1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
28 #include "hash.h"
29 #include "tree-log.h"
30 #include "disk-io.h"
31 #include "print-tree.h"
32 #include "volumes.h"
33 #include "raid56.h"
34 #include "locking.h"
35 #include "free-space-cache.h"
36 #include "math.h"
37 #include "sysfs.h"
38 #include "qgroup.h"
39 
40 #undef SCRAMBLE_DELAYED_REFS
41 
42 /*
43  * control flags for do_chunk_alloc's force field
44  * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45  * if we really need one.
46  *
47  * CHUNK_ALLOC_LIMITED means to only try and allocate one
48  * if we have very few chunks already allocated.  This is
49  * used as part of the clustering code to help make sure
50  * we have a good pool of storage to cluster in, without
51  * filling the FS with empty chunks
52  *
53  * CHUNK_ALLOC_FORCE means it must try to allocate one
54  *
55  */
56 enum {
57 	CHUNK_ALLOC_NO_FORCE = 0,
58 	CHUNK_ALLOC_LIMITED = 1,
59 	CHUNK_ALLOC_FORCE = 2,
60 };
61 
62 /*
63  * Control how reservations are dealt with.
64  *
65  * RESERVE_FREE - freeing a reservation.
66  * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
67  *   ENOSPC accounting
68  * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69  *   bytes_may_use as the ENOSPC accounting is done elsewhere
70  */
71 enum {
72 	RESERVE_FREE = 0,
73 	RESERVE_ALLOC = 1,
74 	RESERVE_ALLOC_NO_ACCOUNT = 2,
75 };
76 
77 static int update_block_group(struct btrfs_trans_handle *trans,
78 			      struct btrfs_root *root, u64 bytenr,
79 			      u64 num_bytes, int alloc);
80 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
81 				struct btrfs_root *root,
82 				u64 bytenr, u64 num_bytes, u64 parent,
83 				u64 root_objectid, u64 owner_objectid,
84 				u64 owner_offset, int refs_to_drop,
85 				struct btrfs_delayed_extent_op *extra_op,
86 				int no_quota);
87 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
88 				    struct extent_buffer *leaf,
89 				    struct btrfs_extent_item *ei);
90 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
91 				      struct btrfs_root *root,
92 				      u64 parent, u64 root_objectid,
93 				      u64 flags, u64 owner, u64 offset,
94 				      struct btrfs_key *ins, int ref_mod);
95 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
96 				     struct btrfs_root *root,
97 				     u64 parent, u64 root_objectid,
98 				     u64 flags, struct btrfs_disk_key *key,
99 				     int level, struct btrfs_key *ins,
100 				     int no_quota);
101 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
102 			  struct btrfs_root *extent_root, u64 flags,
103 			  int force);
104 static int find_next_key(struct btrfs_path *path, int level,
105 			 struct btrfs_key *key);
106 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
107 			    int dump_block_groups);
108 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
109 				       u64 num_bytes, int reserve,
110 				       int delalloc);
111 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
112 			       u64 num_bytes);
113 int btrfs_pin_extent(struct btrfs_root *root,
114 		     u64 bytenr, u64 num_bytes, int reserved);
115 
116 static noinline int
block_group_cache_done(struct btrfs_block_group_cache * cache)117 block_group_cache_done(struct btrfs_block_group_cache *cache)
118 {
119 	smp_mb();
120 	return cache->cached == BTRFS_CACHE_FINISHED ||
121 		cache->cached == BTRFS_CACHE_ERROR;
122 }
123 
block_group_bits(struct btrfs_block_group_cache * cache,u64 bits)124 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
125 {
126 	return (cache->flags & bits) == bits;
127 }
128 
btrfs_get_block_group(struct btrfs_block_group_cache * cache)129 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
130 {
131 	atomic_inc(&cache->count);
132 }
133 
btrfs_put_block_group(struct btrfs_block_group_cache * cache)134 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
135 {
136 	if (atomic_dec_and_test(&cache->count)) {
137 		WARN_ON(cache->pinned > 0);
138 		WARN_ON(cache->reserved > 0);
139 		kfree(cache->free_space_ctl);
140 		kfree(cache);
141 	}
142 }
143 
144 /*
145  * this adds the block group to the fs_info rb tree for the block group
146  * cache
147  */
btrfs_add_block_group_cache(struct btrfs_fs_info * info,struct btrfs_block_group_cache * block_group)148 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
149 				struct btrfs_block_group_cache *block_group)
150 {
151 	struct rb_node **p;
152 	struct rb_node *parent = NULL;
153 	struct btrfs_block_group_cache *cache;
154 
155 	spin_lock(&info->block_group_cache_lock);
156 	p = &info->block_group_cache_tree.rb_node;
157 
158 	while (*p) {
159 		parent = *p;
160 		cache = rb_entry(parent, struct btrfs_block_group_cache,
161 				 cache_node);
162 		if (block_group->key.objectid < cache->key.objectid) {
163 			p = &(*p)->rb_left;
164 		} else if (block_group->key.objectid > cache->key.objectid) {
165 			p = &(*p)->rb_right;
166 		} else {
167 			spin_unlock(&info->block_group_cache_lock);
168 			return -EEXIST;
169 		}
170 	}
171 
172 	rb_link_node(&block_group->cache_node, parent, p);
173 	rb_insert_color(&block_group->cache_node,
174 			&info->block_group_cache_tree);
175 
176 	if (info->first_logical_byte > block_group->key.objectid)
177 		info->first_logical_byte = block_group->key.objectid;
178 
179 	spin_unlock(&info->block_group_cache_lock);
180 
181 	return 0;
182 }
183 
184 /*
185  * This will return the block group at or after bytenr if contains is 0, else
186  * it will return the block group that contains the bytenr
187  */
188 static struct btrfs_block_group_cache *
block_group_cache_tree_search(struct btrfs_fs_info * info,u64 bytenr,int contains)189 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
190 			      int contains)
191 {
192 	struct btrfs_block_group_cache *cache, *ret = NULL;
193 	struct rb_node *n;
194 	u64 end, start;
195 
196 	spin_lock(&info->block_group_cache_lock);
197 	n = info->block_group_cache_tree.rb_node;
198 
199 	while (n) {
200 		cache = rb_entry(n, struct btrfs_block_group_cache,
201 				 cache_node);
202 		end = cache->key.objectid + cache->key.offset - 1;
203 		start = cache->key.objectid;
204 
205 		if (bytenr < start) {
206 			if (!contains && (!ret || start < ret->key.objectid))
207 				ret = cache;
208 			n = n->rb_left;
209 		} else if (bytenr > start) {
210 			if (contains && bytenr <= end) {
211 				ret = cache;
212 				break;
213 			}
214 			n = n->rb_right;
215 		} else {
216 			ret = cache;
217 			break;
218 		}
219 	}
220 	if (ret) {
221 		btrfs_get_block_group(ret);
222 		if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
223 			info->first_logical_byte = ret->key.objectid;
224 	}
225 	spin_unlock(&info->block_group_cache_lock);
226 
227 	return ret;
228 }
229 
add_excluded_extent(struct btrfs_root * root,u64 start,u64 num_bytes)230 static int add_excluded_extent(struct btrfs_root *root,
231 			       u64 start, u64 num_bytes)
232 {
233 	u64 end = start + num_bytes - 1;
234 	set_extent_bits(&root->fs_info->freed_extents[0],
235 			start, end, EXTENT_UPTODATE, GFP_NOFS);
236 	set_extent_bits(&root->fs_info->freed_extents[1],
237 			start, end, EXTENT_UPTODATE, GFP_NOFS);
238 	return 0;
239 }
240 
free_excluded_extents(struct btrfs_root * root,struct btrfs_block_group_cache * cache)241 static void free_excluded_extents(struct btrfs_root *root,
242 				  struct btrfs_block_group_cache *cache)
243 {
244 	u64 start, end;
245 
246 	start = cache->key.objectid;
247 	end = start + cache->key.offset - 1;
248 
249 	clear_extent_bits(&root->fs_info->freed_extents[0],
250 			  start, end, EXTENT_UPTODATE, GFP_NOFS);
251 	clear_extent_bits(&root->fs_info->freed_extents[1],
252 			  start, end, EXTENT_UPTODATE, GFP_NOFS);
253 }
254 
exclude_super_stripes(struct btrfs_root * root,struct btrfs_block_group_cache * cache)255 static int exclude_super_stripes(struct btrfs_root *root,
256 				 struct btrfs_block_group_cache *cache)
257 {
258 	u64 bytenr;
259 	u64 *logical;
260 	int stripe_len;
261 	int i, nr, ret;
262 
263 	if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
264 		stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
265 		cache->bytes_super += stripe_len;
266 		ret = add_excluded_extent(root, cache->key.objectid,
267 					  stripe_len);
268 		if (ret)
269 			return ret;
270 	}
271 
272 	for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
273 		bytenr = btrfs_sb_offset(i);
274 		ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
275 				       cache->key.objectid, bytenr,
276 				       0, &logical, &nr, &stripe_len);
277 		if (ret)
278 			return ret;
279 
280 		while (nr--) {
281 			u64 start, len;
282 
283 			if (logical[nr] > cache->key.objectid +
284 			    cache->key.offset)
285 				continue;
286 
287 			if (logical[nr] + stripe_len <= cache->key.objectid)
288 				continue;
289 
290 			start = logical[nr];
291 			if (start < cache->key.objectid) {
292 				start = cache->key.objectid;
293 				len = (logical[nr] + stripe_len) - start;
294 			} else {
295 				len = min_t(u64, stripe_len,
296 					    cache->key.objectid +
297 					    cache->key.offset - start);
298 			}
299 
300 			cache->bytes_super += len;
301 			ret = add_excluded_extent(root, start, len);
302 			if (ret) {
303 				kfree(logical);
304 				return ret;
305 			}
306 		}
307 
308 		kfree(logical);
309 	}
310 	return 0;
311 }
312 
313 static struct btrfs_caching_control *
get_caching_control(struct btrfs_block_group_cache * cache)314 get_caching_control(struct btrfs_block_group_cache *cache)
315 {
316 	struct btrfs_caching_control *ctl;
317 
318 	spin_lock(&cache->lock);
319 	if (!cache->caching_ctl) {
320 		spin_unlock(&cache->lock);
321 		return NULL;
322 	}
323 
324 	ctl = cache->caching_ctl;
325 	atomic_inc(&ctl->count);
326 	spin_unlock(&cache->lock);
327 	return ctl;
328 }
329 
put_caching_control(struct btrfs_caching_control * ctl)330 static void put_caching_control(struct btrfs_caching_control *ctl)
331 {
332 	if (atomic_dec_and_test(&ctl->count))
333 		kfree(ctl);
334 }
335 
336 /*
337  * this is only called by cache_block_group, since we could have freed extents
338  * we need to check the pinned_extents for any extents that can't be used yet
339  * since their free space will be released as soon as the transaction commits.
340  */
add_new_free_space(struct btrfs_block_group_cache * block_group,struct btrfs_fs_info * info,u64 start,u64 end)341 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
342 			      struct btrfs_fs_info *info, u64 start, u64 end)
343 {
344 	u64 extent_start, extent_end, size, total_added = 0;
345 	int ret;
346 
347 	while (start < end) {
348 		ret = find_first_extent_bit(info->pinned_extents, start,
349 					    &extent_start, &extent_end,
350 					    EXTENT_DIRTY | EXTENT_UPTODATE,
351 					    NULL);
352 		if (ret)
353 			break;
354 
355 		if (extent_start <= start) {
356 			start = extent_end + 1;
357 		} else if (extent_start > start && extent_start < end) {
358 			size = extent_start - start;
359 			total_added += size;
360 			ret = btrfs_add_free_space(block_group, start,
361 						   size);
362 			BUG_ON(ret); /* -ENOMEM or logic error */
363 			start = extent_end + 1;
364 		} else {
365 			break;
366 		}
367 	}
368 
369 	if (start < end) {
370 		size = end - start;
371 		total_added += size;
372 		ret = btrfs_add_free_space(block_group, start, size);
373 		BUG_ON(ret); /* -ENOMEM or logic error */
374 	}
375 
376 	return total_added;
377 }
378 
caching_thread(struct btrfs_work * work)379 static noinline void caching_thread(struct btrfs_work *work)
380 {
381 	struct btrfs_block_group_cache *block_group;
382 	struct btrfs_fs_info *fs_info;
383 	struct btrfs_caching_control *caching_ctl;
384 	struct btrfs_root *extent_root;
385 	struct btrfs_path *path;
386 	struct extent_buffer *leaf;
387 	struct btrfs_key key;
388 	u64 total_found = 0;
389 	u64 last = 0;
390 	u32 nritems;
391 	int ret = -ENOMEM;
392 
393 	caching_ctl = container_of(work, struct btrfs_caching_control, work);
394 	block_group = caching_ctl->block_group;
395 	fs_info = block_group->fs_info;
396 	extent_root = fs_info->extent_root;
397 
398 	path = btrfs_alloc_path();
399 	if (!path)
400 		goto out;
401 
402 	last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
403 
404 	/*
405 	 * We don't want to deadlock with somebody trying to allocate a new
406 	 * extent for the extent root while also trying to search the extent
407 	 * root to add free space.  So we skip locking and search the commit
408 	 * root, since its read-only
409 	 */
410 	path->skip_locking = 1;
411 	path->search_commit_root = 1;
412 	path->reada = 1;
413 
414 	key.objectid = last;
415 	key.offset = 0;
416 	key.type = BTRFS_EXTENT_ITEM_KEY;
417 again:
418 	mutex_lock(&caching_ctl->mutex);
419 	/* need to make sure the commit_root doesn't disappear */
420 	down_read(&fs_info->commit_root_sem);
421 
422 next:
423 	ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
424 	if (ret < 0)
425 		goto err;
426 
427 	leaf = path->nodes[0];
428 	nritems = btrfs_header_nritems(leaf);
429 
430 	while (1) {
431 		if (btrfs_fs_closing(fs_info) > 1) {
432 			last = (u64)-1;
433 			break;
434 		}
435 
436 		if (path->slots[0] < nritems) {
437 			btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
438 		} else {
439 			ret = find_next_key(path, 0, &key);
440 			if (ret)
441 				break;
442 
443 			if (need_resched() ||
444 			    rwsem_is_contended(&fs_info->commit_root_sem)) {
445 				caching_ctl->progress = last;
446 				btrfs_release_path(path);
447 				up_read(&fs_info->commit_root_sem);
448 				mutex_unlock(&caching_ctl->mutex);
449 				cond_resched();
450 				goto again;
451 			}
452 
453 			ret = btrfs_next_leaf(extent_root, path);
454 			if (ret < 0)
455 				goto err;
456 			if (ret)
457 				break;
458 			leaf = path->nodes[0];
459 			nritems = btrfs_header_nritems(leaf);
460 			continue;
461 		}
462 
463 		if (key.objectid < last) {
464 			key.objectid = last;
465 			key.offset = 0;
466 			key.type = BTRFS_EXTENT_ITEM_KEY;
467 
468 			caching_ctl->progress = last;
469 			btrfs_release_path(path);
470 			goto next;
471 		}
472 
473 		if (key.objectid < block_group->key.objectid) {
474 			path->slots[0]++;
475 			continue;
476 		}
477 
478 		if (key.objectid >= block_group->key.objectid +
479 		    block_group->key.offset)
480 			break;
481 
482 		if (key.type == BTRFS_EXTENT_ITEM_KEY ||
483 		    key.type == BTRFS_METADATA_ITEM_KEY) {
484 			total_found += add_new_free_space(block_group,
485 							  fs_info, last,
486 							  key.objectid);
487 			if (key.type == BTRFS_METADATA_ITEM_KEY)
488 				last = key.objectid +
489 					fs_info->tree_root->nodesize;
490 			else
491 				last = key.objectid + key.offset;
492 
493 			if (total_found > (1024 * 1024 * 2)) {
494 				total_found = 0;
495 				wake_up(&caching_ctl->wait);
496 			}
497 		}
498 		path->slots[0]++;
499 	}
500 	ret = 0;
501 
502 	total_found += add_new_free_space(block_group, fs_info, last,
503 					  block_group->key.objectid +
504 					  block_group->key.offset);
505 	caching_ctl->progress = (u64)-1;
506 
507 	spin_lock(&block_group->lock);
508 	block_group->caching_ctl = NULL;
509 	block_group->cached = BTRFS_CACHE_FINISHED;
510 	spin_unlock(&block_group->lock);
511 
512 err:
513 	btrfs_free_path(path);
514 	up_read(&fs_info->commit_root_sem);
515 
516 	free_excluded_extents(extent_root, block_group);
517 
518 	mutex_unlock(&caching_ctl->mutex);
519 out:
520 	if (ret) {
521 		spin_lock(&block_group->lock);
522 		block_group->caching_ctl = NULL;
523 		block_group->cached = BTRFS_CACHE_ERROR;
524 		spin_unlock(&block_group->lock);
525 	}
526 	wake_up(&caching_ctl->wait);
527 
528 	put_caching_control(caching_ctl);
529 	btrfs_put_block_group(block_group);
530 }
531 
cache_block_group(struct btrfs_block_group_cache * cache,int load_cache_only)532 static int cache_block_group(struct btrfs_block_group_cache *cache,
533 			     int load_cache_only)
534 {
535 	DEFINE_WAIT(wait);
536 	struct btrfs_fs_info *fs_info = cache->fs_info;
537 	struct btrfs_caching_control *caching_ctl;
538 	int ret = 0;
539 
540 	caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
541 	if (!caching_ctl)
542 		return -ENOMEM;
543 
544 	INIT_LIST_HEAD(&caching_ctl->list);
545 	mutex_init(&caching_ctl->mutex);
546 	init_waitqueue_head(&caching_ctl->wait);
547 	caching_ctl->block_group = cache;
548 	caching_ctl->progress = cache->key.objectid;
549 	atomic_set(&caching_ctl->count, 1);
550 	btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
551 			caching_thread, NULL, NULL);
552 
553 	spin_lock(&cache->lock);
554 	/*
555 	 * This should be a rare occasion, but this could happen I think in the
556 	 * case where one thread starts to load the space cache info, and then
557 	 * some other thread starts a transaction commit which tries to do an
558 	 * allocation while the other thread is still loading the space cache
559 	 * info.  The previous loop should have kept us from choosing this block
560 	 * group, but if we've moved to the state where we will wait on caching
561 	 * block groups we need to first check if we're doing a fast load here,
562 	 * so we can wait for it to finish, otherwise we could end up allocating
563 	 * from a block group who's cache gets evicted for one reason or
564 	 * another.
565 	 */
566 	while (cache->cached == BTRFS_CACHE_FAST) {
567 		struct btrfs_caching_control *ctl;
568 
569 		ctl = cache->caching_ctl;
570 		atomic_inc(&ctl->count);
571 		prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
572 		spin_unlock(&cache->lock);
573 
574 		schedule();
575 
576 		finish_wait(&ctl->wait, &wait);
577 		put_caching_control(ctl);
578 		spin_lock(&cache->lock);
579 	}
580 
581 	if (cache->cached != BTRFS_CACHE_NO) {
582 		spin_unlock(&cache->lock);
583 		kfree(caching_ctl);
584 		return 0;
585 	}
586 	WARN_ON(cache->caching_ctl);
587 	cache->caching_ctl = caching_ctl;
588 	cache->cached = BTRFS_CACHE_FAST;
589 	spin_unlock(&cache->lock);
590 
591 	if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
592 		mutex_lock(&caching_ctl->mutex);
593 		ret = load_free_space_cache(fs_info, cache);
594 
595 		spin_lock(&cache->lock);
596 		if (ret == 1) {
597 			cache->caching_ctl = NULL;
598 			cache->cached = BTRFS_CACHE_FINISHED;
599 			cache->last_byte_to_unpin = (u64)-1;
600 			caching_ctl->progress = (u64)-1;
601 		} else {
602 			if (load_cache_only) {
603 				cache->caching_ctl = NULL;
604 				cache->cached = BTRFS_CACHE_NO;
605 			} else {
606 				cache->cached = BTRFS_CACHE_STARTED;
607 				cache->has_caching_ctl = 1;
608 			}
609 		}
610 		spin_unlock(&cache->lock);
611 		mutex_unlock(&caching_ctl->mutex);
612 
613 		wake_up(&caching_ctl->wait);
614 		if (ret == 1) {
615 			put_caching_control(caching_ctl);
616 			free_excluded_extents(fs_info->extent_root, cache);
617 			return 0;
618 		}
619 	} else {
620 		/*
621 		 * We are not going to do the fast caching, set cached to the
622 		 * appropriate value and wakeup any waiters.
623 		 */
624 		spin_lock(&cache->lock);
625 		if (load_cache_only) {
626 			cache->caching_ctl = NULL;
627 			cache->cached = BTRFS_CACHE_NO;
628 		} else {
629 			cache->cached = BTRFS_CACHE_STARTED;
630 			cache->has_caching_ctl = 1;
631 		}
632 		spin_unlock(&cache->lock);
633 		wake_up(&caching_ctl->wait);
634 	}
635 
636 	if (load_cache_only) {
637 		put_caching_control(caching_ctl);
638 		return 0;
639 	}
640 
641 	down_write(&fs_info->commit_root_sem);
642 	atomic_inc(&caching_ctl->count);
643 	list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
644 	up_write(&fs_info->commit_root_sem);
645 
646 	btrfs_get_block_group(cache);
647 
648 	btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
649 
650 	return ret;
651 }
652 
653 /*
654  * return the block group that starts at or after bytenr
655  */
656 static struct btrfs_block_group_cache *
btrfs_lookup_first_block_group(struct btrfs_fs_info * info,u64 bytenr)657 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
658 {
659 	struct btrfs_block_group_cache *cache;
660 
661 	cache = block_group_cache_tree_search(info, bytenr, 0);
662 
663 	return cache;
664 }
665 
666 /*
667  * return the block group that contains the given bytenr
668  */
btrfs_lookup_block_group(struct btrfs_fs_info * info,u64 bytenr)669 struct btrfs_block_group_cache *btrfs_lookup_block_group(
670 						 struct btrfs_fs_info *info,
671 						 u64 bytenr)
672 {
673 	struct btrfs_block_group_cache *cache;
674 
675 	cache = block_group_cache_tree_search(info, bytenr, 1);
676 
677 	return cache;
678 }
679 
__find_space_info(struct btrfs_fs_info * info,u64 flags)680 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
681 						  u64 flags)
682 {
683 	struct list_head *head = &info->space_info;
684 	struct btrfs_space_info *found;
685 
686 	flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
687 
688 	rcu_read_lock();
689 	list_for_each_entry_rcu(found, head, list) {
690 		if (found->flags & flags) {
691 			rcu_read_unlock();
692 			return found;
693 		}
694 	}
695 	rcu_read_unlock();
696 	return NULL;
697 }
698 
699 /*
700  * after adding space to the filesystem, we need to clear the full flags
701  * on all the space infos.
702  */
btrfs_clear_space_info_full(struct btrfs_fs_info * info)703 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
704 {
705 	struct list_head *head = &info->space_info;
706 	struct btrfs_space_info *found;
707 
708 	rcu_read_lock();
709 	list_for_each_entry_rcu(found, head, list)
710 		found->full = 0;
711 	rcu_read_unlock();
712 }
713 
714 /* simple helper to search for an existing data extent at a given offset */
btrfs_lookup_data_extent(struct btrfs_root * root,u64 start,u64 len)715 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
716 {
717 	int ret;
718 	struct btrfs_key key;
719 	struct btrfs_path *path;
720 
721 	path = btrfs_alloc_path();
722 	if (!path)
723 		return -ENOMEM;
724 
725 	key.objectid = start;
726 	key.offset = len;
727 	key.type = BTRFS_EXTENT_ITEM_KEY;
728 	ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
729 				0, 0);
730 	btrfs_free_path(path);
731 	return ret;
732 }
733 
734 /*
735  * helper function to lookup reference count and flags of a tree block.
736  *
737  * the head node for delayed ref is used to store the sum of all the
738  * reference count modifications queued up in the rbtree. the head
739  * node may also store the extent flags to set. This way you can check
740  * to see what the reference count and extent flags would be if all of
741  * the delayed refs are not processed.
742  */
btrfs_lookup_extent_info(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 bytenr,u64 offset,int metadata,u64 * refs,u64 * flags)743 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
744 			     struct btrfs_root *root, u64 bytenr,
745 			     u64 offset, int metadata, u64 *refs, u64 *flags)
746 {
747 	struct btrfs_delayed_ref_head *head;
748 	struct btrfs_delayed_ref_root *delayed_refs;
749 	struct btrfs_path *path;
750 	struct btrfs_extent_item *ei;
751 	struct extent_buffer *leaf;
752 	struct btrfs_key key;
753 	u32 item_size;
754 	u64 num_refs;
755 	u64 extent_flags;
756 	int ret;
757 
758 	/*
759 	 * If we don't have skinny metadata, don't bother doing anything
760 	 * different
761 	 */
762 	if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
763 		offset = root->nodesize;
764 		metadata = 0;
765 	}
766 
767 	path = btrfs_alloc_path();
768 	if (!path)
769 		return -ENOMEM;
770 
771 	if (!trans) {
772 		path->skip_locking = 1;
773 		path->search_commit_root = 1;
774 	}
775 
776 search_again:
777 	key.objectid = bytenr;
778 	key.offset = offset;
779 	if (metadata)
780 		key.type = BTRFS_METADATA_ITEM_KEY;
781 	else
782 		key.type = BTRFS_EXTENT_ITEM_KEY;
783 
784 	ret = btrfs_search_slot(trans, root->fs_info->extent_root,
785 				&key, path, 0, 0);
786 	if (ret < 0)
787 		goto out_free;
788 
789 	if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
790 		if (path->slots[0]) {
791 			path->slots[0]--;
792 			btrfs_item_key_to_cpu(path->nodes[0], &key,
793 					      path->slots[0]);
794 			if (key.objectid == bytenr &&
795 			    key.type == BTRFS_EXTENT_ITEM_KEY &&
796 			    key.offset == root->nodesize)
797 				ret = 0;
798 		}
799 	}
800 
801 	if (ret == 0) {
802 		leaf = path->nodes[0];
803 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
804 		if (item_size >= sizeof(*ei)) {
805 			ei = btrfs_item_ptr(leaf, path->slots[0],
806 					    struct btrfs_extent_item);
807 			num_refs = btrfs_extent_refs(leaf, ei);
808 			extent_flags = btrfs_extent_flags(leaf, ei);
809 		} else {
810 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
811 			struct btrfs_extent_item_v0 *ei0;
812 			BUG_ON(item_size != sizeof(*ei0));
813 			ei0 = btrfs_item_ptr(leaf, path->slots[0],
814 					     struct btrfs_extent_item_v0);
815 			num_refs = btrfs_extent_refs_v0(leaf, ei0);
816 			/* FIXME: this isn't correct for data */
817 			extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
818 #else
819 			BUG();
820 #endif
821 		}
822 		BUG_ON(num_refs == 0);
823 	} else {
824 		num_refs = 0;
825 		extent_flags = 0;
826 		ret = 0;
827 	}
828 
829 	if (!trans)
830 		goto out;
831 
832 	delayed_refs = &trans->transaction->delayed_refs;
833 	spin_lock(&delayed_refs->lock);
834 	head = btrfs_find_delayed_ref_head(trans, bytenr);
835 	if (head) {
836 		if (!mutex_trylock(&head->mutex)) {
837 			atomic_inc(&head->node.refs);
838 			spin_unlock(&delayed_refs->lock);
839 
840 			btrfs_release_path(path);
841 
842 			/*
843 			 * Mutex was contended, block until it's released and try
844 			 * again
845 			 */
846 			mutex_lock(&head->mutex);
847 			mutex_unlock(&head->mutex);
848 			btrfs_put_delayed_ref(&head->node);
849 			goto search_again;
850 		}
851 		spin_lock(&head->lock);
852 		if (head->extent_op && head->extent_op->update_flags)
853 			extent_flags |= head->extent_op->flags_to_set;
854 		else
855 			BUG_ON(num_refs == 0);
856 
857 		num_refs += head->node.ref_mod;
858 		spin_unlock(&head->lock);
859 		mutex_unlock(&head->mutex);
860 	}
861 	spin_unlock(&delayed_refs->lock);
862 out:
863 	WARN_ON(num_refs == 0);
864 	if (refs)
865 		*refs = num_refs;
866 	if (flags)
867 		*flags = extent_flags;
868 out_free:
869 	btrfs_free_path(path);
870 	return ret;
871 }
872 
873 /*
874  * Back reference rules.  Back refs have three main goals:
875  *
876  * 1) differentiate between all holders of references to an extent so that
877  *    when a reference is dropped we can make sure it was a valid reference
878  *    before freeing the extent.
879  *
880  * 2) Provide enough information to quickly find the holders of an extent
881  *    if we notice a given block is corrupted or bad.
882  *
883  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
884  *    maintenance.  This is actually the same as #2, but with a slightly
885  *    different use case.
886  *
887  * There are two kinds of back refs. The implicit back refs is optimized
888  * for pointers in non-shared tree blocks. For a given pointer in a block,
889  * back refs of this kind provide information about the block's owner tree
890  * and the pointer's key. These information allow us to find the block by
891  * b-tree searching. The full back refs is for pointers in tree blocks not
892  * referenced by their owner trees. The location of tree block is recorded
893  * in the back refs. Actually the full back refs is generic, and can be
894  * used in all cases the implicit back refs is used. The major shortcoming
895  * of the full back refs is its overhead. Every time a tree block gets
896  * COWed, we have to update back refs entry for all pointers in it.
897  *
898  * For a newly allocated tree block, we use implicit back refs for
899  * pointers in it. This means most tree related operations only involve
900  * implicit back refs. For a tree block created in old transaction, the
901  * only way to drop a reference to it is COW it. So we can detect the
902  * event that tree block loses its owner tree's reference and do the
903  * back refs conversion.
904  *
905  * When a tree block is COW'd through a tree, there are four cases:
906  *
907  * The reference count of the block is one and the tree is the block's
908  * owner tree. Nothing to do in this case.
909  *
910  * The reference count of the block is one and the tree is not the
911  * block's owner tree. In this case, full back refs is used for pointers
912  * in the block. Remove these full back refs, add implicit back refs for
913  * every pointers in the new block.
914  *
915  * The reference count of the block is greater than one and the tree is
916  * the block's owner tree. In this case, implicit back refs is used for
917  * pointers in the block. Add full back refs for every pointers in the
918  * block, increase lower level extents' reference counts. The original
919  * implicit back refs are entailed to the new block.
920  *
921  * The reference count of the block is greater than one and the tree is
922  * not the block's owner tree. Add implicit back refs for every pointer in
923  * the new block, increase lower level extents' reference count.
924  *
925  * Back Reference Key composing:
926  *
927  * The key objectid corresponds to the first byte in the extent,
928  * The key type is used to differentiate between types of back refs.
929  * There are different meanings of the key offset for different types
930  * of back refs.
931  *
932  * File extents can be referenced by:
933  *
934  * - multiple snapshots, subvolumes, or different generations in one subvol
935  * - different files inside a single subvolume
936  * - different offsets inside a file (bookend extents in file.c)
937  *
938  * The extent ref structure for the implicit back refs has fields for:
939  *
940  * - Objectid of the subvolume root
941  * - objectid of the file holding the reference
942  * - original offset in the file
943  * - how many bookend extents
944  *
945  * The key offset for the implicit back refs is hash of the first
946  * three fields.
947  *
948  * The extent ref structure for the full back refs has field for:
949  *
950  * - number of pointers in the tree leaf
951  *
952  * The key offset for the implicit back refs is the first byte of
953  * the tree leaf
954  *
955  * When a file extent is allocated, The implicit back refs is used.
956  * the fields are filled in:
957  *
958  *     (root_key.objectid, inode objectid, offset in file, 1)
959  *
960  * When a file extent is removed file truncation, we find the
961  * corresponding implicit back refs and check the following fields:
962  *
963  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
964  *
965  * Btree extents can be referenced by:
966  *
967  * - Different subvolumes
968  *
969  * Both the implicit back refs and the full back refs for tree blocks
970  * only consist of key. The key offset for the implicit back refs is
971  * objectid of block's owner tree. The key offset for the full back refs
972  * is the first byte of parent block.
973  *
974  * When implicit back refs is used, information about the lowest key and
975  * level of the tree block are required. These information are stored in
976  * tree block info structure.
977  */
978 
979 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
convert_extent_item_v0(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,u64 owner,u32 extra_size)980 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
981 				  struct btrfs_root *root,
982 				  struct btrfs_path *path,
983 				  u64 owner, u32 extra_size)
984 {
985 	struct btrfs_extent_item *item;
986 	struct btrfs_extent_item_v0 *ei0;
987 	struct btrfs_extent_ref_v0 *ref0;
988 	struct btrfs_tree_block_info *bi;
989 	struct extent_buffer *leaf;
990 	struct btrfs_key key;
991 	struct btrfs_key found_key;
992 	u32 new_size = sizeof(*item);
993 	u64 refs;
994 	int ret;
995 
996 	leaf = path->nodes[0];
997 	BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
998 
999 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1000 	ei0 = btrfs_item_ptr(leaf, path->slots[0],
1001 			     struct btrfs_extent_item_v0);
1002 	refs = btrfs_extent_refs_v0(leaf, ei0);
1003 
1004 	if (owner == (u64)-1) {
1005 		while (1) {
1006 			if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1007 				ret = btrfs_next_leaf(root, path);
1008 				if (ret < 0)
1009 					return ret;
1010 				BUG_ON(ret > 0); /* Corruption */
1011 				leaf = path->nodes[0];
1012 			}
1013 			btrfs_item_key_to_cpu(leaf, &found_key,
1014 					      path->slots[0]);
1015 			BUG_ON(key.objectid != found_key.objectid);
1016 			if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1017 				path->slots[0]++;
1018 				continue;
1019 			}
1020 			ref0 = btrfs_item_ptr(leaf, path->slots[0],
1021 					      struct btrfs_extent_ref_v0);
1022 			owner = btrfs_ref_objectid_v0(leaf, ref0);
1023 			break;
1024 		}
1025 	}
1026 	btrfs_release_path(path);
1027 
1028 	if (owner < BTRFS_FIRST_FREE_OBJECTID)
1029 		new_size += sizeof(*bi);
1030 
1031 	new_size -= sizeof(*ei0);
1032 	ret = btrfs_search_slot(trans, root, &key, path,
1033 				new_size + extra_size, 1);
1034 	if (ret < 0)
1035 		return ret;
1036 	BUG_ON(ret); /* Corruption */
1037 
1038 	btrfs_extend_item(root, path, new_size);
1039 
1040 	leaf = path->nodes[0];
1041 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1042 	btrfs_set_extent_refs(leaf, item, refs);
1043 	/* FIXME: get real generation */
1044 	btrfs_set_extent_generation(leaf, item, 0);
1045 	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1046 		btrfs_set_extent_flags(leaf, item,
1047 				       BTRFS_EXTENT_FLAG_TREE_BLOCK |
1048 				       BTRFS_BLOCK_FLAG_FULL_BACKREF);
1049 		bi = (struct btrfs_tree_block_info *)(item + 1);
1050 		/* FIXME: get first key of the block */
1051 		memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1052 		btrfs_set_tree_block_level(leaf, bi, (int)owner);
1053 	} else {
1054 		btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1055 	}
1056 	btrfs_mark_buffer_dirty(leaf);
1057 	return 0;
1058 }
1059 #endif
1060 
hash_extent_data_ref(u64 root_objectid,u64 owner,u64 offset)1061 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1062 {
1063 	u32 high_crc = ~(u32)0;
1064 	u32 low_crc = ~(u32)0;
1065 	__le64 lenum;
1066 
1067 	lenum = cpu_to_le64(root_objectid);
1068 	high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1069 	lenum = cpu_to_le64(owner);
1070 	low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1071 	lenum = cpu_to_le64(offset);
1072 	low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1073 
1074 	return ((u64)high_crc << 31) ^ (u64)low_crc;
1075 }
1076 
hash_extent_data_ref_item(struct extent_buffer * leaf,struct btrfs_extent_data_ref * ref)1077 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1078 				     struct btrfs_extent_data_ref *ref)
1079 {
1080 	return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1081 				    btrfs_extent_data_ref_objectid(leaf, ref),
1082 				    btrfs_extent_data_ref_offset(leaf, ref));
1083 }
1084 
match_extent_data_ref(struct extent_buffer * leaf,struct btrfs_extent_data_ref * ref,u64 root_objectid,u64 owner,u64 offset)1085 static int match_extent_data_ref(struct extent_buffer *leaf,
1086 				 struct btrfs_extent_data_ref *ref,
1087 				 u64 root_objectid, u64 owner, u64 offset)
1088 {
1089 	if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1090 	    btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1091 	    btrfs_extent_data_ref_offset(leaf, ref) != offset)
1092 		return 0;
1093 	return 1;
1094 }
1095 
lookup_extent_data_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,u64 bytenr,u64 parent,u64 root_objectid,u64 owner,u64 offset)1096 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1097 					   struct btrfs_root *root,
1098 					   struct btrfs_path *path,
1099 					   u64 bytenr, u64 parent,
1100 					   u64 root_objectid,
1101 					   u64 owner, u64 offset)
1102 {
1103 	struct btrfs_key key;
1104 	struct btrfs_extent_data_ref *ref;
1105 	struct extent_buffer *leaf;
1106 	u32 nritems;
1107 	int ret;
1108 	int recow;
1109 	int err = -ENOENT;
1110 
1111 	key.objectid = bytenr;
1112 	if (parent) {
1113 		key.type = BTRFS_SHARED_DATA_REF_KEY;
1114 		key.offset = parent;
1115 	} else {
1116 		key.type = BTRFS_EXTENT_DATA_REF_KEY;
1117 		key.offset = hash_extent_data_ref(root_objectid,
1118 						  owner, offset);
1119 	}
1120 again:
1121 	recow = 0;
1122 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1123 	if (ret < 0) {
1124 		err = ret;
1125 		goto fail;
1126 	}
1127 
1128 	if (parent) {
1129 		if (!ret)
1130 			return 0;
1131 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1132 		key.type = BTRFS_EXTENT_REF_V0_KEY;
1133 		btrfs_release_path(path);
1134 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1135 		if (ret < 0) {
1136 			err = ret;
1137 			goto fail;
1138 		}
1139 		if (!ret)
1140 			return 0;
1141 #endif
1142 		goto fail;
1143 	}
1144 
1145 	leaf = path->nodes[0];
1146 	nritems = btrfs_header_nritems(leaf);
1147 	while (1) {
1148 		if (path->slots[0] >= nritems) {
1149 			ret = btrfs_next_leaf(root, path);
1150 			if (ret < 0)
1151 				err = ret;
1152 			if (ret)
1153 				goto fail;
1154 
1155 			leaf = path->nodes[0];
1156 			nritems = btrfs_header_nritems(leaf);
1157 			recow = 1;
1158 		}
1159 
1160 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1161 		if (key.objectid != bytenr ||
1162 		    key.type != BTRFS_EXTENT_DATA_REF_KEY)
1163 			goto fail;
1164 
1165 		ref = btrfs_item_ptr(leaf, path->slots[0],
1166 				     struct btrfs_extent_data_ref);
1167 
1168 		if (match_extent_data_ref(leaf, ref, root_objectid,
1169 					  owner, offset)) {
1170 			if (recow) {
1171 				btrfs_release_path(path);
1172 				goto again;
1173 			}
1174 			err = 0;
1175 			break;
1176 		}
1177 		path->slots[0]++;
1178 	}
1179 fail:
1180 	return err;
1181 }
1182 
insert_extent_data_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,u64 bytenr,u64 parent,u64 root_objectid,u64 owner,u64 offset,int refs_to_add)1183 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1184 					   struct btrfs_root *root,
1185 					   struct btrfs_path *path,
1186 					   u64 bytenr, u64 parent,
1187 					   u64 root_objectid, u64 owner,
1188 					   u64 offset, int refs_to_add)
1189 {
1190 	struct btrfs_key key;
1191 	struct extent_buffer *leaf;
1192 	u32 size;
1193 	u32 num_refs;
1194 	int ret;
1195 
1196 	key.objectid = bytenr;
1197 	if (parent) {
1198 		key.type = BTRFS_SHARED_DATA_REF_KEY;
1199 		key.offset = parent;
1200 		size = sizeof(struct btrfs_shared_data_ref);
1201 	} else {
1202 		key.type = BTRFS_EXTENT_DATA_REF_KEY;
1203 		key.offset = hash_extent_data_ref(root_objectid,
1204 						  owner, offset);
1205 		size = sizeof(struct btrfs_extent_data_ref);
1206 	}
1207 
1208 	ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1209 	if (ret && ret != -EEXIST)
1210 		goto fail;
1211 
1212 	leaf = path->nodes[0];
1213 	if (parent) {
1214 		struct btrfs_shared_data_ref *ref;
1215 		ref = btrfs_item_ptr(leaf, path->slots[0],
1216 				     struct btrfs_shared_data_ref);
1217 		if (ret == 0) {
1218 			btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1219 		} else {
1220 			num_refs = btrfs_shared_data_ref_count(leaf, ref);
1221 			num_refs += refs_to_add;
1222 			btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1223 		}
1224 	} else {
1225 		struct btrfs_extent_data_ref *ref;
1226 		while (ret == -EEXIST) {
1227 			ref = btrfs_item_ptr(leaf, path->slots[0],
1228 					     struct btrfs_extent_data_ref);
1229 			if (match_extent_data_ref(leaf, ref, root_objectid,
1230 						  owner, offset))
1231 				break;
1232 			btrfs_release_path(path);
1233 			key.offset++;
1234 			ret = btrfs_insert_empty_item(trans, root, path, &key,
1235 						      size);
1236 			if (ret && ret != -EEXIST)
1237 				goto fail;
1238 
1239 			leaf = path->nodes[0];
1240 		}
1241 		ref = btrfs_item_ptr(leaf, path->slots[0],
1242 				     struct btrfs_extent_data_ref);
1243 		if (ret == 0) {
1244 			btrfs_set_extent_data_ref_root(leaf, ref,
1245 						       root_objectid);
1246 			btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1247 			btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1248 			btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1249 		} else {
1250 			num_refs = btrfs_extent_data_ref_count(leaf, ref);
1251 			num_refs += refs_to_add;
1252 			btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1253 		}
1254 	}
1255 	btrfs_mark_buffer_dirty(leaf);
1256 	ret = 0;
1257 fail:
1258 	btrfs_release_path(path);
1259 	return ret;
1260 }
1261 
remove_extent_data_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int refs_to_drop,int * last_ref)1262 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1263 					   struct btrfs_root *root,
1264 					   struct btrfs_path *path,
1265 					   int refs_to_drop, int *last_ref)
1266 {
1267 	struct btrfs_key key;
1268 	struct btrfs_extent_data_ref *ref1 = NULL;
1269 	struct btrfs_shared_data_ref *ref2 = NULL;
1270 	struct extent_buffer *leaf;
1271 	u32 num_refs = 0;
1272 	int ret = 0;
1273 
1274 	leaf = path->nodes[0];
1275 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1276 
1277 	if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1278 		ref1 = btrfs_item_ptr(leaf, path->slots[0],
1279 				      struct btrfs_extent_data_ref);
1280 		num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1281 	} else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1282 		ref2 = btrfs_item_ptr(leaf, path->slots[0],
1283 				      struct btrfs_shared_data_ref);
1284 		num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1285 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1286 	} else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1287 		struct btrfs_extent_ref_v0 *ref0;
1288 		ref0 = btrfs_item_ptr(leaf, path->slots[0],
1289 				      struct btrfs_extent_ref_v0);
1290 		num_refs = btrfs_ref_count_v0(leaf, ref0);
1291 #endif
1292 	} else {
1293 		BUG();
1294 	}
1295 
1296 	BUG_ON(num_refs < refs_to_drop);
1297 	num_refs -= refs_to_drop;
1298 
1299 	if (num_refs == 0) {
1300 		ret = btrfs_del_item(trans, root, path);
1301 		*last_ref = 1;
1302 	} else {
1303 		if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1304 			btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1305 		else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1306 			btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1307 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1308 		else {
1309 			struct btrfs_extent_ref_v0 *ref0;
1310 			ref0 = btrfs_item_ptr(leaf, path->slots[0],
1311 					struct btrfs_extent_ref_v0);
1312 			btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1313 		}
1314 #endif
1315 		btrfs_mark_buffer_dirty(leaf);
1316 	}
1317 	return ret;
1318 }
1319 
extent_data_ref_count(struct btrfs_root * root,struct btrfs_path * path,struct btrfs_extent_inline_ref * iref)1320 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1321 					  struct btrfs_path *path,
1322 					  struct btrfs_extent_inline_ref *iref)
1323 {
1324 	struct btrfs_key key;
1325 	struct extent_buffer *leaf;
1326 	struct btrfs_extent_data_ref *ref1;
1327 	struct btrfs_shared_data_ref *ref2;
1328 	u32 num_refs = 0;
1329 
1330 	leaf = path->nodes[0];
1331 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1332 	if (iref) {
1333 		if (btrfs_extent_inline_ref_type(leaf, iref) ==
1334 		    BTRFS_EXTENT_DATA_REF_KEY) {
1335 			ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1336 			num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1337 		} else {
1338 			ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1339 			num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1340 		}
1341 	} else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1342 		ref1 = btrfs_item_ptr(leaf, path->slots[0],
1343 				      struct btrfs_extent_data_ref);
1344 		num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1345 	} else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1346 		ref2 = btrfs_item_ptr(leaf, path->slots[0],
1347 				      struct btrfs_shared_data_ref);
1348 		num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1349 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1350 	} else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1351 		struct btrfs_extent_ref_v0 *ref0;
1352 		ref0 = btrfs_item_ptr(leaf, path->slots[0],
1353 				      struct btrfs_extent_ref_v0);
1354 		num_refs = btrfs_ref_count_v0(leaf, ref0);
1355 #endif
1356 	} else {
1357 		WARN_ON(1);
1358 	}
1359 	return num_refs;
1360 }
1361 
lookup_tree_block_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,u64 bytenr,u64 parent,u64 root_objectid)1362 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1363 					  struct btrfs_root *root,
1364 					  struct btrfs_path *path,
1365 					  u64 bytenr, u64 parent,
1366 					  u64 root_objectid)
1367 {
1368 	struct btrfs_key key;
1369 	int ret;
1370 
1371 	key.objectid = bytenr;
1372 	if (parent) {
1373 		key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1374 		key.offset = parent;
1375 	} else {
1376 		key.type = BTRFS_TREE_BLOCK_REF_KEY;
1377 		key.offset = root_objectid;
1378 	}
1379 
1380 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1381 	if (ret > 0)
1382 		ret = -ENOENT;
1383 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1384 	if (ret == -ENOENT && parent) {
1385 		btrfs_release_path(path);
1386 		key.type = BTRFS_EXTENT_REF_V0_KEY;
1387 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1388 		if (ret > 0)
1389 			ret = -ENOENT;
1390 	}
1391 #endif
1392 	return ret;
1393 }
1394 
insert_tree_block_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,u64 bytenr,u64 parent,u64 root_objectid)1395 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1396 					  struct btrfs_root *root,
1397 					  struct btrfs_path *path,
1398 					  u64 bytenr, u64 parent,
1399 					  u64 root_objectid)
1400 {
1401 	struct btrfs_key key;
1402 	int ret;
1403 
1404 	key.objectid = bytenr;
1405 	if (parent) {
1406 		key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1407 		key.offset = parent;
1408 	} else {
1409 		key.type = BTRFS_TREE_BLOCK_REF_KEY;
1410 		key.offset = root_objectid;
1411 	}
1412 
1413 	ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1414 	btrfs_release_path(path);
1415 	return ret;
1416 }
1417 
extent_ref_type(u64 parent,u64 owner)1418 static inline int extent_ref_type(u64 parent, u64 owner)
1419 {
1420 	int type;
1421 	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1422 		if (parent > 0)
1423 			type = BTRFS_SHARED_BLOCK_REF_KEY;
1424 		else
1425 			type = BTRFS_TREE_BLOCK_REF_KEY;
1426 	} else {
1427 		if (parent > 0)
1428 			type = BTRFS_SHARED_DATA_REF_KEY;
1429 		else
1430 			type = BTRFS_EXTENT_DATA_REF_KEY;
1431 	}
1432 	return type;
1433 }
1434 
find_next_key(struct btrfs_path * path,int level,struct btrfs_key * key)1435 static int find_next_key(struct btrfs_path *path, int level,
1436 			 struct btrfs_key *key)
1437 
1438 {
1439 	for (; level < BTRFS_MAX_LEVEL; level++) {
1440 		if (!path->nodes[level])
1441 			break;
1442 		if (path->slots[level] + 1 >=
1443 		    btrfs_header_nritems(path->nodes[level]))
1444 			continue;
1445 		if (level == 0)
1446 			btrfs_item_key_to_cpu(path->nodes[level], key,
1447 					      path->slots[level] + 1);
1448 		else
1449 			btrfs_node_key_to_cpu(path->nodes[level], key,
1450 					      path->slots[level] + 1);
1451 		return 0;
1452 	}
1453 	return 1;
1454 }
1455 
1456 /*
1457  * look for inline back ref. if back ref is found, *ref_ret is set
1458  * to the address of inline back ref, and 0 is returned.
1459  *
1460  * if back ref isn't found, *ref_ret is set to the address where it
1461  * should be inserted, and -ENOENT is returned.
1462  *
1463  * if insert is true and there are too many inline back refs, the path
1464  * points to the extent item, and -EAGAIN is returned.
1465  *
1466  * NOTE: inline back refs are ordered in the same way that back ref
1467  *	 items in the tree are ordered.
1468  */
1469 static noinline_for_stack
lookup_inline_extent_backref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct btrfs_extent_inline_ref ** ref_ret,u64 bytenr,u64 num_bytes,u64 parent,u64 root_objectid,u64 owner,u64 offset,int insert)1470 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1471 				 struct btrfs_root *root,
1472 				 struct btrfs_path *path,
1473 				 struct btrfs_extent_inline_ref **ref_ret,
1474 				 u64 bytenr, u64 num_bytes,
1475 				 u64 parent, u64 root_objectid,
1476 				 u64 owner, u64 offset, int insert)
1477 {
1478 	struct btrfs_key key;
1479 	struct extent_buffer *leaf;
1480 	struct btrfs_extent_item *ei;
1481 	struct btrfs_extent_inline_ref *iref;
1482 	u64 flags;
1483 	u64 item_size;
1484 	unsigned long ptr;
1485 	unsigned long end;
1486 	int extra_size;
1487 	int type;
1488 	int want;
1489 	int ret;
1490 	int err = 0;
1491 	bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1492 						 SKINNY_METADATA);
1493 
1494 	key.objectid = bytenr;
1495 	key.type = BTRFS_EXTENT_ITEM_KEY;
1496 	key.offset = num_bytes;
1497 
1498 	want = extent_ref_type(parent, owner);
1499 	if (insert) {
1500 		extra_size = btrfs_extent_inline_ref_size(want);
1501 		path->keep_locks = 1;
1502 	} else
1503 		extra_size = -1;
1504 
1505 	/*
1506 	 * Owner is our parent level, so we can just add one to get the level
1507 	 * for the block we are interested in.
1508 	 */
1509 	if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1510 		key.type = BTRFS_METADATA_ITEM_KEY;
1511 		key.offset = owner;
1512 	}
1513 
1514 again:
1515 	ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1516 	if (ret < 0) {
1517 		err = ret;
1518 		goto out;
1519 	}
1520 
1521 	/*
1522 	 * We may be a newly converted file system which still has the old fat
1523 	 * extent entries for metadata, so try and see if we have one of those.
1524 	 */
1525 	if (ret > 0 && skinny_metadata) {
1526 		skinny_metadata = false;
1527 		if (path->slots[0]) {
1528 			path->slots[0]--;
1529 			btrfs_item_key_to_cpu(path->nodes[0], &key,
1530 					      path->slots[0]);
1531 			if (key.objectid == bytenr &&
1532 			    key.type == BTRFS_EXTENT_ITEM_KEY &&
1533 			    key.offset == num_bytes)
1534 				ret = 0;
1535 		}
1536 		if (ret) {
1537 			key.objectid = bytenr;
1538 			key.type = BTRFS_EXTENT_ITEM_KEY;
1539 			key.offset = num_bytes;
1540 			btrfs_release_path(path);
1541 			goto again;
1542 		}
1543 	}
1544 
1545 	if (ret && !insert) {
1546 		err = -ENOENT;
1547 		goto out;
1548 	} else if (WARN_ON(ret)) {
1549 		err = -EIO;
1550 		goto out;
1551 	}
1552 
1553 	leaf = path->nodes[0];
1554 	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1555 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1556 	if (item_size < sizeof(*ei)) {
1557 		if (!insert) {
1558 			err = -ENOENT;
1559 			goto out;
1560 		}
1561 		ret = convert_extent_item_v0(trans, root, path, owner,
1562 					     extra_size);
1563 		if (ret < 0) {
1564 			err = ret;
1565 			goto out;
1566 		}
1567 		leaf = path->nodes[0];
1568 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1569 	}
1570 #endif
1571 	BUG_ON(item_size < sizeof(*ei));
1572 
1573 	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1574 	flags = btrfs_extent_flags(leaf, ei);
1575 
1576 	ptr = (unsigned long)(ei + 1);
1577 	end = (unsigned long)ei + item_size;
1578 
1579 	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1580 		ptr += sizeof(struct btrfs_tree_block_info);
1581 		BUG_ON(ptr > end);
1582 	}
1583 
1584 	err = -ENOENT;
1585 	while (1) {
1586 		if (ptr >= end) {
1587 			WARN_ON(ptr > end);
1588 			break;
1589 		}
1590 		iref = (struct btrfs_extent_inline_ref *)ptr;
1591 		type = btrfs_extent_inline_ref_type(leaf, iref);
1592 		if (want < type)
1593 			break;
1594 		if (want > type) {
1595 			ptr += btrfs_extent_inline_ref_size(type);
1596 			continue;
1597 		}
1598 
1599 		if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1600 			struct btrfs_extent_data_ref *dref;
1601 			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1602 			if (match_extent_data_ref(leaf, dref, root_objectid,
1603 						  owner, offset)) {
1604 				err = 0;
1605 				break;
1606 			}
1607 			if (hash_extent_data_ref_item(leaf, dref) <
1608 			    hash_extent_data_ref(root_objectid, owner, offset))
1609 				break;
1610 		} else {
1611 			u64 ref_offset;
1612 			ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1613 			if (parent > 0) {
1614 				if (parent == ref_offset) {
1615 					err = 0;
1616 					break;
1617 				}
1618 				if (ref_offset < parent)
1619 					break;
1620 			} else {
1621 				if (root_objectid == ref_offset) {
1622 					err = 0;
1623 					break;
1624 				}
1625 				if (ref_offset < root_objectid)
1626 					break;
1627 			}
1628 		}
1629 		ptr += btrfs_extent_inline_ref_size(type);
1630 	}
1631 	if (err == -ENOENT && insert) {
1632 		if (item_size + extra_size >=
1633 		    BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1634 			err = -EAGAIN;
1635 			goto out;
1636 		}
1637 		/*
1638 		 * To add new inline back ref, we have to make sure
1639 		 * there is no corresponding back ref item.
1640 		 * For simplicity, we just do not add new inline back
1641 		 * ref if there is any kind of item for this block
1642 		 */
1643 		if (find_next_key(path, 0, &key) == 0 &&
1644 		    key.objectid == bytenr &&
1645 		    key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1646 			err = -EAGAIN;
1647 			goto out;
1648 		}
1649 	}
1650 	*ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1651 out:
1652 	if (insert) {
1653 		path->keep_locks = 0;
1654 		btrfs_unlock_up_safe(path, 1);
1655 	}
1656 	return err;
1657 }
1658 
1659 /*
1660  * helper to add new inline back ref
1661  */
1662 static noinline_for_stack
setup_inline_extent_backref(struct btrfs_root * root,struct btrfs_path * path,struct btrfs_extent_inline_ref * iref,u64 parent,u64 root_objectid,u64 owner,u64 offset,int refs_to_add,struct btrfs_delayed_extent_op * extent_op)1663 void setup_inline_extent_backref(struct btrfs_root *root,
1664 				 struct btrfs_path *path,
1665 				 struct btrfs_extent_inline_ref *iref,
1666 				 u64 parent, u64 root_objectid,
1667 				 u64 owner, u64 offset, int refs_to_add,
1668 				 struct btrfs_delayed_extent_op *extent_op)
1669 {
1670 	struct extent_buffer *leaf;
1671 	struct btrfs_extent_item *ei;
1672 	unsigned long ptr;
1673 	unsigned long end;
1674 	unsigned long item_offset;
1675 	u64 refs;
1676 	int size;
1677 	int type;
1678 
1679 	leaf = path->nodes[0];
1680 	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1681 	item_offset = (unsigned long)iref - (unsigned long)ei;
1682 
1683 	type = extent_ref_type(parent, owner);
1684 	size = btrfs_extent_inline_ref_size(type);
1685 
1686 	btrfs_extend_item(root, path, size);
1687 
1688 	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1689 	refs = btrfs_extent_refs(leaf, ei);
1690 	refs += refs_to_add;
1691 	btrfs_set_extent_refs(leaf, ei, refs);
1692 	if (extent_op)
1693 		__run_delayed_extent_op(extent_op, leaf, ei);
1694 
1695 	ptr = (unsigned long)ei + item_offset;
1696 	end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1697 	if (ptr < end - size)
1698 		memmove_extent_buffer(leaf, ptr + size, ptr,
1699 				      end - size - ptr);
1700 
1701 	iref = (struct btrfs_extent_inline_ref *)ptr;
1702 	btrfs_set_extent_inline_ref_type(leaf, iref, type);
1703 	if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1704 		struct btrfs_extent_data_ref *dref;
1705 		dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1706 		btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1707 		btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1708 		btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1709 		btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1710 	} else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1711 		struct btrfs_shared_data_ref *sref;
1712 		sref = (struct btrfs_shared_data_ref *)(iref + 1);
1713 		btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1714 		btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1715 	} else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1716 		btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1717 	} else {
1718 		btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1719 	}
1720 	btrfs_mark_buffer_dirty(leaf);
1721 }
1722 
lookup_extent_backref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct btrfs_extent_inline_ref ** ref_ret,u64 bytenr,u64 num_bytes,u64 parent,u64 root_objectid,u64 owner,u64 offset)1723 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1724 				 struct btrfs_root *root,
1725 				 struct btrfs_path *path,
1726 				 struct btrfs_extent_inline_ref **ref_ret,
1727 				 u64 bytenr, u64 num_bytes, u64 parent,
1728 				 u64 root_objectid, u64 owner, u64 offset)
1729 {
1730 	int ret;
1731 
1732 	ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1733 					   bytenr, num_bytes, parent,
1734 					   root_objectid, owner, offset, 0);
1735 	if (ret != -ENOENT)
1736 		return ret;
1737 
1738 	btrfs_release_path(path);
1739 	*ref_ret = NULL;
1740 
1741 	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1742 		ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1743 					    root_objectid);
1744 	} else {
1745 		ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1746 					     root_objectid, owner, offset);
1747 	}
1748 	return ret;
1749 }
1750 
1751 /*
1752  * helper to update/remove inline back ref
1753  */
1754 static noinline_for_stack
update_inline_extent_backref(struct btrfs_root * root,struct btrfs_path * path,struct btrfs_extent_inline_ref * iref,int refs_to_mod,struct btrfs_delayed_extent_op * extent_op,int * last_ref)1755 void update_inline_extent_backref(struct btrfs_root *root,
1756 				  struct btrfs_path *path,
1757 				  struct btrfs_extent_inline_ref *iref,
1758 				  int refs_to_mod,
1759 				  struct btrfs_delayed_extent_op *extent_op,
1760 				  int *last_ref)
1761 {
1762 	struct extent_buffer *leaf;
1763 	struct btrfs_extent_item *ei;
1764 	struct btrfs_extent_data_ref *dref = NULL;
1765 	struct btrfs_shared_data_ref *sref = NULL;
1766 	unsigned long ptr;
1767 	unsigned long end;
1768 	u32 item_size;
1769 	int size;
1770 	int type;
1771 	u64 refs;
1772 
1773 	leaf = path->nodes[0];
1774 	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1775 	refs = btrfs_extent_refs(leaf, ei);
1776 	WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1777 	refs += refs_to_mod;
1778 	btrfs_set_extent_refs(leaf, ei, refs);
1779 	if (extent_op)
1780 		__run_delayed_extent_op(extent_op, leaf, ei);
1781 
1782 	type = btrfs_extent_inline_ref_type(leaf, iref);
1783 
1784 	if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1785 		dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1786 		refs = btrfs_extent_data_ref_count(leaf, dref);
1787 	} else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1788 		sref = (struct btrfs_shared_data_ref *)(iref + 1);
1789 		refs = btrfs_shared_data_ref_count(leaf, sref);
1790 	} else {
1791 		refs = 1;
1792 		BUG_ON(refs_to_mod != -1);
1793 	}
1794 
1795 	BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1796 	refs += refs_to_mod;
1797 
1798 	if (refs > 0) {
1799 		if (type == BTRFS_EXTENT_DATA_REF_KEY)
1800 			btrfs_set_extent_data_ref_count(leaf, dref, refs);
1801 		else
1802 			btrfs_set_shared_data_ref_count(leaf, sref, refs);
1803 	} else {
1804 		*last_ref = 1;
1805 		size =  btrfs_extent_inline_ref_size(type);
1806 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1807 		ptr = (unsigned long)iref;
1808 		end = (unsigned long)ei + item_size;
1809 		if (ptr + size < end)
1810 			memmove_extent_buffer(leaf, ptr, ptr + size,
1811 					      end - ptr - size);
1812 		item_size -= size;
1813 		btrfs_truncate_item(root, path, item_size, 1);
1814 	}
1815 	btrfs_mark_buffer_dirty(leaf);
1816 }
1817 
1818 static noinline_for_stack
insert_inline_extent_backref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,u64 bytenr,u64 num_bytes,u64 parent,u64 root_objectid,u64 owner,u64 offset,int refs_to_add,struct btrfs_delayed_extent_op * extent_op)1819 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1820 				 struct btrfs_root *root,
1821 				 struct btrfs_path *path,
1822 				 u64 bytenr, u64 num_bytes, u64 parent,
1823 				 u64 root_objectid, u64 owner,
1824 				 u64 offset, int refs_to_add,
1825 				 struct btrfs_delayed_extent_op *extent_op)
1826 {
1827 	struct btrfs_extent_inline_ref *iref;
1828 	int ret;
1829 
1830 	ret = lookup_inline_extent_backref(trans, root, path, &iref,
1831 					   bytenr, num_bytes, parent,
1832 					   root_objectid, owner, offset, 1);
1833 	if (ret == 0) {
1834 		BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1835 		update_inline_extent_backref(root, path, iref,
1836 					     refs_to_add, extent_op, NULL);
1837 	} else if (ret == -ENOENT) {
1838 		setup_inline_extent_backref(root, path, iref, parent,
1839 					    root_objectid, owner, offset,
1840 					    refs_to_add, extent_op);
1841 		ret = 0;
1842 	}
1843 	return ret;
1844 }
1845 
insert_extent_backref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,u64 bytenr,u64 parent,u64 root_objectid,u64 owner,u64 offset,int refs_to_add)1846 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1847 				 struct btrfs_root *root,
1848 				 struct btrfs_path *path,
1849 				 u64 bytenr, u64 parent, u64 root_objectid,
1850 				 u64 owner, u64 offset, int refs_to_add)
1851 {
1852 	int ret;
1853 	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1854 		BUG_ON(refs_to_add != 1);
1855 		ret = insert_tree_block_ref(trans, root, path, bytenr,
1856 					    parent, root_objectid);
1857 	} else {
1858 		ret = insert_extent_data_ref(trans, root, path, bytenr,
1859 					     parent, root_objectid,
1860 					     owner, offset, refs_to_add);
1861 	}
1862 	return ret;
1863 }
1864 
remove_extent_backref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct btrfs_extent_inline_ref * iref,int refs_to_drop,int is_data,int * last_ref)1865 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1866 				 struct btrfs_root *root,
1867 				 struct btrfs_path *path,
1868 				 struct btrfs_extent_inline_ref *iref,
1869 				 int refs_to_drop, int is_data, int *last_ref)
1870 {
1871 	int ret = 0;
1872 
1873 	BUG_ON(!is_data && refs_to_drop != 1);
1874 	if (iref) {
1875 		update_inline_extent_backref(root, path, iref,
1876 					     -refs_to_drop, NULL, last_ref);
1877 	} else if (is_data) {
1878 		ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1879 					     last_ref);
1880 	} else {
1881 		*last_ref = 1;
1882 		ret = btrfs_del_item(trans, root, path);
1883 	}
1884 	return ret;
1885 }
1886 
btrfs_issue_discard(struct block_device * bdev,u64 start,u64 len)1887 static int btrfs_issue_discard(struct block_device *bdev,
1888 				u64 start, u64 len)
1889 {
1890 	return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1891 }
1892 
btrfs_discard_extent(struct btrfs_root * root,u64 bytenr,u64 num_bytes,u64 * actual_bytes)1893 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1894 			 u64 num_bytes, u64 *actual_bytes)
1895 {
1896 	int ret;
1897 	u64 discarded_bytes = 0;
1898 	struct btrfs_bio *bbio = NULL;
1899 
1900 
1901 	/* Tell the block device(s) that the sectors can be discarded */
1902 	ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1903 			      bytenr, &num_bytes, &bbio, 0);
1904 	/* Error condition is -ENOMEM */
1905 	if (!ret) {
1906 		struct btrfs_bio_stripe *stripe = bbio->stripes;
1907 		int i;
1908 
1909 
1910 		for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1911 			if (!stripe->dev->can_discard)
1912 				continue;
1913 
1914 			ret = btrfs_issue_discard(stripe->dev->bdev,
1915 						  stripe->physical,
1916 						  stripe->length);
1917 			if (!ret)
1918 				discarded_bytes += stripe->length;
1919 			else if (ret != -EOPNOTSUPP)
1920 				break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1921 
1922 			/*
1923 			 * Just in case we get back EOPNOTSUPP for some reason,
1924 			 * just ignore the return value so we don't screw up
1925 			 * people calling discard_extent.
1926 			 */
1927 			ret = 0;
1928 		}
1929 		btrfs_put_bbio(bbio);
1930 	}
1931 
1932 	if (actual_bytes)
1933 		*actual_bytes = discarded_bytes;
1934 
1935 
1936 	if (ret == -EOPNOTSUPP)
1937 		ret = 0;
1938 	return ret;
1939 }
1940 
1941 /* Can return -ENOMEM */
btrfs_inc_extent_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 bytenr,u64 num_bytes,u64 parent,u64 root_objectid,u64 owner,u64 offset,int no_quota)1942 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1943 			 struct btrfs_root *root,
1944 			 u64 bytenr, u64 num_bytes, u64 parent,
1945 			 u64 root_objectid, u64 owner, u64 offset,
1946 			 int no_quota)
1947 {
1948 	int ret;
1949 	struct btrfs_fs_info *fs_info = root->fs_info;
1950 
1951 	BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1952 	       root_objectid == BTRFS_TREE_LOG_OBJECTID);
1953 
1954 	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1955 		ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1956 					num_bytes,
1957 					parent, root_objectid, (int)owner,
1958 					BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1959 	} else {
1960 		ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1961 					num_bytes,
1962 					parent, root_objectid, owner, offset,
1963 					BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1964 	}
1965 	return ret;
1966 }
1967 
__btrfs_inc_extent_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 bytenr,u64 num_bytes,u64 parent,u64 root_objectid,u64 owner,u64 offset,int refs_to_add,int no_quota,struct btrfs_delayed_extent_op * extent_op)1968 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1969 				  struct btrfs_root *root,
1970 				  u64 bytenr, u64 num_bytes,
1971 				  u64 parent, u64 root_objectid,
1972 				  u64 owner, u64 offset, int refs_to_add,
1973 				  int no_quota,
1974 				  struct btrfs_delayed_extent_op *extent_op)
1975 {
1976 	struct btrfs_fs_info *fs_info = root->fs_info;
1977 	struct btrfs_path *path;
1978 	struct extent_buffer *leaf;
1979 	struct btrfs_extent_item *item;
1980 	struct btrfs_key key;
1981 	u64 refs;
1982 	int ret;
1983 	enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_ADD_EXCL;
1984 
1985 	path = btrfs_alloc_path();
1986 	if (!path)
1987 		return -ENOMEM;
1988 
1989 	if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
1990 		no_quota = 1;
1991 
1992 	path->reada = 1;
1993 	path->leave_spinning = 1;
1994 	/* this will setup the path even if it fails to insert the back ref */
1995 	ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
1996 					   bytenr, num_bytes, parent,
1997 					   root_objectid, owner, offset,
1998 					   refs_to_add, extent_op);
1999 	if ((ret < 0 && ret != -EAGAIN) || (!ret && no_quota))
2000 		goto out;
2001 	/*
2002 	 * Ok we were able to insert an inline extent and it appears to be a new
2003 	 * reference, deal with the qgroup accounting.
2004 	 */
2005 	if (!ret && !no_quota) {
2006 		ASSERT(root->fs_info->quota_enabled);
2007 		leaf = path->nodes[0];
2008 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2009 		item = btrfs_item_ptr(leaf, path->slots[0],
2010 				      struct btrfs_extent_item);
2011 		if (btrfs_extent_refs(leaf, item) > (u64)refs_to_add)
2012 			type = BTRFS_QGROUP_OPER_ADD_SHARED;
2013 		btrfs_release_path(path);
2014 
2015 		ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
2016 					      bytenr, num_bytes, type, 0);
2017 		goto out;
2018 	}
2019 
2020 	/*
2021 	 * Ok we had -EAGAIN which means we didn't have space to insert and
2022 	 * inline extent ref, so just update the reference count and add a
2023 	 * normal backref.
2024 	 */
2025 	leaf = path->nodes[0];
2026 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2027 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2028 	refs = btrfs_extent_refs(leaf, item);
2029 	if (refs)
2030 		type = BTRFS_QGROUP_OPER_ADD_SHARED;
2031 	btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2032 	if (extent_op)
2033 		__run_delayed_extent_op(extent_op, leaf, item);
2034 
2035 	btrfs_mark_buffer_dirty(leaf);
2036 	btrfs_release_path(path);
2037 
2038 	if (!no_quota) {
2039 		ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
2040 					      bytenr, num_bytes, type, 0);
2041 		if (ret)
2042 			goto out;
2043 	}
2044 
2045 	path->reada = 1;
2046 	path->leave_spinning = 1;
2047 	/* now insert the actual backref */
2048 	ret = insert_extent_backref(trans, root->fs_info->extent_root,
2049 				    path, bytenr, parent, root_objectid,
2050 				    owner, offset, refs_to_add);
2051 	if (ret)
2052 		btrfs_abort_transaction(trans, root, ret);
2053 out:
2054 	btrfs_free_path(path);
2055 	return ret;
2056 }
2057 
run_delayed_data_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_delayed_ref_node * node,struct btrfs_delayed_extent_op * extent_op,int insert_reserved)2058 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2059 				struct btrfs_root *root,
2060 				struct btrfs_delayed_ref_node *node,
2061 				struct btrfs_delayed_extent_op *extent_op,
2062 				int insert_reserved)
2063 {
2064 	int ret = 0;
2065 	struct btrfs_delayed_data_ref *ref;
2066 	struct btrfs_key ins;
2067 	u64 parent = 0;
2068 	u64 ref_root = 0;
2069 	u64 flags = 0;
2070 
2071 	ins.objectid = node->bytenr;
2072 	ins.offset = node->num_bytes;
2073 	ins.type = BTRFS_EXTENT_ITEM_KEY;
2074 
2075 	ref = btrfs_delayed_node_to_data_ref(node);
2076 	trace_run_delayed_data_ref(node, ref, node->action);
2077 
2078 	if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2079 		parent = ref->parent;
2080 	ref_root = ref->root;
2081 
2082 	if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2083 		if (extent_op)
2084 			flags |= extent_op->flags_to_set;
2085 		ret = alloc_reserved_file_extent(trans, root,
2086 						 parent, ref_root, flags,
2087 						 ref->objectid, ref->offset,
2088 						 &ins, node->ref_mod);
2089 	} else if (node->action == BTRFS_ADD_DELAYED_REF) {
2090 		ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2091 					     node->num_bytes, parent,
2092 					     ref_root, ref->objectid,
2093 					     ref->offset, node->ref_mod,
2094 					     node->no_quota, extent_op);
2095 	} else if (node->action == BTRFS_DROP_DELAYED_REF) {
2096 		ret = __btrfs_free_extent(trans, root, node->bytenr,
2097 					  node->num_bytes, parent,
2098 					  ref_root, ref->objectid,
2099 					  ref->offset, node->ref_mod,
2100 					  extent_op, node->no_quota);
2101 	} else {
2102 		BUG();
2103 	}
2104 	return ret;
2105 }
2106 
__run_delayed_extent_op(struct btrfs_delayed_extent_op * extent_op,struct extent_buffer * leaf,struct btrfs_extent_item * ei)2107 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2108 				    struct extent_buffer *leaf,
2109 				    struct btrfs_extent_item *ei)
2110 {
2111 	u64 flags = btrfs_extent_flags(leaf, ei);
2112 	if (extent_op->update_flags) {
2113 		flags |= extent_op->flags_to_set;
2114 		btrfs_set_extent_flags(leaf, ei, flags);
2115 	}
2116 
2117 	if (extent_op->update_key) {
2118 		struct btrfs_tree_block_info *bi;
2119 		BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2120 		bi = (struct btrfs_tree_block_info *)(ei + 1);
2121 		btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2122 	}
2123 }
2124 
run_delayed_extent_op(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_delayed_ref_node * node,struct btrfs_delayed_extent_op * extent_op)2125 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2126 				 struct btrfs_root *root,
2127 				 struct btrfs_delayed_ref_node *node,
2128 				 struct btrfs_delayed_extent_op *extent_op)
2129 {
2130 	struct btrfs_key key;
2131 	struct btrfs_path *path;
2132 	struct btrfs_extent_item *ei;
2133 	struct extent_buffer *leaf;
2134 	u32 item_size;
2135 	int ret;
2136 	int err = 0;
2137 	int metadata = !extent_op->is_data;
2138 
2139 	if (trans->aborted)
2140 		return 0;
2141 
2142 	if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2143 		metadata = 0;
2144 
2145 	path = btrfs_alloc_path();
2146 	if (!path)
2147 		return -ENOMEM;
2148 
2149 	key.objectid = node->bytenr;
2150 
2151 	if (metadata) {
2152 		key.type = BTRFS_METADATA_ITEM_KEY;
2153 		key.offset = extent_op->level;
2154 	} else {
2155 		key.type = BTRFS_EXTENT_ITEM_KEY;
2156 		key.offset = node->num_bytes;
2157 	}
2158 
2159 again:
2160 	path->reada = 1;
2161 	path->leave_spinning = 1;
2162 	ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2163 				path, 0, 1);
2164 	if (ret < 0) {
2165 		err = ret;
2166 		goto out;
2167 	}
2168 	if (ret > 0) {
2169 		if (metadata) {
2170 			if (path->slots[0] > 0) {
2171 				path->slots[0]--;
2172 				btrfs_item_key_to_cpu(path->nodes[0], &key,
2173 						      path->slots[0]);
2174 				if (key.objectid == node->bytenr &&
2175 				    key.type == BTRFS_EXTENT_ITEM_KEY &&
2176 				    key.offset == node->num_bytes)
2177 					ret = 0;
2178 			}
2179 			if (ret > 0) {
2180 				btrfs_release_path(path);
2181 				metadata = 0;
2182 
2183 				key.objectid = node->bytenr;
2184 				key.offset = node->num_bytes;
2185 				key.type = BTRFS_EXTENT_ITEM_KEY;
2186 				goto again;
2187 			}
2188 		} else {
2189 			err = -EIO;
2190 			goto out;
2191 		}
2192 	}
2193 
2194 	leaf = path->nodes[0];
2195 	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2196 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2197 	if (item_size < sizeof(*ei)) {
2198 		ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2199 					     path, (u64)-1, 0);
2200 		if (ret < 0) {
2201 			err = ret;
2202 			goto out;
2203 		}
2204 		leaf = path->nodes[0];
2205 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2206 	}
2207 #endif
2208 	BUG_ON(item_size < sizeof(*ei));
2209 	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2210 	__run_delayed_extent_op(extent_op, leaf, ei);
2211 
2212 	btrfs_mark_buffer_dirty(leaf);
2213 out:
2214 	btrfs_free_path(path);
2215 	return err;
2216 }
2217 
run_delayed_tree_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_delayed_ref_node * node,struct btrfs_delayed_extent_op * extent_op,int insert_reserved)2218 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2219 				struct btrfs_root *root,
2220 				struct btrfs_delayed_ref_node *node,
2221 				struct btrfs_delayed_extent_op *extent_op,
2222 				int insert_reserved)
2223 {
2224 	int ret = 0;
2225 	struct btrfs_delayed_tree_ref *ref;
2226 	struct btrfs_key ins;
2227 	u64 parent = 0;
2228 	u64 ref_root = 0;
2229 	bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2230 						 SKINNY_METADATA);
2231 
2232 	ref = btrfs_delayed_node_to_tree_ref(node);
2233 	trace_run_delayed_tree_ref(node, ref, node->action);
2234 
2235 	if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2236 		parent = ref->parent;
2237 	ref_root = ref->root;
2238 
2239 	ins.objectid = node->bytenr;
2240 	if (skinny_metadata) {
2241 		ins.offset = ref->level;
2242 		ins.type = BTRFS_METADATA_ITEM_KEY;
2243 	} else {
2244 		ins.offset = node->num_bytes;
2245 		ins.type = BTRFS_EXTENT_ITEM_KEY;
2246 	}
2247 
2248 	BUG_ON(node->ref_mod != 1);
2249 	if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2250 		BUG_ON(!extent_op || !extent_op->update_flags);
2251 		ret = alloc_reserved_tree_block(trans, root,
2252 						parent, ref_root,
2253 						extent_op->flags_to_set,
2254 						&extent_op->key,
2255 						ref->level, &ins,
2256 						node->no_quota);
2257 	} else if (node->action == BTRFS_ADD_DELAYED_REF) {
2258 		ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2259 					     node->num_bytes, parent, ref_root,
2260 					     ref->level, 0, 1, node->no_quota,
2261 					     extent_op);
2262 	} else if (node->action == BTRFS_DROP_DELAYED_REF) {
2263 		ret = __btrfs_free_extent(trans, root, node->bytenr,
2264 					  node->num_bytes, parent, ref_root,
2265 					  ref->level, 0, 1, extent_op,
2266 					  node->no_quota);
2267 	} else {
2268 		BUG();
2269 	}
2270 	return ret;
2271 }
2272 
2273 /* helper function to actually process a single delayed ref entry */
run_one_delayed_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_delayed_ref_node * node,struct btrfs_delayed_extent_op * extent_op,int insert_reserved)2274 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2275 			       struct btrfs_root *root,
2276 			       struct btrfs_delayed_ref_node *node,
2277 			       struct btrfs_delayed_extent_op *extent_op,
2278 			       int insert_reserved)
2279 {
2280 	int ret = 0;
2281 
2282 	if (trans->aborted) {
2283 		if (insert_reserved)
2284 			btrfs_pin_extent(root, node->bytenr,
2285 					 node->num_bytes, 1);
2286 		return 0;
2287 	}
2288 
2289 	if (btrfs_delayed_ref_is_head(node)) {
2290 		struct btrfs_delayed_ref_head *head;
2291 		/*
2292 		 * we've hit the end of the chain and we were supposed
2293 		 * to insert this extent into the tree.  But, it got
2294 		 * deleted before we ever needed to insert it, so all
2295 		 * we have to do is clean up the accounting
2296 		 */
2297 		BUG_ON(extent_op);
2298 		head = btrfs_delayed_node_to_head(node);
2299 		trace_run_delayed_ref_head(node, head, node->action);
2300 
2301 		if (insert_reserved) {
2302 			btrfs_pin_extent(root, node->bytenr,
2303 					 node->num_bytes, 1);
2304 			if (head->is_data) {
2305 				ret = btrfs_del_csums(trans, root,
2306 						      node->bytenr,
2307 						      node->num_bytes);
2308 			}
2309 		}
2310 		return ret;
2311 	}
2312 
2313 	if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2314 	    node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2315 		ret = run_delayed_tree_ref(trans, root, node, extent_op,
2316 					   insert_reserved);
2317 	else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2318 		 node->type == BTRFS_SHARED_DATA_REF_KEY)
2319 		ret = run_delayed_data_ref(trans, root, node, extent_op,
2320 					   insert_reserved);
2321 	else
2322 		BUG();
2323 	return ret;
2324 }
2325 
2326 static noinline struct btrfs_delayed_ref_node *
select_delayed_ref(struct btrfs_delayed_ref_head * head)2327 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2328 {
2329 	struct rb_node *node;
2330 	struct btrfs_delayed_ref_node *ref, *last = NULL;;
2331 
2332 	/*
2333 	 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2334 	 * this prevents ref count from going down to zero when
2335 	 * there still are pending delayed ref.
2336 	 */
2337 	node = rb_first(&head->ref_root);
2338 	while (node) {
2339 		ref = rb_entry(node, struct btrfs_delayed_ref_node,
2340 				rb_node);
2341 		if (ref->action == BTRFS_ADD_DELAYED_REF)
2342 			return ref;
2343 		else if (last == NULL)
2344 			last = ref;
2345 		node = rb_next(node);
2346 	}
2347 	return last;
2348 }
2349 
2350 /*
2351  * Returns 0 on success or if called with an already aborted transaction.
2352  * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2353  */
__btrfs_run_delayed_refs(struct btrfs_trans_handle * trans,struct btrfs_root * root,unsigned long nr)2354 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2355 					     struct btrfs_root *root,
2356 					     unsigned long nr)
2357 {
2358 	struct btrfs_delayed_ref_root *delayed_refs;
2359 	struct btrfs_delayed_ref_node *ref;
2360 	struct btrfs_delayed_ref_head *locked_ref = NULL;
2361 	struct btrfs_delayed_extent_op *extent_op;
2362 	struct btrfs_fs_info *fs_info = root->fs_info;
2363 	ktime_t start = ktime_get();
2364 	int ret;
2365 	unsigned long count = 0;
2366 	unsigned long actual_count = 0;
2367 	int must_insert_reserved = 0;
2368 
2369 	delayed_refs = &trans->transaction->delayed_refs;
2370 	while (1) {
2371 		if (!locked_ref) {
2372 			if (count >= nr)
2373 				break;
2374 
2375 			spin_lock(&delayed_refs->lock);
2376 			locked_ref = btrfs_select_ref_head(trans);
2377 			if (!locked_ref) {
2378 				spin_unlock(&delayed_refs->lock);
2379 				break;
2380 			}
2381 
2382 			/* grab the lock that says we are going to process
2383 			 * all the refs for this head */
2384 			ret = btrfs_delayed_ref_lock(trans, locked_ref);
2385 			spin_unlock(&delayed_refs->lock);
2386 			/*
2387 			 * we may have dropped the spin lock to get the head
2388 			 * mutex lock, and that might have given someone else
2389 			 * time to free the head.  If that's true, it has been
2390 			 * removed from our list and we can move on.
2391 			 */
2392 			if (ret == -EAGAIN) {
2393 				locked_ref = NULL;
2394 				count++;
2395 				continue;
2396 			}
2397 		}
2398 
2399 		/*
2400 		 * We need to try and merge add/drops of the same ref since we
2401 		 * can run into issues with relocate dropping the implicit ref
2402 		 * and then it being added back again before the drop can
2403 		 * finish.  If we merged anything we need to re-loop so we can
2404 		 * get a good ref.
2405 		 */
2406 		spin_lock(&locked_ref->lock);
2407 		btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2408 					 locked_ref);
2409 
2410 		/*
2411 		 * locked_ref is the head node, so we have to go one
2412 		 * node back for any delayed ref updates
2413 		 */
2414 		ref = select_delayed_ref(locked_ref);
2415 
2416 		if (ref && ref->seq &&
2417 		    btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2418 			spin_unlock(&locked_ref->lock);
2419 			btrfs_delayed_ref_unlock(locked_ref);
2420 			spin_lock(&delayed_refs->lock);
2421 			locked_ref->processing = 0;
2422 			delayed_refs->num_heads_ready++;
2423 			spin_unlock(&delayed_refs->lock);
2424 			locked_ref = NULL;
2425 			cond_resched();
2426 			count++;
2427 			continue;
2428 		}
2429 
2430 		/*
2431 		 * record the must insert reserved flag before we
2432 		 * drop the spin lock.
2433 		 */
2434 		must_insert_reserved = locked_ref->must_insert_reserved;
2435 		locked_ref->must_insert_reserved = 0;
2436 
2437 		extent_op = locked_ref->extent_op;
2438 		locked_ref->extent_op = NULL;
2439 
2440 		if (!ref) {
2441 
2442 
2443 			/* All delayed refs have been processed, Go ahead
2444 			 * and send the head node to run_one_delayed_ref,
2445 			 * so that any accounting fixes can happen
2446 			 */
2447 			ref = &locked_ref->node;
2448 
2449 			if (extent_op && must_insert_reserved) {
2450 				btrfs_free_delayed_extent_op(extent_op);
2451 				extent_op = NULL;
2452 			}
2453 
2454 			if (extent_op) {
2455 				spin_unlock(&locked_ref->lock);
2456 				ret = run_delayed_extent_op(trans, root,
2457 							    ref, extent_op);
2458 				btrfs_free_delayed_extent_op(extent_op);
2459 
2460 				if (ret) {
2461 					/*
2462 					 * Need to reset must_insert_reserved if
2463 					 * there was an error so the abort stuff
2464 					 * can cleanup the reserved space
2465 					 * properly.
2466 					 */
2467 					if (must_insert_reserved)
2468 						locked_ref->must_insert_reserved = 1;
2469 					locked_ref->processing = 0;
2470 					btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2471 					btrfs_delayed_ref_unlock(locked_ref);
2472 					return ret;
2473 				}
2474 				continue;
2475 			}
2476 
2477 			/*
2478 			 * Need to drop our head ref lock and re-aqcuire the
2479 			 * delayed ref lock and then re-check to make sure
2480 			 * nobody got added.
2481 			 */
2482 			spin_unlock(&locked_ref->lock);
2483 			spin_lock(&delayed_refs->lock);
2484 			spin_lock(&locked_ref->lock);
2485 			if (rb_first(&locked_ref->ref_root) ||
2486 			    locked_ref->extent_op) {
2487 				spin_unlock(&locked_ref->lock);
2488 				spin_unlock(&delayed_refs->lock);
2489 				continue;
2490 			}
2491 			ref->in_tree = 0;
2492 			delayed_refs->num_heads--;
2493 			rb_erase(&locked_ref->href_node,
2494 				 &delayed_refs->href_root);
2495 			spin_unlock(&delayed_refs->lock);
2496 		} else {
2497 			actual_count++;
2498 			ref->in_tree = 0;
2499 			rb_erase(&ref->rb_node, &locked_ref->ref_root);
2500 		}
2501 		atomic_dec(&delayed_refs->num_entries);
2502 
2503 		if (!btrfs_delayed_ref_is_head(ref)) {
2504 			/*
2505 			 * when we play the delayed ref, also correct the
2506 			 * ref_mod on head
2507 			 */
2508 			switch (ref->action) {
2509 			case BTRFS_ADD_DELAYED_REF:
2510 			case BTRFS_ADD_DELAYED_EXTENT:
2511 				locked_ref->node.ref_mod -= ref->ref_mod;
2512 				break;
2513 			case BTRFS_DROP_DELAYED_REF:
2514 				locked_ref->node.ref_mod += ref->ref_mod;
2515 				break;
2516 			default:
2517 				WARN_ON(1);
2518 			}
2519 		}
2520 		spin_unlock(&locked_ref->lock);
2521 
2522 		ret = run_one_delayed_ref(trans, root, ref, extent_op,
2523 					  must_insert_reserved);
2524 
2525 		btrfs_free_delayed_extent_op(extent_op);
2526 		if (ret) {
2527 			locked_ref->processing = 0;
2528 			btrfs_delayed_ref_unlock(locked_ref);
2529 			btrfs_put_delayed_ref(ref);
2530 			btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2531 			return ret;
2532 		}
2533 
2534 		/*
2535 		 * If this node is a head, that means all the refs in this head
2536 		 * have been dealt with, and we will pick the next head to deal
2537 		 * with, so we must unlock the head and drop it from the cluster
2538 		 * list before we release it.
2539 		 */
2540 		if (btrfs_delayed_ref_is_head(ref)) {
2541 			if (locked_ref->is_data &&
2542 			    locked_ref->total_ref_mod < 0) {
2543 				spin_lock(&delayed_refs->lock);
2544 				delayed_refs->pending_csums -= ref->num_bytes;
2545 				spin_unlock(&delayed_refs->lock);
2546 			}
2547 			btrfs_delayed_ref_unlock(locked_ref);
2548 			locked_ref = NULL;
2549 		}
2550 		btrfs_put_delayed_ref(ref);
2551 		count++;
2552 		cond_resched();
2553 	}
2554 
2555 	/*
2556 	 * We don't want to include ref heads since we can have empty ref heads
2557 	 * and those will drastically skew our runtime down since we just do
2558 	 * accounting, no actual extent tree updates.
2559 	 */
2560 	if (actual_count > 0) {
2561 		u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2562 		u64 avg;
2563 
2564 		/*
2565 		 * We weigh the current average higher than our current runtime
2566 		 * to avoid large swings in the average.
2567 		 */
2568 		spin_lock(&delayed_refs->lock);
2569 		avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2570 		fs_info->avg_delayed_ref_runtime = avg >> 2;	/* div by 4 */
2571 		spin_unlock(&delayed_refs->lock);
2572 	}
2573 	return 0;
2574 }
2575 
2576 #ifdef SCRAMBLE_DELAYED_REFS
2577 /*
2578  * Normally delayed refs get processed in ascending bytenr order. This
2579  * correlates in most cases to the order added. To expose dependencies on this
2580  * order, we start to process the tree in the middle instead of the beginning
2581  */
find_middle(struct rb_root * root)2582 static u64 find_middle(struct rb_root *root)
2583 {
2584 	struct rb_node *n = root->rb_node;
2585 	struct btrfs_delayed_ref_node *entry;
2586 	int alt = 1;
2587 	u64 middle;
2588 	u64 first = 0, last = 0;
2589 
2590 	n = rb_first(root);
2591 	if (n) {
2592 		entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2593 		first = entry->bytenr;
2594 	}
2595 	n = rb_last(root);
2596 	if (n) {
2597 		entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2598 		last = entry->bytenr;
2599 	}
2600 	n = root->rb_node;
2601 
2602 	while (n) {
2603 		entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2604 		WARN_ON(!entry->in_tree);
2605 
2606 		middle = entry->bytenr;
2607 
2608 		if (alt)
2609 			n = n->rb_left;
2610 		else
2611 			n = n->rb_right;
2612 
2613 		alt = 1 - alt;
2614 	}
2615 	return middle;
2616 }
2617 #endif
2618 
heads_to_leaves(struct btrfs_root * root,u64 heads)2619 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2620 {
2621 	u64 num_bytes;
2622 
2623 	num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2624 			     sizeof(struct btrfs_extent_inline_ref));
2625 	if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2626 		num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2627 
2628 	/*
2629 	 * We don't ever fill up leaves all the way so multiply by 2 just to be
2630 	 * closer to what we're really going to want to ouse.
2631 	 */
2632 	return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2633 }
2634 
2635 /*
2636  * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2637  * would require to store the csums for that many bytes.
2638  */
btrfs_csum_bytes_to_leaves(struct btrfs_root * root,u64 csum_bytes)2639 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2640 {
2641 	u64 csum_size;
2642 	u64 num_csums_per_leaf;
2643 	u64 num_csums;
2644 
2645 	csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2646 	num_csums_per_leaf = div64_u64(csum_size,
2647 			(u64)btrfs_super_csum_size(root->fs_info->super_copy));
2648 	num_csums = div64_u64(csum_bytes, root->sectorsize);
2649 	num_csums += num_csums_per_leaf - 1;
2650 	num_csums = div64_u64(num_csums, num_csums_per_leaf);
2651 	return num_csums;
2652 }
2653 
btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle * trans,struct btrfs_root * root)2654 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2655 				       struct btrfs_root *root)
2656 {
2657 	struct btrfs_block_rsv *global_rsv;
2658 	u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2659 	u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2660 	u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2661 	u64 num_bytes, num_dirty_bgs_bytes;
2662 	int ret = 0;
2663 
2664 	num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2665 	num_heads = heads_to_leaves(root, num_heads);
2666 	if (num_heads > 1)
2667 		num_bytes += (num_heads - 1) * root->nodesize;
2668 	num_bytes <<= 1;
2669 	num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2670 	num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2671 							     num_dirty_bgs);
2672 	global_rsv = &root->fs_info->global_block_rsv;
2673 
2674 	/*
2675 	 * If we can't allocate any more chunks lets make sure we have _lots_ of
2676 	 * wiggle room since running delayed refs can create more delayed refs.
2677 	 */
2678 	if (global_rsv->space_info->full) {
2679 		num_dirty_bgs_bytes <<= 1;
2680 		num_bytes <<= 1;
2681 	}
2682 
2683 	spin_lock(&global_rsv->lock);
2684 	if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2685 		ret = 1;
2686 	spin_unlock(&global_rsv->lock);
2687 	return ret;
2688 }
2689 
btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle * trans,struct btrfs_root * root)2690 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2691 				       struct btrfs_root *root)
2692 {
2693 	struct btrfs_fs_info *fs_info = root->fs_info;
2694 	u64 num_entries =
2695 		atomic_read(&trans->transaction->delayed_refs.num_entries);
2696 	u64 avg_runtime;
2697 	u64 val;
2698 
2699 	smp_mb();
2700 	avg_runtime = fs_info->avg_delayed_ref_runtime;
2701 	val = num_entries * avg_runtime;
2702 	if (num_entries * avg_runtime >= NSEC_PER_SEC)
2703 		return 1;
2704 	if (val >= NSEC_PER_SEC / 2)
2705 		return 2;
2706 
2707 	return btrfs_check_space_for_delayed_refs(trans, root);
2708 }
2709 
2710 struct async_delayed_refs {
2711 	struct btrfs_root *root;
2712 	int count;
2713 	int error;
2714 	int sync;
2715 	struct completion wait;
2716 	struct btrfs_work work;
2717 };
2718 
delayed_ref_async_start(struct btrfs_work * work)2719 static void delayed_ref_async_start(struct btrfs_work *work)
2720 {
2721 	struct async_delayed_refs *async;
2722 	struct btrfs_trans_handle *trans;
2723 	int ret;
2724 
2725 	async = container_of(work, struct async_delayed_refs, work);
2726 
2727 	trans = btrfs_join_transaction(async->root);
2728 	if (IS_ERR(trans)) {
2729 		async->error = PTR_ERR(trans);
2730 		goto done;
2731 	}
2732 
2733 	/*
2734 	 * trans->sync means that when we call end_transaciton, we won't
2735 	 * wait on delayed refs
2736 	 */
2737 	trans->sync = true;
2738 	ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2739 	if (ret)
2740 		async->error = ret;
2741 
2742 	ret = btrfs_end_transaction(trans, async->root);
2743 	if (ret && !async->error)
2744 		async->error = ret;
2745 done:
2746 	if (async->sync)
2747 		complete(&async->wait);
2748 	else
2749 		kfree(async);
2750 }
2751 
btrfs_async_run_delayed_refs(struct btrfs_root * root,unsigned long count,int wait)2752 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2753 				 unsigned long count, int wait)
2754 {
2755 	struct async_delayed_refs *async;
2756 	int ret;
2757 
2758 	async = kmalloc(sizeof(*async), GFP_NOFS);
2759 	if (!async)
2760 		return -ENOMEM;
2761 
2762 	async->root = root->fs_info->tree_root;
2763 	async->count = count;
2764 	async->error = 0;
2765 	if (wait)
2766 		async->sync = 1;
2767 	else
2768 		async->sync = 0;
2769 	init_completion(&async->wait);
2770 
2771 	btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2772 			delayed_ref_async_start, NULL, NULL);
2773 
2774 	btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2775 
2776 	if (wait) {
2777 		wait_for_completion(&async->wait);
2778 		ret = async->error;
2779 		kfree(async);
2780 		return ret;
2781 	}
2782 	return 0;
2783 }
2784 
2785 /*
2786  * this starts processing the delayed reference count updates and
2787  * extent insertions we have queued up so far.  count can be
2788  * 0, which means to process everything in the tree at the start
2789  * of the run (but not newly added entries), or it can be some target
2790  * number you'd like to process.
2791  *
2792  * Returns 0 on success or if called with an aborted transaction
2793  * Returns <0 on error and aborts the transaction
2794  */
btrfs_run_delayed_refs(struct btrfs_trans_handle * trans,struct btrfs_root * root,unsigned long count)2795 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2796 			   struct btrfs_root *root, unsigned long count)
2797 {
2798 	struct rb_node *node;
2799 	struct btrfs_delayed_ref_root *delayed_refs;
2800 	struct btrfs_delayed_ref_head *head;
2801 	int ret;
2802 	int run_all = count == (unsigned long)-1;
2803 
2804 	/* We'll clean this up in btrfs_cleanup_transaction */
2805 	if (trans->aborted)
2806 		return 0;
2807 
2808 	if (root == root->fs_info->extent_root)
2809 		root = root->fs_info->tree_root;
2810 
2811 	delayed_refs = &trans->transaction->delayed_refs;
2812 	if (count == 0)
2813 		count = atomic_read(&delayed_refs->num_entries) * 2;
2814 
2815 again:
2816 #ifdef SCRAMBLE_DELAYED_REFS
2817 	delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2818 #endif
2819 	ret = __btrfs_run_delayed_refs(trans, root, count);
2820 	if (ret < 0) {
2821 		btrfs_abort_transaction(trans, root, ret);
2822 		return ret;
2823 	}
2824 
2825 	if (run_all) {
2826 		if (!list_empty(&trans->new_bgs))
2827 			btrfs_create_pending_block_groups(trans, root);
2828 
2829 		spin_lock(&delayed_refs->lock);
2830 		node = rb_first(&delayed_refs->href_root);
2831 		if (!node) {
2832 			spin_unlock(&delayed_refs->lock);
2833 			goto out;
2834 		}
2835 		count = (unsigned long)-1;
2836 
2837 		while (node) {
2838 			head = rb_entry(node, struct btrfs_delayed_ref_head,
2839 					href_node);
2840 			if (btrfs_delayed_ref_is_head(&head->node)) {
2841 				struct btrfs_delayed_ref_node *ref;
2842 
2843 				ref = &head->node;
2844 				atomic_inc(&ref->refs);
2845 
2846 				spin_unlock(&delayed_refs->lock);
2847 				/*
2848 				 * Mutex was contended, block until it's
2849 				 * released and try again
2850 				 */
2851 				mutex_lock(&head->mutex);
2852 				mutex_unlock(&head->mutex);
2853 
2854 				btrfs_put_delayed_ref(ref);
2855 				cond_resched();
2856 				goto again;
2857 			} else {
2858 				WARN_ON(1);
2859 			}
2860 			node = rb_next(node);
2861 		}
2862 		spin_unlock(&delayed_refs->lock);
2863 		cond_resched();
2864 		goto again;
2865 	}
2866 out:
2867 	ret = btrfs_delayed_qgroup_accounting(trans, root->fs_info);
2868 	if (ret)
2869 		return ret;
2870 	assert_qgroups_uptodate(trans);
2871 	return 0;
2872 }
2873 
btrfs_set_disk_extent_flags(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 bytenr,u64 num_bytes,u64 flags,int level,int is_data)2874 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2875 				struct btrfs_root *root,
2876 				u64 bytenr, u64 num_bytes, u64 flags,
2877 				int level, int is_data)
2878 {
2879 	struct btrfs_delayed_extent_op *extent_op;
2880 	int ret;
2881 
2882 	extent_op = btrfs_alloc_delayed_extent_op();
2883 	if (!extent_op)
2884 		return -ENOMEM;
2885 
2886 	extent_op->flags_to_set = flags;
2887 	extent_op->update_flags = 1;
2888 	extent_op->update_key = 0;
2889 	extent_op->is_data = is_data ? 1 : 0;
2890 	extent_op->level = level;
2891 
2892 	ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2893 					  num_bytes, extent_op);
2894 	if (ret)
2895 		btrfs_free_delayed_extent_op(extent_op);
2896 	return ret;
2897 }
2898 
check_delayed_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,u64 objectid,u64 offset,u64 bytenr)2899 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2900 				      struct btrfs_root *root,
2901 				      struct btrfs_path *path,
2902 				      u64 objectid, u64 offset, u64 bytenr)
2903 {
2904 	struct btrfs_delayed_ref_head *head;
2905 	struct btrfs_delayed_ref_node *ref;
2906 	struct btrfs_delayed_data_ref *data_ref;
2907 	struct btrfs_delayed_ref_root *delayed_refs;
2908 	struct rb_node *node;
2909 	int ret = 0;
2910 
2911 	delayed_refs = &trans->transaction->delayed_refs;
2912 	spin_lock(&delayed_refs->lock);
2913 	head = btrfs_find_delayed_ref_head(trans, bytenr);
2914 	if (!head) {
2915 		spin_unlock(&delayed_refs->lock);
2916 		return 0;
2917 	}
2918 
2919 	if (!mutex_trylock(&head->mutex)) {
2920 		atomic_inc(&head->node.refs);
2921 		spin_unlock(&delayed_refs->lock);
2922 
2923 		btrfs_release_path(path);
2924 
2925 		/*
2926 		 * Mutex was contended, block until it's released and let
2927 		 * caller try again
2928 		 */
2929 		mutex_lock(&head->mutex);
2930 		mutex_unlock(&head->mutex);
2931 		btrfs_put_delayed_ref(&head->node);
2932 		return -EAGAIN;
2933 	}
2934 	spin_unlock(&delayed_refs->lock);
2935 
2936 	spin_lock(&head->lock);
2937 	node = rb_first(&head->ref_root);
2938 	while (node) {
2939 		ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2940 		node = rb_next(node);
2941 
2942 		/* If it's a shared ref we know a cross reference exists */
2943 		if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2944 			ret = 1;
2945 			break;
2946 		}
2947 
2948 		data_ref = btrfs_delayed_node_to_data_ref(ref);
2949 
2950 		/*
2951 		 * If our ref doesn't match the one we're currently looking at
2952 		 * then we have a cross reference.
2953 		 */
2954 		if (data_ref->root != root->root_key.objectid ||
2955 		    data_ref->objectid != objectid ||
2956 		    data_ref->offset != offset) {
2957 			ret = 1;
2958 			break;
2959 		}
2960 	}
2961 	spin_unlock(&head->lock);
2962 	mutex_unlock(&head->mutex);
2963 	return ret;
2964 }
2965 
check_committed_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,u64 objectid,u64 offset,u64 bytenr)2966 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2967 					struct btrfs_root *root,
2968 					struct btrfs_path *path,
2969 					u64 objectid, u64 offset, u64 bytenr)
2970 {
2971 	struct btrfs_root *extent_root = root->fs_info->extent_root;
2972 	struct extent_buffer *leaf;
2973 	struct btrfs_extent_data_ref *ref;
2974 	struct btrfs_extent_inline_ref *iref;
2975 	struct btrfs_extent_item *ei;
2976 	struct btrfs_key key;
2977 	u32 item_size;
2978 	int ret;
2979 
2980 	key.objectid = bytenr;
2981 	key.offset = (u64)-1;
2982 	key.type = BTRFS_EXTENT_ITEM_KEY;
2983 
2984 	ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2985 	if (ret < 0)
2986 		goto out;
2987 	BUG_ON(ret == 0); /* Corruption */
2988 
2989 	ret = -ENOENT;
2990 	if (path->slots[0] == 0)
2991 		goto out;
2992 
2993 	path->slots[0]--;
2994 	leaf = path->nodes[0];
2995 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2996 
2997 	if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2998 		goto out;
2999 
3000 	ret = 1;
3001 	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3002 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3003 	if (item_size < sizeof(*ei)) {
3004 		WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3005 		goto out;
3006 	}
3007 #endif
3008 	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3009 
3010 	if (item_size != sizeof(*ei) +
3011 	    btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3012 		goto out;
3013 
3014 	if (btrfs_extent_generation(leaf, ei) <=
3015 	    btrfs_root_last_snapshot(&root->root_item))
3016 		goto out;
3017 
3018 	iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3019 	if (btrfs_extent_inline_ref_type(leaf, iref) !=
3020 	    BTRFS_EXTENT_DATA_REF_KEY)
3021 		goto out;
3022 
3023 	ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3024 	if (btrfs_extent_refs(leaf, ei) !=
3025 	    btrfs_extent_data_ref_count(leaf, ref) ||
3026 	    btrfs_extent_data_ref_root(leaf, ref) !=
3027 	    root->root_key.objectid ||
3028 	    btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3029 	    btrfs_extent_data_ref_offset(leaf, ref) != offset)
3030 		goto out;
3031 
3032 	ret = 0;
3033 out:
3034 	return ret;
3035 }
3036 
btrfs_cross_ref_exist(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 objectid,u64 offset,u64 bytenr)3037 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3038 			  struct btrfs_root *root,
3039 			  u64 objectid, u64 offset, u64 bytenr)
3040 {
3041 	struct btrfs_path *path;
3042 	int ret;
3043 	int ret2;
3044 
3045 	path = btrfs_alloc_path();
3046 	if (!path)
3047 		return -ENOENT;
3048 
3049 	do {
3050 		ret = check_committed_ref(trans, root, path, objectid,
3051 					  offset, bytenr);
3052 		if (ret && ret != -ENOENT)
3053 			goto out;
3054 
3055 		ret2 = check_delayed_ref(trans, root, path, objectid,
3056 					 offset, bytenr);
3057 	} while (ret2 == -EAGAIN);
3058 
3059 	if (ret2 && ret2 != -ENOENT) {
3060 		ret = ret2;
3061 		goto out;
3062 	}
3063 
3064 	if (ret != -ENOENT || ret2 != -ENOENT)
3065 		ret = 0;
3066 out:
3067 	btrfs_free_path(path);
3068 	if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3069 		WARN_ON(ret > 0);
3070 	return ret;
3071 }
3072 
__btrfs_mod_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * buf,int full_backref,int inc)3073 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3074 			   struct btrfs_root *root,
3075 			   struct extent_buffer *buf,
3076 			   int full_backref, int inc)
3077 {
3078 	u64 bytenr;
3079 	u64 num_bytes;
3080 	u64 parent;
3081 	u64 ref_root;
3082 	u32 nritems;
3083 	struct btrfs_key key;
3084 	struct btrfs_file_extent_item *fi;
3085 	int i;
3086 	int level;
3087 	int ret = 0;
3088 	int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3089 			    u64, u64, u64, u64, u64, u64, int);
3090 
3091 
3092 	if (btrfs_test_is_dummy_root(root))
3093 		return 0;
3094 
3095 	ref_root = btrfs_header_owner(buf);
3096 	nritems = btrfs_header_nritems(buf);
3097 	level = btrfs_header_level(buf);
3098 
3099 	if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3100 		return 0;
3101 
3102 	if (inc)
3103 		process_func = btrfs_inc_extent_ref;
3104 	else
3105 		process_func = btrfs_free_extent;
3106 
3107 	if (full_backref)
3108 		parent = buf->start;
3109 	else
3110 		parent = 0;
3111 
3112 	for (i = 0; i < nritems; i++) {
3113 		if (level == 0) {
3114 			btrfs_item_key_to_cpu(buf, &key, i);
3115 			if (key.type != BTRFS_EXTENT_DATA_KEY)
3116 				continue;
3117 			fi = btrfs_item_ptr(buf, i,
3118 					    struct btrfs_file_extent_item);
3119 			if (btrfs_file_extent_type(buf, fi) ==
3120 			    BTRFS_FILE_EXTENT_INLINE)
3121 				continue;
3122 			bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3123 			if (bytenr == 0)
3124 				continue;
3125 
3126 			num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3127 			key.offset -= btrfs_file_extent_offset(buf, fi);
3128 			ret = process_func(trans, root, bytenr, num_bytes,
3129 					   parent, ref_root, key.objectid,
3130 					   key.offset, 1);
3131 			if (ret)
3132 				goto fail;
3133 		} else {
3134 			bytenr = btrfs_node_blockptr(buf, i);
3135 			num_bytes = root->nodesize;
3136 			ret = process_func(trans, root, bytenr, num_bytes,
3137 					   parent, ref_root, level - 1, 0,
3138 					   1);
3139 			if (ret)
3140 				goto fail;
3141 		}
3142 	}
3143 	return 0;
3144 fail:
3145 	return ret;
3146 }
3147 
btrfs_inc_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * buf,int full_backref)3148 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3149 		  struct extent_buffer *buf, int full_backref)
3150 {
3151 	return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3152 }
3153 
btrfs_dec_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * buf,int full_backref)3154 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3155 		  struct extent_buffer *buf, int full_backref)
3156 {
3157 	return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3158 }
3159 
write_one_cache_group(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct btrfs_block_group_cache * cache)3160 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3161 				 struct btrfs_root *root,
3162 				 struct btrfs_path *path,
3163 				 struct btrfs_block_group_cache *cache)
3164 {
3165 	int ret;
3166 	struct btrfs_root *extent_root = root->fs_info->extent_root;
3167 	unsigned long bi;
3168 	struct extent_buffer *leaf;
3169 
3170 	ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3171 	if (ret) {
3172 		if (ret > 0)
3173 			ret = -ENOENT;
3174 		goto fail;
3175 	}
3176 
3177 	leaf = path->nodes[0];
3178 	bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3179 	write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3180 	btrfs_mark_buffer_dirty(leaf);
3181 fail:
3182 	btrfs_release_path(path);
3183 	return ret;
3184 
3185 }
3186 
3187 static struct btrfs_block_group_cache *
next_block_group(struct btrfs_root * root,struct btrfs_block_group_cache * cache)3188 next_block_group(struct btrfs_root *root,
3189 		 struct btrfs_block_group_cache *cache)
3190 {
3191 	struct rb_node *node;
3192 
3193 	spin_lock(&root->fs_info->block_group_cache_lock);
3194 
3195 	/* If our block group was removed, we need a full search. */
3196 	if (RB_EMPTY_NODE(&cache->cache_node)) {
3197 		const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3198 
3199 		spin_unlock(&root->fs_info->block_group_cache_lock);
3200 		btrfs_put_block_group(cache);
3201 		cache = btrfs_lookup_first_block_group(root->fs_info,
3202 						       next_bytenr);
3203 		return cache;
3204 	}
3205 	node = rb_next(&cache->cache_node);
3206 	btrfs_put_block_group(cache);
3207 	if (node) {
3208 		cache = rb_entry(node, struct btrfs_block_group_cache,
3209 				 cache_node);
3210 		btrfs_get_block_group(cache);
3211 	} else
3212 		cache = NULL;
3213 	spin_unlock(&root->fs_info->block_group_cache_lock);
3214 	return cache;
3215 }
3216 
cache_save_setup(struct btrfs_block_group_cache * block_group,struct btrfs_trans_handle * trans,struct btrfs_path * path)3217 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3218 			    struct btrfs_trans_handle *trans,
3219 			    struct btrfs_path *path)
3220 {
3221 	struct btrfs_root *root = block_group->fs_info->tree_root;
3222 	struct inode *inode = NULL;
3223 	u64 alloc_hint = 0;
3224 	int dcs = BTRFS_DC_ERROR;
3225 	u64 num_pages = 0;
3226 	int retries = 0;
3227 	int ret = 0;
3228 
3229 	/*
3230 	 * If this block group is smaller than 100 megs don't bother caching the
3231 	 * block group.
3232 	 */
3233 	if (block_group->key.offset < (100 * 1024 * 1024)) {
3234 		spin_lock(&block_group->lock);
3235 		block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3236 		spin_unlock(&block_group->lock);
3237 		return 0;
3238 	}
3239 
3240 	if (trans->aborted)
3241 		return 0;
3242 again:
3243 	inode = lookup_free_space_inode(root, block_group, path);
3244 	if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3245 		ret = PTR_ERR(inode);
3246 		btrfs_release_path(path);
3247 		goto out;
3248 	}
3249 
3250 	if (IS_ERR(inode)) {
3251 		BUG_ON(retries);
3252 		retries++;
3253 
3254 		if (block_group->ro)
3255 			goto out_free;
3256 
3257 		ret = create_free_space_inode(root, trans, block_group, path);
3258 		if (ret)
3259 			goto out_free;
3260 		goto again;
3261 	}
3262 
3263 	/* We've already setup this transaction, go ahead and exit */
3264 	if (block_group->cache_generation == trans->transid &&
3265 	    i_size_read(inode)) {
3266 		dcs = BTRFS_DC_SETUP;
3267 		goto out_put;
3268 	}
3269 
3270 	/*
3271 	 * We want to set the generation to 0, that way if anything goes wrong
3272 	 * from here on out we know not to trust this cache when we load up next
3273 	 * time.
3274 	 */
3275 	BTRFS_I(inode)->generation = 0;
3276 	ret = btrfs_update_inode(trans, root, inode);
3277 	if (ret) {
3278 		/*
3279 		 * So theoretically we could recover from this, simply set the
3280 		 * super cache generation to 0 so we know to invalidate the
3281 		 * cache, but then we'd have to keep track of the block groups
3282 		 * that fail this way so we know we _have_ to reset this cache
3283 		 * before the next commit or risk reading stale cache.  So to
3284 		 * limit our exposure to horrible edge cases lets just abort the
3285 		 * transaction, this only happens in really bad situations
3286 		 * anyway.
3287 		 */
3288 		btrfs_abort_transaction(trans, root, ret);
3289 		goto out_put;
3290 	}
3291 	WARN_ON(ret);
3292 
3293 	if (i_size_read(inode) > 0) {
3294 		ret = btrfs_check_trunc_cache_free_space(root,
3295 					&root->fs_info->global_block_rsv);
3296 		if (ret)
3297 			goto out_put;
3298 
3299 		ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3300 		if (ret)
3301 			goto out_put;
3302 	}
3303 
3304 	spin_lock(&block_group->lock);
3305 	if (block_group->cached != BTRFS_CACHE_FINISHED ||
3306 	    !btrfs_test_opt(root, SPACE_CACHE)) {
3307 		/*
3308 		 * don't bother trying to write stuff out _if_
3309 		 * a) we're not cached,
3310 		 * b) we're with nospace_cache mount option.
3311 		 */
3312 		dcs = BTRFS_DC_WRITTEN;
3313 		spin_unlock(&block_group->lock);
3314 		goto out_put;
3315 	}
3316 	spin_unlock(&block_group->lock);
3317 
3318 	/*
3319 	 * Try to preallocate enough space based on how big the block group is.
3320 	 * Keep in mind this has to include any pinned space which could end up
3321 	 * taking up quite a bit since it's not folded into the other space
3322 	 * cache.
3323 	 */
3324 	num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3325 	if (!num_pages)
3326 		num_pages = 1;
3327 
3328 	num_pages *= 16;
3329 	num_pages *= PAGE_CACHE_SIZE;
3330 
3331 	ret = btrfs_check_data_free_space(inode, num_pages, num_pages);
3332 	if (ret)
3333 		goto out_put;
3334 
3335 	ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3336 					      num_pages, num_pages,
3337 					      &alloc_hint);
3338 	if (!ret)
3339 		dcs = BTRFS_DC_SETUP;
3340 	btrfs_free_reserved_data_space(inode, num_pages);
3341 
3342 out_put:
3343 	iput(inode);
3344 out_free:
3345 	btrfs_release_path(path);
3346 out:
3347 	spin_lock(&block_group->lock);
3348 	if (!ret && dcs == BTRFS_DC_SETUP)
3349 		block_group->cache_generation = trans->transid;
3350 	block_group->disk_cache_state = dcs;
3351 	spin_unlock(&block_group->lock);
3352 
3353 	return ret;
3354 }
3355 
btrfs_setup_space_cache(struct btrfs_trans_handle * trans,struct btrfs_root * root)3356 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3357 			    struct btrfs_root *root)
3358 {
3359 	struct btrfs_block_group_cache *cache, *tmp;
3360 	struct btrfs_transaction *cur_trans = trans->transaction;
3361 	struct btrfs_path *path;
3362 
3363 	if (list_empty(&cur_trans->dirty_bgs) ||
3364 	    !btrfs_test_opt(root, SPACE_CACHE))
3365 		return 0;
3366 
3367 	path = btrfs_alloc_path();
3368 	if (!path)
3369 		return -ENOMEM;
3370 
3371 	/* Could add new block groups, use _safe just in case */
3372 	list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3373 				 dirty_list) {
3374 		if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3375 			cache_save_setup(cache, trans, path);
3376 	}
3377 
3378 	btrfs_free_path(path);
3379 	return 0;
3380 }
3381 
3382 /*
3383  * transaction commit does final block group cache writeback during a
3384  * critical section where nothing is allowed to change the FS.  This is
3385  * required in order for the cache to actually match the block group,
3386  * but can introduce a lot of latency into the commit.
3387  *
3388  * So, btrfs_start_dirty_block_groups is here to kick off block group
3389  * cache IO.  There's a chance we'll have to redo some of it if the
3390  * block group changes again during the commit, but it greatly reduces
3391  * the commit latency by getting rid of the easy block groups while
3392  * we're still allowing others to join the commit.
3393  */
btrfs_start_dirty_block_groups(struct btrfs_trans_handle * trans,struct btrfs_root * root)3394 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3395 				   struct btrfs_root *root)
3396 {
3397 	struct btrfs_block_group_cache *cache;
3398 	struct btrfs_transaction *cur_trans = trans->transaction;
3399 	int ret = 0;
3400 	int should_put;
3401 	struct btrfs_path *path = NULL;
3402 	LIST_HEAD(dirty);
3403 	struct list_head *io = &cur_trans->io_bgs;
3404 	int num_started = 0;
3405 	int loops = 0;
3406 
3407 	spin_lock(&cur_trans->dirty_bgs_lock);
3408 	if (list_empty(&cur_trans->dirty_bgs)) {
3409 		spin_unlock(&cur_trans->dirty_bgs_lock);
3410 		return 0;
3411 	}
3412 	list_splice_init(&cur_trans->dirty_bgs, &dirty);
3413 	spin_unlock(&cur_trans->dirty_bgs_lock);
3414 
3415 again:
3416 	/*
3417 	 * make sure all the block groups on our dirty list actually
3418 	 * exist
3419 	 */
3420 	btrfs_create_pending_block_groups(trans, root);
3421 
3422 	if (!path) {
3423 		path = btrfs_alloc_path();
3424 		if (!path)
3425 			return -ENOMEM;
3426 	}
3427 
3428 	/*
3429 	 * cache_write_mutex is here only to save us from balance or automatic
3430 	 * removal of empty block groups deleting this block group while we are
3431 	 * writing out the cache
3432 	 */
3433 	mutex_lock(&trans->transaction->cache_write_mutex);
3434 	while (!list_empty(&dirty)) {
3435 		cache = list_first_entry(&dirty,
3436 					 struct btrfs_block_group_cache,
3437 					 dirty_list);
3438 		/*
3439 		 * this can happen if something re-dirties a block
3440 		 * group that is already under IO.  Just wait for it to
3441 		 * finish and then do it all again
3442 		 */
3443 		if (!list_empty(&cache->io_list)) {
3444 			list_del_init(&cache->io_list);
3445 			btrfs_wait_cache_io(root, trans, cache,
3446 					    &cache->io_ctl, path,
3447 					    cache->key.objectid);
3448 			btrfs_put_block_group(cache);
3449 		}
3450 
3451 
3452 		/*
3453 		 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3454 		 * if it should update the cache_state.  Don't delete
3455 		 * until after we wait.
3456 		 *
3457 		 * Since we're not running in the commit critical section
3458 		 * we need the dirty_bgs_lock to protect from update_block_group
3459 		 */
3460 		spin_lock(&cur_trans->dirty_bgs_lock);
3461 		list_del_init(&cache->dirty_list);
3462 		spin_unlock(&cur_trans->dirty_bgs_lock);
3463 
3464 		should_put = 1;
3465 
3466 		cache_save_setup(cache, trans, path);
3467 
3468 		if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3469 			cache->io_ctl.inode = NULL;
3470 			ret = btrfs_write_out_cache(root, trans, cache, path);
3471 			if (ret == 0 && cache->io_ctl.inode) {
3472 				num_started++;
3473 				should_put = 0;
3474 
3475 				/*
3476 				 * the cache_write_mutex is protecting
3477 				 * the io_list
3478 				 */
3479 				list_add_tail(&cache->io_list, io);
3480 			} else {
3481 				/*
3482 				 * if we failed to write the cache, the
3483 				 * generation will be bad and life goes on
3484 				 */
3485 				ret = 0;
3486 			}
3487 		}
3488 		if (!ret) {
3489 			ret = write_one_cache_group(trans, root, path, cache);
3490 			/*
3491 			 * Our block group might still be attached to the list
3492 			 * of new block groups in the transaction handle of some
3493 			 * other task (struct btrfs_trans_handle->new_bgs). This
3494 			 * means its block group item isn't yet in the extent
3495 			 * tree. If this happens ignore the error, as we will
3496 			 * try again later in the critical section of the
3497 			 * transaction commit.
3498 			 */
3499 			if (ret == -ENOENT) {
3500 				ret = 0;
3501 				spin_lock(&cur_trans->dirty_bgs_lock);
3502 				if (list_empty(&cache->dirty_list)) {
3503 					list_add_tail(&cache->dirty_list,
3504 						      &cur_trans->dirty_bgs);
3505 					btrfs_get_block_group(cache);
3506 				}
3507 				spin_unlock(&cur_trans->dirty_bgs_lock);
3508 			} else if (ret) {
3509 				btrfs_abort_transaction(trans, root, ret);
3510 			}
3511 		}
3512 
3513 		/* if its not on the io list, we need to put the block group */
3514 		if (should_put)
3515 			btrfs_put_block_group(cache);
3516 
3517 		if (ret)
3518 			break;
3519 
3520 		/*
3521 		 * Avoid blocking other tasks for too long. It might even save
3522 		 * us from writing caches for block groups that are going to be
3523 		 * removed.
3524 		 */
3525 		mutex_unlock(&trans->transaction->cache_write_mutex);
3526 		mutex_lock(&trans->transaction->cache_write_mutex);
3527 	}
3528 	mutex_unlock(&trans->transaction->cache_write_mutex);
3529 
3530 	/*
3531 	 * go through delayed refs for all the stuff we've just kicked off
3532 	 * and then loop back (just once)
3533 	 */
3534 	ret = btrfs_run_delayed_refs(trans, root, 0);
3535 	if (!ret && loops == 0) {
3536 		loops++;
3537 		spin_lock(&cur_trans->dirty_bgs_lock);
3538 		list_splice_init(&cur_trans->dirty_bgs, &dirty);
3539 		/*
3540 		 * dirty_bgs_lock protects us from concurrent block group
3541 		 * deletes too (not just cache_write_mutex).
3542 		 */
3543 		if (!list_empty(&dirty)) {
3544 			spin_unlock(&cur_trans->dirty_bgs_lock);
3545 			goto again;
3546 		}
3547 		spin_unlock(&cur_trans->dirty_bgs_lock);
3548 	}
3549 
3550 	btrfs_free_path(path);
3551 	return ret;
3552 }
3553 
btrfs_write_dirty_block_groups(struct btrfs_trans_handle * trans,struct btrfs_root * root)3554 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3555 				   struct btrfs_root *root)
3556 {
3557 	struct btrfs_block_group_cache *cache;
3558 	struct btrfs_transaction *cur_trans = trans->transaction;
3559 	int ret = 0;
3560 	int should_put;
3561 	struct btrfs_path *path;
3562 	struct list_head *io = &cur_trans->io_bgs;
3563 	int num_started = 0;
3564 
3565 	path = btrfs_alloc_path();
3566 	if (!path)
3567 		return -ENOMEM;
3568 
3569 	/*
3570 	 * We don't need the lock here since we are protected by the transaction
3571 	 * commit.  We want to do the cache_save_setup first and then run the
3572 	 * delayed refs to make sure we have the best chance at doing this all
3573 	 * in one shot.
3574 	 */
3575 	while (!list_empty(&cur_trans->dirty_bgs)) {
3576 		cache = list_first_entry(&cur_trans->dirty_bgs,
3577 					 struct btrfs_block_group_cache,
3578 					 dirty_list);
3579 
3580 		/*
3581 		 * this can happen if cache_save_setup re-dirties a block
3582 		 * group that is already under IO.  Just wait for it to
3583 		 * finish and then do it all again
3584 		 */
3585 		if (!list_empty(&cache->io_list)) {
3586 			list_del_init(&cache->io_list);
3587 			btrfs_wait_cache_io(root, trans, cache,
3588 					    &cache->io_ctl, path,
3589 					    cache->key.objectid);
3590 			btrfs_put_block_group(cache);
3591 		}
3592 
3593 		/*
3594 		 * don't remove from the dirty list until after we've waited
3595 		 * on any pending IO
3596 		 */
3597 		list_del_init(&cache->dirty_list);
3598 		should_put = 1;
3599 
3600 		cache_save_setup(cache, trans, path);
3601 
3602 		if (!ret)
3603 			ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3604 
3605 		if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3606 			cache->io_ctl.inode = NULL;
3607 			ret = btrfs_write_out_cache(root, trans, cache, path);
3608 			if (ret == 0 && cache->io_ctl.inode) {
3609 				num_started++;
3610 				should_put = 0;
3611 				list_add_tail(&cache->io_list, io);
3612 			} else {
3613 				/*
3614 				 * if we failed to write the cache, the
3615 				 * generation will be bad and life goes on
3616 				 */
3617 				ret = 0;
3618 			}
3619 		}
3620 		if (!ret) {
3621 			ret = write_one_cache_group(trans, root, path, cache);
3622 			if (ret)
3623 				btrfs_abort_transaction(trans, root, ret);
3624 		}
3625 
3626 		/* if its not on the io list, we need to put the block group */
3627 		if (should_put)
3628 			btrfs_put_block_group(cache);
3629 	}
3630 
3631 	while (!list_empty(io)) {
3632 		cache = list_first_entry(io, struct btrfs_block_group_cache,
3633 					 io_list);
3634 		list_del_init(&cache->io_list);
3635 		btrfs_wait_cache_io(root, trans, cache,
3636 				    &cache->io_ctl, path, cache->key.objectid);
3637 		btrfs_put_block_group(cache);
3638 	}
3639 
3640 	btrfs_free_path(path);
3641 	return ret;
3642 }
3643 
btrfs_extent_readonly(struct btrfs_root * root,u64 bytenr)3644 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3645 {
3646 	struct btrfs_block_group_cache *block_group;
3647 	int readonly = 0;
3648 
3649 	block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3650 	if (!block_group || block_group->ro)
3651 		readonly = 1;
3652 	if (block_group)
3653 		btrfs_put_block_group(block_group);
3654 	return readonly;
3655 }
3656 
alloc_name(u64 flags)3657 static const char *alloc_name(u64 flags)
3658 {
3659 	switch (flags) {
3660 	case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3661 		return "mixed";
3662 	case BTRFS_BLOCK_GROUP_METADATA:
3663 		return "metadata";
3664 	case BTRFS_BLOCK_GROUP_DATA:
3665 		return "data";
3666 	case BTRFS_BLOCK_GROUP_SYSTEM:
3667 		return "system";
3668 	default:
3669 		WARN_ON(1);
3670 		return "invalid-combination";
3671 	};
3672 }
3673 
update_space_info(struct btrfs_fs_info * info,u64 flags,u64 total_bytes,u64 bytes_used,struct btrfs_space_info ** space_info)3674 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3675 			     u64 total_bytes, u64 bytes_used,
3676 			     struct btrfs_space_info **space_info)
3677 {
3678 	struct btrfs_space_info *found;
3679 	int i;
3680 	int factor;
3681 	int ret;
3682 
3683 	if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3684 		     BTRFS_BLOCK_GROUP_RAID10))
3685 		factor = 2;
3686 	else
3687 		factor = 1;
3688 
3689 	found = __find_space_info(info, flags);
3690 	if (found) {
3691 		spin_lock(&found->lock);
3692 		found->total_bytes += total_bytes;
3693 		found->disk_total += total_bytes * factor;
3694 		found->bytes_used += bytes_used;
3695 		found->disk_used += bytes_used * factor;
3696 		found->full = 0;
3697 		spin_unlock(&found->lock);
3698 		*space_info = found;
3699 		return 0;
3700 	}
3701 	found = kzalloc(sizeof(*found), GFP_NOFS);
3702 	if (!found)
3703 		return -ENOMEM;
3704 
3705 	ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3706 	if (ret) {
3707 		kfree(found);
3708 		return ret;
3709 	}
3710 
3711 	for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3712 		INIT_LIST_HEAD(&found->block_groups[i]);
3713 	init_rwsem(&found->groups_sem);
3714 	spin_lock_init(&found->lock);
3715 	found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3716 	found->total_bytes = total_bytes;
3717 	found->disk_total = total_bytes * factor;
3718 	found->bytes_used = bytes_used;
3719 	found->disk_used = bytes_used * factor;
3720 	found->bytes_pinned = 0;
3721 	found->bytes_reserved = 0;
3722 	found->bytes_readonly = 0;
3723 	found->bytes_may_use = 0;
3724 	found->full = 0;
3725 	found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3726 	found->chunk_alloc = 0;
3727 	found->flush = 0;
3728 	init_waitqueue_head(&found->wait);
3729 	INIT_LIST_HEAD(&found->ro_bgs);
3730 
3731 	ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3732 				    info->space_info_kobj, "%s",
3733 				    alloc_name(found->flags));
3734 	if (ret) {
3735 		kfree(found);
3736 		return ret;
3737 	}
3738 
3739 	*space_info = found;
3740 	list_add_rcu(&found->list, &info->space_info);
3741 	if (flags & BTRFS_BLOCK_GROUP_DATA)
3742 		info->data_sinfo = found;
3743 
3744 	return ret;
3745 }
3746 
set_avail_alloc_bits(struct btrfs_fs_info * fs_info,u64 flags)3747 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3748 {
3749 	u64 extra_flags = chunk_to_extended(flags) &
3750 				BTRFS_EXTENDED_PROFILE_MASK;
3751 
3752 	write_seqlock(&fs_info->profiles_lock);
3753 	if (flags & BTRFS_BLOCK_GROUP_DATA)
3754 		fs_info->avail_data_alloc_bits |= extra_flags;
3755 	if (flags & BTRFS_BLOCK_GROUP_METADATA)
3756 		fs_info->avail_metadata_alloc_bits |= extra_flags;
3757 	if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3758 		fs_info->avail_system_alloc_bits |= extra_flags;
3759 	write_sequnlock(&fs_info->profiles_lock);
3760 }
3761 
3762 /*
3763  * returns target flags in extended format or 0 if restripe for this
3764  * chunk_type is not in progress
3765  *
3766  * should be called with either volume_mutex or balance_lock held
3767  */
get_restripe_target(struct btrfs_fs_info * fs_info,u64 flags)3768 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3769 {
3770 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3771 	u64 target = 0;
3772 
3773 	if (!bctl)
3774 		return 0;
3775 
3776 	if (flags & BTRFS_BLOCK_GROUP_DATA &&
3777 	    bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3778 		target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3779 	} else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3780 		   bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3781 		target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3782 	} else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3783 		   bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3784 		target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3785 	}
3786 
3787 	return target;
3788 }
3789 
3790 /*
3791  * @flags: available profiles in extended format (see ctree.h)
3792  *
3793  * Returns reduced profile in chunk format.  If profile changing is in
3794  * progress (either running or paused) picks the target profile (if it's
3795  * already available), otherwise falls back to plain reducing.
3796  */
btrfs_reduce_alloc_profile(struct btrfs_root * root,u64 flags)3797 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3798 {
3799 	u64 num_devices = root->fs_info->fs_devices->rw_devices;
3800 	u64 target;
3801 	u64 tmp;
3802 
3803 	/*
3804 	 * see if restripe for this chunk_type is in progress, if so
3805 	 * try to reduce to the target profile
3806 	 */
3807 	spin_lock(&root->fs_info->balance_lock);
3808 	target = get_restripe_target(root->fs_info, flags);
3809 	if (target) {
3810 		/* pick target profile only if it's already available */
3811 		if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3812 			spin_unlock(&root->fs_info->balance_lock);
3813 			return extended_to_chunk(target);
3814 		}
3815 	}
3816 	spin_unlock(&root->fs_info->balance_lock);
3817 
3818 	/* First, mask out the RAID levels which aren't possible */
3819 	if (num_devices == 1)
3820 		flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3821 			   BTRFS_BLOCK_GROUP_RAID5);
3822 	if (num_devices < 3)
3823 		flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3824 	if (num_devices < 4)
3825 		flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3826 
3827 	tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3828 		       BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3829 		       BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3830 	flags &= ~tmp;
3831 
3832 	if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3833 		tmp = BTRFS_BLOCK_GROUP_RAID6;
3834 	else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3835 		tmp = BTRFS_BLOCK_GROUP_RAID5;
3836 	else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3837 		tmp = BTRFS_BLOCK_GROUP_RAID10;
3838 	else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3839 		tmp = BTRFS_BLOCK_GROUP_RAID1;
3840 	else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3841 		tmp = BTRFS_BLOCK_GROUP_RAID0;
3842 
3843 	return extended_to_chunk(flags | tmp);
3844 }
3845 
get_alloc_profile(struct btrfs_root * root,u64 orig_flags)3846 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3847 {
3848 	unsigned seq;
3849 	u64 flags;
3850 
3851 	do {
3852 		flags = orig_flags;
3853 		seq = read_seqbegin(&root->fs_info->profiles_lock);
3854 
3855 		if (flags & BTRFS_BLOCK_GROUP_DATA)
3856 			flags |= root->fs_info->avail_data_alloc_bits;
3857 		else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3858 			flags |= root->fs_info->avail_system_alloc_bits;
3859 		else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3860 			flags |= root->fs_info->avail_metadata_alloc_bits;
3861 	} while (read_seqretry(&root->fs_info->profiles_lock, seq));
3862 
3863 	return btrfs_reduce_alloc_profile(root, flags);
3864 }
3865 
btrfs_get_alloc_profile(struct btrfs_root * root,int data)3866 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3867 {
3868 	u64 flags;
3869 	u64 ret;
3870 
3871 	if (data)
3872 		flags = BTRFS_BLOCK_GROUP_DATA;
3873 	else if (root == root->fs_info->chunk_root)
3874 		flags = BTRFS_BLOCK_GROUP_SYSTEM;
3875 	else
3876 		flags = BTRFS_BLOCK_GROUP_METADATA;
3877 
3878 	ret = get_alloc_profile(root, flags);
3879 	return ret;
3880 }
3881 
3882 /*
3883  * This will check the space that the inode allocates from to make sure we have
3884  * enough space for bytes.
3885  */
btrfs_check_data_free_space(struct inode * inode,u64 bytes,u64 write_bytes)3886 int btrfs_check_data_free_space(struct inode *inode, u64 bytes, u64 write_bytes)
3887 {
3888 	struct btrfs_space_info *data_sinfo;
3889 	struct btrfs_root *root = BTRFS_I(inode)->root;
3890 	struct btrfs_fs_info *fs_info = root->fs_info;
3891 	u64 used;
3892 	int ret = 0;
3893 	int need_commit = 2;
3894 	int have_pinned_space;
3895 
3896 	/* make sure bytes are sectorsize aligned */
3897 	bytes = ALIGN(bytes, root->sectorsize);
3898 
3899 	if (btrfs_is_free_space_inode(inode)) {
3900 		need_commit = 0;
3901 		ASSERT(current->journal_info);
3902 	}
3903 
3904 	data_sinfo = fs_info->data_sinfo;
3905 	if (!data_sinfo)
3906 		goto alloc;
3907 
3908 again:
3909 	/* make sure we have enough space to handle the data first */
3910 	spin_lock(&data_sinfo->lock);
3911 	used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3912 		data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3913 		data_sinfo->bytes_may_use;
3914 
3915 	if (used + bytes > data_sinfo->total_bytes) {
3916 		struct btrfs_trans_handle *trans;
3917 
3918 		/*
3919 		 * if we don't have enough free bytes in this space then we need
3920 		 * to alloc a new chunk.
3921 		 */
3922 		if (!data_sinfo->full) {
3923 			u64 alloc_target;
3924 
3925 			data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3926 			spin_unlock(&data_sinfo->lock);
3927 alloc:
3928 			alloc_target = btrfs_get_alloc_profile(root, 1);
3929 			/*
3930 			 * It is ugly that we don't call nolock join
3931 			 * transaction for the free space inode case here.
3932 			 * But it is safe because we only do the data space
3933 			 * reservation for the free space cache in the
3934 			 * transaction context, the common join transaction
3935 			 * just increase the counter of the current transaction
3936 			 * handler, doesn't try to acquire the trans_lock of
3937 			 * the fs.
3938 			 */
3939 			trans = btrfs_join_transaction(root);
3940 			if (IS_ERR(trans))
3941 				return PTR_ERR(trans);
3942 
3943 			ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3944 					     alloc_target,
3945 					     CHUNK_ALLOC_NO_FORCE);
3946 			btrfs_end_transaction(trans, root);
3947 			if (ret < 0) {
3948 				if (ret != -ENOSPC)
3949 					return ret;
3950 				else {
3951 					have_pinned_space = 1;
3952 					goto commit_trans;
3953 				}
3954 			}
3955 
3956 			if (!data_sinfo)
3957 				data_sinfo = fs_info->data_sinfo;
3958 
3959 			goto again;
3960 		}
3961 
3962 		/*
3963 		 * If we don't have enough pinned space to deal with this
3964 		 * allocation, and no removed chunk in current transaction,
3965 		 * don't bother committing the transaction.
3966 		 */
3967 		have_pinned_space = percpu_counter_compare(
3968 			&data_sinfo->total_bytes_pinned,
3969 			used + bytes - data_sinfo->total_bytes);
3970 		spin_unlock(&data_sinfo->lock);
3971 
3972 		/* commit the current transaction and try again */
3973 commit_trans:
3974 		if (need_commit &&
3975 		    !atomic_read(&root->fs_info->open_ioctl_trans)) {
3976 			need_commit--;
3977 
3978 			if (need_commit > 0) {
3979 				btrfs_start_delalloc_roots(fs_info, 0, -1);
3980 				btrfs_wait_ordered_roots(fs_info, -1);
3981 			}
3982 
3983 			trans = btrfs_join_transaction(root);
3984 			if (IS_ERR(trans))
3985 				return PTR_ERR(trans);
3986 			if (have_pinned_space >= 0 ||
3987 			    trans->transaction->have_free_bgs ||
3988 			    need_commit > 0) {
3989 				ret = btrfs_commit_transaction(trans, root);
3990 				if (ret)
3991 					return ret;
3992 				/*
3993 				 * The cleaner kthread might still be doing iput
3994 				 * operations. Wait for it to finish so that
3995 				 * more space is released.
3996 				 */
3997 				mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
3998 				mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
3999 				goto again;
4000 			} else {
4001 				btrfs_end_transaction(trans, root);
4002 			}
4003 		}
4004 
4005 		trace_btrfs_space_reservation(root->fs_info,
4006 					      "space_info:enospc",
4007 					      data_sinfo->flags, bytes, 1);
4008 		return -ENOSPC;
4009 	}
4010 	ret = btrfs_qgroup_reserve(root, write_bytes);
4011 	if (ret)
4012 		goto out;
4013 	data_sinfo->bytes_may_use += bytes;
4014 	trace_btrfs_space_reservation(root->fs_info, "space_info",
4015 				      data_sinfo->flags, bytes, 1);
4016 out:
4017 	spin_unlock(&data_sinfo->lock);
4018 
4019 	return ret;
4020 }
4021 
4022 /*
4023  * Called if we need to clear a data reservation for this inode.
4024  */
btrfs_free_reserved_data_space(struct inode * inode,u64 bytes)4025 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
4026 {
4027 	struct btrfs_root *root = BTRFS_I(inode)->root;
4028 	struct btrfs_space_info *data_sinfo;
4029 
4030 	/* make sure bytes are sectorsize aligned */
4031 	bytes = ALIGN(bytes, root->sectorsize);
4032 
4033 	data_sinfo = root->fs_info->data_sinfo;
4034 	spin_lock(&data_sinfo->lock);
4035 	WARN_ON(data_sinfo->bytes_may_use < bytes);
4036 	data_sinfo->bytes_may_use -= bytes;
4037 	trace_btrfs_space_reservation(root->fs_info, "space_info",
4038 				      data_sinfo->flags, bytes, 0);
4039 	spin_unlock(&data_sinfo->lock);
4040 }
4041 
force_metadata_allocation(struct btrfs_fs_info * info)4042 static void force_metadata_allocation(struct btrfs_fs_info *info)
4043 {
4044 	struct list_head *head = &info->space_info;
4045 	struct btrfs_space_info *found;
4046 
4047 	rcu_read_lock();
4048 	list_for_each_entry_rcu(found, head, list) {
4049 		if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4050 			found->force_alloc = CHUNK_ALLOC_FORCE;
4051 	}
4052 	rcu_read_unlock();
4053 }
4054 
calc_global_rsv_need_space(struct btrfs_block_rsv * global)4055 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4056 {
4057 	return (global->size << 1);
4058 }
4059 
should_alloc_chunk(struct btrfs_root * root,struct btrfs_space_info * sinfo,int force)4060 static int should_alloc_chunk(struct btrfs_root *root,
4061 			      struct btrfs_space_info *sinfo, int force)
4062 {
4063 	struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4064 	u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4065 	u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4066 	u64 thresh;
4067 
4068 	if (force == CHUNK_ALLOC_FORCE)
4069 		return 1;
4070 
4071 	/*
4072 	 * We need to take into account the global rsv because for all intents
4073 	 * and purposes it's used space.  Don't worry about locking the
4074 	 * global_rsv, it doesn't change except when the transaction commits.
4075 	 */
4076 	if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4077 		num_allocated += calc_global_rsv_need_space(global_rsv);
4078 
4079 	/*
4080 	 * in limited mode, we want to have some free space up to
4081 	 * about 1% of the FS size.
4082 	 */
4083 	if (force == CHUNK_ALLOC_LIMITED) {
4084 		thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4085 		thresh = max_t(u64, 64 * 1024 * 1024,
4086 			       div_factor_fine(thresh, 1));
4087 
4088 		if (num_bytes - num_allocated < thresh)
4089 			return 1;
4090 	}
4091 
4092 	if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4093 		return 0;
4094 	return 1;
4095 }
4096 
get_system_chunk_thresh(struct btrfs_root * root,u64 type)4097 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
4098 {
4099 	u64 num_dev;
4100 
4101 	if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4102 		    BTRFS_BLOCK_GROUP_RAID0 |
4103 		    BTRFS_BLOCK_GROUP_RAID5 |
4104 		    BTRFS_BLOCK_GROUP_RAID6))
4105 		num_dev = root->fs_info->fs_devices->rw_devices;
4106 	else if (type & BTRFS_BLOCK_GROUP_RAID1)
4107 		num_dev = 2;
4108 	else
4109 		num_dev = 1;	/* DUP or single */
4110 
4111 	/* metadata for updaing devices and chunk tree */
4112 	return btrfs_calc_trans_metadata_size(root, num_dev + 1);
4113 }
4114 
check_system_chunk(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 type)4115 static void check_system_chunk(struct btrfs_trans_handle *trans,
4116 			       struct btrfs_root *root, u64 type)
4117 {
4118 	struct btrfs_space_info *info;
4119 	u64 left;
4120 	u64 thresh;
4121 
4122 	info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4123 	spin_lock(&info->lock);
4124 	left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4125 		info->bytes_reserved - info->bytes_readonly;
4126 	spin_unlock(&info->lock);
4127 
4128 	thresh = get_system_chunk_thresh(root, type);
4129 	if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4130 		btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4131 			left, thresh, type);
4132 		dump_space_info(info, 0, 0);
4133 	}
4134 
4135 	if (left < thresh) {
4136 		u64 flags;
4137 
4138 		flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4139 		btrfs_alloc_chunk(trans, root, flags);
4140 	}
4141 }
4142 
do_chunk_alloc(struct btrfs_trans_handle * trans,struct btrfs_root * extent_root,u64 flags,int force)4143 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4144 			  struct btrfs_root *extent_root, u64 flags, int force)
4145 {
4146 	struct btrfs_space_info *space_info;
4147 	struct btrfs_fs_info *fs_info = extent_root->fs_info;
4148 	int wait_for_alloc = 0;
4149 	int ret = 0;
4150 
4151 	/* Don't re-enter if we're already allocating a chunk */
4152 	if (trans->allocating_chunk)
4153 		return -ENOSPC;
4154 
4155 	space_info = __find_space_info(extent_root->fs_info, flags);
4156 	if (!space_info) {
4157 		ret = update_space_info(extent_root->fs_info, flags,
4158 					0, 0, &space_info);
4159 		BUG_ON(ret); /* -ENOMEM */
4160 	}
4161 	BUG_ON(!space_info); /* Logic error */
4162 
4163 again:
4164 	spin_lock(&space_info->lock);
4165 	if (force < space_info->force_alloc)
4166 		force = space_info->force_alloc;
4167 	if (space_info->full) {
4168 		if (should_alloc_chunk(extent_root, space_info, force))
4169 			ret = -ENOSPC;
4170 		else
4171 			ret = 0;
4172 		spin_unlock(&space_info->lock);
4173 		return ret;
4174 	}
4175 
4176 	if (!should_alloc_chunk(extent_root, space_info, force)) {
4177 		spin_unlock(&space_info->lock);
4178 		return 0;
4179 	} else if (space_info->chunk_alloc) {
4180 		wait_for_alloc = 1;
4181 	} else {
4182 		space_info->chunk_alloc = 1;
4183 	}
4184 
4185 	spin_unlock(&space_info->lock);
4186 
4187 	mutex_lock(&fs_info->chunk_mutex);
4188 
4189 	/*
4190 	 * The chunk_mutex is held throughout the entirety of a chunk
4191 	 * allocation, so once we've acquired the chunk_mutex we know that the
4192 	 * other guy is done and we need to recheck and see if we should
4193 	 * allocate.
4194 	 */
4195 	if (wait_for_alloc) {
4196 		mutex_unlock(&fs_info->chunk_mutex);
4197 		wait_for_alloc = 0;
4198 		goto again;
4199 	}
4200 
4201 	trans->allocating_chunk = true;
4202 
4203 	/*
4204 	 * If we have mixed data/metadata chunks we want to make sure we keep
4205 	 * allocating mixed chunks instead of individual chunks.
4206 	 */
4207 	if (btrfs_mixed_space_info(space_info))
4208 		flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4209 
4210 	/*
4211 	 * if we're doing a data chunk, go ahead and make sure that
4212 	 * we keep a reasonable number of metadata chunks allocated in the
4213 	 * FS as well.
4214 	 */
4215 	if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4216 		fs_info->data_chunk_allocations++;
4217 		if (!(fs_info->data_chunk_allocations %
4218 		      fs_info->metadata_ratio))
4219 			force_metadata_allocation(fs_info);
4220 	}
4221 
4222 	/*
4223 	 * Check if we have enough space in SYSTEM chunk because we may need
4224 	 * to update devices.
4225 	 */
4226 	check_system_chunk(trans, extent_root, flags);
4227 
4228 	ret = btrfs_alloc_chunk(trans, extent_root, flags);
4229 	trans->allocating_chunk = false;
4230 
4231 	spin_lock(&space_info->lock);
4232 	if (ret < 0 && ret != -ENOSPC)
4233 		goto out;
4234 	if (ret)
4235 		space_info->full = 1;
4236 	else
4237 		ret = 1;
4238 
4239 	space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4240 out:
4241 	space_info->chunk_alloc = 0;
4242 	spin_unlock(&space_info->lock);
4243 	mutex_unlock(&fs_info->chunk_mutex);
4244 	return ret;
4245 }
4246 
can_overcommit(struct btrfs_root * root,struct btrfs_space_info * space_info,u64 bytes,enum btrfs_reserve_flush_enum flush)4247 static int can_overcommit(struct btrfs_root *root,
4248 			  struct btrfs_space_info *space_info, u64 bytes,
4249 			  enum btrfs_reserve_flush_enum flush)
4250 {
4251 	struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4252 	u64 profile = btrfs_get_alloc_profile(root, 0);
4253 	u64 space_size;
4254 	u64 avail;
4255 	u64 used;
4256 
4257 	used = space_info->bytes_used + space_info->bytes_reserved +
4258 		space_info->bytes_pinned + space_info->bytes_readonly;
4259 
4260 	/*
4261 	 * We only want to allow over committing if we have lots of actual space
4262 	 * free, but if we don't have enough space to handle the global reserve
4263 	 * space then we could end up having a real enospc problem when trying
4264 	 * to allocate a chunk or some other such important allocation.
4265 	 */
4266 	spin_lock(&global_rsv->lock);
4267 	space_size = calc_global_rsv_need_space(global_rsv);
4268 	spin_unlock(&global_rsv->lock);
4269 	if (used + space_size >= space_info->total_bytes)
4270 		return 0;
4271 
4272 	used += space_info->bytes_may_use;
4273 
4274 	spin_lock(&root->fs_info->free_chunk_lock);
4275 	avail = root->fs_info->free_chunk_space;
4276 	spin_unlock(&root->fs_info->free_chunk_lock);
4277 
4278 	/*
4279 	 * If we have dup, raid1 or raid10 then only half of the free
4280 	 * space is actually useable.  For raid56, the space info used
4281 	 * doesn't include the parity drive, so we don't have to
4282 	 * change the math
4283 	 */
4284 	if (profile & (BTRFS_BLOCK_GROUP_DUP |
4285 		       BTRFS_BLOCK_GROUP_RAID1 |
4286 		       BTRFS_BLOCK_GROUP_RAID10))
4287 		avail >>= 1;
4288 
4289 	/*
4290 	 * If we aren't flushing all things, let us overcommit up to
4291 	 * 1/2th of the space. If we can flush, don't let us overcommit
4292 	 * too much, let it overcommit up to 1/8 of the space.
4293 	 */
4294 	if (flush == BTRFS_RESERVE_FLUSH_ALL)
4295 		avail >>= 3;
4296 	else
4297 		avail >>= 1;
4298 
4299 	if (used + bytes < space_info->total_bytes + avail)
4300 		return 1;
4301 	return 0;
4302 }
4303 
btrfs_writeback_inodes_sb_nr(struct btrfs_root * root,unsigned long nr_pages,int nr_items)4304 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4305 					 unsigned long nr_pages, int nr_items)
4306 {
4307 	struct super_block *sb = root->fs_info->sb;
4308 
4309 	if (down_read_trylock(&sb->s_umount)) {
4310 		writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4311 		up_read(&sb->s_umount);
4312 	} else {
4313 		/*
4314 		 * We needn't worry the filesystem going from r/w to r/o though
4315 		 * we don't acquire ->s_umount mutex, because the filesystem
4316 		 * should guarantee the delalloc inodes list be empty after
4317 		 * the filesystem is readonly(all dirty pages are written to
4318 		 * the disk).
4319 		 */
4320 		btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4321 		if (!current->journal_info)
4322 			btrfs_wait_ordered_roots(root->fs_info, nr_items);
4323 	}
4324 }
4325 
calc_reclaim_items_nr(struct btrfs_root * root,u64 to_reclaim)4326 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4327 {
4328 	u64 bytes;
4329 	int nr;
4330 
4331 	bytes = btrfs_calc_trans_metadata_size(root, 1);
4332 	nr = (int)div64_u64(to_reclaim, bytes);
4333 	if (!nr)
4334 		nr = 1;
4335 	return nr;
4336 }
4337 
4338 #define EXTENT_SIZE_PER_ITEM	(256 * 1024)
4339 
4340 /*
4341  * shrink metadata reservation for delalloc
4342  */
shrink_delalloc(struct btrfs_root * root,u64 to_reclaim,u64 orig,bool wait_ordered)4343 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4344 			    bool wait_ordered)
4345 {
4346 	struct btrfs_block_rsv *block_rsv;
4347 	struct btrfs_space_info *space_info;
4348 	struct btrfs_trans_handle *trans;
4349 	u64 delalloc_bytes;
4350 	u64 max_reclaim;
4351 	long time_left;
4352 	unsigned long nr_pages;
4353 	int loops;
4354 	int items;
4355 	enum btrfs_reserve_flush_enum flush;
4356 
4357 	/* Calc the number of the pages we need flush for space reservation */
4358 	items = calc_reclaim_items_nr(root, to_reclaim);
4359 	to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4360 
4361 	trans = (struct btrfs_trans_handle *)current->journal_info;
4362 	block_rsv = &root->fs_info->delalloc_block_rsv;
4363 	space_info = block_rsv->space_info;
4364 
4365 	delalloc_bytes = percpu_counter_sum_positive(
4366 						&root->fs_info->delalloc_bytes);
4367 	if (delalloc_bytes == 0) {
4368 		if (trans)
4369 			return;
4370 		if (wait_ordered)
4371 			btrfs_wait_ordered_roots(root->fs_info, items);
4372 		return;
4373 	}
4374 
4375 	loops = 0;
4376 	while (delalloc_bytes && loops < 3) {
4377 		max_reclaim = min(delalloc_bytes, to_reclaim);
4378 		nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4379 		btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4380 		/*
4381 		 * We need to wait for the async pages to actually start before
4382 		 * we do anything.
4383 		 */
4384 		max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4385 		if (!max_reclaim)
4386 			goto skip_async;
4387 
4388 		if (max_reclaim <= nr_pages)
4389 			max_reclaim = 0;
4390 		else
4391 			max_reclaim -= nr_pages;
4392 
4393 		wait_event(root->fs_info->async_submit_wait,
4394 			   atomic_read(&root->fs_info->async_delalloc_pages) <=
4395 			   (int)max_reclaim);
4396 skip_async:
4397 		if (!trans)
4398 			flush = BTRFS_RESERVE_FLUSH_ALL;
4399 		else
4400 			flush = BTRFS_RESERVE_NO_FLUSH;
4401 		spin_lock(&space_info->lock);
4402 		if (can_overcommit(root, space_info, orig, flush)) {
4403 			spin_unlock(&space_info->lock);
4404 			break;
4405 		}
4406 		spin_unlock(&space_info->lock);
4407 
4408 		loops++;
4409 		if (wait_ordered && !trans) {
4410 			btrfs_wait_ordered_roots(root->fs_info, items);
4411 		} else {
4412 			time_left = schedule_timeout_killable(1);
4413 			if (time_left)
4414 				break;
4415 		}
4416 		delalloc_bytes = percpu_counter_sum_positive(
4417 						&root->fs_info->delalloc_bytes);
4418 	}
4419 }
4420 
4421 /**
4422  * maybe_commit_transaction - possibly commit the transaction if its ok to
4423  * @root - the root we're allocating for
4424  * @bytes - the number of bytes we want to reserve
4425  * @force - force the commit
4426  *
4427  * This will check to make sure that committing the transaction will actually
4428  * get us somewhere and then commit the transaction if it does.  Otherwise it
4429  * will return -ENOSPC.
4430  */
may_commit_transaction(struct btrfs_root * root,struct btrfs_space_info * space_info,u64 bytes,int force)4431 static int may_commit_transaction(struct btrfs_root *root,
4432 				  struct btrfs_space_info *space_info,
4433 				  u64 bytes, int force)
4434 {
4435 	struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4436 	struct btrfs_trans_handle *trans;
4437 
4438 	trans = (struct btrfs_trans_handle *)current->journal_info;
4439 	if (trans)
4440 		return -EAGAIN;
4441 
4442 	if (force)
4443 		goto commit;
4444 
4445 	/* See if there is enough pinned space to make this reservation */
4446 	if (percpu_counter_compare(&space_info->total_bytes_pinned,
4447 				   bytes) >= 0)
4448 		goto commit;
4449 
4450 	/*
4451 	 * See if there is some space in the delayed insertion reservation for
4452 	 * this reservation.
4453 	 */
4454 	if (space_info != delayed_rsv->space_info)
4455 		return -ENOSPC;
4456 
4457 	spin_lock(&delayed_rsv->lock);
4458 	if (percpu_counter_compare(&space_info->total_bytes_pinned,
4459 				   bytes - delayed_rsv->size) >= 0) {
4460 		spin_unlock(&delayed_rsv->lock);
4461 		return -ENOSPC;
4462 	}
4463 	spin_unlock(&delayed_rsv->lock);
4464 
4465 commit:
4466 	trans = btrfs_join_transaction(root);
4467 	if (IS_ERR(trans))
4468 		return -ENOSPC;
4469 
4470 	return btrfs_commit_transaction(trans, root);
4471 }
4472 
4473 enum flush_state {
4474 	FLUSH_DELAYED_ITEMS_NR	=	1,
4475 	FLUSH_DELAYED_ITEMS	=	2,
4476 	FLUSH_DELALLOC		=	3,
4477 	FLUSH_DELALLOC_WAIT	=	4,
4478 	ALLOC_CHUNK		=	5,
4479 	COMMIT_TRANS		=	6,
4480 };
4481 
flush_space(struct btrfs_root * root,struct btrfs_space_info * space_info,u64 num_bytes,u64 orig_bytes,int state)4482 static int flush_space(struct btrfs_root *root,
4483 		       struct btrfs_space_info *space_info, u64 num_bytes,
4484 		       u64 orig_bytes, int state)
4485 {
4486 	struct btrfs_trans_handle *trans;
4487 	int nr;
4488 	int ret = 0;
4489 
4490 	switch (state) {
4491 	case FLUSH_DELAYED_ITEMS_NR:
4492 	case FLUSH_DELAYED_ITEMS:
4493 		if (state == FLUSH_DELAYED_ITEMS_NR)
4494 			nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4495 		else
4496 			nr = -1;
4497 
4498 		trans = btrfs_join_transaction(root);
4499 		if (IS_ERR(trans)) {
4500 			ret = PTR_ERR(trans);
4501 			break;
4502 		}
4503 		ret = btrfs_run_delayed_items_nr(trans, root, nr);
4504 		btrfs_end_transaction(trans, root);
4505 		break;
4506 	case FLUSH_DELALLOC:
4507 	case FLUSH_DELALLOC_WAIT:
4508 		shrink_delalloc(root, num_bytes * 2, orig_bytes,
4509 				state == FLUSH_DELALLOC_WAIT);
4510 		break;
4511 	case ALLOC_CHUNK:
4512 		trans = btrfs_join_transaction(root);
4513 		if (IS_ERR(trans)) {
4514 			ret = PTR_ERR(trans);
4515 			break;
4516 		}
4517 		ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4518 				     btrfs_get_alloc_profile(root, 0),
4519 				     CHUNK_ALLOC_NO_FORCE);
4520 		btrfs_end_transaction(trans, root);
4521 		if (ret == -ENOSPC)
4522 			ret = 0;
4523 		break;
4524 	case COMMIT_TRANS:
4525 		ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4526 		break;
4527 	default:
4528 		ret = -ENOSPC;
4529 		break;
4530 	}
4531 
4532 	return ret;
4533 }
4534 
4535 static inline u64
btrfs_calc_reclaim_metadata_size(struct btrfs_root * root,struct btrfs_space_info * space_info)4536 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4537 				 struct btrfs_space_info *space_info)
4538 {
4539 	u64 used;
4540 	u64 expected;
4541 	u64 to_reclaim;
4542 
4543 	to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4544 				16 * 1024 * 1024);
4545 	spin_lock(&space_info->lock);
4546 	if (can_overcommit(root, space_info, to_reclaim,
4547 			   BTRFS_RESERVE_FLUSH_ALL)) {
4548 		to_reclaim = 0;
4549 		goto out;
4550 	}
4551 
4552 	used = space_info->bytes_used + space_info->bytes_reserved +
4553 	       space_info->bytes_pinned + space_info->bytes_readonly +
4554 	       space_info->bytes_may_use;
4555 	if (can_overcommit(root, space_info, 1024 * 1024,
4556 			   BTRFS_RESERVE_FLUSH_ALL))
4557 		expected = div_factor_fine(space_info->total_bytes, 95);
4558 	else
4559 		expected = div_factor_fine(space_info->total_bytes, 90);
4560 
4561 	if (used > expected)
4562 		to_reclaim = used - expected;
4563 	else
4564 		to_reclaim = 0;
4565 	to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4566 				     space_info->bytes_reserved);
4567 out:
4568 	spin_unlock(&space_info->lock);
4569 
4570 	return to_reclaim;
4571 }
4572 
need_do_async_reclaim(struct btrfs_space_info * space_info,struct btrfs_fs_info * fs_info,u64 used)4573 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4574 					struct btrfs_fs_info *fs_info, u64 used)
4575 {
4576 	u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4577 
4578 	/* If we're just plain full then async reclaim just slows us down. */
4579 	if (space_info->bytes_used >= thresh)
4580 		return 0;
4581 
4582 	return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4583 		!test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4584 }
4585 
btrfs_need_do_async_reclaim(struct btrfs_space_info * space_info,struct btrfs_fs_info * fs_info,int flush_state)4586 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4587 				       struct btrfs_fs_info *fs_info,
4588 				       int flush_state)
4589 {
4590 	u64 used;
4591 
4592 	spin_lock(&space_info->lock);
4593 	/*
4594 	 * We run out of space and have not got any free space via flush_space,
4595 	 * so don't bother doing async reclaim.
4596 	 */
4597 	if (flush_state > COMMIT_TRANS && space_info->full) {
4598 		spin_unlock(&space_info->lock);
4599 		return 0;
4600 	}
4601 
4602 	used = space_info->bytes_used + space_info->bytes_reserved +
4603 	       space_info->bytes_pinned + space_info->bytes_readonly +
4604 	       space_info->bytes_may_use;
4605 	if (need_do_async_reclaim(space_info, fs_info, used)) {
4606 		spin_unlock(&space_info->lock);
4607 		return 1;
4608 	}
4609 	spin_unlock(&space_info->lock);
4610 
4611 	return 0;
4612 }
4613 
btrfs_async_reclaim_metadata_space(struct work_struct * work)4614 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4615 {
4616 	struct btrfs_fs_info *fs_info;
4617 	struct btrfs_space_info *space_info;
4618 	u64 to_reclaim;
4619 	int flush_state;
4620 
4621 	fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4622 	space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4623 
4624 	to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4625 						      space_info);
4626 	if (!to_reclaim)
4627 		return;
4628 
4629 	flush_state = FLUSH_DELAYED_ITEMS_NR;
4630 	do {
4631 		flush_space(fs_info->fs_root, space_info, to_reclaim,
4632 			    to_reclaim, flush_state);
4633 		flush_state++;
4634 		if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4635 						 flush_state))
4636 			return;
4637 	} while (flush_state < COMMIT_TRANS);
4638 }
4639 
btrfs_init_async_reclaim_work(struct work_struct * work)4640 void btrfs_init_async_reclaim_work(struct work_struct *work)
4641 {
4642 	INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4643 }
4644 
4645 /**
4646  * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4647  * @root - the root we're allocating for
4648  * @block_rsv - the block_rsv we're allocating for
4649  * @orig_bytes - the number of bytes we want
4650  * @flush - whether or not we can flush to make our reservation
4651  *
4652  * This will reserve orgi_bytes number of bytes from the space info associated
4653  * with the block_rsv.  If there is not enough space it will make an attempt to
4654  * flush out space to make room.  It will do this by flushing delalloc if
4655  * possible or committing the transaction.  If flush is 0 then no attempts to
4656  * regain reservations will be made and this will fail if there is not enough
4657  * space already.
4658  */
reserve_metadata_bytes(struct btrfs_root * root,struct btrfs_block_rsv * block_rsv,u64 orig_bytes,enum btrfs_reserve_flush_enum flush)4659 static int reserve_metadata_bytes(struct btrfs_root *root,
4660 				  struct btrfs_block_rsv *block_rsv,
4661 				  u64 orig_bytes,
4662 				  enum btrfs_reserve_flush_enum flush)
4663 {
4664 	struct btrfs_space_info *space_info = block_rsv->space_info;
4665 	u64 used;
4666 	u64 num_bytes = orig_bytes;
4667 	int flush_state = FLUSH_DELAYED_ITEMS_NR;
4668 	int ret = 0;
4669 	bool flushing = false;
4670 
4671 again:
4672 	ret = 0;
4673 	spin_lock(&space_info->lock);
4674 	/*
4675 	 * We only want to wait if somebody other than us is flushing and we
4676 	 * are actually allowed to flush all things.
4677 	 */
4678 	while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4679 	       space_info->flush) {
4680 		spin_unlock(&space_info->lock);
4681 		/*
4682 		 * If we have a trans handle we can't wait because the flusher
4683 		 * may have to commit the transaction, which would mean we would
4684 		 * deadlock since we are waiting for the flusher to finish, but
4685 		 * hold the current transaction open.
4686 		 */
4687 		if (current->journal_info)
4688 			return -EAGAIN;
4689 		ret = wait_event_killable(space_info->wait, !space_info->flush);
4690 		/* Must have been killed, return */
4691 		if (ret)
4692 			return -EINTR;
4693 
4694 		spin_lock(&space_info->lock);
4695 	}
4696 
4697 	ret = -ENOSPC;
4698 	used = space_info->bytes_used + space_info->bytes_reserved +
4699 		space_info->bytes_pinned + space_info->bytes_readonly +
4700 		space_info->bytes_may_use;
4701 
4702 	/*
4703 	 * The idea here is that we've not already over-reserved the block group
4704 	 * then we can go ahead and save our reservation first and then start
4705 	 * flushing if we need to.  Otherwise if we've already overcommitted
4706 	 * lets start flushing stuff first and then come back and try to make
4707 	 * our reservation.
4708 	 */
4709 	if (used <= space_info->total_bytes) {
4710 		if (used + orig_bytes <= space_info->total_bytes) {
4711 			space_info->bytes_may_use += orig_bytes;
4712 			trace_btrfs_space_reservation(root->fs_info,
4713 				"space_info", space_info->flags, orig_bytes, 1);
4714 			ret = 0;
4715 		} else {
4716 			/*
4717 			 * Ok set num_bytes to orig_bytes since we aren't
4718 			 * overocmmitted, this way we only try and reclaim what
4719 			 * we need.
4720 			 */
4721 			num_bytes = orig_bytes;
4722 		}
4723 	} else {
4724 		/*
4725 		 * Ok we're over committed, set num_bytes to the overcommitted
4726 		 * amount plus the amount of bytes that we need for this
4727 		 * reservation.
4728 		 */
4729 		num_bytes = used - space_info->total_bytes +
4730 			(orig_bytes * 2);
4731 	}
4732 
4733 	if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4734 		space_info->bytes_may_use += orig_bytes;
4735 		trace_btrfs_space_reservation(root->fs_info, "space_info",
4736 					      space_info->flags, orig_bytes,
4737 					      1);
4738 		ret = 0;
4739 	}
4740 
4741 	/*
4742 	 * Couldn't make our reservation, save our place so while we're trying
4743 	 * to reclaim space we can actually use it instead of somebody else
4744 	 * stealing it from us.
4745 	 *
4746 	 * We make the other tasks wait for the flush only when we can flush
4747 	 * all things.
4748 	 */
4749 	if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4750 		flushing = true;
4751 		space_info->flush = 1;
4752 	} else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4753 		used += orig_bytes;
4754 		/*
4755 		 * We will do the space reservation dance during log replay,
4756 		 * which means we won't have fs_info->fs_root set, so don't do
4757 		 * the async reclaim as we will panic.
4758 		 */
4759 		if (!root->fs_info->log_root_recovering &&
4760 		    need_do_async_reclaim(space_info, root->fs_info, used) &&
4761 		    !work_busy(&root->fs_info->async_reclaim_work))
4762 			queue_work(system_unbound_wq,
4763 				   &root->fs_info->async_reclaim_work);
4764 	}
4765 	spin_unlock(&space_info->lock);
4766 
4767 	if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4768 		goto out;
4769 
4770 	ret = flush_space(root, space_info, num_bytes, orig_bytes,
4771 			  flush_state);
4772 	flush_state++;
4773 
4774 	/*
4775 	 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4776 	 * would happen. So skip delalloc flush.
4777 	 */
4778 	if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4779 	    (flush_state == FLUSH_DELALLOC ||
4780 	     flush_state == FLUSH_DELALLOC_WAIT))
4781 		flush_state = ALLOC_CHUNK;
4782 
4783 	if (!ret)
4784 		goto again;
4785 	else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4786 		 flush_state < COMMIT_TRANS)
4787 		goto again;
4788 	else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4789 		 flush_state <= COMMIT_TRANS)
4790 		goto again;
4791 
4792 out:
4793 	if (ret == -ENOSPC &&
4794 	    unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4795 		struct btrfs_block_rsv *global_rsv =
4796 			&root->fs_info->global_block_rsv;
4797 
4798 		if (block_rsv != global_rsv &&
4799 		    !block_rsv_use_bytes(global_rsv, orig_bytes))
4800 			ret = 0;
4801 	}
4802 	if (ret == -ENOSPC)
4803 		trace_btrfs_space_reservation(root->fs_info,
4804 					      "space_info:enospc",
4805 					      space_info->flags, orig_bytes, 1);
4806 	if (flushing) {
4807 		spin_lock(&space_info->lock);
4808 		space_info->flush = 0;
4809 		wake_up_all(&space_info->wait);
4810 		spin_unlock(&space_info->lock);
4811 	}
4812 	return ret;
4813 }
4814 
get_block_rsv(const struct btrfs_trans_handle * trans,const struct btrfs_root * root)4815 static struct btrfs_block_rsv *get_block_rsv(
4816 					const struct btrfs_trans_handle *trans,
4817 					const struct btrfs_root *root)
4818 {
4819 	struct btrfs_block_rsv *block_rsv = NULL;
4820 
4821 	if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4822 		block_rsv = trans->block_rsv;
4823 
4824 	if (root == root->fs_info->csum_root && trans->adding_csums)
4825 		block_rsv = trans->block_rsv;
4826 
4827 	if (root == root->fs_info->uuid_root)
4828 		block_rsv = trans->block_rsv;
4829 
4830 	if (!block_rsv)
4831 		block_rsv = root->block_rsv;
4832 
4833 	if (!block_rsv)
4834 		block_rsv = &root->fs_info->empty_block_rsv;
4835 
4836 	return block_rsv;
4837 }
4838 
block_rsv_use_bytes(struct btrfs_block_rsv * block_rsv,u64 num_bytes)4839 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4840 			       u64 num_bytes)
4841 {
4842 	int ret = -ENOSPC;
4843 	spin_lock(&block_rsv->lock);
4844 	if (block_rsv->reserved >= num_bytes) {
4845 		block_rsv->reserved -= num_bytes;
4846 		if (block_rsv->reserved < block_rsv->size)
4847 			block_rsv->full = 0;
4848 		ret = 0;
4849 	}
4850 	spin_unlock(&block_rsv->lock);
4851 	return ret;
4852 }
4853 
block_rsv_add_bytes(struct btrfs_block_rsv * block_rsv,u64 num_bytes,int update_size)4854 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4855 				u64 num_bytes, int update_size)
4856 {
4857 	spin_lock(&block_rsv->lock);
4858 	block_rsv->reserved += num_bytes;
4859 	if (update_size)
4860 		block_rsv->size += num_bytes;
4861 	else if (block_rsv->reserved >= block_rsv->size)
4862 		block_rsv->full = 1;
4863 	spin_unlock(&block_rsv->lock);
4864 }
4865 
btrfs_cond_migrate_bytes(struct btrfs_fs_info * fs_info,struct btrfs_block_rsv * dest,u64 num_bytes,int min_factor)4866 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4867 			     struct btrfs_block_rsv *dest, u64 num_bytes,
4868 			     int min_factor)
4869 {
4870 	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4871 	u64 min_bytes;
4872 
4873 	if (global_rsv->space_info != dest->space_info)
4874 		return -ENOSPC;
4875 
4876 	spin_lock(&global_rsv->lock);
4877 	min_bytes = div_factor(global_rsv->size, min_factor);
4878 	if (global_rsv->reserved < min_bytes + num_bytes) {
4879 		spin_unlock(&global_rsv->lock);
4880 		return -ENOSPC;
4881 	}
4882 	global_rsv->reserved -= num_bytes;
4883 	if (global_rsv->reserved < global_rsv->size)
4884 		global_rsv->full = 0;
4885 	spin_unlock(&global_rsv->lock);
4886 
4887 	block_rsv_add_bytes(dest, num_bytes, 1);
4888 	return 0;
4889 }
4890 
block_rsv_release_bytes(struct btrfs_fs_info * fs_info,struct btrfs_block_rsv * block_rsv,struct btrfs_block_rsv * dest,u64 num_bytes)4891 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4892 				    struct btrfs_block_rsv *block_rsv,
4893 				    struct btrfs_block_rsv *dest, u64 num_bytes)
4894 {
4895 	struct btrfs_space_info *space_info = block_rsv->space_info;
4896 
4897 	spin_lock(&block_rsv->lock);
4898 	if (num_bytes == (u64)-1)
4899 		num_bytes = block_rsv->size;
4900 	block_rsv->size -= num_bytes;
4901 	if (block_rsv->reserved >= block_rsv->size) {
4902 		num_bytes = block_rsv->reserved - block_rsv->size;
4903 		block_rsv->reserved = block_rsv->size;
4904 		block_rsv->full = 1;
4905 	} else {
4906 		num_bytes = 0;
4907 	}
4908 	spin_unlock(&block_rsv->lock);
4909 
4910 	if (num_bytes > 0) {
4911 		if (dest) {
4912 			spin_lock(&dest->lock);
4913 			if (!dest->full) {
4914 				u64 bytes_to_add;
4915 
4916 				bytes_to_add = dest->size - dest->reserved;
4917 				bytes_to_add = min(num_bytes, bytes_to_add);
4918 				dest->reserved += bytes_to_add;
4919 				if (dest->reserved >= dest->size)
4920 					dest->full = 1;
4921 				num_bytes -= bytes_to_add;
4922 			}
4923 			spin_unlock(&dest->lock);
4924 		}
4925 		if (num_bytes) {
4926 			spin_lock(&space_info->lock);
4927 			space_info->bytes_may_use -= num_bytes;
4928 			trace_btrfs_space_reservation(fs_info, "space_info",
4929 					space_info->flags, num_bytes, 0);
4930 			spin_unlock(&space_info->lock);
4931 		}
4932 	}
4933 }
4934 
block_rsv_migrate_bytes(struct btrfs_block_rsv * src,struct btrfs_block_rsv * dst,u64 num_bytes)4935 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4936 				   struct btrfs_block_rsv *dst, u64 num_bytes)
4937 {
4938 	int ret;
4939 
4940 	ret = block_rsv_use_bytes(src, num_bytes);
4941 	if (ret)
4942 		return ret;
4943 
4944 	block_rsv_add_bytes(dst, num_bytes, 1);
4945 	return 0;
4946 }
4947 
btrfs_init_block_rsv(struct btrfs_block_rsv * rsv,unsigned short type)4948 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4949 {
4950 	memset(rsv, 0, sizeof(*rsv));
4951 	spin_lock_init(&rsv->lock);
4952 	rsv->type = type;
4953 }
4954 
btrfs_alloc_block_rsv(struct btrfs_root * root,unsigned short type)4955 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4956 					      unsigned short type)
4957 {
4958 	struct btrfs_block_rsv *block_rsv;
4959 	struct btrfs_fs_info *fs_info = root->fs_info;
4960 
4961 	block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4962 	if (!block_rsv)
4963 		return NULL;
4964 
4965 	btrfs_init_block_rsv(block_rsv, type);
4966 	block_rsv->space_info = __find_space_info(fs_info,
4967 						  BTRFS_BLOCK_GROUP_METADATA);
4968 	return block_rsv;
4969 }
4970 
btrfs_free_block_rsv(struct btrfs_root * root,struct btrfs_block_rsv * rsv)4971 void btrfs_free_block_rsv(struct btrfs_root *root,
4972 			  struct btrfs_block_rsv *rsv)
4973 {
4974 	if (!rsv)
4975 		return;
4976 	btrfs_block_rsv_release(root, rsv, (u64)-1);
4977 	kfree(rsv);
4978 }
4979 
__btrfs_free_block_rsv(struct btrfs_block_rsv * rsv)4980 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
4981 {
4982 	kfree(rsv);
4983 }
4984 
btrfs_block_rsv_add(struct btrfs_root * root,struct btrfs_block_rsv * block_rsv,u64 num_bytes,enum btrfs_reserve_flush_enum flush)4985 int btrfs_block_rsv_add(struct btrfs_root *root,
4986 			struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4987 			enum btrfs_reserve_flush_enum flush)
4988 {
4989 	int ret;
4990 
4991 	if (num_bytes == 0)
4992 		return 0;
4993 
4994 	ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4995 	if (!ret) {
4996 		block_rsv_add_bytes(block_rsv, num_bytes, 1);
4997 		return 0;
4998 	}
4999 
5000 	return ret;
5001 }
5002 
btrfs_block_rsv_check(struct btrfs_root * root,struct btrfs_block_rsv * block_rsv,int min_factor)5003 int btrfs_block_rsv_check(struct btrfs_root *root,
5004 			  struct btrfs_block_rsv *block_rsv, int min_factor)
5005 {
5006 	u64 num_bytes = 0;
5007 	int ret = -ENOSPC;
5008 
5009 	if (!block_rsv)
5010 		return 0;
5011 
5012 	spin_lock(&block_rsv->lock);
5013 	num_bytes = div_factor(block_rsv->size, min_factor);
5014 	if (block_rsv->reserved >= num_bytes)
5015 		ret = 0;
5016 	spin_unlock(&block_rsv->lock);
5017 
5018 	return ret;
5019 }
5020 
btrfs_block_rsv_refill(struct btrfs_root * root,struct btrfs_block_rsv * block_rsv,u64 min_reserved,enum btrfs_reserve_flush_enum flush)5021 int btrfs_block_rsv_refill(struct btrfs_root *root,
5022 			   struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5023 			   enum btrfs_reserve_flush_enum flush)
5024 {
5025 	u64 num_bytes = 0;
5026 	int ret = -ENOSPC;
5027 
5028 	if (!block_rsv)
5029 		return 0;
5030 
5031 	spin_lock(&block_rsv->lock);
5032 	num_bytes = min_reserved;
5033 	if (block_rsv->reserved >= num_bytes)
5034 		ret = 0;
5035 	else
5036 		num_bytes -= block_rsv->reserved;
5037 	spin_unlock(&block_rsv->lock);
5038 
5039 	if (!ret)
5040 		return 0;
5041 
5042 	ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5043 	if (!ret) {
5044 		block_rsv_add_bytes(block_rsv, num_bytes, 0);
5045 		return 0;
5046 	}
5047 
5048 	return ret;
5049 }
5050 
btrfs_block_rsv_migrate(struct btrfs_block_rsv * src_rsv,struct btrfs_block_rsv * dst_rsv,u64 num_bytes)5051 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5052 			    struct btrfs_block_rsv *dst_rsv,
5053 			    u64 num_bytes)
5054 {
5055 	return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5056 }
5057 
btrfs_block_rsv_release(struct btrfs_root * root,struct btrfs_block_rsv * block_rsv,u64 num_bytes)5058 void btrfs_block_rsv_release(struct btrfs_root *root,
5059 			     struct btrfs_block_rsv *block_rsv,
5060 			     u64 num_bytes)
5061 {
5062 	struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5063 	if (global_rsv == block_rsv ||
5064 	    block_rsv->space_info != global_rsv->space_info)
5065 		global_rsv = NULL;
5066 	block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5067 				num_bytes);
5068 }
5069 
5070 /*
5071  * helper to calculate size of global block reservation.
5072  * the desired value is sum of space used by extent tree,
5073  * checksum tree and root tree
5074  */
calc_global_metadata_size(struct btrfs_fs_info * fs_info)5075 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5076 {
5077 	struct btrfs_space_info *sinfo;
5078 	u64 num_bytes;
5079 	u64 meta_used;
5080 	u64 data_used;
5081 	int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5082 
5083 	sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5084 	spin_lock(&sinfo->lock);
5085 	data_used = sinfo->bytes_used;
5086 	spin_unlock(&sinfo->lock);
5087 
5088 	sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5089 	spin_lock(&sinfo->lock);
5090 	if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5091 		data_used = 0;
5092 	meta_used = sinfo->bytes_used;
5093 	spin_unlock(&sinfo->lock);
5094 
5095 	num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5096 		    csum_size * 2;
5097 	num_bytes += div_u64(data_used + meta_used, 50);
5098 
5099 	if (num_bytes * 3 > meta_used)
5100 		num_bytes = div_u64(meta_used, 3);
5101 
5102 	return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5103 }
5104 
update_global_block_rsv(struct btrfs_fs_info * fs_info)5105 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5106 {
5107 	struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5108 	struct btrfs_space_info *sinfo = block_rsv->space_info;
5109 	u64 num_bytes;
5110 
5111 	num_bytes = calc_global_metadata_size(fs_info);
5112 
5113 	spin_lock(&sinfo->lock);
5114 	spin_lock(&block_rsv->lock);
5115 
5116 	block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5117 
5118 	num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5119 		    sinfo->bytes_reserved + sinfo->bytes_readonly +
5120 		    sinfo->bytes_may_use;
5121 
5122 	if (sinfo->total_bytes > num_bytes) {
5123 		num_bytes = sinfo->total_bytes - num_bytes;
5124 		block_rsv->reserved += num_bytes;
5125 		sinfo->bytes_may_use += num_bytes;
5126 		trace_btrfs_space_reservation(fs_info, "space_info",
5127 				      sinfo->flags, num_bytes, 1);
5128 	}
5129 
5130 	if (block_rsv->reserved >= block_rsv->size) {
5131 		num_bytes = block_rsv->reserved - block_rsv->size;
5132 		sinfo->bytes_may_use -= num_bytes;
5133 		trace_btrfs_space_reservation(fs_info, "space_info",
5134 				      sinfo->flags, num_bytes, 0);
5135 		block_rsv->reserved = block_rsv->size;
5136 		block_rsv->full = 1;
5137 	}
5138 
5139 	spin_unlock(&block_rsv->lock);
5140 	spin_unlock(&sinfo->lock);
5141 }
5142 
init_global_block_rsv(struct btrfs_fs_info * fs_info)5143 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5144 {
5145 	struct btrfs_space_info *space_info;
5146 
5147 	space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5148 	fs_info->chunk_block_rsv.space_info = space_info;
5149 
5150 	space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5151 	fs_info->global_block_rsv.space_info = space_info;
5152 	fs_info->delalloc_block_rsv.space_info = space_info;
5153 	fs_info->trans_block_rsv.space_info = space_info;
5154 	fs_info->empty_block_rsv.space_info = space_info;
5155 	fs_info->delayed_block_rsv.space_info = space_info;
5156 
5157 	fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5158 	fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5159 	fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5160 	fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5161 	if (fs_info->quota_root)
5162 		fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5163 	fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5164 
5165 	update_global_block_rsv(fs_info);
5166 }
5167 
release_global_block_rsv(struct btrfs_fs_info * fs_info)5168 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5169 {
5170 	block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5171 				(u64)-1);
5172 	WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5173 	WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5174 	WARN_ON(fs_info->trans_block_rsv.size > 0);
5175 	WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5176 	WARN_ON(fs_info->chunk_block_rsv.size > 0);
5177 	WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5178 	WARN_ON(fs_info->delayed_block_rsv.size > 0);
5179 	WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5180 }
5181 
btrfs_trans_release_metadata(struct btrfs_trans_handle * trans,struct btrfs_root * root)5182 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5183 				  struct btrfs_root *root)
5184 {
5185 	if (!trans->block_rsv)
5186 		return;
5187 
5188 	if (!trans->bytes_reserved)
5189 		return;
5190 
5191 	trace_btrfs_space_reservation(root->fs_info, "transaction",
5192 				      trans->transid, trans->bytes_reserved, 0);
5193 	btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5194 	trans->bytes_reserved = 0;
5195 }
5196 
5197 /* Can only return 0 or -ENOSPC */
btrfs_orphan_reserve_metadata(struct btrfs_trans_handle * trans,struct inode * inode)5198 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5199 				  struct inode *inode)
5200 {
5201 	struct btrfs_root *root = BTRFS_I(inode)->root;
5202 	struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5203 	struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5204 
5205 	/*
5206 	 * We need to hold space in order to delete our orphan item once we've
5207 	 * added it, so this takes the reservation so we can release it later
5208 	 * when we are truly done with the orphan item.
5209 	 */
5210 	u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5211 	trace_btrfs_space_reservation(root->fs_info, "orphan",
5212 				      btrfs_ino(inode), num_bytes, 1);
5213 	return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5214 }
5215 
btrfs_orphan_release_metadata(struct inode * inode)5216 void btrfs_orphan_release_metadata(struct inode *inode)
5217 {
5218 	struct btrfs_root *root = BTRFS_I(inode)->root;
5219 	u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5220 	trace_btrfs_space_reservation(root->fs_info, "orphan",
5221 				      btrfs_ino(inode), num_bytes, 0);
5222 	btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5223 }
5224 
5225 /*
5226  * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5227  * root: the root of the parent directory
5228  * rsv: block reservation
5229  * items: the number of items that we need do reservation
5230  * qgroup_reserved: used to return the reserved size in qgroup
5231  *
5232  * This function is used to reserve the space for snapshot/subvolume
5233  * creation and deletion. Those operations are different with the
5234  * common file/directory operations, they change two fs/file trees
5235  * and root tree, the number of items that the qgroup reserves is
5236  * different with the free space reservation. So we can not use
5237  * the space reseravtion mechanism in start_transaction().
5238  */
btrfs_subvolume_reserve_metadata(struct btrfs_root * root,struct btrfs_block_rsv * rsv,int items,u64 * qgroup_reserved,bool use_global_rsv)5239 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5240 				     struct btrfs_block_rsv *rsv,
5241 				     int items,
5242 				     u64 *qgroup_reserved,
5243 				     bool use_global_rsv)
5244 {
5245 	u64 num_bytes;
5246 	int ret;
5247 	struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5248 
5249 	if (root->fs_info->quota_enabled) {
5250 		/* One for parent inode, two for dir entries */
5251 		num_bytes = 3 * root->nodesize;
5252 		ret = btrfs_qgroup_reserve(root, num_bytes);
5253 		if (ret)
5254 			return ret;
5255 	} else {
5256 		num_bytes = 0;
5257 	}
5258 
5259 	*qgroup_reserved = num_bytes;
5260 
5261 	num_bytes = btrfs_calc_trans_metadata_size(root, items);
5262 	rsv->space_info = __find_space_info(root->fs_info,
5263 					    BTRFS_BLOCK_GROUP_METADATA);
5264 	ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5265 				  BTRFS_RESERVE_FLUSH_ALL);
5266 
5267 	if (ret == -ENOSPC && use_global_rsv)
5268 		ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5269 
5270 	if (ret) {
5271 		if (*qgroup_reserved)
5272 			btrfs_qgroup_free(root, *qgroup_reserved);
5273 	}
5274 
5275 	return ret;
5276 }
5277 
btrfs_subvolume_release_metadata(struct btrfs_root * root,struct btrfs_block_rsv * rsv,u64 qgroup_reserved)5278 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5279 				      struct btrfs_block_rsv *rsv,
5280 				      u64 qgroup_reserved)
5281 {
5282 	btrfs_block_rsv_release(root, rsv, (u64)-1);
5283 }
5284 
5285 /**
5286  * drop_outstanding_extent - drop an outstanding extent
5287  * @inode: the inode we're dropping the extent for
5288  * @num_bytes: the number of bytes we're relaseing.
5289  *
5290  * This is called when we are freeing up an outstanding extent, either called
5291  * after an error or after an extent is written.  This will return the number of
5292  * reserved extents that need to be freed.  This must be called with
5293  * BTRFS_I(inode)->lock held.
5294  */
drop_outstanding_extent(struct inode * inode,u64 num_bytes)5295 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5296 {
5297 	unsigned drop_inode_space = 0;
5298 	unsigned dropped_extents = 0;
5299 	unsigned num_extents = 0;
5300 
5301 	num_extents = (unsigned)div64_u64(num_bytes +
5302 					  BTRFS_MAX_EXTENT_SIZE - 1,
5303 					  BTRFS_MAX_EXTENT_SIZE);
5304 	ASSERT(num_extents);
5305 	ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5306 	BTRFS_I(inode)->outstanding_extents -= num_extents;
5307 
5308 	if (BTRFS_I(inode)->outstanding_extents == 0 &&
5309 	    test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5310 			       &BTRFS_I(inode)->runtime_flags))
5311 		drop_inode_space = 1;
5312 
5313 	/*
5314 	 * If we have more or the same amount of outsanding extents than we have
5315 	 * reserved then we need to leave the reserved extents count alone.
5316 	 */
5317 	if (BTRFS_I(inode)->outstanding_extents >=
5318 	    BTRFS_I(inode)->reserved_extents)
5319 		return drop_inode_space;
5320 
5321 	dropped_extents = BTRFS_I(inode)->reserved_extents -
5322 		BTRFS_I(inode)->outstanding_extents;
5323 	BTRFS_I(inode)->reserved_extents -= dropped_extents;
5324 	return dropped_extents + drop_inode_space;
5325 }
5326 
5327 /**
5328  * calc_csum_metadata_size - return the amount of metada space that must be
5329  *	reserved/free'd for the given bytes.
5330  * @inode: the inode we're manipulating
5331  * @num_bytes: the number of bytes in question
5332  * @reserve: 1 if we are reserving space, 0 if we are freeing space
5333  *
5334  * This adjusts the number of csum_bytes in the inode and then returns the
5335  * correct amount of metadata that must either be reserved or freed.  We
5336  * calculate how many checksums we can fit into one leaf and then divide the
5337  * number of bytes that will need to be checksumed by this value to figure out
5338  * how many checksums will be required.  If we are adding bytes then the number
5339  * may go up and we will return the number of additional bytes that must be
5340  * reserved.  If it is going down we will return the number of bytes that must
5341  * be freed.
5342  *
5343  * This must be called with BTRFS_I(inode)->lock held.
5344  */
calc_csum_metadata_size(struct inode * inode,u64 num_bytes,int reserve)5345 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5346 				   int reserve)
5347 {
5348 	struct btrfs_root *root = BTRFS_I(inode)->root;
5349 	u64 old_csums, num_csums;
5350 
5351 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5352 	    BTRFS_I(inode)->csum_bytes == 0)
5353 		return 0;
5354 
5355 	old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5356 	if (reserve)
5357 		BTRFS_I(inode)->csum_bytes += num_bytes;
5358 	else
5359 		BTRFS_I(inode)->csum_bytes -= num_bytes;
5360 	num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5361 
5362 	/* No change, no need to reserve more */
5363 	if (old_csums == num_csums)
5364 		return 0;
5365 
5366 	if (reserve)
5367 		return btrfs_calc_trans_metadata_size(root,
5368 						      num_csums - old_csums);
5369 
5370 	return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5371 }
5372 
btrfs_delalloc_reserve_metadata(struct inode * inode,u64 num_bytes)5373 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5374 {
5375 	struct btrfs_root *root = BTRFS_I(inode)->root;
5376 	struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5377 	u64 to_reserve = 0;
5378 	u64 csum_bytes;
5379 	unsigned nr_extents = 0;
5380 	int extra_reserve = 0;
5381 	enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5382 	int ret = 0;
5383 	bool delalloc_lock = true;
5384 	u64 to_free = 0;
5385 	unsigned dropped;
5386 
5387 	/* If we are a free space inode we need to not flush since we will be in
5388 	 * the middle of a transaction commit.  We also don't need the delalloc
5389 	 * mutex since we won't race with anybody.  We need this mostly to make
5390 	 * lockdep shut its filthy mouth.
5391 	 */
5392 	if (btrfs_is_free_space_inode(inode)) {
5393 		flush = BTRFS_RESERVE_NO_FLUSH;
5394 		delalloc_lock = false;
5395 	}
5396 
5397 	if (flush != BTRFS_RESERVE_NO_FLUSH &&
5398 	    btrfs_transaction_in_commit(root->fs_info))
5399 		schedule_timeout(1);
5400 
5401 	if (delalloc_lock)
5402 		mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5403 
5404 	num_bytes = ALIGN(num_bytes, root->sectorsize);
5405 
5406 	spin_lock(&BTRFS_I(inode)->lock);
5407 	nr_extents = (unsigned)div64_u64(num_bytes +
5408 					 BTRFS_MAX_EXTENT_SIZE - 1,
5409 					 BTRFS_MAX_EXTENT_SIZE);
5410 	BTRFS_I(inode)->outstanding_extents += nr_extents;
5411 	nr_extents = 0;
5412 
5413 	if (BTRFS_I(inode)->outstanding_extents >
5414 	    BTRFS_I(inode)->reserved_extents)
5415 		nr_extents = BTRFS_I(inode)->outstanding_extents -
5416 			BTRFS_I(inode)->reserved_extents;
5417 
5418 	/*
5419 	 * Add an item to reserve for updating the inode when we complete the
5420 	 * delalloc io.
5421 	 */
5422 	if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5423 		      &BTRFS_I(inode)->runtime_flags)) {
5424 		nr_extents++;
5425 		extra_reserve = 1;
5426 	}
5427 
5428 	to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5429 	to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5430 	csum_bytes = BTRFS_I(inode)->csum_bytes;
5431 	spin_unlock(&BTRFS_I(inode)->lock);
5432 
5433 	if (root->fs_info->quota_enabled) {
5434 		ret = btrfs_qgroup_reserve(root, nr_extents * root->nodesize);
5435 		if (ret)
5436 			goto out_fail;
5437 	}
5438 
5439 	ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5440 	if (unlikely(ret)) {
5441 		if (root->fs_info->quota_enabled)
5442 			btrfs_qgroup_free(root, nr_extents * root->nodesize);
5443 		goto out_fail;
5444 	}
5445 
5446 	spin_lock(&BTRFS_I(inode)->lock);
5447 	if (extra_reserve) {
5448 		set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5449 			&BTRFS_I(inode)->runtime_flags);
5450 		nr_extents--;
5451 	}
5452 	BTRFS_I(inode)->reserved_extents += nr_extents;
5453 	spin_unlock(&BTRFS_I(inode)->lock);
5454 
5455 	if (delalloc_lock)
5456 		mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5457 
5458 	if (to_reserve)
5459 		trace_btrfs_space_reservation(root->fs_info, "delalloc",
5460 					      btrfs_ino(inode), to_reserve, 1);
5461 	block_rsv_add_bytes(block_rsv, to_reserve, 1);
5462 
5463 	return 0;
5464 
5465 out_fail:
5466 	spin_lock(&BTRFS_I(inode)->lock);
5467 	dropped = drop_outstanding_extent(inode, num_bytes);
5468 	/*
5469 	 * If the inodes csum_bytes is the same as the original
5470 	 * csum_bytes then we know we haven't raced with any free()ers
5471 	 * so we can just reduce our inodes csum bytes and carry on.
5472 	 */
5473 	if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5474 		calc_csum_metadata_size(inode, num_bytes, 0);
5475 	} else {
5476 		u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5477 		u64 bytes;
5478 
5479 		/*
5480 		 * This is tricky, but first we need to figure out how much we
5481 		 * free'd from any free-ers that occured during this
5482 		 * reservation, so we reset ->csum_bytes to the csum_bytes
5483 		 * before we dropped our lock, and then call the free for the
5484 		 * number of bytes that were freed while we were trying our
5485 		 * reservation.
5486 		 */
5487 		bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5488 		BTRFS_I(inode)->csum_bytes = csum_bytes;
5489 		to_free = calc_csum_metadata_size(inode, bytes, 0);
5490 
5491 
5492 		/*
5493 		 * Now we need to see how much we would have freed had we not
5494 		 * been making this reservation and our ->csum_bytes were not
5495 		 * artificially inflated.
5496 		 */
5497 		BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5498 		bytes = csum_bytes - orig_csum_bytes;
5499 		bytes = calc_csum_metadata_size(inode, bytes, 0);
5500 
5501 		/*
5502 		 * Now reset ->csum_bytes to what it should be.  If bytes is
5503 		 * more than to_free then we would have free'd more space had we
5504 		 * not had an artificially high ->csum_bytes, so we need to free
5505 		 * the remainder.  If bytes is the same or less then we don't
5506 		 * need to do anything, the other free-ers did the correct
5507 		 * thing.
5508 		 */
5509 		BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5510 		if (bytes > to_free)
5511 			to_free = bytes - to_free;
5512 		else
5513 			to_free = 0;
5514 	}
5515 	spin_unlock(&BTRFS_I(inode)->lock);
5516 	if (dropped)
5517 		to_free += btrfs_calc_trans_metadata_size(root, dropped);
5518 
5519 	if (to_free) {
5520 		btrfs_block_rsv_release(root, block_rsv, to_free);
5521 		trace_btrfs_space_reservation(root->fs_info, "delalloc",
5522 					      btrfs_ino(inode), to_free, 0);
5523 	}
5524 	if (delalloc_lock)
5525 		mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5526 	return ret;
5527 }
5528 
5529 /**
5530  * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5531  * @inode: the inode to release the reservation for
5532  * @num_bytes: the number of bytes we're releasing
5533  *
5534  * This will release the metadata reservation for an inode.  This can be called
5535  * once we complete IO for a given set of bytes to release their metadata
5536  * reservations.
5537  */
btrfs_delalloc_release_metadata(struct inode * inode,u64 num_bytes)5538 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5539 {
5540 	struct btrfs_root *root = BTRFS_I(inode)->root;
5541 	u64 to_free = 0;
5542 	unsigned dropped;
5543 
5544 	num_bytes = ALIGN(num_bytes, root->sectorsize);
5545 	spin_lock(&BTRFS_I(inode)->lock);
5546 	dropped = drop_outstanding_extent(inode, num_bytes);
5547 
5548 	if (num_bytes)
5549 		to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5550 	spin_unlock(&BTRFS_I(inode)->lock);
5551 	if (dropped > 0)
5552 		to_free += btrfs_calc_trans_metadata_size(root, dropped);
5553 
5554 	if (btrfs_test_is_dummy_root(root))
5555 		return;
5556 
5557 	trace_btrfs_space_reservation(root->fs_info, "delalloc",
5558 				      btrfs_ino(inode), to_free, 0);
5559 
5560 	btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5561 				to_free);
5562 }
5563 
5564 /**
5565  * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5566  * @inode: inode we're writing to
5567  * @num_bytes: the number of bytes we want to allocate
5568  *
5569  * This will do the following things
5570  *
5571  * o reserve space in the data space info for num_bytes
5572  * o reserve space in the metadata space info based on number of outstanding
5573  *   extents and how much csums will be needed
5574  * o add to the inodes ->delalloc_bytes
5575  * o add it to the fs_info's delalloc inodes list.
5576  *
5577  * This will return 0 for success and -ENOSPC if there is no space left.
5578  */
btrfs_delalloc_reserve_space(struct inode * inode,u64 num_bytes)5579 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5580 {
5581 	int ret;
5582 
5583 	ret = btrfs_check_data_free_space(inode, num_bytes, num_bytes);
5584 	if (ret)
5585 		return ret;
5586 
5587 	ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5588 	if (ret) {
5589 		btrfs_free_reserved_data_space(inode, num_bytes);
5590 		return ret;
5591 	}
5592 
5593 	return 0;
5594 }
5595 
5596 /**
5597  * btrfs_delalloc_release_space - release data and metadata space for delalloc
5598  * @inode: inode we're releasing space for
5599  * @num_bytes: the number of bytes we want to free up
5600  *
5601  * This must be matched with a call to btrfs_delalloc_reserve_space.  This is
5602  * called in the case that we don't need the metadata AND data reservations
5603  * anymore.  So if there is an error or we insert an inline extent.
5604  *
5605  * This function will release the metadata space that was not used and will
5606  * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5607  * list if there are no delalloc bytes left.
5608  */
btrfs_delalloc_release_space(struct inode * inode,u64 num_bytes)5609 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5610 {
5611 	btrfs_delalloc_release_metadata(inode, num_bytes);
5612 	btrfs_free_reserved_data_space(inode, num_bytes);
5613 }
5614 
update_block_group(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 bytenr,u64 num_bytes,int alloc)5615 static int update_block_group(struct btrfs_trans_handle *trans,
5616 			      struct btrfs_root *root, u64 bytenr,
5617 			      u64 num_bytes, int alloc)
5618 {
5619 	struct btrfs_block_group_cache *cache = NULL;
5620 	struct btrfs_fs_info *info = root->fs_info;
5621 	u64 total = num_bytes;
5622 	u64 old_val;
5623 	u64 byte_in_group;
5624 	int factor;
5625 
5626 	/* block accounting for super block */
5627 	spin_lock(&info->delalloc_root_lock);
5628 	old_val = btrfs_super_bytes_used(info->super_copy);
5629 	if (alloc)
5630 		old_val += num_bytes;
5631 	else
5632 		old_val -= num_bytes;
5633 	btrfs_set_super_bytes_used(info->super_copy, old_val);
5634 	spin_unlock(&info->delalloc_root_lock);
5635 
5636 	while (total) {
5637 		cache = btrfs_lookup_block_group(info, bytenr);
5638 		if (!cache)
5639 			return -ENOENT;
5640 		if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5641 				    BTRFS_BLOCK_GROUP_RAID1 |
5642 				    BTRFS_BLOCK_GROUP_RAID10))
5643 			factor = 2;
5644 		else
5645 			factor = 1;
5646 		/*
5647 		 * If this block group has free space cache written out, we
5648 		 * need to make sure to load it if we are removing space.  This
5649 		 * is because we need the unpinning stage to actually add the
5650 		 * space back to the block group, otherwise we will leak space.
5651 		 */
5652 		if (!alloc && cache->cached == BTRFS_CACHE_NO)
5653 			cache_block_group(cache, 1);
5654 
5655 		byte_in_group = bytenr - cache->key.objectid;
5656 		WARN_ON(byte_in_group > cache->key.offset);
5657 
5658 		spin_lock(&cache->space_info->lock);
5659 		spin_lock(&cache->lock);
5660 
5661 		if (btrfs_test_opt(root, SPACE_CACHE) &&
5662 		    cache->disk_cache_state < BTRFS_DC_CLEAR)
5663 			cache->disk_cache_state = BTRFS_DC_CLEAR;
5664 
5665 		old_val = btrfs_block_group_used(&cache->item);
5666 		num_bytes = min(total, cache->key.offset - byte_in_group);
5667 		if (alloc) {
5668 			old_val += num_bytes;
5669 			btrfs_set_block_group_used(&cache->item, old_val);
5670 			cache->reserved -= num_bytes;
5671 			cache->space_info->bytes_reserved -= num_bytes;
5672 			cache->space_info->bytes_used += num_bytes;
5673 			cache->space_info->disk_used += num_bytes * factor;
5674 			spin_unlock(&cache->lock);
5675 			spin_unlock(&cache->space_info->lock);
5676 		} else {
5677 			old_val -= num_bytes;
5678 			btrfs_set_block_group_used(&cache->item, old_val);
5679 			cache->pinned += num_bytes;
5680 			cache->space_info->bytes_pinned += num_bytes;
5681 			cache->space_info->bytes_used -= num_bytes;
5682 			cache->space_info->disk_used -= num_bytes * factor;
5683 			spin_unlock(&cache->lock);
5684 			spin_unlock(&cache->space_info->lock);
5685 
5686 			set_extent_dirty(info->pinned_extents,
5687 					 bytenr, bytenr + num_bytes - 1,
5688 					 GFP_NOFS | __GFP_NOFAIL);
5689 			/*
5690 			 * No longer have used bytes in this block group, queue
5691 			 * it for deletion.
5692 			 */
5693 			if (old_val == 0) {
5694 				spin_lock(&info->unused_bgs_lock);
5695 				if (list_empty(&cache->bg_list)) {
5696 					btrfs_get_block_group(cache);
5697 					list_add_tail(&cache->bg_list,
5698 						      &info->unused_bgs);
5699 				}
5700 				spin_unlock(&info->unused_bgs_lock);
5701 			}
5702 		}
5703 
5704 		spin_lock(&trans->transaction->dirty_bgs_lock);
5705 		if (list_empty(&cache->dirty_list)) {
5706 			list_add_tail(&cache->dirty_list,
5707 				      &trans->transaction->dirty_bgs);
5708 				trans->transaction->num_dirty_bgs++;
5709 			btrfs_get_block_group(cache);
5710 		}
5711 		spin_unlock(&trans->transaction->dirty_bgs_lock);
5712 
5713 		btrfs_put_block_group(cache);
5714 		total -= num_bytes;
5715 		bytenr += num_bytes;
5716 	}
5717 	return 0;
5718 }
5719 
first_logical_byte(struct btrfs_root * root,u64 search_start)5720 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5721 {
5722 	struct btrfs_block_group_cache *cache;
5723 	u64 bytenr;
5724 
5725 	spin_lock(&root->fs_info->block_group_cache_lock);
5726 	bytenr = root->fs_info->first_logical_byte;
5727 	spin_unlock(&root->fs_info->block_group_cache_lock);
5728 
5729 	if (bytenr < (u64)-1)
5730 		return bytenr;
5731 
5732 	cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5733 	if (!cache)
5734 		return 0;
5735 
5736 	bytenr = cache->key.objectid;
5737 	btrfs_put_block_group(cache);
5738 
5739 	return bytenr;
5740 }
5741 
pin_down_extent(struct btrfs_root * root,struct btrfs_block_group_cache * cache,u64 bytenr,u64 num_bytes,int reserved)5742 static int pin_down_extent(struct btrfs_root *root,
5743 			   struct btrfs_block_group_cache *cache,
5744 			   u64 bytenr, u64 num_bytes, int reserved)
5745 {
5746 	spin_lock(&cache->space_info->lock);
5747 	spin_lock(&cache->lock);
5748 	cache->pinned += num_bytes;
5749 	cache->space_info->bytes_pinned += num_bytes;
5750 	if (reserved) {
5751 		cache->reserved -= num_bytes;
5752 		cache->space_info->bytes_reserved -= num_bytes;
5753 	}
5754 	spin_unlock(&cache->lock);
5755 	spin_unlock(&cache->space_info->lock);
5756 
5757 	set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5758 			 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5759 	if (reserved)
5760 		trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5761 	return 0;
5762 }
5763 
5764 /*
5765  * this function must be called within transaction
5766  */
btrfs_pin_extent(struct btrfs_root * root,u64 bytenr,u64 num_bytes,int reserved)5767 int btrfs_pin_extent(struct btrfs_root *root,
5768 		     u64 bytenr, u64 num_bytes, int reserved)
5769 {
5770 	struct btrfs_block_group_cache *cache;
5771 
5772 	cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5773 	BUG_ON(!cache); /* Logic error */
5774 
5775 	pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5776 
5777 	btrfs_put_block_group(cache);
5778 	return 0;
5779 }
5780 
5781 /*
5782  * this function must be called within transaction
5783  */
btrfs_pin_extent_for_log_replay(struct btrfs_root * root,u64 bytenr,u64 num_bytes)5784 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5785 				    u64 bytenr, u64 num_bytes)
5786 {
5787 	struct btrfs_block_group_cache *cache;
5788 	int ret;
5789 
5790 	cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5791 	if (!cache)
5792 		return -EINVAL;
5793 
5794 	/*
5795 	 * pull in the free space cache (if any) so that our pin
5796 	 * removes the free space from the cache.  We have load_only set
5797 	 * to one because the slow code to read in the free extents does check
5798 	 * the pinned extents.
5799 	 */
5800 	cache_block_group(cache, 1);
5801 
5802 	pin_down_extent(root, cache, bytenr, num_bytes, 0);
5803 
5804 	/* remove us from the free space cache (if we're there at all) */
5805 	ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5806 	btrfs_put_block_group(cache);
5807 	return ret;
5808 }
5809 
__exclude_logged_extent(struct btrfs_root * root,u64 start,u64 num_bytes)5810 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5811 {
5812 	int ret;
5813 	struct btrfs_block_group_cache *block_group;
5814 	struct btrfs_caching_control *caching_ctl;
5815 
5816 	block_group = btrfs_lookup_block_group(root->fs_info, start);
5817 	if (!block_group)
5818 		return -EINVAL;
5819 
5820 	cache_block_group(block_group, 0);
5821 	caching_ctl = get_caching_control(block_group);
5822 
5823 	if (!caching_ctl) {
5824 		/* Logic error */
5825 		BUG_ON(!block_group_cache_done(block_group));
5826 		ret = btrfs_remove_free_space(block_group, start, num_bytes);
5827 	} else {
5828 		mutex_lock(&caching_ctl->mutex);
5829 
5830 		if (start >= caching_ctl->progress) {
5831 			ret = add_excluded_extent(root, start, num_bytes);
5832 		} else if (start + num_bytes <= caching_ctl->progress) {
5833 			ret = btrfs_remove_free_space(block_group,
5834 						      start, num_bytes);
5835 		} else {
5836 			num_bytes = caching_ctl->progress - start;
5837 			ret = btrfs_remove_free_space(block_group,
5838 						      start, num_bytes);
5839 			if (ret)
5840 				goto out_lock;
5841 
5842 			num_bytes = (start + num_bytes) -
5843 				caching_ctl->progress;
5844 			start = caching_ctl->progress;
5845 			ret = add_excluded_extent(root, start, num_bytes);
5846 		}
5847 out_lock:
5848 		mutex_unlock(&caching_ctl->mutex);
5849 		put_caching_control(caching_ctl);
5850 	}
5851 	btrfs_put_block_group(block_group);
5852 	return ret;
5853 }
5854 
btrfs_exclude_logged_extents(struct btrfs_root * log,struct extent_buffer * eb)5855 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5856 				 struct extent_buffer *eb)
5857 {
5858 	struct btrfs_file_extent_item *item;
5859 	struct btrfs_key key;
5860 	int found_type;
5861 	int i;
5862 
5863 	if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5864 		return 0;
5865 
5866 	for (i = 0; i < btrfs_header_nritems(eb); i++) {
5867 		btrfs_item_key_to_cpu(eb, &key, i);
5868 		if (key.type != BTRFS_EXTENT_DATA_KEY)
5869 			continue;
5870 		item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5871 		found_type = btrfs_file_extent_type(eb, item);
5872 		if (found_type == BTRFS_FILE_EXTENT_INLINE)
5873 			continue;
5874 		if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5875 			continue;
5876 		key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5877 		key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5878 		__exclude_logged_extent(log, key.objectid, key.offset);
5879 	}
5880 
5881 	return 0;
5882 }
5883 
5884 /**
5885  * btrfs_update_reserved_bytes - update the block_group and space info counters
5886  * @cache:	The cache we are manipulating
5887  * @num_bytes:	The number of bytes in question
5888  * @reserve:	One of the reservation enums
5889  * @delalloc:   The blocks are allocated for the delalloc write
5890  *
5891  * This is called by the allocator when it reserves space, or by somebody who is
5892  * freeing space that was never actually used on disk.  For example if you
5893  * reserve some space for a new leaf in transaction A and before transaction A
5894  * commits you free that leaf, you call this with reserve set to 0 in order to
5895  * clear the reservation.
5896  *
5897  * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5898  * ENOSPC accounting.  For data we handle the reservation through clearing the
5899  * delalloc bits in the io_tree.  We have to do this since we could end up
5900  * allocating less disk space for the amount of data we have reserved in the
5901  * case of compression.
5902  *
5903  * If this is a reservation and the block group has become read only we cannot
5904  * make the reservation and return -EAGAIN, otherwise this function always
5905  * succeeds.
5906  */
btrfs_update_reserved_bytes(struct btrfs_block_group_cache * cache,u64 num_bytes,int reserve,int delalloc)5907 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5908 				       u64 num_bytes, int reserve, int delalloc)
5909 {
5910 	struct btrfs_space_info *space_info = cache->space_info;
5911 	int ret = 0;
5912 
5913 	spin_lock(&space_info->lock);
5914 	spin_lock(&cache->lock);
5915 	if (reserve != RESERVE_FREE) {
5916 		if (cache->ro) {
5917 			ret = -EAGAIN;
5918 		} else {
5919 			cache->reserved += num_bytes;
5920 			space_info->bytes_reserved += num_bytes;
5921 			if (reserve == RESERVE_ALLOC) {
5922 				trace_btrfs_space_reservation(cache->fs_info,
5923 						"space_info", space_info->flags,
5924 						num_bytes, 0);
5925 				space_info->bytes_may_use -= num_bytes;
5926 			}
5927 
5928 			if (delalloc)
5929 				cache->delalloc_bytes += num_bytes;
5930 		}
5931 	} else {
5932 		if (cache->ro)
5933 			space_info->bytes_readonly += num_bytes;
5934 		cache->reserved -= num_bytes;
5935 		space_info->bytes_reserved -= num_bytes;
5936 
5937 		if (delalloc)
5938 			cache->delalloc_bytes -= num_bytes;
5939 	}
5940 	spin_unlock(&cache->lock);
5941 	spin_unlock(&space_info->lock);
5942 	return ret;
5943 }
5944 
btrfs_prepare_extent_commit(struct btrfs_trans_handle * trans,struct btrfs_root * root)5945 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5946 				struct btrfs_root *root)
5947 {
5948 	struct btrfs_fs_info *fs_info = root->fs_info;
5949 	struct btrfs_caching_control *next;
5950 	struct btrfs_caching_control *caching_ctl;
5951 	struct btrfs_block_group_cache *cache;
5952 
5953 	down_write(&fs_info->commit_root_sem);
5954 
5955 	list_for_each_entry_safe(caching_ctl, next,
5956 				 &fs_info->caching_block_groups, list) {
5957 		cache = caching_ctl->block_group;
5958 		if (block_group_cache_done(cache)) {
5959 			cache->last_byte_to_unpin = (u64)-1;
5960 			list_del_init(&caching_ctl->list);
5961 			put_caching_control(caching_ctl);
5962 		} else {
5963 			cache->last_byte_to_unpin = caching_ctl->progress;
5964 		}
5965 	}
5966 
5967 	if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5968 		fs_info->pinned_extents = &fs_info->freed_extents[1];
5969 	else
5970 		fs_info->pinned_extents = &fs_info->freed_extents[0];
5971 
5972 	up_write(&fs_info->commit_root_sem);
5973 
5974 	update_global_block_rsv(fs_info);
5975 }
5976 
unpin_extent_range(struct btrfs_root * root,u64 start,u64 end,const bool return_free_space)5977 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
5978 			      const bool return_free_space)
5979 {
5980 	struct btrfs_fs_info *fs_info = root->fs_info;
5981 	struct btrfs_block_group_cache *cache = NULL;
5982 	struct btrfs_space_info *space_info;
5983 	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5984 	u64 len;
5985 	bool readonly;
5986 
5987 	while (start <= end) {
5988 		readonly = false;
5989 		if (!cache ||
5990 		    start >= cache->key.objectid + cache->key.offset) {
5991 			if (cache)
5992 				btrfs_put_block_group(cache);
5993 			cache = btrfs_lookup_block_group(fs_info, start);
5994 			BUG_ON(!cache); /* Logic error */
5995 		}
5996 
5997 		len = cache->key.objectid + cache->key.offset - start;
5998 		len = min(len, end + 1 - start);
5999 
6000 		if (start < cache->last_byte_to_unpin) {
6001 			len = min(len, cache->last_byte_to_unpin - start);
6002 			if (return_free_space)
6003 				btrfs_add_free_space(cache, start, len);
6004 		}
6005 
6006 		start += len;
6007 		space_info = cache->space_info;
6008 
6009 		spin_lock(&space_info->lock);
6010 		spin_lock(&cache->lock);
6011 		cache->pinned -= len;
6012 		space_info->bytes_pinned -= len;
6013 		percpu_counter_add(&space_info->total_bytes_pinned, -len);
6014 		if (cache->ro) {
6015 			space_info->bytes_readonly += len;
6016 			readonly = true;
6017 		}
6018 		spin_unlock(&cache->lock);
6019 		if (!readonly && global_rsv->space_info == space_info) {
6020 			spin_lock(&global_rsv->lock);
6021 			if (!global_rsv->full) {
6022 				len = min(len, global_rsv->size -
6023 					  global_rsv->reserved);
6024 				global_rsv->reserved += len;
6025 				space_info->bytes_may_use += len;
6026 				if (global_rsv->reserved >= global_rsv->size)
6027 					global_rsv->full = 1;
6028 			}
6029 			spin_unlock(&global_rsv->lock);
6030 		}
6031 		spin_unlock(&space_info->lock);
6032 	}
6033 
6034 	if (cache)
6035 		btrfs_put_block_group(cache);
6036 	return 0;
6037 }
6038 
btrfs_finish_extent_commit(struct btrfs_trans_handle * trans,struct btrfs_root * root)6039 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6040 			       struct btrfs_root *root)
6041 {
6042 	struct btrfs_fs_info *fs_info = root->fs_info;
6043 	struct extent_io_tree *unpin;
6044 	u64 start;
6045 	u64 end;
6046 	int ret;
6047 
6048 	if (trans->aborted)
6049 		return 0;
6050 
6051 	if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6052 		unpin = &fs_info->freed_extents[1];
6053 	else
6054 		unpin = &fs_info->freed_extents[0];
6055 
6056 	while (1) {
6057 		mutex_lock(&fs_info->unused_bg_unpin_mutex);
6058 		ret = find_first_extent_bit(unpin, 0, &start, &end,
6059 					    EXTENT_DIRTY, NULL);
6060 		if (ret) {
6061 			mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6062 			break;
6063 		}
6064 
6065 		if (btrfs_test_opt(root, DISCARD))
6066 			ret = btrfs_discard_extent(root, start,
6067 						   end + 1 - start, NULL);
6068 
6069 		clear_extent_dirty(unpin, start, end, GFP_NOFS);
6070 		unpin_extent_range(root, start, end, true);
6071 		mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6072 		cond_resched();
6073 	}
6074 
6075 	return 0;
6076 }
6077 
add_pinned_bytes(struct btrfs_fs_info * fs_info,u64 num_bytes,u64 owner,u64 root_objectid)6078 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6079 			     u64 owner, u64 root_objectid)
6080 {
6081 	struct btrfs_space_info *space_info;
6082 	u64 flags;
6083 
6084 	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6085 		if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6086 			flags = BTRFS_BLOCK_GROUP_SYSTEM;
6087 		else
6088 			flags = BTRFS_BLOCK_GROUP_METADATA;
6089 	} else {
6090 		flags = BTRFS_BLOCK_GROUP_DATA;
6091 	}
6092 
6093 	space_info = __find_space_info(fs_info, flags);
6094 	BUG_ON(!space_info); /* Logic bug */
6095 	percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6096 }
6097 
6098 
__btrfs_free_extent(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 bytenr,u64 num_bytes,u64 parent,u64 root_objectid,u64 owner_objectid,u64 owner_offset,int refs_to_drop,struct btrfs_delayed_extent_op * extent_op,int no_quota)6099 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6100 				struct btrfs_root *root,
6101 				u64 bytenr, u64 num_bytes, u64 parent,
6102 				u64 root_objectid, u64 owner_objectid,
6103 				u64 owner_offset, int refs_to_drop,
6104 				struct btrfs_delayed_extent_op *extent_op,
6105 				int no_quota)
6106 {
6107 	struct btrfs_key key;
6108 	struct btrfs_path *path;
6109 	struct btrfs_fs_info *info = root->fs_info;
6110 	struct btrfs_root *extent_root = info->extent_root;
6111 	struct extent_buffer *leaf;
6112 	struct btrfs_extent_item *ei;
6113 	struct btrfs_extent_inline_ref *iref;
6114 	int ret;
6115 	int is_data;
6116 	int extent_slot = 0;
6117 	int found_extent = 0;
6118 	int num_to_del = 1;
6119 	u32 item_size;
6120 	u64 refs;
6121 	int last_ref = 0;
6122 	enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_SUB_EXCL;
6123 	bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6124 						 SKINNY_METADATA);
6125 
6126 	if (!info->quota_enabled || !is_fstree(root_objectid))
6127 		no_quota = 1;
6128 
6129 	path = btrfs_alloc_path();
6130 	if (!path)
6131 		return -ENOMEM;
6132 
6133 	path->reada = 1;
6134 	path->leave_spinning = 1;
6135 
6136 	is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6137 	BUG_ON(!is_data && refs_to_drop != 1);
6138 
6139 	if (is_data)
6140 		skinny_metadata = 0;
6141 
6142 	ret = lookup_extent_backref(trans, extent_root, path, &iref,
6143 				    bytenr, num_bytes, parent,
6144 				    root_objectid, owner_objectid,
6145 				    owner_offset);
6146 	if (ret == 0) {
6147 		extent_slot = path->slots[0];
6148 		while (extent_slot >= 0) {
6149 			btrfs_item_key_to_cpu(path->nodes[0], &key,
6150 					      extent_slot);
6151 			if (key.objectid != bytenr)
6152 				break;
6153 			if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6154 			    key.offset == num_bytes) {
6155 				found_extent = 1;
6156 				break;
6157 			}
6158 			if (key.type == BTRFS_METADATA_ITEM_KEY &&
6159 			    key.offset == owner_objectid) {
6160 				found_extent = 1;
6161 				break;
6162 			}
6163 			if (path->slots[0] - extent_slot > 5)
6164 				break;
6165 			extent_slot--;
6166 		}
6167 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6168 		item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6169 		if (found_extent && item_size < sizeof(*ei))
6170 			found_extent = 0;
6171 #endif
6172 		if (!found_extent) {
6173 			BUG_ON(iref);
6174 			ret = remove_extent_backref(trans, extent_root, path,
6175 						    NULL, refs_to_drop,
6176 						    is_data, &last_ref);
6177 			if (ret) {
6178 				btrfs_abort_transaction(trans, extent_root, ret);
6179 				goto out;
6180 			}
6181 			btrfs_release_path(path);
6182 			path->leave_spinning = 1;
6183 
6184 			key.objectid = bytenr;
6185 			key.type = BTRFS_EXTENT_ITEM_KEY;
6186 			key.offset = num_bytes;
6187 
6188 			if (!is_data && skinny_metadata) {
6189 				key.type = BTRFS_METADATA_ITEM_KEY;
6190 				key.offset = owner_objectid;
6191 			}
6192 
6193 			ret = btrfs_search_slot(trans, extent_root,
6194 						&key, path, -1, 1);
6195 			if (ret > 0 && skinny_metadata && path->slots[0]) {
6196 				/*
6197 				 * Couldn't find our skinny metadata item,
6198 				 * see if we have ye olde extent item.
6199 				 */
6200 				path->slots[0]--;
6201 				btrfs_item_key_to_cpu(path->nodes[0], &key,
6202 						      path->slots[0]);
6203 				if (key.objectid == bytenr &&
6204 				    key.type == BTRFS_EXTENT_ITEM_KEY &&
6205 				    key.offset == num_bytes)
6206 					ret = 0;
6207 			}
6208 
6209 			if (ret > 0 && skinny_metadata) {
6210 				skinny_metadata = false;
6211 				key.objectid = bytenr;
6212 				key.type = BTRFS_EXTENT_ITEM_KEY;
6213 				key.offset = num_bytes;
6214 				btrfs_release_path(path);
6215 				ret = btrfs_search_slot(trans, extent_root,
6216 							&key, path, -1, 1);
6217 			}
6218 
6219 			if (ret) {
6220 				btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6221 					ret, bytenr);
6222 				if (ret > 0)
6223 					btrfs_print_leaf(extent_root,
6224 							 path->nodes[0]);
6225 			}
6226 			if (ret < 0) {
6227 				btrfs_abort_transaction(trans, extent_root, ret);
6228 				goto out;
6229 			}
6230 			extent_slot = path->slots[0];
6231 		}
6232 	} else if (WARN_ON(ret == -ENOENT)) {
6233 		btrfs_print_leaf(extent_root, path->nodes[0]);
6234 		btrfs_err(info,
6235 			"unable to find ref byte nr %llu parent %llu root %llu  owner %llu offset %llu",
6236 			bytenr, parent, root_objectid, owner_objectid,
6237 			owner_offset);
6238 		btrfs_abort_transaction(trans, extent_root, ret);
6239 		goto out;
6240 	} else {
6241 		btrfs_abort_transaction(trans, extent_root, ret);
6242 		goto out;
6243 	}
6244 
6245 	leaf = path->nodes[0];
6246 	item_size = btrfs_item_size_nr(leaf, extent_slot);
6247 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6248 	if (item_size < sizeof(*ei)) {
6249 		BUG_ON(found_extent || extent_slot != path->slots[0]);
6250 		ret = convert_extent_item_v0(trans, extent_root, path,
6251 					     owner_objectid, 0);
6252 		if (ret < 0) {
6253 			btrfs_abort_transaction(trans, extent_root, ret);
6254 			goto out;
6255 		}
6256 
6257 		btrfs_release_path(path);
6258 		path->leave_spinning = 1;
6259 
6260 		key.objectid = bytenr;
6261 		key.type = BTRFS_EXTENT_ITEM_KEY;
6262 		key.offset = num_bytes;
6263 
6264 		ret = btrfs_search_slot(trans, extent_root, &key, path,
6265 					-1, 1);
6266 		if (ret) {
6267 			btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6268 				ret, bytenr);
6269 			btrfs_print_leaf(extent_root, path->nodes[0]);
6270 		}
6271 		if (ret < 0) {
6272 			btrfs_abort_transaction(trans, extent_root, ret);
6273 			goto out;
6274 		}
6275 
6276 		extent_slot = path->slots[0];
6277 		leaf = path->nodes[0];
6278 		item_size = btrfs_item_size_nr(leaf, extent_slot);
6279 	}
6280 #endif
6281 	BUG_ON(item_size < sizeof(*ei));
6282 	ei = btrfs_item_ptr(leaf, extent_slot,
6283 			    struct btrfs_extent_item);
6284 	if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6285 	    key.type == BTRFS_EXTENT_ITEM_KEY) {
6286 		struct btrfs_tree_block_info *bi;
6287 		BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6288 		bi = (struct btrfs_tree_block_info *)(ei + 1);
6289 		WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6290 	}
6291 
6292 	refs = btrfs_extent_refs(leaf, ei);
6293 	if (refs < refs_to_drop) {
6294 		btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6295 			  "for bytenr %Lu", refs_to_drop, refs, bytenr);
6296 		ret = -EINVAL;
6297 		btrfs_abort_transaction(trans, extent_root, ret);
6298 		goto out;
6299 	}
6300 	refs -= refs_to_drop;
6301 
6302 	if (refs > 0) {
6303 		type = BTRFS_QGROUP_OPER_SUB_SHARED;
6304 		if (extent_op)
6305 			__run_delayed_extent_op(extent_op, leaf, ei);
6306 		/*
6307 		 * In the case of inline back ref, reference count will
6308 		 * be updated by remove_extent_backref
6309 		 */
6310 		if (iref) {
6311 			BUG_ON(!found_extent);
6312 		} else {
6313 			btrfs_set_extent_refs(leaf, ei, refs);
6314 			btrfs_mark_buffer_dirty(leaf);
6315 		}
6316 		if (found_extent) {
6317 			ret = remove_extent_backref(trans, extent_root, path,
6318 						    iref, refs_to_drop,
6319 						    is_data, &last_ref);
6320 			if (ret) {
6321 				btrfs_abort_transaction(trans, extent_root, ret);
6322 				goto out;
6323 			}
6324 		}
6325 		add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6326 				 root_objectid);
6327 	} else {
6328 		if (found_extent) {
6329 			BUG_ON(is_data && refs_to_drop !=
6330 			       extent_data_ref_count(root, path, iref));
6331 			if (iref) {
6332 				BUG_ON(path->slots[0] != extent_slot);
6333 			} else {
6334 				BUG_ON(path->slots[0] != extent_slot + 1);
6335 				path->slots[0] = extent_slot;
6336 				num_to_del = 2;
6337 			}
6338 		}
6339 
6340 		last_ref = 1;
6341 		ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6342 				      num_to_del);
6343 		if (ret) {
6344 			btrfs_abort_transaction(trans, extent_root, ret);
6345 			goto out;
6346 		}
6347 		btrfs_release_path(path);
6348 
6349 		if (is_data) {
6350 			ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6351 			if (ret) {
6352 				btrfs_abort_transaction(trans, extent_root, ret);
6353 				goto out;
6354 			}
6355 		}
6356 
6357 		ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6358 		if (ret) {
6359 			btrfs_abort_transaction(trans, extent_root, ret);
6360 			goto out;
6361 		}
6362 	}
6363 	btrfs_release_path(path);
6364 
6365 	/* Deal with the quota accounting */
6366 	if (!ret && last_ref && !no_quota) {
6367 		int mod_seq = 0;
6368 
6369 		if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
6370 		    type == BTRFS_QGROUP_OPER_SUB_SHARED)
6371 			mod_seq = 1;
6372 
6373 		ret = btrfs_qgroup_record_ref(trans, info, root_objectid,
6374 					      bytenr, num_bytes, type,
6375 					      mod_seq);
6376 	}
6377 out:
6378 	btrfs_free_path(path);
6379 	return ret;
6380 }
6381 
6382 /*
6383  * when we free an block, it is possible (and likely) that we free the last
6384  * delayed ref for that extent as well.  This searches the delayed ref tree for
6385  * a given extent, and if there are no other delayed refs to be processed, it
6386  * removes it from the tree.
6387  */
check_ref_cleanup(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 bytenr)6388 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6389 				      struct btrfs_root *root, u64 bytenr)
6390 {
6391 	struct btrfs_delayed_ref_head *head;
6392 	struct btrfs_delayed_ref_root *delayed_refs;
6393 	int ret = 0;
6394 
6395 	delayed_refs = &trans->transaction->delayed_refs;
6396 	spin_lock(&delayed_refs->lock);
6397 	head = btrfs_find_delayed_ref_head(trans, bytenr);
6398 	if (!head)
6399 		goto out_delayed_unlock;
6400 
6401 	spin_lock(&head->lock);
6402 	if (rb_first(&head->ref_root))
6403 		goto out;
6404 
6405 	if (head->extent_op) {
6406 		if (!head->must_insert_reserved)
6407 			goto out;
6408 		btrfs_free_delayed_extent_op(head->extent_op);
6409 		head->extent_op = NULL;
6410 	}
6411 
6412 	/*
6413 	 * waiting for the lock here would deadlock.  If someone else has it
6414 	 * locked they are already in the process of dropping it anyway
6415 	 */
6416 	if (!mutex_trylock(&head->mutex))
6417 		goto out;
6418 
6419 	/*
6420 	 * at this point we have a head with no other entries.  Go
6421 	 * ahead and process it.
6422 	 */
6423 	head->node.in_tree = 0;
6424 	rb_erase(&head->href_node, &delayed_refs->href_root);
6425 
6426 	atomic_dec(&delayed_refs->num_entries);
6427 
6428 	/*
6429 	 * we don't take a ref on the node because we're removing it from the
6430 	 * tree, so we just steal the ref the tree was holding.
6431 	 */
6432 	delayed_refs->num_heads--;
6433 	if (head->processing == 0)
6434 		delayed_refs->num_heads_ready--;
6435 	head->processing = 0;
6436 	spin_unlock(&head->lock);
6437 	spin_unlock(&delayed_refs->lock);
6438 
6439 	BUG_ON(head->extent_op);
6440 	if (head->must_insert_reserved)
6441 		ret = 1;
6442 
6443 	mutex_unlock(&head->mutex);
6444 	btrfs_put_delayed_ref(&head->node);
6445 	return ret;
6446 out:
6447 	spin_unlock(&head->lock);
6448 
6449 out_delayed_unlock:
6450 	spin_unlock(&delayed_refs->lock);
6451 	return 0;
6452 }
6453 
btrfs_free_tree_block(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * buf,u64 parent,int last_ref)6454 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6455 			   struct btrfs_root *root,
6456 			   struct extent_buffer *buf,
6457 			   u64 parent, int last_ref)
6458 {
6459 	int pin = 1;
6460 	int ret;
6461 
6462 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6463 		ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6464 					buf->start, buf->len,
6465 					parent, root->root_key.objectid,
6466 					btrfs_header_level(buf),
6467 					BTRFS_DROP_DELAYED_REF, NULL, 0);
6468 		BUG_ON(ret); /* -ENOMEM */
6469 	}
6470 
6471 	if (!last_ref)
6472 		return;
6473 
6474 	if (btrfs_header_generation(buf) == trans->transid) {
6475 		struct btrfs_block_group_cache *cache;
6476 
6477 		if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6478 			ret = check_ref_cleanup(trans, root, buf->start);
6479 			if (!ret)
6480 				goto out;
6481 		}
6482 
6483 		cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6484 
6485 		if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6486 			pin_down_extent(root, cache, buf->start, buf->len, 1);
6487 			btrfs_put_block_group(cache);
6488 			goto out;
6489 		}
6490 
6491 		WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6492 
6493 		btrfs_add_free_space(cache, buf->start, buf->len);
6494 		btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6495 		btrfs_put_block_group(cache);
6496 		trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6497 		pin = 0;
6498 	}
6499 out:
6500 	if (pin)
6501 		add_pinned_bytes(root->fs_info, buf->len,
6502 				 btrfs_header_level(buf),
6503 				 root->root_key.objectid);
6504 
6505 	/*
6506 	 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6507 	 * anymore.
6508 	 */
6509 	clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6510 }
6511 
6512 /* Can return -ENOMEM */
btrfs_free_extent(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 bytenr,u64 num_bytes,u64 parent,u64 root_objectid,u64 owner,u64 offset,int no_quota)6513 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6514 		      u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6515 		      u64 owner, u64 offset, int no_quota)
6516 {
6517 	int ret;
6518 	struct btrfs_fs_info *fs_info = root->fs_info;
6519 
6520 	if (btrfs_test_is_dummy_root(root))
6521 		return 0;
6522 
6523 	add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6524 
6525 	/*
6526 	 * tree log blocks never actually go into the extent allocation
6527 	 * tree, just update pinning info and exit early.
6528 	 */
6529 	if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6530 		WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6531 		/* unlocks the pinned mutex */
6532 		btrfs_pin_extent(root, bytenr, num_bytes, 1);
6533 		ret = 0;
6534 	} else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6535 		ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6536 					num_bytes,
6537 					parent, root_objectid, (int)owner,
6538 					BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6539 	} else {
6540 		ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6541 						num_bytes,
6542 						parent, root_objectid, owner,
6543 						offset, BTRFS_DROP_DELAYED_REF,
6544 						NULL, no_quota);
6545 	}
6546 	return ret;
6547 }
6548 
6549 /*
6550  * when we wait for progress in the block group caching, its because
6551  * our allocation attempt failed at least once.  So, we must sleep
6552  * and let some progress happen before we try again.
6553  *
6554  * This function will sleep at least once waiting for new free space to
6555  * show up, and then it will check the block group free space numbers
6556  * for our min num_bytes.  Another option is to have it go ahead
6557  * and look in the rbtree for a free extent of a given size, but this
6558  * is a good start.
6559  *
6560  * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6561  * any of the information in this block group.
6562  */
6563 static noinline void
wait_block_group_cache_progress(struct btrfs_block_group_cache * cache,u64 num_bytes)6564 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6565 				u64 num_bytes)
6566 {
6567 	struct btrfs_caching_control *caching_ctl;
6568 
6569 	caching_ctl = get_caching_control(cache);
6570 	if (!caching_ctl)
6571 		return;
6572 
6573 	wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6574 		   (cache->free_space_ctl->free_space >= num_bytes));
6575 
6576 	put_caching_control(caching_ctl);
6577 }
6578 
6579 static noinline int
wait_block_group_cache_done(struct btrfs_block_group_cache * cache)6580 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6581 {
6582 	struct btrfs_caching_control *caching_ctl;
6583 	int ret = 0;
6584 
6585 	caching_ctl = get_caching_control(cache);
6586 	if (!caching_ctl)
6587 		return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6588 
6589 	wait_event(caching_ctl->wait, block_group_cache_done(cache));
6590 	if (cache->cached == BTRFS_CACHE_ERROR)
6591 		ret = -EIO;
6592 	put_caching_control(caching_ctl);
6593 	return ret;
6594 }
6595 
__get_raid_index(u64 flags)6596 int __get_raid_index(u64 flags)
6597 {
6598 	if (flags & BTRFS_BLOCK_GROUP_RAID10)
6599 		return BTRFS_RAID_RAID10;
6600 	else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6601 		return BTRFS_RAID_RAID1;
6602 	else if (flags & BTRFS_BLOCK_GROUP_DUP)
6603 		return BTRFS_RAID_DUP;
6604 	else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6605 		return BTRFS_RAID_RAID0;
6606 	else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6607 		return BTRFS_RAID_RAID5;
6608 	else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6609 		return BTRFS_RAID_RAID6;
6610 
6611 	return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6612 }
6613 
get_block_group_index(struct btrfs_block_group_cache * cache)6614 int get_block_group_index(struct btrfs_block_group_cache *cache)
6615 {
6616 	return __get_raid_index(cache->flags);
6617 }
6618 
6619 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6620 	[BTRFS_RAID_RAID10]	= "raid10",
6621 	[BTRFS_RAID_RAID1]	= "raid1",
6622 	[BTRFS_RAID_DUP]	= "dup",
6623 	[BTRFS_RAID_RAID0]	= "raid0",
6624 	[BTRFS_RAID_SINGLE]	= "single",
6625 	[BTRFS_RAID_RAID5]	= "raid5",
6626 	[BTRFS_RAID_RAID6]	= "raid6",
6627 };
6628 
get_raid_name(enum btrfs_raid_types type)6629 static const char *get_raid_name(enum btrfs_raid_types type)
6630 {
6631 	if (type >= BTRFS_NR_RAID_TYPES)
6632 		return NULL;
6633 
6634 	return btrfs_raid_type_names[type];
6635 }
6636 
6637 enum btrfs_loop_type {
6638 	LOOP_CACHING_NOWAIT = 0,
6639 	LOOP_CACHING_WAIT = 1,
6640 	LOOP_ALLOC_CHUNK = 2,
6641 	LOOP_NO_EMPTY_SIZE = 3,
6642 };
6643 
6644 static inline void
btrfs_lock_block_group(struct btrfs_block_group_cache * cache,int delalloc)6645 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6646 		       int delalloc)
6647 {
6648 	if (delalloc)
6649 		down_read(&cache->data_rwsem);
6650 }
6651 
6652 static inline void
btrfs_grab_block_group(struct btrfs_block_group_cache * cache,int delalloc)6653 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6654 		       int delalloc)
6655 {
6656 	btrfs_get_block_group(cache);
6657 	if (delalloc)
6658 		down_read(&cache->data_rwsem);
6659 }
6660 
6661 static struct btrfs_block_group_cache *
btrfs_lock_cluster(struct btrfs_block_group_cache * block_group,struct btrfs_free_cluster * cluster,int delalloc)6662 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6663 		   struct btrfs_free_cluster *cluster,
6664 		   int delalloc)
6665 {
6666 	struct btrfs_block_group_cache *used_bg;
6667 	bool locked = false;
6668 again:
6669 	spin_lock(&cluster->refill_lock);
6670 	if (locked) {
6671 		if (used_bg == cluster->block_group)
6672 			return used_bg;
6673 
6674 		up_read(&used_bg->data_rwsem);
6675 		btrfs_put_block_group(used_bg);
6676 	}
6677 
6678 	used_bg = cluster->block_group;
6679 	if (!used_bg)
6680 		return NULL;
6681 
6682 	if (used_bg == block_group)
6683 		return used_bg;
6684 
6685 	btrfs_get_block_group(used_bg);
6686 
6687 	if (!delalloc)
6688 		return used_bg;
6689 
6690 	if (down_read_trylock(&used_bg->data_rwsem))
6691 		return used_bg;
6692 
6693 	spin_unlock(&cluster->refill_lock);
6694 	down_read(&used_bg->data_rwsem);
6695 	locked = true;
6696 	goto again;
6697 }
6698 
6699 static inline void
btrfs_release_block_group(struct btrfs_block_group_cache * cache,int delalloc)6700 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6701 			 int delalloc)
6702 {
6703 	if (delalloc)
6704 		up_read(&cache->data_rwsem);
6705 	btrfs_put_block_group(cache);
6706 }
6707 
6708 /*
6709  * walks the btree of allocated extents and find a hole of a given size.
6710  * The key ins is changed to record the hole:
6711  * ins->objectid == start position
6712  * ins->flags = BTRFS_EXTENT_ITEM_KEY
6713  * ins->offset == the size of the hole.
6714  * Any available blocks before search_start are skipped.
6715  *
6716  * If there is no suitable free space, we will record the max size of
6717  * the free space extent currently.
6718  */
find_free_extent(struct btrfs_root * orig_root,u64 num_bytes,u64 empty_size,u64 hint_byte,struct btrfs_key * ins,u64 flags,int delalloc)6719 static noinline int find_free_extent(struct btrfs_root *orig_root,
6720 				     u64 num_bytes, u64 empty_size,
6721 				     u64 hint_byte, struct btrfs_key *ins,
6722 				     u64 flags, int delalloc)
6723 {
6724 	int ret = 0;
6725 	struct btrfs_root *root = orig_root->fs_info->extent_root;
6726 	struct btrfs_free_cluster *last_ptr = NULL;
6727 	struct btrfs_block_group_cache *block_group = NULL;
6728 	u64 search_start = 0;
6729 	u64 max_extent_size = 0;
6730 	int empty_cluster = 2 * 1024 * 1024;
6731 	struct btrfs_space_info *space_info;
6732 	int loop = 0;
6733 	int index = __get_raid_index(flags);
6734 	int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6735 		RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6736 	bool failed_cluster_refill = false;
6737 	bool failed_alloc = false;
6738 	bool use_cluster = true;
6739 	bool have_caching_bg = false;
6740 
6741 	WARN_ON(num_bytes < root->sectorsize);
6742 	ins->type = BTRFS_EXTENT_ITEM_KEY;
6743 	ins->objectid = 0;
6744 	ins->offset = 0;
6745 
6746 	trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6747 
6748 	space_info = __find_space_info(root->fs_info, flags);
6749 	if (!space_info) {
6750 		btrfs_err(root->fs_info, "No space info for %llu", flags);
6751 		return -ENOSPC;
6752 	}
6753 
6754 	/*
6755 	 * If the space info is for both data and metadata it means we have a
6756 	 * small filesystem and we can't use the clustering stuff.
6757 	 */
6758 	if (btrfs_mixed_space_info(space_info))
6759 		use_cluster = false;
6760 
6761 	if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6762 		last_ptr = &root->fs_info->meta_alloc_cluster;
6763 		if (!btrfs_test_opt(root, SSD))
6764 			empty_cluster = 64 * 1024;
6765 	}
6766 
6767 	if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6768 	    btrfs_test_opt(root, SSD)) {
6769 		last_ptr = &root->fs_info->data_alloc_cluster;
6770 	}
6771 
6772 	if (last_ptr) {
6773 		spin_lock(&last_ptr->lock);
6774 		if (last_ptr->block_group)
6775 			hint_byte = last_ptr->window_start;
6776 		spin_unlock(&last_ptr->lock);
6777 	}
6778 
6779 	search_start = max(search_start, first_logical_byte(root, 0));
6780 	search_start = max(search_start, hint_byte);
6781 
6782 	if (!last_ptr)
6783 		empty_cluster = 0;
6784 
6785 	if (search_start == hint_byte) {
6786 		block_group = btrfs_lookup_block_group(root->fs_info,
6787 						       search_start);
6788 		/*
6789 		 * we don't want to use the block group if it doesn't match our
6790 		 * allocation bits, or if its not cached.
6791 		 *
6792 		 * However if we are re-searching with an ideal block group
6793 		 * picked out then we don't care that the block group is cached.
6794 		 */
6795 		if (block_group && block_group_bits(block_group, flags) &&
6796 		    block_group->cached != BTRFS_CACHE_NO) {
6797 			down_read(&space_info->groups_sem);
6798 			if (list_empty(&block_group->list) ||
6799 			    block_group->ro) {
6800 				/*
6801 				 * someone is removing this block group,
6802 				 * we can't jump into the have_block_group
6803 				 * target because our list pointers are not
6804 				 * valid
6805 				 */
6806 				btrfs_put_block_group(block_group);
6807 				up_read(&space_info->groups_sem);
6808 			} else {
6809 				index = get_block_group_index(block_group);
6810 				btrfs_lock_block_group(block_group, delalloc);
6811 				goto have_block_group;
6812 			}
6813 		} else if (block_group) {
6814 			btrfs_put_block_group(block_group);
6815 		}
6816 	}
6817 search:
6818 	have_caching_bg = false;
6819 	down_read(&space_info->groups_sem);
6820 	list_for_each_entry(block_group, &space_info->block_groups[index],
6821 			    list) {
6822 		u64 offset;
6823 		int cached;
6824 
6825 		btrfs_grab_block_group(block_group, delalloc);
6826 		search_start = block_group->key.objectid;
6827 
6828 		/*
6829 		 * this can happen if we end up cycling through all the
6830 		 * raid types, but we want to make sure we only allocate
6831 		 * for the proper type.
6832 		 */
6833 		if (!block_group_bits(block_group, flags)) {
6834 		    u64 extra = BTRFS_BLOCK_GROUP_DUP |
6835 				BTRFS_BLOCK_GROUP_RAID1 |
6836 				BTRFS_BLOCK_GROUP_RAID5 |
6837 				BTRFS_BLOCK_GROUP_RAID6 |
6838 				BTRFS_BLOCK_GROUP_RAID10;
6839 
6840 			/*
6841 			 * if they asked for extra copies and this block group
6842 			 * doesn't provide them, bail.  This does allow us to
6843 			 * fill raid0 from raid1.
6844 			 */
6845 			if ((flags & extra) && !(block_group->flags & extra))
6846 				goto loop;
6847 		}
6848 
6849 have_block_group:
6850 		cached = block_group_cache_done(block_group);
6851 		if (unlikely(!cached)) {
6852 			ret = cache_block_group(block_group, 0);
6853 			BUG_ON(ret < 0);
6854 			ret = 0;
6855 		}
6856 
6857 		if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6858 			goto loop;
6859 		if (unlikely(block_group->ro))
6860 			goto loop;
6861 
6862 		/*
6863 		 * Ok we want to try and use the cluster allocator, so
6864 		 * lets look there
6865 		 */
6866 		if (last_ptr) {
6867 			struct btrfs_block_group_cache *used_block_group;
6868 			unsigned long aligned_cluster;
6869 			/*
6870 			 * the refill lock keeps out other
6871 			 * people trying to start a new cluster
6872 			 */
6873 			used_block_group = btrfs_lock_cluster(block_group,
6874 							      last_ptr,
6875 							      delalloc);
6876 			if (!used_block_group)
6877 				goto refill_cluster;
6878 
6879 			if (used_block_group != block_group &&
6880 			    (used_block_group->ro ||
6881 			     !block_group_bits(used_block_group, flags)))
6882 				goto release_cluster;
6883 
6884 			offset = btrfs_alloc_from_cluster(used_block_group,
6885 						last_ptr,
6886 						num_bytes,
6887 						used_block_group->key.objectid,
6888 						&max_extent_size);
6889 			if (offset) {
6890 				/* we have a block, we're done */
6891 				spin_unlock(&last_ptr->refill_lock);
6892 				trace_btrfs_reserve_extent_cluster(root,
6893 						used_block_group,
6894 						search_start, num_bytes);
6895 				if (used_block_group != block_group) {
6896 					btrfs_release_block_group(block_group,
6897 								  delalloc);
6898 					block_group = used_block_group;
6899 				}
6900 				goto checks;
6901 			}
6902 
6903 			WARN_ON(last_ptr->block_group != used_block_group);
6904 release_cluster:
6905 			/* If we are on LOOP_NO_EMPTY_SIZE, we can't
6906 			 * set up a new clusters, so lets just skip it
6907 			 * and let the allocator find whatever block
6908 			 * it can find.  If we reach this point, we
6909 			 * will have tried the cluster allocator
6910 			 * plenty of times and not have found
6911 			 * anything, so we are likely way too
6912 			 * fragmented for the clustering stuff to find
6913 			 * anything.
6914 			 *
6915 			 * However, if the cluster is taken from the
6916 			 * current block group, release the cluster
6917 			 * first, so that we stand a better chance of
6918 			 * succeeding in the unclustered
6919 			 * allocation.  */
6920 			if (loop >= LOOP_NO_EMPTY_SIZE &&
6921 			    used_block_group != block_group) {
6922 				spin_unlock(&last_ptr->refill_lock);
6923 				btrfs_release_block_group(used_block_group,
6924 							  delalloc);
6925 				goto unclustered_alloc;
6926 			}
6927 
6928 			/*
6929 			 * this cluster didn't work out, free it and
6930 			 * start over
6931 			 */
6932 			btrfs_return_cluster_to_free_space(NULL, last_ptr);
6933 
6934 			if (used_block_group != block_group)
6935 				btrfs_release_block_group(used_block_group,
6936 							  delalloc);
6937 refill_cluster:
6938 			if (loop >= LOOP_NO_EMPTY_SIZE) {
6939 				spin_unlock(&last_ptr->refill_lock);
6940 				goto unclustered_alloc;
6941 			}
6942 
6943 			aligned_cluster = max_t(unsigned long,
6944 						empty_cluster + empty_size,
6945 					      block_group->full_stripe_len);
6946 
6947 			/* allocate a cluster in this block group */
6948 			ret = btrfs_find_space_cluster(root, block_group,
6949 						       last_ptr, search_start,
6950 						       num_bytes,
6951 						       aligned_cluster);
6952 			if (ret == 0) {
6953 				/*
6954 				 * now pull our allocation out of this
6955 				 * cluster
6956 				 */
6957 				offset = btrfs_alloc_from_cluster(block_group,
6958 							last_ptr,
6959 							num_bytes,
6960 							search_start,
6961 							&max_extent_size);
6962 				if (offset) {
6963 					/* we found one, proceed */
6964 					spin_unlock(&last_ptr->refill_lock);
6965 					trace_btrfs_reserve_extent_cluster(root,
6966 						block_group, search_start,
6967 						num_bytes);
6968 					goto checks;
6969 				}
6970 			} else if (!cached && loop > LOOP_CACHING_NOWAIT
6971 				   && !failed_cluster_refill) {
6972 				spin_unlock(&last_ptr->refill_lock);
6973 
6974 				failed_cluster_refill = true;
6975 				wait_block_group_cache_progress(block_group,
6976 				       num_bytes + empty_cluster + empty_size);
6977 				goto have_block_group;
6978 			}
6979 
6980 			/*
6981 			 * at this point we either didn't find a cluster
6982 			 * or we weren't able to allocate a block from our
6983 			 * cluster.  Free the cluster we've been trying
6984 			 * to use, and go to the next block group
6985 			 */
6986 			btrfs_return_cluster_to_free_space(NULL, last_ptr);
6987 			spin_unlock(&last_ptr->refill_lock);
6988 			goto loop;
6989 		}
6990 
6991 unclustered_alloc:
6992 		spin_lock(&block_group->free_space_ctl->tree_lock);
6993 		if (cached &&
6994 		    block_group->free_space_ctl->free_space <
6995 		    num_bytes + empty_cluster + empty_size) {
6996 			if (block_group->free_space_ctl->free_space >
6997 			    max_extent_size)
6998 				max_extent_size =
6999 					block_group->free_space_ctl->free_space;
7000 			spin_unlock(&block_group->free_space_ctl->tree_lock);
7001 			goto loop;
7002 		}
7003 		spin_unlock(&block_group->free_space_ctl->tree_lock);
7004 
7005 		offset = btrfs_find_space_for_alloc(block_group, search_start,
7006 						    num_bytes, empty_size,
7007 						    &max_extent_size);
7008 		/*
7009 		 * If we didn't find a chunk, and we haven't failed on this
7010 		 * block group before, and this block group is in the middle of
7011 		 * caching and we are ok with waiting, then go ahead and wait
7012 		 * for progress to be made, and set failed_alloc to true.
7013 		 *
7014 		 * If failed_alloc is true then we've already waited on this
7015 		 * block group once and should move on to the next block group.
7016 		 */
7017 		if (!offset && !failed_alloc && !cached &&
7018 		    loop > LOOP_CACHING_NOWAIT) {
7019 			wait_block_group_cache_progress(block_group,
7020 						num_bytes + empty_size);
7021 			failed_alloc = true;
7022 			goto have_block_group;
7023 		} else if (!offset) {
7024 			if (!cached)
7025 				have_caching_bg = true;
7026 			goto loop;
7027 		}
7028 checks:
7029 		search_start = ALIGN(offset, root->stripesize);
7030 
7031 		/* move on to the next group */
7032 		if (search_start + num_bytes >
7033 		    block_group->key.objectid + block_group->key.offset) {
7034 			btrfs_add_free_space(block_group, offset, num_bytes);
7035 			goto loop;
7036 		}
7037 
7038 		if (offset < search_start)
7039 			btrfs_add_free_space(block_group, offset,
7040 					     search_start - offset);
7041 		BUG_ON(offset > search_start);
7042 
7043 		ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7044 						  alloc_type, delalloc);
7045 		if (ret == -EAGAIN) {
7046 			btrfs_add_free_space(block_group, offset, num_bytes);
7047 			goto loop;
7048 		}
7049 
7050 		/* we are all good, lets return */
7051 		ins->objectid = search_start;
7052 		ins->offset = num_bytes;
7053 
7054 		trace_btrfs_reserve_extent(orig_root, block_group,
7055 					   search_start, num_bytes);
7056 		btrfs_release_block_group(block_group, delalloc);
7057 		break;
7058 loop:
7059 		failed_cluster_refill = false;
7060 		failed_alloc = false;
7061 		BUG_ON(index != get_block_group_index(block_group));
7062 		btrfs_release_block_group(block_group, delalloc);
7063 	}
7064 	up_read(&space_info->groups_sem);
7065 
7066 	if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7067 		goto search;
7068 
7069 	if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7070 		goto search;
7071 
7072 	/*
7073 	 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7074 	 *			caching kthreads as we move along
7075 	 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7076 	 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7077 	 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7078 	 *			again
7079 	 */
7080 	if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7081 		index = 0;
7082 		loop++;
7083 		if (loop == LOOP_ALLOC_CHUNK) {
7084 			struct btrfs_trans_handle *trans;
7085 			int exist = 0;
7086 
7087 			trans = current->journal_info;
7088 			if (trans)
7089 				exist = 1;
7090 			else
7091 				trans = btrfs_join_transaction(root);
7092 
7093 			if (IS_ERR(trans)) {
7094 				ret = PTR_ERR(trans);
7095 				goto out;
7096 			}
7097 
7098 			ret = do_chunk_alloc(trans, root, flags,
7099 					     CHUNK_ALLOC_FORCE);
7100 			/*
7101 			 * Do not bail out on ENOSPC since we
7102 			 * can do more things.
7103 			 */
7104 			if (ret < 0 && ret != -ENOSPC)
7105 				btrfs_abort_transaction(trans,
7106 							root, ret);
7107 			else
7108 				ret = 0;
7109 			if (!exist)
7110 				btrfs_end_transaction(trans, root);
7111 			if (ret)
7112 				goto out;
7113 		}
7114 
7115 		if (loop == LOOP_NO_EMPTY_SIZE) {
7116 			empty_size = 0;
7117 			empty_cluster = 0;
7118 		}
7119 
7120 		goto search;
7121 	} else if (!ins->objectid) {
7122 		ret = -ENOSPC;
7123 	} else if (ins->objectid) {
7124 		ret = 0;
7125 	}
7126 out:
7127 	if (ret == -ENOSPC)
7128 		ins->offset = max_extent_size;
7129 	return ret;
7130 }
7131 
dump_space_info(struct btrfs_space_info * info,u64 bytes,int dump_block_groups)7132 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7133 			    int dump_block_groups)
7134 {
7135 	struct btrfs_block_group_cache *cache;
7136 	int index = 0;
7137 
7138 	spin_lock(&info->lock);
7139 	printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7140 	       info->flags,
7141 	       info->total_bytes - info->bytes_used - info->bytes_pinned -
7142 	       info->bytes_reserved - info->bytes_readonly,
7143 	       (info->full) ? "" : "not ");
7144 	printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7145 	       "reserved=%llu, may_use=%llu, readonly=%llu\n",
7146 	       info->total_bytes, info->bytes_used, info->bytes_pinned,
7147 	       info->bytes_reserved, info->bytes_may_use,
7148 	       info->bytes_readonly);
7149 	spin_unlock(&info->lock);
7150 
7151 	if (!dump_block_groups)
7152 		return;
7153 
7154 	down_read(&info->groups_sem);
7155 again:
7156 	list_for_each_entry(cache, &info->block_groups[index], list) {
7157 		spin_lock(&cache->lock);
7158 		printk(KERN_INFO "BTRFS: "
7159 			   "block group %llu has %llu bytes, "
7160 			   "%llu used %llu pinned %llu reserved %s\n",
7161 		       cache->key.objectid, cache->key.offset,
7162 		       btrfs_block_group_used(&cache->item), cache->pinned,
7163 		       cache->reserved, cache->ro ? "[readonly]" : "");
7164 		btrfs_dump_free_space(cache, bytes);
7165 		spin_unlock(&cache->lock);
7166 	}
7167 	if (++index < BTRFS_NR_RAID_TYPES)
7168 		goto again;
7169 	up_read(&info->groups_sem);
7170 }
7171 
btrfs_reserve_extent(struct btrfs_root * root,u64 num_bytes,u64 min_alloc_size,u64 empty_size,u64 hint_byte,struct btrfs_key * ins,int is_data,int delalloc)7172 int btrfs_reserve_extent(struct btrfs_root *root,
7173 			 u64 num_bytes, u64 min_alloc_size,
7174 			 u64 empty_size, u64 hint_byte,
7175 			 struct btrfs_key *ins, int is_data, int delalloc)
7176 {
7177 	bool final_tried = false;
7178 	u64 flags;
7179 	int ret;
7180 
7181 	flags = btrfs_get_alloc_profile(root, is_data);
7182 again:
7183 	WARN_ON(num_bytes < root->sectorsize);
7184 	ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7185 			       flags, delalloc);
7186 
7187 	if (ret == -ENOSPC) {
7188 		if (!final_tried && ins->offset) {
7189 			num_bytes = min(num_bytes >> 1, ins->offset);
7190 			num_bytes = round_down(num_bytes, root->sectorsize);
7191 			num_bytes = max(num_bytes, min_alloc_size);
7192 			if (num_bytes == min_alloc_size)
7193 				final_tried = true;
7194 			goto again;
7195 		} else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7196 			struct btrfs_space_info *sinfo;
7197 
7198 			sinfo = __find_space_info(root->fs_info, flags);
7199 			btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7200 				flags, num_bytes);
7201 			if (sinfo)
7202 				dump_space_info(sinfo, num_bytes, 1);
7203 		}
7204 	}
7205 
7206 	return ret;
7207 }
7208 
__btrfs_free_reserved_extent(struct btrfs_root * root,u64 start,u64 len,int pin,int delalloc)7209 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7210 					u64 start, u64 len,
7211 					int pin, int delalloc)
7212 {
7213 	struct btrfs_block_group_cache *cache;
7214 	int ret = 0;
7215 
7216 	cache = btrfs_lookup_block_group(root->fs_info, start);
7217 	if (!cache) {
7218 		btrfs_err(root->fs_info, "Unable to find block group for %llu",
7219 			start);
7220 		return -ENOSPC;
7221 	}
7222 
7223 	if (pin)
7224 		pin_down_extent(root, cache, start, len, 1);
7225 	else {
7226 		if (btrfs_test_opt(root, DISCARD))
7227 			ret = btrfs_discard_extent(root, start, len, NULL);
7228 		btrfs_add_free_space(cache, start, len);
7229 		btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7230 	}
7231 
7232 	btrfs_put_block_group(cache);
7233 
7234 	trace_btrfs_reserved_extent_free(root, start, len);
7235 
7236 	return ret;
7237 }
7238 
btrfs_free_reserved_extent(struct btrfs_root * root,u64 start,u64 len,int delalloc)7239 int btrfs_free_reserved_extent(struct btrfs_root *root,
7240 			       u64 start, u64 len, int delalloc)
7241 {
7242 	return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7243 }
7244 
btrfs_free_and_pin_reserved_extent(struct btrfs_root * root,u64 start,u64 len)7245 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7246 				       u64 start, u64 len)
7247 {
7248 	return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7249 }
7250 
alloc_reserved_file_extent(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 parent,u64 root_objectid,u64 flags,u64 owner,u64 offset,struct btrfs_key * ins,int ref_mod)7251 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7252 				      struct btrfs_root *root,
7253 				      u64 parent, u64 root_objectid,
7254 				      u64 flags, u64 owner, u64 offset,
7255 				      struct btrfs_key *ins, int ref_mod)
7256 {
7257 	int ret;
7258 	struct btrfs_fs_info *fs_info = root->fs_info;
7259 	struct btrfs_extent_item *extent_item;
7260 	struct btrfs_extent_inline_ref *iref;
7261 	struct btrfs_path *path;
7262 	struct extent_buffer *leaf;
7263 	int type;
7264 	u32 size;
7265 
7266 	if (parent > 0)
7267 		type = BTRFS_SHARED_DATA_REF_KEY;
7268 	else
7269 		type = BTRFS_EXTENT_DATA_REF_KEY;
7270 
7271 	size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7272 
7273 	path = btrfs_alloc_path();
7274 	if (!path)
7275 		return -ENOMEM;
7276 
7277 	path->leave_spinning = 1;
7278 	ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7279 				      ins, size);
7280 	if (ret) {
7281 		btrfs_free_path(path);
7282 		return ret;
7283 	}
7284 
7285 	leaf = path->nodes[0];
7286 	extent_item = btrfs_item_ptr(leaf, path->slots[0],
7287 				     struct btrfs_extent_item);
7288 	btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7289 	btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7290 	btrfs_set_extent_flags(leaf, extent_item,
7291 			       flags | BTRFS_EXTENT_FLAG_DATA);
7292 
7293 	iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7294 	btrfs_set_extent_inline_ref_type(leaf, iref, type);
7295 	if (parent > 0) {
7296 		struct btrfs_shared_data_ref *ref;
7297 		ref = (struct btrfs_shared_data_ref *)(iref + 1);
7298 		btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7299 		btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7300 	} else {
7301 		struct btrfs_extent_data_ref *ref;
7302 		ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7303 		btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7304 		btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7305 		btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7306 		btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7307 	}
7308 
7309 	btrfs_mark_buffer_dirty(path->nodes[0]);
7310 	btrfs_free_path(path);
7311 
7312 	/* Always set parent to 0 here since its exclusive anyway. */
7313 	ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7314 				      ins->objectid, ins->offset,
7315 				      BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7316 	if (ret)
7317 		return ret;
7318 
7319 	ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7320 	if (ret) { /* -ENOENT, logic error */
7321 		btrfs_err(fs_info, "update block group failed for %llu %llu",
7322 			ins->objectid, ins->offset);
7323 		BUG();
7324 	}
7325 	trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7326 	return ret;
7327 }
7328 
alloc_reserved_tree_block(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 parent,u64 root_objectid,u64 flags,struct btrfs_disk_key * key,int level,struct btrfs_key * ins,int no_quota)7329 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7330 				     struct btrfs_root *root,
7331 				     u64 parent, u64 root_objectid,
7332 				     u64 flags, struct btrfs_disk_key *key,
7333 				     int level, struct btrfs_key *ins,
7334 				     int no_quota)
7335 {
7336 	int ret;
7337 	struct btrfs_fs_info *fs_info = root->fs_info;
7338 	struct btrfs_extent_item *extent_item;
7339 	struct btrfs_tree_block_info *block_info;
7340 	struct btrfs_extent_inline_ref *iref;
7341 	struct btrfs_path *path;
7342 	struct extent_buffer *leaf;
7343 	u32 size = sizeof(*extent_item) + sizeof(*iref);
7344 	u64 num_bytes = ins->offset;
7345 	bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7346 						 SKINNY_METADATA);
7347 
7348 	if (!skinny_metadata)
7349 		size += sizeof(*block_info);
7350 
7351 	path = btrfs_alloc_path();
7352 	if (!path) {
7353 		btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7354 						   root->nodesize);
7355 		return -ENOMEM;
7356 	}
7357 
7358 	path->leave_spinning = 1;
7359 	ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7360 				      ins, size);
7361 	if (ret) {
7362 		btrfs_free_path(path);
7363 		btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7364 						   root->nodesize);
7365 		return ret;
7366 	}
7367 
7368 	leaf = path->nodes[0];
7369 	extent_item = btrfs_item_ptr(leaf, path->slots[0],
7370 				     struct btrfs_extent_item);
7371 	btrfs_set_extent_refs(leaf, extent_item, 1);
7372 	btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7373 	btrfs_set_extent_flags(leaf, extent_item,
7374 			       flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7375 
7376 	if (skinny_metadata) {
7377 		iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7378 		num_bytes = root->nodesize;
7379 	} else {
7380 		block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7381 		btrfs_set_tree_block_key(leaf, block_info, key);
7382 		btrfs_set_tree_block_level(leaf, block_info, level);
7383 		iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7384 	}
7385 
7386 	if (parent > 0) {
7387 		BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7388 		btrfs_set_extent_inline_ref_type(leaf, iref,
7389 						 BTRFS_SHARED_BLOCK_REF_KEY);
7390 		btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7391 	} else {
7392 		btrfs_set_extent_inline_ref_type(leaf, iref,
7393 						 BTRFS_TREE_BLOCK_REF_KEY);
7394 		btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7395 	}
7396 
7397 	btrfs_mark_buffer_dirty(leaf);
7398 	btrfs_free_path(path);
7399 
7400 	if (!no_quota) {
7401 		ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7402 					      ins->objectid, num_bytes,
7403 					      BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7404 		if (ret)
7405 			return ret;
7406 	}
7407 
7408 	ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7409 				 1);
7410 	if (ret) { /* -ENOENT, logic error */
7411 		btrfs_err(fs_info, "update block group failed for %llu %llu",
7412 			ins->objectid, ins->offset);
7413 		BUG();
7414 	}
7415 
7416 	trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7417 	return ret;
7418 }
7419 
btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 root_objectid,u64 owner,u64 offset,struct btrfs_key * ins)7420 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7421 				     struct btrfs_root *root,
7422 				     u64 root_objectid, u64 owner,
7423 				     u64 offset, struct btrfs_key *ins)
7424 {
7425 	int ret;
7426 
7427 	BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7428 
7429 	ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7430 					 ins->offset, 0,
7431 					 root_objectid, owner, offset,
7432 					 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7433 	return ret;
7434 }
7435 
7436 /*
7437  * this is used by the tree logging recovery code.  It records that
7438  * an extent has been allocated and makes sure to clear the free
7439  * space cache bits as well
7440  */
btrfs_alloc_logged_file_extent(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 root_objectid,u64 owner,u64 offset,struct btrfs_key * ins)7441 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7442 				   struct btrfs_root *root,
7443 				   u64 root_objectid, u64 owner, u64 offset,
7444 				   struct btrfs_key *ins)
7445 {
7446 	int ret;
7447 	struct btrfs_block_group_cache *block_group;
7448 
7449 	/*
7450 	 * Mixed block groups will exclude before processing the log so we only
7451 	 * need to do the exlude dance if this fs isn't mixed.
7452 	 */
7453 	if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7454 		ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7455 		if (ret)
7456 			return ret;
7457 	}
7458 
7459 	block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7460 	if (!block_group)
7461 		return -EINVAL;
7462 
7463 	ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7464 					  RESERVE_ALLOC_NO_ACCOUNT, 0);
7465 	BUG_ON(ret); /* logic error */
7466 	ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7467 					 0, owner, offset, ins, 1);
7468 	btrfs_put_block_group(block_group);
7469 	return ret;
7470 }
7471 
7472 static struct extent_buffer *
btrfs_init_new_buffer(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 bytenr,int level)7473 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7474 		      u64 bytenr, int level)
7475 {
7476 	struct extent_buffer *buf;
7477 
7478 	buf = btrfs_find_create_tree_block(root, bytenr);
7479 	if (!buf)
7480 		return ERR_PTR(-ENOMEM);
7481 	btrfs_set_header_generation(buf, trans->transid);
7482 	btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7483 	btrfs_tree_lock(buf);
7484 	clean_tree_block(trans, root->fs_info, buf);
7485 	clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7486 
7487 	btrfs_set_lock_blocking(buf);
7488 	btrfs_set_buffer_uptodate(buf);
7489 
7490 	if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7491 		buf->log_index = root->log_transid % 2;
7492 		/*
7493 		 * we allow two log transactions at a time, use different
7494 		 * EXENT bit to differentiate dirty pages.
7495 		 */
7496 		if (buf->log_index == 0)
7497 			set_extent_dirty(&root->dirty_log_pages, buf->start,
7498 					buf->start + buf->len - 1, GFP_NOFS);
7499 		else
7500 			set_extent_new(&root->dirty_log_pages, buf->start,
7501 					buf->start + buf->len - 1, GFP_NOFS);
7502 	} else {
7503 		buf->log_index = -1;
7504 		set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7505 			 buf->start + buf->len - 1, GFP_NOFS);
7506 	}
7507 	trans->blocks_used++;
7508 	/* this returns a buffer locked for blocking */
7509 	return buf;
7510 }
7511 
7512 static struct btrfs_block_rsv *
use_block_rsv(struct btrfs_trans_handle * trans,struct btrfs_root * root,u32 blocksize)7513 use_block_rsv(struct btrfs_trans_handle *trans,
7514 	      struct btrfs_root *root, u32 blocksize)
7515 {
7516 	struct btrfs_block_rsv *block_rsv;
7517 	struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7518 	int ret;
7519 	bool global_updated = false;
7520 
7521 	block_rsv = get_block_rsv(trans, root);
7522 
7523 	if (unlikely(block_rsv->size == 0))
7524 		goto try_reserve;
7525 again:
7526 	ret = block_rsv_use_bytes(block_rsv, blocksize);
7527 	if (!ret)
7528 		return block_rsv;
7529 
7530 	if (block_rsv->failfast)
7531 		return ERR_PTR(ret);
7532 
7533 	if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7534 		global_updated = true;
7535 		update_global_block_rsv(root->fs_info);
7536 		goto again;
7537 	}
7538 
7539 	if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7540 		static DEFINE_RATELIMIT_STATE(_rs,
7541 				DEFAULT_RATELIMIT_INTERVAL * 10,
7542 				/*DEFAULT_RATELIMIT_BURST*/ 1);
7543 		if (__ratelimit(&_rs))
7544 			WARN(1, KERN_DEBUG
7545 				"BTRFS: block rsv returned %d\n", ret);
7546 	}
7547 try_reserve:
7548 	ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7549 				     BTRFS_RESERVE_NO_FLUSH);
7550 	if (!ret)
7551 		return block_rsv;
7552 	/*
7553 	 * If we couldn't reserve metadata bytes try and use some from
7554 	 * the global reserve if its space type is the same as the global
7555 	 * reservation.
7556 	 */
7557 	if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7558 	    block_rsv->space_info == global_rsv->space_info) {
7559 		ret = block_rsv_use_bytes(global_rsv, blocksize);
7560 		if (!ret)
7561 			return global_rsv;
7562 	}
7563 	return ERR_PTR(ret);
7564 }
7565 
unuse_block_rsv(struct btrfs_fs_info * fs_info,struct btrfs_block_rsv * block_rsv,u32 blocksize)7566 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7567 			    struct btrfs_block_rsv *block_rsv, u32 blocksize)
7568 {
7569 	block_rsv_add_bytes(block_rsv, blocksize, 0);
7570 	block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7571 }
7572 
7573 /*
7574  * finds a free extent and does all the dirty work required for allocation
7575  * returns the key for the extent through ins, and a tree buffer for
7576  * the first block of the extent through buf.
7577  *
7578  * returns the tree buffer or an ERR_PTR on error.
7579  */
btrfs_alloc_tree_block(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 parent,u64 root_objectid,struct btrfs_disk_key * key,int level,u64 hint,u64 empty_size)7580 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7581 					struct btrfs_root *root,
7582 					u64 parent, u64 root_objectid,
7583 					struct btrfs_disk_key *key, int level,
7584 					u64 hint, u64 empty_size)
7585 {
7586 	struct btrfs_key ins;
7587 	struct btrfs_block_rsv *block_rsv;
7588 	struct extent_buffer *buf;
7589 	struct btrfs_delayed_extent_op *extent_op;
7590 	u64 flags = 0;
7591 	int ret;
7592 	u32 blocksize = root->nodesize;
7593 	bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7594 						 SKINNY_METADATA);
7595 
7596 	if (btrfs_test_is_dummy_root(root)) {
7597 		buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7598 					    level);
7599 		if (!IS_ERR(buf))
7600 			root->alloc_bytenr += blocksize;
7601 		return buf;
7602 	}
7603 
7604 	block_rsv = use_block_rsv(trans, root, blocksize);
7605 	if (IS_ERR(block_rsv))
7606 		return ERR_CAST(block_rsv);
7607 
7608 	ret = btrfs_reserve_extent(root, blocksize, blocksize,
7609 				   empty_size, hint, &ins, 0, 0);
7610 	if (ret)
7611 		goto out_unuse;
7612 
7613 	buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7614 	if (IS_ERR(buf)) {
7615 		ret = PTR_ERR(buf);
7616 		goto out_free_reserved;
7617 	}
7618 
7619 	if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7620 		if (parent == 0)
7621 			parent = ins.objectid;
7622 		flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7623 	} else
7624 		BUG_ON(parent > 0);
7625 
7626 	if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7627 		extent_op = btrfs_alloc_delayed_extent_op();
7628 		if (!extent_op) {
7629 			ret = -ENOMEM;
7630 			goto out_free_buf;
7631 		}
7632 		if (key)
7633 			memcpy(&extent_op->key, key, sizeof(extent_op->key));
7634 		else
7635 			memset(&extent_op->key, 0, sizeof(extent_op->key));
7636 		extent_op->flags_to_set = flags;
7637 		if (skinny_metadata)
7638 			extent_op->update_key = 0;
7639 		else
7640 			extent_op->update_key = 1;
7641 		extent_op->update_flags = 1;
7642 		extent_op->is_data = 0;
7643 		extent_op->level = level;
7644 
7645 		ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7646 						 ins.objectid, ins.offset,
7647 						 parent, root_objectid, level,
7648 						 BTRFS_ADD_DELAYED_EXTENT,
7649 						 extent_op, 0);
7650 		if (ret)
7651 			goto out_free_delayed;
7652 	}
7653 	return buf;
7654 
7655 out_free_delayed:
7656 	btrfs_free_delayed_extent_op(extent_op);
7657 out_free_buf:
7658 	free_extent_buffer(buf);
7659 out_free_reserved:
7660 	btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
7661 out_unuse:
7662 	unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7663 	return ERR_PTR(ret);
7664 }
7665 
7666 struct walk_control {
7667 	u64 refs[BTRFS_MAX_LEVEL];
7668 	u64 flags[BTRFS_MAX_LEVEL];
7669 	struct btrfs_key update_progress;
7670 	int stage;
7671 	int level;
7672 	int shared_level;
7673 	int update_ref;
7674 	int keep_locks;
7675 	int reada_slot;
7676 	int reada_count;
7677 	int for_reloc;
7678 };
7679 
7680 #define DROP_REFERENCE	1
7681 #define UPDATE_BACKREF	2
7682 
reada_walk_down(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct walk_control * wc,struct btrfs_path * path)7683 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7684 				     struct btrfs_root *root,
7685 				     struct walk_control *wc,
7686 				     struct btrfs_path *path)
7687 {
7688 	u64 bytenr;
7689 	u64 generation;
7690 	u64 refs;
7691 	u64 flags;
7692 	u32 nritems;
7693 	u32 blocksize;
7694 	struct btrfs_key key;
7695 	struct extent_buffer *eb;
7696 	int ret;
7697 	int slot;
7698 	int nread = 0;
7699 
7700 	if (path->slots[wc->level] < wc->reada_slot) {
7701 		wc->reada_count = wc->reada_count * 2 / 3;
7702 		wc->reada_count = max(wc->reada_count, 2);
7703 	} else {
7704 		wc->reada_count = wc->reada_count * 3 / 2;
7705 		wc->reada_count = min_t(int, wc->reada_count,
7706 					BTRFS_NODEPTRS_PER_BLOCK(root));
7707 	}
7708 
7709 	eb = path->nodes[wc->level];
7710 	nritems = btrfs_header_nritems(eb);
7711 	blocksize = root->nodesize;
7712 
7713 	for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7714 		if (nread >= wc->reada_count)
7715 			break;
7716 
7717 		cond_resched();
7718 		bytenr = btrfs_node_blockptr(eb, slot);
7719 		generation = btrfs_node_ptr_generation(eb, slot);
7720 
7721 		if (slot == path->slots[wc->level])
7722 			goto reada;
7723 
7724 		if (wc->stage == UPDATE_BACKREF &&
7725 		    generation <= root->root_key.offset)
7726 			continue;
7727 
7728 		/* We don't lock the tree block, it's OK to be racy here */
7729 		ret = btrfs_lookup_extent_info(trans, root, bytenr,
7730 					       wc->level - 1, 1, &refs,
7731 					       &flags);
7732 		/* We don't care about errors in readahead. */
7733 		if (ret < 0)
7734 			continue;
7735 		BUG_ON(refs == 0);
7736 
7737 		if (wc->stage == DROP_REFERENCE) {
7738 			if (refs == 1)
7739 				goto reada;
7740 
7741 			if (wc->level == 1 &&
7742 			    (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7743 				continue;
7744 			if (!wc->update_ref ||
7745 			    generation <= root->root_key.offset)
7746 				continue;
7747 			btrfs_node_key_to_cpu(eb, &key, slot);
7748 			ret = btrfs_comp_cpu_keys(&key,
7749 						  &wc->update_progress);
7750 			if (ret < 0)
7751 				continue;
7752 		} else {
7753 			if (wc->level == 1 &&
7754 			    (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7755 				continue;
7756 		}
7757 reada:
7758 		readahead_tree_block(root, bytenr);
7759 		nread++;
7760 	}
7761 	wc->reada_slot = slot;
7762 }
7763 
account_leaf_items(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * eb)7764 static int account_leaf_items(struct btrfs_trans_handle *trans,
7765 			      struct btrfs_root *root,
7766 			      struct extent_buffer *eb)
7767 {
7768 	int nr = btrfs_header_nritems(eb);
7769 	int i, extent_type, ret;
7770 	struct btrfs_key key;
7771 	struct btrfs_file_extent_item *fi;
7772 	u64 bytenr, num_bytes;
7773 
7774 	for (i = 0; i < nr; i++) {
7775 		btrfs_item_key_to_cpu(eb, &key, i);
7776 
7777 		if (key.type != BTRFS_EXTENT_DATA_KEY)
7778 			continue;
7779 
7780 		fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7781 		/* filter out non qgroup-accountable extents  */
7782 		extent_type = btrfs_file_extent_type(eb, fi);
7783 
7784 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7785 			continue;
7786 
7787 		bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7788 		if (!bytenr)
7789 			continue;
7790 
7791 		num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7792 
7793 		ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7794 					      root->objectid,
7795 					      bytenr, num_bytes,
7796 					      BTRFS_QGROUP_OPER_SUB_SUBTREE, 0);
7797 		if (ret)
7798 			return ret;
7799 	}
7800 	return 0;
7801 }
7802 
7803 /*
7804  * Walk up the tree from the bottom, freeing leaves and any interior
7805  * nodes which have had all slots visited. If a node (leaf or
7806  * interior) is freed, the node above it will have it's slot
7807  * incremented. The root node will never be freed.
7808  *
7809  * At the end of this function, we should have a path which has all
7810  * slots incremented to the next position for a search. If we need to
7811  * read a new node it will be NULL and the node above it will have the
7812  * correct slot selected for a later read.
7813  *
7814  * If we increment the root nodes slot counter past the number of
7815  * elements, 1 is returned to signal completion of the search.
7816  */
adjust_slots_upwards(struct btrfs_root * root,struct btrfs_path * path,int root_level)7817 static int adjust_slots_upwards(struct btrfs_root *root,
7818 				struct btrfs_path *path, int root_level)
7819 {
7820 	int level = 0;
7821 	int nr, slot;
7822 	struct extent_buffer *eb;
7823 
7824 	if (root_level == 0)
7825 		return 1;
7826 
7827 	while (level <= root_level) {
7828 		eb = path->nodes[level];
7829 		nr = btrfs_header_nritems(eb);
7830 		path->slots[level]++;
7831 		slot = path->slots[level];
7832 		if (slot >= nr || level == 0) {
7833 			/*
7834 			 * Don't free the root -  we will detect this
7835 			 * condition after our loop and return a
7836 			 * positive value for caller to stop walking the tree.
7837 			 */
7838 			if (level != root_level) {
7839 				btrfs_tree_unlock_rw(eb, path->locks[level]);
7840 				path->locks[level] = 0;
7841 
7842 				free_extent_buffer(eb);
7843 				path->nodes[level] = NULL;
7844 				path->slots[level] = 0;
7845 			}
7846 		} else {
7847 			/*
7848 			 * We have a valid slot to walk back down
7849 			 * from. Stop here so caller can process these
7850 			 * new nodes.
7851 			 */
7852 			break;
7853 		}
7854 
7855 		level++;
7856 	}
7857 
7858 	eb = path->nodes[root_level];
7859 	if (path->slots[root_level] >= btrfs_header_nritems(eb))
7860 		return 1;
7861 
7862 	return 0;
7863 }
7864 
7865 /*
7866  * root_eb is the subtree root and is locked before this function is called.
7867  */
account_shared_subtree(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * root_eb,u64 root_gen,int root_level)7868 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7869 				  struct btrfs_root *root,
7870 				  struct extent_buffer *root_eb,
7871 				  u64 root_gen,
7872 				  int root_level)
7873 {
7874 	int ret = 0;
7875 	int level;
7876 	struct extent_buffer *eb = root_eb;
7877 	struct btrfs_path *path = NULL;
7878 
7879 	BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7880 	BUG_ON(root_eb == NULL);
7881 
7882 	if (!root->fs_info->quota_enabled)
7883 		return 0;
7884 
7885 	if (!extent_buffer_uptodate(root_eb)) {
7886 		ret = btrfs_read_buffer(root_eb, root_gen);
7887 		if (ret)
7888 			goto out;
7889 	}
7890 
7891 	if (root_level == 0) {
7892 		ret = account_leaf_items(trans, root, root_eb);
7893 		goto out;
7894 	}
7895 
7896 	path = btrfs_alloc_path();
7897 	if (!path)
7898 		return -ENOMEM;
7899 
7900 	/*
7901 	 * Walk down the tree.  Missing extent blocks are filled in as
7902 	 * we go. Metadata is accounted every time we read a new
7903 	 * extent block.
7904 	 *
7905 	 * When we reach a leaf, we account for file extent items in it,
7906 	 * walk back up the tree (adjusting slot pointers as we go)
7907 	 * and restart the search process.
7908 	 */
7909 	extent_buffer_get(root_eb); /* For path */
7910 	path->nodes[root_level] = root_eb;
7911 	path->slots[root_level] = 0;
7912 	path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
7913 walk_down:
7914 	level = root_level;
7915 	while (level >= 0) {
7916 		if (path->nodes[level] == NULL) {
7917 			int parent_slot;
7918 			u64 child_gen;
7919 			u64 child_bytenr;
7920 
7921 			/* We need to get child blockptr/gen from
7922 			 * parent before we can read it. */
7923 			eb = path->nodes[level + 1];
7924 			parent_slot = path->slots[level + 1];
7925 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
7926 			child_gen = btrfs_node_ptr_generation(eb, parent_slot);
7927 
7928 			eb = read_tree_block(root, child_bytenr, child_gen);
7929 			if (!eb || !extent_buffer_uptodate(eb)) {
7930 				ret = -EIO;
7931 				goto out;
7932 			}
7933 
7934 			path->nodes[level] = eb;
7935 			path->slots[level] = 0;
7936 
7937 			btrfs_tree_read_lock(eb);
7938 			btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
7939 			path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
7940 
7941 			ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7942 						root->objectid,
7943 						child_bytenr,
7944 						root->nodesize,
7945 						BTRFS_QGROUP_OPER_SUB_SUBTREE,
7946 						0);
7947 			if (ret)
7948 				goto out;
7949 
7950 		}
7951 
7952 		if (level == 0) {
7953 			ret = account_leaf_items(trans, root, path->nodes[level]);
7954 			if (ret)
7955 				goto out;
7956 
7957 			/* Nonzero return here means we completed our search */
7958 			ret = adjust_slots_upwards(root, path, root_level);
7959 			if (ret)
7960 				break;
7961 
7962 			/* Restart search with new slots */
7963 			goto walk_down;
7964 		}
7965 
7966 		level--;
7967 	}
7968 
7969 	ret = 0;
7970 out:
7971 	btrfs_free_path(path);
7972 
7973 	return ret;
7974 }
7975 
7976 /*
7977  * helper to process tree block while walking down the tree.
7978  *
7979  * when wc->stage == UPDATE_BACKREF, this function updates
7980  * back refs for pointers in the block.
7981  *
7982  * NOTE: return value 1 means we should stop walking down.
7983  */
walk_down_proc(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct walk_control * wc,int lookup_info)7984 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7985 				   struct btrfs_root *root,
7986 				   struct btrfs_path *path,
7987 				   struct walk_control *wc, int lookup_info)
7988 {
7989 	int level = wc->level;
7990 	struct extent_buffer *eb = path->nodes[level];
7991 	u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7992 	int ret;
7993 
7994 	if (wc->stage == UPDATE_BACKREF &&
7995 	    btrfs_header_owner(eb) != root->root_key.objectid)
7996 		return 1;
7997 
7998 	/*
7999 	 * when reference count of tree block is 1, it won't increase
8000 	 * again. once full backref flag is set, we never clear it.
8001 	 */
8002 	if (lookup_info &&
8003 	    ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8004 	     (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8005 		BUG_ON(!path->locks[level]);
8006 		ret = btrfs_lookup_extent_info(trans, root,
8007 					       eb->start, level, 1,
8008 					       &wc->refs[level],
8009 					       &wc->flags[level]);
8010 		BUG_ON(ret == -ENOMEM);
8011 		if (ret)
8012 			return ret;
8013 		BUG_ON(wc->refs[level] == 0);
8014 	}
8015 
8016 	if (wc->stage == DROP_REFERENCE) {
8017 		if (wc->refs[level] > 1)
8018 			return 1;
8019 
8020 		if (path->locks[level] && !wc->keep_locks) {
8021 			btrfs_tree_unlock_rw(eb, path->locks[level]);
8022 			path->locks[level] = 0;
8023 		}
8024 		return 0;
8025 	}
8026 
8027 	/* wc->stage == UPDATE_BACKREF */
8028 	if (!(wc->flags[level] & flag)) {
8029 		BUG_ON(!path->locks[level]);
8030 		ret = btrfs_inc_ref(trans, root, eb, 1);
8031 		BUG_ON(ret); /* -ENOMEM */
8032 		ret = btrfs_dec_ref(trans, root, eb, 0);
8033 		BUG_ON(ret); /* -ENOMEM */
8034 		ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8035 						  eb->len, flag,
8036 						  btrfs_header_level(eb), 0);
8037 		BUG_ON(ret); /* -ENOMEM */
8038 		wc->flags[level] |= flag;
8039 	}
8040 
8041 	/*
8042 	 * the block is shared by multiple trees, so it's not good to
8043 	 * keep the tree lock
8044 	 */
8045 	if (path->locks[level] && level > 0) {
8046 		btrfs_tree_unlock_rw(eb, path->locks[level]);
8047 		path->locks[level] = 0;
8048 	}
8049 	return 0;
8050 }
8051 
8052 /*
8053  * helper to process tree block pointer.
8054  *
8055  * when wc->stage == DROP_REFERENCE, this function checks
8056  * reference count of the block pointed to. if the block
8057  * is shared and we need update back refs for the subtree
8058  * rooted at the block, this function changes wc->stage to
8059  * UPDATE_BACKREF. if the block is shared and there is no
8060  * need to update back, this function drops the reference
8061  * to the block.
8062  *
8063  * NOTE: return value 1 means we should stop walking down.
8064  */
do_walk_down(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct walk_control * wc,int * lookup_info)8065 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8066 				 struct btrfs_root *root,
8067 				 struct btrfs_path *path,
8068 				 struct walk_control *wc, int *lookup_info)
8069 {
8070 	u64 bytenr;
8071 	u64 generation;
8072 	u64 parent;
8073 	u32 blocksize;
8074 	struct btrfs_key key;
8075 	struct extent_buffer *next;
8076 	int level = wc->level;
8077 	int reada = 0;
8078 	int ret = 0;
8079 	bool need_account = false;
8080 
8081 	generation = btrfs_node_ptr_generation(path->nodes[level],
8082 					       path->slots[level]);
8083 	/*
8084 	 * if the lower level block was created before the snapshot
8085 	 * was created, we know there is no need to update back refs
8086 	 * for the subtree
8087 	 */
8088 	if (wc->stage == UPDATE_BACKREF &&
8089 	    generation <= root->root_key.offset) {
8090 		*lookup_info = 1;
8091 		return 1;
8092 	}
8093 
8094 	bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8095 	blocksize = root->nodesize;
8096 
8097 	next = btrfs_find_tree_block(root->fs_info, bytenr);
8098 	if (!next) {
8099 		next = btrfs_find_create_tree_block(root, bytenr);
8100 		if (!next)
8101 			return -ENOMEM;
8102 		btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8103 					       level - 1);
8104 		reada = 1;
8105 	}
8106 	btrfs_tree_lock(next);
8107 	btrfs_set_lock_blocking(next);
8108 
8109 	ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8110 				       &wc->refs[level - 1],
8111 				       &wc->flags[level - 1]);
8112 	if (ret < 0) {
8113 		btrfs_tree_unlock(next);
8114 		return ret;
8115 	}
8116 
8117 	if (unlikely(wc->refs[level - 1] == 0)) {
8118 		btrfs_err(root->fs_info, "Missing references.");
8119 		BUG();
8120 	}
8121 	*lookup_info = 0;
8122 
8123 	if (wc->stage == DROP_REFERENCE) {
8124 		if (wc->refs[level - 1] > 1) {
8125 			need_account = true;
8126 			if (level == 1 &&
8127 			    (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8128 				goto skip;
8129 
8130 			if (!wc->update_ref ||
8131 			    generation <= root->root_key.offset)
8132 				goto skip;
8133 
8134 			btrfs_node_key_to_cpu(path->nodes[level], &key,
8135 					      path->slots[level]);
8136 			ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8137 			if (ret < 0)
8138 				goto skip;
8139 
8140 			wc->stage = UPDATE_BACKREF;
8141 			wc->shared_level = level - 1;
8142 		}
8143 	} else {
8144 		if (level == 1 &&
8145 		    (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8146 			goto skip;
8147 	}
8148 
8149 	if (!btrfs_buffer_uptodate(next, generation, 0)) {
8150 		btrfs_tree_unlock(next);
8151 		free_extent_buffer(next);
8152 		next = NULL;
8153 		*lookup_info = 1;
8154 	}
8155 
8156 	if (!next) {
8157 		if (reada && level == 1)
8158 			reada_walk_down(trans, root, wc, path);
8159 		next = read_tree_block(root, bytenr, generation);
8160 		if (!next || !extent_buffer_uptodate(next)) {
8161 			free_extent_buffer(next);
8162 			return -EIO;
8163 		}
8164 		btrfs_tree_lock(next);
8165 		btrfs_set_lock_blocking(next);
8166 	}
8167 
8168 	level--;
8169 	BUG_ON(level != btrfs_header_level(next));
8170 	path->nodes[level] = next;
8171 	path->slots[level] = 0;
8172 	path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8173 	wc->level = level;
8174 	if (wc->level == 1)
8175 		wc->reada_slot = 0;
8176 	return 0;
8177 skip:
8178 	wc->refs[level - 1] = 0;
8179 	wc->flags[level - 1] = 0;
8180 	if (wc->stage == DROP_REFERENCE) {
8181 		if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8182 			parent = path->nodes[level]->start;
8183 		} else {
8184 			BUG_ON(root->root_key.objectid !=
8185 			       btrfs_header_owner(path->nodes[level]));
8186 			parent = 0;
8187 		}
8188 
8189 		if (need_account) {
8190 			ret = account_shared_subtree(trans, root, next,
8191 						     generation, level - 1);
8192 			if (ret) {
8193 				printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8194 					"%d accounting shared subtree. Quota "
8195 					"is out of sync, rescan required.\n",
8196 					root->fs_info->sb->s_id, ret);
8197 			}
8198 		}
8199 		ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8200 				root->root_key.objectid, level - 1, 0, 0);
8201 		BUG_ON(ret); /* -ENOMEM */
8202 	}
8203 	btrfs_tree_unlock(next);
8204 	free_extent_buffer(next);
8205 	*lookup_info = 1;
8206 	return 1;
8207 }
8208 
8209 /*
8210  * helper to process tree block while walking up the tree.
8211  *
8212  * when wc->stage == DROP_REFERENCE, this function drops
8213  * reference count on the block.
8214  *
8215  * when wc->stage == UPDATE_BACKREF, this function changes
8216  * wc->stage back to DROP_REFERENCE if we changed wc->stage
8217  * to UPDATE_BACKREF previously while processing the block.
8218  *
8219  * NOTE: return value 1 means we should stop walking up.
8220  */
walk_up_proc(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct walk_control * wc)8221 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8222 				 struct btrfs_root *root,
8223 				 struct btrfs_path *path,
8224 				 struct walk_control *wc)
8225 {
8226 	int ret;
8227 	int level = wc->level;
8228 	struct extent_buffer *eb = path->nodes[level];
8229 	u64 parent = 0;
8230 
8231 	if (wc->stage == UPDATE_BACKREF) {
8232 		BUG_ON(wc->shared_level < level);
8233 		if (level < wc->shared_level)
8234 			goto out;
8235 
8236 		ret = find_next_key(path, level + 1, &wc->update_progress);
8237 		if (ret > 0)
8238 			wc->update_ref = 0;
8239 
8240 		wc->stage = DROP_REFERENCE;
8241 		wc->shared_level = -1;
8242 		path->slots[level] = 0;
8243 
8244 		/*
8245 		 * check reference count again if the block isn't locked.
8246 		 * we should start walking down the tree again if reference
8247 		 * count is one.
8248 		 */
8249 		if (!path->locks[level]) {
8250 			BUG_ON(level == 0);
8251 			btrfs_tree_lock(eb);
8252 			btrfs_set_lock_blocking(eb);
8253 			path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8254 
8255 			ret = btrfs_lookup_extent_info(trans, root,
8256 						       eb->start, level, 1,
8257 						       &wc->refs[level],
8258 						       &wc->flags[level]);
8259 			if (ret < 0) {
8260 				btrfs_tree_unlock_rw(eb, path->locks[level]);
8261 				path->locks[level] = 0;
8262 				return ret;
8263 			}
8264 			BUG_ON(wc->refs[level] == 0);
8265 			if (wc->refs[level] == 1) {
8266 				btrfs_tree_unlock_rw(eb, path->locks[level]);
8267 				path->locks[level] = 0;
8268 				return 1;
8269 			}
8270 		}
8271 	}
8272 
8273 	/* wc->stage == DROP_REFERENCE */
8274 	BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8275 
8276 	if (wc->refs[level] == 1) {
8277 		if (level == 0) {
8278 			if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8279 				ret = btrfs_dec_ref(trans, root, eb, 1);
8280 			else
8281 				ret = btrfs_dec_ref(trans, root, eb, 0);
8282 			BUG_ON(ret); /* -ENOMEM */
8283 			ret = account_leaf_items(trans, root, eb);
8284 			if (ret) {
8285 				printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8286 					"%d accounting leaf items. Quota "
8287 					"is out of sync, rescan required.\n",
8288 					root->fs_info->sb->s_id, ret);
8289 			}
8290 		}
8291 		/* make block locked assertion in clean_tree_block happy */
8292 		if (!path->locks[level] &&
8293 		    btrfs_header_generation(eb) == trans->transid) {
8294 			btrfs_tree_lock(eb);
8295 			btrfs_set_lock_blocking(eb);
8296 			path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8297 		}
8298 		clean_tree_block(trans, root->fs_info, eb);
8299 	}
8300 
8301 	if (eb == root->node) {
8302 		if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8303 			parent = eb->start;
8304 		else
8305 			BUG_ON(root->root_key.objectid !=
8306 			       btrfs_header_owner(eb));
8307 	} else {
8308 		if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8309 			parent = path->nodes[level + 1]->start;
8310 		else
8311 			BUG_ON(root->root_key.objectid !=
8312 			       btrfs_header_owner(path->nodes[level + 1]));
8313 	}
8314 
8315 	btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8316 out:
8317 	wc->refs[level] = 0;
8318 	wc->flags[level] = 0;
8319 	return 0;
8320 }
8321 
walk_down_tree(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct walk_control * wc)8322 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8323 				   struct btrfs_root *root,
8324 				   struct btrfs_path *path,
8325 				   struct walk_control *wc)
8326 {
8327 	int level = wc->level;
8328 	int lookup_info = 1;
8329 	int ret;
8330 
8331 	while (level >= 0) {
8332 		ret = walk_down_proc(trans, root, path, wc, lookup_info);
8333 		if (ret > 0)
8334 			break;
8335 
8336 		if (level == 0)
8337 			break;
8338 
8339 		if (path->slots[level] >=
8340 		    btrfs_header_nritems(path->nodes[level]))
8341 			break;
8342 
8343 		ret = do_walk_down(trans, root, path, wc, &lookup_info);
8344 		if (ret > 0) {
8345 			path->slots[level]++;
8346 			continue;
8347 		} else if (ret < 0)
8348 			return ret;
8349 		level = wc->level;
8350 	}
8351 	return 0;
8352 }
8353 
walk_up_tree(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct walk_control * wc,int max_level)8354 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8355 				 struct btrfs_root *root,
8356 				 struct btrfs_path *path,
8357 				 struct walk_control *wc, int max_level)
8358 {
8359 	int level = wc->level;
8360 	int ret;
8361 
8362 	path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8363 	while (level < max_level && path->nodes[level]) {
8364 		wc->level = level;
8365 		if (path->slots[level] + 1 <
8366 		    btrfs_header_nritems(path->nodes[level])) {
8367 			path->slots[level]++;
8368 			return 0;
8369 		} else {
8370 			ret = walk_up_proc(trans, root, path, wc);
8371 			if (ret > 0)
8372 				return 0;
8373 
8374 			if (path->locks[level]) {
8375 				btrfs_tree_unlock_rw(path->nodes[level],
8376 						     path->locks[level]);
8377 				path->locks[level] = 0;
8378 			}
8379 			free_extent_buffer(path->nodes[level]);
8380 			path->nodes[level] = NULL;
8381 			level++;
8382 		}
8383 	}
8384 	return 1;
8385 }
8386 
8387 /*
8388  * drop a subvolume tree.
8389  *
8390  * this function traverses the tree freeing any blocks that only
8391  * referenced by the tree.
8392  *
8393  * when a shared tree block is found. this function decreases its
8394  * reference count by one. if update_ref is true, this function
8395  * also make sure backrefs for the shared block and all lower level
8396  * blocks are properly updated.
8397  *
8398  * If called with for_reloc == 0, may exit early with -EAGAIN
8399  */
btrfs_drop_snapshot(struct btrfs_root * root,struct btrfs_block_rsv * block_rsv,int update_ref,int for_reloc)8400 int btrfs_drop_snapshot(struct btrfs_root *root,
8401 			 struct btrfs_block_rsv *block_rsv, int update_ref,
8402 			 int for_reloc)
8403 {
8404 	struct btrfs_path *path;
8405 	struct btrfs_trans_handle *trans;
8406 	struct btrfs_root *tree_root = root->fs_info->tree_root;
8407 	struct btrfs_root_item *root_item = &root->root_item;
8408 	struct walk_control *wc;
8409 	struct btrfs_key key;
8410 	int err = 0;
8411 	int ret;
8412 	int level;
8413 	bool root_dropped = false;
8414 
8415 	btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8416 
8417 	path = btrfs_alloc_path();
8418 	if (!path) {
8419 		err = -ENOMEM;
8420 		goto out;
8421 	}
8422 
8423 	wc = kzalloc(sizeof(*wc), GFP_NOFS);
8424 	if (!wc) {
8425 		btrfs_free_path(path);
8426 		err = -ENOMEM;
8427 		goto out;
8428 	}
8429 
8430 	trans = btrfs_start_transaction(tree_root, 0);
8431 	if (IS_ERR(trans)) {
8432 		err = PTR_ERR(trans);
8433 		goto out_free;
8434 	}
8435 
8436 	if (block_rsv)
8437 		trans->block_rsv = block_rsv;
8438 
8439 	if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8440 		level = btrfs_header_level(root->node);
8441 		path->nodes[level] = btrfs_lock_root_node(root);
8442 		btrfs_set_lock_blocking(path->nodes[level]);
8443 		path->slots[level] = 0;
8444 		path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8445 		memset(&wc->update_progress, 0,
8446 		       sizeof(wc->update_progress));
8447 	} else {
8448 		btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8449 		memcpy(&wc->update_progress, &key,
8450 		       sizeof(wc->update_progress));
8451 
8452 		level = root_item->drop_level;
8453 		BUG_ON(level == 0);
8454 		path->lowest_level = level;
8455 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8456 		path->lowest_level = 0;
8457 		if (ret < 0) {
8458 			err = ret;
8459 			goto out_end_trans;
8460 		}
8461 		WARN_ON(ret > 0);
8462 
8463 		/*
8464 		 * unlock our path, this is safe because only this
8465 		 * function is allowed to delete this snapshot
8466 		 */
8467 		btrfs_unlock_up_safe(path, 0);
8468 
8469 		level = btrfs_header_level(root->node);
8470 		while (1) {
8471 			btrfs_tree_lock(path->nodes[level]);
8472 			btrfs_set_lock_blocking(path->nodes[level]);
8473 			path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8474 
8475 			ret = btrfs_lookup_extent_info(trans, root,
8476 						path->nodes[level]->start,
8477 						level, 1, &wc->refs[level],
8478 						&wc->flags[level]);
8479 			if (ret < 0) {
8480 				err = ret;
8481 				goto out_end_trans;
8482 			}
8483 			BUG_ON(wc->refs[level] == 0);
8484 
8485 			if (level == root_item->drop_level)
8486 				break;
8487 
8488 			btrfs_tree_unlock(path->nodes[level]);
8489 			path->locks[level] = 0;
8490 			WARN_ON(wc->refs[level] != 1);
8491 			level--;
8492 		}
8493 	}
8494 
8495 	wc->level = level;
8496 	wc->shared_level = -1;
8497 	wc->stage = DROP_REFERENCE;
8498 	wc->update_ref = update_ref;
8499 	wc->keep_locks = 0;
8500 	wc->for_reloc = for_reloc;
8501 	wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8502 
8503 	while (1) {
8504 
8505 		ret = walk_down_tree(trans, root, path, wc);
8506 		if (ret < 0) {
8507 			err = ret;
8508 			break;
8509 		}
8510 
8511 		ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8512 		if (ret < 0) {
8513 			err = ret;
8514 			break;
8515 		}
8516 
8517 		if (ret > 0) {
8518 			BUG_ON(wc->stage != DROP_REFERENCE);
8519 			break;
8520 		}
8521 
8522 		if (wc->stage == DROP_REFERENCE) {
8523 			level = wc->level;
8524 			btrfs_node_key(path->nodes[level],
8525 				       &root_item->drop_progress,
8526 				       path->slots[level]);
8527 			root_item->drop_level = level;
8528 		}
8529 
8530 		BUG_ON(wc->level == 0);
8531 		if (btrfs_should_end_transaction(trans, tree_root) ||
8532 		    (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8533 			ret = btrfs_update_root(trans, tree_root,
8534 						&root->root_key,
8535 						root_item);
8536 			if (ret) {
8537 				btrfs_abort_transaction(trans, tree_root, ret);
8538 				err = ret;
8539 				goto out_end_trans;
8540 			}
8541 
8542 			/*
8543 			 * Qgroup update accounting is run from
8544 			 * delayed ref handling. This usually works
8545 			 * out because delayed refs are normally the
8546 			 * only way qgroup updates are added. However,
8547 			 * we may have added updates during our tree
8548 			 * walk so run qgroups here to make sure we
8549 			 * don't lose any updates.
8550 			 */
8551 			ret = btrfs_delayed_qgroup_accounting(trans,
8552 							      root->fs_info);
8553 			if (ret)
8554 				printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8555 						   "running qgroup updates "
8556 						   "during snapshot delete. "
8557 						   "Quota is out of sync, "
8558 						   "rescan required.\n", ret);
8559 
8560 			btrfs_end_transaction_throttle(trans, tree_root);
8561 			if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8562 				pr_debug("BTRFS: drop snapshot early exit\n");
8563 				err = -EAGAIN;
8564 				goto out_free;
8565 			}
8566 
8567 			trans = btrfs_start_transaction(tree_root, 0);
8568 			if (IS_ERR(trans)) {
8569 				err = PTR_ERR(trans);
8570 				goto out_free;
8571 			}
8572 			if (block_rsv)
8573 				trans->block_rsv = block_rsv;
8574 		}
8575 	}
8576 	btrfs_release_path(path);
8577 	if (err)
8578 		goto out_end_trans;
8579 
8580 	ret = btrfs_del_root(trans, tree_root, &root->root_key);
8581 	if (ret) {
8582 		btrfs_abort_transaction(trans, tree_root, ret);
8583 		goto out_end_trans;
8584 	}
8585 
8586 	if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8587 		ret = btrfs_find_root(tree_root, &root->root_key, path,
8588 				      NULL, NULL);
8589 		if (ret < 0) {
8590 			btrfs_abort_transaction(trans, tree_root, ret);
8591 			err = ret;
8592 			goto out_end_trans;
8593 		} else if (ret > 0) {
8594 			/* if we fail to delete the orphan item this time
8595 			 * around, it'll get picked up the next time.
8596 			 *
8597 			 * The most common failure here is just -ENOENT.
8598 			 */
8599 			btrfs_del_orphan_item(trans, tree_root,
8600 					      root->root_key.objectid);
8601 		}
8602 	}
8603 
8604 	if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8605 		btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8606 	} else {
8607 		free_extent_buffer(root->node);
8608 		free_extent_buffer(root->commit_root);
8609 		btrfs_put_fs_root(root);
8610 	}
8611 	root_dropped = true;
8612 out_end_trans:
8613 	ret = btrfs_delayed_qgroup_accounting(trans, tree_root->fs_info);
8614 	if (ret)
8615 		printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8616 				   "running qgroup updates "
8617 				   "during snapshot delete. "
8618 				   "Quota is out of sync, "
8619 				   "rescan required.\n", ret);
8620 
8621 	btrfs_end_transaction_throttle(trans, tree_root);
8622 out_free:
8623 	kfree(wc);
8624 	btrfs_free_path(path);
8625 out:
8626 	/*
8627 	 * So if we need to stop dropping the snapshot for whatever reason we
8628 	 * need to make sure to add it back to the dead root list so that we
8629 	 * keep trying to do the work later.  This also cleans up roots if we
8630 	 * don't have it in the radix (like when we recover after a power fail
8631 	 * or unmount) so we don't leak memory.
8632 	 */
8633 	if (!for_reloc && root_dropped == false)
8634 		btrfs_add_dead_root(root);
8635 	if (err && err != -EAGAIN)
8636 		btrfs_std_error(root->fs_info, err);
8637 	return err;
8638 }
8639 
8640 /*
8641  * drop subtree rooted at tree block 'node'.
8642  *
8643  * NOTE: this function will unlock and release tree block 'node'
8644  * only used by relocation code
8645  */
btrfs_drop_subtree(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * node,struct extent_buffer * parent)8646 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8647 			struct btrfs_root *root,
8648 			struct extent_buffer *node,
8649 			struct extent_buffer *parent)
8650 {
8651 	struct btrfs_path *path;
8652 	struct walk_control *wc;
8653 	int level;
8654 	int parent_level;
8655 	int ret = 0;
8656 	int wret;
8657 
8658 	BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8659 
8660 	path = btrfs_alloc_path();
8661 	if (!path)
8662 		return -ENOMEM;
8663 
8664 	wc = kzalloc(sizeof(*wc), GFP_NOFS);
8665 	if (!wc) {
8666 		btrfs_free_path(path);
8667 		return -ENOMEM;
8668 	}
8669 
8670 	btrfs_assert_tree_locked(parent);
8671 	parent_level = btrfs_header_level(parent);
8672 	extent_buffer_get(parent);
8673 	path->nodes[parent_level] = parent;
8674 	path->slots[parent_level] = btrfs_header_nritems(parent);
8675 
8676 	btrfs_assert_tree_locked(node);
8677 	level = btrfs_header_level(node);
8678 	path->nodes[level] = node;
8679 	path->slots[level] = 0;
8680 	path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8681 
8682 	wc->refs[parent_level] = 1;
8683 	wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8684 	wc->level = level;
8685 	wc->shared_level = -1;
8686 	wc->stage = DROP_REFERENCE;
8687 	wc->update_ref = 0;
8688 	wc->keep_locks = 1;
8689 	wc->for_reloc = 1;
8690 	wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8691 
8692 	while (1) {
8693 		wret = walk_down_tree(trans, root, path, wc);
8694 		if (wret < 0) {
8695 			ret = wret;
8696 			break;
8697 		}
8698 
8699 		wret = walk_up_tree(trans, root, path, wc, parent_level);
8700 		if (wret < 0)
8701 			ret = wret;
8702 		if (wret != 0)
8703 			break;
8704 	}
8705 
8706 	kfree(wc);
8707 	btrfs_free_path(path);
8708 	return ret;
8709 }
8710 
update_block_group_flags(struct btrfs_root * root,u64 flags)8711 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8712 {
8713 	u64 num_devices;
8714 	u64 stripped;
8715 
8716 	/*
8717 	 * if restripe for this chunk_type is on pick target profile and
8718 	 * return, otherwise do the usual balance
8719 	 */
8720 	stripped = get_restripe_target(root->fs_info, flags);
8721 	if (stripped)
8722 		return extended_to_chunk(stripped);
8723 
8724 	num_devices = root->fs_info->fs_devices->rw_devices;
8725 
8726 	stripped = BTRFS_BLOCK_GROUP_RAID0 |
8727 		BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8728 		BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8729 
8730 	if (num_devices == 1) {
8731 		stripped |= BTRFS_BLOCK_GROUP_DUP;
8732 		stripped = flags & ~stripped;
8733 
8734 		/* turn raid0 into single device chunks */
8735 		if (flags & BTRFS_BLOCK_GROUP_RAID0)
8736 			return stripped;
8737 
8738 		/* turn mirroring into duplication */
8739 		if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8740 			     BTRFS_BLOCK_GROUP_RAID10))
8741 			return stripped | BTRFS_BLOCK_GROUP_DUP;
8742 	} else {
8743 		/* they already had raid on here, just return */
8744 		if (flags & stripped)
8745 			return flags;
8746 
8747 		stripped |= BTRFS_BLOCK_GROUP_DUP;
8748 		stripped = flags & ~stripped;
8749 
8750 		/* switch duplicated blocks with raid1 */
8751 		if (flags & BTRFS_BLOCK_GROUP_DUP)
8752 			return stripped | BTRFS_BLOCK_GROUP_RAID1;
8753 
8754 		/* this is drive concat, leave it alone */
8755 	}
8756 
8757 	return flags;
8758 }
8759 
set_block_group_ro(struct btrfs_block_group_cache * cache,int force)8760 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8761 {
8762 	struct btrfs_space_info *sinfo = cache->space_info;
8763 	u64 num_bytes;
8764 	u64 min_allocable_bytes;
8765 	int ret = -ENOSPC;
8766 
8767 
8768 	/*
8769 	 * We need some metadata space and system metadata space for
8770 	 * allocating chunks in some corner cases until we force to set
8771 	 * it to be readonly.
8772 	 */
8773 	if ((sinfo->flags &
8774 	     (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8775 	    !force)
8776 		min_allocable_bytes = 1 * 1024 * 1024;
8777 	else
8778 		min_allocable_bytes = 0;
8779 
8780 	spin_lock(&sinfo->lock);
8781 	spin_lock(&cache->lock);
8782 
8783 	if (cache->ro) {
8784 		ret = 0;
8785 		goto out;
8786 	}
8787 
8788 	num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8789 		    cache->bytes_super - btrfs_block_group_used(&cache->item);
8790 
8791 	if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8792 	    sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8793 	    min_allocable_bytes <= sinfo->total_bytes) {
8794 		sinfo->bytes_readonly += num_bytes;
8795 		cache->ro = 1;
8796 		list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8797 		ret = 0;
8798 	}
8799 out:
8800 	spin_unlock(&cache->lock);
8801 	spin_unlock(&sinfo->lock);
8802 	return ret;
8803 }
8804 
btrfs_set_block_group_ro(struct btrfs_root * root,struct btrfs_block_group_cache * cache)8805 int btrfs_set_block_group_ro(struct btrfs_root *root,
8806 			     struct btrfs_block_group_cache *cache)
8807 
8808 {
8809 	struct btrfs_trans_handle *trans;
8810 	u64 alloc_flags;
8811 	int ret;
8812 
8813 	BUG_ON(cache->ro);
8814 
8815 again:
8816 	trans = btrfs_join_transaction(root);
8817 	if (IS_ERR(trans))
8818 		return PTR_ERR(trans);
8819 
8820 	/*
8821 	 * we're not allowed to set block groups readonly after the dirty
8822 	 * block groups cache has started writing.  If it already started,
8823 	 * back off and let this transaction commit
8824 	 */
8825 	mutex_lock(&root->fs_info->ro_block_group_mutex);
8826 	if (trans->transaction->dirty_bg_run) {
8827 		u64 transid = trans->transid;
8828 
8829 		mutex_unlock(&root->fs_info->ro_block_group_mutex);
8830 		btrfs_end_transaction(trans, root);
8831 
8832 		ret = btrfs_wait_for_commit(root, transid);
8833 		if (ret)
8834 			return ret;
8835 		goto again;
8836 	}
8837 
8838 	/*
8839 	 * if we are changing raid levels, try to allocate a corresponding
8840 	 * block group with the new raid level.
8841 	 */
8842 	alloc_flags = update_block_group_flags(root, cache->flags);
8843 	if (alloc_flags != cache->flags) {
8844 		ret = do_chunk_alloc(trans, root, alloc_flags,
8845 				     CHUNK_ALLOC_FORCE);
8846 		/*
8847 		 * ENOSPC is allowed here, we may have enough space
8848 		 * already allocated at the new raid level to
8849 		 * carry on
8850 		 */
8851 		if (ret == -ENOSPC)
8852 			ret = 0;
8853 		if (ret < 0)
8854 			goto out;
8855 	}
8856 
8857 	ret = set_block_group_ro(cache, 0);
8858 	if (!ret)
8859 		goto out;
8860 	alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8861 	ret = do_chunk_alloc(trans, root, alloc_flags,
8862 			     CHUNK_ALLOC_FORCE);
8863 	if (ret < 0)
8864 		goto out;
8865 	ret = set_block_group_ro(cache, 0);
8866 out:
8867 	if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8868 		alloc_flags = update_block_group_flags(root, cache->flags);
8869 		lock_chunks(root->fs_info->chunk_root);
8870 		check_system_chunk(trans, root, alloc_flags);
8871 		unlock_chunks(root->fs_info->chunk_root);
8872 	}
8873 	mutex_unlock(&root->fs_info->ro_block_group_mutex);
8874 
8875 	btrfs_end_transaction(trans, root);
8876 	return ret;
8877 }
8878 
btrfs_force_chunk_alloc(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 type)8879 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8880 			    struct btrfs_root *root, u64 type)
8881 {
8882 	u64 alloc_flags = get_alloc_profile(root, type);
8883 	return do_chunk_alloc(trans, root, alloc_flags,
8884 			      CHUNK_ALLOC_FORCE);
8885 }
8886 
8887 /*
8888  * helper to account the unused space of all the readonly block group in the
8889  * space_info. takes mirrors into account.
8890  */
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info * sinfo)8891 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8892 {
8893 	struct btrfs_block_group_cache *block_group;
8894 	u64 free_bytes = 0;
8895 	int factor;
8896 
8897 	/* It's df, we don't care if it's racey */
8898 	if (list_empty(&sinfo->ro_bgs))
8899 		return 0;
8900 
8901 	spin_lock(&sinfo->lock);
8902 	list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8903 		spin_lock(&block_group->lock);
8904 
8905 		if (!block_group->ro) {
8906 			spin_unlock(&block_group->lock);
8907 			continue;
8908 		}
8909 
8910 		if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8911 					  BTRFS_BLOCK_GROUP_RAID10 |
8912 					  BTRFS_BLOCK_GROUP_DUP))
8913 			factor = 2;
8914 		else
8915 			factor = 1;
8916 
8917 		free_bytes += (block_group->key.offset -
8918 			       btrfs_block_group_used(&block_group->item)) *
8919 			       factor;
8920 
8921 		spin_unlock(&block_group->lock);
8922 	}
8923 	spin_unlock(&sinfo->lock);
8924 
8925 	return free_bytes;
8926 }
8927 
btrfs_set_block_group_rw(struct btrfs_root * root,struct btrfs_block_group_cache * cache)8928 void btrfs_set_block_group_rw(struct btrfs_root *root,
8929 			      struct btrfs_block_group_cache *cache)
8930 {
8931 	struct btrfs_space_info *sinfo = cache->space_info;
8932 	u64 num_bytes;
8933 
8934 	BUG_ON(!cache->ro);
8935 
8936 	spin_lock(&sinfo->lock);
8937 	spin_lock(&cache->lock);
8938 	num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8939 		    cache->bytes_super - btrfs_block_group_used(&cache->item);
8940 	sinfo->bytes_readonly -= num_bytes;
8941 	cache->ro = 0;
8942 	list_del_init(&cache->ro_list);
8943 	spin_unlock(&cache->lock);
8944 	spin_unlock(&sinfo->lock);
8945 }
8946 
8947 /*
8948  * checks to see if its even possible to relocate this block group.
8949  *
8950  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8951  * ok to go ahead and try.
8952  */
btrfs_can_relocate(struct btrfs_root * root,u64 bytenr)8953 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8954 {
8955 	struct btrfs_block_group_cache *block_group;
8956 	struct btrfs_space_info *space_info;
8957 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8958 	struct btrfs_device *device;
8959 	struct btrfs_trans_handle *trans;
8960 	u64 min_free;
8961 	u64 dev_min = 1;
8962 	u64 dev_nr = 0;
8963 	u64 target;
8964 	int index;
8965 	int full = 0;
8966 	int ret = 0;
8967 
8968 	block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8969 
8970 	/* odd, couldn't find the block group, leave it alone */
8971 	if (!block_group)
8972 		return -1;
8973 
8974 	min_free = btrfs_block_group_used(&block_group->item);
8975 
8976 	/* no bytes used, we're good */
8977 	if (!min_free)
8978 		goto out;
8979 
8980 	space_info = block_group->space_info;
8981 	spin_lock(&space_info->lock);
8982 
8983 	full = space_info->full;
8984 
8985 	/*
8986 	 * if this is the last block group we have in this space, we can't
8987 	 * relocate it unless we're able to allocate a new chunk below.
8988 	 *
8989 	 * Otherwise, we need to make sure we have room in the space to handle
8990 	 * all of the extents from this block group.  If we can, we're good
8991 	 */
8992 	if ((space_info->total_bytes != block_group->key.offset) &&
8993 	    (space_info->bytes_used + space_info->bytes_reserved +
8994 	     space_info->bytes_pinned + space_info->bytes_readonly +
8995 	     min_free < space_info->total_bytes)) {
8996 		spin_unlock(&space_info->lock);
8997 		goto out;
8998 	}
8999 	spin_unlock(&space_info->lock);
9000 
9001 	/*
9002 	 * ok we don't have enough space, but maybe we have free space on our
9003 	 * devices to allocate new chunks for relocation, so loop through our
9004 	 * alloc devices and guess if we have enough space.  if this block
9005 	 * group is going to be restriped, run checks against the target
9006 	 * profile instead of the current one.
9007 	 */
9008 	ret = -1;
9009 
9010 	/*
9011 	 * index:
9012 	 *      0: raid10
9013 	 *      1: raid1
9014 	 *      2: dup
9015 	 *      3: raid0
9016 	 *      4: single
9017 	 */
9018 	target = get_restripe_target(root->fs_info, block_group->flags);
9019 	if (target) {
9020 		index = __get_raid_index(extended_to_chunk(target));
9021 	} else {
9022 		/*
9023 		 * this is just a balance, so if we were marked as full
9024 		 * we know there is no space for a new chunk
9025 		 */
9026 		if (full)
9027 			goto out;
9028 
9029 		index = get_block_group_index(block_group);
9030 	}
9031 
9032 	if (index == BTRFS_RAID_RAID10) {
9033 		dev_min = 4;
9034 		/* Divide by 2 */
9035 		min_free >>= 1;
9036 	} else if (index == BTRFS_RAID_RAID1) {
9037 		dev_min = 2;
9038 	} else if (index == BTRFS_RAID_DUP) {
9039 		/* Multiply by 2 */
9040 		min_free <<= 1;
9041 	} else if (index == BTRFS_RAID_RAID0) {
9042 		dev_min = fs_devices->rw_devices;
9043 		min_free = div64_u64(min_free, dev_min);
9044 	}
9045 
9046 	/* We need to do this so that we can look at pending chunks */
9047 	trans = btrfs_join_transaction(root);
9048 	if (IS_ERR(trans)) {
9049 		ret = PTR_ERR(trans);
9050 		goto out;
9051 	}
9052 
9053 	mutex_lock(&root->fs_info->chunk_mutex);
9054 	list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9055 		u64 dev_offset;
9056 
9057 		/*
9058 		 * check to make sure we can actually find a chunk with enough
9059 		 * space to fit our block group in.
9060 		 */
9061 		if (device->total_bytes > device->bytes_used + min_free &&
9062 		    !device->is_tgtdev_for_dev_replace) {
9063 			ret = find_free_dev_extent(trans, device, min_free,
9064 						   &dev_offset, NULL);
9065 			if (!ret)
9066 				dev_nr++;
9067 
9068 			if (dev_nr >= dev_min)
9069 				break;
9070 
9071 			ret = -1;
9072 		}
9073 	}
9074 	mutex_unlock(&root->fs_info->chunk_mutex);
9075 	btrfs_end_transaction(trans, root);
9076 out:
9077 	btrfs_put_block_group(block_group);
9078 	return ret;
9079 }
9080 
find_first_block_group(struct btrfs_root * root,struct btrfs_path * path,struct btrfs_key * key)9081 static int find_first_block_group(struct btrfs_root *root,
9082 		struct btrfs_path *path, struct btrfs_key *key)
9083 {
9084 	int ret = 0;
9085 	struct btrfs_key found_key;
9086 	struct extent_buffer *leaf;
9087 	int slot;
9088 
9089 	ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9090 	if (ret < 0)
9091 		goto out;
9092 
9093 	while (1) {
9094 		slot = path->slots[0];
9095 		leaf = path->nodes[0];
9096 		if (slot >= btrfs_header_nritems(leaf)) {
9097 			ret = btrfs_next_leaf(root, path);
9098 			if (ret == 0)
9099 				continue;
9100 			if (ret < 0)
9101 				goto out;
9102 			break;
9103 		}
9104 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
9105 
9106 		if (found_key.objectid >= key->objectid &&
9107 		    found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9108 			ret = 0;
9109 			goto out;
9110 		}
9111 		path->slots[0]++;
9112 	}
9113 out:
9114 	return ret;
9115 }
9116 
btrfs_put_block_group_cache(struct btrfs_fs_info * info)9117 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9118 {
9119 	struct btrfs_block_group_cache *block_group;
9120 	u64 last = 0;
9121 
9122 	while (1) {
9123 		struct inode *inode;
9124 
9125 		block_group = btrfs_lookup_first_block_group(info, last);
9126 		while (block_group) {
9127 			spin_lock(&block_group->lock);
9128 			if (block_group->iref)
9129 				break;
9130 			spin_unlock(&block_group->lock);
9131 			block_group = next_block_group(info->tree_root,
9132 						       block_group);
9133 		}
9134 		if (!block_group) {
9135 			if (last == 0)
9136 				break;
9137 			last = 0;
9138 			continue;
9139 		}
9140 
9141 		inode = block_group->inode;
9142 		block_group->iref = 0;
9143 		block_group->inode = NULL;
9144 		spin_unlock(&block_group->lock);
9145 		iput(inode);
9146 		last = block_group->key.objectid + block_group->key.offset;
9147 		btrfs_put_block_group(block_group);
9148 	}
9149 }
9150 
btrfs_free_block_groups(struct btrfs_fs_info * info)9151 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9152 {
9153 	struct btrfs_block_group_cache *block_group;
9154 	struct btrfs_space_info *space_info;
9155 	struct btrfs_caching_control *caching_ctl;
9156 	struct rb_node *n;
9157 
9158 	down_write(&info->commit_root_sem);
9159 	while (!list_empty(&info->caching_block_groups)) {
9160 		caching_ctl = list_entry(info->caching_block_groups.next,
9161 					 struct btrfs_caching_control, list);
9162 		list_del(&caching_ctl->list);
9163 		put_caching_control(caching_ctl);
9164 	}
9165 	up_write(&info->commit_root_sem);
9166 
9167 	spin_lock(&info->unused_bgs_lock);
9168 	while (!list_empty(&info->unused_bgs)) {
9169 		block_group = list_first_entry(&info->unused_bgs,
9170 					       struct btrfs_block_group_cache,
9171 					       bg_list);
9172 		list_del_init(&block_group->bg_list);
9173 		btrfs_put_block_group(block_group);
9174 	}
9175 	spin_unlock(&info->unused_bgs_lock);
9176 
9177 	spin_lock(&info->block_group_cache_lock);
9178 	while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9179 		block_group = rb_entry(n, struct btrfs_block_group_cache,
9180 				       cache_node);
9181 		rb_erase(&block_group->cache_node,
9182 			 &info->block_group_cache_tree);
9183 		RB_CLEAR_NODE(&block_group->cache_node);
9184 		spin_unlock(&info->block_group_cache_lock);
9185 
9186 		down_write(&block_group->space_info->groups_sem);
9187 		list_del(&block_group->list);
9188 		up_write(&block_group->space_info->groups_sem);
9189 
9190 		if (block_group->cached == BTRFS_CACHE_STARTED)
9191 			wait_block_group_cache_done(block_group);
9192 
9193 		/*
9194 		 * We haven't cached this block group, which means we could
9195 		 * possibly have excluded extents on this block group.
9196 		 */
9197 		if (block_group->cached == BTRFS_CACHE_NO ||
9198 		    block_group->cached == BTRFS_CACHE_ERROR)
9199 			free_excluded_extents(info->extent_root, block_group);
9200 
9201 		btrfs_remove_free_space_cache(block_group);
9202 		btrfs_put_block_group(block_group);
9203 
9204 		spin_lock(&info->block_group_cache_lock);
9205 	}
9206 	spin_unlock(&info->block_group_cache_lock);
9207 
9208 	/* now that all the block groups are freed, go through and
9209 	 * free all the space_info structs.  This is only called during
9210 	 * the final stages of unmount, and so we know nobody is
9211 	 * using them.  We call synchronize_rcu() once before we start,
9212 	 * just to be on the safe side.
9213 	 */
9214 	synchronize_rcu();
9215 
9216 	release_global_block_rsv(info);
9217 
9218 	while (!list_empty(&info->space_info)) {
9219 		int i;
9220 
9221 		space_info = list_entry(info->space_info.next,
9222 					struct btrfs_space_info,
9223 					list);
9224 		if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9225 			if (WARN_ON(space_info->bytes_pinned > 0 ||
9226 			    space_info->bytes_reserved > 0 ||
9227 			    space_info->bytes_may_use > 0)) {
9228 				dump_space_info(space_info, 0, 0);
9229 			}
9230 		}
9231 		list_del(&space_info->list);
9232 		for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9233 			struct kobject *kobj;
9234 			kobj = space_info->block_group_kobjs[i];
9235 			space_info->block_group_kobjs[i] = NULL;
9236 			if (kobj) {
9237 				kobject_del(kobj);
9238 				kobject_put(kobj);
9239 			}
9240 		}
9241 		kobject_del(&space_info->kobj);
9242 		kobject_put(&space_info->kobj);
9243 	}
9244 	return 0;
9245 }
9246 
__link_block_group(struct btrfs_space_info * space_info,struct btrfs_block_group_cache * cache)9247 static void __link_block_group(struct btrfs_space_info *space_info,
9248 			       struct btrfs_block_group_cache *cache)
9249 {
9250 	int index = get_block_group_index(cache);
9251 	bool first = false;
9252 
9253 	down_write(&space_info->groups_sem);
9254 	if (list_empty(&space_info->block_groups[index]))
9255 		first = true;
9256 	list_add_tail(&cache->list, &space_info->block_groups[index]);
9257 	up_write(&space_info->groups_sem);
9258 
9259 	if (first) {
9260 		struct raid_kobject *rkobj;
9261 		int ret;
9262 
9263 		rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9264 		if (!rkobj)
9265 			goto out_err;
9266 		rkobj->raid_type = index;
9267 		kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9268 		ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9269 				  "%s", get_raid_name(index));
9270 		if (ret) {
9271 			kobject_put(&rkobj->kobj);
9272 			goto out_err;
9273 		}
9274 		space_info->block_group_kobjs[index] = &rkobj->kobj;
9275 	}
9276 
9277 	return;
9278 out_err:
9279 	pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9280 }
9281 
9282 static struct btrfs_block_group_cache *
btrfs_create_block_group_cache(struct btrfs_root * root,u64 start,u64 size)9283 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9284 {
9285 	struct btrfs_block_group_cache *cache;
9286 
9287 	cache = kzalloc(sizeof(*cache), GFP_NOFS);
9288 	if (!cache)
9289 		return NULL;
9290 
9291 	cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9292 					GFP_NOFS);
9293 	if (!cache->free_space_ctl) {
9294 		kfree(cache);
9295 		return NULL;
9296 	}
9297 
9298 	cache->key.objectid = start;
9299 	cache->key.offset = size;
9300 	cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9301 
9302 	cache->sectorsize = root->sectorsize;
9303 	cache->fs_info = root->fs_info;
9304 	cache->full_stripe_len = btrfs_full_stripe_len(root,
9305 					       &root->fs_info->mapping_tree,
9306 					       start);
9307 	atomic_set(&cache->count, 1);
9308 	spin_lock_init(&cache->lock);
9309 	init_rwsem(&cache->data_rwsem);
9310 	INIT_LIST_HEAD(&cache->list);
9311 	INIT_LIST_HEAD(&cache->cluster_list);
9312 	INIT_LIST_HEAD(&cache->bg_list);
9313 	INIT_LIST_HEAD(&cache->ro_list);
9314 	INIT_LIST_HEAD(&cache->dirty_list);
9315 	INIT_LIST_HEAD(&cache->io_list);
9316 	btrfs_init_free_space_ctl(cache);
9317 	atomic_set(&cache->trimming, 0);
9318 
9319 	return cache;
9320 }
9321 
btrfs_read_block_groups(struct btrfs_root * root)9322 int btrfs_read_block_groups(struct btrfs_root *root)
9323 {
9324 	struct btrfs_path *path;
9325 	int ret;
9326 	struct btrfs_block_group_cache *cache;
9327 	struct btrfs_fs_info *info = root->fs_info;
9328 	struct btrfs_space_info *space_info;
9329 	struct btrfs_key key;
9330 	struct btrfs_key found_key;
9331 	struct extent_buffer *leaf;
9332 	int need_clear = 0;
9333 	u64 cache_gen;
9334 
9335 	root = info->extent_root;
9336 	key.objectid = 0;
9337 	key.offset = 0;
9338 	key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9339 	path = btrfs_alloc_path();
9340 	if (!path)
9341 		return -ENOMEM;
9342 	path->reada = 1;
9343 
9344 	cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9345 	if (btrfs_test_opt(root, SPACE_CACHE) &&
9346 	    btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9347 		need_clear = 1;
9348 	if (btrfs_test_opt(root, CLEAR_CACHE))
9349 		need_clear = 1;
9350 
9351 	while (1) {
9352 		ret = find_first_block_group(root, path, &key);
9353 		if (ret > 0)
9354 			break;
9355 		if (ret != 0)
9356 			goto error;
9357 
9358 		leaf = path->nodes[0];
9359 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9360 
9361 		cache = btrfs_create_block_group_cache(root, found_key.objectid,
9362 						       found_key.offset);
9363 		if (!cache) {
9364 			ret = -ENOMEM;
9365 			goto error;
9366 		}
9367 
9368 		if (need_clear) {
9369 			/*
9370 			 * When we mount with old space cache, we need to
9371 			 * set BTRFS_DC_CLEAR and set dirty flag.
9372 			 *
9373 			 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9374 			 *    truncate the old free space cache inode and
9375 			 *    setup a new one.
9376 			 * b) Setting 'dirty flag' makes sure that we flush
9377 			 *    the new space cache info onto disk.
9378 			 */
9379 			if (btrfs_test_opt(root, SPACE_CACHE))
9380 				cache->disk_cache_state = BTRFS_DC_CLEAR;
9381 		}
9382 
9383 		read_extent_buffer(leaf, &cache->item,
9384 				   btrfs_item_ptr_offset(leaf, path->slots[0]),
9385 				   sizeof(cache->item));
9386 		cache->flags = btrfs_block_group_flags(&cache->item);
9387 
9388 		key.objectid = found_key.objectid + found_key.offset;
9389 		btrfs_release_path(path);
9390 
9391 		/*
9392 		 * We need to exclude the super stripes now so that the space
9393 		 * info has super bytes accounted for, otherwise we'll think
9394 		 * we have more space than we actually do.
9395 		 */
9396 		ret = exclude_super_stripes(root, cache);
9397 		if (ret) {
9398 			/*
9399 			 * We may have excluded something, so call this just in
9400 			 * case.
9401 			 */
9402 			free_excluded_extents(root, cache);
9403 			btrfs_put_block_group(cache);
9404 			goto error;
9405 		}
9406 
9407 		/*
9408 		 * check for two cases, either we are full, and therefore
9409 		 * don't need to bother with the caching work since we won't
9410 		 * find any space, or we are empty, and we can just add all
9411 		 * the space in and be done with it.  This saves us _alot_ of
9412 		 * time, particularly in the full case.
9413 		 */
9414 		if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9415 			cache->last_byte_to_unpin = (u64)-1;
9416 			cache->cached = BTRFS_CACHE_FINISHED;
9417 			free_excluded_extents(root, cache);
9418 		} else if (btrfs_block_group_used(&cache->item) == 0) {
9419 			cache->last_byte_to_unpin = (u64)-1;
9420 			cache->cached = BTRFS_CACHE_FINISHED;
9421 			add_new_free_space(cache, root->fs_info,
9422 					   found_key.objectid,
9423 					   found_key.objectid +
9424 					   found_key.offset);
9425 			free_excluded_extents(root, cache);
9426 		}
9427 
9428 		ret = btrfs_add_block_group_cache(root->fs_info, cache);
9429 		if (ret) {
9430 			btrfs_remove_free_space_cache(cache);
9431 			btrfs_put_block_group(cache);
9432 			goto error;
9433 		}
9434 
9435 		ret = update_space_info(info, cache->flags, found_key.offset,
9436 					btrfs_block_group_used(&cache->item),
9437 					&space_info);
9438 		if (ret) {
9439 			btrfs_remove_free_space_cache(cache);
9440 			spin_lock(&info->block_group_cache_lock);
9441 			rb_erase(&cache->cache_node,
9442 				 &info->block_group_cache_tree);
9443 			RB_CLEAR_NODE(&cache->cache_node);
9444 			spin_unlock(&info->block_group_cache_lock);
9445 			btrfs_put_block_group(cache);
9446 			goto error;
9447 		}
9448 
9449 		cache->space_info = space_info;
9450 		spin_lock(&cache->space_info->lock);
9451 		cache->space_info->bytes_readonly += cache->bytes_super;
9452 		spin_unlock(&cache->space_info->lock);
9453 
9454 		__link_block_group(space_info, cache);
9455 
9456 		set_avail_alloc_bits(root->fs_info, cache->flags);
9457 		if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9458 			set_block_group_ro(cache, 1);
9459 		} else if (btrfs_block_group_used(&cache->item) == 0) {
9460 			spin_lock(&info->unused_bgs_lock);
9461 			/* Should always be true but just in case. */
9462 			if (list_empty(&cache->bg_list)) {
9463 				btrfs_get_block_group(cache);
9464 				list_add_tail(&cache->bg_list,
9465 					      &info->unused_bgs);
9466 			}
9467 			spin_unlock(&info->unused_bgs_lock);
9468 		}
9469 	}
9470 
9471 	list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9472 		if (!(get_alloc_profile(root, space_info->flags) &
9473 		      (BTRFS_BLOCK_GROUP_RAID10 |
9474 		       BTRFS_BLOCK_GROUP_RAID1 |
9475 		       BTRFS_BLOCK_GROUP_RAID5 |
9476 		       BTRFS_BLOCK_GROUP_RAID6 |
9477 		       BTRFS_BLOCK_GROUP_DUP)))
9478 			continue;
9479 		/*
9480 		 * avoid allocating from un-mirrored block group if there are
9481 		 * mirrored block groups.
9482 		 */
9483 		list_for_each_entry(cache,
9484 				&space_info->block_groups[BTRFS_RAID_RAID0],
9485 				list)
9486 			set_block_group_ro(cache, 1);
9487 		list_for_each_entry(cache,
9488 				&space_info->block_groups[BTRFS_RAID_SINGLE],
9489 				list)
9490 			set_block_group_ro(cache, 1);
9491 	}
9492 
9493 	init_global_block_rsv(info);
9494 	ret = 0;
9495 error:
9496 	btrfs_free_path(path);
9497 	return ret;
9498 }
9499 
btrfs_create_pending_block_groups(struct btrfs_trans_handle * trans,struct btrfs_root * root)9500 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9501 				       struct btrfs_root *root)
9502 {
9503 	struct btrfs_block_group_cache *block_group, *tmp;
9504 	struct btrfs_root *extent_root = root->fs_info->extent_root;
9505 	struct btrfs_block_group_item item;
9506 	struct btrfs_key key;
9507 	int ret = 0;
9508 
9509 	list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9510 		if (ret)
9511 			goto next;
9512 
9513 		spin_lock(&block_group->lock);
9514 		memcpy(&item, &block_group->item, sizeof(item));
9515 		memcpy(&key, &block_group->key, sizeof(key));
9516 		spin_unlock(&block_group->lock);
9517 
9518 		ret = btrfs_insert_item(trans, extent_root, &key, &item,
9519 					sizeof(item));
9520 		if (ret)
9521 			btrfs_abort_transaction(trans, extent_root, ret);
9522 		ret = btrfs_finish_chunk_alloc(trans, extent_root,
9523 					       key.objectid, key.offset);
9524 		if (ret)
9525 			btrfs_abort_transaction(trans, extent_root, ret);
9526 next:
9527 		list_del_init(&block_group->bg_list);
9528 	}
9529 }
9530 
btrfs_make_block_group(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 bytes_used,u64 type,u64 chunk_objectid,u64 chunk_offset,u64 size)9531 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9532 			   struct btrfs_root *root, u64 bytes_used,
9533 			   u64 type, u64 chunk_objectid, u64 chunk_offset,
9534 			   u64 size)
9535 {
9536 	int ret;
9537 	struct btrfs_root *extent_root;
9538 	struct btrfs_block_group_cache *cache;
9539 
9540 	extent_root = root->fs_info->extent_root;
9541 
9542 	btrfs_set_log_full_commit(root->fs_info, trans);
9543 
9544 	cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9545 	if (!cache)
9546 		return -ENOMEM;
9547 
9548 	btrfs_set_block_group_used(&cache->item, bytes_used);
9549 	btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9550 	btrfs_set_block_group_flags(&cache->item, type);
9551 
9552 	cache->flags = type;
9553 	cache->last_byte_to_unpin = (u64)-1;
9554 	cache->cached = BTRFS_CACHE_FINISHED;
9555 	ret = exclude_super_stripes(root, cache);
9556 	if (ret) {
9557 		/*
9558 		 * We may have excluded something, so call this just in
9559 		 * case.
9560 		 */
9561 		free_excluded_extents(root, cache);
9562 		btrfs_put_block_group(cache);
9563 		return ret;
9564 	}
9565 
9566 	add_new_free_space(cache, root->fs_info, chunk_offset,
9567 			   chunk_offset + size);
9568 
9569 	free_excluded_extents(root, cache);
9570 
9571 	ret = btrfs_add_block_group_cache(root->fs_info, cache);
9572 	if (ret) {
9573 		btrfs_remove_free_space_cache(cache);
9574 		btrfs_put_block_group(cache);
9575 		return ret;
9576 	}
9577 
9578 	ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9579 				&cache->space_info);
9580 	if (ret) {
9581 		btrfs_remove_free_space_cache(cache);
9582 		spin_lock(&root->fs_info->block_group_cache_lock);
9583 		rb_erase(&cache->cache_node,
9584 			 &root->fs_info->block_group_cache_tree);
9585 		RB_CLEAR_NODE(&cache->cache_node);
9586 		spin_unlock(&root->fs_info->block_group_cache_lock);
9587 		btrfs_put_block_group(cache);
9588 		return ret;
9589 	}
9590 	update_global_block_rsv(root->fs_info);
9591 
9592 	spin_lock(&cache->space_info->lock);
9593 	cache->space_info->bytes_readonly += cache->bytes_super;
9594 	spin_unlock(&cache->space_info->lock);
9595 
9596 	__link_block_group(cache->space_info, cache);
9597 
9598 	list_add_tail(&cache->bg_list, &trans->new_bgs);
9599 
9600 	set_avail_alloc_bits(extent_root->fs_info, type);
9601 
9602 	return 0;
9603 }
9604 
clear_avail_alloc_bits(struct btrfs_fs_info * fs_info,u64 flags)9605 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9606 {
9607 	u64 extra_flags = chunk_to_extended(flags) &
9608 				BTRFS_EXTENDED_PROFILE_MASK;
9609 
9610 	write_seqlock(&fs_info->profiles_lock);
9611 	if (flags & BTRFS_BLOCK_GROUP_DATA)
9612 		fs_info->avail_data_alloc_bits &= ~extra_flags;
9613 	if (flags & BTRFS_BLOCK_GROUP_METADATA)
9614 		fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9615 	if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9616 		fs_info->avail_system_alloc_bits &= ~extra_flags;
9617 	write_sequnlock(&fs_info->profiles_lock);
9618 }
9619 
btrfs_remove_block_group(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 group_start,struct extent_map * em)9620 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9621 			     struct btrfs_root *root, u64 group_start,
9622 			     struct extent_map *em)
9623 {
9624 	struct btrfs_path *path;
9625 	struct btrfs_block_group_cache *block_group;
9626 	struct btrfs_free_cluster *cluster;
9627 	struct btrfs_root *tree_root = root->fs_info->tree_root;
9628 	struct btrfs_key key;
9629 	struct inode *inode;
9630 	struct kobject *kobj = NULL;
9631 	int ret;
9632 	int index;
9633 	int factor;
9634 	struct btrfs_caching_control *caching_ctl = NULL;
9635 	bool remove_em;
9636 
9637 	root = root->fs_info->extent_root;
9638 
9639 	block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9640 	BUG_ON(!block_group);
9641 	BUG_ON(!block_group->ro);
9642 
9643 	/*
9644 	 * Free the reserved super bytes from this block group before
9645 	 * remove it.
9646 	 */
9647 	free_excluded_extents(root, block_group);
9648 
9649 	memcpy(&key, &block_group->key, sizeof(key));
9650 	index = get_block_group_index(block_group);
9651 	if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9652 				  BTRFS_BLOCK_GROUP_RAID1 |
9653 				  BTRFS_BLOCK_GROUP_RAID10))
9654 		factor = 2;
9655 	else
9656 		factor = 1;
9657 
9658 	/* make sure this block group isn't part of an allocation cluster */
9659 	cluster = &root->fs_info->data_alloc_cluster;
9660 	spin_lock(&cluster->refill_lock);
9661 	btrfs_return_cluster_to_free_space(block_group, cluster);
9662 	spin_unlock(&cluster->refill_lock);
9663 
9664 	/*
9665 	 * make sure this block group isn't part of a metadata
9666 	 * allocation cluster
9667 	 */
9668 	cluster = &root->fs_info->meta_alloc_cluster;
9669 	spin_lock(&cluster->refill_lock);
9670 	btrfs_return_cluster_to_free_space(block_group, cluster);
9671 	spin_unlock(&cluster->refill_lock);
9672 
9673 	path = btrfs_alloc_path();
9674 	if (!path) {
9675 		ret = -ENOMEM;
9676 		goto out;
9677 	}
9678 
9679 	/*
9680 	 * get the inode first so any iput calls done for the io_list
9681 	 * aren't the final iput (no unlinks allowed now)
9682 	 */
9683 	inode = lookup_free_space_inode(tree_root, block_group, path);
9684 
9685 	mutex_lock(&trans->transaction->cache_write_mutex);
9686 	/*
9687 	 * make sure our free spache cache IO is done before remove the
9688 	 * free space inode
9689 	 */
9690 	spin_lock(&trans->transaction->dirty_bgs_lock);
9691 	if (!list_empty(&block_group->io_list)) {
9692 		list_del_init(&block_group->io_list);
9693 
9694 		WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
9695 
9696 		spin_unlock(&trans->transaction->dirty_bgs_lock);
9697 		btrfs_wait_cache_io(root, trans, block_group,
9698 				    &block_group->io_ctl, path,
9699 				    block_group->key.objectid);
9700 		btrfs_put_block_group(block_group);
9701 		spin_lock(&trans->transaction->dirty_bgs_lock);
9702 	}
9703 
9704 	if (!list_empty(&block_group->dirty_list)) {
9705 		list_del_init(&block_group->dirty_list);
9706 		btrfs_put_block_group(block_group);
9707 	}
9708 	spin_unlock(&trans->transaction->dirty_bgs_lock);
9709 	mutex_unlock(&trans->transaction->cache_write_mutex);
9710 
9711 	if (!IS_ERR(inode)) {
9712 		ret = btrfs_orphan_add(trans, inode);
9713 		if (ret) {
9714 			btrfs_add_delayed_iput(inode);
9715 			goto out;
9716 		}
9717 		clear_nlink(inode);
9718 		/* One for the block groups ref */
9719 		spin_lock(&block_group->lock);
9720 		if (block_group->iref) {
9721 			block_group->iref = 0;
9722 			block_group->inode = NULL;
9723 			spin_unlock(&block_group->lock);
9724 			iput(inode);
9725 		} else {
9726 			spin_unlock(&block_group->lock);
9727 		}
9728 		/* One for our lookup ref */
9729 		btrfs_add_delayed_iput(inode);
9730 	}
9731 
9732 	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9733 	key.offset = block_group->key.objectid;
9734 	key.type = 0;
9735 
9736 	ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9737 	if (ret < 0)
9738 		goto out;
9739 	if (ret > 0)
9740 		btrfs_release_path(path);
9741 	if (ret == 0) {
9742 		ret = btrfs_del_item(trans, tree_root, path);
9743 		if (ret)
9744 			goto out;
9745 		btrfs_release_path(path);
9746 	}
9747 
9748 	spin_lock(&root->fs_info->block_group_cache_lock);
9749 	rb_erase(&block_group->cache_node,
9750 		 &root->fs_info->block_group_cache_tree);
9751 	RB_CLEAR_NODE(&block_group->cache_node);
9752 
9753 	if (root->fs_info->first_logical_byte == block_group->key.objectid)
9754 		root->fs_info->first_logical_byte = (u64)-1;
9755 	spin_unlock(&root->fs_info->block_group_cache_lock);
9756 
9757 	down_write(&block_group->space_info->groups_sem);
9758 	/*
9759 	 * we must use list_del_init so people can check to see if they
9760 	 * are still on the list after taking the semaphore
9761 	 */
9762 	list_del_init(&block_group->list);
9763 	if (list_empty(&block_group->space_info->block_groups[index])) {
9764 		kobj = block_group->space_info->block_group_kobjs[index];
9765 		block_group->space_info->block_group_kobjs[index] = NULL;
9766 		clear_avail_alloc_bits(root->fs_info, block_group->flags);
9767 	}
9768 	up_write(&block_group->space_info->groups_sem);
9769 	if (kobj) {
9770 		kobject_del(kobj);
9771 		kobject_put(kobj);
9772 	}
9773 
9774 	if (block_group->has_caching_ctl)
9775 		caching_ctl = get_caching_control(block_group);
9776 	if (block_group->cached == BTRFS_CACHE_STARTED)
9777 		wait_block_group_cache_done(block_group);
9778 	if (block_group->has_caching_ctl) {
9779 		down_write(&root->fs_info->commit_root_sem);
9780 		if (!caching_ctl) {
9781 			struct btrfs_caching_control *ctl;
9782 
9783 			list_for_each_entry(ctl,
9784 				    &root->fs_info->caching_block_groups, list)
9785 				if (ctl->block_group == block_group) {
9786 					caching_ctl = ctl;
9787 					atomic_inc(&caching_ctl->count);
9788 					break;
9789 				}
9790 		}
9791 		if (caching_ctl)
9792 			list_del_init(&caching_ctl->list);
9793 		up_write(&root->fs_info->commit_root_sem);
9794 		if (caching_ctl) {
9795 			/* Once for the caching bgs list and once for us. */
9796 			put_caching_control(caching_ctl);
9797 			put_caching_control(caching_ctl);
9798 		}
9799 	}
9800 
9801 	spin_lock(&trans->transaction->dirty_bgs_lock);
9802 	if (!list_empty(&block_group->dirty_list)) {
9803 		WARN_ON(1);
9804 	}
9805 	if (!list_empty(&block_group->io_list)) {
9806 		WARN_ON(1);
9807 	}
9808 	spin_unlock(&trans->transaction->dirty_bgs_lock);
9809 	btrfs_remove_free_space_cache(block_group);
9810 
9811 	spin_lock(&block_group->space_info->lock);
9812 	list_del_init(&block_group->ro_list);
9813 
9814 	if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
9815 		WARN_ON(block_group->space_info->total_bytes
9816 			< block_group->key.offset);
9817 		WARN_ON(block_group->space_info->bytes_readonly
9818 			< block_group->key.offset);
9819 		WARN_ON(block_group->space_info->disk_total
9820 			< block_group->key.offset * factor);
9821 	}
9822 	block_group->space_info->total_bytes -= block_group->key.offset;
9823 	block_group->space_info->bytes_readonly -= block_group->key.offset;
9824 	block_group->space_info->disk_total -= block_group->key.offset * factor;
9825 
9826 	spin_unlock(&block_group->space_info->lock);
9827 
9828 	memcpy(&key, &block_group->key, sizeof(key));
9829 
9830 	lock_chunks(root);
9831 	if (!list_empty(&em->list)) {
9832 		/* We're in the transaction->pending_chunks list. */
9833 		free_extent_map(em);
9834 	}
9835 	spin_lock(&block_group->lock);
9836 	block_group->removed = 1;
9837 	/*
9838 	 * At this point trimming can't start on this block group, because we
9839 	 * removed the block group from the tree fs_info->block_group_cache_tree
9840 	 * so no one can't find it anymore and even if someone already got this
9841 	 * block group before we removed it from the rbtree, they have already
9842 	 * incremented block_group->trimming - if they didn't, they won't find
9843 	 * any free space entries because we already removed them all when we
9844 	 * called btrfs_remove_free_space_cache().
9845 	 *
9846 	 * And we must not remove the extent map from the fs_info->mapping_tree
9847 	 * to prevent the same logical address range and physical device space
9848 	 * ranges from being reused for a new block group. This is because our
9849 	 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9850 	 * completely transactionless, so while it is trimming a range the
9851 	 * currently running transaction might finish and a new one start,
9852 	 * allowing for new block groups to be created that can reuse the same
9853 	 * physical device locations unless we take this special care.
9854 	 */
9855 	remove_em = (atomic_read(&block_group->trimming) == 0);
9856 	/*
9857 	 * Make sure a trimmer task always sees the em in the pinned_chunks list
9858 	 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9859 	 * before checking block_group->removed).
9860 	 */
9861 	if (!remove_em) {
9862 		/*
9863 		 * Our em might be in trans->transaction->pending_chunks which
9864 		 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9865 		 * and so is the fs_info->pinned_chunks list.
9866 		 *
9867 		 * So at this point we must be holding the chunk_mutex to avoid
9868 		 * any races with chunk allocation (more specifically at
9869 		 * volumes.c:contains_pending_extent()), to ensure it always
9870 		 * sees the em, either in the pending_chunks list or in the
9871 		 * pinned_chunks list.
9872 		 */
9873 		list_move_tail(&em->list, &root->fs_info->pinned_chunks);
9874 	}
9875 	spin_unlock(&block_group->lock);
9876 
9877 	if (remove_em) {
9878 		struct extent_map_tree *em_tree;
9879 
9880 		em_tree = &root->fs_info->mapping_tree.map_tree;
9881 		write_lock(&em_tree->lock);
9882 		/*
9883 		 * The em might be in the pending_chunks list, so make sure the
9884 		 * chunk mutex is locked, since remove_extent_mapping() will
9885 		 * delete us from that list.
9886 		 */
9887 		remove_extent_mapping(em_tree, em);
9888 		write_unlock(&em_tree->lock);
9889 		/* once for the tree */
9890 		free_extent_map(em);
9891 	}
9892 
9893 	unlock_chunks(root);
9894 
9895 	btrfs_put_block_group(block_group);
9896 	btrfs_put_block_group(block_group);
9897 
9898 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9899 	if (ret > 0)
9900 		ret = -EIO;
9901 	if (ret < 0)
9902 		goto out;
9903 
9904 	ret = btrfs_del_item(trans, root, path);
9905 out:
9906 	btrfs_free_path(path);
9907 	return ret;
9908 }
9909 
9910 /*
9911  * Process the unused_bgs list and remove any that don't have any allocated
9912  * space inside of them.
9913  */
btrfs_delete_unused_bgs(struct btrfs_fs_info * fs_info)9914 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
9915 {
9916 	struct btrfs_block_group_cache *block_group;
9917 	struct btrfs_space_info *space_info;
9918 	struct btrfs_root *root = fs_info->extent_root;
9919 	struct btrfs_trans_handle *trans;
9920 	int ret = 0;
9921 
9922 	if (!fs_info->open)
9923 		return;
9924 
9925 	spin_lock(&fs_info->unused_bgs_lock);
9926 	while (!list_empty(&fs_info->unused_bgs)) {
9927 		u64 start, end;
9928 
9929 		block_group = list_first_entry(&fs_info->unused_bgs,
9930 					       struct btrfs_block_group_cache,
9931 					       bg_list);
9932 		space_info = block_group->space_info;
9933 		list_del_init(&block_group->bg_list);
9934 		if (ret || btrfs_mixed_space_info(space_info)) {
9935 			btrfs_put_block_group(block_group);
9936 			continue;
9937 		}
9938 		spin_unlock(&fs_info->unused_bgs_lock);
9939 
9940 		/* Don't want to race with allocators so take the groups_sem */
9941 		down_write(&space_info->groups_sem);
9942 		spin_lock(&block_group->lock);
9943 		if (block_group->reserved ||
9944 		    btrfs_block_group_used(&block_group->item) ||
9945 		    block_group->ro) {
9946 			/*
9947 			 * We want to bail if we made new allocations or have
9948 			 * outstanding allocations in this block group.  We do
9949 			 * the ro check in case balance is currently acting on
9950 			 * this block group.
9951 			 */
9952 			spin_unlock(&block_group->lock);
9953 			up_write(&space_info->groups_sem);
9954 			goto next;
9955 		}
9956 		spin_unlock(&block_group->lock);
9957 
9958 		/* We don't want to force the issue, only flip if it's ok. */
9959 		ret = set_block_group_ro(block_group, 0);
9960 		up_write(&space_info->groups_sem);
9961 		if (ret < 0) {
9962 			ret = 0;
9963 			goto next;
9964 		}
9965 
9966 		/*
9967 		 * Want to do this before we do anything else so we can recover
9968 		 * properly if we fail to join the transaction.
9969 		 */
9970 		/* 1 for btrfs_orphan_reserve_metadata() */
9971 		trans = btrfs_start_transaction(root, 1);
9972 		if (IS_ERR(trans)) {
9973 			btrfs_set_block_group_rw(root, block_group);
9974 			ret = PTR_ERR(trans);
9975 			goto next;
9976 		}
9977 
9978 		/*
9979 		 * We could have pending pinned extents for this block group,
9980 		 * just delete them, we don't care about them anymore.
9981 		 */
9982 		start = block_group->key.objectid;
9983 		end = start + block_group->key.offset - 1;
9984 		/*
9985 		 * Hold the unused_bg_unpin_mutex lock to avoid racing with
9986 		 * btrfs_finish_extent_commit(). If we are at transaction N,
9987 		 * another task might be running finish_extent_commit() for the
9988 		 * previous transaction N - 1, and have seen a range belonging
9989 		 * to the block group in freed_extents[] before we were able to
9990 		 * clear the whole block group range from freed_extents[]. This
9991 		 * means that task can lookup for the block group after we
9992 		 * unpinned it from freed_extents[] and removed it, leading to
9993 		 * a BUG_ON() at btrfs_unpin_extent_range().
9994 		 */
9995 		mutex_lock(&fs_info->unused_bg_unpin_mutex);
9996 		ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
9997 				  EXTENT_DIRTY, GFP_NOFS);
9998 		if (ret) {
9999 			mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10000 			btrfs_set_block_group_rw(root, block_group);
10001 			goto end_trans;
10002 		}
10003 		ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10004 				  EXTENT_DIRTY, GFP_NOFS);
10005 		if (ret) {
10006 			mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10007 			btrfs_set_block_group_rw(root, block_group);
10008 			goto end_trans;
10009 		}
10010 		mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10011 
10012 		/* Reset pinned so btrfs_put_block_group doesn't complain */
10013 		spin_lock(&space_info->lock);
10014 		spin_lock(&block_group->lock);
10015 
10016 		space_info->bytes_pinned -= block_group->pinned;
10017 		space_info->bytes_readonly += block_group->pinned;
10018 		percpu_counter_add(&space_info->total_bytes_pinned,
10019 				   -block_group->pinned);
10020 		block_group->pinned = 0;
10021 
10022 		spin_unlock(&block_group->lock);
10023 		spin_unlock(&space_info->lock);
10024 
10025 		/*
10026 		 * Btrfs_remove_chunk will abort the transaction if things go
10027 		 * horribly wrong.
10028 		 */
10029 		ret = btrfs_remove_chunk(trans, root,
10030 					 block_group->key.objectid);
10031 end_trans:
10032 		btrfs_end_transaction(trans, root);
10033 next:
10034 		btrfs_put_block_group(block_group);
10035 		spin_lock(&fs_info->unused_bgs_lock);
10036 	}
10037 	spin_unlock(&fs_info->unused_bgs_lock);
10038 }
10039 
btrfs_init_space_info(struct btrfs_fs_info * fs_info)10040 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10041 {
10042 	struct btrfs_space_info *space_info;
10043 	struct btrfs_super_block *disk_super;
10044 	u64 features;
10045 	u64 flags;
10046 	int mixed = 0;
10047 	int ret;
10048 
10049 	disk_super = fs_info->super_copy;
10050 	if (!btrfs_super_root(disk_super))
10051 		return 1;
10052 
10053 	features = btrfs_super_incompat_flags(disk_super);
10054 	if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10055 		mixed = 1;
10056 
10057 	flags = BTRFS_BLOCK_GROUP_SYSTEM;
10058 	ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10059 	if (ret)
10060 		goto out;
10061 
10062 	if (mixed) {
10063 		flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10064 		ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10065 	} else {
10066 		flags = BTRFS_BLOCK_GROUP_METADATA;
10067 		ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10068 		if (ret)
10069 			goto out;
10070 
10071 		flags = BTRFS_BLOCK_GROUP_DATA;
10072 		ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10073 	}
10074 out:
10075 	return ret;
10076 }
10077 
btrfs_error_unpin_extent_range(struct btrfs_root * root,u64 start,u64 end)10078 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10079 {
10080 	return unpin_extent_range(root, start, end, false);
10081 }
10082 
btrfs_trim_fs(struct btrfs_root * root,struct fstrim_range * range)10083 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10084 {
10085 	struct btrfs_fs_info *fs_info = root->fs_info;
10086 	struct btrfs_block_group_cache *cache = NULL;
10087 	u64 group_trimmed;
10088 	u64 start;
10089 	u64 end;
10090 	u64 trimmed = 0;
10091 	u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10092 	int ret = 0;
10093 
10094 	/*
10095 	 * try to trim all FS space, our block group may start from non-zero.
10096 	 */
10097 	if (range->len == total_bytes)
10098 		cache = btrfs_lookup_first_block_group(fs_info, range->start);
10099 	else
10100 		cache = btrfs_lookup_block_group(fs_info, range->start);
10101 
10102 	while (cache) {
10103 		if (cache->key.objectid >= (range->start + range->len)) {
10104 			btrfs_put_block_group(cache);
10105 			break;
10106 		}
10107 
10108 		start = max(range->start, cache->key.objectid);
10109 		end = min(range->start + range->len,
10110 				cache->key.objectid + cache->key.offset);
10111 
10112 		if (end - start >= range->minlen) {
10113 			if (!block_group_cache_done(cache)) {
10114 				ret = cache_block_group(cache, 0);
10115 				if (ret) {
10116 					btrfs_put_block_group(cache);
10117 					break;
10118 				}
10119 				ret = wait_block_group_cache_done(cache);
10120 				if (ret) {
10121 					btrfs_put_block_group(cache);
10122 					break;
10123 				}
10124 			}
10125 			ret = btrfs_trim_block_group(cache,
10126 						     &group_trimmed,
10127 						     start,
10128 						     end,
10129 						     range->minlen);
10130 
10131 			trimmed += group_trimmed;
10132 			if (ret) {
10133 				btrfs_put_block_group(cache);
10134 				break;
10135 			}
10136 		}
10137 
10138 		cache = next_block_group(fs_info->tree_root, cache);
10139 	}
10140 
10141 	range->len = trimmed;
10142 	return ret;
10143 }
10144 
10145 /*
10146  * btrfs_{start,end}_write_no_snapshoting() are similar to
10147  * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10148  * data into the page cache through nocow before the subvolume is snapshoted,
10149  * but flush the data into disk after the snapshot creation, or to prevent
10150  * operations while snapshoting is ongoing and that cause the snapshot to be
10151  * inconsistent (writes followed by expanding truncates for example).
10152  */
btrfs_end_write_no_snapshoting(struct btrfs_root * root)10153 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10154 {
10155 	percpu_counter_dec(&root->subv_writers->counter);
10156 	/*
10157 	 * Make sure counter is updated before we wake up
10158 	 * waiters.
10159 	 */
10160 	smp_mb();
10161 	if (waitqueue_active(&root->subv_writers->wait))
10162 		wake_up(&root->subv_writers->wait);
10163 }
10164 
btrfs_start_write_no_snapshoting(struct btrfs_root * root)10165 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10166 {
10167 	if (atomic_read(&root->will_be_snapshoted))
10168 		return 0;
10169 
10170 	percpu_counter_inc(&root->subv_writers->counter);
10171 	/*
10172 	 * Make sure counter is updated before we check for snapshot creation.
10173 	 */
10174 	smp_mb();
10175 	if (atomic_read(&root->will_be_snapshoted)) {
10176 		btrfs_end_write_no_snapshoting(root);
10177 		return 0;
10178 	}
10179 	return 1;
10180 }
10181