1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5 
6 #include <linux/sched.h>
7 #include <linux/sched/signal.h>
8 #include <linux/pagemap.h>
9 #include <linux/writeback.h>
10 #include <linux/blkdev.h>
11 #include <linux/sort.h>
12 #include <linux/rcupdate.h>
13 #include <linux/kthread.h>
14 #include <linux/slab.h>
15 #include <linux/ratelimit.h>
16 #include <linux/percpu_counter.h>
17 #include <linux/lockdep.h>
18 #include <linux/crc32c.h>
19 #include "misc.h"
20 #include "tree-log.h"
21 #include "disk-io.h"
22 #include "print-tree.h"
23 #include "volumes.h"
24 #include "raid56.h"
25 #include "locking.h"
26 #include "free-space-cache.h"
27 #include "free-space-tree.h"
28 #include "sysfs.h"
29 #include "qgroup.h"
30 #include "ref-verify.h"
31 #include "space-info.h"
32 #include "block-rsv.h"
33 #include "delalloc-space.h"
34 #include "block-group.h"
35 
36 #undef SCRAMBLE_DELAYED_REFS
37 
38 
39 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
40 			       struct btrfs_delayed_ref_node *node, u64 parent,
41 			       u64 root_objectid, u64 owner_objectid,
42 			       u64 owner_offset, int refs_to_drop,
43 			       struct btrfs_delayed_extent_op *extra_op);
44 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
45 				    struct extent_buffer *leaf,
46 				    struct btrfs_extent_item *ei);
47 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
48 				      u64 parent, u64 root_objectid,
49 				      u64 flags, u64 owner, u64 offset,
50 				      struct btrfs_key *ins, int ref_mod);
51 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
52 				     struct btrfs_delayed_ref_node *node,
53 				     struct btrfs_delayed_extent_op *extent_op);
54 static int find_next_key(struct btrfs_path *path, int level,
55 			 struct btrfs_key *key);
56 
block_group_bits(struct btrfs_block_group_cache * cache,u64 bits)57 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
58 {
59 	return (cache->flags & bits) == bits;
60 }
61 
btrfs_add_excluded_extent(struct btrfs_fs_info * fs_info,u64 start,u64 num_bytes)62 int btrfs_add_excluded_extent(struct btrfs_fs_info *fs_info,
63 			      u64 start, u64 num_bytes)
64 {
65 	u64 end = start + num_bytes - 1;
66 	set_extent_bits(&fs_info->freed_extents[0],
67 			start, end, EXTENT_UPTODATE);
68 	set_extent_bits(&fs_info->freed_extents[1],
69 			start, end, EXTENT_UPTODATE);
70 	return 0;
71 }
72 
btrfs_free_excluded_extents(struct btrfs_block_group_cache * cache)73 void btrfs_free_excluded_extents(struct btrfs_block_group_cache *cache)
74 {
75 	struct btrfs_fs_info *fs_info = cache->fs_info;
76 	u64 start, end;
77 
78 	start = cache->key.objectid;
79 	end = start + cache->key.offset - 1;
80 
81 	clear_extent_bits(&fs_info->freed_extents[0],
82 			  start, end, EXTENT_UPTODATE);
83 	clear_extent_bits(&fs_info->freed_extents[1],
84 			  start, end, EXTENT_UPTODATE);
85 }
86 
generic_ref_to_space_flags(struct btrfs_ref * ref)87 static u64 generic_ref_to_space_flags(struct btrfs_ref *ref)
88 {
89 	if (ref->type == BTRFS_REF_METADATA) {
90 		if (ref->tree_ref.root == BTRFS_CHUNK_TREE_OBJECTID)
91 			return BTRFS_BLOCK_GROUP_SYSTEM;
92 		else
93 			return BTRFS_BLOCK_GROUP_METADATA;
94 	}
95 	return BTRFS_BLOCK_GROUP_DATA;
96 }
97 
add_pinned_bytes(struct btrfs_fs_info * fs_info,struct btrfs_ref * ref)98 static void add_pinned_bytes(struct btrfs_fs_info *fs_info,
99 			     struct btrfs_ref *ref)
100 {
101 	struct btrfs_space_info *space_info;
102 	u64 flags = generic_ref_to_space_flags(ref);
103 
104 	space_info = btrfs_find_space_info(fs_info, flags);
105 	ASSERT(space_info);
106 	percpu_counter_add_batch(&space_info->total_bytes_pinned, ref->len,
107 		    BTRFS_TOTAL_BYTES_PINNED_BATCH);
108 }
109 
sub_pinned_bytes(struct btrfs_fs_info * fs_info,struct btrfs_ref * ref)110 static void sub_pinned_bytes(struct btrfs_fs_info *fs_info,
111 			     struct btrfs_ref *ref)
112 {
113 	struct btrfs_space_info *space_info;
114 	u64 flags = generic_ref_to_space_flags(ref);
115 
116 	space_info = btrfs_find_space_info(fs_info, flags);
117 	ASSERT(space_info);
118 	percpu_counter_add_batch(&space_info->total_bytes_pinned, -ref->len,
119 		    BTRFS_TOTAL_BYTES_PINNED_BATCH);
120 }
121 
122 /* simple helper to search for an existing data extent at a given offset */
btrfs_lookup_data_extent(struct btrfs_fs_info * fs_info,u64 start,u64 len)123 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
124 {
125 	int ret;
126 	struct btrfs_key key;
127 	struct btrfs_path *path;
128 
129 	path = btrfs_alloc_path();
130 	if (!path)
131 		return -ENOMEM;
132 
133 	key.objectid = start;
134 	key.offset = len;
135 	key.type = BTRFS_EXTENT_ITEM_KEY;
136 	ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
137 	btrfs_free_path(path);
138 	return ret;
139 }
140 
141 /*
142  * helper function to lookup reference count and flags of a tree block.
143  *
144  * the head node for delayed ref is used to store the sum of all the
145  * reference count modifications queued up in the rbtree. the head
146  * node may also store the extent flags to set. This way you can check
147  * to see what the reference count and extent flags would be if all of
148  * the delayed refs are not processed.
149  */
btrfs_lookup_extent_info(struct btrfs_trans_handle * trans,struct btrfs_fs_info * fs_info,u64 bytenr,u64 offset,int metadata,u64 * refs,u64 * flags)150 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
151 			     struct btrfs_fs_info *fs_info, u64 bytenr,
152 			     u64 offset, int metadata, u64 *refs, u64 *flags)
153 {
154 	struct btrfs_delayed_ref_head *head;
155 	struct btrfs_delayed_ref_root *delayed_refs;
156 	struct btrfs_path *path;
157 	struct btrfs_extent_item *ei;
158 	struct extent_buffer *leaf;
159 	struct btrfs_key key;
160 	u32 item_size;
161 	u64 num_refs;
162 	u64 extent_flags;
163 	int ret;
164 
165 	/*
166 	 * If we don't have skinny metadata, don't bother doing anything
167 	 * different
168 	 */
169 	if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
170 		offset = fs_info->nodesize;
171 		metadata = 0;
172 	}
173 
174 	path = btrfs_alloc_path();
175 	if (!path)
176 		return -ENOMEM;
177 
178 	if (!trans) {
179 		path->skip_locking = 1;
180 		path->search_commit_root = 1;
181 	}
182 
183 search_again:
184 	key.objectid = bytenr;
185 	key.offset = offset;
186 	if (metadata)
187 		key.type = BTRFS_METADATA_ITEM_KEY;
188 	else
189 		key.type = BTRFS_EXTENT_ITEM_KEY;
190 
191 	ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
192 	if (ret < 0)
193 		goto out_free;
194 
195 	if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
196 		if (path->slots[0]) {
197 			path->slots[0]--;
198 			btrfs_item_key_to_cpu(path->nodes[0], &key,
199 					      path->slots[0]);
200 			if (key.objectid == bytenr &&
201 			    key.type == BTRFS_EXTENT_ITEM_KEY &&
202 			    key.offset == fs_info->nodesize)
203 				ret = 0;
204 		}
205 	}
206 
207 	if (ret == 0) {
208 		leaf = path->nodes[0];
209 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
210 		if (item_size >= sizeof(*ei)) {
211 			ei = btrfs_item_ptr(leaf, path->slots[0],
212 					    struct btrfs_extent_item);
213 			num_refs = btrfs_extent_refs(leaf, ei);
214 			extent_flags = btrfs_extent_flags(leaf, ei);
215 		} else {
216 			ret = -EINVAL;
217 			btrfs_print_v0_err(fs_info);
218 			if (trans)
219 				btrfs_abort_transaction(trans, ret);
220 			else
221 				btrfs_handle_fs_error(fs_info, ret, NULL);
222 
223 			goto out_free;
224 		}
225 
226 		BUG_ON(num_refs == 0);
227 	} else {
228 		num_refs = 0;
229 		extent_flags = 0;
230 		ret = 0;
231 	}
232 
233 	if (!trans)
234 		goto out;
235 
236 	delayed_refs = &trans->transaction->delayed_refs;
237 	spin_lock(&delayed_refs->lock);
238 	head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
239 	if (head) {
240 		if (!mutex_trylock(&head->mutex)) {
241 			refcount_inc(&head->refs);
242 			spin_unlock(&delayed_refs->lock);
243 
244 			btrfs_release_path(path);
245 
246 			/*
247 			 * Mutex was contended, block until it's released and try
248 			 * again
249 			 */
250 			mutex_lock(&head->mutex);
251 			mutex_unlock(&head->mutex);
252 			btrfs_put_delayed_ref_head(head);
253 			goto search_again;
254 		}
255 		spin_lock(&head->lock);
256 		if (head->extent_op && head->extent_op->update_flags)
257 			extent_flags |= head->extent_op->flags_to_set;
258 		else
259 			BUG_ON(num_refs == 0);
260 
261 		num_refs += head->ref_mod;
262 		spin_unlock(&head->lock);
263 		mutex_unlock(&head->mutex);
264 	}
265 	spin_unlock(&delayed_refs->lock);
266 out:
267 	WARN_ON(num_refs == 0);
268 	if (refs)
269 		*refs = num_refs;
270 	if (flags)
271 		*flags = extent_flags;
272 out_free:
273 	btrfs_free_path(path);
274 	return ret;
275 }
276 
277 /*
278  * Back reference rules.  Back refs have three main goals:
279  *
280  * 1) differentiate between all holders of references to an extent so that
281  *    when a reference is dropped we can make sure it was a valid reference
282  *    before freeing the extent.
283  *
284  * 2) Provide enough information to quickly find the holders of an extent
285  *    if we notice a given block is corrupted or bad.
286  *
287  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
288  *    maintenance.  This is actually the same as #2, but with a slightly
289  *    different use case.
290  *
291  * There are two kinds of back refs. The implicit back refs is optimized
292  * for pointers in non-shared tree blocks. For a given pointer in a block,
293  * back refs of this kind provide information about the block's owner tree
294  * and the pointer's key. These information allow us to find the block by
295  * b-tree searching. The full back refs is for pointers in tree blocks not
296  * referenced by their owner trees. The location of tree block is recorded
297  * in the back refs. Actually the full back refs is generic, and can be
298  * used in all cases the implicit back refs is used. The major shortcoming
299  * of the full back refs is its overhead. Every time a tree block gets
300  * COWed, we have to update back refs entry for all pointers in it.
301  *
302  * For a newly allocated tree block, we use implicit back refs for
303  * pointers in it. This means most tree related operations only involve
304  * implicit back refs. For a tree block created in old transaction, the
305  * only way to drop a reference to it is COW it. So we can detect the
306  * event that tree block loses its owner tree's reference and do the
307  * back refs conversion.
308  *
309  * When a tree block is COWed through a tree, there are four cases:
310  *
311  * The reference count of the block is one and the tree is the block's
312  * owner tree. Nothing to do in this case.
313  *
314  * The reference count of the block is one and the tree is not the
315  * block's owner tree. In this case, full back refs is used for pointers
316  * in the block. Remove these full back refs, add implicit back refs for
317  * every pointers in the new block.
318  *
319  * The reference count of the block is greater than one and the tree is
320  * the block's owner tree. In this case, implicit back refs is used for
321  * pointers in the block. Add full back refs for every pointers in the
322  * block, increase lower level extents' reference counts. The original
323  * implicit back refs are entailed to the new block.
324  *
325  * The reference count of the block is greater than one and the tree is
326  * not the block's owner tree. Add implicit back refs for every pointer in
327  * the new block, increase lower level extents' reference count.
328  *
329  * Back Reference Key composing:
330  *
331  * The key objectid corresponds to the first byte in the extent,
332  * The key type is used to differentiate between types of back refs.
333  * There are different meanings of the key offset for different types
334  * of back refs.
335  *
336  * File extents can be referenced by:
337  *
338  * - multiple snapshots, subvolumes, or different generations in one subvol
339  * - different files inside a single subvolume
340  * - different offsets inside a file (bookend extents in file.c)
341  *
342  * The extent ref structure for the implicit back refs has fields for:
343  *
344  * - Objectid of the subvolume root
345  * - objectid of the file holding the reference
346  * - original offset in the file
347  * - how many bookend extents
348  *
349  * The key offset for the implicit back refs is hash of the first
350  * three fields.
351  *
352  * The extent ref structure for the full back refs has field for:
353  *
354  * - number of pointers in the tree leaf
355  *
356  * The key offset for the implicit back refs is the first byte of
357  * the tree leaf
358  *
359  * When a file extent is allocated, The implicit back refs is used.
360  * the fields are filled in:
361  *
362  *     (root_key.objectid, inode objectid, offset in file, 1)
363  *
364  * When a file extent is removed file truncation, we find the
365  * corresponding implicit back refs and check the following fields:
366  *
367  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
368  *
369  * Btree extents can be referenced by:
370  *
371  * - Different subvolumes
372  *
373  * Both the implicit back refs and the full back refs for tree blocks
374  * only consist of key. The key offset for the implicit back refs is
375  * objectid of block's owner tree. The key offset for the full back refs
376  * is the first byte of parent block.
377  *
378  * When implicit back refs is used, information about the lowest key and
379  * level of the tree block are required. These information are stored in
380  * tree block info structure.
381  */
382 
383 /*
384  * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
385  * is_data == BTRFS_REF_TYPE_DATA, data type is requiried,
386  * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
387  */
btrfs_get_extent_inline_ref_type(const struct extent_buffer * eb,struct btrfs_extent_inline_ref * iref,enum btrfs_inline_ref_type is_data)388 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
389 				     struct btrfs_extent_inline_ref *iref,
390 				     enum btrfs_inline_ref_type is_data)
391 {
392 	int type = btrfs_extent_inline_ref_type(eb, iref);
393 	u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
394 
395 	if (type == BTRFS_TREE_BLOCK_REF_KEY ||
396 	    type == BTRFS_SHARED_BLOCK_REF_KEY ||
397 	    type == BTRFS_SHARED_DATA_REF_KEY ||
398 	    type == BTRFS_EXTENT_DATA_REF_KEY) {
399 		if (is_data == BTRFS_REF_TYPE_BLOCK) {
400 			if (type == BTRFS_TREE_BLOCK_REF_KEY)
401 				return type;
402 			if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
403 				ASSERT(eb->fs_info);
404 				/*
405 				 * Every shared one has parent tree
406 				 * block, which must be aligned to
407 				 * nodesize.
408 				 */
409 				if (offset &&
410 				    IS_ALIGNED(offset, eb->fs_info->nodesize))
411 					return type;
412 			}
413 		} else if (is_data == BTRFS_REF_TYPE_DATA) {
414 			if (type == BTRFS_EXTENT_DATA_REF_KEY)
415 				return type;
416 			if (type == BTRFS_SHARED_DATA_REF_KEY) {
417 				ASSERT(eb->fs_info);
418 				/*
419 				 * Every shared one has parent tree
420 				 * block, which must be aligned to
421 				 * nodesize.
422 				 */
423 				if (offset &&
424 				    IS_ALIGNED(offset, eb->fs_info->nodesize))
425 					return type;
426 			}
427 		} else {
428 			ASSERT(is_data == BTRFS_REF_TYPE_ANY);
429 			return type;
430 		}
431 	}
432 
433 	btrfs_print_leaf((struct extent_buffer *)eb);
434 	btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d",
435 		  eb->start, type);
436 	WARN_ON(1);
437 
438 	return BTRFS_REF_TYPE_INVALID;
439 }
440 
hash_extent_data_ref(u64 root_objectid,u64 owner,u64 offset)441 u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
442 {
443 	u32 high_crc = ~(u32)0;
444 	u32 low_crc = ~(u32)0;
445 	__le64 lenum;
446 
447 	lenum = cpu_to_le64(root_objectid);
448 	high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
449 	lenum = cpu_to_le64(owner);
450 	low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
451 	lenum = cpu_to_le64(offset);
452 	low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
453 
454 	return ((u64)high_crc << 31) ^ (u64)low_crc;
455 }
456 
hash_extent_data_ref_item(struct extent_buffer * leaf,struct btrfs_extent_data_ref * ref)457 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
458 				     struct btrfs_extent_data_ref *ref)
459 {
460 	return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
461 				    btrfs_extent_data_ref_objectid(leaf, ref),
462 				    btrfs_extent_data_ref_offset(leaf, ref));
463 }
464 
match_extent_data_ref(struct extent_buffer * leaf,struct btrfs_extent_data_ref * ref,u64 root_objectid,u64 owner,u64 offset)465 static int match_extent_data_ref(struct extent_buffer *leaf,
466 				 struct btrfs_extent_data_ref *ref,
467 				 u64 root_objectid, u64 owner, u64 offset)
468 {
469 	if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
470 	    btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
471 	    btrfs_extent_data_ref_offset(leaf, ref) != offset)
472 		return 0;
473 	return 1;
474 }
475 
lookup_extent_data_ref(struct btrfs_trans_handle * trans,struct btrfs_path * path,u64 bytenr,u64 parent,u64 root_objectid,u64 owner,u64 offset)476 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
477 					   struct btrfs_path *path,
478 					   u64 bytenr, u64 parent,
479 					   u64 root_objectid,
480 					   u64 owner, u64 offset)
481 {
482 	struct btrfs_root *root = trans->fs_info->extent_root;
483 	struct btrfs_key key;
484 	struct btrfs_extent_data_ref *ref;
485 	struct extent_buffer *leaf;
486 	u32 nritems;
487 	int ret;
488 	int recow;
489 	int err = -ENOENT;
490 
491 	key.objectid = bytenr;
492 	if (parent) {
493 		key.type = BTRFS_SHARED_DATA_REF_KEY;
494 		key.offset = parent;
495 	} else {
496 		key.type = BTRFS_EXTENT_DATA_REF_KEY;
497 		key.offset = hash_extent_data_ref(root_objectid,
498 						  owner, offset);
499 	}
500 again:
501 	recow = 0;
502 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
503 	if (ret < 0) {
504 		err = ret;
505 		goto fail;
506 	}
507 
508 	if (parent) {
509 		if (!ret)
510 			return 0;
511 		goto fail;
512 	}
513 
514 	leaf = path->nodes[0];
515 	nritems = btrfs_header_nritems(leaf);
516 	while (1) {
517 		if (path->slots[0] >= nritems) {
518 			ret = btrfs_next_leaf(root, path);
519 			if (ret < 0)
520 				err = ret;
521 			if (ret)
522 				goto fail;
523 
524 			leaf = path->nodes[0];
525 			nritems = btrfs_header_nritems(leaf);
526 			recow = 1;
527 		}
528 
529 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
530 		if (key.objectid != bytenr ||
531 		    key.type != BTRFS_EXTENT_DATA_REF_KEY)
532 			goto fail;
533 
534 		ref = btrfs_item_ptr(leaf, path->slots[0],
535 				     struct btrfs_extent_data_ref);
536 
537 		if (match_extent_data_ref(leaf, ref, root_objectid,
538 					  owner, offset)) {
539 			if (recow) {
540 				btrfs_release_path(path);
541 				goto again;
542 			}
543 			err = 0;
544 			break;
545 		}
546 		path->slots[0]++;
547 	}
548 fail:
549 	return err;
550 }
551 
insert_extent_data_ref(struct btrfs_trans_handle * trans,struct btrfs_path * path,u64 bytenr,u64 parent,u64 root_objectid,u64 owner,u64 offset,int refs_to_add)552 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
553 					   struct btrfs_path *path,
554 					   u64 bytenr, u64 parent,
555 					   u64 root_objectid, u64 owner,
556 					   u64 offset, int refs_to_add)
557 {
558 	struct btrfs_root *root = trans->fs_info->extent_root;
559 	struct btrfs_key key;
560 	struct extent_buffer *leaf;
561 	u32 size;
562 	u32 num_refs;
563 	int ret;
564 
565 	key.objectid = bytenr;
566 	if (parent) {
567 		key.type = BTRFS_SHARED_DATA_REF_KEY;
568 		key.offset = parent;
569 		size = sizeof(struct btrfs_shared_data_ref);
570 	} else {
571 		key.type = BTRFS_EXTENT_DATA_REF_KEY;
572 		key.offset = hash_extent_data_ref(root_objectid,
573 						  owner, offset);
574 		size = sizeof(struct btrfs_extent_data_ref);
575 	}
576 
577 	ret = btrfs_insert_empty_item(trans, root, path, &key, size);
578 	if (ret && ret != -EEXIST)
579 		goto fail;
580 
581 	leaf = path->nodes[0];
582 	if (parent) {
583 		struct btrfs_shared_data_ref *ref;
584 		ref = btrfs_item_ptr(leaf, path->slots[0],
585 				     struct btrfs_shared_data_ref);
586 		if (ret == 0) {
587 			btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
588 		} else {
589 			num_refs = btrfs_shared_data_ref_count(leaf, ref);
590 			num_refs += refs_to_add;
591 			btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
592 		}
593 	} else {
594 		struct btrfs_extent_data_ref *ref;
595 		while (ret == -EEXIST) {
596 			ref = btrfs_item_ptr(leaf, path->slots[0],
597 					     struct btrfs_extent_data_ref);
598 			if (match_extent_data_ref(leaf, ref, root_objectid,
599 						  owner, offset))
600 				break;
601 			btrfs_release_path(path);
602 			key.offset++;
603 			ret = btrfs_insert_empty_item(trans, root, path, &key,
604 						      size);
605 			if (ret && ret != -EEXIST)
606 				goto fail;
607 
608 			leaf = path->nodes[0];
609 		}
610 		ref = btrfs_item_ptr(leaf, path->slots[0],
611 				     struct btrfs_extent_data_ref);
612 		if (ret == 0) {
613 			btrfs_set_extent_data_ref_root(leaf, ref,
614 						       root_objectid);
615 			btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
616 			btrfs_set_extent_data_ref_offset(leaf, ref, offset);
617 			btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
618 		} else {
619 			num_refs = btrfs_extent_data_ref_count(leaf, ref);
620 			num_refs += refs_to_add;
621 			btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
622 		}
623 	}
624 	btrfs_mark_buffer_dirty(leaf);
625 	ret = 0;
626 fail:
627 	btrfs_release_path(path);
628 	return ret;
629 }
630 
remove_extent_data_ref(struct btrfs_trans_handle * trans,struct btrfs_path * path,int refs_to_drop,int * last_ref)631 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
632 					   struct btrfs_path *path,
633 					   int refs_to_drop, int *last_ref)
634 {
635 	struct btrfs_key key;
636 	struct btrfs_extent_data_ref *ref1 = NULL;
637 	struct btrfs_shared_data_ref *ref2 = NULL;
638 	struct extent_buffer *leaf;
639 	u32 num_refs = 0;
640 	int ret = 0;
641 
642 	leaf = path->nodes[0];
643 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
644 
645 	if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
646 		ref1 = btrfs_item_ptr(leaf, path->slots[0],
647 				      struct btrfs_extent_data_ref);
648 		num_refs = btrfs_extent_data_ref_count(leaf, ref1);
649 	} else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
650 		ref2 = btrfs_item_ptr(leaf, path->slots[0],
651 				      struct btrfs_shared_data_ref);
652 		num_refs = btrfs_shared_data_ref_count(leaf, ref2);
653 	} else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
654 		btrfs_print_v0_err(trans->fs_info);
655 		btrfs_abort_transaction(trans, -EINVAL);
656 		return -EINVAL;
657 	} else {
658 		BUG();
659 	}
660 
661 	BUG_ON(num_refs < refs_to_drop);
662 	num_refs -= refs_to_drop;
663 
664 	if (num_refs == 0) {
665 		ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
666 		*last_ref = 1;
667 	} else {
668 		if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
669 			btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
670 		else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
671 			btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
672 		btrfs_mark_buffer_dirty(leaf);
673 	}
674 	return ret;
675 }
676 
extent_data_ref_count(struct btrfs_path * path,struct btrfs_extent_inline_ref * iref)677 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
678 					  struct btrfs_extent_inline_ref *iref)
679 {
680 	struct btrfs_key key;
681 	struct extent_buffer *leaf;
682 	struct btrfs_extent_data_ref *ref1;
683 	struct btrfs_shared_data_ref *ref2;
684 	u32 num_refs = 0;
685 	int type;
686 
687 	leaf = path->nodes[0];
688 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
689 
690 	BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
691 	if (iref) {
692 		/*
693 		 * If type is invalid, we should have bailed out earlier than
694 		 * this call.
695 		 */
696 		type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
697 		ASSERT(type != BTRFS_REF_TYPE_INVALID);
698 		if (type == BTRFS_EXTENT_DATA_REF_KEY) {
699 			ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
700 			num_refs = btrfs_extent_data_ref_count(leaf, ref1);
701 		} else {
702 			ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
703 			num_refs = btrfs_shared_data_ref_count(leaf, ref2);
704 		}
705 	} else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
706 		ref1 = btrfs_item_ptr(leaf, path->slots[0],
707 				      struct btrfs_extent_data_ref);
708 		num_refs = btrfs_extent_data_ref_count(leaf, ref1);
709 	} else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
710 		ref2 = btrfs_item_ptr(leaf, path->slots[0],
711 				      struct btrfs_shared_data_ref);
712 		num_refs = btrfs_shared_data_ref_count(leaf, ref2);
713 	} else {
714 		WARN_ON(1);
715 	}
716 	return num_refs;
717 }
718 
lookup_tree_block_ref(struct btrfs_trans_handle * trans,struct btrfs_path * path,u64 bytenr,u64 parent,u64 root_objectid)719 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
720 					  struct btrfs_path *path,
721 					  u64 bytenr, u64 parent,
722 					  u64 root_objectid)
723 {
724 	struct btrfs_root *root = trans->fs_info->extent_root;
725 	struct btrfs_key key;
726 	int ret;
727 
728 	key.objectid = bytenr;
729 	if (parent) {
730 		key.type = BTRFS_SHARED_BLOCK_REF_KEY;
731 		key.offset = parent;
732 	} else {
733 		key.type = BTRFS_TREE_BLOCK_REF_KEY;
734 		key.offset = root_objectid;
735 	}
736 
737 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
738 	if (ret > 0)
739 		ret = -ENOENT;
740 	return ret;
741 }
742 
insert_tree_block_ref(struct btrfs_trans_handle * trans,struct btrfs_path * path,u64 bytenr,u64 parent,u64 root_objectid)743 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
744 					  struct btrfs_path *path,
745 					  u64 bytenr, u64 parent,
746 					  u64 root_objectid)
747 {
748 	struct btrfs_key key;
749 	int ret;
750 
751 	key.objectid = bytenr;
752 	if (parent) {
753 		key.type = BTRFS_SHARED_BLOCK_REF_KEY;
754 		key.offset = parent;
755 	} else {
756 		key.type = BTRFS_TREE_BLOCK_REF_KEY;
757 		key.offset = root_objectid;
758 	}
759 
760 	ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root,
761 				      path, &key, 0);
762 	btrfs_release_path(path);
763 	return ret;
764 }
765 
extent_ref_type(u64 parent,u64 owner)766 static inline int extent_ref_type(u64 parent, u64 owner)
767 {
768 	int type;
769 	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
770 		if (parent > 0)
771 			type = BTRFS_SHARED_BLOCK_REF_KEY;
772 		else
773 			type = BTRFS_TREE_BLOCK_REF_KEY;
774 	} else {
775 		if (parent > 0)
776 			type = BTRFS_SHARED_DATA_REF_KEY;
777 		else
778 			type = BTRFS_EXTENT_DATA_REF_KEY;
779 	}
780 	return type;
781 }
782 
find_next_key(struct btrfs_path * path,int level,struct btrfs_key * key)783 static int find_next_key(struct btrfs_path *path, int level,
784 			 struct btrfs_key *key)
785 
786 {
787 	for (; level < BTRFS_MAX_LEVEL; level++) {
788 		if (!path->nodes[level])
789 			break;
790 		if (path->slots[level] + 1 >=
791 		    btrfs_header_nritems(path->nodes[level]))
792 			continue;
793 		if (level == 0)
794 			btrfs_item_key_to_cpu(path->nodes[level], key,
795 					      path->slots[level] + 1);
796 		else
797 			btrfs_node_key_to_cpu(path->nodes[level], key,
798 					      path->slots[level] + 1);
799 		return 0;
800 	}
801 	return 1;
802 }
803 
804 /*
805  * look for inline back ref. if back ref is found, *ref_ret is set
806  * to the address of inline back ref, and 0 is returned.
807  *
808  * if back ref isn't found, *ref_ret is set to the address where it
809  * should be inserted, and -ENOENT is returned.
810  *
811  * if insert is true and there are too many inline back refs, the path
812  * points to the extent item, and -EAGAIN is returned.
813  *
814  * NOTE: inline back refs are ordered in the same way that back ref
815  *	 items in the tree are ordered.
816  */
817 static noinline_for_stack
lookup_inline_extent_backref(struct btrfs_trans_handle * trans,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)818 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
819 				 struct btrfs_path *path,
820 				 struct btrfs_extent_inline_ref **ref_ret,
821 				 u64 bytenr, u64 num_bytes,
822 				 u64 parent, u64 root_objectid,
823 				 u64 owner, u64 offset, int insert)
824 {
825 	struct btrfs_fs_info *fs_info = trans->fs_info;
826 	struct btrfs_root *root = fs_info->extent_root;
827 	struct btrfs_key key;
828 	struct extent_buffer *leaf;
829 	struct btrfs_extent_item *ei;
830 	struct btrfs_extent_inline_ref *iref;
831 	u64 flags;
832 	u64 item_size;
833 	unsigned long ptr;
834 	unsigned long end;
835 	int extra_size;
836 	int type;
837 	int want;
838 	int ret;
839 	int err = 0;
840 	bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
841 	int needed;
842 
843 	key.objectid = bytenr;
844 	key.type = BTRFS_EXTENT_ITEM_KEY;
845 	key.offset = num_bytes;
846 
847 	want = extent_ref_type(parent, owner);
848 	if (insert) {
849 		extra_size = btrfs_extent_inline_ref_size(want);
850 		path->keep_locks = 1;
851 	} else
852 		extra_size = -1;
853 
854 	/*
855 	 * Owner is our level, so we can just add one to get the level for the
856 	 * block we are interested in.
857 	 */
858 	if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
859 		key.type = BTRFS_METADATA_ITEM_KEY;
860 		key.offset = owner;
861 	}
862 
863 again:
864 	ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
865 	if (ret < 0) {
866 		err = ret;
867 		goto out;
868 	}
869 
870 	/*
871 	 * We may be a newly converted file system which still has the old fat
872 	 * extent entries for metadata, so try and see if we have one of those.
873 	 */
874 	if (ret > 0 && skinny_metadata) {
875 		skinny_metadata = false;
876 		if (path->slots[0]) {
877 			path->slots[0]--;
878 			btrfs_item_key_to_cpu(path->nodes[0], &key,
879 					      path->slots[0]);
880 			if (key.objectid == bytenr &&
881 			    key.type == BTRFS_EXTENT_ITEM_KEY &&
882 			    key.offset == num_bytes)
883 				ret = 0;
884 		}
885 		if (ret) {
886 			key.objectid = bytenr;
887 			key.type = BTRFS_EXTENT_ITEM_KEY;
888 			key.offset = num_bytes;
889 			btrfs_release_path(path);
890 			goto again;
891 		}
892 	}
893 
894 	if (ret && !insert) {
895 		err = -ENOENT;
896 		goto out;
897 	} else if (WARN_ON(ret)) {
898 		err = -EIO;
899 		goto out;
900 	}
901 
902 	leaf = path->nodes[0];
903 	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
904 	if (unlikely(item_size < sizeof(*ei))) {
905 		err = -EINVAL;
906 		btrfs_print_v0_err(fs_info);
907 		btrfs_abort_transaction(trans, err);
908 		goto out;
909 	}
910 
911 	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
912 	flags = btrfs_extent_flags(leaf, ei);
913 
914 	ptr = (unsigned long)(ei + 1);
915 	end = (unsigned long)ei + item_size;
916 
917 	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
918 		ptr += sizeof(struct btrfs_tree_block_info);
919 		BUG_ON(ptr > end);
920 	}
921 
922 	if (owner >= BTRFS_FIRST_FREE_OBJECTID)
923 		needed = BTRFS_REF_TYPE_DATA;
924 	else
925 		needed = BTRFS_REF_TYPE_BLOCK;
926 
927 	err = -ENOENT;
928 	while (1) {
929 		if (ptr >= end) {
930 			WARN_ON(ptr > end);
931 			break;
932 		}
933 		iref = (struct btrfs_extent_inline_ref *)ptr;
934 		type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
935 		if (type == BTRFS_REF_TYPE_INVALID) {
936 			err = -EUCLEAN;
937 			goto out;
938 		}
939 
940 		if (want < type)
941 			break;
942 		if (want > type) {
943 			ptr += btrfs_extent_inline_ref_size(type);
944 			continue;
945 		}
946 
947 		if (type == BTRFS_EXTENT_DATA_REF_KEY) {
948 			struct btrfs_extent_data_ref *dref;
949 			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
950 			if (match_extent_data_ref(leaf, dref, root_objectid,
951 						  owner, offset)) {
952 				err = 0;
953 				break;
954 			}
955 			if (hash_extent_data_ref_item(leaf, dref) <
956 			    hash_extent_data_ref(root_objectid, owner, offset))
957 				break;
958 		} else {
959 			u64 ref_offset;
960 			ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
961 			if (parent > 0) {
962 				if (parent == ref_offset) {
963 					err = 0;
964 					break;
965 				}
966 				if (ref_offset < parent)
967 					break;
968 			} else {
969 				if (root_objectid == ref_offset) {
970 					err = 0;
971 					break;
972 				}
973 				if (ref_offset < root_objectid)
974 					break;
975 			}
976 		}
977 		ptr += btrfs_extent_inline_ref_size(type);
978 	}
979 	if (err == -ENOENT && insert) {
980 		if (item_size + extra_size >=
981 		    BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
982 			err = -EAGAIN;
983 			goto out;
984 		}
985 		/*
986 		 * To add new inline back ref, we have to make sure
987 		 * there is no corresponding back ref item.
988 		 * For simplicity, we just do not add new inline back
989 		 * ref if there is any kind of item for this block
990 		 */
991 		if (find_next_key(path, 0, &key) == 0 &&
992 		    key.objectid == bytenr &&
993 		    key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
994 			err = -EAGAIN;
995 			goto out;
996 		}
997 	}
998 	*ref_ret = (struct btrfs_extent_inline_ref *)ptr;
999 out:
1000 	if (insert) {
1001 		path->keep_locks = 0;
1002 		btrfs_unlock_up_safe(path, 1);
1003 	}
1004 	return err;
1005 }
1006 
1007 /*
1008  * helper to add new inline back ref
1009  */
1010 static noinline_for_stack
setup_inline_extent_backref(struct btrfs_fs_info * fs_info,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)1011 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1012 				 struct btrfs_path *path,
1013 				 struct btrfs_extent_inline_ref *iref,
1014 				 u64 parent, u64 root_objectid,
1015 				 u64 owner, u64 offset, int refs_to_add,
1016 				 struct btrfs_delayed_extent_op *extent_op)
1017 {
1018 	struct extent_buffer *leaf;
1019 	struct btrfs_extent_item *ei;
1020 	unsigned long ptr;
1021 	unsigned long end;
1022 	unsigned long item_offset;
1023 	u64 refs;
1024 	int size;
1025 	int type;
1026 
1027 	leaf = path->nodes[0];
1028 	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1029 	item_offset = (unsigned long)iref - (unsigned long)ei;
1030 
1031 	type = extent_ref_type(parent, owner);
1032 	size = btrfs_extent_inline_ref_size(type);
1033 
1034 	btrfs_extend_item(path, size);
1035 
1036 	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1037 	refs = btrfs_extent_refs(leaf, ei);
1038 	refs += refs_to_add;
1039 	btrfs_set_extent_refs(leaf, ei, refs);
1040 	if (extent_op)
1041 		__run_delayed_extent_op(extent_op, leaf, ei);
1042 
1043 	ptr = (unsigned long)ei + item_offset;
1044 	end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1045 	if (ptr < end - size)
1046 		memmove_extent_buffer(leaf, ptr + size, ptr,
1047 				      end - size - ptr);
1048 
1049 	iref = (struct btrfs_extent_inline_ref *)ptr;
1050 	btrfs_set_extent_inline_ref_type(leaf, iref, type);
1051 	if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1052 		struct btrfs_extent_data_ref *dref;
1053 		dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1054 		btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1055 		btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1056 		btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1057 		btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1058 	} else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1059 		struct btrfs_shared_data_ref *sref;
1060 		sref = (struct btrfs_shared_data_ref *)(iref + 1);
1061 		btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1062 		btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1063 	} else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1064 		btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1065 	} else {
1066 		btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1067 	}
1068 	btrfs_mark_buffer_dirty(leaf);
1069 }
1070 
lookup_extent_backref(struct btrfs_trans_handle * trans,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)1071 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1072 				 struct btrfs_path *path,
1073 				 struct btrfs_extent_inline_ref **ref_ret,
1074 				 u64 bytenr, u64 num_bytes, u64 parent,
1075 				 u64 root_objectid, u64 owner, u64 offset)
1076 {
1077 	int ret;
1078 
1079 	ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1080 					   num_bytes, parent, root_objectid,
1081 					   owner, offset, 0);
1082 	if (ret != -ENOENT)
1083 		return ret;
1084 
1085 	btrfs_release_path(path);
1086 	*ref_ret = NULL;
1087 
1088 	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1089 		ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1090 					    root_objectid);
1091 	} else {
1092 		ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1093 					     root_objectid, owner, offset);
1094 	}
1095 	return ret;
1096 }
1097 
1098 /*
1099  * helper to update/remove inline back ref
1100  */
1101 static noinline_for_stack
update_inline_extent_backref(struct btrfs_path * path,struct btrfs_extent_inline_ref * iref,int refs_to_mod,struct btrfs_delayed_extent_op * extent_op,int * last_ref)1102 void update_inline_extent_backref(struct btrfs_path *path,
1103 				  struct btrfs_extent_inline_ref *iref,
1104 				  int refs_to_mod,
1105 				  struct btrfs_delayed_extent_op *extent_op,
1106 				  int *last_ref)
1107 {
1108 	struct extent_buffer *leaf = path->nodes[0];
1109 	struct btrfs_extent_item *ei;
1110 	struct btrfs_extent_data_ref *dref = NULL;
1111 	struct btrfs_shared_data_ref *sref = NULL;
1112 	unsigned long ptr;
1113 	unsigned long end;
1114 	u32 item_size;
1115 	int size;
1116 	int type;
1117 	u64 refs;
1118 
1119 	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1120 	refs = btrfs_extent_refs(leaf, ei);
1121 	WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1122 	refs += refs_to_mod;
1123 	btrfs_set_extent_refs(leaf, ei, refs);
1124 	if (extent_op)
1125 		__run_delayed_extent_op(extent_op, leaf, ei);
1126 
1127 	/*
1128 	 * If type is invalid, we should have bailed out after
1129 	 * lookup_inline_extent_backref().
1130 	 */
1131 	type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1132 	ASSERT(type != BTRFS_REF_TYPE_INVALID);
1133 
1134 	if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1135 		dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1136 		refs = btrfs_extent_data_ref_count(leaf, dref);
1137 	} else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1138 		sref = (struct btrfs_shared_data_ref *)(iref + 1);
1139 		refs = btrfs_shared_data_ref_count(leaf, sref);
1140 	} else {
1141 		refs = 1;
1142 		BUG_ON(refs_to_mod != -1);
1143 	}
1144 
1145 	BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1146 	refs += refs_to_mod;
1147 
1148 	if (refs > 0) {
1149 		if (type == BTRFS_EXTENT_DATA_REF_KEY)
1150 			btrfs_set_extent_data_ref_count(leaf, dref, refs);
1151 		else
1152 			btrfs_set_shared_data_ref_count(leaf, sref, refs);
1153 	} else {
1154 		*last_ref = 1;
1155 		size =  btrfs_extent_inline_ref_size(type);
1156 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1157 		ptr = (unsigned long)iref;
1158 		end = (unsigned long)ei + item_size;
1159 		if (ptr + size < end)
1160 			memmove_extent_buffer(leaf, ptr, ptr + size,
1161 					      end - ptr - size);
1162 		item_size -= size;
1163 		btrfs_truncate_item(path, item_size, 1);
1164 	}
1165 	btrfs_mark_buffer_dirty(leaf);
1166 }
1167 
1168 static noinline_for_stack
insert_inline_extent_backref(struct btrfs_trans_handle * trans,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)1169 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1170 				 struct btrfs_path *path,
1171 				 u64 bytenr, u64 num_bytes, u64 parent,
1172 				 u64 root_objectid, u64 owner,
1173 				 u64 offset, int refs_to_add,
1174 				 struct btrfs_delayed_extent_op *extent_op)
1175 {
1176 	struct btrfs_extent_inline_ref *iref;
1177 	int ret;
1178 
1179 	ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1180 					   num_bytes, parent, root_objectid,
1181 					   owner, offset, 1);
1182 	if (ret == 0) {
1183 		BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1184 		update_inline_extent_backref(path, iref, refs_to_add,
1185 					     extent_op, NULL);
1186 	} else if (ret == -ENOENT) {
1187 		setup_inline_extent_backref(trans->fs_info, path, iref, parent,
1188 					    root_objectid, owner, offset,
1189 					    refs_to_add, extent_op);
1190 		ret = 0;
1191 	}
1192 	return ret;
1193 }
1194 
insert_extent_backref(struct btrfs_trans_handle * trans,struct btrfs_path * path,u64 bytenr,u64 parent,u64 root_objectid,u64 owner,u64 offset,int refs_to_add)1195 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1196 				 struct btrfs_path *path,
1197 				 u64 bytenr, u64 parent, u64 root_objectid,
1198 				 u64 owner, u64 offset, int refs_to_add)
1199 {
1200 	int ret;
1201 	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1202 		BUG_ON(refs_to_add != 1);
1203 		ret = insert_tree_block_ref(trans, path, bytenr, parent,
1204 					    root_objectid);
1205 	} else {
1206 		ret = insert_extent_data_ref(trans, path, bytenr, parent,
1207 					     root_objectid, owner, offset,
1208 					     refs_to_add);
1209 	}
1210 	return ret;
1211 }
1212 
remove_extent_backref(struct btrfs_trans_handle * trans,struct btrfs_path * path,struct btrfs_extent_inline_ref * iref,int refs_to_drop,int is_data,int * last_ref)1213 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1214 				 struct btrfs_path *path,
1215 				 struct btrfs_extent_inline_ref *iref,
1216 				 int refs_to_drop, int is_data, int *last_ref)
1217 {
1218 	int ret = 0;
1219 
1220 	BUG_ON(!is_data && refs_to_drop != 1);
1221 	if (iref) {
1222 		update_inline_extent_backref(path, iref, -refs_to_drop, NULL,
1223 					     last_ref);
1224 	} else if (is_data) {
1225 		ret = remove_extent_data_ref(trans, path, refs_to_drop,
1226 					     last_ref);
1227 	} else {
1228 		*last_ref = 1;
1229 		ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
1230 	}
1231 	return ret;
1232 }
1233 
btrfs_issue_discard(struct block_device * bdev,u64 start,u64 len,u64 * discarded_bytes)1234 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1235 			       u64 *discarded_bytes)
1236 {
1237 	int j, ret = 0;
1238 	u64 bytes_left, end;
1239 	u64 aligned_start = ALIGN(start, 1 << 9);
1240 
1241 	if (WARN_ON(start != aligned_start)) {
1242 		len -= aligned_start - start;
1243 		len = round_down(len, 1 << 9);
1244 		start = aligned_start;
1245 	}
1246 
1247 	*discarded_bytes = 0;
1248 
1249 	if (!len)
1250 		return 0;
1251 
1252 	end = start + len;
1253 	bytes_left = len;
1254 
1255 	/* Skip any superblocks on this device. */
1256 	for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1257 		u64 sb_start = btrfs_sb_offset(j);
1258 		u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1259 		u64 size = sb_start - start;
1260 
1261 		if (!in_range(sb_start, start, bytes_left) &&
1262 		    !in_range(sb_end, start, bytes_left) &&
1263 		    !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1264 			continue;
1265 
1266 		/*
1267 		 * Superblock spans beginning of range.  Adjust start and
1268 		 * try again.
1269 		 */
1270 		if (sb_start <= start) {
1271 			start += sb_end - start;
1272 			if (start > end) {
1273 				bytes_left = 0;
1274 				break;
1275 			}
1276 			bytes_left = end - start;
1277 			continue;
1278 		}
1279 
1280 		if (size) {
1281 			ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1282 						   GFP_NOFS, 0);
1283 			if (!ret)
1284 				*discarded_bytes += size;
1285 			else if (ret != -EOPNOTSUPP)
1286 				return ret;
1287 		}
1288 
1289 		start = sb_end;
1290 		if (start > end) {
1291 			bytes_left = 0;
1292 			break;
1293 		}
1294 		bytes_left = end - start;
1295 	}
1296 
1297 	if (bytes_left) {
1298 		ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
1299 					   GFP_NOFS, 0);
1300 		if (!ret)
1301 			*discarded_bytes += bytes_left;
1302 	}
1303 	return ret;
1304 }
1305 
btrfs_discard_extent(struct btrfs_fs_info * fs_info,u64 bytenr,u64 num_bytes,u64 * actual_bytes)1306 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1307 			 u64 num_bytes, u64 *actual_bytes)
1308 {
1309 	int ret;
1310 	u64 discarded_bytes = 0;
1311 	struct btrfs_bio *bbio = NULL;
1312 
1313 
1314 	/*
1315 	 * Avoid races with device replace and make sure our bbio has devices
1316 	 * associated to its stripes that don't go away while we are discarding.
1317 	 */
1318 	btrfs_bio_counter_inc_blocked(fs_info);
1319 	/* Tell the block device(s) that the sectors can be discarded */
1320 	ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
1321 			      &bbio, 0);
1322 	/* Error condition is -ENOMEM */
1323 	if (!ret) {
1324 		struct btrfs_bio_stripe *stripe = bbio->stripes;
1325 		int i;
1326 
1327 
1328 		for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1329 			u64 bytes;
1330 			struct request_queue *req_q;
1331 
1332 			if (!stripe->dev->bdev) {
1333 				ASSERT(btrfs_test_opt(fs_info, DEGRADED));
1334 				continue;
1335 			}
1336 			req_q = bdev_get_queue(stripe->dev->bdev);
1337 			if (!blk_queue_discard(req_q))
1338 				continue;
1339 
1340 			ret = btrfs_issue_discard(stripe->dev->bdev,
1341 						  stripe->physical,
1342 						  stripe->length,
1343 						  &bytes);
1344 			if (!ret)
1345 				discarded_bytes += bytes;
1346 			else if (ret != -EOPNOTSUPP)
1347 				break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1348 
1349 			/*
1350 			 * Just in case we get back EOPNOTSUPP for some reason,
1351 			 * just ignore the return value so we don't screw up
1352 			 * people calling discard_extent.
1353 			 */
1354 			ret = 0;
1355 		}
1356 		btrfs_put_bbio(bbio);
1357 	}
1358 	btrfs_bio_counter_dec(fs_info);
1359 
1360 	if (actual_bytes)
1361 		*actual_bytes = discarded_bytes;
1362 
1363 
1364 	if (ret == -EOPNOTSUPP)
1365 		ret = 0;
1366 	return ret;
1367 }
1368 
1369 /* Can return -ENOMEM */
btrfs_inc_extent_ref(struct btrfs_trans_handle * trans,struct btrfs_ref * generic_ref)1370 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1371 			 struct btrfs_ref *generic_ref)
1372 {
1373 	struct btrfs_fs_info *fs_info = trans->fs_info;
1374 	int old_ref_mod, new_ref_mod;
1375 	int ret;
1376 
1377 	ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
1378 	       generic_ref->action);
1379 	BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
1380 	       generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID);
1381 
1382 	if (generic_ref->type == BTRFS_REF_METADATA)
1383 		ret = btrfs_add_delayed_tree_ref(trans, generic_ref,
1384 				NULL, &old_ref_mod, &new_ref_mod);
1385 	else
1386 		ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0,
1387 						 &old_ref_mod, &new_ref_mod);
1388 
1389 	btrfs_ref_tree_mod(fs_info, generic_ref);
1390 
1391 	if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0)
1392 		sub_pinned_bytes(fs_info, generic_ref);
1393 
1394 	return ret;
1395 }
1396 
1397 /*
1398  * __btrfs_inc_extent_ref - insert backreference for a given extent
1399  *
1400  * @trans:	    Handle of transaction
1401  *
1402  * @node:	    The delayed ref node used to get the bytenr/length for
1403  *		    extent whose references are incremented.
1404  *
1405  * @parent:	    If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
1406  *		    BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
1407  *		    bytenr of the parent block. Since new extents are always
1408  *		    created with indirect references, this will only be the case
1409  *		    when relocating a shared extent. In that case, root_objectid
1410  *		    will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must
1411  *		    be 0
1412  *
1413  * @root_objectid:  The id of the root where this modification has originated,
1414  *		    this can be either one of the well-known metadata trees or
1415  *		    the subvolume id which references this extent.
1416  *
1417  * @owner:	    For data extents it is the inode number of the owning file.
1418  *		    For metadata extents this parameter holds the level in the
1419  *		    tree of the extent.
1420  *
1421  * @offset:	    For metadata extents the offset is ignored and is currently
1422  *		    always passed as 0. For data extents it is the fileoffset
1423  *		    this extent belongs to.
1424  *
1425  * @refs_to_add     Number of references to add
1426  *
1427  * @extent_op       Pointer to a structure, holding information necessary when
1428  *                  updating a tree block's flags
1429  *
1430  */
__btrfs_inc_extent_ref(struct btrfs_trans_handle * trans,struct btrfs_delayed_ref_node * node,u64 parent,u64 root_objectid,u64 owner,u64 offset,int refs_to_add,struct btrfs_delayed_extent_op * extent_op)1431 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1432 				  struct btrfs_delayed_ref_node *node,
1433 				  u64 parent, u64 root_objectid,
1434 				  u64 owner, u64 offset, int refs_to_add,
1435 				  struct btrfs_delayed_extent_op *extent_op)
1436 {
1437 	struct btrfs_path *path;
1438 	struct extent_buffer *leaf;
1439 	struct btrfs_extent_item *item;
1440 	struct btrfs_key key;
1441 	u64 bytenr = node->bytenr;
1442 	u64 num_bytes = node->num_bytes;
1443 	u64 refs;
1444 	int ret;
1445 
1446 	path = btrfs_alloc_path();
1447 	if (!path)
1448 		return -ENOMEM;
1449 
1450 	path->reada = READA_FORWARD;
1451 	path->leave_spinning = 1;
1452 	/* this will setup the path even if it fails to insert the back ref */
1453 	ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
1454 					   parent, root_objectid, owner,
1455 					   offset, refs_to_add, extent_op);
1456 	if ((ret < 0 && ret != -EAGAIN) || !ret)
1457 		goto out;
1458 
1459 	/*
1460 	 * Ok we had -EAGAIN which means we didn't have space to insert and
1461 	 * inline extent ref, so just update the reference count and add a
1462 	 * normal backref.
1463 	 */
1464 	leaf = path->nodes[0];
1465 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1466 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1467 	refs = btrfs_extent_refs(leaf, item);
1468 	btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1469 	if (extent_op)
1470 		__run_delayed_extent_op(extent_op, leaf, item);
1471 
1472 	btrfs_mark_buffer_dirty(leaf);
1473 	btrfs_release_path(path);
1474 
1475 	path->reada = READA_FORWARD;
1476 	path->leave_spinning = 1;
1477 	/* now insert the actual backref */
1478 	ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid,
1479 				    owner, offset, refs_to_add);
1480 	if (ret)
1481 		btrfs_abort_transaction(trans, ret);
1482 out:
1483 	btrfs_free_path(path);
1484 	return ret;
1485 }
1486 
run_delayed_data_ref(struct btrfs_trans_handle * trans,struct btrfs_delayed_ref_node * node,struct btrfs_delayed_extent_op * extent_op,int insert_reserved)1487 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1488 				struct btrfs_delayed_ref_node *node,
1489 				struct btrfs_delayed_extent_op *extent_op,
1490 				int insert_reserved)
1491 {
1492 	int ret = 0;
1493 	struct btrfs_delayed_data_ref *ref;
1494 	struct btrfs_key ins;
1495 	u64 parent = 0;
1496 	u64 ref_root = 0;
1497 	u64 flags = 0;
1498 
1499 	ins.objectid = node->bytenr;
1500 	ins.offset = node->num_bytes;
1501 	ins.type = BTRFS_EXTENT_ITEM_KEY;
1502 
1503 	ref = btrfs_delayed_node_to_data_ref(node);
1504 	trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
1505 
1506 	if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1507 		parent = ref->parent;
1508 	ref_root = ref->root;
1509 
1510 	if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1511 		if (extent_op)
1512 			flags |= extent_op->flags_to_set;
1513 		ret = alloc_reserved_file_extent(trans, parent, ref_root,
1514 						 flags, ref->objectid,
1515 						 ref->offset, &ins,
1516 						 node->ref_mod);
1517 	} else if (node->action == BTRFS_ADD_DELAYED_REF) {
1518 		ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1519 					     ref->objectid, ref->offset,
1520 					     node->ref_mod, extent_op);
1521 	} else if (node->action == BTRFS_DROP_DELAYED_REF) {
1522 		ret = __btrfs_free_extent(trans, node, parent,
1523 					  ref_root, ref->objectid,
1524 					  ref->offset, node->ref_mod,
1525 					  extent_op);
1526 	} else {
1527 		BUG();
1528 	}
1529 	return ret;
1530 }
1531 
__run_delayed_extent_op(struct btrfs_delayed_extent_op * extent_op,struct extent_buffer * leaf,struct btrfs_extent_item * ei)1532 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1533 				    struct extent_buffer *leaf,
1534 				    struct btrfs_extent_item *ei)
1535 {
1536 	u64 flags = btrfs_extent_flags(leaf, ei);
1537 	if (extent_op->update_flags) {
1538 		flags |= extent_op->flags_to_set;
1539 		btrfs_set_extent_flags(leaf, ei, flags);
1540 	}
1541 
1542 	if (extent_op->update_key) {
1543 		struct btrfs_tree_block_info *bi;
1544 		BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1545 		bi = (struct btrfs_tree_block_info *)(ei + 1);
1546 		btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1547 	}
1548 }
1549 
run_delayed_extent_op(struct btrfs_trans_handle * trans,struct btrfs_delayed_ref_head * head,struct btrfs_delayed_extent_op * extent_op)1550 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1551 				 struct btrfs_delayed_ref_head *head,
1552 				 struct btrfs_delayed_extent_op *extent_op)
1553 {
1554 	struct btrfs_fs_info *fs_info = trans->fs_info;
1555 	struct btrfs_key key;
1556 	struct btrfs_path *path;
1557 	struct btrfs_extent_item *ei;
1558 	struct extent_buffer *leaf;
1559 	u32 item_size;
1560 	int ret;
1561 	int err = 0;
1562 	int metadata = !extent_op->is_data;
1563 
1564 	if (trans->aborted)
1565 		return 0;
1566 
1567 	if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1568 		metadata = 0;
1569 
1570 	path = btrfs_alloc_path();
1571 	if (!path)
1572 		return -ENOMEM;
1573 
1574 	key.objectid = head->bytenr;
1575 
1576 	if (metadata) {
1577 		key.type = BTRFS_METADATA_ITEM_KEY;
1578 		key.offset = extent_op->level;
1579 	} else {
1580 		key.type = BTRFS_EXTENT_ITEM_KEY;
1581 		key.offset = head->num_bytes;
1582 	}
1583 
1584 again:
1585 	path->reada = READA_FORWARD;
1586 	path->leave_spinning = 1;
1587 	ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
1588 	if (ret < 0) {
1589 		err = ret;
1590 		goto out;
1591 	}
1592 	if (ret > 0) {
1593 		if (metadata) {
1594 			if (path->slots[0] > 0) {
1595 				path->slots[0]--;
1596 				btrfs_item_key_to_cpu(path->nodes[0], &key,
1597 						      path->slots[0]);
1598 				if (key.objectid == head->bytenr &&
1599 				    key.type == BTRFS_EXTENT_ITEM_KEY &&
1600 				    key.offset == head->num_bytes)
1601 					ret = 0;
1602 			}
1603 			if (ret > 0) {
1604 				btrfs_release_path(path);
1605 				metadata = 0;
1606 
1607 				key.objectid = head->bytenr;
1608 				key.offset = head->num_bytes;
1609 				key.type = BTRFS_EXTENT_ITEM_KEY;
1610 				goto again;
1611 			}
1612 		} else {
1613 			err = -EIO;
1614 			goto out;
1615 		}
1616 	}
1617 
1618 	leaf = path->nodes[0];
1619 	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1620 
1621 	if (unlikely(item_size < sizeof(*ei))) {
1622 		err = -EINVAL;
1623 		btrfs_print_v0_err(fs_info);
1624 		btrfs_abort_transaction(trans, err);
1625 		goto out;
1626 	}
1627 
1628 	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1629 	__run_delayed_extent_op(extent_op, leaf, ei);
1630 
1631 	btrfs_mark_buffer_dirty(leaf);
1632 out:
1633 	btrfs_free_path(path);
1634 	return err;
1635 }
1636 
run_delayed_tree_ref(struct btrfs_trans_handle * trans,struct btrfs_delayed_ref_node * node,struct btrfs_delayed_extent_op * extent_op,int insert_reserved)1637 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1638 				struct btrfs_delayed_ref_node *node,
1639 				struct btrfs_delayed_extent_op *extent_op,
1640 				int insert_reserved)
1641 {
1642 	int ret = 0;
1643 	struct btrfs_delayed_tree_ref *ref;
1644 	u64 parent = 0;
1645 	u64 ref_root = 0;
1646 
1647 	ref = btrfs_delayed_node_to_tree_ref(node);
1648 	trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
1649 
1650 	if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1651 		parent = ref->parent;
1652 	ref_root = ref->root;
1653 
1654 	if (node->ref_mod != 1) {
1655 		btrfs_err(trans->fs_info,
1656 	"btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
1657 			  node->bytenr, node->ref_mod, node->action, ref_root,
1658 			  parent);
1659 		return -EIO;
1660 	}
1661 	if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1662 		BUG_ON(!extent_op || !extent_op->update_flags);
1663 		ret = alloc_reserved_tree_block(trans, node, extent_op);
1664 	} else if (node->action == BTRFS_ADD_DELAYED_REF) {
1665 		ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1666 					     ref->level, 0, 1, extent_op);
1667 	} else if (node->action == BTRFS_DROP_DELAYED_REF) {
1668 		ret = __btrfs_free_extent(trans, node, parent, ref_root,
1669 					  ref->level, 0, 1, extent_op);
1670 	} else {
1671 		BUG();
1672 	}
1673 	return ret;
1674 }
1675 
1676 /* helper function to actually process a single delayed ref entry */
run_one_delayed_ref(struct btrfs_trans_handle * trans,struct btrfs_delayed_ref_node * node,struct btrfs_delayed_extent_op * extent_op,int insert_reserved)1677 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1678 			       struct btrfs_delayed_ref_node *node,
1679 			       struct btrfs_delayed_extent_op *extent_op,
1680 			       int insert_reserved)
1681 {
1682 	int ret = 0;
1683 
1684 	if (trans->aborted) {
1685 		if (insert_reserved)
1686 			btrfs_pin_extent(trans->fs_info, node->bytenr,
1687 					 node->num_bytes, 1);
1688 		return 0;
1689 	}
1690 
1691 	if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1692 	    node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1693 		ret = run_delayed_tree_ref(trans, node, extent_op,
1694 					   insert_reserved);
1695 	else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1696 		 node->type == BTRFS_SHARED_DATA_REF_KEY)
1697 		ret = run_delayed_data_ref(trans, node, extent_op,
1698 					   insert_reserved);
1699 	else
1700 		BUG();
1701 	if (ret && insert_reserved)
1702 		btrfs_pin_extent(trans->fs_info, node->bytenr,
1703 				 node->num_bytes, 1);
1704 	return ret;
1705 }
1706 
1707 static inline struct btrfs_delayed_ref_node *
select_delayed_ref(struct btrfs_delayed_ref_head * head)1708 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1709 {
1710 	struct btrfs_delayed_ref_node *ref;
1711 
1712 	if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
1713 		return NULL;
1714 
1715 	/*
1716 	 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1717 	 * This is to prevent a ref count from going down to zero, which deletes
1718 	 * the extent item from the extent tree, when there still are references
1719 	 * to add, which would fail because they would not find the extent item.
1720 	 */
1721 	if (!list_empty(&head->ref_add_list))
1722 		return list_first_entry(&head->ref_add_list,
1723 				struct btrfs_delayed_ref_node, add_list);
1724 
1725 	ref = rb_entry(rb_first_cached(&head->ref_tree),
1726 		       struct btrfs_delayed_ref_node, ref_node);
1727 	ASSERT(list_empty(&ref->add_list));
1728 	return ref;
1729 }
1730 
unselect_delayed_ref_head(struct btrfs_delayed_ref_root * delayed_refs,struct btrfs_delayed_ref_head * head)1731 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
1732 				      struct btrfs_delayed_ref_head *head)
1733 {
1734 	spin_lock(&delayed_refs->lock);
1735 	head->processing = 0;
1736 	delayed_refs->num_heads_ready++;
1737 	spin_unlock(&delayed_refs->lock);
1738 	btrfs_delayed_ref_unlock(head);
1739 }
1740 
cleanup_extent_op(struct btrfs_delayed_ref_head * head)1741 static struct btrfs_delayed_extent_op *cleanup_extent_op(
1742 				struct btrfs_delayed_ref_head *head)
1743 {
1744 	struct btrfs_delayed_extent_op *extent_op = head->extent_op;
1745 
1746 	if (!extent_op)
1747 		return NULL;
1748 
1749 	if (head->must_insert_reserved) {
1750 		head->extent_op = NULL;
1751 		btrfs_free_delayed_extent_op(extent_op);
1752 		return NULL;
1753 	}
1754 	return extent_op;
1755 }
1756 
run_and_cleanup_extent_op(struct btrfs_trans_handle * trans,struct btrfs_delayed_ref_head * head)1757 static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
1758 				     struct btrfs_delayed_ref_head *head)
1759 {
1760 	struct btrfs_delayed_extent_op *extent_op;
1761 	int ret;
1762 
1763 	extent_op = cleanup_extent_op(head);
1764 	if (!extent_op)
1765 		return 0;
1766 	head->extent_op = NULL;
1767 	spin_unlock(&head->lock);
1768 	ret = run_delayed_extent_op(trans, head, extent_op);
1769 	btrfs_free_delayed_extent_op(extent_op);
1770 	return ret ? ret : 1;
1771 }
1772 
btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info * fs_info,struct btrfs_delayed_ref_root * delayed_refs,struct btrfs_delayed_ref_head * head)1773 void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
1774 				  struct btrfs_delayed_ref_root *delayed_refs,
1775 				  struct btrfs_delayed_ref_head *head)
1776 {
1777 	int nr_items = 1;	/* Dropping this ref head update. */
1778 
1779 	if (head->total_ref_mod < 0) {
1780 		struct btrfs_space_info *space_info;
1781 		u64 flags;
1782 
1783 		if (head->is_data)
1784 			flags = BTRFS_BLOCK_GROUP_DATA;
1785 		else if (head->is_system)
1786 			flags = BTRFS_BLOCK_GROUP_SYSTEM;
1787 		else
1788 			flags = BTRFS_BLOCK_GROUP_METADATA;
1789 		space_info = btrfs_find_space_info(fs_info, flags);
1790 		ASSERT(space_info);
1791 		percpu_counter_add_batch(&space_info->total_bytes_pinned,
1792 				   -head->num_bytes,
1793 				   BTRFS_TOTAL_BYTES_PINNED_BATCH);
1794 
1795 		/*
1796 		 * We had csum deletions accounted for in our delayed refs rsv,
1797 		 * we need to drop the csum leaves for this update from our
1798 		 * delayed_refs_rsv.
1799 		 */
1800 		if (head->is_data) {
1801 			spin_lock(&delayed_refs->lock);
1802 			delayed_refs->pending_csums -= head->num_bytes;
1803 			spin_unlock(&delayed_refs->lock);
1804 			nr_items += btrfs_csum_bytes_to_leaves(fs_info,
1805 				head->num_bytes);
1806 		}
1807 	}
1808 
1809 	btrfs_delayed_refs_rsv_release(fs_info, nr_items);
1810 }
1811 
cleanup_ref_head(struct btrfs_trans_handle * trans,struct btrfs_delayed_ref_head * head)1812 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
1813 			    struct btrfs_delayed_ref_head *head)
1814 {
1815 
1816 	struct btrfs_fs_info *fs_info = trans->fs_info;
1817 	struct btrfs_delayed_ref_root *delayed_refs;
1818 	int ret;
1819 
1820 	delayed_refs = &trans->transaction->delayed_refs;
1821 
1822 	ret = run_and_cleanup_extent_op(trans, head);
1823 	if (ret < 0) {
1824 		unselect_delayed_ref_head(delayed_refs, head);
1825 		btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
1826 		return ret;
1827 	} else if (ret) {
1828 		return ret;
1829 	}
1830 
1831 	/*
1832 	 * Need to drop our head ref lock and re-acquire the delayed ref lock
1833 	 * and then re-check to make sure nobody got added.
1834 	 */
1835 	spin_unlock(&head->lock);
1836 	spin_lock(&delayed_refs->lock);
1837 	spin_lock(&head->lock);
1838 	if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
1839 		spin_unlock(&head->lock);
1840 		spin_unlock(&delayed_refs->lock);
1841 		return 1;
1842 	}
1843 	btrfs_delete_ref_head(delayed_refs, head);
1844 	spin_unlock(&head->lock);
1845 	spin_unlock(&delayed_refs->lock);
1846 
1847 	if (head->must_insert_reserved) {
1848 		btrfs_pin_extent(fs_info, head->bytenr,
1849 				 head->num_bytes, 1);
1850 		if (head->is_data) {
1851 			ret = btrfs_del_csums(trans, fs_info, head->bytenr,
1852 					      head->num_bytes);
1853 		}
1854 	}
1855 
1856 	btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
1857 
1858 	trace_run_delayed_ref_head(fs_info, head, 0);
1859 	btrfs_delayed_ref_unlock(head);
1860 	btrfs_put_delayed_ref_head(head);
1861 	return 0;
1862 }
1863 
btrfs_obtain_ref_head(struct btrfs_trans_handle * trans)1864 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
1865 					struct btrfs_trans_handle *trans)
1866 {
1867 	struct btrfs_delayed_ref_root *delayed_refs =
1868 		&trans->transaction->delayed_refs;
1869 	struct btrfs_delayed_ref_head *head = NULL;
1870 	int ret;
1871 
1872 	spin_lock(&delayed_refs->lock);
1873 	head = btrfs_select_ref_head(delayed_refs);
1874 	if (!head) {
1875 		spin_unlock(&delayed_refs->lock);
1876 		return head;
1877 	}
1878 
1879 	/*
1880 	 * Grab the lock that says we are going to process all the refs for
1881 	 * this head
1882 	 */
1883 	ret = btrfs_delayed_ref_lock(delayed_refs, head);
1884 	spin_unlock(&delayed_refs->lock);
1885 
1886 	/*
1887 	 * We may have dropped the spin lock to get the head mutex lock, and
1888 	 * that might have given someone else time to free the head.  If that's
1889 	 * true, it has been removed from our list and we can move on.
1890 	 */
1891 	if (ret == -EAGAIN)
1892 		head = ERR_PTR(-EAGAIN);
1893 
1894 	return head;
1895 }
1896 
btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle * trans,struct btrfs_delayed_ref_head * locked_ref,unsigned long * run_refs)1897 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
1898 				    struct btrfs_delayed_ref_head *locked_ref,
1899 				    unsigned long *run_refs)
1900 {
1901 	struct btrfs_fs_info *fs_info = trans->fs_info;
1902 	struct btrfs_delayed_ref_root *delayed_refs;
1903 	struct btrfs_delayed_extent_op *extent_op;
1904 	struct btrfs_delayed_ref_node *ref;
1905 	int must_insert_reserved = 0;
1906 	int ret;
1907 
1908 	delayed_refs = &trans->transaction->delayed_refs;
1909 
1910 	lockdep_assert_held(&locked_ref->mutex);
1911 	lockdep_assert_held(&locked_ref->lock);
1912 
1913 	while ((ref = select_delayed_ref(locked_ref))) {
1914 		if (ref->seq &&
1915 		    btrfs_check_delayed_seq(fs_info, ref->seq)) {
1916 			spin_unlock(&locked_ref->lock);
1917 			unselect_delayed_ref_head(delayed_refs, locked_ref);
1918 			return -EAGAIN;
1919 		}
1920 
1921 		(*run_refs)++;
1922 		ref->in_tree = 0;
1923 		rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
1924 		RB_CLEAR_NODE(&ref->ref_node);
1925 		if (!list_empty(&ref->add_list))
1926 			list_del(&ref->add_list);
1927 		/*
1928 		 * When we play the delayed ref, also correct the ref_mod on
1929 		 * head
1930 		 */
1931 		switch (ref->action) {
1932 		case BTRFS_ADD_DELAYED_REF:
1933 		case BTRFS_ADD_DELAYED_EXTENT:
1934 			locked_ref->ref_mod -= ref->ref_mod;
1935 			break;
1936 		case BTRFS_DROP_DELAYED_REF:
1937 			locked_ref->ref_mod += ref->ref_mod;
1938 			break;
1939 		default:
1940 			WARN_ON(1);
1941 		}
1942 		atomic_dec(&delayed_refs->num_entries);
1943 
1944 		/*
1945 		 * Record the must_insert_reserved flag before we drop the
1946 		 * spin lock.
1947 		 */
1948 		must_insert_reserved = locked_ref->must_insert_reserved;
1949 		locked_ref->must_insert_reserved = 0;
1950 
1951 		extent_op = locked_ref->extent_op;
1952 		locked_ref->extent_op = NULL;
1953 		spin_unlock(&locked_ref->lock);
1954 
1955 		ret = run_one_delayed_ref(trans, ref, extent_op,
1956 					  must_insert_reserved);
1957 
1958 		btrfs_free_delayed_extent_op(extent_op);
1959 		if (ret) {
1960 			unselect_delayed_ref_head(delayed_refs, locked_ref);
1961 			btrfs_put_delayed_ref(ref);
1962 			btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
1963 				    ret);
1964 			return ret;
1965 		}
1966 
1967 		btrfs_put_delayed_ref(ref);
1968 		cond_resched();
1969 
1970 		spin_lock(&locked_ref->lock);
1971 		btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
1972 	}
1973 
1974 	return 0;
1975 }
1976 
1977 /*
1978  * Returns 0 on success or if called with an already aborted transaction.
1979  * Returns -ENOMEM or -EIO on failure and will abort the transaction.
1980  */
__btrfs_run_delayed_refs(struct btrfs_trans_handle * trans,unsigned long nr)1981 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1982 					     unsigned long nr)
1983 {
1984 	struct btrfs_fs_info *fs_info = trans->fs_info;
1985 	struct btrfs_delayed_ref_root *delayed_refs;
1986 	struct btrfs_delayed_ref_head *locked_ref = NULL;
1987 	ktime_t start = ktime_get();
1988 	int ret;
1989 	unsigned long count = 0;
1990 	unsigned long actual_count = 0;
1991 
1992 	delayed_refs = &trans->transaction->delayed_refs;
1993 	do {
1994 		if (!locked_ref) {
1995 			locked_ref = btrfs_obtain_ref_head(trans);
1996 			if (IS_ERR_OR_NULL(locked_ref)) {
1997 				if (PTR_ERR(locked_ref) == -EAGAIN) {
1998 					continue;
1999 				} else {
2000 					break;
2001 				}
2002 			}
2003 			count++;
2004 		}
2005 		/*
2006 		 * We need to try and merge add/drops of the same ref since we
2007 		 * can run into issues with relocate dropping the implicit ref
2008 		 * and then it being added back again before the drop can
2009 		 * finish.  If we merged anything we need to re-loop so we can
2010 		 * get a good ref.
2011 		 * Or we can get node references of the same type that weren't
2012 		 * merged when created due to bumps in the tree mod seq, and
2013 		 * we need to merge them to prevent adding an inline extent
2014 		 * backref before dropping it (triggering a BUG_ON at
2015 		 * insert_inline_extent_backref()).
2016 		 */
2017 		spin_lock(&locked_ref->lock);
2018 		btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2019 
2020 		ret = btrfs_run_delayed_refs_for_head(trans, locked_ref,
2021 						      &actual_count);
2022 		if (ret < 0 && ret != -EAGAIN) {
2023 			/*
2024 			 * Error, btrfs_run_delayed_refs_for_head already
2025 			 * unlocked everything so just bail out
2026 			 */
2027 			return ret;
2028 		} else if (!ret) {
2029 			/*
2030 			 * Success, perform the usual cleanup of a processed
2031 			 * head
2032 			 */
2033 			ret = cleanup_ref_head(trans, locked_ref);
2034 			if (ret > 0 ) {
2035 				/* We dropped our lock, we need to loop. */
2036 				ret = 0;
2037 				continue;
2038 			} else if (ret) {
2039 				return ret;
2040 			}
2041 		}
2042 
2043 		/*
2044 		 * Either success case or btrfs_run_delayed_refs_for_head
2045 		 * returned -EAGAIN, meaning we need to select another head
2046 		 */
2047 
2048 		locked_ref = NULL;
2049 		cond_resched();
2050 	} while ((nr != -1 && count < nr) || locked_ref);
2051 
2052 	/*
2053 	 * We don't want to include ref heads since we can have empty ref heads
2054 	 * and those will drastically skew our runtime down since we just do
2055 	 * accounting, no actual extent tree updates.
2056 	 */
2057 	if (actual_count > 0) {
2058 		u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2059 		u64 avg;
2060 
2061 		/*
2062 		 * We weigh the current average higher than our current runtime
2063 		 * to avoid large swings in the average.
2064 		 */
2065 		spin_lock(&delayed_refs->lock);
2066 		avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2067 		fs_info->avg_delayed_ref_runtime = avg >> 2;	/* div by 4 */
2068 		spin_unlock(&delayed_refs->lock);
2069 	}
2070 	return 0;
2071 }
2072 
2073 #ifdef SCRAMBLE_DELAYED_REFS
2074 /*
2075  * Normally delayed refs get processed in ascending bytenr order. This
2076  * correlates in most cases to the order added. To expose dependencies on this
2077  * order, we start to process the tree in the middle instead of the beginning
2078  */
find_middle(struct rb_root * root)2079 static u64 find_middle(struct rb_root *root)
2080 {
2081 	struct rb_node *n = root->rb_node;
2082 	struct btrfs_delayed_ref_node *entry;
2083 	int alt = 1;
2084 	u64 middle;
2085 	u64 first = 0, last = 0;
2086 
2087 	n = rb_first(root);
2088 	if (n) {
2089 		entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2090 		first = entry->bytenr;
2091 	}
2092 	n = rb_last(root);
2093 	if (n) {
2094 		entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2095 		last = entry->bytenr;
2096 	}
2097 	n = root->rb_node;
2098 
2099 	while (n) {
2100 		entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2101 		WARN_ON(!entry->in_tree);
2102 
2103 		middle = entry->bytenr;
2104 
2105 		if (alt)
2106 			n = n->rb_left;
2107 		else
2108 			n = n->rb_right;
2109 
2110 		alt = 1 - alt;
2111 	}
2112 	return middle;
2113 }
2114 #endif
2115 
heads_to_leaves(struct btrfs_fs_info * fs_info,u64 heads)2116 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2117 {
2118 	u64 num_bytes;
2119 
2120 	num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2121 			     sizeof(struct btrfs_extent_inline_ref));
2122 	if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2123 		num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2124 
2125 	/*
2126 	 * We don't ever fill up leaves all the way so multiply by 2 just to be
2127 	 * closer to what we're really going to want to use.
2128 	 */
2129 	return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2130 }
2131 
2132 /*
2133  * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2134  * would require to store the csums for that many bytes.
2135  */
btrfs_csum_bytes_to_leaves(struct btrfs_fs_info * fs_info,u64 csum_bytes)2136 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2137 {
2138 	u64 csum_size;
2139 	u64 num_csums_per_leaf;
2140 	u64 num_csums;
2141 
2142 	csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2143 	num_csums_per_leaf = div64_u64(csum_size,
2144 			(u64)btrfs_super_csum_size(fs_info->super_copy));
2145 	num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2146 	num_csums += num_csums_per_leaf - 1;
2147 	num_csums = div64_u64(num_csums, num_csums_per_leaf);
2148 	return num_csums;
2149 }
2150 
2151 /*
2152  * this starts processing the delayed reference count updates and
2153  * extent insertions we have queued up so far.  count can be
2154  * 0, which means to process everything in the tree at the start
2155  * of the run (but not newly added entries), or it can be some target
2156  * number you'd like to process.
2157  *
2158  * Returns 0 on success or if called with an aborted transaction
2159  * Returns <0 on error and aborts the transaction
2160  */
btrfs_run_delayed_refs(struct btrfs_trans_handle * trans,unsigned long count)2161 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2162 			   unsigned long count)
2163 {
2164 	struct btrfs_fs_info *fs_info = trans->fs_info;
2165 	struct rb_node *node;
2166 	struct btrfs_delayed_ref_root *delayed_refs;
2167 	struct btrfs_delayed_ref_head *head;
2168 	int ret;
2169 	int run_all = count == (unsigned long)-1;
2170 
2171 	/* We'll clean this up in btrfs_cleanup_transaction */
2172 	if (trans->aborted)
2173 		return 0;
2174 
2175 	if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2176 		return 0;
2177 
2178 	delayed_refs = &trans->transaction->delayed_refs;
2179 	if (count == 0)
2180 		count = atomic_read(&delayed_refs->num_entries) * 2;
2181 
2182 again:
2183 #ifdef SCRAMBLE_DELAYED_REFS
2184 	delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2185 #endif
2186 	ret = __btrfs_run_delayed_refs(trans, count);
2187 	if (ret < 0) {
2188 		btrfs_abort_transaction(trans, ret);
2189 		return ret;
2190 	}
2191 
2192 	if (run_all) {
2193 		btrfs_create_pending_block_groups(trans);
2194 
2195 		spin_lock(&delayed_refs->lock);
2196 		node = rb_first_cached(&delayed_refs->href_root);
2197 		if (!node) {
2198 			spin_unlock(&delayed_refs->lock);
2199 			goto out;
2200 		}
2201 		head = rb_entry(node, struct btrfs_delayed_ref_head,
2202 				href_node);
2203 		refcount_inc(&head->refs);
2204 		spin_unlock(&delayed_refs->lock);
2205 
2206 		/* Mutex was contended, block until it's released and retry. */
2207 		mutex_lock(&head->mutex);
2208 		mutex_unlock(&head->mutex);
2209 
2210 		btrfs_put_delayed_ref_head(head);
2211 		cond_resched();
2212 		goto again;
2213 	}
2214 out:
2215 	return 0;
2216 }
2217 
btrfs_set_disk_extent_flags(struct btrfs_trans_handle * trans,u64 bytenr,u64 num_bytes,u64 flags,int level,int is_data)2218 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2219 				u64 bytenr, u64 num_bytes, u64 flags,
2220 				int level, int is_data)
2221 {
2222 	struct btrfs_delayed_extent_op *extent_op;
2223 	int ret;
2224 
2225 	extent_op = btrfs_alloc_delayed_extent_op();
2226 	if (!extent_op)
2227 		return -ENOMEM;
2228 
2229 	extent_op->flags_to_set = flags;
2230 	extent_op->update_flags = true;
2231 	extent_op->update_key = false;
2232 	extent_op->is_data = is_data ? true : false;
2233 	extent_op->level = level;
2234 
2235 	ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2236 	if (ret)
2237 		btrfs_free_delayed_extent_op(extent_op);
2238 	return ret;
2239 }
2240 
check_delayed_ref(struct btrfs_root * root,struct btrfs_path * path,u64 objectid,u64 offset,u64 bytenr)2241 static noinline int check_delayed_ref(struct btrfs_root *root,
2242 				      struct btrfs_path *path,
2243 				      u64 objectid, u64 offset, u64 bytenr)
2244 {
2245 	struct btrfs_delayed_ref_head *head;
2246 	struct btrfs_delayed_ref_node *ref;
2247 	struct btrfs_delayed_data_ref *data_ref;
2248 	struct btrfs_delayed_ref_root *delayed_refs;
2249 	struct btrfs_transaction *cur_trans;
2250 	struct rb_node *node;
2251 	int ret = 0;
2252 
2253 	spin_lock(&root->fs_info->trans_lock);
2254 	cur_trans = root->fs_info->running_transaction;
2255 	if (cur_trans)
2256 		refcount_inc(&cur_trans->use_count);
2257 	spin_unlock(&root->fs_info->trans_lock);
2258 	if (!cur_trans)
2259 		return 0;
2260 
2261 	delayed_refs = &cur_trans->delayed_refs;
2262 	spin_lock(&delayed_refs->lock);
2263 	head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
2264 	if (!head) {
2265 		spin_unlock(&delayed_refs->lock);
2266 		btrfs_put_transaction(cur_trans);
2267 		return 0;
2268 	}
2269 
2270 	if (!mutex_trylock(&head->mutex)) {
2271 		refcount_inc(&head->refs);
2272 		spin_unlock(&delayed_refs->lock);
2273 
2274 		btrfs_release_path(path);
2275 
2276 		/*
2277 		 * Mutex was contended, block until it's released and let
2278 		 * caller try again
2279 		 */
2280 		mutex_lock(&head->mutex);
2281 		mutex_unlock(&head->mutex);
2282 		btrfs_put_delayed_ref_head(head);
2283 		btrfs_put_transaction(cur_trans);
2284 		return -EAGAIN;
2285 	}
2286 	spin_unlock(&delayed_refs->lock);
2287 
2288 	spin_lock(&head->lock);
2289 	/*
2290 	 * XXX: We should replace this with a proper search function in the
2291 	 * future.
2292 	 */
2293 	for (node = rb_first_cached(&head->ref_tree); node;
2294 	     node = rb_next(node)) {
2295 		ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
2296 		/* If it's a shared ref we know a cross reference exists */
2297 		if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2298 			ret = 1;
2299 			break;
2300 		}
2301 
2302 		data_ref = btrfs_delayed_node_to_data_ref(ref);
2303 
2304 		/*
2305 		 * If our ref doesn't match the one we're currently looking at
2306 		 * then we have a cross reference.
2307 		 */
2308 		if (data_ref->root != root->root_key.objectid ||
2309 		    data_ref->objectid != objectid ||
2310 		    data_ref->offset != offset) {
2311 			ret = 1;
2312 			break;
2313 		}
2314 	}
2315 	spin_unlock(&head->lock);
2316 	mutex_unlock(&head->mutex);
2317 	btrfs_put_transaction(cur_trans);
2318 	return ret;
2319 }
2320 
check_committed_ref(struct btrfs_root * root,struct btrfs_path * path,u64 objectid,u64 offset,u64 bytenr)2321 static noinline int check_committed_ref(struct btrfs_root *root,
2322 					struct btrfs_path *path,
2323 					u64 objectid, u64 offset, u64 bytenr)
2324 {
2325 	struct btrfs_fs_info *fs_info = root->fs_info;
2326 	struct btrfs_root *extent_root = fs_info->extent_root;
2327 	struct extent_buffer *leaf;
2328 	struct btrfs_extent_data_ref *ref;
2329 	struct btrfs_extent_inline_ref *iref;
2330 	struct btrfs_extent_item *ei;
2331 	struct btrfs_key key;
2332 	u32 item_size;
2333 	int type;
2334 	int ret;
2335 
2336 	key.objectid = bytenr;
2337 	key.offset = (u64)-1;
2338 	key.type = BTRFS_EXTENT_ITEM_KEY;
2339 
2340 	ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2341 	if (ret < 0)
2342 		goto out;
2343 	BUG_ON(ret == 0); /* Corruption */
2344 
2345 	ret = -ENOENT;
2346 	if (path->slots[0] == 0)
2347 		goto out;
2348 
2349 	path->slots[0]--;
2350 	leaf = path->nodes[0];
2351 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2352 
2353 	if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2354 		goto out;
2355 
2356 	ret = 1;
2357 	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2358 	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2359 
2360 	/* If extent item has more than 1 inline ref then it's shared */
2361 	if (item_size != sizeof(*ei) +
2362 	    btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2363 		goto out;
2364 
2365 	/* If extent created before last snapshot => it's definitely shared */
2366 	if (btrfs_extent_generation(leaf, ei) <=
2367 	    btrfs_root_last_snapshot(&root->root_item))
2368 		goto out;
2369 
2370 	iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2371 
2372 	/* If this extent has SHARED_DATA_REF then it's shared */
2373 	type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2374 	if (type != BTRFS_EXTENT_DATA_REF_KEY)
2375 		goto out;
2376 
2377 	ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2378 	if (btrfs_extent_refs(leaf, ei) !=
2379 	    btrfs_extent_data_ref_count(leaf, ref) ||
2380 	    btrfs_extent_data_ref_root(leaf, ref) !=
2381 	    root->root_key.objectid ||
2382 	    btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2383 	    btrfs_extent_data_ref_offset(leaf, ref) != offset)
2384 		goto out;
2385 
2386 	ret = 0;
2387 out:
2388 	return ret;
2389 }
2390 
btrfs_cross_ref_exist(struct btrfs_root * root,u64 objectid,u64 offset,u64 bytenr)2391 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
2392 			  u64 bytenr)
2393 {
2394 	struct btrfs_path *path;
2395 	int ret;
2396 
2397 	path = btrfs_alloc_path();
2398 	if (!path)
2399 		return -ENOMEM;
2400 
2401 	do {
2402 		ret = check_committed_ref(root, path, objectid,
2403 					  offset, bytenr);
2404 		if (ret && ret != -ENOENT)
2405 			goto out;
2406 
2407 		ret = check_delayed_ref(root, path, objectid, offset, bytenr);
2408 	} while (ret == -EAGAIN);
2409 
2410 out:
2411 	btrfs_free_path(path);
2412 	if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2413 		WARN_ON(ret > 0);
2414 	return ret;
2415 }
2416 
__btrfs_mod_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * buf,int full_backref,int inc)2417 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2418 			   struct btrfs_root *root,
2419 			   struct extent_buffer *buf,
2420 			   int full_backref, int inc)
2421 {
2422 	struct btrfs_fs_info *fs_info = root->fs_info;
2423 	u64 bytenr;
2424 	u64 num_bytes;
2425 	u64 parent;
2426 	u64 ref_root;
2427 	u32 nritems;
2428 	struct btrfs_key key;
2429 	struct btrfs_file_extent_item *fi;
2430 	struct btrfs_ref generic_ref = { 0 };
2431 	bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC);
2432 	int i;
2433 	int action;
2434 	int level;
2435 	int ret = 0;
2436 
2437 	if (btrfs_is_testing(fs_info))
2438 		return 0;
2439 
2440 	ref_root = btrfs_header_owner(buf);
2441 	nritems = btrfs_header_nritems(buf);
2442 	level = btrfs_header_level(buf);
2443 
2444 	if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
2445 		return 0;
2446 
2447 	if (full_backref)
2448 		parent = buf->start;
2449 	else
2450 		parent = 0;
2451 	if (inc)
2452 		action = BTRFS_ADD_DELAYED_REF;
2453 	else
2454 		action = BTRFS_DROP_DELAYED_REF;
2455 
2456 	for (i = 0; i < nritems; i++) {
2457 		if (level == 0) {
2458 			btrfs_item_key_to_cpu(buf, &key, i);
2459 			if (key.type != BTRFS_EXTENT_DATA_KEY)
2460 				continue;
2461 			fi = btrfs_item_ptr(buf, i,
2462 					    struct btrfs_file_extent_item);
2463 			if (btrfs_file_extent_type(buf, fi) ==
2464 			    BTRFS_FILE_EXTENT_INLINE)
2465 				continue;
2466 			bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2467 			if (bytenr == 0)
2468 				continue;
2469 
2470 			num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2471 			key.offset -= btrfs_file_extent_offset(buf, fi);
2472 			btrfs_init_generic_ref(&generic_ref, action, bytenr,
2473 					       num_bytes, parent);
2474 			generic_ref.real_root = root->root_key.objectid;
2475 			btrfs_init_data_ref(&generic_ref, ref_root, key.objectid,
2476 					    key.offset);
2477 			generic_ref.skip_qgroup = for_reloc;
2478 			if (inc)
2479 				ret = btrfs_inc_extent_ref(trans, &generic_ref);
2480 			else
2481 				ret = btrfs_free_extent(trans, &generic_ref);
2482 			if (ret)
2483 				goto fail;
2484 		} else {
2485 			bytenr = btrfs_node_blockptr(buf, i);
2486 			num_bytes = fs_info->nodesize;
2487 			btrfs_init_generic_ref(&generic_ref, action, bytenr,
2488 					       num_bytes, parent);
2489 			generic_ref.real_root = root->root_key.objectid;
2490 			btrfs_init_tree_ref(&generic_ref, level - 1, ref_root);
2491 			generic_ref.skip_qgroup = for_reloc;
2492 			if (inc)
2493 				ret = btrfs_inc_extent_ref(trans, &generic_ref);
2494 			else
2495 				ret = btrfs_free_extent(trans, &generic_ref);
2496 			if (ret)
2497 				goto fail;
2498 		}
2499 	}
2500 	return 0;
2501 fail:
2502 	return ret;
2503 }
2504 
btrfs_inc_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * buf,int full_backref)2505 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2506 		  struct extent_buffer *buf, int full_backref)
2507 {
2508 	return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2509 }
2510 
btrfs_dec_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * buf,int full_backref)2511 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2512 		  struct extent_buffer *buf, int full_backref)
2513 {
2514 	return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2515 }
2516 
btrfs_extent_readonly(struct btrfs_fs_info * fs_info,u64 bytenr)2517 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
2518 {
2519 	struct btrfs_block_group_cache *block_group;
2520 	int readonly = 0;
2521 
2522 	block_group = btrfs_lookup_block_group(fs_info, bytenr);
2523 	if (!block_group || block_group->ro)
2524 		readonly = 1;
2525 	if (block_group)
2526 		btrfs_put_block_group(block_group);
2527 	return readonly;
2528 }
2529 
get_alloc_profile_by_root(struct btrfs_root * root,int data)2530 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
2531 {
2532 	struct btrfs_fs_info *fs_info = root->fs_info;
2533 	u64 flags;
2534 	u64 ret;
2535 
2536 	if (data)
2537 		flags = BTRFS_BLOCK_GROUP_DATA;
2538 	else if (root == fs_info->chunk_root)
2539 		flags = BTRFS_BLOCK_GROUP_SYSTEM;
2540 	else
2541 		flags = BTRFS_BLOCK_GROUP_METADATA;
2542 
2543 	ret = btrfs_get_alloc_profile(fs_info, flags);
2544 	return ret;
2545 }
2546 
first_logical_byte(struct btrfs_fs_info * fs_info,u64 search_start)2547 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
2548 {
2549 	struct btrfs_block_group_cache *cache;
2550 	u64 bytenr;
2551 
2552 	spin_lock(&fs_info->block_group_cache_lock);
2553 	bytenr = fs_info->first_logical_byte;
2554 	spin_unlock(&fs_info->block_group_cache_lock);
2555 
2556 	if (bytenr < (u64)-1)
2557 		return bytenr;
2558 
2559 	cache = btrfs_lookup_first_block_group(fs_info, search_start);
2560 	if (!cache)
2561 		return 0;
2562 
2563 	bytenr = cache->key.objectid;
2564 	btrfs_put_block_group(cache);
2565 
2566 	return bytenr;
2567 }
2568 
pin_down_extent(struct btrfs_block_group_cache * cache,u64 bytenr,u64 num_bytes,int reserved)2569 static int pin_down_extent(struct btrfs_block_group_cache *cache,
2570 			   u64 bytenr, u64 num_bytes, int reserved)
2571 {
2572 	struct btrfs_fs_info *fs_info = cache->fs_info;
2573 
2574 	spin_lock(&cache->space_info->lock);
2575 	spin_lock(&cache->lock);
2576 	cache->pinned += num_bytes;
2577 	btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info,
2578 					     num_bytes);
2579 	if (reserved) {
2580 		cache->reserved -= num_bytes;
2581 		cache->space_info->bytes_reserved -= num_bytes;
2582 	}
2583 	spin_unlock(&cache->lock);
2584 	spin_unlock(&cache->space_info->lock);
2585 
2586 	percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
2587 		    num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
2588 	set_extent_dirty(fs_info->pinned_extents, bytenr,
2589 			 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
2590 	return 0;
2591 }
2592 
2593 /*
2594  * this function must be called within transaction
2595  */
btrfs_pin_extent(struct btrfs_fs_info * fs_info,u64 bytenr,u64 num_bytes,int reserved)2596 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
2597 		     u64 bytenr, u64 num_bytes, int reserved)
2598 {
2599 	struct btrfs_block_group_cache *cache;
2600 
2601 	cache = btrfs_lookup_block_group(fs_info, bytenr);
2602 	BUG_ON(!cache); /* Logic error */
2603 
2604 	pin_down_extent(cache, bytenr, num_bytes, reserved);
2605 
2606 	btrfs_put_block_group(cache);
2607 	return 0;
2608 }
2609 
2610 /*
2611  * this function must be called within transaction
2612  */
btrfs_pin_extent_for_log_replay(struct btrfs_fs_info * fs_info,u64 bytenr,u64 num_bytes)2613 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
2614 				    u64 bytenr, u64 num_bytes)
2615 {
2616 	struct btrfs_block_group_cache *cache;
2617 	int ret;
2618 
2619 	cache = btrfs_lookup_block_group(fs_info, bytenr);
2620 	if (!cache)
2621 		return -EINVAL;
2622 
2623 	/*
2624 	 * pull in the free space cache (if any) so that our pin
2625 	 * removes the free space from the cache.  We have load_only set
2626 	 * to one because the slow code to read in the free extents does check
2627 	 * the pinned extents.
2628 	 */
2629 	btrfs_cache_block_group(cache, 1);
2630 
2631 	pin_down_extent(cache, bytenr, num_bytes, 0);
2632 
2633 	/* remove us from the free space cache (if we're there at all) */
2634 	ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
2635 	btrfs_put_block_group(cache);
2636 	return ret;
2637 }
2638 
__exclude_logged_extent(struct btrfs_fs_info * fs_info,u64 start,u64 num_bytes)2639 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
2640 				   u64 start, u64 num_bytes)
2641 {
2642 	int ret;
2643 	struct btrfs_block_group_cache *block_group;
2644 	struct btrfs_caching_control *caching_ctl;
2645 
2646 	block_group = btrfs_lookup_block_group(fs_info, start);
2647 	if (!block_group)
2648 		return -EINVAL;
2649 
2650 	btrfs_cache_block_group(block_group, 0);
2651 	caching_ctl = btrfs_get_caching_control(block_group);
2652 
2653 	if (!caching_ctl) {
2654 		/* Logic error */
2655 		BUG_ON(!btrfs_block_group_cache_done(block_group));
2656 		ret = btrfs_remove_free_space(block_group, start, num_bytes);
2657 	} else {
2658 		mutex_lock(&caching_ctl->mutex);
2659 
2660 		if (start >= caching_ctl->progress) {
2661 			ret = btrfs_add_excluded_extent(fs_info, start,
2662 							num_bytes);
2663 		} else if (start + num_bytes <= caching_ctl->progress) {
2664 			ret = btrfs_remove_free_space(block_group,
2665 						      start, num_bytes);
2666 		} else {
2667 			num_bytes = caching_ctl->progress - start;
2668 			ret = btrfs_remove_free_space(block_group,
2669 						      start, num_bytes);
2670 			if (ret)
2671 				goto out_lock;
2672 
2673 			num_bytes = (start + num_bytes) -
2674 				caching_ctl->progress;
2675 			start = caching_ctl->progress;
2676 			ret = btrfs_add_excluded_extent(fs_info, start,
2677 							num_bytes);
2678 		}
2679 out_lock:
2680 		mutex_unlock(&caching_ctl->mutex);
2681 		btrfs_put_caching_control(caching_ctl);
2682 	}
2683 	btrfs_put_block_group(block_group);
2684 	return ret;
2685 }
2686 
btrfs_exclude_logged_extents(struct extent_buffer * eb)2687 int btrfs_exclude_logged_extents(struct extent_buffer *eb)
2688 {
2689 	struct btrfs_fs_info *fs_info = eb->fs_info;
2690 	struct btrfs_file_extent_item *item;
2691 	struct btrfs_key key;
2692 	int found_type;
2693 	int i;
2694 	int ret = 0;
2695 
2696 	if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
2697 		return 0;
2698 
2699 	for (i = 0; i < btrfs_header_nritems(eb); i++) {
2700 		btrfs_item_key_to_cpu(eb, &key, i);
2701 		if (key.type != BTRFS_EXTENT_DATA_KEY)
2702 			continue;
2703 		item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
2704 		found_type = btrfs_file_extent_type(eb, item);
2705 		if (found_type == BTRFS_FILE_EXTENT_INLINE)
2706 			continue;
2707 		if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
2708 			continue;
2709 		key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
2710 		key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
2711 		ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
2712 		if (ret)
2713 			break;
2714 	}
2715 
2716 	return ret;
2717 }
2718 
2719 static void
btrfs_inc_block_group_reservations(struct btrfs_block_group_cache * bg)2720 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
2721 {
2722 	atomic_inc(&bg->reservations);
2723 }
2724 
btrfs_prepare_extent_commit(struct btrfs_fs_info * fs_info)2725 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
2726 {
2727 	struct btrfs_caching_control *next;
2728 	struct btrfs_caching_control *caching_ctl;
2729 	struct btrfs_block_group_cache *cache;
2730 
2731 	down_write(&fs_info->commit_root_sem);
2732 
2733 	list_for_each_entry_safe(caching_ctl, next,
2734 				 &fs_info->caching_block_groups, list) {
2735 		cache = caching_ctl->block_group;
2736 		if (btrfs_block_group_cache_done(cache)) {
2737 			cache->last_byte_to_unpin = (u64)-1;
2738 			list_del_init(&caching_ctl->list);
2739 			btrfs_put_caching_control(caching_ctl);
2740 		} else {
2741 			cache->last_byte_to_unpin = caching_ctl->progress;
2742 		}
2743 	}
2744 
2745 	if (fs_info->pinned_extents == &fs_info->freed_extents[0])
2746 		fs_info->pinned_extents = &fs_info->freed_extents[1];
2747 	else
2748 		fs_info->pinned_extents = &fs_info->freed_extents[0];
2749 
2750 	up_write(&fs_info->commit_root_sem);
2751 
2752 	btrfs_update_global_block_rsv(fs_info);
2753 }
2754 
2755 /*
2756  * Returns the free cluster for the given space info and sets empty_cluster to
2757  * what it should be based on the mount options.
2758  */
2759 static struct btrfs_free_cluster *
fetch_cluster_info(struct btrfs_fs_info * fs_info,struct btrfs_space_info * space_info,u64 * empty_cluster)2760 fetch_cluster_info(struct btrfs_fs_info *fs_info,
2761 		   struct btrfs_space_info *space_info, u64 *empty_cluster)
2762 {
2763 	struct btrfs_free_cluster *ret = NULL;
2764 
2765 	*empty_cluster = 0;
2766 	if (btrfs_mixed_space_info(space_info))
2767 		return ret;
2768 
2769 	if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
2770 		ret = &fs_info->meta_alloc_cluster;
2771 		if (btrfs_test_opt(fs_info, SSD))
2772 			*empty_cluster = SZ_2M;
2773 		else
2774 			*empty_cluster = SZ_64K;
2775 	} else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
2776 		   btrfs_test_opt(fs_info, SSD_SPREAD)) {
2777 		*empty_cluster = SZ_2M;
2778 		ret = &fs_info->data_alloc_cluster;
2779 	}
2780 
2781 	return ret;
2782 }
2783 
unpin_extent_range(struct btrfs_fs_info * fs_info,u64 start,u64 end,const bool return_free_space)2784 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
2785 			      u64 start, u64 end,
2786 			      const bool return_free_space)
2787 {
2788 	struct btrfs_block_group_cache *cache = NULL;
2789 	struct btrfs_space_info *space_info;
2790 	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2791 	struct btrfs_free_cluster *cluster = NULL;
2792 	u64 len;
2793 	u64 total_unpinned = 0;
2794 	u64 empty_cluster = 0;
2795 	bool readonly;
2796 
2797 	while (start <= end) {
2798 		readonly = false;
2799 		if (!cache ||
2800 		    start >= cache->key.objectid + cache->key.offset) {
2801 			if (cache)
2802 				btrfs_put_block_group(cache);
2803 			total_unpinned = 0;
2804 			cache = btrfs_lookup_block_group(fs_info, start);
2805 			BUG_ON(!cache); /* Logic error */
2806 
2807 			cluster = fetch_cluster_info(fs_info,
2808 						     cache->space_info,
2809 						     &empty_cluster);
2810 			empty_cluster <<= 1;
2811 		}
2812 
2813 		len = cache->key.objectid + cache->key.offset - start;
2814 		len = min(len, end + 1 - start);
2815 
2816 		if (start < cache->last_byte_to_unpin) {
2817 			len = min(len, cache->last_byte_to_unpin - start);
2818 			if (return_free_space)
2819 				btrfs_add_free_space(cache, start, len);
2820 		}
2821 
2822 		start += len;
2823 		total_unpinned += len;
2824 		space_info = cache->space_info;
2825 
2826 		/*
2827 		 * If this space cluster has been marked as fragmented and we've
2828 		 * unpinned enough in this block group to potentially allow a
2829 		 * cluster to be created inside of it go ahead and clear the
2830 		 * fragmented check.
2831 		 */
2832 		if (cluster && cluster->fragmented &&
2833 		    total_unpinned > empty_cluster) {
2834 			spin_lock(&cluster->lock);
2835 			cluster->fragmented = 0;
2836 			spin_unlock(&cluster->lock);
2837 		}
2838 
2839 		spin_lock(&space_info->lock);
2840 		spin_lock(&cache->lock);
2841 		cache->pinned -= len;
2842 		btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len);
2843 		space_info->max_extent_size = 0;
2844 		percpu_counter_add_batch(&space_info->total_bytes_pinned,
2845 			    -len, BTRFS_TOTAL_BYTES_PINNED_BATCH);
2846 		if (cache->ro) {
2847 			space_info->bytes_readonly += len;
2848 			readonly = true;
2849 		}
2850 		spin_unlock(&cache->lock);
2851 		if (!readonly && return_free_space &&
2852 		    global_rsv->space_info == space_info) {
2853 			u64 to_add = len;
2854 
2855 			spin_lock(&global_rsv->lock);
2856 			if (!global_rsv->full) {
2857 				to_add = min(len, global_rsv->size -
2858 					     global_rsv->reserved);
2859 				global_rsv->reserved += to_add;
2860 				btrfs_space_info_update_bytes_may_use(fs_info,
2861 						space_info, to_add);
2862 				if (global_rsv->reserved >= global_rsv->size)
2863 					global_rsv->full = 1;
2864 				len -= to_add;
2865 			}
2866 			spin_unlock(&global_rsv->lock);
2867 			/* Add to any tickets we may have */
2868 			if (len)
2869 				btrfs_try_granting_tickets(fs_info,
2870 							   space_info);
2871 		}
2872 		spin_unlock(&space_info->lock);
2873 	}
2874 
2875 	if (cache)
2876 		btrfs_put_block_group(cache);
2877 	return 0;
2878 }
2879 
btrfs_finish_extent_commit(struct btrfs_trans_handle * trans)2880 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
2881 {
2882 	struct btrfs_fs_info *fs_info = trans->fs_info;
2883 	struct btrfs_block_group_cache *block_group, *tmp;
2884 	struct list_head *deleted_bgs;
2885 	struct extent_io_tree *unpin;
2886 	u64 start;
2887 	u64 end;
2888 	int ret;
2889 
2890 	if (fs_info->pinned_extents == &fs_info->freed_extents[0])
2891 		unpin = &fs_info->freed_extents[1];
2892 	else
2893 		unpin = &fs_info->freed_extents[0];
2894 
2895 	while (!trans->aborted) {
2896 		struct extent_state *cached_state = NULL;
2897 
2898 		mutex_lock(&fs_info->unused_bg_unpin_mutex);
2899 		ret = find_first_extent_bit(unpin, 0, &start, &end,
2900 					    EXTENT_DIRTY, &cached_state);
2901 		if (ret) {
2902 			mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2903 			break;
2904 		}
2905 
2906 		if (btrfs_test_opt(fs_info, DISCARD))
2907 			ret = btrfs_discard_extent(fs_info, start,
2908 						   end + 1 - start, NULL);
2909 
2910 		clear_extent_dirty(unpin, start, end, &cached_state);
2911 		unpin_extent_range(fs_info, start, end, true);
2912 		mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2913 		free_extent_state(cached_state);
2914 		cond_resched();
2915 	}
2916 
2917 	/*
2918 	 * Transaction is finished.  We don't need the lock anymore.  We
2919 	 * do need to clean up the block groups in case of a transaction
2920 	 * abort.
2921 	 */
2922 	deleted_bgs = &trans->transaction->deleted_bgs;
2923 	list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
2924 		u64 trimmed = 0;
2925 
2926 		ret = -EROFS;
2927 		if (!trans->aborted)
2928 			ret = btrfs_discard_extent(fs_info,
2929 						   block_group->key.objectid,
2930 						   block_group->key.offset,
2931 						   &trimmed);
2932 
2933 		list_del_init(&block_group->bg_list);
2934 		btrfs_put_block_group_trimming(block_group);
2935 		btrfs_put_block_group(block_group);
2936 
2937 		if (ret) {
2938 			const char *errstr = btrfs_decode_error(ret);
2939 			btrfs_warn(fs_info,
2940 			   "discard failed while removing blockgroup: errno=%d %s",
2941 				   ret, errstr);
2942 		}
2943 	}
2944 
2945 	return 0;
2946 }
2947 
__btrfs_free_extent(struct btrfs_trans_handle * trans,struct btrfs_delayed_ref_node * node,u64 parent,u64 root_objectid,u64 owner_objectid,u64 owner_offset,int refs_to_drop,struct btrfs_delayed_extent_op * extent_op)2948 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2949 			       struct btrfs_delayed_ref_node *node, u64 parent,
2950 			       u64 root_objectid, u64 owner_objectid,
2951 			       u64 owner_offset, int refs_to_drop,
2952 			       struct btrfs_delayed_extent_op *extent_op)
2953 {
2954 	struct btrfs_fs_info *info = trans->fs_info;
2955 	struct btrfs_key key;
2956 	struct btrfs_path *path;
2957 	struct btrfs_root *extent_root = info->extent_root;
2958 	struct extent_buffer *leaf;
2959 	struct btrfs_extent_item *ei;
2960 	struct btrfs_extent_inline_ref *iref;
2961 	int ret;
2962 	int is_data;
2963 	int extent_slot = 0;
2964 	int found_extent = 0;
2965 	int num_to_del = 1;
2966 	u32 item_size;
2967 	u64 refs;
2968 	u64 bytenr = node->bytenr;
2969 	u64 num_bytes = node->num_bytes;
2970 	int last_ref = 0;
2971 	bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
2972 
2973 	path = btrfs_alloc_path();
2974 	if (!path)
2975 		return -ENOMEM;
2976 
2977 	path->reada = READA_FORWARD;
2978 	path->leave_spinning = 1;
2979 
2980 	is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2981 	BUG_ON(!is_data && refs_to_drop != 1);
2982 
2983 	if (is_data)
2984 		skinny_metadata = false;
2985 
2986 	ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
2987 				    parent, root_objectid, owner_objectid,
2988 				    owner_offset);
2989 	if (ret == 0) {
2990 		extent_slot = path->slots[0];
2991 		while (extent_slot >= 0) {
2992 			btrfs_item_key_to_cpu(path->nodes[0], &key,
2993 					      extent_slot);
2994 			if (key.objectid != bytenr)
2995 				break;
2996 			if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2997 			    key.offset == num_bytes) {
2998 				found_extent = 1;
2999 				break;
3000 			}
3001 			if (key.type == BTRFS_METADATA_ITEM_KEY &&
3002 			    key.offset == owner_objectid) {
3003 				found_extent = 1;
3004 				break;
3005 			}
3006 			if (path->slots[0] - extent_slot > 5)
3007 				break;
3008 			extent_slot--;
3009 		}
3010 
3011 		if (!found_extent) {
3012 			BUG_ON(iref);
3013 			ret = remove_extent_backref(trans, path, NULL,
3014 						    refs_to_drop,
3015 						    is_data, &last_ref);
3016 			if (ret) {
3017 				btrfs_abort_transaction(trans, ret);
3018 				goto out;
3019 			}
3020 			btrfs_release_path(path);
3021 			path->leave_spinning = 1;
3022 
3023 			key.objectid = bytenr;
3024 			key.type = BTRFS_EXTENT_ITEM_KEY;
3025 			key.offset = num_bytes;
3026 
3027 			if (!is_data && skinny_metadata) {
3028 				key.type = BTRFS_METADATA_ITEM_KEY;
3029 				key.offset = owner_objectid;
3030 			}
3031 
3032 			ret = btrfs_search_slot(trans, extent_root,
3033 						&key, path, -1, 1);
3034 			if (ret > 0 && skinny_metadata && path->slots[0]) {
3035 				/*
3036 				 * Couldn't find our skinny metadata item,
3037 				 * see if we have ye olde extent item.
3038 				 */
3039 				path->slots[0]--;
3040 				btrfs_item_key_to_cpu(path->nodes[0], &key,
3041 						      path->slots[0]);
3042 				if (key.objectid == bytenr &&
3043 				    key.type == BTRFS_EXTENT_ITEM_KEY &&
3044 				    key.offset == num_bytes)
3045 					ret = 0;
3046 			}
3047 
3048 			if (ret > 0 && skinny_metadata) {
3049 				skinny_metadata = false;
3050 				key.objectid = bytenr;
3051 				key.type = BTRFS_EXTENT_ITEM_KEY;
3052 				key.offset = num_bytes;
3053 				btrfs_release_path(path);
3054 				ret = btrfs_search_slot(trans, extent_root,
3055 							&key, path, -1, 1);
3056 			}
3057 
3058 			if (ret) {
3059 				btrfs_err(info,
3060 					  "umm, got %d back from search, was looking for %llu",
3061 					  ret, bytenr);
3062 				if (ret > 0)
3063 					btrfs_print_leaf(path->nodes[0]);
3064 			}
3065 			if (ret < 0) {
3066 				btrfs_abort_transaction(trans, ret);
3067 				goto out;
3068 			}
3069 			extent_slot = path->slots[0];
3070 		}
3071 	} else if (WARN_ON(ret == -ENOENT)) {
3072 		btrfs_print_leaf(path->nodes[0]);
3073 		btrfs_err(info,
3074 			"unable to find ref byte nr %llu parent %llu root %llu  owner %llu offset %llu",
3075 			bytenr, parent, root_objectid, owner_objectid,
3076 			owner_offset);
3077 		btrfs_abort_transaction(trans, ret);
3078 		goto out;
3079 	} else {
3080 		btrfs_abort_transaction(trans, ret);
3081 		goto out;
3082 	}
3083 
3084 	leaf = path->nodes[0];
3085 	item_size = btrfs_item_size_nr(leaf, extent_slot);
3086 	if (unlikely(item_size < sizeof(*ei))) {
3087 		ret = -EINVAL;
3088 		btrfs_print_v0_err(info);
3089 		btrfs_abort_transaction(trans, ret);
3090 		goto out;
3091 	}
3092 	ei = btrfs_item_ptr(leaf, extent_slot,
3093 			    struct btrfs_extent_item);
3094 	if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
3095 	    key.type == BTRFS_EXTENT_ITEM_KEY) {
3096 		struct btrfs_tree_block_info *bi;
3097 		BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3098 		bi = (struct btrfs_tree_block_info *)(ei + 1);
3099 		WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3100 	}
3101 
3102 	refs = btrfs_extent_refs(leaf, ei);
3103 	if (refs < refs_to_drop) {
3104 		btrfs_err(info,
3105 			  "trying to drop %d refs but we only have %Lu for bytenr %Lu",
3106 			  refs_to_drop, refs, bytenr);
3107 		ret = -EINVAL;
3108 		btrfs_abort_transaction(trans, ret);
3109 		goto out;
3110 	}
3111 	refs -= refs_to_drop;
3112 
3113 	if (refs > 0) {
3114 		if (extent_op)
3115 			__run_delayed_extent_op(extent_op, leaf, ei);
3116 		/*
3117 		 * In the case of inline back ref, reference count will
3118 		 * be updated by remove_extent_backref
3119 		 */
3120 		if (iref) {
3121 			BUG_ON(!found_extent);
3122 		} else {
3123 			btrfs_set_extent_refs(leaf, ei, refs);
3124 			btrfs_mark_buffer_dirty(leaf);
3125 		}
3126 		if (found_extent) {
3127 			ret = remove_extent_backref(trans, path, iref,
3128 						    refs_to_drop, is_data,
3129 						    &last_ref);
3130 			if (ret) {
3131 				btrfs_abort_transaction(trans, ret);
3132 				goto out;
3133 			}
3134 		}
3135 	} else {
3136 		if (found_extent) {
3137 			BUG_ON(is_data && refs_to_drop !=
3138 			       extent_data_ref_count(path, iref));
3139 			if (iref) {
3140 				BUG_ON(path->slots[0] != extent_slot);
3141 			} else {
3142 				BUG_ON(path->slots[0] != extent_slot + 1);
3143 				path->slots[0] = extent_slot;
3144 				num_to_del = 2;
3145 			}
3146 		}
3147 
3148 		last_ref = 1;
3149 		ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3150 				      num_to_del);
3151 		if (ret) {
3152 			btrfs_abort_transaction(trans, ret);
3153 			goto out;
3154 		}
3155 		btrfs_release_path(path);
3156 
3157 		if (is_data) {
3158 			ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
3159 			if (ret) {
3160 				btrfs_abort_transaction(trans, ret);
3161 				goto out;
3162 			}
3163 		}
3164 
3165 		ret = add_to_free_space_tree(trans, bytenr, num_bytes);
3166 		if (ret) {
3167 			btrfs_abort_transaction(trans, ret);
3168 			goto out;
3169 		}
3170 
3171 		ret = btrfs_update_block_group(trans, bytenr, num_bytes, 0);
3172 		if (ret) {
3173 			btrfs_abort_transaction(trans, ret);
3174 			goto out;
3175 		}
3176 	}
3177 	btrfs_release_path(path);
3178 
3179 out:
3180 	btrfs_free_path(path);
3181 	return ret;
3182 }
3183 
3184 /*
3185  * when we free an block, it is possible (and likely) that we free the last
3186  * delayed ref for that extent as well.  This searches the delayed ref tree for
3187  * a given extent, and if there are no other delayed refs to be processed, it
3188  * removes it from the tree.
3189  */
check_ref_cleanup(struct btrfs_trans_handle * trans,u64 bytenr)3190 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3191 				      u64 bytenr)
3192 {
3193 	struct btrfs_delayed_ref_head *head;
3194 	struct btrfs_delayed_ref_root *delayed_refs;
3195 	int ret = 0;
3196 
3197 	delayed_refs = &trans->transaction->delayed_refs;
3198 	spin_lock(&delayed_refs->lock);
3199 	head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3200 	if (!head)
3201 		goto out_delayed_unlock;
3202 
3203 	spin_lock(&head->lock);
3204 	if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
3205 		goto out;
3206 
3207 	if (cleanup_extent_op(head) != NULL)
3208 		goto out;
3209 
3210 	/*
3211 	 * waiting for the lock here would deadlock.  If someone else has it
3212 	 * locked they are already in the process of dropping it anyway
3213 	 */
3214 	if (!mutex_trylock(&head->mutex))
3215 		goto out;
3216 
3217 	btrfs_delete_ref_head(delayed_refs, head);
3218 	head->processing = 0;
3219 
3220 	spin_unlock(&head->lock);
3221 	spin_unlock(&delayed_refs->lock);
3222 
3223 	BUG_ON(head->extent_op);
3224 	if (head->must_insert_reserved)
3225 		ret = 1;
3226 
3227 	btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
3228 	mutex_unlock(&head->mutex);
3229 	btrfs_put_delayed_ref_head(head);
3230 	return ret;
3231 out:
3232 	spin_unlock(&head->lock);
3233 
3234 out_delayed_unlock:
3235 	spin_unlock(&delayed_refs->lock);
3236 	return 0;
3237 }
3238 
btrfs_free_tree_block(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * buf,u64 parent,int last_ref)3239 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
3240 			   struct btrfs_root *root,
3241 			   struct extent_buffer *buf,
3242 			   u64 parent, int last_ref)
3243 {
3244 	struct btrfs_fs_info *fs_info = root->fs_info;
3245 	struct btrfs_ref generic_ref = { 0 };
3246 	int pin = 1;
3247 	int ret;
3248 
3249 	btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF,
3250 			       buf->start, buf->len, parent);
3251 	btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf),
3252 			    root->root_key.objectid);
3253 
3254 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
3255 		int old_ref_mod, new_ref_mod;
3256 
3257 		btrfs_ref_tree_mod(fs_info, &generic_ref);
3258 		ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL,
3259 						 &old_ref_mod, &new_ref_mod);
3260 		BUG_ON(ret); /* -ENOMEM */
3261 		pin = old_ref_mod >= 0 && new_ref_mod < 0;
3262 	}
3263 
3264 	if (last_ref && btrfs_header_generation(buf) == trans->transid) {
3265 		struct btrfs_block_group_cache *cache;
3266 
3267 		if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
3268 			ret = check_ref_cleanup(trans, buf->start);
3269 			if (!ret)
3270 				goto out;
3271 		}
3272 
3273 		pin = 0;
3274 		cache = btrfs_lookup_block_group(fs_info, buf->start);
3275 
3276 		if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3277 			pin_down_extent(cache, buf->start, buf->len, 1);
3278 			btrfs_put_block_group(cache);
3279 			goto out;
3280 		}
3281 
3282 		WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
3283 
3284 		btrfs_add_free_space(cache, buf->start, buf->len);
3285 		btrfs_free_reserved_bytes(cache, buf->len, 0);
3286 		btrfs_put_block_group(cache);
3287 		trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
3288 	}
3289 out:
3290 	if (pin)
3291 		add_pinned_bytes(fs_info, &generic_ref);
3292 
3293 	if (last_ref) {
3294 		/*
3295 		 * Deleting the buffer, clear the corrupt flag since it doesn't
3296 		 * matter anymore.
3297 		 */
3298 		clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
3299 	}
3300 }
3301 
3302 /* Can return -ENOMEM */
btrfs_free_extent(struct btrfs_trans_handle * trans,struct btrfs_ref * ref)3303 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
3304 {
3305 	struct btrfs_fs_info *fs_info = trans->fs_info;
3306 	int old_ref_mod, new_ref_mod;
3307 	int ret;
3308 
3309 	if (btrfs_is_testing(fs_info))
3310 		return 0;
3311 
3312 	/*
3313 	 * tree log blocks never actually go into the extent allocation
3314 	 * tree, just update pinning info and exit early.
3315 	 */
3316 	if ((ref->type == BTRFS_REF_METADATA &&
3317 	     ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) ||
3318 	    (ref->type == BTRFS_REF_DATA &&
3319 	     ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) {
3320 		/* unlocks the pinned mutex */
3321 		btrfs_pin_extent(fs_info, ref->bytenr, ref->len, 1);
3322 		old_ref_mod = new_ref_mod = 0;
3323 		ret = 0;
3324 	} else if (ref->type == BTRFS_REF_METADATA) {
3325 		ret = btrfs_add_delayed_tree_ref(trans, ref, NULL,
3326 						 &old_ref_mod, &new_ref_mod);
3327 	} else {
3328 		ret = btrfs_add_delayed_data_ref(trans, ref, 0,
3329 						 &old_ref_mod, &new_ref_mod);
3330 	}
3331 
3332 	if (!((ref->type == BTRFS_REF_METADATA &&
3333 	       ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) ||
3334 	      (ref->type == BTRFS_REF_DATA &&
3335 	       ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)))
3336 		btrfs_ref_tree_mod(fs_info, ref);
3337 
3338 	if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0)
3339 		add_pinned_bytes(fs_info, ref);
3340 
3341 	return ret;
3342 }
3343 
3344 enum btrfs_loop_type {
3345 	LOOP_CACHING_NOWAIT,
3346 	LOOP_CACHING_WAIT,
3347 	LOOP_ALLOC_CHUNK,
3348 	LOOP_NO_EMPTY_SIZE,
3349 };
3350 
3351 static inline void
btrfs_lock_block_group(struct btrfs_block_group_cache * cache,int delalloc)3352 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
3353 		       int delalloc)
3354 {
3355 	if (delalloc)
3356 		down_read(&cache->data_rwsem);
3357 }
3358 
3359 static inline void
btrfs_grab_block_group(struct btrfs_block_group_cache * cache,int delalloc)3360 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
3361 		       int delalloc)
3362 {
3363 	btrfs_get_block_group(cache);
3364 	if (delalloc)
3365 		down_read(&cache->data_rwsem);
3366 }
3367 
3368 static struct btrfs_block_group_cache *
btrfs_lock_cluster(struct btrfs_block_group_cache * block_group,struct btrfs_free_cluster * cluster,int delalloc)3369 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
3370 		   struct btrfs_free_cluster *cluster,
3371 		   int delalloc)
3372 {
3373 	struct btrfs_block_group_cache *used_bg = NULL;
3374 
3375 	spin_lock(&cluster->refill_lock);
3376 	while (1) {
3377 		used_bg = cluster->block_group;
3378 		if (!used_bg)
3379 			return NULL;
3380 
3381 		if (used_bg == block_group)
3382 			return used_bg;
3383 
3384 		btrfs_get_block_group(used_bg);
3385 
3386 		if (!delalloc)
3387 			return used_bg;
3388 
3389 		if (down_read_trylock(&used_bg->data_rwsem))
3390 			return used_bg;
3391 
3392 		spin_unlock(&cluster->refill_lock);
3393 
3394 		/* We should only have one-level nested. */
3395 		down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
3396 
3397 		spin_lock(&cluster->refill_lock);
3398 		if (used_bg == cluster->block_group)
3399 			return used_bg;
3400 
3401 		up_read(&used_bg->data_rwsem);
3402 		btrfs_put_block_group(used_bg);
3403 	}
3404 }
3405 
3406 static inline void
btrfs_release_block_group(struct btrfs_block_group_cache * cache,int delalloc)3407 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
3408 			 int delalloc)
3409 {
3410 	if (delalloc)
3411 		up_read(&cache->data_rwsem);
3412 	btrfs_put_block_group(cache);
3413 }
3414 
3415 /*
3416  * Structure used internally for find_free_extent() function.  Wraps needed
3417  * parameters.
3418  */
3419 struct find_free_extent_ctl {
3420 	/* Basic allocation info */
3421 	u64 ram_bytes;
3422 	u64 num_bytes;
3423 	u64 empty_size;
3424 	u64 flags;
3425 	int delalloc;
3426 
3427 	/* Where to start the search inside the bg */
3428 	u64 search_start;
3429 
3430 	/* For clustered allocation */
3431 	u64 empty_cluster;
3432 
3433 	bool have_caching_bg;
3434 	bool orig_have_caching_bg;
3435 
3436 	/* RAID index, converted from flags */
3437 	int index;
3438 
3439 	/*
3440 	 * Current loop number, check find_free_extent_update_loop() for details
3441 	 */
3442 	int loop;
3443 
3444 	/*
3445 	 * Whether we're refilling a cluster, if true we need to re-search
3446 	 * current block group but don't try to refill the cluster again.
3447 	 */
3448 	bool retry_clustered;
3449 
3450 	/*
3451 	 * Whether we're updating free space cache, if true we need to re-search
3452 	 * current block group but don't try updating free space cache again.
3453 	 */
3454 	bool retry_unclustered;
3455 
3456 	/* If current block group is cached */
3457 	int cached;
3458 
3459 	/* Max contiguous hole found */
3460 	u64 max_extent_size;
3461 
3462 	/* Total free space from free space cache, not always contiguous */
3463 	u64 total_free_space;
3464 
3465 	/* Found result */
3466 	u64 found_offset;
3467 };
3468 
3469 
3470 /*
3471  * Helper function for find_free_extent().
3472  *
3473  * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3474  * Return -EAGAIN to inform caller that we need to re-search this block group
3475  * Return >0 to inform caller that we find nothing
3476  * Return 0 means we have found a location and set ffe_ctl->found_offset.
3477  */
find_free_extent_clustered(struct btrfs_block_group_cache * bg,struct btrfs_free_cluster * last_ptr,struct find_free_extent_ctl * ffe_ctl,struct btrfs_block_group_cache ** cluster_bg_ret)3478 static int find_free_extent_clustered(struct btrfs_block_group_cache *bg,
3479 		struct btrfs_free_cluster *last_ptr,
3480 		struct find_free_extent_ctl *ffe_ctl,
3481 		struct btrfs_block_group_cache **cluster_bg_ret)
3482 {
3483 	struct btrfs_block_group_cache *cluster_bg;
3484 	u64 aligned_cluster;
3485 	u64 offset;
3486 	int ret;
3487 
3488 	cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
3489 	if (!cluster_bg)
3490 		goto refill_cluster;
3491 	if (cluster_bg != bg && (cluster_bg->ro ||
3492 	    !block_group_bits(cluster_bg, ffe_ctl->flags)))
3493 		goto release_cluster;
3494 
3495 	offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
3496 			ffe_ctl->num_bytes, cluster_bg->key.objectid,
3497 			&ffe_ctl->max_extent_size);
3498 	if (offset) {
3499 		/* We have a block, we're done */
3500 		spin_unlock(&last_ptr->refill_lock);
3501 		trace_btrfs_reserve_extent_cluster(cluster_bg,
3502 				ffe_ctl->search_start, ffe_ctl->num_bytes);
3503 		*cluster_bg_ret = cluster_bg;
3504 		ffe_ctl->found_offset = offset;
3505 		return 0;
3506 	}
3507 	WARN_ON(last_ptr->block_group != cluster_bg);
3508 
3509 release_cluster:
3510 	/*
3511 	 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3512 	 * lets just skip it and let the allocator find whatever block it can
3513 	 * find. If we reach this point, we will have tried the cluster
3514 	 * allocator plenty of times and not have found anything, so we are
3515 	 * likely way too fragmented for the clustering stuff to find anything.
3516 	 *
3517 	 * However, if the cluster is taken from the current block group,
3518 	 * release the cluster first, so that we stand a better chance of
3519 	 * succeeding in the unclustered allocation.
3520 	 */
3521 	if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
3522 		spin_unlock(&last_ptr->refill_lock);
3523 		btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3524 		return -ENOENT;
3525 	}
3526 
3527 	/* This cluster didn't work out, free it and start over */
3528 	btrfs_return_cluster_to_free_space(NULL, last_ptr);
3529 
3530 	if (cluster_bg != bg)
3531 		btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3532 
3533 refill_cluster:
3534 	if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
3535 		spin_unlock(&last_ptr->refill_lock);
3536 		return -ENOENT;
3537 	}
3538 
3539 	aligned_cluster = max_t(u64,
3540 			ffe_ctl->empty_cluster + ffe_ctl->empty_size,
3541 			bg->full_stripe_len);
3542 	ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start,
3543 			ffe_ctl->num_bytes, aligned_cluster);
3544 	if (ret == 0) {
3545 		/* Now pull our allocation out of this cluster */
3546 		offset = btrfs_alloc_from_cluster(bg, last_ptr,
3547 				ffe_ctl->num_bytes, ffe_ctl->search_start,
3548 				&ffe_ctl->max_extent_size);
3549 		if (offset) {
3550 			/* We found one, proceed */
3551 			spin_unlock(&last_ptr->refill_lock);
3552 			trace_btrfs_reserve_extent_cluster(bg,
3553 					ffe_ctl->search_start,
3554 					ffe_ctl->num_bytes);
3555 			ffe_ctl->found_offset = offset;
3556 			return 0;
3557 		}
3558 	} else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
3559 		   !ffe_ctl->retry_clustered) {
3560 		spin_unlock(&last_ptr->refill_lock);
3561 
3562 		ffe_ctl->retry_clustered = true;
3563 		btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3564 				ffe_ctl->empty_cluster + ffe_ctl->empty_size);
3565 		return -EAGAIN;
3566 	}
3567 	/*
3568 	 * At this point we either didn't find a cluster or we weren't able to
3569 	 * allocate a block from our cluster.  Free the cluster we've been
3570 	 * trying to use, and go to the next block group.
3571 	 */
3572 	btrfs_return_cluster_to_free_space(NULL, last_ptr);
3573 	spin_unlock(&last_ptr->refill_lock);
3574 	return 1;
3575 }
3576 
3577 /*
3578  * Return >0 to inform caller that we find nothing
3579  * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3580  * Return -EAGAIN to inform caller that we need to re-search this block group
3581  */
find_free_extent_unclustered(struct btrfs_block_group_cache * bg,struct btrfs_free_cluster * last_ptr,struct find_free_extent_ctl * ffe_ctl)3582 static int find_free_extent_unclustered(struct btrfs_block_group_cache *bg,
3583 		struct btrfs_free_cluster *last_ptr,
3584 		struct find_free_extent_ctl *ffe_ctl)
3585 {
3586 	u64 offset;
3587 
3588 	/*
3589 	 * We are doing an unclustered allocation, set the fragmented flag so
3590 	 * we don't bother trying to setup a cluster again until we get more
3591 	 * space.
3592 	 */
3593 	if (unlikely(last_ptr)) {
3594 		spin_lock(&last_ptr->lock);
3595 		last_ptr->fragmented = 1;
3596 		spin_unlock(&last_ptr->lock);
3597 	}
3598 	if (ffe_ctl->cached) {
3599 		struct btrfs_free_space_ctl *free_space_ctl;
3600 
3601 		free_space_ctl = bg->free_space_ctl;
3602 		spin_lock(&free_space_ctl->tree_lock);
3603 		if (free_space_ctl->free_space <
3604 		    ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
3605 		    ffe_ctl->empty_size) {
3606 			ffe_ctl->total_free_space = max_t(u64,
3607 					ffe_ctl->total_free_space,
3608 					free_space_ctl->free_space);
3609 			spin_unlock(&free_space_ctl->tree_lock);
3610 			return 1;
3611 		}
3612 		spin_unlock(&free_space_ctl->tree_lock);
3613 	}
3614 
3615 	offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
3616 			ffe_ctl->num_bytes, ffe_ctl->empty_size,
3617 			&ffe_ctl->max_extent_size);
3618 
3619 	/*
3620 	 * If we didn't find a chunk, and we haven't failed on this block group
3621 	 * before, and this block group is in the middle of caching and we are
3622 	 * ok with waiting, then go ahead and wait for progress to be made, and
3623 	 * set @retry_unclustered to true.
3624 	 *
3625 	 * If @retry_unclustered is true then we've already waited on this
3626 	 * block group once and should move on to the next block group.
3627 	 */
3628 	if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
3629 	    ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
3630 		btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3631 						      ffe_ctl->empty_size);
3632 		ffe_ctl->retry_unclustered = true;
3633 		return -EAGAIN;
3634 	} else if (!offset) {
3635 		return 1;
3636 	}
3637 	ffe_ctl->found_offset = offset;
3638 	return 0;
3639 }
3640 
3641 /*
3642  * Return >0 means caller needs to re-search for free extent
3643  * Return 0 means we have the needed free extent.
3644  * Return <0 means we failed to locate any free extent.
3645  */
find_free_extent_update_loop(struct btrfs_fs_info * fs_info,struct btrfs_free_cluster * last_ptr,struct btrfs_key * ins,struct find_free_extent_ctl * ffe_ctl,int full_search,bool use_cluster)3646 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
3647 					struct btrfs_free_cluster *last_ptr,
3648 					struct btrfs_key *ins,
3649 					struct find_free_extent_ctl *ffe_ctl,
3650 					int full_search, bool use_cluster)
3651 {
3652 	struct btrfs_root *root = fs_info->extent_root;
3653 	int ret;
3654 
3655 	if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
3656 	    ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
3657 		ffe_ctl->orig_have_caching_bg = true;
3658 
3659 	if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT &&
3660 	    ffe_ctl->have_caching_bg)
3661 		return 1;
3662 
3663 	if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES)
3664 		return 1;
3665 
3666 	if (ins->objectid) {
3667 		if (!use_cluster && last_ptr) {
3668 			spin_lock(&last_ptr->lock);
3669 			last_ptr->window_start = ins->objectid;
3670 			spin_unlock(&last_ptr->lock);
3671 		}
3672 		return 0;
3673 	}
3674 
3675 	/*
3676 	 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
3677 	 *			caching kthreads as we move along
3678 	 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
3679 	 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
3680 	 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
3681 	 *		       again
3682 	 */
3683 	if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
3684 		ffe_ctl->index = 0;
3685 		if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) {
3686 			/*
3687 			 * We want to skip the LOOP_CACHING_WAIT step if we
3688 			 * don't have any uncached bgs and we've already done a
3689 			 * full search through.
3690 			 */
3691 			if (ffe_ctl->orig_have_caching_bg || !full_search)
3692 				ffe_ctl->loop = LOOP_CACHING_WAIT;
3693 			else
3694 				ffe_ctl->loop = LOOP_ALLOC_CHUNK;
3695 		} else {
3696 			ffe_ctl->loop++;
3697 		}
3698 
3699 		if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
3700 			struct btrfs_trans_handle *trans;
3701 			int exist = 0;
3702 
3703 			trans = current->journal_info;
3704 			if (trans)
3705 				exist = 1;
3706 			else
3707 				trans = btrfs_join_transaction(root);
3708 
3709 			if (IS_ERR(trans)) {
3710 				ret = PTR_ERR(trans);
3711 				return ret;
3712 			}
3713 
3714 			ret = btrfs_chunk_alloc(trans, ffe_ctl->flags,
3715 						CHUNK_ALLOC_FORCE);
3716 
3717 			/*
3718 			 * If we can't allocate a new chunk we've already looped
3719 			 * through at least once, move on to the NO_EMPTY_SIZE
3720 			 * case.
3721 			 */
3722 			if (ret == -ENOSPC)
3723 				ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
3724 
3725 			/* Do not bail out on ENOSPC since we can do more. */
3726 			if (ret < 0 && ret != -ENOSPC)
3727 				btrfs_abort_transaction(trans, ret);
3728 			else
3729 				ret = 0;
3730 			if (!exist)
3731 				btrfs_end_transaction(trans);
3732 			if (ret)
3733 				return ret;
3734 		}
3735 
3736 		if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
3737 			/*
3738 			 * Don't loop again if we already have no empty_size and
3739 			 * no empty_cluster.
3740 			 */
3741 			if (ffe_ctl->empty_size == 0 &&
3742 			    ffe_ctl->empty_cluster == 0)
3743 				return -ENOSPC;
3744 			ffe_ctl->empty_size = 0;
3745 			ffe_ctl->empty_cluster = 0;
3746 		}
3747 		return 1;
3748 	}
3749 	return -ENOSPC;
3750 }
3751 
3752 /*
3753  * walks the btree of allocated extents and find a hole of a given size.
3754  * The key ins is changed to record the hole:
3755  * ins->objectid == start position
3756  * ins->flags = BTRFS_EXTENT_ITEM_KEY
3757  * ins->offset == the size of the hole.
3758  * Any available blocks before search_start are skipped.
3759  *
3760  * If there is no suitable free space, we will record the max size of
3761  * the free space extent currently.
3762  *
3763  * The overall logic and call chain:
3764  *
3765  * find_free_extent()
3766  * |- Iterate through all block groups
3767  * |  |- Get a valid block group
3768  * |  |- Try to do clustered allocation in that block group
3769  * |  |- Try to do unclustered allocation in that block group
3770  * |  |- Check if the result is valid
3771  * |  |  |- If valid, then exit
3772  * |  |- Jump to next block group
3773  * |
3774  * |- Push harder to find free extents
3775  *    |- If not found, re-iterate all block groups
3776  */
find_free_extent(struct btrfs_fs_info * fs_info,u64 ram_bytes,u64 num_bytes,u64 empty_size,u64 hint_byte,struct btrfs_key * ins,u64 flags,int delalloc)3777 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
3778 				u64 ram_bytes, u64 num_bytes, u64 empty_size,
3779 				u64 hint_byte, struct btrfs_key *ins,
3780 				u64 flags, int delalloc)
3781 {
3782 	int ret = 0;
3783 	struct btrfs_free_cluster *last_ptr = NULL;
3784 	struct btrfs_block_group_cache *block_group = NULL;
3785 	struct find_free_extent_ctl ffe_ctl = {0};
3786 	struct btrfs_space_info *space_info;
3787 	bool use_cluster = true;
3788 	bool full_search = false;
3789 
3790 	WARN_ON(num_bytes < fs_info->sectorsize);
3791 
3792 	ffe_ctl.ram_bytes = ram_bytes;
3793 	ffe_ctl.num_bytes = num_bytes;
3794 	ffe_ctl.empty_size = empty_size;
3795 	ffe_ctl.flags = flags;
3796 	ffe_ctl.search_start = 0;
3797 	ffe_ctl.retry_clustered = false;
3798 	ffe_ctl.retry_unclustered = false;
3799 	ffe_ctl.delalloc = delalloc;
3800 	ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags);
3801 	ffe_ctl.have_caching_bg = false;
3802 	ffe_ctl.orig_have_caching_bg = false;
3803 	ffe_ctl.found_offset = 0;
3804 
3805 	ins->type = BTRFS_EXTENT_ITEM_KEY;
3806 	ins->objectid = 0;
3807 	ins->offset = 0;
3808 
3809 	trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
3810 
3811 	space_info = btrfs_find_space_info(fs_info, flags);
3812 	if (!space_info) {
3813 		btrfs_err(fs_info, "No space info for %llu", flags);
3814 		return -ENOSPC;
3815 	}
3816 
3817 	/*
3818 	 * If our free space is heavily fragmented we may not be able to make
3819 	 * big contiguous allocations, so instead of doing the expensive search
3820 	 * for free space, simply return ENOSPC with our max_extent_size so we
3821 	 * can go ahead and search for a more manageable chunk.
3822 	 *
3823 	 * If our max_extent_size is large enough for our allocation simply
3824 	 * disable clustering since we will likely not be able to find enough
3825 	 * space to create a cluster and induce latency trying.
3826 	 */
3827 	if (unlikely(space_info->max_extent_size)) {
3828 		spin_lock(&space_info->lock);
3829 		if (space_info->max_extent_size &&
3830 		    num_bytes > space_info->max_extent_size) {
3831 			ins->offset = space_info->max_extent_size;
3832 			spin_unlock(&space_info->lock);
3833 			return -ENOSPC;
3834 		} else if (space_info->max_extent_size) {
3835 			use_cluster = false;
3836 		}
3837 		spin_unlock(&space_info->lock);
3838 	}
3839 
3840 	last_ptr = fetch_cluster_info(fs_info, space_info,
3841 				      &ffe_ctl.empty_cluster);
3842 	if (last_ptr) {
3843 		spin_lock(&last_ptr->lock);
3844 		if (last_ptr->block_group)
3845 			hint_byte = last_ptr->window_start;
3846 		if (last_ptr->fragmented) {
3847 			/*
3848 			 * We still set window_start so we can keep track of the
3849 			 * last place we found an allocation to try and save
3850 			 * some time.
3851 			 */
3852 			hint_byte = last_ptr->window_start;
3853 			use_cluster = false;
3854 		}
3855 		spin_unlock(&last_ptr->lock);
3856 	}
3857 
3858 	ffe_ctl.search_start = max(ffe_ctl.search_start,
3859 				   first_logical_byte(fs_info, 0));
3860 	ffe_ctl.search_start = max(ffe_ctl.search_start, hint_byte);
3861 	if (ffe_ctl.search_start == hint_byte) {
3862 		block_group = btrfs_lookup_block_group(fs_info,
3863 						       ffe_ctl.search_start);
3864 		/*
3865 		 * we don't want to use the block group if it doesn't match our
3866 		 * allocation bits, or if its not cached.
3867 		 *
3868 		 * However if we are re-searching with an ideal block group
3869 		 * picked out then we don't care that the block group is cached.
3870 		 */
3871 		if (block_group && block_group_bits(block_group, flags) &&
3872 		    block_group->cached != BTRFS_CACHE_NO) {
3873 			down_read(&space_info->groups_sem);
3874 			if (list_empty(&block_group->list) ||
3875 			    block_group->ro) {
3876 				/*
3877 				 * someone is removing this block group,
3878 				 * we can't jump into the have_block_group
3879 				 * target because our list pointers are not
3880 				 * valid
3881 				 */
3882 				btrfs_put_block_group(block_group);
3883 				up_read(&space_info->groups_sem);
3884 			} else {
3885 				ffe_ctl.index = btrfs_bg_flags_to_raid_index(
3886 						block_group->flags);
3887 				btrfs_lock_block_group(block_group, delalloc);
3888 				goto have_block_group;
3889 			}
3890 		} else if (block_group) {
3891 			btrfs_put_block_group(block_group);
3892 		}
3893 	}
3894 search:
3895 	ffe_ctl.have_caching_bg = false;
3896 	if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) ||
3897 	    ffe_ctl.index == 0)
3898 		full_search = true;
3899 	down_read(&space_info->groups_sem);
3900 	list_for_each_entry(block_group,
3901 			    &space_info->block_groups[ffe_ctl.index], list) {
3902 		/* If the block group is read-only, we can skip it entirely. */
3903 		if (unlikely(block_group->ro))
3904 			continue;
3905 
3906 		btrfs_grab_block_group(block_group, delalloc);
3907 		ffe_ctl.search_start = block_group->key.objectid;
3908 
3909 		/*
3910 		 * this can happen if we end up cycling through all the
3911 		 * raid types, but we want to make sure we only allocate
3912 		 * for the proper type.
3913 		 */
3914 		if (!block_group_bits(block_group, flags)) {
3915 			u64 extra = BTRFS_BLOCK_GROUP_DUP |
3916 				BTRFS_BLOCK_GROUP_RAID1_MASK |
3917 				BTRFS_BLOCK_GROUP_RAID56_MASK |
3918 				BTRFS_BLOCK_GROUP_RAID10;
3919 
3920 			/*
3921 			 * if they asked for extra copies and this block group
3922 			 * doesn't provide them, bail.  This does allow us to
3923 			 * fill raid0 from raid1.
3924 			 */
3925 			if ((flags & extra) && !(block_group->flags & extra))
3926 				goto loop;
3927 
3928 			/*
3929 			 * This block group has different flags than we want.
3930 			 * It's possible that we have MIXED_GROUP flag but no
3931 			 * block group is mixed.  Just skip such block group.
3932 			 */
3933 			btrfs_release_block_group(block_group, delalloc);
3934 			continue;
3935 		}
3936 
3937 have_block_group:
3938 		ffe_ctl.cached = btrfs_block_group_cache_done(block_group);
3939 		if (unlikely(!ffe_ctl.cached)) {
3940 			ffe_ctl.have_caching_bg = true;
3941 			ret = btrfs_cache_block_group(block_group, 0);
3942 			BUG_ON(ret < 0);
3943 			ret = 0;
3944 		}
3945 
3946 		if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
3947 			goto loop;
3948 
3949 		/*
3950 		 * Ok we want to try and use the cluster allocator, so
3951 		 * lets look there
3952 		 */
3953 		if (last_ptr && use_cluster) {
3954 			struct btrfs_block_group_cache *cluster_bg = NULL;
3955 
3956 			ret = find_free_extent_clustered(block_group, last_ptr,
3957 							 &ffe_ctl, &cluster_bg);
3958 
3959 			if (ret == 0) {
3960 				if (cluster_bg && cluster_bg != block_group) {
3961 					btrfs_release_block_group(block_group,
3962 								  delalloc);
3963 					block_group = cluster_bg;
3964 				}
3965 				goto checks;
3966 			} else if (ret == -EAGAIN) {
3967 				goto have_block_group;
3968 			} else if (ret > 0) {
3969 				goto loop;
3970 			}
3971 			/* ret == -ENOENT case falls through */
3972 		}
3973 
3974 		ret = find_free_extent_unclustered(block_group, last_ptr,
3975 						   &ffe_ctl);
3976 		if (ret == -EAGAIN)
3977 			goto have_block_group;
3978 		else if (ret > 0)
3979 			goto loop;
3980 		/* ret == 0 case falls through */
3981 checks:
3982 		ffe_ctl.search_start = round_up(ffe_ctl.found_offset,
3983 					     fs_info->stripesize);
3984 
3985 		/* move on to the next group */
3986 		if (ffe_ctl.search_start + num_bytes >
3987 		    block_group->key.objectid + block_group->key.offset) {
3988 			btrfs_add_free_space(block_group, ffe_ctl.found_offset,
3989 					     num_bytes);
3990 			goto loop;
3991 		}
3992 
3993 		if (ffe_ctl.found_offset < ffe_ctl.search_start)
3994 			btrfs_add_free_space(block_group, ffe_ctl.found_offset,
3995 				ffe_ctl.search_start - ffe_ctl.found_offset);
3996 
3997 		ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
3998 				num_bytes, delalloc);
3999 		if (ret == -EAGAIN) {
4000 			btrfs_add_free_space(block_group, ffe_ctl.found_offset,
4001 					     num_bytes);
4002 			goto loop;
4003 		}
4004 		btrfs_inc_block_group_reservations(block_group);
4005 
4006 		/* we are all good, lets return */
4007 		ins->objectid = ffe_ctl.search_start;
4008 		ins->offset = num_bytes;
4009 
4010 		trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start,
4011 					   num_bytes);
4012 		btrfs_release_block_group(block_group, delalloc);
4013 		break;
4014 loop:
4015 		ffe_ctl.retry_clustered = false;
4016 		ffe_ctl.retry_unclustered = false;
4017 		BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
4018 		       ffe_ctl.index);
4019 		btrfs_release_block_group(block_group, delalloc);
4020 		cond_resched();
4021 	}
4022 	up_read(&space_info->groups_sem);
4023 
4024 	ret = find_free_extent_update_loop(fs_info, last_ptr, ins, &ffe_ctl,
4025 					   full_search, use_cluster);
4026 	if (ret > 0)
4027 		goto search;
4028 
4029 	if (ret == -ENOSPC) {
4030 		/*
4031 		 * Use ffe_ctl->total_free_space as fallback if we can't find
4032 		 * any contiguous hole.
4033 		 */
4034 		if (!ffe_ctl.max_extent_size)
4035 			ffe_ctl.max_extent_size = ffe_ctl.total_free_space;
4036 		spin_lock(&space_info->lock);
4037 		space_info->max_extent_size = ffe_ctl.max_extent_size;
4038 		spin_unlock(&space_info->lock);
4039 		ins->offset = ffe_ctl.max_extent_size;
4040 	}
4041 	return ret;
4042 }
4043 
4044 /*
4045  * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
4046  *			  hole that is at least as big as @num_bytes.
4047  *
4048  * @root           -	The root that will contain this extent
4049  *
4050  * @ram_bytes      -	The amount of space in ram that @num_bytes take. This
4051  *			is used for accounting purposes. This value differs
4052  *			from @num_bytes only in the case of compressed extents.
4053  *
4054  * @num_bytes      -	Number of bytes to allocate on-disk.
4055  *
4056  * @min_alloc_size -	Indicates the minimum amount of space that the
4057  *			allocator should try to satisfy. In some cases
4058  *			@num_bytes may be larger than what is required and if
4059  *			the filesystem is fragmented then allocation fails.
4060  *			However, the presence of @min_alloc_size gives a
4061  *			chance to try and satisfy the smaller allocation.
4062  *
4063  * @empty_size     -	A hint that you plan on doing more COW. This is the
4064  *			size in bytes the allocator should try to find free
4065  *			next to the block it returns.  This is just a hint and
4066  *			may be ignored by the allocator.
4067  *
4068  * @hint_byte      -	Hint to the allocator to start searching above the byte
4069  *			address passed. It might be ignored.
4070  *
4071  * @ins            -	This key is modified to record the found hole. It will
4072  *			have the following values:
4073  *			ins->objectid == start position
4074  *			ins->flags = BTRFS_EXTENT_ITEM_KEY
4075  *			ins->offset == the size of the hole.
4076  *
4077  * @is_data        -	Boolean flag indicating whether an extent is
4078  *			allocated for data (true) or metadata (false)
4079  *
4080  * @delalloc       -	Boolean flag indicating whether this allocation is for
4081  *			delalloc or not. If 'true' data_rwsem of block groups
4082  *			is going to be acquired.
4083  *
4084  *
4085  * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4086  * case -ENOSPC is returned then @ins->offset will contain the size of the
4087  * largest available hole the allocator managed to find.
4088  */
btrfs_reserve_extent(struct btrfs_root * root,u64 ram_bytes,u64 num_bytes,u64 min_alloc_size,u64 empty_size,u64 hint_byte,struct btrfs_key * ins,int is_data,int delalloc)4089 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
4090 			 u64 num_bytes, u64 min_alloc_size,
4091 			 u64 empty_size, u64 hint_byte,
4092 			 struct btrfs_key *ins, int is_data, int delalloc)
4093 {
4094 	struct btrfs_fs_info *fs_info = root->fs_info;
4095 	bool final_tried = num_bytes == min_alloc_size;
4096 	u64 flags;
4097 	int ret;
4098 
4099 	flags = get_alloc_profile_by_root(root, is_data);
4100 again:
4101 	WARN_ON(num_bytes < fs_info->sectorsize);
4102 	ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
4103 			       hint_byte, ins, flags, delalloc);
4104 	if (!ret && !is_data) {
4105 		btrfs_dec_block_group_reservations(fs_info, ins->objectid);
4106 	} else if (ret == -ENOSPC) {
4107 		if (!final_tried && ins->offset) {
4108 			num_bytes = min(num_bytes >> 1, ins->offset);
4109 			num_bytes = round_down(num_bytes,
4110 					       fs_info->sectorsize);
4111 			num_bytes = max(num_bytes, min_alloc_size);
4112 			ram_bytes = num_bytes;
4113 			if (num_bytes == min_alloc_size)
4114 				final_tried = true;
4115 			goto again;
4116 		} else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4117 			struct btrfs_space_info *sinfo;
4118 
4119 			sinfo = btrfs_find_space_info(fs_info, flags);
4120 			btrfs_err(fs_info,
4121 				  "allocation failed flags %llu, wanted %llu",
4122 				  flags, num_bytes);
4123 			if (sinfo)
4124 				btrfs_dump_space_info(fs_info, sinfo,
4125 						      num_bytes, 1);
4126 		}
4127 	}
4128 
4129 	return ret;
4130 }
4131 
__btrfs_free_reserved_extent(struct btrfs_fs_info * fs_info,u64 start,u64 len,int pin,int delalloc)4132 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
4133 					u64 start, u64 len,
4134 					int pin, int delalloc)
4135 {
4136 	struct btrfs_block_group_cache *cache;
4137 	int ret = 0;
4138 
4139 	cache = btrfs_lookup_block_group(fs_info, start);
4140 	if (!cache) {
4141 		btrfs_err(fs_info, "Unable to find block group for %llu",
4142 			  start);
4143 		return -ENOSPC;
4144 	}
4145 
4146 	if (pin)
4147 		pin_down_extent(cache, start, len, 1);
4148 	else {
4149 		if (btrfs_test_opt(fs_info, DISCARD))
4150 			ret = btrfs_discard_extent(fs_info, start, len, NULL);
4151 		btrfs_add_free_space(cache, start, len);
4152 		btrfs_free_reserved_bytes(cache, len, delalloc);
4153 		trace_btrfs_reserved_extent_free(fs_info, start, len);
4154 	}
4155 
4156 	btrfs_put_block_group(cache);
4157 	return ret;
4158 }
4159 
btrfs_free_reserved_extent(struct btrfs_fs_info * fs_info,u64 start,u64 len,int delalloc)4160 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
4161 			       u64 start, u64 len, int delalloc)
4162 {
4163 	return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
4164 }
4165 
btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info * fs_info,u64 start,u64 len)4166 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
4167 				       u64 start, u64 len)
4168 {
4169 	return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
4170 }
4171 
alloc_reserved_file_extent(struct btrfs_trans_handle * trans,u64 parent,u64 root_objectid,u64 flags,u64 owner,u64 offset,struct btrfs_key * ins,int ref_mod)4172 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4173 				      u64 parent, u64 root_objectid,
4174 				      u64 flags, u64 owner, u64 offset,
4175 				      struct btrfs_key *ins, int ref_mod)
4176 {
4177 	struct btrfs_fs_info *fs_info = trans->fs_info;
4178 	int ret;
4179 	struct btrfs_extent_item *extent_item;
4180 	struct btrfs_extent_inline_ref *iref;
4181 	struct btrfs_path *path;
4182 	struct extent_buffer *leaf;
4183 	int type;
4184 	u32 size;
4185 
4186 	if (parent > 0)
4187 		type = BTRFS_SHARED_DATA_REF_KEY;
4188 	else
4189 		type = BTRFS_EXTENT_DATA_REF_KEY;
4190 
4191 	size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4192 
4193 	path = btrfs_alloc_path();
4194 	if (!path)
4195 		return -ENOMEM;
4196 
4197 	path->leave_spinning = 1;
4198 	ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4199 				      ins, size);
4200 	if (ret) {
4201 		btrfs_free_path(path);
4202 		return ret;
4203 	}
4204 
4205 	leaf = path->nodes[0];
4206 	extent_item = btrfs_item_ptr(leaf, path->slots[0],
4207 				     struct btrfs_extent_item);
4208 	btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4209 	btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4210 	btrfs_set_extent_flags(leaf, extent_item,
4211 			       flags | BTRFS_EXTENT_FLAG_DATA);
4212 
4213 	iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4214 	btrfs_set_extent_inline_ref_type(leaf, iref, type);
4215 	if (parent > 0) {
4216 		struct btrfs_shared_data_ref *ref;
4217 		ref = (struct btrfs_shared_data_ref *)(iref + 1);
4218 		btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4219 		btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4220 	} else {
4221 		struct btrfs_extent_data_ref *ref;
4222 		ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4223 		btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4224 		btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4225 		btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4226 		btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4227 	}
4228 
4229 	btrfs_mark_buffer_dirty(path->nodes[0]);
4230 	btrfs_free_path(path);
4231 
4232 	ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
4233 	if (ret)
4234 		return ret;
4235 
4236 	ret = btrfs_update_block_group(trans, ins->objectid, ins->offset, 1);
4237 	if (ret) { /* -ENOENT, logic error */
4238 		btrfs_err(fs_info, "update block group failed for %llu %llu",
4239 			ins->objectid, ins->offset);
4240 		BUG();
4241 	}
4242 	trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
4243 	return ret;
4244 }
4245 
alloc_reserved_tree_block(struct btrfs_trans_handle * trans,struct btrfs_delayed_ref_node * node,struct btrfs_delayed_extent_op * extent_op)4246 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4247 				     struct btrfs_delayed_ref_node *node,
4248 				     struct btrfs_delayed_extent_op *extent_op)
4249 {
4250 	struct btrfs_fs_info *fs_info = trans->fs_info;
4251 	int ret;
4252 	struct btrfs_extent_item *extent_item;
4253 	struct btrfs_key extent_key;
4254 	struct btrfs_tree_block_info *block_info;
4255 	struct btrfs_extent_inline_ref *iref;
4256 	struct btrfs_path *path;
4257 	struct extent_buffer *leaf;
4258 	struct btrfs_delayed_tree_ref *ref;
4259 	u32 size = sizeof(*extent_item) + sizeof(*iref);
4260 	u64 num_bytes;
4261 	u64 flags = extent_op->flags_to_set;
4262 	bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4263 
4264 	ref = btrfs_delayed_node_to_tree_ref(node);
4265 
4266 	extent_key.objectid = node->bytenr;
4267 	if (skinny_metadata) {
4268 		extent_key.offset = ref->level;
4269 		extent_key.type = BTRFS_METADATA_ITEM_KEY;
4270 		num_bytes = fs_info->nodesize;
4271 	} else {
4272 		extent_key.offset = node->num_bytes;
4273 		extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4274 		size += sizeof(*block_info);
4275 		num_bytes = node->num_bytes;
4276 	}
4277 
4278 	path = btrfs_alloc_path();
4279 	if (!path)
4280 		return -ENOMEM;
4281 
4282 	path->leave_spinning = 1;
4283 	ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4284 				      &extent_key, size);
4285 	if (ret) {
4286 		btrfs_free_path(path);
4287 		return ret;
4288 	}
4289 
4290 	leaf = path->nodes[0];
4291 	extent_item = btrfs_item_ptr(leaf, path->slots[0],
4292 				     struct btrfs_extent_item);
4293 	btrfs_set_extent_refs(leaf, extent_item, 1);
4294 	btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4295 	btrfs_set_extent_flags(leaf, extent_item,
4296 			       flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4297 
4298 	if (skinny_metadata) {
4299 		iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4300 	} else {
4301 		block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4302 		btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
4303 		btrfs_set_tree_block_level(leaf, block_info, ref->level);
4304 		iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4305 	}
4306 
4307 	if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
4308 		BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4309 		btrfs_set_extent_inline_ref_type(leaf, iref,
4310 						 BTRFS_SHARED_BLOCK_REF_KEY);
4311 		btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
4312 	} else {
4313 		btrfs_set_extent_inline_ref_type(leaf, iref,
4314 						 BTRFS_TREE_BLOCK_REF_KEY);
4315 		btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
4316 	}
4317 
4318 	btrfs_mark_buffer_dirty(leaf);
4319 	btrfs_free_path(path);
4320 
4321 	ret = remove_from_free_space_tree(trans, extent_key.objectid,
4322 					  num_bytes);
4323 	if (ret)
4324 		return ret;
4325 
4326 	ret = btrfs_update_block_group(trans, extent_key.objectid,
4327 				       fs_info->nodesize, 1);
4328 	if (ret) { /* -ENOENT, logic error */
4329 		btrfs_err(fs_info, "update block group failed for %llu %llu",
4330 			extent_key.objectid, extent_key.offset);
4331 		BUG();
4332 	}
4333 
4334 	trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
4335 					  fs_info->nodesize);
4336 	return ret;
4337 }
4338 
btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 owner,u64 offset,u64 ram_bytes,struct btrfs_key * ins)4339 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4340 				     struct btrfs_root *root, u64 owner,
4341 				     u64 offset, u64 ram_bytes,
4342 				     struct btrfs_key *ins)
4343 {
4344 	struct btrfs_ref generic_ref = { 0 };
4345 	int ret;
4346 
4347 	BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4348 
4349 	btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4350 			       ins->objectid, ins->offset, 0);
4351 	btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset);
4352 	btrfs_ref_tree_mod(root->fs_info, &generic_ref);
4353 	ret = btrfs_add_delayed_data_ref(trans, &generic_ref,
4354 					 ram_bytes, NULL, NULL);
4355 	return ret;
4356 }
4357 
4358 /*
4359  * this is used by the tree logging recovery code.  It records that
4360  * an extent has been allocated and makes sure to clear the free
4361  * space cache bits as well
4362  */
btrfs_alloc_logged_file_extent(struct btrfs_trans_handle * trans,u64 root_objectid,u64 owner,u64 offset,struct btrfs_key * ins)4363 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4364 				   u64 root_objectid, u64 owner, u64 offset,
4365 				   struct btrfs_key *ins)
4366 {
4367 	struct btrfs_fs_info *fs_info = trans->fs_info;
4368 	int ret;
4369 	struct btrfs_block_group_cache *block_group;
4370 	struct btrfs_space_info *space_info;
4371 
4372 	/*
4373 	 * Mixed block groups will exclude before processing the log so we only
4374 	 * need to do the exclude dance if this fs isn't mixed.
4375 	 */
4376 	if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
4377 		ret = __exclude_logged_extent(fs_info, ins->objectid,
4378 					      ins->offset);
4379 		if (ret)
4380 			return ret;
4381 	}
4382 
4383 	block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
4384 	if (!block_group)
4385 		return -EINVAL;
4386 
4387 	space_info = block_group->space_info;
4388 	spin_lock(&space_info->lock);
4389 	spin_lock(&block_group->lock);
4390 	space_info->bytes_reserved += ins->offset;
4391 	block_group->reserved += ins->offset;
4392 	spin_unlock(&block_group->lock);
4393 	spin_unlock(&space_info->lock);
4394 
4395 	ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
4396 					 offset, ins, 1);
4397 	btrfs_put_block_group(block_group);
4398 	return ret;
4399 }
4400 
4401 static struct extent_buffer *
btrfs_init_new_buffer(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 bytenr,int level,u64 owner)4402 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4403 		      u64 bytenr, int level, u64 owner)
4404 {
4405 	struct btrfs_fs_info *fs_info = root->fs_info;
4406 	struct extent_buffer *buf;
4407 
4408 	buf = btrfs_find_create_tree_block(fs_info, bytenr);
4409 	if (IS_ERR(buf))
4410 		return buf;
4411 
4412 	/*
4413 	 * Extra safety check in case the extent tree is corrupted and extent
4414 	 * allocator chooses to use a tree block which is already used and
4415 	 * locked.
4416 	 */
4417 	if (buf->lock_owner == current->pid) {
4418 		btrfs_err_rl(fs_info,
4419 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
4420 			buf->start, btrfs_header_owner(buf), current->pid);
4421 		free_extent_buffer(buf);
4422 		return ERR_PTR(-EUCLEAN);
4423 	}
4424 
4425 	btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
4426 	btrfs_tree_lock(buf);
4427 	btrfs_clean_tree_block(buf);
4428 	clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
4429 
4430 	btrfs_set_lock_blocking_write(buf);
4431 	set_extent_buffer_uptodate(buf);
4432 
4433 	memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
4434 	btrfs_set_header_level(buf, level);
4435 	btrfs_set_header_bytenr(buf, buf->start);
4436 	btrfs_set_header_generation(buf, trans->transid);
4437 	btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
4438 	btrfs_set_header_owner(buf, owner);
4439 	write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
4440 	write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
4441 	if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4442 		buf->log_index = root->log_transid % 2;
4443 		/*
4444 		 * we allow two log transactions at a time, use different
4445 		 * EXTENT bit to differentiate dirty pages.
4446 		 */
4447 		if (buf->log_index == 0)
4448 			set_extent_dirty(&root->dirty_log_pages, buf->start,
4449 					buf->start + buf->len - 1, GFP_NOFS);
4450 		else
4451 			set_extent_new(&root->dirty_log_pages, buf->start,
4452 					buf->start + buf->len - 1);
4453 	} else {
4454 		buf->log_index = -1;
4455 		set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4456 			 buf->start + buf->len - 1, GFP_NOFS);
4457 	}
4458 	trans->dirty = true;
4459 	/* this returns a buffer locked for blocking */
4460 	return buf;
4461 }
4462 
4463 /*
4464  * finds a free extent and does all the dirty work required for allocation
4465  * returns the tree buffer or an ERR_PTR on error.
4466  */
btrfs_alloc_tree_block(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 parent,u64 root_objectid,const struct btrfs_disk_key * key,int level,u64 hint,u64 empty_size)4467 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
4468 					     struct btrfs_root *root,
4469 					     u64 parent, u64 root_objectid,
4470 					     const struct btrfs_disk_key *key,
4471 					     int level, u64 hint,
4472 					     u64 empty_size)
4473 {
4474 	struct btrfs_fs_info *fs_info = root->fs_info;
4475 	struct btrfs_key ins;
4476 	struct btrfs_block_rsv *block_rsv;
4477 	struct extent_buffer *buf;
4478 	struct btrfs_delayed_extent_op *extent_op;
4479 	struct btrfs_ref generic_ref = { 0 };
4480 	u64 flags = 0;
4481 	int ret;
4482 	u32 blocksize = fs_info->nodesize;
4483 	bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4484 
4485 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4486 	if (btrfs_is_testing(fs_info)) {
4487 		buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
4488 					    level, root_objectid);
4489 		if (!IS_ERR(buf))
4490 			root->alloc_bytenr += blocksize;
4491 		return buf;
4492 	}
4493 #endif
4494 
4495 	block_rsv = btrfs_use_block_rsv(trans, root, blocksize);
4496 	if (IS_ERR(block_rsv))
4497 		return ERR_CAST(block_rsv);
4498 
4499 	ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
4500 				   empty_size, hint, &ins, 0, 0);
4501 	if (ret)
4502 		goto out_unuse;
4503 
4504 	buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
4505 				    root_objectid);
4506 	if (IS_ERR(buf)) {
4507 		ret = PTR_ERR(buf);
4508 		goto out_free_reserved;
4509 	}
4510 
4511 	if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4512 		if (parent == 0)
4513 			parent = ins.objectid;
4514 		flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4515 	} else
4516 		BUG_ON(parent > 0);
4517 
4518 	if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4519 		extent_op = btrfs_alloc_delayed_extent_op();
4520 		if (!extent_op) {
4521 			ret = -ENOMEM;
4522 			goto out_free_buf;
4523 		}
4524 		if (key)
4525 			memcpy(&extent_op->key, key, sizeof(extent_op->key));
4526 		else
4527 			memset(&extent_op->key, 0, sizeof(extent_op->key));
4528 		extent_op->flags_to_set = flags;
4529 		extent_op->update_key = skinny_metadata ? false : true;
4530 		extent_op->update_flags = true;
4531 		extent_op->is_data = false;
4532 		extent_op->level = level;
4533 
4534 		btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4535 				       ins.objectid, ins.offset, parent);
4536 		generic_ref.real_root = root->root_key.objectid;
4537 		btrfs_init_tree_ref(&generic_ref, level, root_objectid);
4538 		btrfs_ref_tree_mod(fs_info, &generic_ref);
4539 		ret = btrfs_add_delayed_tree_ref(trans, &generic_ref,
4540 						 extent_op, NULL, NULL);
4541 		if (ret)
4542 			goto out_free_delayed;
4543 	}
4544 	return buf;
4545 
4546 out_free_delayed:
4547 	btrfs_free_delayed_extent_op(extent_op);
4548 out_free_buf:
4549 	free_extent_buffer(buf);
4550 out_free_reserved:
4551 	btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
4552 out_unuse:
4553 	btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
4554 	return ERR_PTR(ret);
4555 }
4556 
4557 struct walk_control {
4558 	u64 refs[BTRFS_MAX_LEVEL];
4559 	u64 flags[BTRFS_MAX_LEVEL];
4560 	struct btrfs_key update_progress;
4561 	struct btrfs_key drop_progress;
4562 	int drop_level;
4563 	int stage;
4564 	int level;
4565 	int shared_level;
4566 	int update_ref;
4567 	int keep_locks;
4568 	int reada_slot;
4569 	int reada_count;
4570 	int restarted;
4571 };
4572 
4573 #define DROP_REFERENCE	1
4574 #define UPDATE_BACKREF	2
4575 
reada_walk_down(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct walk_control * wc,struct btrfs_path * path)4576 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4577 				     struct btrfs_root *root,
4578 				     struct walk_control *wc,
4579 				     struct btrfs_path *path)
4580 {
4581 	struct btrfs_fs_info *fs_info = root->fs_info;
4582 	u64 bytenr;
4583 	u64 generation;
4584 	u64 refs;
4585 	u64 flags;
4586 	u32 nritems;
4587 	struct btrfs_key key;
4588 	struct extent_buffer *eb;
4589 	int ret;
4590 	int slot;
4591 	int nread = 0;
4592 
4593 	if (path->slots[wc->level] < wc->reada_slot) {
4594 		wc->reada_count = wc->reada_count * 2 / 3;
4595 		wc->reada_count = max(wc->reada_count, 2);
4596 	} else {
4597 		wc->reada_count = wc->reada_count * 3 / 2;
4598 		wc->reada_count = min_t(int, wc->reada_count,
4599 					BTRFS_NODEPTRS_PER_BLOCK(fs_info));
4600 	}
4601 
4602 	eb = path->nodes[wc->level];
4603 	nritems = btrfs_header_nritems(eb);
4604 
4605 	for (slot = path->slots[wc->level]; slot < nritems; slot++) {
4606 		if (nread >= wc->reada_count)
4607 			break;
4608 
4609 		cond_resched();
4610 		bytenr = btrfs_node_blockptr(eb, slot);
4611 		generation = btrfs_node_ptr_generation(eb, slot);
4612 
4613 		if (slot == path->slots[wc->level])
4614 			goto reada;
4615 
4616 		if (wc->stage == UPDATE_BACKREF &&
4617 		    generation <= root->root_key.offset)
4618 			continue;
4619 
4620 		/* We don't lock the tree block, it's OK to be racy here */
4621 		ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
4622 					       wc->level - 1, 1, &refs,
4623 					       &flags);
4624 		/* We don't care about errors in readahead. */
4625 		if (ret < 0)
4626 			continue;
4627 		BUG_ON(refs == 0);
4628 
4629 		if (wc->stage == DROP_REFERENCE) {
4630 			if (refs == 1)
4631 				goto reada;
4632 
4633 			if (wc->level == 1 &&
4634 			    (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4635 				continue;
4636 			if (!wc->update_ref ||
4637 			    generation <= root->root_key.offset)
4638 				continue;
4639 			btrfs_node_key_to_cpu(eb, &key, slot);
4640 			ret = btrfs_comp_cpu_keys(&key,
4641 						  &wc->update_progress);
4642 			if (ret < 0)
4643 				continue;
4644 		} else {
4645 			if (wc->level == 1 &&
4646 			    (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4647 				continue;
4648 		}
4649 reada:
4650 		readahead_tree_block(fs_info, bytenr);
4651 		nread++;
4652 	}
4653 	wc->reada_slot = slot;
4654 }
4655 
4656 /*
4657  * helper to process tree block while walking down the tree.
4658  *
4659  * when wc->stage == UPDATE_BACKREF, this function updates
4660  * back refs for pointers in the block.
4661  *
4662  * NOTE: return value 1 means we should stop walking down.
4663  */
walk_down_proc(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct walk_control * wc,int lookup_info)4664 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4665 				   struct btrfs_root *root,
4666 				   struct btrfs_path *path,
4667 				   struct walk_control *wc, int lookup_info)
4668 {
4669 	struct btrfs_fs_info *fs_info = root->fs_info;
4670 	int level = wc->level;
4671 	struct extent_buffer *eb = path->nodes[level];
4672 	u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4673 	int ret;
4674 
4675 	if (wc->stage == UPDATE_BACKREF &&
4676 	    btrfs_header_owner(eb) != root->root_key.objectid)
4677 		return 1;
4678 
4679 	/*
4680 	 * when reference count of tree block is 1, it won't increase
4681 	 * again. once full backref flag is set, we never clear it.
4682 	 */
4683 	if (lookup_info &&
4684 	    ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
4685 	     (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
4686 		BUG_ON(!path->locks[level]);
4687 		ret = btrfs_lookup_extent_info(trans, fs_info,
4688 					       eb->start, level, 1,
4689 					       &wc->refs[level],
4690 					       &wc->flags[level]);
4691 		BUG_ON(ret == -ENOMEM);
4692 		if (ret)
4693 			return ret;
4694 		BUG_ON(wc->refs[level] == 0);
4695 	}
4696 
4697 	if (wc->stage == DROP_REFERENCE) {
4698 		if (wc->refs[level] > 1)
4699 			return 1;
4700 
4701 		if (path->locks[level] && !wc->keep_locks) {
4702 			btrfs_tree_unlock_rw(eb, path->locks[level]);
4703 			path->locks[level] = 0;
4704 		}
4705 		return 0;
4706 	}
4707 
4708 	/* wc->stage == UPDATE_BACKREF */
4709 	if (!(wc->flags[level] & flag)) {
4710 		BUG_ON(!path->locks[level]);
4711 		ret = btrfs_inc_ref(trans, root, eb, 1);
4712 		BUG_ON(ret); /* -ENOMEM */
4713 		ret = btrfs_dec_ref(trans, root, eb, 0);
4714 		BUG_ON(ret); /* -ENOMEM */
4715 		ret = btrfs_set_disk_extent_flags(trans, eb->start,
4716 						  eb->len, flag,
4717 						  btrfs_header_level(eb), 0);
4718 		BUG_ON(ret); /* -ENOMEM */
4719 		wc->flags[level] |= flag;
4720 	}
4721 
4722 	/*
4723 	 * the block is shared by multiple trees, so it's not good to
4724 	 * keep the tree lock
4725 	 */
4726 	if (path->locks[level] && level > 0) {
4727 		btrfs_tree_unlock_rw(eb, path->locks[level]);
4728 		path->locks[level] = 0;
4729 	}
4730 	return 0;
4731 }
4732 
4733 /*
4734  * This is used to verify a ref exists for this root to deal with a bug where we
4735  * would have a drop_progress key that hadn't been updated properly.
4736  */
check_ref_exists(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 bytenr,u64 parent,int level)4737 static int check_ref_exists(struct btrfs_trans_handle *trans,
4738 			    struct btrfs_root *root, u64 bytenr, u64 parent,
4739 			    int level)
4740 {
4741 	struct btrfs_path *path;
4742 	struct btrfs_extent_inline_ref *iref;
4743 	int ret;
4744 
4745 	path = btrfs_alloc_path();
4746 	if (!path)
4747 		return -ENOMEM;
4748 
4749 	ret = lookup_extent_backref(trans, path, &iref, bytenr,
4750 				    root->fs_info->nodesize, parent,
4751 				    root->root_key.objectid, level, 0);
4752 	btrfs_free_path(path);
4753 	if (ret == -ENOENT)
4754 		return 0;
4755 	if (ret < 0)
4756 		return ret;
4757 	return 1;
4758 }
4759 
4760 /*
4761  * helper to process tree block pointer.
4762  *
4763  * when wc->stage == DROP_REFERENCE, this function checks
4764  * reference count of the block pointed to. if the block
4765  * is shared and we need update back refs for the subtree
4766  * rooted at the block, this function changes wc->stage to
4767  * UPDATE_BACKREF. if the block is shared and there is no
4768  * need to update back, this function drops the reference
4769  * to the block.
4770  *
4771  * NOTE: return value 1 means we should stop walking down.
4772  */
do_walk_down(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct walk_control * wc,int * lookup_info)4773 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
4774 				 struct btrfs_root *root,
4775 				 struct btrfs_path *path,
4776 				 struct walk_control *wc, int *lookup_info)
4777 {
4778 	struct btrfs_fs_info *fs_info = root->fs_info;
4779 	u64 bytenr;
4780 	u64 generation;
4781 	u64 parent;
4782 	struct btrfs_key key;
4783 	struct btrfs_key first_key;
4784 	struct btrfs_ref ref = { 0 };
4785 	struct extent_buffer *next;
4786 	int level = wc->level;
4787 	int reada = 0;
4788 	int ret = 0;
4789 	bool need_account = false;
4790 
4791 	generation = btrfs_node_ptr_generation(path->nodes[level],
4792 					       path->slots[level]);
4793 	/*
4794 	 * if the lower level block was created before the snapshot
4795 	 * was created, we know there is no need to update back refs
4796 	 * for the subtree
4797 	 */
4798 	if (wc->stage == UPDATE_BACKREF &&
4799 	    generation <= root->root_key.offset) {
4800 		*lookup_info = 1;
4801 		return 1;
4802 	}
4803 
4804 	bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
4805 	btrfs_node_key_to_cpu(path->nodes[level], &first_key,
4806 			      path->slots[level]);
4807 
4808 	next = find_extent_buffer(fs_info, bytenr);
4809 	if (!next) {
4810 		next = btrfs_find_create_tree_block(fs_info, bytenr);
4811 		if (IS_ERR(next))
4812 			return PTR_ERR(next);
4813 
4814 		btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
4815 					       level - 1);
4816 		reada = 1;
4817 	}
4818 	btrfs_tree_lock(next);
4819 	btrfs_set_lock_blocking_write(next);
4820 
4821 	ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
4822 				       &wc->refs[level - 1],
4823 				       &wc->flags[level - 1]);
4824 	if (ret < 0)
4825 		goto out_unlock;
4826 
4827 	if (unlikely(wc->refs[level - 1] == 0)) {
4828 		btrfs_err(fs_info, "Missing references.");
4829 		ret = -EIO;
4830 		goto out_unlock;
4831 	}
4832 	*lookup_info = 0;
4833 
4834 	if (wc->stage == DROP_REFERENCE) {
4835 		if (wc->refs[level - 1] > 1) {
4836 			need_account = true;
4837 			if (level == 1 &&
4838 			    (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4839 				goto skip;
4840 
4841 			if (!wc->update_ref ||
4842 			    generation <= root->root_key.offset)
4843 				goto skip;
4844 
4845 			btrfs_node_key_to_cpu(path->nodes[level], &key,
4846 					      path->slots[level]);
4847 			ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
4848 			if (ret < 0)
4849 				goto skip;
4850 
4851 			wc->stage = UPDATE_BACKREF;
4852 			wc->shared_level = level - 1;
4853 		}
4854 	} else {
4855 		if (level == 1 &&
4856 		    (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4857 			goto skip;
4858 	}
4859 
4860 	if (!btrfs_buffer_uptodate(next, generation, 0)) {
4861 		btrfs_tree_unlock(next);
4862 		free_extent_buffer(next);
4863 		next = NULL;
4864 		*lookup_info = 1;
4865 	}
4866 
4867 	if (!next) {
4868 		if (reada && level == 1)
4869 			reada_walk_down(trans, root, wc, path);
4870 		next = read_tree_block(fs_info, bytenr, generation, level - 1,
4871 				       &first_key);
4872 		if (IS_ERR(next)) {
4873 			return PTR_ERR(next);
4874 		} else if (!extent_buffer_uptodate(next)) {
4875 			free_extent_buffer(next);
4876 			return -EIO;
4877 		}
4878 		btrfs_tree_lock(next);
4879 		btrfs_set_lock_blocking_write(next);
4880 	}
4881 
4882 	level--;
4883 	ASSERT(level == btrfs_header_level(next));
4884 	if (level != btrfs_header_level(next)) {
4885 		btrfs_err(root->fs_info, "mismatched level");
4886 		ret = -EIO;
4887 		goto out_unlock;
4888 	}
4889 	path->nodes[level] = next;
4890 	path->slots[level] = 0;
4891 	path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
4892 	wc->level = level;
4893 	if (wc->level == 1)
4894 		wc->reada_slot = 0;
4895 	return 0;
4896 skip:
4897 	wc->refs[level - 1] = 0;
4898 	wc->flags[level - 1] = 0;
4899 	if (wc->stage == DROP_REFERENCE) {
4900 		if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
4901 			parent = path->nodes[level]->start;
4902 		} else {
4903 			ASSERT(root->root_key.objectid ==
4904 			       btrfs_header_owner(path->nodes[level]));
4905 			if (root->root_key.objectid !=
4906 			    btrfs_header_owner(path->nodes[level])) {
4907 				btrfs_err(root->fs_info,
4908 						"mismatched block owner");
4909 				ret = -EIO;
4910 				goto out_unlock;
4911 			}
4912 			parent = 0;
4913 		}
4914 
4915 		/*
4916 		 * If we had a drop_progress we need to verify the refs are set
4917 		 * as expected.  If we find our ref then we know that from here
4918 		 * on out everything should be correct, and we can clear the
4919 		 * ->restarted flag.
4920 		 */
4921 		if (wc->restarted) {
4922 			ret = check_ref_exists(trans, root, bytenr, parent,
4923 					       level - 1);
4924 			if (ret < 0)
4925 				goto out_unlock;
4926 			if (ret == 0)
4927 				goto no_delete;
4928 			ret = 0;
4929 			wc->restarted = 0;
4930 		}
4931 
4932 		/*
4933 		 * Reloc tree doesn't contribute to qgroup numbers, and we have
4934 		 * already accounted them at merge time (replace_path),
4935 		 * thus we could skip expensive subtree trace here.
4936 		 */
4937 		if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
4938 		    need_account) {
4939 			ret = btrfs_qgroup_trace_subtree(trans, next,
4940 							 generation, level - 1);
4941 			if (ret) {
4942 				btrfs_err_rl(fs_info,
4943 					     "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
4944 					     ret);
4945 			}
4946 		}
4947 
4948 		/*
4949 		 * We need to update the next key in our walk control so we can
4950 		 * update the drop_progress key accordingly.  We don't care if
4951 		 * find_next_key doesn't find a key because that means we're at
4952 		 * the end and are going to clean up now.
4953 		 */
4954 		wc->drop_level = level;
4955 		find_next_key(path, level, &wc->drop_progress);
4956 
4957 		btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
4958 				       fs_info->nodesize, parent);
4959 		btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid);
4960 		ret = btrfs_free_extent(trans, &ref);
4961 		if (ret)
4962 			goto out_unlock;
4963 	}
4964 no_delete:
4965 	*lookup_info = 1;
4966 	ret = 1;
4967 
4968 out_unlock:
4969 	btrfs_tree_unlock(next);
4970 	free_extent_buffer(next);
4971 
4972 	return ret;
4973 }
4974 
4975 /*
4976  * helper to process tree block while walking up the tree.
4977  *
4978  * when wc->stage == DROP_REFERENCE, this function drops
4979  * reference count on the block.
4980  *
4981  * when wc->stage == UPDATE_BACKREF, this function changes
4982  * wc->stage back to DROP_REFERENCE if we changed wc->stage
4983  * to UPDATE_BACKREF previously while processing the block.
4984  *
4985  * NOTE: return value 1 means we should stop walking up.
4986  */
walk_up_proc(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct walk_control * wc)4987 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
4988 				 struct btrfs_root *root,
4989 				 struct btrfs_path *path,
4990 				 struct walk_control *wc)
4991 {
4992 	struct btrfs_fs_info *fs_info = root->fs_info;
4993 	int ret;
4994 	int level = wc->level;
4995 	struct extent_buffer *eb = path->nodes[level];
4996 	u64 parent = 0;
4997 
4998 	if (wc->stage == UPDATE_BACKREF) {
4999 		BUG_ON(wc->shared_level < level);
5000 		if (level < wc->shared_level)
5001 			goto out;
5002 
5003 		ret = find_next_key(path, level + 1, &wc->update_progress);
5004 		if (ret > 0)
5005 			wc->update_ref = 0;
5006 
5007 		wc->stage = DROP_REFERENCE;
5008 		wc->shared_level = -1;
5009 		path->slots[level] = 0;
5010 
5011 		/*
5012 		 * check reference count again if the block isn't locked.
5013 		 * we should start walking down the tree again if reference
5014 		 * count is one.
5015 		 */
5016 		if (!path->locks[level]) {
5017 			BUG_ON(level == 0);
5018 			btrfs_tree_lock(eb);
5019 			btrfs_set_lock_blocking_write(eb);
5020 			path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
5021 
5022 			ret = btrfs_lookup_extent_info(trans, fs_info,
5023 						       eb->start, level, 1,
5024 						       &wc->refs[level],
5025 						       &wc->flags[level]);
5026 			if (ret < 0) {
5027 				btrfs_tree_unlock_rw(eb, path->locks[level]);
5028 				path->locks[level] = 0;
5029 				return ret;
5030 			}
5031 			BUG_ON(wc->refs[level] == 0);
5032 			if (wc->refs[level] == 1) {
5033 				btrfs_tree_unlock_rw(eb, path->locks[level]);
5034 				path->locks[level] = 0;
5035 				return 1;
5036 			}
5037 		}
5038 	}
5039 
5040 	/* wc->stage == DROP_REFERENCE */
5041 	BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5042 
5043 	if (wc->refs[level] == 1) {
5044 		if (level == 0) {
5045 			if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5046 				ret = btrfs_dec_ref(trans, root, eb, 1);
5047 			else
5048 				ret = btrfs_dec_ref(trans, root, eb, 0);
5049 			BUG_ON(ret); /* -ENOMEM */
5050 			if (is_fstree(root->root_key.objectid)) {
5051 				ret = btrfs_qgroup_trace_leaf_items(trans, eb);
5052 				if (ret) {
5053 					btrfs_err_rl(fs_info,
5054 	"error %d accounting leaf items, quota is out of sync, rescan required",
5055 					     ret);
5056 				}
5057 			}
5058 		}
5059 		/* make block locked assertion in btrfs_clean_tree_block happy */
5060 		if (!path->locks[level] &&
5061 		    btrfs_header_generation(eb) == trans->transid) {
5062 			btrfs_tree_lock(eb);
5063 			btrfs_set_lock_blocking_write(eb);
5064 			path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
5065 		}
5066 		btrfs_clean_tree_block(eb);
5067 	}
5068 
5069 	if (eb == root->node) {
5070 		if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5071 			parent = eb->start;
5072 		else if (root->root_key.objectid != btrfs_header_owner(eb))
5073 			goto owner_mismatch;
5074 	} else {
5075 		if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5076 			parent = path->nodes[level + 1]->start;
5077 		else if (root->root_key.objectid !=
5078 			 btrfs_header_owner(path->nodes[level + 1]))
5079 			goto owner_mismatch;
5080 	}
5081 
5082 	btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
5083 out:
5084 	wc->refs[level] = 0;
5085 	wc->flags[level] = 0;
5086 	return 0;
5087 
5088 owner_mismatch:
5089 	btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
5090 		     btrfs_header_owner(eb), root->root_key.objectid);
5091 	return -EUCLEAN;
5092 }
5093 
walk_down_tree(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct walk_control * wc)5094 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5095 				   struct btrfs_root *root,
5096 				   struct btrfs_path *path,
5097 				   struct walk_control *wc)
5098 {
5099 	int level = wc->level;
5100 	int lookup_info = 1;
5101 	int ret;
5102 
5103 	while (level >= 0) {
5104 		ret = walk_down_proc(trans, root, path, wc, lookup_info);
5105 		if (ret > 0)
5106 			break;
5107 
5108 		if (level == 0)
5109 			break;
5110 
5111 		if (path->slots[level] >=
5112 		    btrfs_header_nritems(path->nodes[level]))
5113 			break;
5114 
5115 		ret = do_walk_down(trans, root, path, wc, &lookup_info);
5116 		if (ret > 0) {
5117 			path->slots[level]++;
5118 			continue;
5119 		} else if (ret < 0)
5120 			return ret;
5121 		level = wc->level;
5122 	}
5123 	return 0;
5124 }
5125 
walk_up_tree(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct walk_control * wc,int max_level)5126 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5127 				 struct btrfs_root *root,
5128 				 struct btrfs_path *path,
5129 				 struct walk_control *wc, int max_level)
5130 {
5131 	int level = wc->level;
5132 	int ret;
5133 
5134 	path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5135 	while (level < max_level && path->nodes[level]) {
5136 		wc->level = level;
5137 		if (path->slots[level] + 1 <
5138 		    btrfs_header_nritems(path->nodes[level])) {
5139 			path->slots[level]++;
5140 			return 0;
5141 		} else {
5142 			ret = walk_up_proc(trans, root, path, wc);
5143 			if (ret > 0)
5144 				return 0;
5145 			if (ret < 0)
5146 				return ret;
5147 
5148 			if (path->locks[level]) {
5149 				btrfs_tree_unlock_rw(path->nodes[level],
5150 						     path->locks[level]);
5151 				path->locks[level] = 0;
5152 			}
5153 			free_extent_buffer(path->nodes[level]);
5154 			path->nodes[level] = NULL;
5155 			level++;
5156 		}
5157 	}
5158 	return 1;
5159 }
5160 
5161 /*
5162  * drop a subvolume tree.
5163  *
5164  * this function traverses the tree freeing any blocks that only
5165  * referenced by the tree.
5166  *
5167  * when a shared tree block is found. this function decreases its
5168  * reference count by one. if update_ref is true, this function
5169  * also make sure backrefs for the shared block and all lower level
5170  * blocks are properly updated.
5171  *
5172  * If called with for_reloc == 0, may exit early with -EAGAIN
5173  */
btrfs_drop_snapshot(struct btrfs_root * root,struct btrfs_block_rsv * block_rsv,int update_ref,int for_reloc)5174 int btrfs_drop_snapshot(struct btrfs_root *root,
5175 			 struct btrfs_block_rsv *block_rsv, int update_ref,
5176 			 int for_reloc)
5177 {
5178 	struct btrfs_fs_info *fs_info = root->fs_info;
5179 	struct btrfs_path *path;
5180 	struct btrfs_trans_handle *trans;
5181 	struct btrfs_root *tree_root = fs_info->tree_root;
5182 	struct btrfs_root_item *root_item = &root->root_item;
5183 	struct walk_control *wc;
5184 	struct btrfs_key key;
5185 	int err = 0;
5186 	int ret;
5187 	int level;
5188 	bool root_dropped = false;
5189 
5190 	btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
5191 
5192 	path = btrfs_alloc_path();
5193 	if (!path) {
5194 		err = -ENOMEM;
5195 		goto out;
5196 	}
5197 
5198 	wc = kzalloc(sizeof(*wc), GFP_NOFS);
5199 	if (!wc) {
5200 		btrfs_free_path(path);
5201 		err = -ENOMEM;
5202 		goto out;
5203 	}
5204 
5205 	trans = btrfs_start_transaction(tree_root, 0);
5206 	if (IS_ERR(trans)) {
5207 		err = PTR_ERR(trans);
5208 		goto out_free;
5209 	}
5210 
5211 	err = btrfs_run_delayed_items(trans);
5212 	if (err)
5213 		goto out_end_trans;
5214 
5215 	if (block_rsv)
5216 		trans->block_rsv = block_rsv;
5217 
5218 	/*
5219 	 * This will help us catch people modifying the fs tree while we're
5220 	 * dropping it.  It is unsafe to mess with the fs tree while it's being
5221 	 * dropped as we unlock the root node and parent nodes as we walk down
5222 	 * the tree, assuming nothing will change.  If something does change
5223 	 * then we'll have stale information and drop references to blocks we've
5224 	 * already dropped.
5225 	 */
5226 	set_bit(BTRFS_ROOT_DELETING, &root->state);
5227 	if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5228 		level = btrfs_header_level(root->node);
5229 		path->nodes[level] = btrfs_lock_root_node(root);
5230 		btrfs_set_lock_blocking_write(path->nodes[level]);
5231 		path->slots[level] = 0;
5232 		path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
5233 		memset(&wc->update_progress, 0,
5234 		       sizeof(wc->update_progress));
5235 	} else {
5236 		btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5237 		memcpy(&wc->update_progress, &key,
5238 		       sizeof(wc->update_progress));
5239 
5240 		level = root_item->drop_level;
5241 		BUG_ON(level == 0);
5242 		path->lowest_level = level;
5243 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5244 		path->lowest_level = 0;
5245 		if (ret < 0) {
5246 			err = ret;
5247 			goto out_end_trans;
5248 		}
5249 		WARN_ON(ret > 0);
5250 
5251 		/*
5252 		 * unlock our path, this is safe because only this
5253 		 * function is allowed to delete this snapshot
5254 		 */
5255 		btrfs_unlock_up_safe(path, 0);
5256 
5257 		level = btrfs_header_level(root->node);
5258 		while (1) {
5259 			btrfs_tree_lock(path->nodes[level]);
5260 			btrfs_set_lock_blocking_write(path->nodes[level]);
5261 			path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
5262 
5263 			ret = btrfs_lookup_extent_info(trans, fs_info,
5264 						path->nodes[level]->start,
5265 						level, 1, &wc->refs[level],
5266 						&wc->flags[level]);
5267 			if (ret < 0) {
5268 				err = ret;
5269 				goto out_end_trans;
5270 			}
5271 			BUG_ON(wc->refs[level] == 0);
5272 
5273 			if (level == root_item->drop_level)
5274 				break;
5275 
5276 			btrfs_tree_unlock(path->nodes[level]);
5277 			path->locks[level] = 0;
5278 			WARN_ON(wc->refs[level] != 1);
5279 			level--;
5280 		}
5281 	}
5282 
5283 	wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
5284 	wc->level = level;
5285 	wc->shared_level = -1;
5286 	wc->stage = DROP_REFERENCE;
5287 	wc->update_ref = update_ref;
5288 	wc->keep_locks = 0;
5289 	wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5290 
5291 	while (1) {
5292 
5293 		ret = walk_down_tree(trans, root, path, wc);
5294 		if (ret < 0) {
5295 			err = ret;
5296 			break;
5297 		}
5298 
5299 		ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5300 		if (ret < 0) {
5301 			err = ret;
5302 			break;
5303 		}
5304 
5305 		if (ret > 0) {
5306 			BUG_ON(wc->stage != DROP_REFERENCE);
5307 			break;
5308 		}
5309 
5310 		if (wc->stage == DROP_REFERENCE) {
5311 			wc->drop_level = wc->level;
5312 			btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
5313 					      &wc->drop_progress,
5314 					      path->slots[wc->drop_level]);
5315 		}
5316 		btrfs_cpu_key_to_disk(&root_item->drop_progress,
5317 				      &wc->drop_progress);
5318 		root_item->drop_level = wc->drop_level;
5319 
5320 		BUG_ON(wc->level == 0);
5321 		if (btrfs_should_end_transaction(trans) ||
5322 		    (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
5323 			ret = btrfs_update_root(trans, tree_root,
5324 						&root->root_key,
5325 						root_item);
5326 			if (ret) {
5327 				btrfs_abort_transaction(trans, ret);
5328 				err = ret;
5329 				goto out_end_trans;
5330 			}
5331 
5332 			btrfs_end_transaction_throttle(trans);
5333 			if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
5334 				btrfs_debug(fs_info,
5335 					    "drop snapshot early exit");
5336 				err = -EAGAIN;
5337 				goto out_free;
5338 			}
5339 
5340 			trans = btrfs_start_transaction(tree_root, 0);
5341 			if (IS_ERR(trans)) {
5342 				err = PTR_ERR(trans);
5343 				goto out_free;
5344 			}
5345 			if (block_rsv)
5346 				trans->block_rsv = block_rsv;
5347 		}
5348 	}
5349 	btrfs_release_path(path);
5350 	if (err)
5351 		goto out_end_trans;
5352 
5353 	ret = btrfs_del_root(trans, &root->root_key);
5354 	if (ret) {
5355 		btrfs_abort_transaction(trans, ret);
5356 		err = ret;
5357 		goto out_end_trans;
5358 	}
5359 
5360 	if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5361 		ret = btrfs_find_root(tree_root, &root->root_key, path,
5362 				      NULL, NULL);
5363 		if (ret < 0) {
5364 			btrfs_abort_transaction(trans, ret);
5365 			err = ret;
5366 			goto out_end_trans;
5367 		} else if (ret > 0) {
5368 			/* if we fail to delete the orphan item this time
5369 			 * around, it'll get picked up the next time.
5370 			 *
5371 			 * The most common failure here is just -ENOENT.
5372 			 */
5373 			btrfs_del_orphan_item(trans, tree_root,
5374 					      root->root_key.objectid);
5375 		}
5376 	}
5377 
5378 	if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
5379 		btrfs_add_dropped_root(trans, root);
5380 	} else {
5381 		free_extent_buffer(root->node);
5382 		free_extent_buffer(root->commit_root);
5383 		btrfs_put_fs_root(root);
5384 	}
5385 	root_dropped = true;
5386 out_end_trans:
5387 	btrfs_end_transaction_throttle(trans);
5388 out_free:
5389 	kfree(wc);
5390 	btrfs_free_path(path);
5391 out:
5392 	/*
5393 	 * So if we need to stop dropping the snapshot for whatever reason we
5394 	 * need to make sure to add it back to the dead root list so that we
5395 	 * keep trying to do the work later.  This also cleans up roots if we
5396 	 * don't have it in the radix (like when we recover after a power fail
5397 	 * or unmount) so we don't leak memory.
5398 	 */
5399 	if (!for_reloc && !root_dropped)
5400 		btrfs_add_dead_root(root);
5401 	if (err && err != -EAGAIN)
5402 		btrfs_handle_fs_error(fs_info, err, NULL);
5403 	return err;
5404 }
5405 
5406 /*
5407  * drop subtree rooted at tree block 'node'.
5408  *
5409  * NOTE: this function will unlock and release tree block 'node'
5410  * only used by relocation code
5411  */
btrfs_drop_subtree(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * node,struct extent_buffer * parent)5412 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5413 			struct btrfs_root *root,
5414 			struct extent_buffer *node,
5415 			struct extent_buffer *parent)
5416 {
5417 	struct btrfs_fs_info *fs_info = root->fs_info;
5418 	struct btrfs_path *path;
5419 	struct walk_control *wc;
5420 	int level;
5421 	int parent_level;
5422 	int ret = 0;
5423 	int wret;
5424 
5425 	BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5426 
5427 	path = btrfs_alloc_path();
5428 	if (!path)
5429 		return -ENOMEM;
5430 
5431 	wc = kzalloc(sizeof(*wc), GFP_NOFS);
5432 	if (!wc) {
5433 		btrfs_free_path(path);
5434 		return -ENOMEM;
5435 	}
5436 
5437 	btrfs_assert_tree_locked(parent);
5438 	parent_level = btrfs_header_level(parent);
5439 	extent_buffer_get(parent);
5440 	path->nodes[parent_level] = parent;
5441 	path->slots[parent_level] = btrfs_header_nritems(parent);
5442 
5443 	btrfs_assert_tree_locked(node);
5444 	level = btrfs_header_level(node);
5445 	path->nodes[level] = node;
5446 	path->slots[level] = 0;
5447 	path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
5448 
5449 	wc->refs[parent_level] = 1;
5450 	wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5451 	wc->level = level;
5452 	wc->shared_level = -1;
5453 	wc->stage = DROP_REFERENCE;
5454 	wc->update_ref = 0;
5455 	wc->keep_locks = 1;
5456 	wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5457 
5458 	while (1) {
5459 		wret = walk_down_tree(trans, root, path, wc);
5460 		if (wret < 0) {
5461 			ret = wret;
5462 			break;
5463 		}
5464 
5465 		wret = walk_up_tree(trans, root, path, wc, parent_level);
5466 		if (wret < 0)
5467 			ret = wret;
5468 		if (wret != 0)
5469 			break;
5470 	}
5471 
5472 	kfree(wc);
5473 	btrfs_free_path(path);
5474 	return ret;
5475 }
5476 
5477 /*
5478  * helper to account the unused space of all the readonly block group in the
5479  * space_info. takes mirrors into account.
5480  */
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info * sinfo)5481 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
5482 {
5483 	struct btrfs_block_group_cache *block_group;
5484 	u64 free_bytes = 0;
5485 	int factor;
5486 
5487 	/* It's df, we don't care if it's racy */
5488 	if (list_empty(&sinfo->ro_bgs))
5489 		return 0;
5490 
5491 	spin_lock(&sinfo->lock);
5492 	list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
5493 		spin_lock(&block_group->lock);
5494 
5495 		if (!block_group->ro) {
5496 			spin_unlock(&block_group->lock);
5497 			continue;
5498 		}
5499 
5500 		factor = btrfs_bg_type_to_factor(block_group->flags);
5501 		free_bytes += (block_group->key.offset -
5502 			       btrfs_block_group_used(&block_group->item)) *
5503 			       factor;
5504 
5505 		spin_unlock(&block_group->lock);
5506 	}
5507 	spin_unlock(&sinfo->lock);
5508 
5509 	return free_bytes;
5510 }
5511 
btrfs_error_unpin_extent_range(struct btrfs_fs_info * fs_info,u64 start,u64 end)5512 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
5513 				   u64 start, u64 end)
5514 {
5515 	return unpin_extent_range(fs_info, start, end, false);
5516 }
5517 
5518 /*
5519  * It used to be that old block groups would be left around forever.
5520  * Iterating over them would be enough to trim unused space.  Since we
5521  * now automatically remove them, we also need to iterate over unallocated
5522  * space.
5523  *
5524  * We don't want a transaction for this since the discard may take a
5525  * substantial amount of time.  We don't require that a transaction be
5526  * running, but we do need to take a running transaction into account
5527  * to ensure that we're not discarding chunks that were released or
5528  * allocated in the current transaction.
5529  *
5530  * Holding the chunks lock will prevent other threads from allocating
5531  * or releasing chunks, but it won't prevent a running transaction
5532  * from committing and releasing the memory that the pending chunks
5533  * list head uses.  For that, we need to take a reference to the
5534  * transaction and hold the commit root sem.  We only need to hold
5535  * it while performing the free space search since we have already
5536  * held back allocations.
5537  */
btrfs_trim_free_extents(struct btrfs_device * device,u64 * trimmed)5538 static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed)
5539 {
5540 	u64 start = SZ_1M, len = 0, end = 0;
5541 	int ret;
5542 
5543 	*trimmed = 0;
5544 
5545 	/* Discard not supported = nothing to do. */
5546 	if (!blk_queue_discard(bdev_get_queue(device->bdev)))
5547 		return 0;
5548 
5549 	/* Not writable = nothing to do. */
5550 	if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
5551 		return 0;
5552 
5553 	/* No free space = nothing to do. */
5554 	if (device->total_bytes <= device->bytes_used)
5555 		return 0;
5556 
5557 	ret = 0;
5558 
5559 	while (1) {
5560 		struct btrfs_fs_info *fs_info = device->fs_info;
5561 		u64 bytes;
5562 
5563 		ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
5564 		if (ret)
5565 			break;
5566 
5567 		find_first_clear_extent_bit(&device->alloc_state, start,
5568 					    &start, &end,
5569 					    CHUNK_TRIMMED | CHUNK_ALLOCATED);
5570 
5571 		/* Ensure we skip the reserved area in the first 1M */
5572 		start = max_t(u64, start, SZ_1M);
5573 
5574 		/*
5575 		 * If find_first_clear_extent_bit find a range that spans the
5576 		 * end of the device it will set end to -1, in this case it's up
5577 		 * to the caller to trim the value to the size of the device.
5578 		 */
5579 		end = min(end, device->total_bytes - 1);
5580 
5581 		len = end - start + 1;
5582 
5583 		/* We didn't find any extents */
5584 		if (!len) {
5585 			mutex_unlock(&fs_info->chunk_mutex);
5586 			ret = 0;
5587 			break;
5588 		}
5589 
5590 		ret = btrfs_issue_discard(device->bdev, start, len,
5591 					  &bytes);
5592 		if (!ret)
5593 			set_extent_bits(&device->alloc_state, start,
5594 					start + bytes - 1,
5595 					CHUNK_TRIMMED);
5596 		mutex_unlock(&fs_info->chunk_mutex);
5597 
5598 		if (ret)
5599 			break;
5600 
5601 		start += len;
5602 		*trimmed += bytes;
5603 
5604 		if (fatal_signal_pending(current)) {
5605 			ret = -ERESTARTSYS;
5606 			break;
5607 		}
5608 
5609 		cond_resched();
5610 	}
5611 
5612 	return ret;
5613 }
5614 
5615 /*
5616  * Trim the whole filesystem by:
5617  * 1) trimming the free space in each block group
5618  * 2) trimming the unallocated space on each device
5619  *
5620  * This will also continue trimming even if a block group or device encounters
5621  * an error.  The return value will be the last error, or 0 if nothing bad
5622  * happens.
5623  */
btrfs_trim_fs(struct btrfs_fs_info * fs_info,struct fstrim_range * range)5624 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
5625 {
5626 	struct btrfs_block_group_cache *cache = NULL;
5627 	struct btrfs_device *device;
5628 	struct list_head *devices;
5629 	u64 group_trimmed;
5630 	u64 range_end = U64_MAX;
5631 	u64 start;
5632 	u64 end;
5633 	u64 trimmed = 0;
5634 	u64 bg_failed = 0;
5635 	u64 dev_failed = 0;
5636 	int bg_ret = 0;
5637 	int dev_ret = 0;
5638 	int ret = 0;
5639 
5640 	/*
5641 	 * Check range overflow if range->len is set.
5642 	 * The default range->len is U64_MAX.
5643 	 */
5644 	if (range->len != U64_MAX &&
5645 	    check_add_overflow(range->start, range->len, &range_end))
5646 		return -EINVAL;
5647 
5648 	cache = btrfs_lookup_first_block_group(fs_info, range->start);
5649 	for (; cache; cache = btrfs_next_block_group(cache)) {
5650 		if (cache->key.objectid >= range_end) {
5651 			btrfs_put_block_group(cache);
5652 			break;
5653 		}
5654 
5655 		start = max(range->start, cache->key.objectid);
5656 		end = min(range_end, cache->key.objectid + cache->key.offset);
5657 
5658 		if (end - start >= range->minlen) {
5659 			if (!btrfs_block_group_cache_done(cache)) {
5660 				ret = btrfs_cache_block_group(cache, 0);
5661 				if (ret) {
5662 					bg_failed++;
5663 					bg_ret = ret;
5664 					continue;
5665 				}
5666 				ret = btrfs_wait_block_group_cache_done(cache);
5667 				if (ret) {
5668 					bg_failed++;
5669 					bg_ret = ret;
5670 					continue;
5671 				}
5672 			}
5673 			ret = btrfs_trim_block_group(cache,
5674 						     &group_trimmed,
5675 						     start,
5676 						     end,
5677 						     range->minlen);
5678 
5679 			trimmed += group_trimmed;
5680 			if (ret) {
5681 				bg_failed++;
5682 				bg_ret = ret;
5683 				continue;
5684 			}
5685 		}
5686 	}
5687 
5688 	if (bg_failed)
5689 		btrfs_warn(fs_info,
5690 			"failed to trim %llu block group(s), last error %d",
5691 			bg_failed, bg_ret);
5692 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
5693 	devices = &fs_info->fs_devices->devices;
5694 	list_for_each_entry(device, devices, dev_list) {
5695 		ret = btrfs_trim_free_extents(device, &group_trimmed);
5696 		if (ret) {
5697 			dev_failed++;
5698 			dev_ret = ret;
5699 			break;
5700 		}
5701 
5702 		trimmed += group_trimmed;
5703 	}
5704 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5705 
5706 	if (dev_failed)
5707 		btrfs_warn(fs_info,
5708 			"failed to trim %llu device(s), last error %d",
5709 			dev_failed, dev_ret);
5710 	range->len = trimmed;
5711 	if (bg_ret)
5712 		return bg_ret;
5713 	return dev_ret;
5714 }
5715 
5716 /*
5717  * btrfs_{start,end}_write_no_snapshotting() are similar to
5718  * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
5719  * data into the page cache through nocow before the subvolume is snapshoted,
5720  * but flush the data into disk after the snapshot creation, or to prevent
5721  * operations while snapshotting is ongoing and that cause the snapshot to be
5722  * inconsistent (writes followed by expanding truncates for example).
5723  */
btrfs_end_write_no_snapshotting(struct btrfs_root * root)5724 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
5725 {
5726 	percpu_counter_dec(&root->subv_writers->counter);
5727 	cond_wake_up(&root->subv_writers->wait);
5728 }
5729 
btrfs_start_write_no_snapshotting(struct btrfs_root * root)5730 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
5731 {
5732 	if (atomic_read(&root->will_be_snapshotted))
5733 		return 0;
5734 
5735 	percpu_counter_inc(&root->subv_writers->counter);
5736 	/*
5737 	 * Make sure counter is updated before we check for snapshot creation.
5738 	 */
5739 	smp_mb();
5740 	if (atomic_read(&root->will_be_snapshotted)) {
5741 		btrfs_end_write_no_snapshotting(root);
5742 		return 0;
5743 	}
5744 	return 1;
5745 }
5746 
btrfs_wait_for_snapshot_creation(struct btrfs_root * root)5747 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
5748 {
5749 	while (true) {
5750 		int ret;
5751 
5752 		ret = btrfs_start_write_no_snapshotting(root);
5753 		if (ret)
5754 			break;
5755 		wait_var_event(&root->will_be_snapshotted,
5756 			       !atomic_read(&root->will_be_snapshotted));
5757 	}
5758 }
5759