706	bg_reclaim_threshold = READ_ONCE(sinfo->bg_reclaim_threshold);
2707
2708	spin_lock(lock: &ctl->tree_lock);
2709	if (!used)
2710	to_free = size;
2711	else if (initial)
2712	to_free = block_group->zone_capacity;
2713	else if (offset >= block_group->alloc_offset)
2714	to_free = size;
2715	else if (offset + size <= block_group->alloc_offset)
2716	to_free = 0;
2717	else
2718	to_free = offset + size - block_group->alloc_offset;
2719	to_unusable = size - to_free;
2720
2721	ctl->free_space += to_free;
2722	/*
2723	* If the block group is read-only, we should account freed space into
2724	* bytes_readonly.
2725	*/
2726	if (!block_group->ro)
2727	block_group->zone_unusable += to_unusable;
2728	spin_unlock(lock: &ctl->tree_lock);
2729	if (!used) {
2730	spin_lock(lock: &block_group->lock);
2731	block_group->alloc_offset -= size;
2732	spin_unlock(lock: &block_group->lock);
2733	}
2734
2735	reclaimable_unusable = block_group->zone_unusable -
2736	(block_group->length - block_group->zone_capacity);
2737	/* All the region is now unusable. Mark it as unused and reclaim */
2738	if (block_group->zone_unusable == block_group->length) {
2739	btrfs_mark_bg_unused(bg: block_group);
2740	} else if (bg_reclaim_threshold &&
2741	reclaimable_unusable >=
2742	mult_perc(num: block_group->zone_capacity, percent: bg_reclaim_threshold)) {
2743	btrfs_mark_bg_to_reclaim(bg: block_group);
2744	}
2745
2746	return 0;
2747	}
2748
2749	int btrfs_add_free_space(struct btrfs_block_group *block_group,
2750	u64 bytenr, u64 size)
2751	{
2752	enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2753
2754	if (btrfs_is_zoned(fs_info: block_group->fs_info))
2755	return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2756	used: true);
2757
2758	if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC))
2759	trim_state = BTRFS_TRIM_STATE_TRIMMED;
2760
2761	return __btrfs_add_free_space(block_group, offset: bytenr, bytes: size, trim_state);
2762	}
2763
2764	int btrfs_add_free_space_unused(struct btrfs_block_group *block_group,
2765	u64 bytenr, u64 size)
2766	{
2767	if (btrfs_is_zoned(fs_info: block_group->fs_info))
2768	return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2769	used: false);
2770
2771	return btrfs_add_free_space(block_group, bytenr, size);
2772	}
2773
2774	/*
2775	* This is a subtle distinction because when adding free space back in general,
2776	* we want it to be added as untrimmed for async. But in the case where we add
2777	* it on loading of a block group, we want to consider it trimmed.
2778	*/
2779	int btrfs_add_free_space_async_trimmed(struct btrfs_block_group *block_group,
2780	u64 bytenr, u64 size)
2781	{
2782	enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2783
2784	if (btrfs_is_zoned(fs_info: block_group->fs_info))
2785	return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2786	used: true);
2787
2788	if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC) ||
2789	btrfs_test_opt(block_group->fs_info, DISCARD_ASYNC))
2790	trim_state = BTRFS_TRIM_STATE_TRIMMED;
2791
2792	return __btrfs_add_free_space(block_group, offset: bytenr, bytes: size, trim_state);
2793	}
2794
2795	int btrfs_remove_free_space(struct btrfs_block_group *block_group,
2796	u64 offset, u64 bytes)
2797	{
2798	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2799	struct btrfs_free_space *info;
2800	int ret;
2801	bool re_search = false;
2802
2803	if (btrfs_is_zoned(fs_info: block_group->fs_info)) {
2804	/*
2805	* This can happen with conventional zones when replaying log.
2806	* Since the allocation info of tree-log nodes are not recorded
2807	* to the extent-tree, calculate_alloc_pointer() failed to
2808	* advance the allocation pointer after last allocated tree log
2809	* node blocks.
2810	*
2811	* This function is called from
2812	* btrfs_pin_extent_for_log_replay() when replaying the log.
2813	* Advance the pointer not to overwrite the tree-log nodes.
2814	*/
2815	if (block_group->start + block_group->alloc_offset <
2816	offset + bytes) {
2817	block_group->alloc_offset =
2818	offset + bytes - block_group->start;
2819	}
2820	return 0;
2821	}
2822
2823	spin_lock(lock: &ctl->tree_lock);
2824
2825	again:
2826	ret = 0;
2827	if (!bytes)
2828	goto out_lock;
2829
2830	info = tree_search_offset(ctl, offset, bitmap_only: 0, fuzzy: 0);
2831	if (!info) {
2832	/*
2833	* oops didn't find an extent that matched the space we wanted
2834	* to remove, look for a bitmap instead
2835	*/
2836	info = tree_search_offset(ctl, offset: offset_to_bitmap(ctl, offset),
2837	bitmap_only: 1, fuzzy: 0);
2838	if (!info) {
2839	/*
2840	* If we found a partial bit of our free space in a
2841	* bitmap but then couldn't find the other part this may
2842	* be a problem, so WARN about it.
2843	*/
2844	WARN_ON(re_search);
2845	goto out_lock;
2846	}
2847	}
2848
2849	re_search = false;
2850	if (!info->bitmap) {
2851	unlink_free_space(ctl, info, update_stat: true);
2852	if (offset == info->offset) {
2853	u64 to_free = min(bytes, info->bytes);
2854
2855	info->bytes -= to_free;
2856	info->offset += to_free;
2857	if (info->bytes) {
2858	ret = link_free_space(ctl, info);
2859	WARN_ON(ret);
2860	} else {
2861	kmem_cache_free(s: btrfs_free_space_cachep, objp: info);
2862	}
2863
2864	offset += to_free;
2865	bytes -= to_free;
2866	goto again;
2867	} else {
2868	u64 old_end = info->bytes + info->offset;
2869
2870	info->bytes = offset - info->offset;
2871	ret = link_free_space(ctl, info);
2872	WARN_ON(ret);
2873	if (ret)
2874	goto out_lock;
2875
2876	/* Not enough bytes in this entry to satisfy us */
2877	if (old_end < offset + bytes) {
2878	bytes -= old_end - offset;
2879	offset = old_end;
2880	goto again;
2881	} else if (old_end == offset + bytes) {
2882	/* all done */
2883	goto out_lock;
2884	}
2885	spin_unlock(lock: &ctl->tree_lock);
2886
2887	ret = __btrfs_add_free_space(block_group,
2888	offset: offset + bytes,
2889	bytes: old_end - (offset + bytes),
2890	trim_state: info->trim_state);
2891	WARN_ON(ret);
2892	goto out;
2893	}
2894	}
2895
2896	ret = remove_from_bitmap(ctl, bitmap_info: info, offset: &offset, bytes: &bytes);
2897	if (ret == -EAGAIN) {
2898	re_search = true;
2899	goto again;
2900	}
2901	out_lock:
2902	btrfs_discard_update_discardable(block_group);
2903	spin_unlock(lock: &ctl->tree_lock);
2904	out:
2905	return ret;
2906	}
2907
2908	void btrfs_dump_free_space(struct btrfs_block_group *block_group,
2909	u64 bytes)
2910	{
2911	struct btrfs_fs_info *fs_info = block_group->fs_info;
2912	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2913	struct btrfs_free_space *info;
2914	struct rb_node *n;
2915	int count = 0;
2916
2917	/*
2918	* Zoned btrfs does not use free space tree and cluster. Just print
2919	* out the free space after the allocation offset.
2920	*/
2921	if (btrfs_is_zoned(fs_info)) {
2922	btrfs_info(fs_info, "free space %llu active %d",
2923	block_group->zone_capacity - block_group->alloc_offset,
2924	test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
2925	&block_group->runtime_flags));
2926	return;
2927	}
2928
2929	spin_lock(lock: &ctl->tree_lock);
2930	for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2931	info = rb_entry(n, struct btrfs_free_space, offset_index);
2932	if (info->bytes >= bytes && !block_group->ro)
2933	count++;
2934	btrfs_crit(fs_info, "entry offset %llu, bytes %llu, bitmap %s",
2935	info->offset, info->bytes,
2936	(info->bitmap) ? "yes" : "no");
2937	}
2938	spin_unlock(lock: &ctl->tree_lock);
2939	btrfs_info(fs_info, "block group has cluster?: %s",
2940	list_empty(&block_group->cluster_list) ? "no" : "yes");
2941	btrfs_info(fs_info,
2942	"%d free space entries at or bigger than %llu bytes",
2943	count, bytes);
2944	}
2945
2946	void btrfs_init_free_space_ctl(struct btrfs_block_group *block_group,
2947	struct btrfs_free_space_ctl *ctl)
2948	{
2949	struct btrfs_fs_info *fs_info = block_group->fs_info;
2950
2951	spin_lock_init(&ctl->tree_lock);
2952	ctl->unit = fs_info->sectorsize;
2953	ctl->start = block_group->start;
2954	ctl->block_group = block_group;
2955	ctl->op = &free_space_op;
2956	ctl->free_space_bytes = RB_ROOT_CACHED;
2957	INIT_LIST_HEAD(list: &ctl->trimming_ranges);
2958	mutex_init(&ctl->cache_writeout_mutex);
2959
2960	/*
2961	* we only want to have 32k of ram per block group for keeping
2962	* track of free space, and if we pass 1/2 of that we want to
2963	* start converting things over to using bitmaps
2964	*/
2965	ctl->extents_thresh = (SZ_32K / 2) / sizeof(struct btrfs_free_space);
2966	}
2967
2968	/*
2969	* for a given cluster, put all of its extents back into the free
2970	* space cache. If the block group passed doesn't match the block group
2971	* pointed to by the cluster, someone else raced in and freed the
2972	* cluster already. In that case, we just return without changing anything
2973	*/
2974	static void __btrfs_return_cluster_to_free_space(
2975	struct btrfs_block_group *block_group,
2976	struct btrfs_free_cluster *cluster)
2977	{
2978	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2979	struct rb_node *node;
2980
2981	lockdep_assert_held(&ctl->tree_lock);
2982
2983	spin_lock(lock: &cluster->lock);
2984	if (cluster->block_group != block_group) {
2985	spin_unlock(lock: &cluster->lock);
2986	return;
2987	}
2988
2989	cluster->block_group = NULL;
2990	cluster->window_start = 0;
2991	list_del_init(entry: &cluster->block_group_list);
2992
2993	node = rb_first(&cluster->root);
2994	while (node) {
2995	struct btrfs_free_space *entry;
2996
2997	entry = rb_entry(node, struct btrfs_free_space, offset_index);
2998	node = rb_next(&entry->offset_index);
2999	rb_erase(&entry->offset_index, &cluster->root);
3000	RB_CLEAR_NODE(&entry->offset_index);
3001
3002	if (!entry->bitmap) {
3003	/* Merging treats extents as if they were new */
3004	if (!btrfs_free_space_trimmed(info: entry)) {
3005	ctl->discardable_extents[BTRFS_STAT_CURR]--;
3006	ctl->discardable_bytes[BTRFS_STAT_CURR] -=
3007	entry->bytes;
3008	}
3009
3010	try_merge_free_space(ctl, info: entry, update_stat: false);
3011	steal_from_bitmap(ctl, info: entry, update_stat: false);
3012
3013	/* As we insert directly, update these statistics */
3014	if (!btrfs_free_space_trimmed(info: entry)) {
3015	ctl->discardable_extents[BTRFS_STAT_CURR]++;
3016	ctl->discardable_bytes[BTRFS_STAT_CURR] +=
3017	entry->bytes;
3018	}
3019	}
3020	tree_insert_offset(ctl, NULL, new_entry: entry);
3021	rb_add_cached(node: &entry->bytes_index, tree: &ctl->free_space_bytes,
3022	less: entry_less);
3023	}
3024	cluster->root = RB_ROOT;
3025	spin_unlock(lock: &cluster->lock);
3026	btrfs_put_block_group(cache: block_group);
3027	}
3028
3029	void btrfs_remove_free_space_cache(struct btrfs_block_group *block_group)
3030	{
3031	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3032	struct btrfs_free_cluster *cluster;
3033	struct list_head *head;
3034
3035	spin_lock(lock: &ctl->tree_lock);
3036	while ((head = block_group->cluster_list.next) !=
3037	&block_group->cluster_list) {
3038	cluster = list_entry(head, struct btrfs_free_cluster,
3039	block_group_list);
3040
3041	WARN_ON(cluster->block_group != block_group);
3042	__btrfs_return_cluster_to_free_space(block_group, cluster);
3043
3044	cond_resched_lock(&ctl->tree_lock);
3045	}
3046	__btrfs_remove_free_space_cache(ctl);
3047	btrfs_discard_update_discardable(block_group);
3048	spin_unlock(lock: &ctl->tree_lock);
3049
3050	}
3051
3052	/*
3053	* Walk @block_group's free space rb_tree to determine if everything is trimmed.
3054	*/
3055	bool btrfs_is_free_space_trimmed(struct btrfs_block_group *block_group)
3056	{
3057	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3058	struct btrfs_free_space *info;
3059	struct rb_node *node;
3060	bool ret = true;
3061
3062	spin_lock(lock: &ctl->tree_lock);
3063	node = rb_first(&ctl->free_space_offset);
3064
3065	while (node) {
3066	info = rb_entry(node, struct btrfs_free_space, offset_index);
3067
3068	if (!btrfs_free_space_trimmed(info)) {
3069	ret = false;
3070	break;
3071	}
3072
3073	node = rb_next(node);
3074	}
3075
3076	spin_unlock(lock: &ctl->tree_lock);
3077	return ret;
3078	}
3079
3080	u64 btrfs_find_space_for_alloc(struct btrfs_block_group *block_group,
3081	u64 offset, u64 bytes, u64 empty_size,
3082	u64 *max_extent_size)
3083	{
3084	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3085	struct btrfs_discard_ctl *discard_ctl =
3086	&block_group->fs_info->discard_ctl;
3087	struct btrfs_free_space *entry = NULL;
3088	u64 bytes_search = bytes + empty_size;
3089	u64 ret = 0;
3090	u64 align_gap = 0;
3091	u64 align_gap_len = 0;
3092	enum btrfs_trim_state align_gap_trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3093	bool use_bytes_index = (offset == block_group->start);
3094
3095	ASSERT(!btrfs_is_zoned(block_group->fs_info));
3096
3097	spin_lock(lock: &ctl->tree_lock);
3098	entry = find_free_space(ctl, offset: &offset, bytes: &bytes_search,
3099	align: block_group->full_stripe_len, max_extent_size,
3100	use_bytes_index);
3101	if (!entry)
3102	goto out;
3103
3104	ret = offset;
3105	if (entry->bitmap) {
3106	bitmap_clear_bits(ctl, info: entry, offset, bytes, update_stat: true);
3107
3108	if (!btrfs_free_space_trimmed(info: entry))
3109	atomic64_add(i: bytes, v: &discard_ctl->discard_bytes_saved);
3110
3111	if (!entry->bytes)
3112	free_bitmap(ctl, bitmap_info: entry);
3113	} else {
3114	unlink_free_space(ctl, info: entry, update_stat: true);
3115	align_gap_len = offset - entry->offset;
3116	align_gap = entry->offset;
3117	align_gap_trim_state = entry->trim_state;
3118
3119	if (!btrfs_free_space_trimmed(info: entry))
3120	atomic64_add(i: bytes, v: &discard_ctl->discard_bytes_saved);
3121
3122	entry->offset = offset + bytes;
3123	WARN_ON(entry->bytes < bytes + align_gap_len);
3124
3125	entry->bytes -= bytes + align_gap_len;
3126	if (!entry->bytes)
3127	kmem_cache_free(s: btrfs_free_space_cachep, objp: entry);
3128	else
3129	link_free_space(ctl, info: entry);
3130	}
3131	out:
3132	btrfs_discard_update_discardable(block_group);
3133	spin_unlock(lock: &ctl->tree_lock);
3134
3135	if (align_gap_len)
3136	__btrfs_add_free_space(block_group, offset: align_gap, bytes: align_gap_len,
3137	trim_state: align_gap_trim_state);
3138	return ret;
3139	}
3140
3141	/*
3142	* given a cluster, put all of its extents back into the free space
3143	* cache. If a block group is passed, this function will only free
3144	* a cluster that belongs to the passed block group.
3145	*
3146	* Otherwise, it'll get a reference on the block group pointed to by the
3147	* cluster and remove the cluster from it.
3148	*/
3149	void btrfs_return_cluster_to_free_space(
3150	struct btrfs_block_group *block_group,
3151	struct btrfs_free_cluster *cluster)
3152	{
3153	struct btrfs_free_space_ctl *ctl;
3154
3155	/* first, get a safe pointer to the block group */
3156	spin_lock(lock: &cluster->lock);
3157	if (!block_group) {
3158	block_group = cluster->block_group;
3159	if (!block_group) {
3160	spin_unlock(lock: &cluster->lock);
3161	return;
3162	}
3163	} else if (cluster->block_group != block_group) {
3164	/* someone else has already freed it don't redo their work */
3165	spin_unlock(lock: &cluster->lock);
3166	return;
3167	}
3168	btrfs_get_block_group(cache: block_group);
3169	spin_unlock(lock: &cluster->lock);
3170
3171	ctl = block_group->free_space_ctl;
3172
3173	/* now return any extents the cluster had on it */
3174	spin_lock(lock: &ctl->tree_lock);
3175	__btrfs_return_cluster_to_free_space(block_group, cluster);
3176	spin_unlock(lock: &ctl->tree_lock);
3177
3178	btrfs_discard_queue_work(discard_ctl: &block_group->fs_info->discard_ctl, block_group);
3179
3180	/* finally drop our ref */
3181	btrfs_put_block_group(cache: block_group);
3182	}
3183
3184	static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group *block_group,
3185	struct btrfs_free_cluster *cluster,
3186	struct btrfs_free_space *entry,
3187	u64 bytes, u64 min_start,
3188	u64 *max_extent_size)
3189	{
3190	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3191	int err;
3192	u64 search_start = cluster->window_start;
3193	u64 search_bytes = bytes;
3194	u64 ret = 0;
3195
3196	search_start = min_start;
3197	search_bytes = bytes;
3198
3199	err = search_bitmap(ctl, bitmap_info: entry, offset: &search_start, bytes: &search_bytes, for_alloc: true);
3200	if (err) {
3201	*max_extent_size = max(get_max_extent_size(entry),
3202	*max_extent_size);
3203	return 0;
3204	}
3205
3206	ret = search_start;
3207	bitmap_clear_bits(ctl, info: entry, offset: ret, bytes, update_stat: false);
3208
3209	return ret;
3210	}
3211
3212	/*
3213	* given a cluster, try to allocate 'bytes' from it, returns 0
3214	* if it couldn't find anything suitably large, or a logical disk offset
3215	* if things worked out
3216	*/
3217	u64 btrfs_alloc_from_cluster(struct btrfs_block_group *block_group,
3218	struct btrfs_free_cluster *cluster, u64 bytes,
3219	u64 min_start, u64 *max_extent_size)
3220	{
3221	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3222	struct btrfs_discard_ctl *discard_ctl =
3223	&block_group->fs_info->discard_ctl;
3224	struct btrfs_free_space *entry = NULL;
3225	struct rb_node *node;
3226	u64 ret = 0;
3227
3228	ASSERT(!btrfs_is_zoned(block_group->fs_info));
3229
3230	spin_lock(lock: &cluster->lock);
3231	if (bytes > cluster->max_size)
3232	goto out;
3233
3234	if (cluster->block_group != block_group)
3235	goto out;
3236
3237	node = rb_first(&cluster->root);
3238	if (!node)
3239	goto out;
3240
3241	entry = rb_entry(node, struct btrfs_free_space, offset_index);
3242	while (1) {
3243	if (entry->bytes < bytes)
3244	*max_extent_size = max(get_max_extent_size(entry),
3245	*max_extent_size);
3246
3247	if (entry->bytes < bytes ||
3248	(!entry->bitmap && entry->offset < min_start)) {
3249	node = rb_next(&entry->offset_index);
3250	if (!node)
3251	break;
3252	entry = rb_entry(node, struct btrfs_free_space,
3253	offset_index);
3254	continue;
3255	}
3256
3257	if (entry->bitmap) {
3258	ret = btrfs_alloc_from_bitmap(block_group,
3259	cluster, entry, bytes,
3260	min_start: cluster->window_start,
3261	max_extent_size);
3262	if (ret == 0) {
3263	node = rb_next(&entry->offset_index);
3264	if (!node)
3265	break;
3266	entry = rb_entry(node, struct btrfs_free_space,
3267	offset_index);
3268	continue;
3269	}
3270	cluster->window_start += bytes;
3271	} else {
3272	ret = entry->offset;
3273
3274	entry->offset += bytes;
3275	entry->bytes -= bytes;
3276	}
3277
3278	break;
3279	}
3280	out:
3281	spin_unlock(lock: &cluster->lock);
3282
3283	if (!ret)
3284	return 0;
3285
3286	spin_lock(lock: &ctl->tree_lock);
3287
3288	if (!btrfs_free_space_trimmed(info: entry))
3289	atomic64_add(i: bytes, v: &discard_ctl->discard_bytes_saved);
3290
3291	ctl->free_space -= bytes;
3292	if (!entry->bitmap && !btrfs_free_space_trimmed(info: entry))
3293	ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
3294
3295	spin_lock(lock: &cluster->lock);
3296	if (entry->bytes == 0) {
3297	rb_erase(&entry->offset_index, &cluster->root);
3298	ctl->free_extents--;
3299	if (entry->bitmap) {
3300	kmem_cache_free(s: btrfs_free_space_bitmap_cachep,
3301	objp: entry->bitmap);
3302	ctl->total_bitmaps--;
3303	recalculate_thresholds(ctl);
3304	} else if (!btrfs_free_space_trimmed(info: entry)) {
3305	ctl->discardable_extents[BTRFS_STAT_CURR]--;
3306	}
3307	kmem_cache_free(s: btrfs_free_space_cachep, objp: entry);
3308	}
3309
3310	spin_unlock(lock: &cluster->lock);
3311	spin_unlock(lock: &ctl->tree_lock);
3312
3313	return ret;
3314	}
3315
3316	static int btrfs_bitmap_cluster(struct btrfs_block_group *block_group,
3317	struct btrfs_free_space *entry,
3318	struct btrfs_free_cluster *cluster,
3319	u64 offset, u64 bytes,
3320	u64 cont1_bytes, u64 min_bytes)
3321	{
3322	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3323	unsigned long next_zero;
3324	unsigned long i;
3325	unsigned long want_bits;
3326	unsigned long min_bits;
3327	unsigned long found_bits;
3328	unsigned long max_bits = 0;
3329	unsigned long start = 0;
3330	unsigned long total_found = 0;
3331	int ret;
3332
3333	lockdep_assert_held(&ctl->tree_lock);
3334
3335	i = offset_to_bit(bitmap_start: entry->offset, unit: ctl->unit,
3336	max_t(u64, offset, entry->offset));
3337	want_bits = bytes_to_bits(bytes, unit: ctl->unit);
3338	min_bits = bytes_to_bits(bytes: min_bytes, unit: ctl->unit);
3339
3340	/*
3341	* Don't bother looking for a cluster in this bitmap if it's heavily
3342	* fragmented.
3343	*/
3344	if (entry->max_extent_size &&
3345	entry->max_extent_size < cont1_bytes)
3346	return -ENOSPC;
3347	again:
3348	found_bits = 0;
3349	for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
3350	next_zero = find_next_zero_bit(addr: entry->bitmap,
3351	BITS_PER_BITMAP, offset: i);
3352	if (next_zero - i >= min_bits) {
3353	found_bits = next_zero - i;
3354	if (found_bits > max_bits)
3355	max_bits = found_bits;
3356	break;
3357	}
3358	if (next_zero - i > max_bits)
3359	max_bits = next_zero - i;
3360	i = next_zero;
3361	}
3362
3363	if (!found_bits) {
3364	entry->max_extent_size = (u64)max_bits * ctl->unit;
3365	return -ENOSPC;
3366	}
3367
3368	if (!total_found) {
3369	start = i;
3370	cluster->max_size = 0;
3371	}
3372
3373	total_found += found_bits;
3374
3375	if (cluster->max_size < found_bits * ctl->unit)
3376	cluster->max_size = found_bits * ctl->unit;
3377
3378	if (total_found < want_bits || cluster->max_size < cont1_bytes) {
3379	i = next_zero + 1;
3380	goto again;
3381	}
3382
3383	cluster->window_start = start * ctl->unit + entry->offset;
3384	rb_erase(&entry->offset_index, &ctl->free_space_offset);
3385	rb_erase_cached(node: &entry->bytes_index, root: &ctl->free_space_bytes);
3386
3387	/*
3388	* We need to know if we're currently on the normal space index when we
3389	* manipulate the bitmap so that we know we need to remove and re-insert
3390	* it into the space_index tree. Clear the bytes_index node here so the
3391	* bitmap manipulation helpers know not to mess with the space_index
3392	* until this bitmap entry is added back into the normal cache.
3393	*/
3394	RB_CLEAR_NODE(&entry->bytes_index);
3395
3396	ret = tree_insert_offset(ctl, cluster, new_entry: entry);
3397	ASSERT(!ret); /* -EEXIST; Logic error */
3398
3399	trace_btrfs_setup_cluster(block_group, cluster,
3400	size: total_found * ctl->unit, bitmap: 1);
3401	return 0;
3402	}
3403
3404	/*
3405	* This searches the block group for just extents to fill the cluster with.
3406	* Try to find a cluster with at least bytes total bytes, at least one
3407	* extent of cont1_bytes, and other clusters of at least min_bytes.
3408	*/
3409	static noinline int
3410	setup_cluster_no_bitmap(struct btrfs_block_group *block_group,
3411	struct btrfs_free_cluster *cluster,
3412	struct list_head *bitmaps, u64 offset, u64 bytes,
3413	u64 cont1_bytes, u64 min_bytes)
3414	{
3415	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3416	struct btrfs_free_space *first = NULL;
3417	struct btrfs_free_space *entry = NULL;
3418	struct btrfs_free_space *last;
3419	struct rb_node *node;
3420	u64 window_free;
3421	u64 max_extent;
3422	u64 total_size = 0;
3423
3424	lockdep_assert_held(&ctl->tree_lock);
3425
3426	entry = tree_search_offset(ctl, offset, bitmap_only: 0, fuzzy: 1);
3427	if (!entry)
3428	return -ENOSPC;
3429
3430	/*
3431	* We don't want bitmaps, so just move along until we find a normal
3432	* extent entry.
3433	*/
3434	while (entry->bitmap || entry->bytes < min_bytes) {
3435	if (entry->bitmap && list_empty(head: &entry->list))
3436	list_add_tail(new: &entry->list, head: bitmaps);
3437	node = rb_next(&entry->offset_index);
3438	if (!node)
3439	return -ENOSPC;
3440	entry = rb_entry(node, struct btrfs_free_space, offset_index);
3441	}
3442
3443	window_free = entry->bytes;
3444	max_extent = entry->bytes;
3445	first = entry;
3446	last = entry;
3447
3448	for (node = rb_next(&entry->offset_index); node;
3449	node = rb_next(&entry->offset_index)) {
3450	entry = rb_entry(node, struct btrfs_free_space, offset_index);
3451
3452	if (entry->bitmap) {
3453	if (list_empty(head: &entry->list))
3454	list_add_tail(new: &entry->list, head: bitmaps);
3455	continue;
3456	}
3457
3458	if (entry->bytes < min_bytes)
3459	continue;
3460
3461	last = entry;
3462	window_free += entry->bytes;
3463	if (entry->bytes > max_extent)
3464	max_extent = entry->bytes;
3465	}
3466
3467	if (window_free < bytes || max_extent < cont1_bytes)
3468	return -ENOSPC;
3469
3470	cluster->window_start = first->offset;
3471
3472	node = &first->offset_index;
3473
3474	/*
3475	* now we've found our entries, pull them out of the free space
3476	* cache and put them into the cluster rbtree
3477	*/
3478	do {
3479	int ret;
3480
3481	entry = rb_entry(node, struct btrfs_free_space, offset_index);
3482	node = rb_next(&entry->offset_index);
3483	if (entry->bitmap || entry->bytes < min_bytes)
3484	continue;
3485
3486	rb_erase(&entry->offset_index, &ctl->free_space_offset);
3487	rb_erase_cached(node: &entry->bytes_index, root: &ctl->free_space_bytes);
3488	ret = tree_insert_offset(ctl, cluster, new_entry: entry);
3489	total_size += entry->bytes;
3490	ASSERT(!ret); /* -EEXIST; Logic error */
3491	} while (node && entry != last);
3492
3493	cluster->max_size = max_extent;
3494	trace_btrfs_setup_cluster(block_group, cluster, size: total_size, bitmap: 0);
3495	return 0;
3496	}
3497
3498	/*
3499	* This specifically looks for bitmaps that may work in the cluster, we assume
3500	* that we have already failed to find extents that will work.
3501	*/
3502	static noinline int
3503	setup_cluster_bitmap(struct btrfs_block_group *block_group,
3504	struct btrfs_free_cluster *cluster,
3505	struct list_head *bitmaps, u64 offset, u64 bytes,
3506	u64 cont1_bytes, u64 min_bytes)
3507	{
3508	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3509	struct btrfs_free_space *entry = NULL;
3510	int ret = -ENOSPC;
3511	u64 bitmap_offset = offset_to_bitmap(ctl, offset);
3512
3513	if (ctl->total_bitmaps == 0)
3514	return -ENOSPC;
3515
3516	/*
3517	* The bitmap that covers offset won't be in the list unless offset
3518	* is just its start offset.
3519	*/
3520	if (!list_empty(head: bitmaps))
3521	entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
3522
3523	if (!entry || entry->offset != bitmap_offset) {
3524	entry = tree_search_offset(ctl, offset: bitmap_offset, bitmap_only: 1, fuzzy: 0);
3525	if (entry && list_empty(head: &entry->list))
3526	list_add(new: &entry->list, head: bitmaps);
3527	}
3528
3529	list_for_each_entry(entry, bitmaps, list) {
3530	if (entry->bytes < bytes)
3531	continue;
3532	ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
3533	bytes, cont1_bytes, min_bytes);
3534	if (!ret)
3535	return 0;
3536	}
3537
3538	/*
3539	* The bitmaps list has all the bitmaps that record free space
3540	* starting after offset, so no more search is required.
3541	*/
3542	return -ENOSPC;
3543	}
3544
3545	/*
3546	* here we try to find a cluster of blocks in a block group. The goal
3547	* is to find at least bytes+empty_size.
3548	* We might not find them all in one contiguous area.
3549	*
3550	* returns zero and sets up cluster if things worked out, otherwise
3551	* it returns -enospc
3552	*/
3553	int btrfs_find_space_cluster(struct btrfs_block_group *block_group,
3554	struct btrfs_free_cluster *cluster,
3555	u64 offset, u64 bytes, u64 empty_size)
3556	{
3557	struct btrfs_fs_info *fs_info = block_group->fs_info;
3558	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3559	struct btrfs_free_space *entry, *tmp;
3560	LIST_HEAD(bitmaps);
3561	u64 min_bytes;
3562	u64 cont1_bytes;
3563	int ret;
3564
3565	/*
3566	* Choose the minimum extent size we'll require for this
3567	* cluster. For SSD_SPREAD, don't allow any fragmentation.
3568	* For metadata, allow allocates with smaller extents. For
3569	* data, keep it dense.
3570	*/
3571	if (btrfs_test_opt(fs_info, SSD_SPREAD)) {
3572	cont1_bytes = bytes + empty_size;
3573	min_bytes = cont1_bytes;
3574	} else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3575	cont1_bytes = bytes;
3576	min_bytes = fs_info->sectorsize;
3577	} else {
3578	cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3579	min_bytes = fs_info->sectorsize;
3580	}
3581
3582	spin_lock(lock: &ctl->tree_lock);
3583
3584	/*
3585	* If we know we don't have enough space to make a cluster don't even
3586	* bother doing all the work to try and find one.
3587	*/
3588	if (ctl->free_space < bytes) {
3589	spin_unlock(lock: &ctl->tree_lock);
3590	return -ENOSPC;
3591	}
3592
3593	spin_lock(lock: &cluster->lock);
3594
3595	/* someone already found a cluster, hooray */
3596	if (cluster->block_group) {
3597	ret = 0;
3598	goto out;
3599	}
3600
3601	trace_btrfs_find_cluster(block_group, start: offset, bytes, empty_size,
3602	min_bytes);
3603
3604	ret = setup_cluster_no_bitmap(block_group, cluster, bitmaps: &bitmaps, offset,
3605	bytes: bytes + empty_size,
3606	cont1_bytes, min_bytes);
3607	if (ret)
3608	ret = setup_cluster_bitmap(block_group, cluster, bitmaps: &bitmaps,
3609	offset, bytes: bytes + empty_size,
3610	cont1_bytes, min_bytes);
3611
3612	/* Clear our temporary list */
3613	list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3614	list_del_init(entry: &entry->list);
3615
3616	if (!ret) {
3617	btrfs_get_block_group(cache: block_group);
3618	list_add_tail(new: &cluster->block_group_list,
3619	head: &block_group->cluster_list);
3620	cluster->block_group = block_group;
3621	} else {
3622	trace_btrfs_failed_cluster_setup(block_group);
3623	}
3624	out:
3625	spin_unlock(lock: &cluster->lock);
3626	spin_unlock(lock: &ctl->tree_lock);
3627
3628	return ret;
3629	}
3630
3631	/*
3632	* simple code to zero out a cluster
3633	*/
3634	void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3635	{
3636	spin_lock_init(&cluster->lock);
3637	spin_lock_init(&cluster->refill_lock);
3638	cluster->root = RB_ROOT;
3639	cluster->max_size = 0;
3640	cluster->fragmented = false;
3641	INIT_LIST_HEAD(list: &cluster->block_group_list);
3642	cluster->block_group = NULL;
3643	}
3644
3645	static int do_trimming(struct btrfs_block_group *block_group,
3646	u64 *total_trimmed, u64 start, u64 bytes,
3647	u64 reserved_start, u64 reserved_bytes,
3648	enum btrfs_trim_state reserved_trim_state,
3649	struct btrfs_trim_range *trim_entry)
3650	{
3651	struct btrfs_space_info *space_info = block_group->space_info;
3652	struct btrfs_fs_info *fs_info = block_group->fs_info;
3653	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3654	int ret;
3655	int update = 0;
3656	const u64 end = start + bytes;
3657	const u64 reserved_end = reserved_start + reserved_bytes;
3658	enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3659	u64 trimmed = 0;
3660
3661	spin_lock(lock: &space_info->lock);
3662	spin_lock(lock: &block_group->lock);
3663	if (!block_group->ro) {
3664	block_group->reserved += reserved_bytes;
3665	space_info->bytes_reserved += reserved_bytes;
3666	update = 1;
3667	}
3668	spin_unlock(lock: &block_group->lock);
3669	spin_unlock(lock: &space_info->lock);
3670
3671	ret = btrfs_discard_extent(fs_info, bytenr: start, num_bytes: bytes, actual_bytes: &trimmed);
3672	if (!ret) {
3673	*total_trimmed += trimmed;
3674	trim_state = BTRFS_TRIM_STATE_TRIMMED;
3675	}
3676
3677	mutex_lock(&ctl->cache_writeout_mutex);
3678	if (reserved_start < start)
3679	__btrfs_add_free_space(block_group, offset: reserved_start,
3680	bytes: start - reserved_start,
3681	trim_state: reserved_trim_state);
3682	if (end < reserved_end)
3683	__btrfs_add_free_space(block_group, offset: end, bytes: reserved_end - end,
3684	trim_state: reserved_trim_state);
3685	__btrfs_add_free_space(block_group, offset: start, bytes, trim_state);
3686	list_del(entry: &trim_entry->list);
3687	mutex_unlock(lock: &ctl->cache_writeout_mutex);
3688
3689	if (update) {
3690	spin_lock(lock: &space_info->lock);
3691	spin_lock(lock: &block_group->lock);
3692	if (block_group->ro)
3693	space_info->bytes_readonly += reserved_bytes;
3694	block_group->reserved -= reserved_bytes;
3695	space_info->bytes_reserved -= reserved_bytes;
3696	spin_unlock(lock: &block_group->lock);
3697	spin_unlock(lock: &space_info->lock);
3698	}
3699
3700	return ret;
3701	}
3702
3703	/*
3704	* If @async is set, then we will trim 1 region and return.
3705	*/
3706	static int trim_no_bitmap(struct btrfs_block_group *block_group,
3707	u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3708	bool async)
3709	{
3710	struct btrfs_discard_ctl *discard_ctl =
3711	&block_group->fs_info->discard_ctl;
3712	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3713	struct btrfs_free_space *entry;
3714	struct rb_node *node;
3715	int ret = 0;
3716	u64 extent_start;
3717	u64 extent_bytes;
3718	enum btrfs_trim_state extent_trim_state;
3719	u64 bytes;
3720	const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3721
3722	while (start < end) {
3723	struct btrfs_trim_range trim_entry;
3724
3725	mutex_lock(&ctl->cache_writeout_mutex);
3726	spin_lock(lock: &ctl->tree_lock);
3727
3728	if (ctl->free_space < minlen)
3729	goto out_unlock;
3730
3731	entry = tree_search_offset(ctl, offset: start, bitmap_only: 0, fuzzy: 1);
3732	if (!entry)
3733	goto out_unlock;
3734
3735	/* Skip bitmaps and if async, already trimmed entries */
3736	while (entry->bitmap ||
3737	(async && btrfs_free_space_trimmed(info: entry))) {
3738	node = rb_next(&entry->offset_index);
3739	if (!node)
3740	goto out_unlock;
3741	entry = rb_entry(node, struct btrfs_free_space,
3742	offset_index);
3743	}
3744
3745	if (entry->offset >= end)
3746	goto out_unlock;
3747
3748	extent_start = entry->offset;
3749	extent_bytes = entry->bytes;
3750	extent_trim_state = entry->trim_state;
3751	if (async) {
3752	start = entry->offset;
3753	bytes = entry->bytes;
3754	if (bytes < minlen) {
3755	spin_unlock(lock: &ctl->tree_lock);
3756	mutex_unlock(lock: &ctl->cache_writeout_mutex);
3757	goto next;
3758	}
3759	unlink_free_space(ctl, info: entry, update_stat: true);
3760	/*
3761	* Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3762	* If X < BTRFS_ASYNC_DISCARD_MIN_FILTER, we won't trim
3763	* X when we come back around. So trim it now.
3764	*/
3765	if (max_discard_size &&
3766	bytes >= (max_discard_size +
3767	BTRFS_ASYNC_DISCARD_MIN_FILTER)) {
3768	bytes = max_discard_size;
3769	extent_bytes = max_discard_size;
3770	entry->offset += max_discard_size;
3771	entry->bytes -= max_discard_size;
3772	link_free_space(ctl, info: entry);
3773	} else {
3774	kmem_cache_free(s: btrfs_free_space_cachep, objp: entry);
3775	}
3776	} else {
3777	start = max(start, extent_start);
3778	bytes = min(extent_start + extent_bytes, end) - start;
3779	if (bytes < minlen) {
3780	spin_unlock(lock: &ctl->tree_lock);
3781	mutex_unlock(lock: &ctl->cache_writeout_mutex);
3782	goto next;
3783	}
3784
3785	unlink_free_space(ctl, info: entry, update_stat: true);
3786	kmem_cache_free(s: btrfs_free_space_cachep, objp: entry);
3787	}
3788
3789	spin_unlock(lock: &ctl->tree_lock);
3790	trim_entry.start = extent_start;
3791	trim_entry.bytes = extent_bytes;
3792	list_add_tail(new: &trim_entry.list, head: &ctl->trimming_ranges);
3793	mutex_unlock(lock: &ctl->cache_writeout_mutex);
3794
3795	ret = do_trimming(block_group, total_trimmed, start, bytes,
3796	reserved_start: extent_start, reserved_bytes: extent_bytes, reserved_trim_state: extent_trim_state,
3797	trim_entry: &trim_entry);
3798	if (ret) {
3799	block_group->discard_cursor = start + bytes;
3800	break;
3801	}
3802	next:
3803	start += bytes;
3804	block_group->discard_cursor = start;
3805	if (async && *total_trimmed)
3806	break;
3807
3808	if (fatal_signal_pending(current)) {
3809	ret = -ERESTARTSYS;
3810	break;
3811	}
3812
3813	cond_resched();
3814	}
3815
3816	return ret;
3817
3818	out_unlock:
3819	block_group->discard_cursor = btrfs_block_group_end(block_group);
3820	spin_unlock(lock: &ctl->tree_lock);
3821	mutex_unlock(lock: &ctl->cache_writeout_mutex);
3822
3823	return ret;
3824	}
3825
3826	/*
3827	* If we break out of trimming a bitmap prematurely, we should reset the
3828	* trimming bit. In a rather contrieved case, it's possible to race here so
3829	* reset the state to BTRFS_TRIM_STATE_UNTRIMMED.
3830	*
3831	* start = start of bitmap
3832	* end = near end of bitmap
3833	*
3834	* Thread 1: Thread 2:
3835	* trim_bitmaps(start)
3836	* trim_bitmaps(end)
3837	* end_trimming_bitmap()
3838	* reset_trimming_bitmap()
3839	*/
3840	static void reset_trimming_bitmap(struct btrfs_free_space_ctl *ctl, u64 offset)
3841	{
3842	struct btrfs_free_space *entry;
3843
3844	spin_lock(lock: &ctl->tree_lock);
3845	entry = tree_search_offset(ctl, offset, bitmap_only: 1, fuzzy: 0);
3846	if (entry) {
3847	if (btrfs_free_space_trimmed(info: entry)) {
3848	ctl->discardable_extents[BTRFS_STAT_CURR] +=
3849	entry->bitmap_extents;
3850	ctl->discardable_bytes[BTRFS_STAT_CURR] += entry->bytes;
3851	}
3852	entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3853	}
3854
3855	spin_unlock(lock: &ctl->tree_lock);
3856	}
3857
3858	static void end_trimming_bitmap(struct btrfs_free_space_ctl *ctl,
3859	struct btrfs_free_space *entry)
3860	{
3861	if (btrfs_free_space_trimming_bitmap(info: entry)) {
3862	entry->trim_state = BTRFS_TRIM_STATE_TRIMMED;
3863	ctl->discardable_extents[BTRFS_STAT_CURR] -=
3864	entry->bitmap_extents;
3865	ctl->discardable_bytes[BTRFS_STAT_CURR] -= entry->bytes;
3866	}
3867	}
3868
3869	/*
3870	* If @async is set, then we will trim 1 region and return.
3871	*/
3872	static int trim_bitmaps(struct btrfs_block_group *block_group,
3873	u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3874	u64 maxlen, bool async)
3875	{
3876	struct btrfs_discard_ctl *discard_ctl =
3877	&block_group->fs_info->discard_ctl;
3878	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3879	struct btrfs_free_space *entry;
3880	int ret = 0;
3881	int ret2;
3882	u64 bytes;
3883	u64 offset = offset_to_bitmap(ctl, offset: start);
3884	const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3885
3886	while (offset < end) {
3887	bool next_bitmap = false;
3888	struct btrfs_trim_range trim_entry;
3889
3890	mutex_lock(&ctl->cache_writeout_mutex);
3891	spin_lock(lock: &ctl->tree_lock);
3892
3893	if (ctl->free_space < minlen) {
3894	block_group->discard_cursor =
3895	btrfs_block_group_end(block_group);
3896	spin_unlock(lock: &ctl->tree_lock);
3897	mutex_unlock(lock: &ctl->cache_writeout_mutex);
3898	break;
3899	}
3900
3901	entry = tree_search_offset(ctl, offset, bitmap_only: 1, fuzzy: 0);
3902	/*
3903	* Bitmaps are marked trimmed lossily now to prevent constant
3904	* discarding of the same bitmap (the reason why we are bound
3905	* by the filters). So, retrim the block group bitmaps when we
3906	* are preparing to punt to the unused_bgs list. This uses
3907	* @minlen to determine if we are in BTRFS_DISCARD_INDEX_UNUSED
3908	* which is the only discard index which sets minlen to 0.
3909	*/
3910	if (!entry || (async && minlen && start == offset &&
3911	btrfs_free_space_trimmed(info: entry))) {
3912	spin_unlock(lock: &ctl->tree_lock);
3913	mutex_unlock(lock: &ctl->cache_writeout_mutex);
3914	next_bitmap = true;
3915	goto next;
3916	}
3917
3918	/*
3919	* Async discard bitmap trimming begins at by setting the start
3920	* to be key.objectid and the offset_to_bitmap() aligns to the
3921	* start of the bitmap. This lets us know we are fully
3922	* scanning the bitmap rather than only some portion of it.
3923	*/
3924	if (start == offset)
3925	entry->trim_state = BTRFS_TRIM_STATE_TRIMMING;
3926
3927	bytes = minlen;
3928	ret2 = search_bitmap(ctl, bitmap_info: entry, offset: &start, bytes: &bytes, for_alloc: false);
3929	if (ret2 || start >= end) {
3930	/*
3931	* We lossily consider a bitmap trimmed if we only skip
3932	* over regions <= BTRFS_ASYNC_DISCARD_MIN_FILTER.
3933	*/
3934	if (ret2 && minlen <= BTRFS_ASYNC_DISCARD_MIN_FILTER)
3935	end_trimming_bitmap(ctl, entry);
3936	else
3937	entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3938	spin_unlock(lock: &ctl->tree_lock);
3939	mutex_unlock(lock: &ctl->cache_writeout_mutex);
3940	next_bitmap = true;
3941	goto next;
3942	}
3943
3944	/*
3945	* We already trimmed a region, but are using the locking above
3946	* to reset the trim_state.
3947	*/
3948	if (async && *total_trimmed) {
3949	spin_unlock(lock: &ctl->tree_lock);
3950	mutex_unlock(lock: &ctl->cache_writeout_mutex);
3951	goto out;
3952	}
3953
3954	bytes = min(bytes, end - start);
3955	if (bytes < minlen || (async && maxlen && bytes > maxlen)) {
3956	spin_unlock(lock: &ctl->tree_lock);
3957	mutex_unlock(lock: &ctl->cache_writeout_mutex);
3958	goto next;
3959	}
3960
3961	/*
3962	* Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3963	* If X < @minlen, we won't trim X when we come back around.
3964	* So trim it now. We differ here from trimming extents as we
3965	* don't keep individual state per bit.
3966	*/
3967	if (async &&
3968	max_discard_size &&
3969	bytes > (max_discard_size + minlen))
3970	bytes = max_discard_size;
3971
3972	bitmap_clear_bits(ctl, info: entry, offset: start, bytes, update_stat: true);
3973	if (entry->bytes == 0)
3974	free_bitmap(ctl, bitmap_info: entry);
3975
3976	spin_unlock(lock: &ctl->tree_lock);
3977	trim_entry.start = start;
3978	trim_entry.bytes = bytes;
3979	list_add_tail(new: &trim_entry.list, head: &ctl->trimming_ranges);
3980	mutex_unlock(lock: &ctl->cache_writeout_mutex);
3981
3982	ret = do_trimming(block_group, total_trimmed, start, bytes,
3983	reserved_start: start, reserved_bytes: bytes, reserved_trim_state: 0, trim_entry: &trim_entry);
3984	if (ret) {
3985	reset_trimming_bitmap(ctl, offset);
3986	block_group->discard_cursor =
3987	btrfs_block_group_end(block_group);
3988	break;
3989	}
3990	next:
3991	if (next_bitmap) {
3992	offset += BITS_PER_BITMAP * ctl->unit;
3993	start = offset;
3994	} else {
3995	start += bytes;
3996	}
3997	block_group->discard_cursor = start;
3998
3999	if (fatal_signal_pending(current)) {
4000	if (start != offset)
4001	reset_trimming_bitmap(ctl, offset);
4002	ret = -ERESTARTSYS;
4003	break;
4004	}
4005
4006	cond_resched();
4007	}
4008
4009	if (offset >= end)
4010	block_group->discard_cursor = end;
4011
4012	out:
4013	return ret;
4014	}
4015
4016	int btrfs_trim_block_group(struct btrfs_block_group *block_group,
4017	u64 *trimmed, u64 start, u64 end, u64 minlen)
4018	{
4019	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
4020	int ret;
4021	u64 rem = 0;
4022
4023	ASSERT(!btrfs_is_zoned(block_group->fs_info));
4024
4025	*trimmed = 0;
4026
4027	spin_lock(lock: &block_group->lock);
4028	if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4029	spin_unlock(lock: &block_group->lock);
4030	return 0;
4031	}
4032	btrfs_freeze_block_group(cache: block_group);
4033	spin_unlock(lock: &block_group->lock);
4034
4035	ret = trim_no_bitmap(block_group, total_trimmed: trimmed, start, end, minlen, async: false);
4036	if (ret)
4037	goto out;
4038
4039	ret = trim_bitmaps(block_group, total_trimmed: trimmed, start, end, minlen, maxlen: 0, async: false);
4040	div64_u64_rem(dividend: end, BITS_PER_BITMAP * ctl->unit, remainder: &rem);
4041	/* If we ended in the middle of a bitmap, reset the trimming flag */
4042	if (rem)
4043	reset_trimming_bitmap(ctl, offset: offset_to_bitmap(ctl, offset: end));
4044	out:
4045	btrfs_unfreeze_block_group(cache: block_group);
4046	return ret;
4047	}
4048
4049	int btrfs_trim_block_group_extents(struct btrfs_block_group *block_group,
4050	u64 *trimmed, u64 start, u64 end, u64 minlen,
4051	bool async)
4052	{
4053	int ret;
4054
4055	*trimmed = 0;
4056
4057	spin_lock(lock: &block_group->lock);
4058	if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4059	spin_unlock(lock: &block_group->lock);
4060	return 0;
4061	}
4062	btrfs_freeze_block_group(cache: block_group);
4063	spin_unlock(lock: &block_group->lock);
4064
4065	ret = trim_no_bitmap(block_group, total_trimmed: trimmed, start, end, minlen, async);
4066	btrfs_unfreeze_block_group(cache: block_group);
4067
4068	return ret;
4069	}
4070
4071	int btrfs_trim_block_group_bitmaps(struct btrfs_block_group *block_group,
4072	u64 *trimmed, u64 start, u64 end, u64 minlen,
4073	u64 maxlen, bool async)
4074	{
4075	int ret;
4076
4077	*trimmed = 0;
4078
4079	spin_lock(lock: &block_group->lock);
4080	if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4081	spin_unlock(lock: &block_group->lock);
4082	return 0;
4083	}
4084	btrfs_freeze_block_group(cache: block_group);
4085	spin_unlock(lock: &block_group->lock);
4086
4087	ret = trim_bitmaps(block_group, total_trimmed: trimmed, start, end, minlen, maxlen,
4088	async);
4089
4090	btrfs_unfreeze_block_group(cache: block_group);
4091
4092	return ret;
4093	}
4094
4095	bool btrfs_free_space_cache_v1_active(struct btrfs_fs_info *fs_info)
4096	{
4097	return btrfs_super_cache_generation(s: fs_info->super_copy);
4098	}
4099
4100	static int cleanup_free_space_cache_v1(struct btrfs_fs_info *fs_info,
4101	struct btrfs_trans_handle *trans)
4102	{
4103	struct btrfs_block_group *block_group;
4104	struct rb_node *node;
4105	int ret = 0;
4106
4107	btrfs_info(fs_info, "cleaning free space cache v1");
4108
4109	node = rb_first_cached(&fs_info->block_group_cache_tree);
4110	while (node) {
4111	block_group = rb_entry(node, struct btrfs_block_group, cache_node);
4112	ret = btrfs_remove_free_space_inode(trans, NULL, block_group);
4113	if (ret)
4114	goto out;
4115	node = rb_next(node);
4116	}
4117	out:
4118	return ret;
4119	}
4120
4121	int btrfs_set_free_space_cache_v1_active(struct btrfs_fs_info *fs_info, bool active)
4122	{
4123	struct btrfs_trans_handle *trans;
4124	int ret;
4125
4126	/*
4127	* update_super_roots will appropriately set or unset
4128	* super_copy->cache_generation based on SPACE_CACHE and
4129	* BTRFS_FS_CLEANUP_SPACE_CACHE_V1. For this reason, we need a
4130	* transaction commit whether we are enabling space cache v1 and don't
4131	* have any other work to do, or are disabling it and removing free
4132	* space inodes.
4133	*/
4134	trans = btrfs_start_transaction(root: fs_info->tree_root, num_items: 0);
4135	if (IS_ERR(ptr: trans))
4136	return PTR_ERR(ptr: trans);
4137
4138	if (!active) {
4139	set_bit(nr: BTRFS_FS_CLEANUP_SPACE_CACHE_V1, addr: &fs_info->flags);
4140	ret = cleanup_free_space_cache_v1(fs_info, trans);
4141	if (ret) {
4142	btrfs_abort_transaction(trans, ret);
4143	btrfs_end_transaction(trans);
4144	goto out;
4145	}
4146	}
4147
4148	ret = btrfs_commit_transaction(trans);
4149	out:
4150	clear_bit(nr: BTRFS_FS_CLEANUP_SPACE_CACHE_V1, addr: &fs_info->flags);
4151
4152	return ret;
4153	}
4154
4155	int __init btrfs_free_space_init(void)
4156	{
4157	btrfs_free_space_cachep = KMEM_CACHE(btrfs_free_space, 0);
4158	if (!btrfs_free_space_cachep)
4159	return -ENOMEM;
4160
4161	btrfs_free_space_bitmap_cachep = kmem_cache_create(name: "btrfs_free_space_bitmap",
4162	PAGE_SIZE, PAGE_SIZE,
4163	flags: 0, NULL);
4164	if (!btrfs_free_space_bitmap_cachep) {
4165	kmem_cache_destroy(s: btrfs_free_space_cachep);
4166	return -ENOMEM;
4167	}
4168
4169	return 0;
4170	}
4171
4172	void __cold btrfs_free_space_exit(void)
4173	{
4174	kmem_cache_destroy(s: btrfs_free_space_cachep);
4175	kmem_cache_destroy(s: btrfs_free_space_bitmap_cachep);
4176	}
4177
4178	#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4179	/*
4180	* Use this if you need to make a bitmap or extent entry specifically, it
4181	* doesn't do any of the merging that add_free_space does, this acts a lot like
4182	* how the free space cache loading stuff works, so you can get really weird
4183	* configurations.
4184	*/
4185	int test_add_free_space_entry(struct btrfs_block_group *cache,
4186	u64 offset, u64 bytes, bool bitmap)
4187	{
4188	struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4189	struct btrfs_free_space *info = NULL, *bitmap_info;
4190	void *map = NULL;
4191	enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_TRIMMED;
4192	u64 bytes_added;
4193	int ret;
4194
4195	again:
4196	if (!info) {
4197	info = kmem_cache_zalloc(k: btrfs_free_space_cachep, GFP_NOFS);
4198	if (!info)
4199	return -ENOMEM;
4200	}
4201
4202	if (!bitmap) {
4203	spin_lock(lock: &ctl->tree_lock);
4204	info->offset = offset;
4205	info->bytes = bytes;
4206	info->max_extent_size = 0;
4207	ret = link_free_space(ctl, info);
4208	spin_unlock(lock: &ctl->tree_lock);
4209	if (ret)
4210	kmem_cache_free(s: btrfs_free_space_cachep, objp: info);
4211	return ret;
4212	}
4213
4214	if (!map) {
4215	map = kmem_cache_zalloc(k: btrfs_free_space_bitmap_cachep, GFP_NOFS);
4216	if (!map) {
4217	kmem_cache_free(s: btrfs_free_space_cachep, objp: info);
4218	return -ENOMEM;
4219	}
4220	}
4221
4222	spin_lock(lock: &ctl->tree_lock);
4223	bitmap_info = tree_search_offset(ctl, offset: offset_to_bitmap(ctl, offset),
4224	bitmap_only: 1, fuzzy: 0);
4225	if (!bitmap_info) {
4226	info->bitmap = map;
4227	map = NULL;
4228	add_new_bitmap(ctl, info, offset);
4229	bitmap_info = info;
4230	info = NULL;
4231	}
4232
4233	bytes_added = add_bytes_to_bitmap(ctl, info: bitmap_info, offset, bytes,
4234	trim_state);
4235
4236	bytes -= bytes_added;
4237	offset += bytes_added;
4238	spin_unlock(lock: &ctl->tree_lock);
4239
4240	if (bytes)
4241	goto again;
4242
4243	if (info)
4244	kmem_cache_free(s: btrfs_free_space_cachep, objp: info);
4245	if (map)
4246	kmem_cache_free(s: btrfs_free_space_bitmap_cachep, objp: map);
4247	return 0;
4248	}
4249
4250	/*
4251	* Checks to see if the given range is in the free space cache. This is really
4252	* just used to check the absence of space, so if there is free space in the
4253	* range at all we will return 1.
4254	*/
4255	int test_check_exists(struct btrfs_block_group *cache,
4256	u64 offset, u64 bytes)
4257	{
4258	struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4259	struct btrfs_free_space *info;
4260	int ret = 0;
4261
4262	spin_lock(lock: &ctl->tree_lock);
4263	info = tree_search_offset(ctl, offset, bitmap_only: 0, fuzzy: 0);
4264	if (!info) {
4265	info = tree_search_offset(ctl, offset: offset_to_bitmap(ctl, offset),
4266	bitmap_only: 1, fuzzy: 0);
4267	if (!info)
4268	goto out;
4269	}
4270
4271	have_info:
4272	if (info->bitmap) {
4273	u64 bit_off, bit_bytes;
4274	struct rb_node *n;
4275	struct btrfs_free_space *tmp;
4276
4277	bit_off = offset;
4278	bit_bytes = ctl->unit;
4279	ret = search_bitmap(ctl, bitmap_info: info, offset: &bit_off, bytes: &bit_bytes, for_alloc: false);
4280	if (!ret) {
4281	if (bit_off == offset) {
4282	ret = 1;
4283	goto out;
4284	} else if (bit_off > offset &&
4285	offset + bytes > bit_off) {
4286	ret = 1;
4287	goto out;
4288	}
4289	}
4290
4291	n = rb_prev(&info->offset_index);
4292	while (n) {
4293	tmp = rb_entry(n, struct btrfs_free_space,
4294	offset_index);
4295	if (tmp->offset + tmp->bytes < offset)
4296	break;
4297	if (offset + bytes < tmp->offset) {
4298	n = rb_prev(&tmp->offset_index);
4299	continue;
4300	}
4301	info = tmp;
4302	goto have_info;
4303	}
4304
4305	n = rb_next(&info->offset_index);
4306	while (n) {
4307	tmp = rb_entry(n, struct btrfs_free_space,
4308	offset_index);
4309	if (offset + bytes < tmp->offset)
4310	break;
4311	if (tmp->offset + tmp->bytes < offset) {
4312	n = rb_next(&tmp->offset_index);
4313	continue;
4314	}
4315	info = tmp;
4316	goto have_info;
4317	}
4318
4319	ret = 0;
4320	goto out;
4321	}
4322
4323	if (info->offset == offset) {
4324	ret = 1;
4325	goto out;
4326	}
4327
4328	if (offset > info->offset && offset < info->offset + info->bytes)
4329	ret = 1;
4330	out:
4331	spin_unlock(lock: &ctl->tree_lock);
4332	return ret;
4333	}
4334	#endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */
4335