• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5 
6 #include <linux/fs.h>
7 #include <linux/slab.h>
8 #include <linux/sched.h>
9 #include <linux/writeback.h>
10 #include <linux/pagemap.h>
11 #include <linux/blkdev.h>
12 #include <linux/uuid.h>
13 #include "misc.h"
14 #include "ctree.h"
15 #include "disk-io.h"
16 #include "transaction.h"
17 #include "locking.h"
18 #include "tree-log.h"
19 #include "inode-map.h"
20 #include "volumes.h"
21 #include "dev-replace.h"
22 #include "qgroup.h"
23 #include "block-group.h"
24 #include "space-info.h"
25 
26 #define BTRFS_ROOT_TRANS_TAG 0
27 
28 /*
29  * Transaction states and transitions
30  *
31  * No running transaction (fs tree blocks are not modified)
32  * |
33  * | To next stage:
34  * |  Call start_transaction() variants. Except btrfs_join_transaction_nostart().
35  * V
36  * Transaction N [[TRANS_STATE_RUNNING]]
37  * |
38  * | New trans handles can be attached to transaction N by calling all
39  * | start_transaction() variants.
40  * |
41  * | To next stage:
42  * |  Call btrfs_commit_transaction() on any trans handle attached to
43  * |  transaction N
44  * V
45  * Transaction N [[TRANS_STATE_COMMIT_START]]
46  * |
47  * | Will wait for previous running transaction to completely finish if there
48  * | is one
49  * |
50  * | Then one of the following happes:
51  * | - Wait for all other trans handle holders to release.
52  * |   The btrfs_commit_transaction() caller will do the commit work.
53  * | - Wait for current transaction to be committed by others.
54  * |   Other btrfs_commit_transaction() caller will do the commit work.
55  * |
56  * | At this stage, only btrfs_join_transaction*() variants can attach
57  * | to this running transaction.
58  * | All other variants will wait for current one to finish and attach to
59  * | transaction N+1.
60  * |
61  * | To next stage:
62  * |  Caller is chosen to commit transaction N, and all other trans handle
63  * |  haven been released.
64  * V
65  * Transaction N [[TRANS_STATE_COMMIT_DOING]]
66  * |
67  * | The heavy lifting transaction work is started.
68  * | From running delayed refs (modifying extent tree) to creating pending
69  * | snapshots, running qgroups.
70  * | In short, modify supporting trees to reflect modifications of subvolume
71  * | trees.
72  * |
73  * | At this stage, all start_transaction() calls will wait for this
74  * | transaction to finish and attach to transaction N+1.
75  * |
76  * | To next stage:
77  * |  Until all supporting trees are updated.
78  * V
79  * Transaction N [[TRANS_STATE_UNBLOCKED]]
80  * |						    Transaction N+1
81  * | All needed trees are modified, thus we only    [[TRANS_STATE_RUNNING]]
82  * | need to write them back to disk and update	    |
83  * | super blocks.				    |
84  * |						    |
85  * | At this stage, new transaction is allowed to   |
86  * | start.					    |
87  * | All new start_transaction() calls will be	    |
88  * | attached to transid N+1.			    |
89  * |						    |
90  * | To next stage:				    |
91  * |  Until all tree blocks are super blocks are    |
92  * |  written to block devices			    |
93  * V						    |
94  * Transaction N [[TRANS_STATE_COMPLETED]]	    V
95  *   All tree blocks and super blocks are written.  Transaction N+1
96  *   This transaction is finished and all its	    [[TRANS_STATE_COMMIT_START]]
97  *   data structures will be cleaned up.	    | Life goes on
98  */
99 static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
100 	[TRANS_STATE_RUNNING]		= 0U,
101 	[TRANS_STATE_COMMIT_START]	= (__TRANS_START | __TRANS_ATTACH),
102 	[TRANS_STATE_COMMIT_DOING]	= (__TRANS_START |
103 					   __TRANS_ATTACH |
104 					   __TRANS_JOIN |
105 					   __TRANS_JOIN_NOSTART),
106 	[TRANS_STATE_UNBLOCKED]		= (__TRANS_START |
107 					   __TRANS_ATTACH |
108 					   __TRANS_JOIN |
109 					   __TRANS_JOIN_NOLOCK |
110 					   __TRANS_JOIN_NOSTART),
111 	[TRANS_STATE_COMPLETED]		= (__TRANS_START |
112 					   __TRANS_ATTACH |
113 					   __TRANS_JOIN |
114 					   __TRANS_JOIN_NOLOCK |
115 					   __TRANS_JOIN_NOSTART),
116 };
117 
btrfs_put_transaction(struct btrfs_transaction * transaction)118 void btrfs_put_transaction(struct btrfs_transaction *transaction)
119 {
120 	WARN_ON(refcount_read(&transaction->use_count) == 0);
121 	if (refcount_dec_and_test(&transaction->use_count)) {
122 		BUG_ON(!list_empty(&transaction->list));
123 		WARN_ON(!RB_EMPTY_ROOT(
124 				&transaction->delayed_refs.href_root.rb_root));
125 		WARN_ON(!RB_EMPTY_ROOT(
126 				&transaction->delayed_refs.dirty_extent_root));
127 		if (transaction->delayed_refs.pending_csums)
128 			btrfs_err(transaction->fs_info,
129 				  "pending csums is %llu",
130 				  transaction->delayed_refs.pending_csums);
131 		/*
132 		 * If any block groups are found in ->deleted_bgs then it's
133 		 * because the transaction was aborted and a commit did not
134 		 * happen (things failed before writing the new superblock
135 		 * and calling btrfs_finish_extent_commit()), so we can not
136 		 * discard the physical locations of the block groups.
137 		 */
138 		while (!list_empty(&transaction->deleted_bgs)) {
139 			struct btrfs_block_group *cache;
140 
141 			cache = list_first_entry(&transaction->deleted_bgs,
142 						 struct btrfs_block_group,
143 						 bg_list);
144 			list_del_init(&cache->bg_list);
145 			btrfs_unfreeze_block_group(cache);
146 			btrfs_put_block_group(cache);
147 		}
148 		WARN_ON(!list_empty(&transaction->dev_update_list));
149 		kfree(transaction);
150 	}
151 }
152 
switch_commit_roots(struct btrfs_trans_handle * trans)153 static noinline void switch_commit_roots(struct btrfs_trans_handle *trans)
154 {
155 	struct btrfs_transaction *cur_trans = trans->transaction;
156 	struct btrfs_fs_info *fs_info = trans->fs_info;
157 	struct btrfs_root *root, *tmp;
158 	struct btrfs_caching_control *caching_ctl, *next;
159 
160 	down_write(&fs_info->commit_root_sem);
161 	list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits,
162 				 dirty_list) {
163 		list_del_init(&root->dirty_list);
164 		free_extent_buffer(root->commit_root);
165 		root->commit_root = btrfs_root_node(root);
166 		if (is_fstree(root->root_key.objectid))
167 			btrfs_unpin_free_ino(root);
168 		extent_io_tree_release(&root->dirty_log_pages);
169 		btrfs_qgroup_clean_swapped_blocks(root);
170 	}
171 
172 	/* We can free old roots now. */
173 	spin_lock(&cur_trans->dropped_roots_lock);
174 	while (!list_empty(&cur_trans->dropped_roots)) {
175 		root = list_first_entry(&cur_trans->dropped_roots,
176 					struct btrfs_root, root_list);
177 		list_del_init(&root->root_list);
178 		spin_unlock(&cur_trans->dropped_roots_lock);
179 		btrfs_free_log(trans, root);
180 		btrfs_drop_and_free_fs_root(fs_info, root);
181 		spin_lock(&cur_trans->dropped_roots_lock);
182 	}
183 	spin_unlock(&cur_trans->dropped_roots_lock);
184 
185 	/*
186 	 * We have to update the last_byte_to_unpin under the commit_root_sem,
187 	 * at the same time we swap out the commit roots.
188 	 *
189 	 * This is because we must have a real view of the last spot the caching
190 	 * kthreads were while caching.  Consider the following views of the
191 	 * extent tree for a block group
192 	 *
193 	 * commit root
194 	 * +----+----+----+----+----+----+----+
195 	 * |\\\\|    |\\\\|\\\\|    |\\\\|\\\\|
196 	 * +----+----+----+----+----+----+----+
197 	 * 0    1    2    3    4    5    6    7
198 	 *
199 	 * new commit root
200 	 * +----+----+----+----+----+----+----+
201 	 * |    |    |    |\\\\|    |    |\\\\|
202 	 * +----+----+----+----+----+----+----+
203 	 * 0    1    2    3    4    5    6    7
204 	 *
205 	 * If the cache_ctl->progress was at 3, then we are only allowed to
206 	 * unpin [0,1) and [2,3], because the caching thread has already
207 	 * processed those extents.  We are not allowed to unpin [5,6), because
208 	 * the caching thread will re-start it's search from 3, and thus find
209 	 * the hole from [4,6) to add to the free space cache.
210 	 */
211 	list_for_each_entry_safe(caching_ctl, next,
212 				 &fs_info->caching_block_groups, list) {
213 		struct btrfs_block_group *cache = caching_ctl->block_group;
214 
215 		if (btrfs_block_group_done(cache)) {
216 			cache->last_byte_to_unpin = (u64)-1;
217 			list_del_init(&caching_ctl->list);
218 			btrfs_put_caching_control(caching_ctl);
219 		} else {
220 			cache->last_byte_to_unpin = caching_ctl->progress;
221 		}
222 	}
223 	up_write(&fs_info->commit_root_sem);
224 }
225 
extwriter_counter_inc(struct btrfs_transaction * trans,unsigned int type)226 static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
227 					 unsigned int type)
228 {
229 	if (type & TRANS_EXTWRITERS)
230 		atomic_inc(&trans->num_extwriters);
231 }
232 
extwriter_counter_dec(struct btrfs_transaction * trans,unsigned int type)233 static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
234 					 unsigned int type)
235 {
236 	if (type & TRANS_EXTWRITERS)
237 		atomic_dec(&trans->num_extwriters);
238 }
239 
extwriter_counter_init(struct btrfs_transaction * trans,unsigned int type)240 static inline void extwriter_counter_init(struct btrfs_transaction *trans,
241 					  unsigned int type)
242 {
243 	atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
244 }
245 
extwriter_counter_read(struct btrfs_transaction * trans)246 static inline int extwriter_counter_read(struct btrfs_transaction *trans)
247 {
248 	return atomic_read(&trans->num_extwriters);
249 }
250 
251 /*
252  * To be called after all the new block groups attached to the transaction
253  * handle have been created (btrfs_create_pending_block_groups()).
254  */
btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle * trans)255 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
256 {
257 	struct btrfs_fs_info *fs_info = trans->fs_info;
258 
259 	if (!trans->chunk_bytes_reserved)
260 		return;
261 
262 	WARN_ON_ONCE(!list_empty(&trans->new_bgs));
263 
264 	btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv,
265 				trans->chunk_bytes_reserved, NULL);
266 	trans->chunk_bytes_reserved = 0;
267 }
268 
269 /*
270  * either allocate a new transaction or hop into the existing one
271  */
join_transaction(struct btrfs_fs_info * fs_info,unsigned int type)272 static noinline int join_transaction(struct btrfs_fs_info *fs_info,
273 				     unsigned int type)
274 {
275 	struct btrfs_transaction *cur_trans;
276 
277 	spin_lock(&fs_info->trans_lock);
278 loop:
279 	/* The file system has been taken offline. No new transactions. */
280 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
281 		spin_unlock(&fs_info->trans_lock);
282 		return -EROFS;
283 	}
284 
285 	cur_trans = fs_info->running_transaction;
286 	if (cur_trans) {
287 		if (TRANS_ABORTED(cur_trans)) {
288 			spin_unlock(&fs_info->trans_lock);
289 			return cur_trans->aborted;
290 		}
291 		if (btrfs_blocked_trans_types[cur_trans->state] & type) {
292 			spin_unlock(&fs_info->trans_lock);
293 			return -EBUSY;
294 		}
295 		refcount_inc(&cur_trans->use_count);
296 		atomic_inc(&cur_trans->num_writers);
297 		extwriter_counter_inc(cur_trans, type);
298 		spin_unlock(&fs_info->trans_lock);
299 		return 0;
300 	}
301 	spin_unlock(&fs_info->trans_lock);
302 
303 	/*
304 	 * If we are ATTACH or TRANS_JOIN_NOSTART, we just want to catch the
305 	 * current transaction, and commit it. If there is no transaction, just
306 	 * return ENOENT.
307 	 */
308 	if (type == TRANS_ATTACH || type == TRANS_JOIN_NOSTART)
309 		return -ENOENT;
310 
311 	/*
312 	 * JOIN_NOLOCK only happens during the transaction commit, so
313 	 * it is impossible that ->running_transaction is NULL
314 	 */
315 	BUG_ON(type == TRANS_JOIN_NOLOCK);
316 
317 	cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS);
318 	if (!cur_trans)
319 		return -ENOMEM;
320 
321 	spin_lock(&fs_info->trans_lock);
322 	if (fs_info->running_transaction) {
323 		/*
324 		 * someone started a transaction after we unlocked.  Make sure
325 		 * to redo the checks above
326 		 */
327 		kfree(cur_trans);
328 		goto loop;
329 	} else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
330 		spin_unlock(&fs_info->trans_lock);
331 		kfree(cur_trans);
332 		return -EROFS;
333 	}
334 
335 	cur_trans->fs_info = fs_info;
336 	atomic_set(&cur_trans->pending_ordered, 0);
337 	init_waitqueue_head(&cur_trans->pending_wait);
338 	atomic_set(&cur_trans->num_writers, 1);
339 	extwriter_counter_init(cur_trans, type);
340 	init_waitqueue_head(&cur_trans->writer_wait);
341 	init_waitqueue_head(&cur_trans->commit_wait);
342 	cur_trans->state = TRANS_STATE_RUNNING;
343 	/*
344 	 * One for this trans handle, one so it will live on until we
345 	 * commit the transaction.
346 	 */
347 	refcount_set(&cur_trans->use_count, 2);
348 	cur_trans->flags = 0;
349 	cur_trans->start_time = ktime_get_seconds();
350 
351 	memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));
352 
353 	cur_trans->delayed_refs.href_root = RB_ROOT_CACHED;
354 	cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
355 	atomic_set(&cur_trans->delayed_refs.num_entries, 0);
356 
357 	/*
358 	 * although the tree mod log is per file system and not per transaction,
359 	 * the log must never go across transaction boundaries.
360 	 */
361 	smp_mb();
362 	if (!list_empty(&fs_info->tree_mod_seq_list))
363 		WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");
364 	if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
365 		WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");
366 	atomic64_set(&fs_info->tree_mod_seq, 0);
367 
368 	spin_lock_init(&cur_trans->delayed_refs.lock);
369 
370 	INIT_LIST_HEAD(&cur_trans->pending_snapshots);
371 	INIT_LIST_HEAD(&cur_trans->dev_update_list);
372 	INIT_LIST_HEAD(&cur_trans->switch_commits);
373 	INIT_LIST_HEAD(&cur_trans->dirty_bgs);
374 	INIT_LIST_HEAD(&cur_trans->io_bgs);
375 	INIT_LIST_HEAD(&cur_trans->dropped_roots);
376 	mutex_init(&cur_trans->cache_write_mutex);
377 	spin_lock_init(&cur_trans->dirty_bgs_lock);
378 	INIT_LIST_HEAD(&cur_trans->deleted_bgs);
379 	spin_lock_init(&cur_trans->dropped_roots_lock);
380 	list_add_tail(&cur_trans->list, &fs_info->trans_list);
381 	extent_io_tree_init(fs_info, &cur_trans->dirty_pages,
382 			IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode);
383 	extent_io_tree_init(fs_info, &cur_trans->pinned_extents,
384 			IO_TREE_FS_PINNED_EXTENTS, NULL);
385 	fs_info->generation++;
386 	cur_trans->transid = fs_info->generation;
387 	fs_info->running_transaction = cur_trans;
388 	cur_trans->aborted = 0;
389 	spin_unlock(&fs_info->trans_lock);
390 
391 	return 0;
392 }
393 
394 /*
395  * This does all the record keeping required to make sure that a shareable root
396  * is properly recorded in a given transaction.  This is required to make sure
397  * the old root from before we joined the transaction is deleted when the
398  * transaction commits.
399  */
record_root_in_trans(struct btrfs_trans_handle * trans,struct btrfs_root * root,int force)400 static int record_root_in_trans(struct btrfs_trans_handle *trans,
401 			       struct btrfs_root *root,
402 			       int force)
403 {
404 	struct btrfs_fs_info *fs_info = root->fs_info;
405 
406 	if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
407 	    root->last_trans < trans->transid) || force) {
408 		WARN_ON(root == fs_info->extent_root);
409 		WARN_ON(!force && root->commit_root != root->node);
410 
411 		/*
412 		 * see below for IN_TRANS_SETUP usage rules
413 		 * we have the reloc mutex held now, so there
414 		 * is only one writer in this function
415 		 */
416 		set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
417 
418 		/* make sure readers find IN_TRANS_SETUP before
419 		 * they find our root->last_trans update
420 		 */
421 		smp_wmb();
422 
423 		spin_lock(&fs_info->fs_roots_radix_lock);
424 		if (root->last_trans == trans->transid && !force) {
425 			spin_unlock(&fs_info->fs_roots_radix_lock);
426 			return 0;
427 		}
428 		radix_tree_tag_set(&fs_info->fs_roots_radix,
429 				   (unsigned long)root->root_key.objectid,
430 				   BTRFS_ROOT_TRANS_TAG);
431 		spin_unlock(&fs_info->fs_roots_radix_lock);
432 		root->last_trans = trans->transid;
433 
434 		/* this is pretty tricky.  We don't want to
435 		 * take the relocation lock in btrfs_record_root_in_trans
436 		 * unless we're really doing the first setup for this root in
437 		 * this transaction.
438 		 *
439 		 * Normally we'd use root->last_trans as a flag to decide
440 		 * if we want to take the expensive mutex.
441 		 *
442 		 * But, we have to set root->last_trans before we
443 		 * init the relocation root, otherwise, we trip over warnings
444 		 * in ctree.c.  The solution used here is to flag ourselves
445 		 * with root IN_TRANS_SETUP.  When this is 1, we're still
446 		 * fixing up the reloc trees and everyone must wait.
447 		 *
448 		 * When this is zero, they can trust root->last_trans and fly
449 		 * through btrfs_record_root_in_trans without having to take the
450 		 * lock.  smp_wmb() makes sure that all the writes above are
451 		 * done before we pop in the zero below
452 		 */
453 		btrfs_init_reloc_root(trans, root);
454 		smp_mb__before_atomic();
455 		clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
456 	}
457 	return 0;
458 }
459 
460 
btrfs_add_dropped_root(struct btrfs_trans_handle * trans,struct btrfs_root * root)461 void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
462 			    struct btrfs_root *root)
463 {
464 	struct btrfs_fs_info *fs_info = root->fs_info;
465 	struct btrfs_transaction *cur_trans = trans->transaction;
466 
467 	/* Add ourselves to the transaction dropped list */
468 	spin_lock(&cur_trans->dropped_roots_lock);
469 	list_add_tail(&root->root_list, &cur_trans->dropped_roots);
470 	spin_unlock(&cur_trans->dropped_roots_lock);
471 
472 	/* Make sure we don't try to update the root at commit time */
473 	spin_lock(&fs_info->fs_roots_radix_lock);
474 	radix_tree_tag_clear(&fs_info->fs_roots_radix,
475 			     (unsigned long)root->root_key.objectid,
476 			     BTRFS_ROOT_TRANS_TAG);
477 	spin_unlock(&fs_info->fs_roots_radix_lock);
478 }
479 
btrfs_record_root_in_trans(struct btrfs_trans_handle * trans,struct btrfs_root * root)480 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
481 			       struct btrfs_root *root)
482 {
483 	struct btrfs_fs_info *fs_info = root->fs_info;
484 
485 	if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
486 		return 0;
487 
488 	/*
489 	 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
490 	 * and barriers
491 	 */
492 	smp_rmb();
493 	if (root->last_trans == trans->transid &&
494 	    !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
495 		return 0;
496 
497 	mutex_lock(&fs_info->reloc_mutex);
498 	record_root_in_trans(trans, root, 0);
499 	mutex_unlock(&fs_info->reloc_mutex);
500 
501 	return 0;
502 }
503 
is_transaction_blocked(struct btrfs_transaction * trans)504 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
505 {
506 	return (trans->state >= TRANS_STATE_COMMIT_START &&
507 		trans->state < TRANS_STATE_UNBLOCKED &&
508 		!TRANS_ABORTED(trans));
509 }
510 
511 /* wait for commit against the current transaction to become unblocked
512  * when this is done, it is safe to start a new transaction, but the current
513  * transaction might not be fully on disk.
514  */
wait_current_trans(struct btrfs_fs_info * fs_info)515 static void wait_current_trans(struct btrfs_fs_info *fs_info)
516 {
517 	struct btrfs_transaction *cur_trans;
518 
519 	spin_lock(&fs_info->trans_lock);
520 	cur_trans = fs_info->running_transaction;
521 	if (cur_trans && is_transaction_blocked(cur_trans)) {
522 		refcount_inc(&cur_trans->use_count);
523 		spin_unlock(&fs_info->trans_lock);
524 
525 		wait_event(fs_info->transaction_wait,
526 			   cur_trans->state >= TRANS_STATE_UNBLOCKED ||
527 			   TRANS_ABORTED(cur_trans));
528 		btrfs_put_transaction(cur_trans);
529 	} else {
530 		spin_unlock(&fs_info->trans_lock);
531 	}
532 }
533 
may_wait_transaction(struct btrfs_fs_info * fs_info,int type)534 static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type)
535 {
536 	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
537 		return 0;
538 
539 	if (type == TRANS_START)
540 		return 1;
541 
542 	return 0;
543 }
544 
need_reserve_reloc_root(struct btrfs_root * root)545 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
546 {
547 	struct btrfs_fs_info *fs_info = root->fs_info;
548 
549 	if (!fs_info->reloc_ctl ||
550 	    !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
551 	    root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
552 	    root->reloc_root)
553 		return false;
554 
555 	return true;
556 }
557 
558 static struct btrfs_trans_handle *
start_transaction(struct btrfs_root * root,unsigned int num_items,unsigned int type,enum btrfs_reserve_flush_enum flush,bool enforce_qgroups)559 start_transaction(struct btrfs_root *root, unsigned int num_items,
560 		  unsigned int type, enum btrfs_reserve_flush_enum flush,
561 		  bool enforce_qgroups)
562 {
563 	struct btrfs_fs_info *fs_info = root->fs_info;
564 	struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
565 	struct btrfs_trans_handle *h;
566 	struct btrfs_transaction *cur_trans;
567 	u64 num_bytes = 0;
568 	u64 qgroup_reserved = 0;
569 	bool reloc_reserved = false;
570 	bool do_chunk_alloc = false;
571 	int ret;
572 
573 	/* Send isn't supposed to start transactions. */
574 	ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
575 
576 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
577 		return ERR_PTR(-EROFS);
578 
579 	if (current->journal_info) {
580 		WARN_ON(type & TRANS_EXTWRITERS);
581 		h = current->journal_info;
582 		refcount_inc(&h->use_count);
583 		WARN_ON(refcount_read(&h->use_count) > 2);
584 		h->orig_rsv = h->block_rsv;
585 		h->block_rsv = NULL;
586 		goto got_it;
587 	}
588 
589 	/*
590 	 * Do the reservation before we join the transaction so we can do all
591 	 * the appropriate flushing if need be.
592 	 */
593 	if (num_items && root != fs_info->chunk_root) {
594 		struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv;
595 		u64 delayed_refs_bytes = 0;
596 
597 		qgroup_reserved = num_items * fs_info->nodesize;
598 		ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved,
599 				enforce_qgroups);
600 		if (ret)
601 			return ERR_PTR(ret);
602 
603 		/*
604 		 * We want to reserve all the bytes we may need all at once, so
605 		 * we only do 1 enospc flushing cycle per transaction start.  We
606 		 * accomplish this by simply assuming we'll do 2 x num_items
607 		 * worth of delayed refs updates in this trans handle, and
608 		 * refill that amount for whatever is missing in the reserve.
609 		 */
610 		num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items);
611 		if (flush == BTRFS_RESERVE_FLUSH_ALL &&
612 		    delayed_refs_rsv->full == 0) {
613 			delayed_refs_bytes = num_bytes;
614 			num_bytes <<= 1;
615 		}
616 
617 		/*
618 		 * Do the reservation for the relocation root creation
619 		 */
620 		if (need_reserve_reloc_root(root)) {
621 			num_bytes += fs_info->nodesize;
622 			reloc_reserved = true;
623 		}
624 
625 		ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush);
626 		if (ret)
627 			goto reserve_fail;
628 		if (delayed_refs_bytes) {
629 			btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv,
630 							  delayed_refs_bytes);
631 			num_bytes -= delayed_refs_bytes;
632 		}
633 
634 		if (rsv->space_info->force_alloc)
635 			do_chunk_alloc = true;
636 	} else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL &&
637 		   !delayed_refs_rsv->full) {
638 		/*
639 		 * Some people call with btrfs_start_transaction(root, 0)
640 		 * because they can be throttled, but have some other mechanism
641 		 * for reserving space.  We still want these guys to refill the
642 		 * delayed block_rsv so just add 1 items worth of reservation
643 		 * here.
644 		 */
645 		ret = btrfs_delayed_refs_rsv_refill(fs_info, flush);
646 		if (ret)
647 			goto reserve_fail;
648 	}
649 again:
650 	h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
651 	if (!h) {
652 		ret = -ENOMEM;
653 		goto alloc_fail;
654 	}
655 
656 	/*
657 	 * If we are JOIN_NOLOCK we're already committing a transaction and
658 	 * waiting on this guy, so we don't need to do the sb_start_intwrite
659 	 * because we're already holding a ref.  We need this because we could
660 	 * have raced in and did an fsync() on a file which can kick a commit
661 	 * and then we deadlock with somebody doing a freeze.
662 	 *
663 	 * If we are ATTACH, it means we just want to catch the current
664 	 * transaction and commit it, so we needn't do sb_start_intwrite().
665 	 */
666 	if (type & __TRANS_FREEZABLE)
667 		sb_start_intwrite(fs_info->sb);
668 
669 	if (may_wait_transaction(fs_info, type))
670 		wait_current_trans(fs_info);
671 
672 	do {
673 		ret = join_transaction(fs_info, type);
674 		if (ret == -EBUSY) {
675 			wait_current_trans(fs_info);
676 			if (unlikely(type == TRANS_ATTACH ||
677 				     type == TRANS_JOIN_NOSTART))
678 				ret = -ENOENT;
679 		}
680 	} while (ret == -EBUSY);
681 
682 	if (ret < 0)
683 		goto join_fail;
684 
685 	cur_trans = fs_info->running_transaction;
686 
687 	h->transid = cur_trans->transid;
688 	h->transaction = cur_trans;
689 	h->root = root;
690 	refcount_set(&h->use_count, 1);
691 	h->fs_info = root->fs_info;
692 
693 	h->type = type;
694 	h->can_flush_pending_bgs = true;
695 	INIT_LIST_HEAD(&h->new_bgs);
696 
697 	smp_mb();
698 	if (cur_trans->state >= TRANS_STATE_COMMIT_START &&
699 	    may_wait_transaction(fs_info, type)) {
700 		current->journal_info = h;
701 		btrfs_commit_transaction(h);
702 		goto again;
703 	}
704 
705 	if (num_bytes) {
706 		trace_btrfs_space_reservation(fs_info, "transaction",
707 					      h->transid, num_bytes, 1);
708 		h->block_rsv = &fs_info->trans_block_rsv;
709 		h->bytes_reserved = num_bytes;
710 		h->reloc_reserved = reloc_reserved;
711 	}
712 
713 got_it:
714 	if (!current->journal_info)
715 		current->journal_info = h;
716 
717 	/*
718 	 * If the space_info is marked ALLOC_FORCE then we'll get upgraded to
719 	 * ALLOC_FORCE the first run through, and then we won't allocate for
720 	 * anybody else who races in later.  We don't care about the return
721 	 * value here.
722 	 */
723 	if (do_chunk_alloc && num_bytes) {
724 		u64 flags = h->block_rsv->space_info->flags;
725 
726 		btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags),
727 				  CHUNK_ALLOC_NO_FORCE);
728 	}
729 
730 	/*
731 	 * btrfs_record_root_in_trans() needs to alloc new extents, and may
732 	 * call btrfs_join_transaction() while we're also starting a
733 	 * transaction.
734 	 *
735 	 * Thus it need to be called after current->journal_info initialized,
736 	 * or we can deadlock.
737 	 */
738 	btrfs_record_root_in_trans(h, root);
739 
740 	return h;
741 
742 join_fail:
743 	if (type & __TRANS_FREEZABLE)
744 		sb_end_intwrite(fs_info->sb);
745 	kmem_cache_free(btrfs_trans_handle_cachep, h);
746 alloc_fail:
747 	if (num_bytes)
748 		btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv,
749 					num_bytes, NULL);
750 reserve_fail:
751 	btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved);
752 	return ERR_PTR(ret);
753 }
754 
btrfs_start_transaction(struct btrfs_root * root,unsigned int num_items)755 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
756 						   unsigned int num_items)
757 {
758 	return start_transaction(root, num_items, TRANS_START,
759 				 BTRFS_RESERVE_FLUSH_ALL, true);
760 }
761 
btrfs_start_transaction_fallback_global_rsv(struct btrfs_root * root,unsigned int num_items)762 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
763 					struct btrfs_root *root,
764 					unsigned int num_items)
765 {
766 	return start_transaction(root, num_items, TRANS_START,
767 				 BTRFS_RESERVE_FLUSH_ALL_STEAL, false);
768 }
769 
btrfs_join_transaction(struct btrfs_root * root)770 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
771 {
772 	return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH,
773 				 true);
774 }
775 
btrfs_join_transaction_spacecache(struct btrfs_root * root)776 struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root)
777 {
778 	return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
779 				 BTRFS_RESERVE_NO_FLUSH, true);
780 }
781 
782 /*
783  * Similar to regular join but it never starts a transaction when none is
784  * running or after waiting for the current one to finish.
785  */
btrfs_join_transaction_nostart(struct btrfs_root * root)786 struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root)
787 {
788 	return start_transaction(root, 0, TRANS_JOIN_NOSTART,
789 				 BTRFS_RESERVE_NO_FLUSH, true);
790 }
791 
792 /*
793  * btrfs_attach_transaction() - catch the running transaction
794  *
795  * It is used when we want to commit the current the transaction, but
796  * don't want to start a new one.
797  *
798  * Note: If this function return -ENOENT, it just means there is no
799  * running transaction. But it is possible that the inactive transaction
800  * is still in the memory, not fully on disk. If you hope there is no
801  * inactive transaction in the fs when -ENOENT is returned, you should
802  * invoke
803  *     btrfs_attach_transaction_barrier()
804  */
btrfs_attach_transaction(struct btrfs_root * root)805 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
806 {
807 	return start_transaction(root, 0, TRANS_ATTACH,
808 				 BTRFS_RESERVE_NO_FLUSH, true);
809 }
810 
811 /*
812  * btrfs_attach_transaction_barrier() - catch the running transaction
813  *
814  * It is similar to the above function, the difference is this one
815  * will wait for all the inactive transactions until they fully
816  * complete.
817  */
818 struct btrfs_trans_handle *
btrfs_attach_transaction_barrier(struct btrfs_root * root)819 btrfs_attach_transaction_barrier(struct btrfs_root *root)
820 {
821 	struct btrfs_trans_handle *trans;
822 
823 	trans = start_transaction(root, 0, TRANS_ATTACH,
824 				  BTRFS_RESERVE_NO_FLUSH, true);
825 	if (trans == ERR_PTR(-ENOENT)) {
826 		int ret;
827 
828 		ret = btrfs_wait_for_commit(root->fs_info, 0);
829 		if (ret)
830 			return ERR_PTR(ret);
831 	}
832 
833 	return trans;
834 }
835 
836 /* wait for a transaction commit to be fully complete */
wait_for_commit(struct btrfs_transaction * commit)837 static noinline void wait_for_commit(struct btrfs_transaction *commit)
838 {
839 	wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
840 }
841 
btrfs_wait_for_commit(struct btrfs_fs_info * fs_info,u64 transid)842 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid)
843 {
844 	struct btrfs_transaction *cur_trans = NULL, *t;
845 	int ret = 0;
846 
847 	if (transid) {
848 		if (transid <= fs_info->last_trans_committed)
849 			goto out;
850 
851 		/* find specified transaction */
852 		spin_lock(&fs_info->trans_lock);
853 		list_for_each_entry(t, &fs_info->trans_list, list) {
854 			if (t->transid == transid) {
855 				cur_trans = t;
856 				refcount_inc(&cur_trans->use_count);
857 				ret = 0;
858 				break;
859 			}
860 			if (t->transid > transid) {
861 				ret = 0;
862 				break;
863 			}
864 		}
865 		spin_unlock(&fs_info->trans_lock);
866 
867 		/*
868 		 * The specified transaction doesn't exist, or we
869 		 * raced with btrfs_commit_transaction
870 		 */
871 		if (!cur_trans) {
872 			if (transid > fs_info->last_trans_committed)
873 				ret = -EINVAL;
874 			goto out;
875 		}
876 	} else {
877 		/* find newest transaction that is committing | committed */
878 		spin_lock(&fs_info->trans_lock);
879 		list_for_each_entry_reverse(t, &fs_info->trans_list,
880 					    list) {
881 			if (t->state >= TRANS_STATE_COMMIT_START) {
882 				if (t->state == TRANS_STATE_COMPLETED)
883 					break;
884 				cur_trans = t;
885 				refcount_inc(&cur_trans->use_count);
886 				break;
887 			}
888 		}
889 		spin_unlock(&fs_info->trans_lock);
890 		if (!cur_trans)
891 			goto out;  /* nothing committing|committed */
892 	}
893 
894 	wait_for_commit(cur_trans);
895 	ret = cur_trans->aborted;
896 	btrfs_put_transaction(cur_trans);
897 out:
898 	return ret;
899 }
900 
btrfs_throttle(struct btrfs_fs_info * fs_info)901 void btrfs_throttle(struct btrfs_fs_info *fs_info)
902 {
903 	wait_current_trans(fs_info);
904 }
905 
should_end_transaction(struct btrfs_trans_handle * trans)906 static int should_end_transaction(struct btrfs_trans_handle *trans)
907 {
908 	struct btrfs_fs_info *fs_info = trans->fs_info;
909 
910 	if (btrfs_check_space_for_delayed_refs(fs_info))
911 		return 1;
912 
913 	return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5);
914 }
915 
btrfs_should_end_transaction(struct btrfs_trans_handle * trans)916 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
917 {
918 	struct btrfs_transaction *cur_trans = trans->transaction;
919 
920 	smp_mb();
921 	if (cur_trans->state >= TRANS_STATE_COMMIT_START ||
922 	    cur_trans->delayed_refs.flushing)
923 		return 1;
924 
925 	return should_end_transaction(trans);
926 }
927 
btrfs_trans_release_metadata(struct btrfs_trans_handle * trans)928 static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans)
929 
930 {
931 	struct btrfs_fs_info *fs_info = trans->fs_info;
932 
933 	if (!trans->block_rsv) {
934 		ASSERT(!trans->bytes_reserved);
935 		return;
936 	}
937 
938 	if (!trans->bytes_reserved)
939 		return;
940 
941 	ASSERT(trans->block_rsv == &fs_info->trans_block_rsv);
942 	trace_btrfs_space_reservation(fs_info, "transaction",
943 				      trans->transid, trans->bytes_reserved, 0);
944 	btrfs_block_rsv_release(fs_info, trans->block_rsv,
945 				trans->bytes_reserved, NULL);
946 	trans->bytes_reserved = 0;
947 }
948 
__btrfs_end_transaction(struct btrfs_trans_handle * trans,int throttle)949 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
950 				   int throttle)
951 {
952 	struct btrfs_fs_info *info = trans->fs_info;
953 	struct btrfs_transaction *cur_trans = trans->transaction;
954 	int err = 0;
955 
956 	if (refcount_read(&trans->use_count) > 1) {
957 		refcount_dec(&trans->use_count);
958 		trans->block_rsv = trans->orig_rsv;
959 		return 0;
960 	}
961 
962 	btrfs_trans_release_metadata(trans);
963 	trans->block_rsv = NULL;
964 
965 	btrfs_create_pending_block_groups(trans);
966 
967 	btrfs_trans_release_chunk_metadata(trans);
968 
969 	if (trans->type & __TRANS_FREEZABLE)
970 		sb_end_intwrite(info->sb);
971 
972 	WARN_ON(cur_trans != info->running_transaction);
973 	WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
974 	atomic_dec(&cur_trans->num_writers);
975 	extwriter_counter_dec(cur_trans, trans->type);
976 
977 	cond_wake_up(&cur_trans->writer_wait);
978 	btrfs_put_transaction(cur_trans);
979 
980 	if (current->journal_info == trans)
981 		current->journal_info = NULL;
982 
983 	if (throttle)
984 		btrfs_run_delayed_iputs(info);
985 
986 	if (TRANS_ABORTED(trans) ||
987 	    test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) {
988 		wake_up_process(info->transaction_kthread);
989 		if (TRANS_ABORTED(trans))
990 			err = trans->aborted;
991 		else
992 			err = -EROFS;
993 	}
994 
995 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
996 	return err;
997 }
998 
btrfs_end_transaction(struct btrfs_trans_handle * trans)999 int btrfs_end_transaction(struct btrfs_trans_handle *trans)
1000 {
1001 	return __btrfs_end_transaction(trans, 0);
1002 }
1003 
btrfs_end_transaction_throttle(struct btrfs_trans_handle * trans)1004 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans)
1005 {
1006 	return __btrfs_end_transaction(trans, 1);
1007 }
1008 
1009 /*
1010  * when btree blocks are allocated, they have some corresponding bits set for
1011  * them in one of two extent_io trees.  This is used to make sure all of
1012  * those extents are sent to disk but does not wait on them
1013  */
btrfs_write_marked_extents(struct btrfs_fs_info * fs_info,struct extent_io_tree * dirty_pages,int mark)1014 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
1015 			       struct extent_io_tree *dirty_pages, int mark)
1016 {
1017 	int err = 0;
1018 	int werr = 0;
1019 	struct address_space *mapping = fs_info->btree_inode->i_mapping;
1020 	struct extent_state *cached_state = NULL;
1021 	u64 start = 0;
1022 	u64 end;
1023 
1024 	atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers);
1025 	while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1026 				      mark, &cached_state)) {
1027 		bool wait_writeback = false;
1028 
1029 		err = convert_extent_bit(dirty_pages, start, end,
1030 					 EXTENT_NEED_WAIT,
1031 					 mark, &cached_state);
1032 		/*
1033 		 * convert_extent_bit can return -ENOMEM, which is most of the
1034 		 * time a temporary error. So when it happens, ignore the error
1035 		 * and wait for writeback of this range to finish - because we
1036 		 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
1037 		 * to __btrfs_wait_marked_extents() would not know that
1038 		 * writeback for this range started and therefore wouldn't
1039 		 * wait for it to finish - we don't want to commit a
1040 		 * superblock that points to btree nodes/leafs for which
1041 		 * writeback hasn't finished yet (and without errors).
1042 		 * We cleanup any entries left in the io tree when committing
1043 		 * the transaction (through extent_io_tree_release()).
1044 		 */
1045 		if (err == -ENOMEM) {
1046 			err = 0;
1047 			wait_writeback = true;
1048 		}
1049 		if (!err)
1050 			err = filemap_fdatawrite_range(mapping, start, end);
1051 		if (err)
1052 			werr = err;
1053 		else if (wait_writeback)
1054 			werr = filemap_fdatawait_range(mapping, start, end);
1055 		free_extent_state(cached_state);
1056 		cached_state = NULL;
1057 		cond_resched();
1058 		start = end + 1;
1059 	}
1060 	atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers);
1061 	return werr;
1062 }
1063 
1064 /*
1065  * when btree blocks are allocated, they have some corresponding bits set for
1066  * them in one of two extent_io trees.  This is used to make sure all of
1067  * those extents are on disk for transaction or log commit.  We wait
1068  * on all the pages and clear them from the dirty pages state tree
1069  */
__btrfs_wait_marked_extents(struct btrfs_fs_info * fs_info,struct extent_io_tree * dirty_pages)1070 static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info,
1071 				       struct extent_io_tree *dirty_pages)
1072 {
1073 	int err = 0;
1074 	int werr = 0;
1075 	struct address_space *mapping = fs_info->btree_inode->i_mapping;
1076 	struct extent_state *cached_state = NULL;
1077 	u64 start = 0;
1078 	u64 end;
1079 
1080 	while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1081 				      EXTENT_NEED_WAIT, &cached_state)) {
1082 		/*
1083 		 * Ignore -ENOMEM errors returned by clear_extent_bit().
1084 		 * When committing the transaction, we'll remove any entries
1085 		 * left in the io tree. For a log commit, we don't remove them
1086 		 * after committing the log because the tree can be accessed
1087 		 * concurrently - we do it only at transaction commit time when
1088 		 * it's safe to do it (through extent_io_tree_release()).
1089 		 */
1090 		err = clear_extent_bit(dirty_pages, start, end,
1091 				       EXTENT_NEED_WAIT, 0, 0, &cached_state);
1092 		if (err == -ENOMEM)
1093 			err = 0;
1094 		if (!err)
1095 			err = filemap_fdatawait_range(mapping, start, end);
1096 		if (err)
1097 			werr = err;
1098 		free_extent_state(cached_state);
1099 		cached_state = NULL;
1100 		cond_resched();
1101 		start = end + 1;
1102 	}
1103 	if (err)
1104 		werr = err;
1105 	return werr;
1106 }
1107 
btrfs_wait_extents(struct btrfs_fs_info * fs_info,struct extent_io_tree * dirty_pages)1108 static int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
1109 		       struct extent_io_tree *dirty_pages)
1110 {
1111 	bool errors = false;
1112 	int err;
1113 
1114 	err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1115 	if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags))
1116 		errors = true;
1117 
1118 	if (errors && !err)
1119 		err = -EIO;
1120 	return err;
1121 }
1122 
btrfs_wait_tree_log_extents(struct btrfs_root * log_root,int mark)1123 int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark)
1124 {
1125 	struct btrfs_fs_info *fs_info = log_root->fs_info;
1126 	struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages;
1127 	bool errors = false;
1128 	int err;
1129 
1130 	ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
1131 
1132 	err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1133 	if ((mark & EXTENT_DIRTY) &&
1134 	    test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags))
1135 		errors = true;
1136 
1137 	if ((mark & EXTENT_NEW) &&
1138 	    test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags))
1139 		errors = true;
1140 
1141 	if (errors && !err)
1142 		err = -EIO;
1143 	return err;
1144 }
1145 
1146 /*
1147  * When btree blocks are allocated the corresponding extents are marked dirty.
1148  * This function ensures such extents are persisted on disk for transaction or
1149  * log commit.
1150  *
1151  * @trans: transaction whose dirty pages we'd like to write
1152  */
btrfs_write_and_wait_transaction(struct btrfs_trans_handle * trans)1153 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans)
1154 {
1155 	int ret;
1156 	int ret2;
1157 	struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages;
1158 	struct btrfs_fs_info *fs_info = trans->fs_info;
1159 	struct blk_plug plug;
1160 
1161 	blk_start_plug(&plug);
1162 	ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY);
1163 	blk_finish_plug(&plug);
1164 	ret2 = btrfs_wait_extents(fs_info, dirty_pages);
1165 
1166 	extent_io_tree_release(&trans->transaction->dirty_pages);
1167 
1168 	if (ret)
1169 		return ret;
1170 	else if (ret2)
1171 		return ret2;
1172 	else
1173 		return 0;
1174 }
1175 
1176 /*
1177  * this is used to update the root pointer in the tree of tree roots.
1178  *
1179  * But, in the case of the extent allocation tree, updating the root
1180  * pointer may allocate blocks which may change the root of the extent
1181  * allocation tree.
1182  *
1183  * So, this loops and repeats and makes sure the cowonly root didn't
1184  * change while the root pointer was being updated in the metadata.
1185  */
update_cowonly_root(struct btrfs_trans_handle * trans,struct btrfs_root * root)1186 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1187 			       struct btrfs_root *root)
1188 {
1189 	int ret;
1190 	u64 old_root_bytenr;
1191 	u64 old_root_used;
1192 	struct btrfs_fs_info *fs_info = root->fs_info;
1193 	struct btrfs_root *tree_root = fs_info->tree_root;
1194 
1195 	old_root_used = btrfs_root_used(&root->root_item);
1196 
1197 	while (1) {
1198 		old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1199 		if (old_root_bytenr == root->node->start &&
1200 		    old_root_used == btrfs_root_used(&root->root_item))
1201 			break;
1202 
1203 		btrfs_set_root_node(&root->root_item, root->node);
1204 		ret = btrfs_update_root(trans, tree_root,
1205 					&root->root_key,
1206 					&root->root_item);
1207 		if (ret)
1208 			return ret;
1209 
1210 		old_root_used = btrfs_root_used(&root->root_item);
1211 	}
1212 
1213 	return 0;
1214 }
1215 
1216 /*
1217  * update all the cowonly tree roots on disk
1218  *
1219  * The error handling in this function may not be obvious. Any of the
1220  * failures will cause the file system to go offline. We still need
1221  * to clean up the delayed refs.
1222  */
commit_cowonly_roots(struct btrfs_trans_handle * trans)1223 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans)
1224 {
1225 	struct btrfs_fs_info *fs_info = trans->fs_info;
1226 	struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1227 	struct list_head *io_bgs = &trans->transaction->io_bgs;
1228 	struct list_head *next;
1229 	struct extent_buffer *eb;
1230 	int ret;
1231 
1232 	eb = btrfs_lock_root_node(fs_info->tree_root);
1233 	ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1234 			      0, &eb, BTRFS_NESTING_COW);
1235 	btrfs_tree_unlock(eb);
1236 	free_extent_buffer(eb);
1237 
1238 	if (ret)
1239 		return ret;
1240 
1241 	ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1242 	if (ret)
1243 		return ret;
1244 
1245 	ret = btrfs_run_dev_stats(trans);
1246 	if (ret)
1247 		return ret;
1248 	ret = btrfs_run_dev_replace(trans);
1249 	if (ret)
1250 		return ret;
1251 	ret = btrfs_run_qgroups(trans);
1252 	if (ret)
1253 		return ret;
1254 
1255 	ret = btrfs_setup_space_cache(trans);
1256 	if (ret)
1257 		return ret;
1258 
1259 	/* run_qgroups might have added some more refs */
1260 	ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1261 	if (ret)
1262 		return ret;
1263 again:
1264 	while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1265 		struct btrfs_root *root;
1266 		next = fs_info->dirty_cowonly_roots.next;
1267 		list_del_init(next);
1268 		root = list_entry(next, struct btrfs_root, dirty_list);
1269 		clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1270 
1271 		if (root != fs_info->extent_root)
1272 			list_add_tail(&root->dirty_list,
1273 				      &trans->transaction->switch_commits);
1274 		ret = update_cowonly_root(trans, root);
1275 		if (ret)
1276 			return ret;
1277 		ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1278 		if (ret)
1279 			return ret;
1280 	}
1281 
1282 	while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1283 		ret = btrfs_write_dirty_block_groups(trans);
1284 		if (ret)
1285 			return ret;
1286 		ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1287 		if (ret)
1288 			return ret;
1289 	}
1290 
1291 	if (!list_empty(&fs_info->dirty_cowonly_roots))
1292 		goto again;
1293 
1294 	list_add_tail(&fs_info->extent_root->dirty_list,
1295 		      &trans->transaction->switch_commits);
1296 
1297 	/* Update dev-replace pointer once everything is committed */
1298 	fs_info->dev_replace.committed_cursor_left =
1299 		fs_info->dev_replace.cursor_left_last_write_of_item;
1300 
1301 	return 0;
1302 }
1303 
1304 /*
1305  * dead roots are old snapshots that need to be deleted.  This allocates
1306  * a dirty root struct and adds it into the list of dead roots that need to
1307  * be deleted
1308  */
btrfs_add_dead_root(struct btrfs_root * root)1309 void btrfs_add_dead_root(struct btrfs_root *root)
1310 {
1311 	struct btrfs_fs_info *fs_info = root->fs_info;
1312 
1313 	spin_lock(&fs_info->trans_lock);
1314 	if (list_empty(&root->root_list)) {
1315 		btrfs_grab_root(root);
1316 		list_add_tail(&root->root_list, &fs_info->dead_roots);
1317 	}
1318 	spin_unlock(&fs_info->trans_lock);
1319 }
1320 
1321 /*
1322  * update all the cowonly tree roots on disk
1323  */
commit_fs_roots(struct btrfs_trans_handle * trans)1324 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
1325 {
1326 	struct btrfs_fs_info *fs_info = trans->fs_info;
1327 	struct btrfs_root *gang[8];
1328 	int i;
1329 	int ret;
1330 
1331 	spin_lock(&fs_info->fs_roots_radix_lock);
1332 	while (1) {
1333 		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1334 						 (void **)gang, 0,
1335 						 ARRAY_SIZE(gang),
1336 						 BTRFS_ROOT_TRANS_TAG);
1337 		if (ret == 0)
1338 			break;
1339 		for (i = 0; i < ret; i++) {
1340 			struct btrfs_root *root = gang[i];
1341 			int ret2;
1342 
1343 			radix_tree_tag_clear(&fs_info->fs_roots_radix,
1344 					(unsigned long)root->root_key.objectid,
1345 					BTRFS_ROOT_TRANS_TAG);
1346 			spin_unlock(&fs_info->fs_roots_radix_lock);
1347 
1348 			btrfs_free_log(trans, root);
1349 			btrfs_update_reloc_root(trans, root);
1350 
1351 			btrfs_save_ino_cache(root, trans);
1352 
1353 			/* see comments in should_cow_block() */
1354 			clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1355 			smp_mb__after_atomic();
1356 
1357 			if (root->commit_root != root->node) {
1358 				list_add_tail(&root->dirty_list,
1359 					&trans->transaction->switch_commits);
1360 				btrfs_set_root_node(&root->root_item,
1361 						    root->node);
1362 			}
1363 
1364 			ret2 = btrfs_update_root(trans, fs_info->tree_root,
1365 						&root->root_key,
1366 						&root->root_item);
1367 			if (ret2)
1368 				return ret2;
1369 			spin_lock(&fs_info->fs_roots_radix_lock);
1370 			btrfs_qgroup_free_meta_all_pertrans(root);
1371 		}
1372 	}
1373 	spin_unlock(&fs_info->fs_roots_radix_lock);
1374 	return 0;
1375 }
1376 
1377 /*
1378  * defrag a given btree.
1379  * Every leaf in the btree is read and defragged.
1380  */
btrfs_defrag_root(struct btrfs_root * root)1381 int btrfs_defrag_root(struct btrfs_root *root)
1382 {
1383 	struct btrfs_fs_info *info = root->fs_info;
1384 	struct btrfs_trans_handle *trans;
1385 	int ret;
1386 
1387 	if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1388 		return 0;
1389 
1390 	while (1) {
1391 		trans = btrfs_start_transaction(root, 0);
1392 		if (IS_ERR(trans)) {
1393 			ret = PTR_ERR(trans);
1394 			break;
1395 		}
1396 
1397 		ret = btrfs_defrag_leaves(trans, root);
1398 
1399 		btrfs_end_transaction(trans);
1400 		btrfs_btree_balance_dirty(info);
1401 		cond_resched();
1402 
1403 		if (btrfs_fs_closing(info) || ret != -EAGAIN)
1404 			break;
1405 
1406 		if (btrfs_defrag_cancelled(info)) {
1407 			btrfs_debug(info, "defrag_root cancelled");
1408 			ret = -EAGAIN;
1409 			break;
1410 		}
1411 	}
1412 	clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1413 	return ret;
1414 }
1415 
1416 /*
1417  * Do all special snapshot related qgroup dirty hack.
1418  *
1419  * Will do all needed qgroup inherit and dirty hack like switch commit
1420  * roots inside one transaction and write all btree into disk, to make
1421  * qgroup works.
1422  */
qgroup_account_snapshot(struct btrfs_trans_handle * trans,struct btrfs_root * src,struct btrfs_root * parent,struct btrfs_qgroup_inherit * inherit,u64 dst_objectid)1423 static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1424 				   struct btrfs_root *src,
1425 				   struct btrfs_root *parent,
1426 				   struct btrfs_qgroup_inherit *inherit,
1427 				   u64 dst_objectid)
1428 {
1429 	struct btrfs_fs_info *fs_info = src->fs_info;
1430 	int ret;
1431 
1432 	/*
1433 	 * Save some performance in the case that qgroups are not
1434 	 * enabled. If this check races with the ioctl, rescan will
1435 	 * kick in anyway.
1436 	 */
1437 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1438 		return 0;
1439 
1440 	/*
1441 	 * Ensure dirty @src will be committed.  Or, after coming
1442 	 * commit_fs_roots() and switch_commit_roots(), any dirty but not
1443 	 * recorded root will never be updated again, causing an outdated root
1444 	 * item.
1445 	 */
1446 	record_root_in_trans(trans, src, 1);
1447 
1448 	/*
1449 	 * We are going to commit transaction, see btrfs_commit_transaction()
1450 	 * comment for reason locking tree_log_mutex
1451 	 */
1452 	mutex_lock(&fs_info->tree_log_mutex);
1453 
1454 	ret = commit_fs_roots(trans);
1455 	if (ret)
1456 		goto out;
1457 	ret = btrfs_qgroup_account_extents(trans);
1458 	if (ret < 0)
1459 		goto out;
1460 
1461 	/* Now qgroup are all updated, we can inherit it to new qgroups */
1462 	ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid,
1463 				   inherit);
1464 	if (ret < 0)
1465 		goto out;
1466 
1467 	/*
1468 	 * Now we do a simplified commit transaction, which will:
1469 	 * 1) commit all subvolume and extent tree
1470 	 *    To ensure all subvolume and extent tree have a valid
1471 	 *    commit_root to accounting later insert_dir_item()
1472 	 * 2) write all btree blocks onto disk
1473 	 *    This is to make sure later btree modification will be cowed
1474 	 *    Or commit_root can be populated and cause wrong qgroup numbers
1475 	 * In this simplified commit, we don't really care about other trees
1476 	 * like chunk and root tree, as they won't affect qgroup.
1477 	 * And we don't write super to avoid half committed status.
1478 	 */
1479 	ret = commit_cowonly_roots(trans);
1480 	if (ret)
1481 		goto out;
1482 	switch_commit_roots(trans);
1483 	ret = btrfs_write_and_wait_transaction(trans);
1484 	if (ret)
1485 		btrfs_handle_fs_error(fs_info, ret,
1486 			"Error while writing out transaction for qgroup");
1487 
1488 out:
1489 	mutex_unlock(&fs_info->tree_log_mutex);
1490 
1491 	/*
1492 	 * Force parent root to be updated, as we recorded it before so its
1493 	 * last_trans == cur_transid.
1494 	 * Or it won't be committed again onto disk after later
1495 	 * insert_dir_item()
1496 	 */
1497 	if (!ret)
1498 		record_root_in_trans(trans, parent, 1);
1499 	return ret;
1500 }
1501 
1502 /*
1503  * new snapshots need to be created at a very specific time in the
1504  * transaction commit.  This does the actual creation.
1505  *
1506  * Note:
1507  * If the error which may affect the commitment of the current transaction
1508  * happens, we should return the error number. If the error which just affect
1509  * the creation of the pending snapshots, just return 0.
1510  */
create_pending_snapshot(struct btrfs_trans_handle * trans,struct btrfs_pending_snapshot * pending)1511 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1512 				   struct btrfs_pending_snapshot *pending)
1513 {
1514 
1515 	struct btrfs_fs_info *fs_info = trans->fs_info;
1516 	struct btrfs_key key;
1517 	struct btrfs_root_item *new_root_item;
1518 	struct btrfs_root *tree_root = fs_info->tree_root;
1519 	struct btrfs_root *root = pending->root;
1520 	struct btrfs_root *parent_root;
1521 	struct btrfs_block_rsv *rsv;
1522 	struct inode *parent_inode;
1523 	struct btrfs_path *path;
1524 	struct btrfs_dir_item *dir_item;
1525 	struct dentry *dentry;
1526 	struct extent_buffer *tmp;
1527 	struct extent_buffer *old;
1528 	struct timespec64 cur_time;
1529 	int ret = 0;
1530 	u64 to_reserve = 0;
1531 	u64 index = 0;
1532 	u64 objectid;
1533 	u64 root_flags;
1534 
1535 	ASSERT(pending->path);
1536 	path = pending->path;
1537 
1538 	ASSERT(pending->root_item);
1539 	new_root_item = pending->root_item;
1540 
1541 	pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1542 	if (pending->error)
1543 		goto no_free_objectid;
1544 
1545 	/*
1546 	 * Make qgroup to skip current new snapshot's qgroupid, as it is
1547 	 * accounted by later btrfs_qgroup_inherit().
1548 	 */
1549 	btrfs_set_skip_qgroup(trans, objectid);
1550 
1551 	btrfs_reloc_pre_snapshot(pending, &to_reserve);
1552 
1553 	if (to_reserve > 0) {
1554 		pending->error = btrfs_block_rsv_add(root,
1555 						     &pending->block_rsv,
1556 						     to_reserve,
1557 						     BTRFS_RESERVE_NO_FLUSH);
1558 		if (pending->error)
1559 			goto clear_skip_qgroup;
1560 	}
1561 
1562 	key.objectid = objectid;
1563 	key.offset = (u64)-1;
1564 	key.type = BTRFS_ROOT_ITEM_KEY;
1565 
1566 	rsv = trans->block_rsv;
1567 	trans->block_rsv = &pending->block_rsv;
1568 	trans->bytes_reserved = trans->block_rsv->reserved;
1569 	trace_btrfs_space_reservation(fs_info, "transaction",
1570 				      trans->transid,
1571 				      trans->bytes_reserved, 1);
1572 	dentry = pending->dentry;
1573 	parent_inode = pending->dir;
1574 	parent_root = BTRFS_I(parent_inode)->root;
1575 	record_root_in_trans(trans, parent_root, 0);
1576 
1577 	cur_time = current_time(parent_inode);
1578 
1579 	/*
1580 	 * insert the directory item
1581 	 */
1582 	ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index);
1583 	BUG_ON(ret); /* -ENOMEM */
1584 
1585 	/* check if there is a file/dir which has the same name. */
1586 	dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1587 					 btrfs_ino(BTRFS_I(parent_inode)),
1588 					 dentry->d_name.name,
1589 					 dentry->d_name.len, 0);
1590 	if (dir_item != NULL && !IS_ERR(dir_item)) {
1591 		pending->error = -EEXIST;
1592 		goto dir_item_existed;
1593 	} else if (IS_ERR(dir_item)) {
1594 		ret = PTR_ERR(dir_item);
1595 		btrfs_abort_transaction(trans, ret);
1596 		goto fail;
1597 	}
1598 	btrfs_release_path(path);
1599 
1600 	/*
1601 	 * pull in the delayed directory update
1602 	 * and the delayed inode item
1603 	 * otherwise we corrupt the FS during
1604 	 * snapshot
1605 	 */
1606 	ret = btrfs_run_delayed_items(trans);
1607 	if (ret) {	/* Transaction aborted */
1608 		btrfs_abort_transaction(trans, ret);
1609 		goto fail;
1610 	}
1611 
1612 	record_root_in_trans(trans, root, 0);
1613 	btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1614 	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1615 	btrfs_check_and_init_root_item(new_root_item);
1616 
1617 	root_flags = btrfs_root_flags(new_root_item);
1618 	if (pending->readonly)
1619 		root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1620 	else
1621 		root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1622 	btrfs_set_root_flags(new_root_item, root_flags);
1623 
1624 	btrfs_set_root_generation_v2(new_root_item,
1625 			trans->transid);
1626 	generate_random_guid(new_root_item->uuid);
1627 	memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1628 			BTRFS_UUID_SIZE);
1629 	if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1630 		memset(new_root_item->received_uuid, 0,
1631 		       sizeof(new_root_item->received_uuid));
1632 		memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1633 		memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1634 		btrfs_set_root_stransid(new_root_item, 0);
1635 		btrfs_set_root_rtransid(new_root_item, 0);
1636 	}
1637 	btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1638 	btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1639 	btrfs_set_root_otransid(new_root_item, trans->transid);
1640 
1641 	old = btrfs_lock_root_node(root);
1642 	ret = btrfs_cow_block(trans, root, old, NULL, 0, &old,
1643 			      BTRFS_NESTING_COW);
1644 	if (ret) {
1645 		btrfs_tree_unlock(old);
1646 		free_extent_buffer(old);
1647 		btrfs_abort_transaction(trans, ret);
1648 		goto fail;
1649 	}
1650 
1651 	btrfs_set_lock_blocking_write(old);
1652 
1653 	ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1654 	/* clean up in any case */
1655 	btrfs_tree_unlock(old);
1656 	free_extent_buffer(old);
1657 	if (ret) {
1658 		btrfs_abort_transaction(trans, ret);
1659 		goto fail;
1660 	}
1661 	/* see comments in should_cow_block() */
1662 	set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1663 	smp_wmb();
1664 
1665 	btrfs_set_root_node(new_root_item, tmp);
1666 	/* record when the snapshot was created in key.offset */
1667 	key.offset = trans->transid;
1668 	ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1669 	btrfs_tree_unlock(tmp);
1670 	free_extent_buffer(tmp);
1671 	if (ret) {
1672 		btrfs_abort_transaction(trans, ret);
1673 		goto fail;
1674 	}
1675 
1676 	/*
1677 	 * insert root back/forward references
1678 	 */
1679 	ret = btrfs_add_root_ref(trans, objectid,
1680 				 parent_root->root_key.objectid,
1681 				 btrfs_ino(BTRFS_I(parent_inode)), index,
1682 				 dentry->d_name.name, dentry->d_name.len);
1683 	if (ret) {
1684 		btrfs_abort_transaction(trans, ret);
1685 		goto fail;
1686 	}
1687 
1688 	key.offset = (u64)-1;
1689 	pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev);
1690 	if (IS_ERR(pending->snap)) {
1691 		ret = PTR_ERR(pending->snap);
1692 		pending->snap = NULL;
1693 		btrfs_abort_transaction(trans, ret);
1694 		goto fail;
1695 	}
1696 
1697 	ret = btrfs_reloc_post_snapshot(trans, pending);
1698 	if (ret) {
1699 		btrfs_abort_transaction(trans, ret);
1700 		goto fail;
1701 	}
1702 
1703 	ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1704 	if (ret) {
1705 		btrfs_abort_transaction(trans, ret);
1706 		goto fail;
1707 	}
1708 
1709 	/*
1710 	 * Do special qgroup accounting for snapshot, as we do some qgroup
1711 	 * snapshot hack to do fast snapshot.
1712 	 * To co-operate with that hack, we do hack again.
1713 	 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1714 	 */
1715 	ret = qgroup_account_snapshot(trans, root, parent_root,
1716 				      pending->inherit, objectid);
1717 	if (ret < 0)
1718 		goto fail;
1719 
1720 	ret = btrfs_insert_dir_item(trans, dentry->d_name.name,
1721 				    dentry->d_name.len, BTRFS_I(parent_inode),
1722 				    &key, BTRFS_FT_DIR, index);
1723 	/* We have check then name at the beginning, so it is impossible. */
1724 	BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1725 	if (ret) {
1726 		btrfs_abort_transaction(trans, ret);
1727 		goto fail;
1728 	}
1729 
1730 	btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size +
1731 					 dentry->d_name.len * 2);
1732 	parent_inode->i_mtime = parent_inode->i_ctime =
1733 		current_time(parent_inode);
1734 	ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
1735 	if (ret) {
1736 		btrfs_abort_transaction(trans, ret);
1737 		goto fail;
1738 	}
1739 	ret = btrfs_uuid_tree_add(trans, new_root_item->uuid,
1740 				  BTRFS_UUID_KEY_SUBVOL,
1741 				  objectid);
1742 	if (ret) {
1743 		btrfs_abort_transaction(trans, ret);
1744 		goto fail;
1745 	}
1746 	if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1747 		ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid,
1748 					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1749 					  objectid);
1750 		if (ret && ret != -EEXIST) {
1751 			btrfs_abort_transaction(trans, ret);
1752 			goto fail;
1753 		}
1754 	}
1755 
1756 	ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1757 	if (ret) {
1758 		btrfs_abort_transaction(trans, ret);
1759 		goto fail;
1760 	}
1761 
1762 fail:
1763 	pending->error = ret;
1764 dir_item_existed:
1765 	trans->block_rsv = rsv;
1766 	trans->bytes_reserved = 0;
1767 clear_skip_qgroup:
1768 	btrfs_clear_skip_qgroup(trans);
1769 no_free_objectid:
1770 	kfree(new_root_item);
1771 	pending->root_item = NULL;
1772 	btrfs_free_path(path);
1773 	pending->path = NULL;
1774 
1775 	return ret;
1776 }
1777 
1778 /*
1779  * create all the snapshots we've scheduled for creation
1780  */
create_pending_snapshots(struct btrfs_trans_handle * trans)1781 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans)
1782 {
1783 	struct btrfs_pending_snapshot *pending, *next;
1784 	struct list_head *head = &trans->transaction->pending_snapshots;
1785 	int ret = 0;
1786 
1787 	list_for_each_entry_safe(pending, next, head, list) {
1788 		list_del(&pending->list);
1789 		ret = create_pending_snapshot(trans, pending);
1790 		if (ret)
1791 			break;
1792 	}
1793 	return ret;
1794 }
1795 
update_super_roots(struct btrfs_fs_info * fs_info)1796 static void update_super_roots(struct btrfs_fs_info *fs_info)
1797 {
1798 	struct btrfs_root_item *root_item;
1799 	struct btrfs_super_block *super;
1800 
1801 	super = fs_info->super_copy;
1802 
1803 	root_item = &fs_info->chunk_root->root_item;
1804 	super->chunk_root = root_item->bytenr;
1805 	super->chunk_root_generation = root_item->generation;
1806 	super->chunk_root_level = root_item->level;
1807 
1808 	root_item = &fs_info->tree_root->root_item;
1809 	super->root = root_item->bytenr;
1810 	super->generation = root_item->generation;
1811 	super->root_level = root_item->level;
1812 	if (btrfs_test_opt(fs_info, SPACE_CACHE))
1813 		super->cache_generation = root_item->generation;
1814 	if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags))
1815 		super->uuid_tree_generation = root_item->generation;
1816 }
1817 
btrfs_transaction_in_commit(struct btrfs_fs_info * info)1818 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1819 {
1820 	struct btrfs_transaction *trans;
1821 	int ret = 0;
1822 
1823 	spin_lock(&info->trans_lock);
1824 	trans = info->running_transaction;
1825 	if (trans)
1826 		ret = (trans->state >= TRANS_STATE_COMMIT_START);
1827 	spin_unlock(&info->trans_lock);
1828 	return ret;
1829 }
1830 
btrfs_transaction_blocked(struct btrfs_fs_info * info)1831 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1832 {
1833 	struct btrfs_transaction *trans;
1834 	int ret = 0;
1835 
1836 	spin_lock(&info->trans_lock);
1837 	trans = info->running_transaction;
1838 	if (trans)
1839 		ret = is_transaction_blocked(trans);
1840 	spin_unlock(&info->trans_lock);
1841 	return ret;
1842 }
1843 
1844 /*
1845  * wait for the current transaction commit to start and block subsequent
1846  * transaction joins
1847  */
wait_current_trans_commit_start(struct btrfs_fs_info * fs_info,struct btrfs_transaction * trans)1848 static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info,
1849 					    struct btrfs_transaction *trans)
1850 {
1851 	wait_event(fs_info->transaction_blocked_wait,
1852 		   trans->state >= TRANS_STATE_COMMIT_START ||
1853 		   TRANS_ABORTED(trans));
1854 }
1855 
1856 /*
1857  * wait for the current transaction to start and then become unblocked.
1858  * caller holds ref.
1859  */
wait_current_trans_commit_start_and_unblock(struct btrfs_fs_info * fs_info,struct btrfs_transaction * trans)1860 static void wait_current_trans_commit_start_and_unblock(
1861 					struct btrfs_fs_info *fs_info,
1862 					struct btrfs_transaction *trans)
1863 {
1864 	wait_event(fs_info->transaction_wait,
1865 		   trans->state >= TRANS_STATE_UNBLOCKED ||
1866 		   TRANS_ABORTED(trans));
1867 }
1868 
1869 /*
1870  * commit transactions asynchronously. once btrfs_commit_transaction_async
1871  * returns, any subsequent transaction will not be allowed to join.
1872  */
1873 struct btrfs_async_commit {
1874 	struct btrfs_trans_handle *newtrans;
1875 	struct work_struct work;
1876 };
1877 
do_async_commit(struct work_struct * work)1878 static void do_async_commit(struct work_struct *work)
1879 {
1880 	struct btrfs_async_commit *ac =
1881 		container_of(work, struct btrfs_async_commit, work);
1882 
1883 	/*
1884 	 * We've got freeze protection passed with the transaction.
1885 	 * Tell lockdep about it.
1886 	 */
1887 	if (ac->newtrans->type & __TRANS_FREEZABLE)
1888 		__sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS);
1889 
1890 	current->journal_info = ac->newtrans;
1891 
1892 	btrfs_commit_transaction(ac->newtrans);
1893 	kfree(ac);
1894 }
1895 
btrfs_commit_transaction_async(struct btrfs_trans_handle * trans,int wait_for_unblock)1896 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1897 				   int wait_for_unblock)
1898 {
1899 	struct btrfs_fs_info *fs_info = trans->fs_info;
1900 	struct btrfs_async_commit *ac;
1901 	struct btrfs_transaction *cur_trans;
1902 
1903 	ac = kmalloc(sizeof(*ac), GFP_NOFS);
1904 	if (!ac)
1905 		return -ENOMEM;
1906 
1907 	INIT_WORK(&ac->work, do_async_commit);
1908 	ac->newtrans = btrfs_join_transaction(trans->root);
1909 	if (IS_ERR(ac->newtrans)) {
1910 		int err = PTR_ERR(ac->newtrans);
1911 		kfree(ac);
1912 		return err;
1913 	}
1914 
1915 	/* take transaction reference */
1916 	cur_trans = trans->transaction;
1917 	refcount_inc(&cur_trans->use_count);
1918 
1919 	btrfs_end_transaction(trans);
1920 
1921 	/*
1922 	 * Tell lockdep we've released the freeze rwsem, since the
1923 	 * async commit thread will be the one to unlock it.
1924 	 */
1925 	if (ac->newtrans->type & __TRANS_FREEZABLE)
1926 		__sb_writers_release(fs_info->sb, SB_FREEZE_FS);
1927 
1928 	schedule_work(&ac->work);
1929 
1930 	/* wait for transaction to start and unblock */
1931 	if (wait_for_unblock)
1932 		wait_current_trans_commit_start_and_unblock(fs_info, cur_trans);
1933 	else
1934 		wait_current_trans_commit_start(fs_info, cur_trans);
1935 
1936 	if (current->journal_info == trans)
1937 		current->journal_info = NULL;
1938 
1939 	btrfs_put_transaction(cur_trans);
1940 	return 0;
1941 }
1942 
1943 
cleanup_transaction(struct btrfs_trans_handle * trans,int err)1944 static void cleanup_transaction(struct btrfs_trans_handle *trans, int err)
1945 {
1946 	struct btrfs_fs_info *fs_info = trans->fs_info;
1947 	struct btrfs_transaction *cur_trans = trans->transaction;
1948 
1949 	WARN_ON(refcount_read(&trans->use_count) > 1);
1950 
1951 	btrfs_abort_transaction(trans, err);
1952 
1953 	spin_lock(&fs_info->trans_lock);
1954 
1955 	/*
1956 	 * If the transaction is removed from the list, it means this
1957 	 * transaction has been committed successfully, so it is impossible
1958 	 * to call the cleanup function.
1959 	 */
1960 	BUG_ON(list_empty(&cur_trans->list));
1961 
1962 	if (cur_trans == fs_info->running_transaction) {
1963 		cur_trans->state = TRANS_STATE_COMMIT_DOING;
1964 		spin_unlock(&fs_info->trans_lock);
1965 		wait_event(cur_trans->writer_wait,
1966 			   atomic_read(&cur_trans->num_writers) == 1);
1967 
1968 		spin_lock(&fs_info->trans_lock);
1969 	}
1970 
1971 	/*
1972 	 * Now that we know no one else is still using the transaction we can
1973 	 * remove the transaction from the list of transactions. This avoids
1974 	 * the transaction kthread from cleaning up the transaction while some
1975 	 * other task is still using it, which could result in a use-after-free
1976 	 * on things like log trees, as it forces the transaction kthread to
1977 	 * wait for this transaction to be cleaned up by us.
1978 	 */
1979 	list_del_init(&cur_trans->list);
1980 
1981 	spin_unlock(&fs_info->trans_lock);
1982 
1983 	btrfs_cleanup_one_transaction(trans->transaction, fs_info);
1984 
1985 	spin_lock(&fs_info->trans_lock);
1986 	if (cur_trans == fs_info->running_transaction)
1987 		fs_info->running_transaction = NULL;
1988 	spin_unlock(&fs_info->trans_lock);
1989 
1990 	if (trans->type & __TRANS_FREEZABLE)
1991 		sb_end_intwrite(fs_info->sb);
1992 	btrfs_put_transaction(cur_trans);
1993 	btrfs_put_transaction(cur_trans);
1994 
1995 	trace_btrfs_transaction_commit(trans->root);
1996 
1997 	if (current->journal_info == trans)
1998 		current->journal_info = NULL;
1999 	btrfs_scrub_cancel(fs_info);
2000 
2001 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
2002 }
2003 
2004 /*
2005  * Release reserved delayed ref space of all pending block groups of the
2006  * transaction and remove them from the list
2007  */
btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle * trans)2008 static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans)
2009 {
2010        struct btrfs_fs_info *fs_info = trans->fs_info;
2011        struct btrfs_block_group *block_group, *tmp;
2012 
2013        list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
2014                btrfs_delayed_refs_rsv_release(fs_info, 1);
2015                list_del_init(&block_group->bg_list);
2016        }
2017 }
2018 
btrfs_start_delalloc_flush(struct btrfs_trans_handle * trans)2019 static inline int btrfs_start_delalloc_flush(struct btrfs_trans_handle *trans)
2020 {
2021 	struct btrfs_fs_info *fs_info = trans->fs_info;
2022 
2023 	/*
2024 	 * We use writeback_inodes_sb here because if we used
2025 	 * btrfs_start_delalloc_roots we would deadlock with fs freeze.
2026 	 * Currently are holding the fs freeze lock, if we do an async flush
2027 	 * we'll do btrfs_join_transaction() and deadlock because we need to
2028 	 * wait for the fs freeze lock.  Using the direct flushing we benefit
2029 	 * from already being in a transaction and our join_transaction doesn't
2030 	 * have to re-take the fs freeze lock.
2031 	 */
2032 	if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) {
2033 		writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC);
2034 	} else {
2035 		struct btrfs_pending_snapshot *pending;
2036 		struct list_head *head = &trans->transaction->pending_snapshots;
2037 
2038 		/*
2039 		 * Flush dellaloc for any root that is going to be snapshotted.
2040 		 * This is done to avoid a corrupted version of files, in the
2041 		 * snapshots, that had both buffered and direct IO writes (even
2042 		 * if they were done sequentially) due to an unordered update of
2043 		 * the inode's size on disk.
2044 		 */
2045 		list_for_each_entry(pending, head, list) {
2046 			int ret;
2047 
2048 			ret = btrfs_start_delalloc_snapshot(pending->root);
2049 			if (ret)
2050 				return ret;
2051 		}
2052 	}
2053 	return 0;
2054 }
2055 
btrfs_wait_delalloc_flush(struct btrfs_trans_handle * trans)2056 static inline void btrfs_wait_delalloc_flush(struct btrfs_trans_handle *trans)
2057 {
2058 	struct btrfs_fs_info *fs_info = trans->fs_info;
2059 
2060 	if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) {
2061 		btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
2062 	} else {
2063 		struct btrfs_pending_snapshot *pending;
2064 		struct list_head *head = &trans->transaction->pending_snapshots;
2065 
2066 		/*
2067 		 * Wait for any dellaloc that we started previously for the roots
2068 		 * that are going to be snapshotted. This is to avoid a corrupted
2069 		 * version of files in the snapshots that had both buffered and
2070 		 * direct IO writes (even if they were done sequentially).
2071 		 */
2072 		list_for_each_entry(pending, head, list)
2073 			btrfs_wait_ordered_extents(pending->root,
2074 						   U64_MAX, 0, U64_MAX);
2075 	}
2076 }
2077 
btrfs_commit_transaction(struct btrfs_trans_handle * trans)2078 int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
2079 {
2080 	struct btrfs_fs_info *fs_info = trans->fs_info;
2081 	struct btrfs_transaction *cur_trans = trans->transaction;
2082 	struct btrfs_transaction *prev_trans = NULL;
2083 	int ret;
2084 
2085 	ASSERT(refcount_read(&trans->use_count) == 1);
2086 
2087 	/*
2088 	 * Some places just start a transaction to commit it.  We need to make
2089 	 * sure that if this commit fails that the abort code actually marks the
2090 	 * transaction as failed, so set trans->dirty to make the abort code do
2091 	 * the right thing.
2092 	 */
2093 	trans->dirty = true;
2094 
2095 	/* Stop the commit early if ->aborted is set */
2096 	if (TRANS_ABORTED(cur_trans)) {
2097 		ret = cur_trans->aborted;
2098 		btrfs_end_transaction(trans);
2099 		return ret;
2100 	}
2101 
2102 	btrfs_trans_release_metadata(trans);
2103 	trans->block_rsv = NULL;
2104 
2105 	/* make a pass through all the delayed refs we have so far
2106 	 * any runnings procs may add more while we are here
2107 	 */
2108 	ret = btrfs_run_delayed_refs(trans, 0);
2109 	if (ret) {
2110 		btrfs_end_transaction(trans);
2111 		return ret;
2112 	}
2113 
2114 	cur_trans = trans->transaction;
2115 
2116 	/*
2117 	 * set the flushing flag so procs in this transaction have to
2118 	 * start sending their work down.
2119 	 */
2120 	cur_trans->delayed_refs.flushing = 1;
2121 	smp_wmb();
2122 
2123 	btrfs_create_pending_block_groups(trans);
2124 
2125 	ret = btrfs_run_delayed_refs(trans, 0);
2126 	if (ret) {
2127 		btrfs_end_transaction(trans);
2128 		return ret;
2129 	}
2130 
2131 	if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
2132 		int run_it = 0;
2133 
2134 		/* this mutex is also taken before trying to set
2135 		 * block groups readonly.  We need to make sure
2136 		 * that nobody has set a block group readonly
2137 		 * after a extents from that block group have been
2138 		 * allocated for cache files.  btrfs_set_block_group_ro
2139 		 * will wait for the transaction to commit if it
2140 		 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
2141 		 *
2142 		 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
2143 		 * only one process starts all the block group IO.  It wouldn't
2144 		 * hurt to have more than one go through, but there's no
2145 		 * real advantage to it either.
2146 		 */
2147 		mutex_lock(&fs_info->ro_block_group_mutex);
2148 		if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
2149 				      &cur_trans->flags))
2150 			run_it = 1;
2151 		mutex_unlock(&fs_info->ro_block_group_mutex);
2152 
2153 		if (run_it) {
2154 			ret = btrfs_start_dirty_block_groups(trans);
2155 			if (ret) {
2156 				btrfs_end_transaction(trans);
2157 				return ret;
2158 			}
2159 		}
2160 	}
2161 
2162 	spin_lock(&fs_info->trans_lock);
2163 	if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
2164 		spin_unlock(&fs_info->trans_lock);
2165 		refcount_inc(&cur_trans->use_count);
2166 		ret = btrfs_end_transaction(trans);
2167 
2168 		wait_for_commit(cur_trans);
2169 
2170 		if (TRANS_ABORTED(cur_trans))
2171 			ret = cur_trans->aborted;
2172 
2173 		btrfs_put_transaction(cur_trans);
2174 
2175 		return ret;
2176 	}
2177 
2178 	cur_trans->state = TRANS_STATE_COMMIT_START;
2179 	wake_up(&fs_info->transaction_blocked_wait);
2180 
2181 	if (cur_trans->list.prev != &fs_info->trans_list) {
2182 		prev_trans = list_entry(cur_trans->list.prev,
2183 					struct btrfs_transaction, list);
2184 		if (prev_trans->state != TRANS_STATE_COMPLETED) {
2185 			refcount_inc(&prev_trans->use_count);
2186 			spin_unlock(&fs_info->trans_lock);
2187 
2188 			wait_for_commit(prev_trans);
2189 			ret = READ_ONCE(prev_trans->aborted);
2190 
2191 			btrfs_put_transaction(prev_trans);
2192 			if (ret)
2193 				goto cleanup_transaction;
2194 		} else {
2195 			spin_unlock(&fs_info->trans_lock);
2196 		}
2197 	} else {
2198 		spin_unlock(&fs_info->trans_lock);
2199 		/*
2200 		 * The previous transaction was aborted and was already removed
2201 		 * from the list of transactions at fs_info->trans_list. So we
2202 		 * abort to prevent writing a new superblock that reflects a
2203 		 * corrupt state (pointing to trees with unwritten nodes/leafs).
2204 		 */
2205 		if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) {
2206 			ret = -EROFS;
2207 			goto cleanup_transaction;
2208 		}
2209 	}
2210 
2211 	extwriter_counter_dec(cur_trans, trans->type);
2212 
2213 	ret = btrfs_start_delalloc_flush(trans);
2214 	if (ret)
2215 		goto cleanup_transaction;
2216 
2217 	ret = btrfs_run_delayed_items(trans);
2218 	if (ret)
2219 		goto cleanup_transaction;
2220 
2221 	wait_event(cur_trans->writer_wait,
2222 		   extwriter_counter_read(cur_trans) == 0);
2223 
2224 	/* some pending stuffs might be added after the previous flush. */
2225 	ret = btrfs_run_delayed_items(trans);
2226 	if (ret)
2227 		goto cleanup_transaction;
2228 
2229 	btrfs_wait_delalloc_flush(trans);
2230 
2231 	/*
2232 	 * Wait for all ordered extents started by a fast fsync that joined this
2233 	 * transaction. Otherwise if this transaction commits before the ordered
2234 	 * extents complete we lose logged data after a power failure.
2235 	 */
2236 	wait_event(cur_trans->pending_wait,
2237 		   atomic_read(&cur_trans->pending_ordered) == 0);
2238 
2239 	btrfs_scrub_pause(fs_info);
2240 	/*
2241 	 * Ok now we need to make sure to block out any other joins while we
2242 	 * commit the transaction.  We could have started a join before setting
2243 	 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2244 	 */
2245 	spin_lock(&fs_info->trans_lock);
2246 	cur_trans->state = TRANS_STATE_COMMIT_DOING;
2247 	spin_unlock(&fs_info->trans_lock);
2248 	wait_event(cur_trans->writer_wait,
2249 		   atomic_read(&cur_trans->num_writers) == 1);
2250 
2251 	if (TRANS_ABORTED(cur_trans)) {
2252 		ret = cur_trans->aborted;
2253 		goto scrub_continue;
2254 	}
2255 	/*
2256 	 * the reloc mutex makes sure that we stop
2257 	 * the balancing code from coming in and moving
2258 	 * extents around in the middle of the commit
2259 	 */
2260 	mutex_lock(&fs_info->reloc_mutex);
2261 
2262 	/*
2263 	 * We needn't worry about the delayed items because we will
2264 	 * deal with them in create_pending_snapshot(), which is the
2265 	 * core function of the snapshot creation.
2266 	 */
2267 	ret = create_pending_snapshots(trans);
2268 	if (ret)
2269 		goto unlock_reloc;
2270 
2271 	/*
2272 	 * We insert the dir indexes of the snapshots and update the inode
2273 	 * of the snapshots' parents after the snapshot creation, so there
2274 	 * are some delayed items which are not dealt with. Now deal with
2275 	 * them.
2276 	 *
2277 	 * We needn't worry that this operation will corrupt the snapshots,
2278 	 * because all the tree which are snapshoted will be forced to COW
2279 	 * the nodes and leaves.
2280 	 */
2281 	ret = btrfs_run_delayed_items(trans);
2282 	if (ret)
2283 		goto unlock_reloc;
2284 
2285 	ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
2286 	if (ret)
2287 		goto unlock_reloc;
2288 
2289 	/*
2290 	 * make sure none of the code above managed to slip in a
2291 	 * delayed item
2292 	 */
2293 	btrfs_assert_delayed_root_empty(fs_info);
2294 
2295 	WARN_ON(cur_trans != trans->transaction);
2296 
2297 	/* btrfs_commit_tree_roots is responsible for getting the
2298 	 * various roots consistent with each other.  Every pointer
2299 	 * in the tree of tree roots has to point to the most up to date
2300 	 * root for every subvolume and other tree.  So, we have to keep
2301 	 * the tree logging code from jumping in and changing any
2302 	 * of the trees.
2303 	 *
2304 	 * At this point in the commit, there can't be any tree-log
2305 	 * writers, but a little lower down we drop the trans mutex
2306 	 * and let new people in.  By holding the tree_log_mutex
2307 	 * from now until after the super is written, we avoid races
2308 	 * with the tree-log code.
2309 	 */
2310 	mutex_lock(&fs_info->tree_log_mutex);
2311 
2312 	ret = commit_fs_roots(trans);
2313 	if (ret)
2314 		goto unlock_tree_log;
2315 
2316 	/*
2317 	 * Since the transaction is done, we can apply the pending changes
2318 	 * before the next transaction.
2319 	 */
2320 	btrfs_apply_pending_changes(fs_info);
2321 
2322 	/* commit_fs_roots gets rid of all the tree log roots, it is now
2323 	 * safe to free the root of tree log roots
2324 	 */
2325 	btrfs_free_log_root_tree(trans, fs_info);
2326 
2327 	/*
2328 	 * commit_fs_roots() can call btrfs_save_ino_cache(), which generates
2329 	 * new delayed refs. Must handle them or qgroup can be wrong.
2330 	 */
2331 	ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
2332 	if (ret)
2333 		goto unlock_tree_log;
2334 
2335 	/*
2336 	 * Since fs roots are all committed, we can get a quite accurate
2337 	 * new_roots. So let's do quota accounting.
2338 	 */
2339 	ret = btrfs_qgroup_account_extents(trans);
2340 	if (ret < 0)
2341 		goto unlock_tree_log;
2342 
2343 	ret = commit_cowonly_roots(trans);
2344 	if (ret)
2345 		goto unlock_tree_log;
2346 
2347 	/*
2348 	 * The tasks which save the space cache and inode cache may also
2349 	 * update ->aborted, check it.
2350 	 */
2351 	if (TRANS_ABORTED(cur_trans)) {
2352 		ret = cur_trans->aborted;
2353 		goto unlock_tree_log;
2354 	}
2355 
2356 	cur_trans = fs_info->running_transaction;
2357 
2358 	btrfs_set_root_node(&fs_info->tree_root->root_item,
2359 			    fs_info->tree_root->node);
2360 	list_add_tail(&fs_info->tree_root->dirty_list,
2361 		      &cur_trans->switch_commits);
2362 
2363 	btrfs_set_root_node(&fs_info->chunk_root->root_item,
2364 			    fs_info->chunk_root->node);
2365 	list_add_tail(&fs_info->chunk_root->dirty_list,
2366 		      &cur_trans->switch_commits);
2367 
2368 	switch_commit_roots(trans);
2369 
2370 	ASSERT(list_empty(&cur_trans->dirty_bgs));
2371 	ASSERT(list_empty(&cur_trans->io_bgs));
2372 	update_super_roots(fs_info);
2373 
2374 	btrfs_set_super_log_root(fs_info->super_copy, 0);
2375 	btrfs_set_super_log_root_level(fs_info->super_copy, 0);
2376 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
2377 	       sizeof(*fs_info->super_copy));
2378 
2379 	btrfs_commit_device_sizes(cur_trans);
2380 
2381 	clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
2382 	clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
2383 
2384 	btrfs_trans_release_chunk_metadata(trans);
2385 
2386 	spin_lock(&fs_info->trans_lock);
2387 	cur_trans->state = TRANS_STATE_UNBLOCKED;
2388 	fs_info->running_transaction = NULL;
2389 	spin_unlock(&fs_info->trans_lock);
2390 	mutex_unlock(&fs_info->reloc_mutex);
2391 
2392 	wake_up(&fs_info->transaction_wait);
2393 
2394 	ret = btrfs_write_and_wait_transaction(trans);
2395 	if (ret) {
2396 		btrfs_handle_fs_error(fs_info, ret,
2397 				      "Error while writing out transaction");
2398 		/*
2399 		 * reloc_mutex has been unlocked, tree_log_mutex is still held
2400 		 * but we can't jump to unlock_tree_log causing double unlock
2401 		 */
2402 		mutex_unlock(&fs_info->tree_log_mutex);
2403 		goto scrub_continue;
2404 	}
2405 
2406 	ret = write_all_supers(fs_info, 0);
2407 	/*
2408 	 * the super is written, we can safely allow the tree-loggers
2409 	 * to go about their business
2410 	 */
2411 	mutex_unlock(&fs_info->tree_log_mutex);
2412 	if (ret)
2413 		goto scrub_continue;
2414 
2415 	btrfs_finish_extent_commit(trans);
2416 
2417 	if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
2418 		btrfs_clear_space_info_full(fs_info);
2419 
2420 	fs_info->last_trans_committed = cur_trans->transid;
2421 	/*
2422 	 * We needn't acquire the lock here because there is no other task
2423 	 * which can change it.
2424 	 */
2425 	cur_trans->state = TRANS_STATE_COMPLETED;
2426 	wake_up(&cur_trans->commit_wait);
2427 
2428 	spin_lock(&fs_info->trans_lock);
2429 	list_del_init(&cur_trans->list);
2430 	spin_unlock(&fs_info->trans_lock);
2431 
2432 	btrfs_put_transaction(cur_trans);
2433 	btrfs_put_transaction(cur_trans);
2434 
2435 	if (trans->type & __TRANS_FREEZABLE)
2436 		sb_end_intwrite(fs_info->sb);
2437 
2438 	trace_btrfs_transaction_commit(trans->root);
2439 
2440 	btrfs_scrub_continue(fs_info);
2441 
2442 	if (current->journal_info == trans)
2443 		current->journal_info = NULL;
2444 
2445 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
2446 
2447 	return ret;
2448 
2449 unlock_tree_log:
2450 	mutex_unlock(&fs_info->tree_log_mutex);
2451 unlock_reloc:
2452 	mutex_unlock(&fs_info->reloc_mutex);
2453 scrub_continue:
2454 	btrfs_scrub_continue(fs_info);
2455 cleanup_transaction:
2456 	btrfs_trans_release_metadata(trans);
2457 	btrfs_cleanup_pending_block_groups(trans);
2458 	btrfs_trans_release_chunk_metadata(trans);
2459 	trans->block_rsv = NULL;
2460 	btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
2461 	if (current->journal_info == trans)
2462 		current->journal_info = NULL;
2463 	cleanup_transaction(trans, ret);
2464 
2465 	return ret;
2466 }
2467 
2468 /*
2469  * return < 0 if error
2470  * 0 if there are no more dead_roots at the time of call
2471  * 1 there are more to be processed, call me again
2472  *
2473  * The return value indicates there are certainly more snapshots to delete, but
2474  * if there comes a new one during processing, it may return 0. We don't mind,
2475  * because btrfs_commit_super will poke cleaner thread and it will process it a
2476  * few seconds later.
2477  */
btrfs_clean_one_deleted_snapshot(struct btrfs_root * root)2478 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
2479 {
2480 	int ret;
2481 	struct btrfs_fs_info *fs_info = root->fs_info;
2482 
2483 	spin_lock(&fs_info->trans_lock);
2484 	if (list_empty(&fs_info->dead_roots)) {
2485 		spin_unlock(&fs_info->trans_lock);
2486 		return 0;
2487 	}
2488 	root = list_first_entry(&fs_info->dead_roots,
2489 			struct btrfs_root, root_list);
2490 	list_del_init(&root->root_list);
2491 	spin_unlock(&fs_info->trans_lock);
2492 
2493 	btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid);
2494 
2495 	btrfs_kill_all_delayed_nodes(root);
2496 	if (root->ino_cache_inode) {
2497 		iput(root->ino_cache_inode);
2498 		root->ino_cache_inode = NULL;
2499 	}
2500 
2501 	if (btrfs_header_backref_rev(root->node) <
2502 			BTRFS_MIXED_BACKREF_REV)
2503 		ret = btrfs_drop_snapshot(root, 0, 0);
2504 	else
2505 		ret = btrfs_drop_snapshot(root, 1, 0);
2506 
2507 	btrfs_put_root(root);
2508 	return (ret < 0) ? 0 : 1;
2509 }
2510 
btrfs_apply_pending_changes(struct btrfs_fs_info * fs_info)2511 void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2512 {
2513 	unsigned long prev;
2514 	unsigned long bit;
2515 
2516 	prev = xchg(&fs_info->pending_changes, 0);
2517 	if (!prev)
2518 		return;
2519 
2520 	bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE;
2521 	if (prev & bit)
2522 		btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2523 	prev &= ~bit;
2524 
2525 	bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE;
2526 	if (prev & bit)
2527 		btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2528 	prev &= ~bit;
2529 
2530 	bit = 1 << BTRFS_PENDING_COMMIT;
2531 	if (prev & bit)
2532 		btrfs_debug(fs_info, "pending commit done");
2533 	prev &= ~bit;
2534 
2535 	if (prev)
2536 		btrfs_warn(fs_info,
2537 			"unknown pending changes left 0x%lx, ignoring", prev);
2538 }
2539