xref: /openbmc/linux/fs/btrfs/transaction.c (revision 360823a09426347ea8f232b0b0b5156d0aed0302)
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/sched/mm.h>
10 #include <linux/writeback.h>
11 #include <linux/pagemap.h>
12 #include <linux/blkdev.h>
13 #include <linux/uuid.h>
14 #include <linux/timekeeping.h>
15 #include "misc.h"
16 #include "ctree.h"
17 #include "disk-io.h"
18 #include "transaction.h"
19 #include "locking.h"
20 #include "tree-log.h"
21 #include "volumes.h"
22 #include "dev-replace.h"
23 #include "qgroup.h"
24 #include "block-group.h"
25 #include "space-info.h"
26 #include "zoned.h"
27 #include "fs.h"
28 #include "accessors.h"
29 #include "extent-tree.h"
30 #include "root-tree.h"
31 #include "defrag.h"
32 #include "dir-item.h"
33 #include "uuid-tree.h"
34 #include "ioctl.h"
35 #include "relocation.h"
36 #include "scrub.h"
37 
38 static struct kmem_cache *btrfs_trans_handle_cachep;
39 
40 /*
41  * Transaction states and transitions
42  *
43  * No running transaction (fs tree blocks are not modified)
44  * |
45  * | To next stage:
46  * |  Call start_transaction() variants. Except btrfs_join_transaction_nostart().
47  * V
48  * Transaction N [[TRANS_STATE_RUNNING]]
49  * |
50  * | New trans handles can be attached to transaction N by calling all
51  * | start_transaction() variants.
52  * |
53  * | To next stage:
54  * |  Call btrfs_commit_transaction() on any trans handle attached to
55  * |  transaction N
56  * V
57  * Transaction N [[TRANS_STATE_COMMIT_PREP]]
58  * |
59  * | If there are simultaneous calls to btrfs_commit_transaction() one will win
60  * | the race and the rest will wait for the winner to commit the transaction.
61  * |
62  * | The winner will wait for previous running transaction to completely finish
63  * | if there is one.
64  * |
65  * Transaction N [[TRANS_STATE_COMMIT_START]]
66  * |
67  * | Then one of the following happens:
68  * | - Wait for all other trans handle holders to release.
69  * |   The btrfs_commit_transaction() caller will do the commit work.
70  * | - Wait for current transaction to be committed by others.
71  * |   Other btrfs_commit_transaction() caller will do the commit work.
72  * |
73  * | At this stage, only btrfs_join_transaction*() variants can attach
74  * | to this running transaction.
75  * | All other variants will wait for current one to finish and attach to
76  * | transaction N+1.
77  * |
78  * | To next stage:
79  * |  Caller is chosen to commit transaction N, and all other trans handle
80  * |  haven been released.
81  * V
82  * Transaction N [[TRANS_STATE_COMMIT_DOING]]
83  * |
84  * | The heavy lifting transaction work is started.
85  * | From running delayed refs (modifying extent tree) to creating pending
86  * | snapshots, running qgroups.
87  * | In short, modify supporting trees to reflect modifications of subvolume
88  * | trees.
89  * |
90  * | At this stage, all start_transaction() calls will wait for this
91  * | transaction to finish and attach to transaction N+1.
92  * |
93  * | To next stage:
94  * |  Until all supporting trees are updated.
95  * V
96  * Transaction N [[TRANS_STATE_UNBLOCKED]]
97  * |						    Transaction N+1
98  * | All needed trees are modified, thus we only    [[TRANS_STATE_RUNNING]]
99  * | need to write them back to disk and update	    |
100  * | super blocks.				    |
101  * |						    |
102  * | At this stage, new transaction is allowed to   |
103  * | start.					    |
104  * | All new start_transaction() calls will be	    |
105  * | attached to transid N+1.			    |
106  * |						    |
107  * | To next stage:				    |
108  * |  Until all tree blocks are super blocks are    |
109  * |  written to block devices			    |
110  * V						    |
111  * Transaction N [[TRANS_STATE_COMPLETED]]	    V
112  *   All tree blocks and super blocks are written.  Transaction N+1
113  *   This transaction is finished and all its	    [[TRANS_STATE_COMMIT_START]]
114  *   data structures will be cleaned up.	    | Life goes on
115  */
116 static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
117 	[TRANS_STATE_RUNNING]		= 0U,
118 	[TRANS_STATE_COMMIT_PREP]	= 0U,
119 	[TRANS_STATE_COMMIT_START]	= (__TRANS_START | __TRANS_ATTACH),
120 	[TRANS_STATE_COMMIT_DOING]	= (__TRANS_START |
121 					   __TRANS_ATTACH |
122 					   __TRANS_JOIN |
123 					   __TRANS_JOIN_NOSTART),
124 	[TRANS_STATE_UNBLOCKED]		= (__TRANS_START |
125 					   __TRANS_ATTACH |
126 					   __TRANS_JOIN |
127 					   __TRANS_JOIN_NOLOCK |
128 					   __TRANS_JOIN_NOSTART),
129 	[TRANS_STATE_SUPER_COMMITTED]	= (__TRANS_START |
130 					   __TRANS_ATTACH |
131 					   __TRANS_JOIN |
132 					   __TRANS_JOIN_NOLOCK |
133 					   __TRANS_JOIN_NOSTART),
134 	[TRANS_STATE_COMPLETED]		= (__TRANS_START |
135 					   __TRANS_ATTACH |
136 					   __TRANS_JOIN |
137 					   __TRANS_JOIN_NOLOCK |
138 					   __TRANS_JOIN_NOSTART),
139 };
140 
btrfs_put_transaction(struct btrfs_transaction * transaction)141 void btrfs_put_transaction(struct btrfs_transaction *transaction)
142 {
143 	WARN_ON(refcount_read(&transaction->use_count) == 0);
144 	if (refcount_dec_and_test(&transaction->use_count)) {
145 		BUG_ON(!list_empty(&transaction->list));
146 		WARN_ON(!RB_EMPTY_ROOT(
147 				&transaction->delayed_refs.href_root.rb_root));
148 		WARN_ON(!RB_EMPTY_ROOT(
149 				&transaction->delayed_refs.dirty_extent_root));
150 		if (transaction->delayed_refs.pending_csums)
151 			btrfs_err(transaction->fs_info,
152 				  "pending csums is %llu",
153 				  transaction->delayed_refs.pending_csums);
154 		/*
155 		 * If any block groups are found in ->deleted_bgs then it's
156 		 * because the transaction was aborted and a commit did not
157 		 * happen (things failed before writing the new superblock
158 		 * and calling btrfs_finish_extent_commit()), so we can not
159 		 * discard the physical locations of the block groups.
160 		 */
161 		while (!list_empty(&transaction->deleted_bgs)) {
162 			struct btrfs_block_group *cache;
163 
164 			cache = list_first_entry(&transaction->deleted_bgs,
165 						 struct btrfs_block_group,
166 						 bg_list);
167 			list_del_init(&cache->bg_list);
168 			btrfs_unfreeze_block_group(cache);
169 			btrfs_put_block_group(cache);
170 		}
171 		WARN_ON(!list_empty(&transaction->dev_update_list));
172 		kfree(transaction);
173 	}
174 }
175 
switch_commit_roots(struct btrfs_trans_handle * trans)176 static noinline void switch_commit_roots(struct btrfs_trans_handle *trans)
177 {
178 	struct btrfs_transaction *cur_trans = trans->transaction;
179 	struct btrfs_fs_info *fs_info = trans->fs_info;
180 	struct btrfs_root *root, *tmp;
181 
182 	/*
183 	 * At this point no one can be using this transaction to modify any tree
184 	 * and no one can start another transaction to modify any tree either.
185 	 */
186 	ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING);
187 
188 	down_write(&fs_info->commit_root_sem);
189 
190 	if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
191 		fs_info->last_reloc_trans = trans->transid;
192 
193 	list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits,
194 				 dirty_list) {
195 		list_del_init(&root->dirty_list);
196 		free_extent_buffer(root->commit_root);
197 		root->commit_root = btrfs_root_node(root);
198 		extent_io_tree_release(&root->dirty_log_pages);
199 		btrfs_qgroup_clean_swapped_blocks(root);
200 	}
201 
202 	/* We can free old roots now. */
203 	spin_lock(&cur_trans->dropped_roots_lock);
204 	while (!list_empty(&cur_trans->dropped_roots)) {
205 		root = list_first_entry(&cur_trans->dropped_roots,
206 					struct btrfs_root, root_list);
207 		list_del_init(&root->root_list);
208 		spin_unlock(&cur_trans->dropped_roots_lock);
209 		btrfs_free_log(trans, root);
210 		btrfs_drop_and_free_fs_root(fs_info, root);
211 		spin_lock(&cur_trans->dropped_roots_lock);
212 	}
213 	spin_unlock(&cur_trans->dropped_roots_lock);
214 
215 	up_write(&fs_info->commit_root_sem);
216 }
217 
extwriter_counter_inc(struct btrfs_transaction * trans,unsigned int type)218 static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
219 					 unsigned int type)
220 {
221 	if (type & TRANS_EXTWRITERS)
222 		atomic_inc(&trans->num_extwriters);
223 }
224 
extwriter_counter_dec(struct btrfs_transaction * trans,unsigned int type)225 static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
226 					 unsigned int type)
227 {
228 	if (type & TRANS_EXTWRITERS)
229 		atomic_dec(&trans->num_extwriters);
230 }
231 
extwriter_counter_init(struct btrfs_transaction * trans,unsigned int type)232 static inline void extwriter_counter_init(struct btrfs_transaction *trans,
233 					  unsigned int type)
234 {
235 	atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
236 }
237 
extwriter_counter_read(struct btrfs_transaction * trans)238 static inline int extwriter_counter_read(struct btrfs_transaction *trans)
239 {
240 	return atomic_read(&trans->num_extwriters);
241 }
242 
243 /*
244  * To be called after doing the chunk btree updates right after allocating a new
245  * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a
246  * chunk after all chunk btree updates and after finishing the second phase of
247  * chunk allocation (btrfs_create_pending_block_groups()) in case some block
248  * group had its chunk item insertion delayed to the second phase.
249  */
btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle * trans)250 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
251 {
252 	struct btrfs_fs_info *fs_info = trans->fs_info;
253 
254 	if (!trans->chunk_bytes_reserved)
255 		return;
256 
257 	btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv,
258 				trans->chunk_bytes_reserved, NULL);
259 	trans->chunk_bytes_reserved = 0;
260 }
261 
262 /*
263  * either allocate a new transaction or hop into the existing one
264  */
join_transaction(struct btrfs_fs_info * fs_info,unsigned int type)265 static noinline int join_transaction(struct btrfs_fs_info *fs_info,
266 				     unsigned int type)
267 {
268 	struct btrfs_transaction *cur_trans;
269 
270 	spin_lock(&fs_info->trans_lock);
271 loop:
272 	/* The file system has been taken offline. No new transactions. */
273 	if (BTRFS_FS_ERROR(fs_info)) {
274 		spin_unlock(&fs_info->trans_lock);
275 		return -EROFS;
276 	}
277 
278 	cur_trans = fs_info->running_transaction;
279 	if (cur_trans) {
280 		if (TRANS_ABORTED(cur_trans)) {
281 			const int abort_error = cur_trans->aborted;
282 
283 			spin_unlock(&fs_info->trans_lock);
284 			return abort_error;
285 		}
286 		if (btrfs_blocked_trans_types[cur_trans->state] & type) {
287 			spin_unlock(&fs_info->trans_lock);
288 			return -EBUSY;
289 		}
290 		refcount_inc(&cur_trans->use_count);
291 		atomic_inc(&cur_trans->num_writers);
292 		extwriter_counter_inc(cur_trans, type);
293 		spin_unlock(&fs_info->trans_lock);
294 		btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers);
295 		btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters);
296 		return 0;
297 	}
298 	spin_unlock(&fs_info->trans_lock);
299 
300 	/*
301 	 * If we are ATTACH or TRANS_JOIN_NOSTART, we just want to catch the
302 	 * current transaction, and commit it. If there is no transaction, just
303 	 * return ENOENT.
304 	 */
305 	if (type == TRANS_ATTACH || type == TRANS_JOIN_NOSTART)
306 		return -ENOENT;
307 
308 	/*
309 	 * JOIN_NOLOCK only happens during the transaction commit, so
310 	 * it is impossible that ->running_transaction is NULL
311 	 */
312 	BUG_ON(type == TRANS_JOIN_NOLOCK);
313 
314 	cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS);
315 	if (!cur_trans)
316 		return -ENOMEM;
317 
318 	btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers);
319 	btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters);
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 		btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
328 		btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
329 		kfree(cur_trans);
330 		goto loop;
331 	} else if (BTRFS_FS_ERROR(fs_info)) {
332 		spin_unlock(&fs_info->trans_lock);
333 		btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
334 		btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
335 		kfree(cur_trans);
336 		return -EROFS;
337 	}
338 
339 	cur_trans->fs_info = fs_info;
340 	atomic_set(&cur_trans->pending_ordered, 0);
341 	init_waitqueue_head(&cur_trans->pending_wait);
342 	atomic_set(&cur_trans->num_writers, 1);
343 	extwriter_counter_init(cur_trans, type);
344 	init_waitqueue_head(&cur_trans->writer_wait);
345 	init_waitqueue_head(&cur_trans->commit_wait);
346 	cur_trans->state = TRANS_STATE_RUNNING;
347 	/*
348 	 * One for this trans handle, one so it will live on until we
349 	 * commit the transaction.
350 	 */
351 	refcount_set(&cur_trans->use_count, 2);
352 	cur_trans->flags = 0;
353 	cur_trans->start_time = ktime_get_seconds();
354 
355 	memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));
356 
357 	cur_trans->delayed_refs.href_root = RB_ROOT_CACHED;
358 	cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
359 	atomic_set(&cur_trans->delayed_refs.num_entries, 0);
360 
361 	/*
362 	 * although the tree mod log is per file system and not per transaction,
363 	 * the log must never go across transaction boundaries.
364 	 */
365 	smp_mb();
366 	if (!list_empty(&fs_info->tree_mod_seq_list))
367 		WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");
368 	if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
369 		WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");
370 	atomic64_set(&fs_info->tree_mod_seq, 0);
371 
372 	spin_lock_init(&cur_trans->delayed_refs.lock);
373 
374 	INIT_LIST_HEAD(&cur_trans->pending_snapshots);
375 	INIT_LIST_HEAD(&cur_trans->dev_update_list);
376 	INIT_LIST_HEAD(&cur_trans->switch_commits);
377 	INIT_LIST_HEAD(&cur_trans->dirty_bgs);
378 	INIT_LIST_HEAD(&cur_trans->io_bgs);
379 	INIT_LIST_HEAD(&cur_trans->dropped_roots);
380 	mutex_init(&cur_trans->cache_write_mutex);
381 	spin_lock_init(&cur_trans->dirty_bgs_lock);
382 	INIT_LIST_HEAD(&cur_trans->deleted_bgs);
383 	spin_lock_init(&cur_trans->dropped_roots_lock);
384 	list_add_tail(&cur_trans->list, &fs_info->trans_list);
385 	extent_io_tree_init(fs_info, &cur_trans->dirty_pages,
386 			IO_TREE_TRANS_DIRTY_PAGES);
387 	extent_io_tree_init(fs_info, &cur_trans->pinned_extents,
388 			IO_TREE_FS_PINNED_EXTENTS);
389 	fs_info->generation++;
390 	cur_trans->transid = fs_info->generation;
391 	fs_info->running_transaction = cur_trans;
392 	cur_trans->aborted = 0;
393 	spin_unlock(&fs_info->trans_lock);
394 
395 	return 0;
396 }
397 
398 /*
399  * This does all the record keeping required to make sure that a shareable root
400  * is properly recorded in a given transaction.  This is required to make sure
401  * the old root from before we joined the transaction is deleted when the
402  * transaction commits.
403  */
record_root_in_trans(struct btrfs_trans_handle * trans,struct btrfs_root * root,int force)404 static int record_root_in_trans(struct btrfs_trans_handle *trans,
405 			       struct btrfs_root *root,
406 			       int force)
407 {
408 	struct btrfs_fs_info *fs_info = root->fs_info;
409 	int ret = 0;
410 
411 	if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
412 	    root->last_trans < trans->transid) || force) {
413 		WARN_ON(!force && root->commit_root != root->node);
414 
415 		/*
416 		 * see below for IN_TRANS_SETUP usage rules
417 		 * we have the reloc mutex held now, so there
418 		 * is only one writer in this function
419 		 */
420 		set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
421 
422 		/* make sure readers find IN_TRANS_SETUP before
423 		 * they find our root->last_trans update
424 		 */
425 		smp_wmb();
426 
427 		spin_lock(&fs_info->fs_roots_radix_lock);
428 		if (root->last_trans == trans->transid && !force) {
429 			spin_unlock(&fs_info->fs_roots_radix_lock);
430 			return 0;
431 		}
432 		radix_tree_tag_set(&fs_info->fs_roots_radix,
433 				   (unsigned long)root->root_key.objectid,
434 				   BTRFS_ROOT_TRANS_TAG);
435 		spin_unlock(&fs_info->fs_roots_radix_lock);
436 		root->last_trans = trans->transid;
437 
438 		/* this is pretty tricky.  We don't want to
439 		 * take the relocation lock in btrfs_record_root_in_trans
440 		 * unless we're really doing the first setup for this root in
441 		 * this transaction.
442 		 *
443 		 * Normally we'd use root->last_trans as a flag to decide
444 		 * if we want to take the expensive mutex.
445 		 *
446 		 * But, we have to set root->last_trans before we
447 		 * init the relocation root, otherwise, we trip over warnings
448 		 * in ctree.c.  The solution used here is to flag ourselves
449 		 * with root IN_TRANS_SETUP.  When this is 1, we're still
450 		 * fixing up the reloc trees and everyone must wait.
451 		 *
452 		 * When this is zero, they can trust root->last_trans and fly
453 		 * through btrfs_record_root_in_trans without having to take the
454 		 * lock.  smp_wmb() makes sure that all the writes above are
455 		 * done before we pop in the zero below
456 		 */
457 		ret = btrfs_init_reloc_root(trans, root);
458 		smp_mb__before_atomic();
459 		clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
460 	}
461 	return ret;
462 }
463 
464 
btrfs_add_dropped_root(struct btrfs_trans_handle * trans,struct btrfs_root * root)465 void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
466 			    struct btrfs_root *root)
467 {
468 	struct btrfs_fs_info *fs_info = root->fs_info;
469 	struct btrfs_transaction *cur_trans = trans->transaction;
470 
471 	/* Add ourselves to the transaction dropped list */
472 	spin_lock(&cur_trans->dropped_roots_lock);
473 	list_add_tail(&root->root_list, &cur_trans->dropped_roots);
474 	spin_unlock(&cur_trans->dropped_roots_lock);
475 
476 	/* Make sure we don't try to update the root at commit time */
477 	spin_lock(&fs_info->fs_roots_radix_lock);
478 	radix_tree_tag_clear(&fs_info->fs_roots_radix,
479 			     (unsigned long)root->root_key.objectid,
480 			     BTRFS_ROOT_TRANS_TAG);
481 	spin_unlock(&fs_info->fs_roots_radix_lock);
482 }
483 
btrfs_record_root_in_trans(struct btrfs_trans_handle * trans,struct btrfs_root * root)484 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
485 			       struct btrfs_root *root)
486 {
487 	struct btrfs_fs_info *fs_info = root->fs_info;
488 	int ret;
489 
490 	if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
491 		return 0;
492 
493 	/*
494 	 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
495 	 * and barriers
496 	 */
497 	smp_rmb();
498 	if (root->last_trans == trans->transid &&
499 	    !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
500 		return 0;
501 
502 	mutex_lock(&fs_info->reloc_mutex);
503 	ret = record_root_in_trans(trans, root, 0);
504 	mutex_unlock(&fs_info->reloc_mutex);
505 
506 	return ret;
507 }
508 
is_transaction_blocked(struct btrfs_transaction * trans)509 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
510 {
511 	return (trans->state >= TRANS_STATE_COMMIT_START &&
512 		trans->state < TRANS_STATE_UNBLOCKED &&
513 		!TRANS_ABORTED(trans));
514 }
515 
516 /* wait for commit against the current transaction to become unblocked
517  * when this is done, it is safe to start a new transaction, but the current
518  * transaction might not be fully on disk.
519  */
wait_current_trans(struct btrfs_fs_info * fs_info)520 static void wait_current_trans(struct btrfs_fs_info *fs_info)
521 {
522 	struct btrfs_transaction *cur_trans;
523 
524 	spin_lock(&fs_info->trans_lock);
525 	cur_trans = fs_info->running_transaction;
526 	if (cur_trans && is_transaction_blocked(cur_trans)) {
527 		refcount_inc(&cur_trans->use_count);
528 		spin_unlock(&fs_info->trans_lock);
529 
530 		btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
531 		wait_event(fs_info->transaction_wait,
532 			   cur_trans->state >= TRANS_STATE_UNBLOCKED ||
533 			   TRANS_ABORTED(cur_trans));
534 		btrfs_put_transaction(cur_trans);
535 	} else {
536 		spin_unlock(&fs_info->trans_lock);
537 	}
538 }
539 
may_wait_transaction(struct btrfs_fs_info * fs_info,int type)540 static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type)
541 {
542 	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
543 		return 0;
544 
545 	if (type == TRANS_START)
546 		return 1;
547 
548 	return 0;
549 }
550 
need_reserve_reloc_root(struct btrfs_root * root)551 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
552 {
553 	struct btrfs_fs_info *fs_info = root->fs_info;
554 
555 	if (!fs_info->reloc_ctl ||
556 	    !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
557 	    root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
558 	    root->reloc_root)
559 		return false;
560 
561 	return true;
562 }
563 
564 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)565 start_transaction(struct btrfs_root *root, unsigned int num_items,
566 		  unsigned int type, enum btrfs_reserve_flush_enum flush,
567 		  bool enforce_qgroups)
568 {
569 	struct btrfs_fs_info *fs_info = root->fs_info;
570 	struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
571 	struct btrfs_trans_handle *h;
572 	struct btrfs_transaction *cur_trans;
573 	u64 num_bytes = 0;
574 	u64 qgroup_reserved = 0;
575 	bool reloc_reserved = false;
576 	bool do_chunk_alloc = false;
577 	int ret;
578 
579 	if (BTRFS_FS_ERROR(fs_info))
580 		return ERR_PTR(-EROFS);
581 
582 	if (current->journal_info) {
583 		WARN_ON(type & TRANS_EXTWRITERS);
584 		h = current->journal_info;
585 		refcount_inc(&h->use_count);
586 		WARN_ON(refcount_read(&h->use_count) > 2);
587 		h->orig_rsv = h->block_rsv;
588 		h->block_rsv = NULL;
589 		goto got_it;
590 	}
591 
592 	/*
593 	 * Do the reservation before we join the transaction so we can do all
594 	 * the appropriate flushing if need be.
595 	 */
596 	if (num_items && root != fs_info->chunk_root) {
597 		struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv;
598 		u64 delayed_refs_bytes = 0;
599 
600 		qgroup_reserved = num_items * fs_info->nodesize;
601 		/*
602 		 * Use prealloc for now, as there might be a currently running
603 		 * transaction that could free this reserved space prematurely
604 		 * by committing.
605 		 */
606 		ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_reserved,
607 							 enforce_qgroups, false);
608 		if (ret)
609 			return ERR_PTR(ret);
610 
611 		/*
612 		 * We want to reserve all the bytes we may need all at once, so
613 		 * we only do 1 enospc flushing cycle per transaction start.  We
614 		 * accomplish this by simply assuming we'll do num_items worth
615 		 * of delayed refs updates in this trans handle, and refill that
616 		 * amount for whatever is missing in the reserve.
617 		 */
618 		num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items);
619 		if (flush == BTRFS_RESERVE_FLUSH_ALL &&
620 		    !btrfs_block_rsv_full(delayed_refs_rsv)) {
621 			delayed_refs_bytes = btrfs_calc_delayed_ref_bytes(fs_info,
622 									  num_items);
623 			num_bytes += delayed_refs_bytes;
624 		}
625 
626 		/*
627 		 * Do the reservation for the relocation root creation
628 		 */
629 		if (need_reserve_reloc_root(root)) {
630 			num_bytes += fs_info->nodesize;
631 			reloc_reserved = true;
632 		}
633 
634 		ret = btrfs_reserve_metadata_bytes(fs_info, rsv, num_bytes, flush);
635 		if (ret)
636 			goto reserve_fail;
637 		if (delayed_refs_bytes) {
638 			btrfs_migrate_to_delayed_refs_rsv(fs_info, delayed_refs_bytes);
639 			num_bytes -= delayed_refs_bytes;
640 		}
641 		btrfs_block_rsv_add_bytes(rsv, num_bytes, true);
642 
643 		if (rsv->space_info->force_alloc)
644 			do_chunk_alloc = true;
645 	} else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL &&
646 		   !btrfs_block_rsv_full(delayed_refs_rsv)) {
647 		/*
648 		 * Some people call with btrfs_start_transaction(root, 0)
649 		 * because they can be throttled, but have some other mechanism
650 		 * for reserving space.  We still want these guys to refill the
651 		 * delayed block_rsv so just add 1 items worth of reservation
652 		 * here.
653 		 */
654 		ret = btrfs_delayed_refs_rsv_refill(fs_info, flush);
655 		if (ret)
656 			goto reserve_fail;
657 	}
658 again:
659 	h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
660 	if (!h) {
661 		ret = -ENOMEM;
662 		goto alloc_fail;
663 	}
664 
665 	/*
666 	 * If we are JOIN_NOLOCK we're already committing a transaction and
667 	 * waiting on this guy, so we don't need to do the sb_start_intwrite
668 	 * because we're already holding a ref.  We need this because we could
669 	 * have raced in and did an fsync() on a file which can kick a commit
670 	 * and then we deadlock with somebody doing a freeze.
671 	 *
672 	 * If we are ATTACH, it means we just want to catch the current
673 	 * transaction and commit it, so we needn't do sb_start_intwrite().
674 	 */
675 	if (type & __TRANS_FREEZABLE)
676 		sb_start_intwrite(fs_info->sb);
677 
678 	if (may_wait_transaction(fs_info, type))
679 		wait_current_trans(fs_info);
680 
681 	do {
682 		ret = join_transaction(fs_info, type);
683 		if (ret == -EBUSY) {
684 			wait_current_trans(fs_info);
685 			if (unlikely(type == TRANS_ATTACH ||
686 				     type == TRANS_JOIN_NOSTART))
687 				ret = -ENOENT;
688 		}
689 	} while (ret == -EBUSY);
690 
691 	if (ret < 0)
692 		goto join_fail;
693 
694 	cur_trans = fs_info->running_transaction;
695 
696 	h->transid = cur_trans->transid;
697 	h->transaction = cur_trans;
698 	refcount_set(&h->use_count, 1);
699 	h->fs_info = root->fs_info;
700 
701 	h->type = type;
702 	INIT_LIST_HEAD(&h->new_bgs);
703 
704 	smp_mb();
705 	if (cur_trans->state >= TRANS_STATE_COMMIT_START &&
706 	    may_wait_transaction(fs_info, type)) {
707 		current->journal_info = h;
708 		btrfs_commit_transaction(h);
709 		goto again;
710 	}
711 
712 	if (num_bytes) {
713 		trace_btrfs_space_reservation(fs_info, "transaction",
714 					      h->transid, num_bytes, 1);
715 		h->block_rsv = &fs_info->trans_block_rsv;
716 		h->bytes_reserved = num_bytes;
717 		h->reloc_reserved = reloc_reserved;
718 	}
719 
720 got_it:
721 	if (!current->journal_info)
722 		current->journal_info = h;
723 
724 	/*
725 	 * If the space_info is marked ALLOC_FORCE then we'll get upgraded to
726 	 * ALLOC_FORCE the first run through, and then we won't allocate for
727 	 * anybody else who races in later.  We don't care about the return
728 	 * value here.
729 	 */
730 	if (do_chunk_alloc && num_bytes) {
731 		u64 flags = h->block_rsv->space_info->flags;
732 
733 		btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags),
734 				  CHUNK_ALLOC_NO_FORCE);
735 	}
736 
737 	/*
738 	 * btrfs_record_root_in_trans() needs to alloc new extents, and may
739 	 * call btrfs_join_transaction() while we're also starting a
740 	 * transaction.
741 	 *
742 	 * Thus it need to be called after current->journal_info initialized,
743 	 * or we can deadlock.
744 	 */
745 	ret = btrfs_record_root_in_trans(h, root);
746 	if (ret) {
747 		/*
748 		 * The transaction handle is fully initialized and linked with
749 		 * other structures so it needs to be ended in case of errors,
750 		 * not just freed.
751 		 */
752 		btrfs_end_transaction(h);
753 		goto reserve_fail;
754 	}
755 	/*
756 	 * Now that we have found a transaction to be a part of, convert the
757 	 * qgroup reservation from prealloc to pertrans. A different transaction
758 	 * can't race in and free our pertrans out from under us.
759 	 */
760 	if (qgroup_reserved)
761 		btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved);
762 
763 	return h;
764 
765 join_fail:
766 	if (type & __TRANS_FREEZABLE)
767 		sb_end_intwrite(fs_info->sb);
768 	kmem_cache_free(btrfs_trans_handle_cachep, h);
769 alloc_fail:
770 	if (num_bytes)
771 		btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv,
772 					num_bytes, NULL);
773 reserve_fail:
774 	btrfs_qgroup_free_meta_prealloc(root, qgroup_reserved);
775 	return ERR_PTR(ret);
776 }
777 
btrfs_start_transaction(struct btrfs_root * root,unsigned int num_items)778 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
779 						   unsigned int num_items)
780 {
781 	return start_transaction(root, num_items, TRANS_START,
782 				 BTRFS_RESERVE_FLUSH_ALL, true);
783 }
784 
btrfs_start_transaction_fallback_global_rsv(struct btrfs_root * root,unsigned int num_items)785 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
786 					struct btrfs_root *root,
787 					unsigned int num_items)
788 {
789 	return start_transaction(root, num_items, TRANS_START,
790 				 BTRFS_RESERVE_FLUSH_ALL_STEAL, false);
791 }
792 
btrfs_join_transaction(struct btrfs_root * root)793 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
794 {
795 	return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH,
796 				 true);
797 }
798 
btrfs_join_transaction_spacecache(struct btrfs_root * root)799 struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root)
800 {
801 	return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
802 				 BTRFS_RESERVE_NO_FLUSH, true);
803 }
804 
805 /*
806  * Similar to regular join but it never starts a transaction when none is
807  * running or when there's a running one at a state >= TRANS_STATE_UNBLOCKED.
808  * This is similar to btrfs_attach_transaction() but it allows the join to
809  * happen if the transaction commit already started but it's not yet in the
810  * "doing" phase (the state is < TRANS_STATE_COMMIT_DOING).
811  */
btrfs_join_transaction_nostart(struct btrfs_root * root)812 struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root)
813 {
814 	return start_transaction(root, 0, TRANS_JOIN_NOSTART,
815 				 BTRFS_RESERVE_NO_FLUSH, true);
816 }
817 
818 /*
819  * btrfs_attach_transaction() - catch the running transaction
820  *
821  * It is used when we want to commit the current the transaction, but
822  * don't want to start a new one.
823  *
824  * Note: If this function return -ENOENT, it just means there is no
825  * running transaction. But it is possible that the inactive transaction
826  * is still in the memory, not fully on disk. If you hope there is no
827  * inactive transaction in the fs when -ENOENT is returned, you should
828  * invoke
829  *     btrfs_attach_transaction_barrier()
830  */
btrfs_attach_transaction(struct btrfs_root * root)831 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
832 {
833 	return start_transaction(root, 0, TRANS_ATTACH,
834 				 BTRFS_RESERVE_NO_FLUSH, true);
835 }
836 
837 /*
838  * btrfs_attach_transaction_barrier() - catch the running transaction
839  *
840  * It is similar to the above function, the difference is this one
841  * will wait for all the inactive transactions until they fully
842  * complete.
843  */
844 struct btrfs_trans_handle *
btrfs_attach_transaction_barrier(struct btrfs_root * root)845 btrfs_attach_transaction_barrier(struct btrfs_root *root)
846 {
847 	struct btrfs_trans_handle *trans;
848 
849 	trans = start_transaction(root, 0, TRANS_ATTACH,
850 				  BTRFS_RESERVE_NO_FLUSH, true);
851 	if (trans == ERR_PTR(-ENOENT)) {
852 		int ret;
853 
854 		ret = btrfs_wait_for_commit(root->fs_info, 0);
855 		if (ret)
856 			return ERR_PTR(ret);
857 	}
858 
859 	return trans;
860 }
861 
862 /* Wait for a transaction commit to reach at least the given state. */
wait_for_commit(struct btrfs_transaction * commit,const enum btrfs_trans_state min_state)863 static noinline void wait_for_commit(struct btrfs_transaction *commit,
864 				     const enum btrfs_trans_state min_state)
865 {
866 	struct btrfs_fs_info *fs_info = commit->fs_info;
867 	u64 transid = commit->transid;
868 	bool put = false;
869 
870 	/*
871 	 * At the moment this function is called with min_state either being
872 	 * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED.
873 	 */
874 	if (min_state == TRANS_STATE_COMPLETED)
875 		btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
876 	else
877 		btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
878 
879 	while (1) {
880 		wait_event(commit->commit_wait, commit->state >= min_state);
881 		if (put)
882 			btrfs_put_transaction(commit);
883 
884 		if (min_state < TRANS_STATE_COMPLETED)
885 			break;
886 
887 		/*
888 		 * A transaction isn't really completed until all of the
889 		 * previous transactions are completed, but with fsync we can
890 		 * end up with SUPER_COMMITTED transactions before a COMPLETED
891 		 * transaction. Wait for those.
892 		 */
893 
894 		spin_lock(&fs_info->trans_lock);
895 		commit = list_first_entry_or_null(&fs_info->trans_list,
896 						  struct btrfs_transaction,
897 						  list);
898 		if (!commit || commit->transid > transid) {
899 			spin_unlock(&fs_info->trans_lock);
900 			break;
901 		}
902 		refcount_inc(&commit->use_count);
903 		put = true;
904 		spin_unlock(&fs_info->trans_lock);
905 	}
906 }
907 
btrfs_wait_for_commit(struct btrfs_fs_info * fs_info,u64 transid)908 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid)
909 {
910 	struct btrfs_transaction *cur_trans = NULL, *t;
911 	int ret = 0;
912 
913 	if (transid) {
914 		if (transid <= fs_info->last_trans_committed)
915 			goto out;
916 
917 		/* find specified transaction */
918 		spin_lock(&fs_info->trans_lock);
919 		list_for_each_entry(t, &fs_info->trans_list, list) {
920 			if (t->transid == transid) {
921 				cur_trans = t;
922 				refcount_inc(&cur_trans->use_count);
923 				ret = 0;
924 				break;
925 			}
926 			if (t->transid > transid) {
927 				ret = 0;
928 				break;
929 			}
930 		}
931 		spin_unlock(&fs_info->trans_lock);
932 
933 		/*
934 		 * The specified transaction doesn't exist, or we
935 		 * raced with btrfs_commit_transaction
936 		 */
937 		if (!cur_trans) {
938 			if (transid > fs_info->last_trans_committed)
939 				ret = -EINVAL;
940 			goto out;
941 		}
942 	} else {
943 		/* find newest transaction that is committing | committed */
944 		spin_lock(&fs_info->trans_lock);
945 		list_for_each_entry_reverse(t, &fs_info->trans_list,
946 					    list) {
947 			if (t->state >= TRANS_STATE_COMMIT_START) {
948 				if (t->state == TRANS_STATE_COMPLETED)
949 					break;
950 				cur_trans = t;
951 				refcount_inc(&cur_trans->use_count);
952 				break;
953 			}
954 		}
955 		spin_unlock(&fs_info->trans_lock);
956 		if (!cur_trans)
957 			goto out;  /* nothing committing|committed */
958 	}
959 
960 	wait_for_commit(cur_trans, TRANS_STATE_COMPLETED);
961 	ret = cur_trans->aborted;
962 	btrfs_put_transaction(cur_trans);
963 out:
964 	return ret;
965 }
966 
btrfs_throttle(struct btrfs_fs_info * fs_info)967 void btrfs_throttle(struct btrfs_fs_info *fs_info)
968 {
969 	wait_current_trans(fs_info);
970 }
971 
btrfs_should_end_transaction(struct btrfs_trans_handle * trans)972 bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
973 {
974 	struct btrfs_transaction *cur_trans = trans->transaction;
975 
976 	if (cur_trans->state >= TRANS_STATE_COMMIT_START ||
977 	    test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags))
978 		return true;
979 
980 	if (btrfs_check_space_for_delayed_refs(trans->fs_info))
981 		return true;
982 
983 	return !!btrfs_block_rsv_check(&trans->fs_info->global_block_rsv, 50);
984 }
985 
btrfs_trans_release_metadata(struct btrfs_trans_handle * trans)986 static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans)
987 
988 {
989 	struct btrfs_fs_info *fs_info = trans->fs_info;
990 
991 	if (!trans->block_rsv) {
992 		ASSERT(!trans->bytes_reserved);
993 		return;
994 	}
995 
996 	if (!trans->bytes_reserved)
997 		return;
998 
999 	ASSERT(trans->block_rsv == &fs_info->trans_block_rsv);
1000 	trace_btrfs_space_reservation(fs_info, "transaction",
1001 				      trans->transid, trans->bytes_reserved, 0);
1002 	btrfs_block_rsv_release(fs_info, trans->block_rsv,
1003 				trans->bytes_reserved, NULL);
1004 	trans->bytes_reserved = 0;
1005 }
1006 
__btrfs_end_transaction(struct btrfs_trans_handle * trans,int throttle)1007 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
1008 				   int throttle)
1009 {
1010 	struct btrfs_fs_info *info = trans->fs_info;
1011 	struct btrfs_transaction *cur_trans = trans->transaction;
1012 	int err = 0;
1013 
1014 	if (refcount_read(&trans->use_count) > 1) {
1015 		refcount_dec(&trans->use_count);
1016 		trans->block_rsv = trans->orig_rsv;
1017 		return 0;
1018 	}
1019 
1020 	btrfs_trans_release_metadata(trans);
1021 	trans->block_rsv = NULL;
1022 
1023 	btrfs_create_pending_block_groups(trans);
1024 
1025 	btrfs_trans_release_chunk_metadata(trans);
1026 
1027 	if (trans->type & __TRANS_FREEZABLE)
1028 		sb_end_intwrite(info->sb);
1029 
1030 	WARN_ON(cur_trans != info->running_transaction);
1031 	WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
1032 	atomic_dec(&cur_trans->num_writers);
1033 	extwriter_counter_dec(cur_trans, trans->type);
1034 
1035 	cond_wake_up(&cur_trans->writer_wait);
1036 
1037 	btrfs_lockdep_release(info, btrfs_trans_num_extwriters);
1038 	btrfs_lockdep_release(info, btrfs_trans_num_writers);
1039 
1040 	btrfs_put_transaction(cur_trans);
1041 
1042 	if (current->journal_info == trans)
1043 		current->journal_info = NULL;
1044 
1045 	if (throttle)
1046 		btrfs_run_delayed_iputs(info);
1047 
1048 	if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) {
1049 		wake_up_process(info->transaction_kthread);
1050 		if (TRANS_ABORTED(trans))
1051 			err = trans->aborted;
1052 		else
1053 			err = -EROFS;
1054 	}
1055 
1056 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
1057 	return err;
1058 }
1059 
btrfs_end_transaction(struct btrfs_trans_handle * trans)1060 int btrfs_end_transaction(struct btrfs_trans_handle *trans)
1061 {
1062 	return __btrfs_end_transaction(trans, 0);
1063 }
1064 
btrfs_end_transaction_throttle(struct btrfs_trans_handle * trans)1065 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans)
1066 {
1067 	return __btrfs_end_transaction(trans, 1);
1068 }
1069 
1070 /*
1071  * when btree blocks are allocated, they have some corresponding bits set for
1072  * them in one of two extent_io trees.  This is used to make sure all of
1073  * those extents are sent to disk but does not wait on them
1074  */
btrfs_write_marked_extents(struct btrfs_fs_info * fs_info,struct extent_io_tree * dirty_pages,int mark)1075 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
1076 			       struct extent_io_tree *dirty_pages, int mark)
1077 {
1078 	int err = 0;
1079 	int werr = 0;
1080 	struct address_space *mapping = fs_info->btree_inode->i_mapping;
1081 	struct extent_state *cached_state = NULL;
1082 	u64 start = 0;
1083 	u64 end;
1084 
1085 	while (find_first_extent_bit(dirty_pages, start, &start, &end,
1086 				     mark, &cached_state)) {
1087 		bool wait_writeback = false;
1088 
1089 		err = convert_extent_bit(dirty_pages, start, end,
1090 					 EXTENT_NEED_WAIT,
1091 					 mark, &cached_state);
1092 		/*
1093 		 * convert_extent_bit can return -ENOMEM, which is most of the
1094 		 * time a temporary error. So when it happens, ignore the error
1095 		 * and wait for writeback of this range to finish - because we
1096 		 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
1097 		 * to __btrfs_wait_marked_extents() would not know that
1098 		 * writeback for this range started and therefore wouldn't
1099 		 * wait for it to finish - we don't want to commit a
1100 		 * superblock that points to btree nodes/leafs for which
1101 		 * writeback hasn't finished yet (and without errors).
1102 		 * We cleanup any entries left in the io tree when committing
1103 		 * the transaction (through extent_io_tree_release()).
1104 		 */
1105 		if (err == -ENOMEM) {
1106 			err = 0;
1107 			wait_writeback = true;
1108 		}
1109 		if (!err)
1110 			err = filemap_fdatawrite_range(mapping, start, end);
1111 		if (err)
1112 			werr = err;
1113 		else if (wait_writeback)
1114 			werr = filemap_fdatawait_range(mapping, start, end);
1115 		free_extent_state(cached_state);
1116 		cached_state = NULL;
1117 		cond_resched();
1118 		start = end + 1;
1119 	}
1120 	return werr;
1121 }
1122 
1123 /*
1124  * when btree blocks are allocated, they have some corresponding bits set for
1125  * them in one of two extent_io trees.  This is used to make sure all of
1126  * those extents are on disk for transaction or log commit.  We wait
1127  * on all the pages and clear them from the dirty pages state tree
1128  */
__btrfs_wait_marked_extents(struct btrfs_fs_info * fs_info,struct extent_io_tree * dirty_pages)1129 static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info,
1130 				       struct extent_io_tree *dirty_pages)
1131 {
1132 	int err = 0;
1133 	int werr = 0;
1134 	struct address_space *mapping = fs_info->btree_inode->i_mapping;
1135 	struct extent_state *cached_state = NULL;
1136 	u64 start = 0;
1137 	u64 end;
1138 
1139 	while (find_first_extent_bit(dirty_pages, start, &start, &end,
1140 				     EXTENT_NEED_WAIT, &cached_state)) {
1141 		/*
1142 		 * Ignore -ENOMEM errors returned by clear_extent_bit().
1143 		 * When committing the transaction, we'll remove any entries
1144 		 * left in the io tree. For a log commit, we don't remove them
1145 		 * after committing the log because the tree can be accessed
1146 		 * concurrently - we do it only at transaction commit time when
1147 		 * it's safe to do it (through extent_io_tree_release()).
1148 		 */
1149 		err = clear_extent_bit(dirty_pages, start, end,
1150 				       EXTENT_NEED_WAIT, &cached_state);
1151 		if (err == -ENOMEM)
1152 			err = 0;
1153 		if (!err)
1154 			err = filemap_fdatawait_range(mapping, start, end);
1155 		if (err)
1156 			werr = err;
1157 		free_extent_state(cached_state);
1158 		cached_state = NULL;
1159 		cond_resched();
1160 		start = end + 1;
1161 	}
1162 	if (err)
1163 		werr = err;
1164 	return werr;
1165 }
1166 
btrfs_wait_extents(struct btrfs_fs_info * fs_info,struct extent_io_tree * dirty_pages)1167 static int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
1168 		       struct extent_io_tree *dirty_pages)
1169 {
1170 	bool errors = false;
1171 	int err;
1172 
1173 	err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1174 	if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags))
1175 		errors = true;
1176 
1177 	if (errors && !err)
1178 		err = -EIO;
1179 	return err;
1180 }
1181 
btrfs_wait_tree_log_extents(struct btrfs_root * log_root,int mark)1182 int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark)
1183 {
1184 	struct btrfs_fs_info *fs_info = log_root->fs_info;
1185 	struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages;
1186 	bool errors = false;
1187 	int err;
1188 
1189 	ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
1190 
1191 	err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1192 	if ((mark & EXTENT_DIRTY) &&
1193 	    test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags))
1194 		errors = true;
1195 
1196 	if ((mark & EXTENT_NEW) &&
1197 	    test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags))
1198 		errors = true;
1199 
1200 	if (errors && !err)
1201 		err = -EIO;
1202 	return err;
1203 }
1204 
1205 /*
1206  * When btree blocks are allocated the corresponding extents are marked dirty.
1207  * This function ensures such extents are persisted on disk for transaction or
1208  * log commit.
1209  *
1210  * @trans: transaction whose dirty pages we'd like to write
1211  */
btrfs_write_and_wait_transaction(struct btrfs_trans_handle * trans)1212 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans)
1213 {
1214 	int ret;
1215 	int ret2;
1216 	struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages;
1217 	struct btrfs_fs_info *fs_info = trans->fs_info;
1218 	struct blk_plug plug;
1219 
1220 	blk_start_plug(&plug);
1221 	ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY);
1222 	blk_finish_plug(&plug);
1223 	ret2 = btrfs_wait_extents(fs_info, dirty_pages);
1224 
1225 	extent_io_tree_release(&trans->transaction->dirty_pages);
1226 
1227 	if (ret)
1228 		return ret;
1229 	else if (ret2)
1230 		return ret2;
1231 	else
1232 		return 0;
1233 }
1234 
1235 /*
1236  * this is used to update the root pointer in the tree of tree roots.
1237  *
1238  * But, in the case of the extent allocation tree, updating the root
1239  * pointer may allocate blocks which may change the root of the extent
1240  * allocation tree.
1241  *
1242  * So, this loops and repeats and makes sure the cowonly root didn't
1243  * change while the root pointer was being updated in the metadata.
1244  */
update_cowonly_root(struct btrfs_trans_handle * trans,struct btrfs_root * root)1245 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1246 			       struct btrfs_root *root)
1247 {
1248 	int ret;
1249 	u64 old_root_bytenr;
1250 	u64 old_root_used;
1251 	struct btrfs_fs_info *fs_info = root->fs_info;
1252 	struct btrfs_root *tree_root = fs_info->tree_root;
1253 
1254 	old_root_used = btrfs_root_used(&root->root_item);
1255 
1256 	while (1) {
1257 		old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1258 		if (old_root_bytenr == root->node->start &&
1259 		    old_root_used == btrfs_root_used(&root->root_item))
1260 			break;
1261 
1262 		btrfs_set_root_node(&root->root_item, root->node);
1263 		ret = btrfs_update_root(trans, tree_root,
1264 					&root->root_key,
1265 					&root->root_item);
1266 		if (ret)
1267 			return ret;
1268 
1269 		old_root_used = btrfs_root_used(&root->root_item);
1270 	}
1271 
1272 	return 0;
1273 }
1274 
1275 /*
1276  * update all the cowonly tree roots on disk
1277  *
1278  * The error handling in this function may not be obvious. Any of the
1279  * failures will cause the file system to go offline. We still need
1280  * to clean up the delayed refs.
1281  */
commit_cowonly_roots(struct btrfs_trans_handle * trans)1282 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans)
1283 {
1284 	struct btrfs_fs_info *fs_info = trans->fs_info;
1285 	struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1286 	struct list_head *io_bgs = &trans->transaction->io_bgs;
1287 	struct list_head *next;
1288 	struct extent_buffer *eb;
1289 	int ret;
1290 
1291 	/*
1292 	 * At this point no one can be using this transaction to modify any tree
1293 	 * and no one can start another transaction to modify any tree either.
1294 	 */
1295 	ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);
1296 
1297 	eb = btrfs_lock_root_node(fs_info->tree_root);
1298 	ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1299 			      0, &eb, BTRFS_NESTING_COW);
1300 	btrfs_tree_unlock(eb);
1301 	free_extent_buffer(eb);
1302 
1303 	if (ret)
1304 		return ret;
1305 
1306 	ret = btrfs_run_dev_stats(trans);
1307 	if (ret)
1308 		return ret;
1309 	ret = btrfs_run_dev_replace(trans);
1310 	if (ret)
1311 		return ret;
1312 	ret = btrfs_run_qgroups(trans);
1313 	if (ret)
1314 		return ret;
1315 
1316 	ret = btrfs_setup_space_cache(trans);
1317 	if (ret)
1318 		return ret;
1319 
1320 again:
1321 	while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1322 		struct btrfs_root *root;
1323 		next = fs_info->dirty_cowonly_roots.next;
1324 		list_del_init(next);
1325 		root = list_entry(next, struct btrfs_root, dirty_list);
1326 		clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1327 
1328 		list_add_tail(&root->dirty_list,
1329 			      &trans->transaction->switch_commits);
1330 		ret = update_cowonly_root(trans, root);
1331 		if (ret)
1332 			return ret;
1333 	}
1334 
1335 	/* Now flush any delayed refs generated by updating all of the roots */
1336 	ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1337 	if (ret)
1338 		return ret;
1339 
1340 	while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1341 		ret = btrfs_write_dirty_block_groups(trans);
1342 		if (ret)
1343 			return ret;
1344 
1345 		/*
1346 		 * We're writing the dirty block groups, which could generate
1347 		 * delayed refs, which could generate more dirty block groups,
1348 		 * so we want to keep this flushing in this loop to make sure
1349 		 * everything gets run.
1350 		 */
1351 		ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1352 		if (ret)
1353 			return ret;
1354 	}
1355 
1356 	if (!list_empty(&fs_info->dirty_cowonly_roots))
1357 		goto again;
1358 
1359 	/* Update dev-replace pointer once everything is committed */
1360 	fs_info->dev_replace.committed_cursor_left =
1361 		fs_info->dev_replace.cursor_left_last_write_of_item;
1362 
1363 	return 0;
1364 }
1365 
1366 /*
1367  * If we had a pending drop we need to see if there are any others left in our
1368  * dead roots list, and if not clear our bit and wake any waiters.
1369  */
btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info * fs_info)1370 void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info)
1371 {
1372 	/*
1373 	 * We put the drop in progress roots at the front of the list, so if the
1374 	 * first entry doesn't have UNFINISHED_DROP set we can wake everybody
1375 	 * up.
1376 	 */
1377 	spin_lock(&fs_info->trans_lock);
1378 	if (!list_empty(&fs_info->dead_roots)) {
1379 		struct btrfs_root *root = list_first_entry(&fs_info->dead_roots,
1380 							   struct btrfs_root,
1381 							   root_list);
1382 		if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) {
1383 			spin_unlock(&fs_info->trans_lock);
1384 			return;
1385 		}
1386 	}
1387 	spin_unlock(&fs_info->trans_lock);
1388 
1389 	btrfs_wake_unfinished_drop(fs_info);
1390 }
1391 
1392 /*
1393  * dead roots are old snapshots that need to be deleted.  This allocates
1394  * a dirty root struct and adds it into the list of dead roots that need to
1395  * be deleted
1396  */
btrfs_add_dead_root(struct btrfs_root * root)1397 void btrfs_add_dead_root(struct btrfs_root *root)
1398 {
1399 	struct btrfs_fs_info *fs_info = root->fs_info;
1400 
1401 	spin_lock(&fs_info->trans_lock);
1402 	if (list_empty(&root->root_list)) {
1403 		btrfs_grab_root(root);
1404 
1405 		/* We want to process the partially complete drops first. */
1406 		if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state))
1407 			list_add(&root->root_list, &fs_info->dead_roots);
1408 		else
1409 			list_add_tail(&root->root_list, &fs_info->dead_roots);
1410 	}
1411 	spin_unlock(&fs_info->trans_lock);
1412 }
1413 
1414 /*
1415  * Update each subvolume root and its relocation root, if it exists, in the tree
1416  * of tree roots. Also free log roots if they exist.
1417  */
commit_fs_roots(struct btrfs_trans_handle * trans)1418 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
1419 {
1420 	struct btrfs_fs_info *fs_info = trans->fs_info;
1421 	struct btrfs_root *gang[8];
1422 	int i;
1423 	int ret;
1424 
1425 	/*
1426 	 * At this point no one can be using this transaction to modify any tree
1427 	 * and no one can start another transaction to modify any tree either.
1428 	 */
1429 	ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);
1430 
1431 	spin_lock(&fs_info->fs_roots_radix_lock);
1432 	while (1) {
1433 		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1434 						 (void **)gang, 0,
1435 						 ARRAY_SIZE(gang),
1436 						 BTRFS_ROOT_TRANS_TAG);
1437 		if (ret == 0)
1438 			break;
1439 		for (i = 0; i < ret; i++) {
1440 			struct btrfs_root *root = gang[i];
1441 			int ret2;
1442 
1443 			/*
1444 			 * At this point we can neither have tasks logging inodes
1445 			 * from a root nor trying to commit a log tree.
1446 			 */
1447 			ASSERT(atomic_read(&root->log_writers) == 0);
1448 			ASSERT(atomic_read(&root->log_commit[0]) == 0);
1449 			ASSERT(atomic_read(&root->log_commit[1]) == 0);
1450 
1451 			radix_tree_tag_clear(&fs_info->fs_roots_radix,
1452 					(unsigned long)root->root_key.objectid,
1453 					BTRFS_ROOT_TRANS_TAG);
1454 			btrfs_qgroup_free_meta_all_pertrans(root);
1455 			spin_unlock(&fs_info->fs_roots_radix_lock);
1456 
1457 			btrfs_free_log(trans, root);
1458 			ret2 = btrfs_update_reloc_root(trans, root);
1459 			if (ret2)
1460 				return ret2;
1461 
1462 			/* see comments in should_cow_block() */
1463 			clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1464 			smp_mb__after_atomic();
1465 
1466 			if (root->commit_root != root->node) {
1467 				list_add_tail(&root->dirty_list,
1468 					&trans->transaction->switch_commits);
1469 				btrfs_set_root_node(&root->root_item,
1470 						    root->node);
1471 			}
1472 
1473 			ret2 = btrfs_update_root(trans, fs_info->tree_root,
1474 						&root->root_key,
1475 						&root->root_item);
1476 			if (ret2)
1477 				return ret2;
1478 			spin_lock(&fs_info->fs_roots_radix_lock);
1479 		}
1480 	}
1481 	spin_unlock(&fs_info->fs_roots_radix_lock);
1482 	return 0;
1483 }
1484 
1485 /*
1486  * defrag a given btree.
1487  * Every leaf in the btree is read and defragged.
1488  */
btrfs_defrag_root(struct btrfs_root * root)1489 int btrfs_defrag_root(struct btrfs_root *root)
1490 {
1491 	struct btrfs_fs_info *info = root->fs_info;
1492 	struct btrfs_trans_handle *trans;
1493 	int ret;
1494 
1495 	if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1496 		return 0;
1497 
1498 	while (1) {
1499 		trans = btrfs_start_transaction(root, 0);
1500 		if (IS_ERR(trans)) {
1501 			ret = PTR_ERR(trans);
1502 			break;
1503 		}
1504 
1505 		ret = btrfs_defrag_leaves(trans, root);
1506 
1507 		btrfs_end_transaction(trans);
1508 		btrfs_btree_balance_dirty(info);
1509 		cond_resched();
1510 
1511 		if (btrfs_fs_closing(info) || ret != -EAGAIN)
1512 			break;
1513 
1514 		if (btrfs_defrag_cancelled(info)) {
1515 			btrfs_debug(info, "defrag_root cancelled");
1516 			ret = -EAGAIN;
1517 			break;
1518 		}
1519 	}
1520 	clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1521 	return ret;
1522 }
1523 
1524 /*
1525  * Do all special snapshot related qgroup dirty hack.
1526  *
1527  * Will do all needed qgroup inherit and dirty hack like switch commit
1528  * roots inside one transaction and write all btree into disk, to make
1529  * qgroup works.
1530  */
qgroup_account_snapshot(struct btrfs_trans_handle * trans,struct btrfs_root * src,struct btrfs_root * parent,struct btrfs_qgroup_inherit * inherit,u64 dst_objectid)1531 static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1532 				   struct btrfs_root *src,
1533 				   struct btrfs_root *parent,
1534 				   struct btrfs_qgroup_inherit *inherit,
1535 				   u64 dst_objectid)
1536 {
1537 	struct btrfs_fs_info *fs_info = src->fs_info;
1538 	int ret;
1539 
1540 	/*
1541 	 * Save some performance in the case that qgroups are not
1542 	 * enabled. If this check races with the ioctl, rescan will
1543 	 * kick in anyway.
1544 	 */
1545 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1546 		return 0;
1547 
1548 	/*
1549 	 * Ensure dirty @src will be committed.  Or, after coming
1550 	 * commit_fs_roots() and switch_commit_roots(), any dirty but not
1551 	 * recorded root will never be updated again, causing an outdated root
1552 	 * item.
1553 	 */
1554 	ret = record_root_in_trans(trans, src, 1);
1555 	if (ret)
1556 		return ret;
1557 
1558 	/*
1559 	 * btrfs_qgroup_inherit relies on a consistent view of the usage for the
1560 	 * src root, so we must run the delayed refs here.
1561 	 *
1562 	 * However this isn't particularly fool proof, because there's no
1563 	 * synchronization keeping us from changing the tree after this point
1564 	 * before we do the qgroup_inherit, or even from making changes while
1565 	 * we're doing the qgroup_inherit.  But that's a problem for the future,
1566 	 * for now flush the delayed refs to narrow the race window where the
1567 	 * qgroup counters could end up wrong.
1568 	 */
1569 	ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1570 	if (ret) {
1571 		btrfs_abort_transaction(trans, ret);
1572 		return ret;
1573 	}
1574 
1575 	ret = commit_fs_roots(trans);
1576 	if (ret)
1577 		goto out;
1578 	ret = btrfs_qgroup_account_extents(trans);
1579 	if (ret < 0)
1580 		goto out;
1581 
1582 	/* Now qgroup are all updated, we can inherit it to new qgroups */
1583 	ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid,
1584 				   inherit);
1585 	if (ret < 0)
1586 		goto out;
1587 
1588 	/*
1589 	 * Now we do a simplified commit transaction, which will:
1590 	 * 1) commit all subvolume and extent tree
1591 	 *    To ensure all subvolume and extent tree have a valid
1592 	 *    commit_root to accounting later insert_dir_item()
1593 	 * 2) write all btree blocks onto disk
1594 	 *    This is to make sure later btree modification will be cowed
1595 	 *    Or commit_root can be populated and cause wrong qgroup numbers
1596 	 * In this simplified commit, we don't really care about other trees
1597 	 * like chunk and root tree, as they won't affect qgroup.
1598 	 * And we don't write super to avoid half committed status.
1599 	 */
1600 	ret = commit_cowonly_roots(trans);
1601 	if (ret)
1602 		goto out;
1603 	switch_commit_roots(trans);
1604 	ret = btrfs_write_and_wait_transaction(trans);
1605 	if (ret)
1606 		btrfs_handle_fs_error(fs_info, ret,
1607 			"Error while writing out transaction for qgroup");
1608 
1609 out:
1610 	/*
1611 	 * Force parent root to be updated, as we recorded it before so its
1612 	 * last_trans == cur_transid.
1613 	 * Or it won't be committed again onto disk after later
1614 	 * insert_dir_item()
1615 	 */
1616 	if (!ret)
1617 		ret = record_root_in_trans(trans, parent, 1);
1618 	return ret;
1619 }
1620 
1621 /*
1622  * new snapshots need to be created at a very specific time in the
1623  * transaction commit.  This does the actual creation.
1624  *
1625  * Note:
1626  * If the error which may affect the commitment of the current transaction
1627  * happens, we should return the error number. If the error which just affect
1628  * the creation of the pending snapshots, just return 0.
1629  */
create_pending_snapshot(struct btrfs_trans_handle * trans,struct btrfs_pending_snapshot * pending)1630 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1631 				   struct btrfs_pending_snapshot *pending)
1632 {
1633 
1634 	struct btrfs_fs_info *fs_info = trans->fs_info;
1635 	struct btrfs_key key;
1636 	struct btrfs_root_item *new_root_item;
1637 	struct btrfs_root *tree_root = fs_info->tree_root;
1638 	struct btrfs_root *root = pending->root;
1639 	struct btrfs_root *parent_root;
1640 	struct btrfs_block_rsv *rsv;
1641 	struct inode *parent_inode = pending->dir;
1642 	struct btrfs_path *path;
1643 	struct btrfs_dir_item *dir_item;
1644 	struct extent_buffer *tmp;
1645 	struct extent_buffer *old;
1646 	struct timespec64 cur_time;
1647 	int ret = 0;
1648 	u64 to_reserve = 0;
1649 	u64 index = 0;
1650 	u64 objectid;
1651 	u64 root_flags;
1652 	unsigned int nofs_flags;
1653 	struct fscrypt_name fname;
1654 
1655 	ASSERT(pending->path);
1656 	path = pending->path;
1657 
1658 	ASSERT(pending->root_item);
1659 	new_root_item = pending->root_item;
1660 
1661 	/*
1662 	 * We're inside a transaction and must make sure that any potential
1663 	 * allocations with GFP_KERNEL in fscrypt won't recurse back to
1664 	 * filesystem.
1665 	 */
1666 	nofs_flags = memalloc_nofs_save();
1667 	pending->error = fscrypt_setup_filename(parent_inode,
1668 						&pending->dentry->d_name, 0,
1669 						&fname);
1670 	memalloc_nofs_restore(nofs_flags);
1671 	if (pending->error)
1672 		goto free_pending;
1673 
1674 	pending->error = btrfs_get_free_objectid(tree_root, &objectid);
1675 	if (pending->error)
1676 		goto free_fname;
1677 
1678 	/*
1679 	 * Make qgroup to skip current new snapshot's qgroupid, as it is
1680 	 * accounted by later btrfs_qgroup_inherit().
1681 	 */
1682 	btrfs_set_skip_qgroup(trans, objectid);
1683 
1684 	btrfs_reloc_pre_snapshot(pending, &to_reserve);
1685 
1686 	if (to_reserve > 0) {
1687 		pending->error = btrfs_block_rsv_add(fs_info,
1688 						     &pending->block_rsv,
1689 						     to_reserve,
1690 						     BTRFS_RESERVE_NO_FLUSH);
1691 		if (pending->error)
1692 			goto clear_skip_qgroup;
1693 	}
1694 
1695 	key.objectid = objectid;
1696 	key.offset = (u64)-1;
1697 	key.type = BTRFS_ROOT_ITEM_KEY;
1698 
1699 	rsv = trans->block_rsv;
1700 	trans->block_rsv = &pending->block_rsv;
1701 	trans->bytes_reserved = trans->block_rsv->reserved;
1702 	trace_btrfs_space_reservation(fs_info, "transaction",
1703 				      trans->transid,
1704 				      trans->bytes_reserved, 1);
1705 	parent_root = BTRFS_I(parent_inode)->root;
1706 	ret = record_root_in_trans(trans, parent_root, 0);
1707 	if (ret)
1708 		goto fail;
1709 	cur_time = current_time(parent_inode);
1710 
1711 	/*
1712 	 * insert the directory item
1713 	 */
1714 	ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index);
1715 	if (ret) {
1716 		btrfs_abort_transaction(trans, ret);
1717 		goto fail;
1718 	}
1719 
1720 	/* check if there is a file/dir which has the same name. */
1721 	dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1722 					 btrfs_ino(BTRFS_I(parent_inode)),
1723 					 &fname.disk_name, 0);
1724 	if (dir_item != NULL && !IS_ERR(dir_item)) {
1725 		pending->error = -EEXIST;
1726 		goto dir_item_existed;
1727 	} else if (IS_ERR(dir_item)) {
1728 		ret = PTR_ERR(dir_item);
1729 		btrfs_abort_transaction(trans, ret);
1730 		goto fail;
1731 	}
1732 	btrfs_release_path(path);
1733 
1734 	/*
1735 	 * pull in the delayed directory update
1736 	 * and the delayed inode item
1737 	 * otherwise we corrupt the FS during
1738 	 * snapshot
1739 	 */
1740 	ret = btrfs_run_delayed_items(trans);
1741 	if (ret) {	/* Transaction aborted */
1742 		btrfs_abort_transaction(trans, ret);
1743 		goto fail;
1744 	}
1745 
1746 	ret = record_root_in_trans(trans, root, 0);
1747 	if (ret) {
1748 		btrfs_abort_transaction(trans, ret);
1749 		goto fail;
1750 	}
1751 	btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1752 	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1753 	btrfs_check_and_init_root_item(new_root_item);
1754 
1755 	root_flags = btrfs_root_flags(new_root_item);
1756 	if (pending->readonly)
1757 		root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1758 	else
1759 		root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1760 	btrfs_set_root_flags(new_root_item, root_flags);
1761 
1762 	btrfs_set_root_generation_v2(new_root_item,
1763 			trans->transid);
1764 	generate_random_guid(new_root_item->uuid);
1765 	memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1766 			BTRFS_UUID_SIZE);
1767 	if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1768 		memset(new_root_item->received_uuid, 0,
1769 		       sizeof(new_root_item->received_uuid));
1770 		memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1771 		memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1772 		btrfs_set_root_stransid(new_root_item, 0);
1773 		btrfs_set_root_rtransid(new_root_item, 0);
1774 	}
1775 	btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1776 	btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1777 	btrfs_set_root_otransid(new_root_item, trans->transid);
1778 
1779 	old = btrfs_lock_root_node(root);
1780 	ret = btrfs_cow_block(trans, root, old, NULL, 0, &old,
1781 			      BTRFS_NESTING_COW);
1782 	if (ret) {
1783 		btrfs_tree_unlock(old);
1784 		free_extent_buffer(old);
1785 		btrfs_abort_transaction(trans, ret);
1786 		goto fail;
1787 	}
1788 
1789 	ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1790 	/* clean up in any case */
1791 	btrfs_tree_unlock(old);
1792 	free_extent_buffer(old);
1793 	if (ret) {
1794 		btrfs_abort_transaction(trans, ret);
1795 		goto fail;
1796 	}
1797 	/* see comments in should_cow_block() */
1798 	set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1799 	smp_wmb();
1800 
1801 	btrfs_set_root_node(new_root_item, tmp);
1802 	/* record when the snapshot was created in key.offset */
1803 	key.offset = trans->transid;
1804 	ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1805 	btrfs_tree_unlock(tmp);
1806 	free_extent_buffer(tmp);
1807 	if (ret) {
1808 		btrfs_abort_transaction(trans, ret);
1809 		goto fail;
1810 	}
1811 
1812 	/*
1813 	 * insert root back/forward references
1814 	 */
1815 	ret = btrfs_add_root_ref(trans, objectid,
1816 				 parent_root->root_key.objectid,
1817 				 btrfs_ino(BTRFS_I(parent_inode)), index,
1818 				 &fname.disk_name);
1819 	if (ret) {
1820 		btrfs_abort_transaction(trans, ret);
1821 		goto fail;
1822 	}
1823 
1824 	key.offset = (u64)-1;
1825 	pending->snap = btrfs_get_new_fs_root(fs_info, objectid, &pending->anon_dev);
1826 	if (IS_ERR(pending->snap)) {
1827 		ret = PTR_ERR(pending->snap);
1828 		pending->snap = NULL;
1829 		btrfs_abort_transaction(trans, ret);
1830 		goto fail;
1831 	}
1832 
1833 	ret = btrfs_reloc_post_snapshot(trans, pending);
1834 	if (ret) {
1835 		btrfs_abort_transaction(trans, ret);
1836 		goto fail;
1837 	}
1838 
1839 	/*
1840 	 * Do special qgroup accounting for snapshot, as we do some qgroup
1841 	 * snapshot hack to do fast snapshot.
1842 	 * To co-operate with that hack, we do hack again.
1843 	 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1844 	 */
1845 	ret = qgroup_account_snapshot(trans, root, parent_root,
1846 				      pending->inherit, objectid);
1847 	if (ret < 0)
1848 		goto fail;
1849 
1850 	ret = btrfs_insert_dir_item(trans, &fname.disk_name,
1851 				    BTRFS_I(parent_inode), &key, BTRFS_FT_DIR,
1852 				    index);
1853 	/* We have check then name at the beginning, so it is impossible. */
1854 	BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1855 	if (ret) {
1856 		btrfs_abort_transaction(trans, ret);
1857 		goto fail;
1858 	}
1859 
1860 	btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size +
1861 						  fname.disk_name.len * 2);
1862 	parent_inode->i_mtime = inode_set_ctime_current(parent_inode);
1863 	ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode));
1864 	if (ret) {
1865 		btrfs_abort_transaction(trans, ret);
1866 		goto fail;
1867 	}
1868 	ret = btrfs_uuid_tree_add(trans, new_root_item->uuid,
1869 				  BTRFS_UUID_KEY_SUBVOL,
1870 				  objectid);
1871 	if (ret) {
1872 		btrfs_abort_transaction(trans, ret);
1873 		goto fail;
1874 	}
1875 	if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1876 		ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid,
1877 					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1878 					  objectid);
1879 		if (ret && ret != -EEXIST) {
1880 			btrfs_abort_transaction(trans, ret);
1881 			goto fail;
1882 		}
1883 	}
1884 
1885 fail:
1886 	pending->error = ret;
1887 dir_item_existed:
1888 	trans->block_rsv = rsv;
1889 	trans->bytes_reserved = 0;
1890 clear_skip_qgroup:
1891 	btrfs_clear_skip_qgroup(trans);
1892 free_fname:
1893 	fscrypt_free_filename(&fname);
1894 free_pending:
1895 	kfree(new_root_item);
1896 	pending->root_item = NULL;
1897 	btrfs_free_path(path);
1898 	pending->path = NULL;
1899 
1900 	return ret;
1901 }
1902 
1903 /*
1904  * create all the snapshots we've scheduled for creation
1905  */
create_pending_snapshots(struct btrfs_trans_handle * trans)1906 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans)
1907 {
1908 	struct btrfs_pending_snapshot *pending, *next;
1909 	struct list_head *head = &trans->transaction->pending_snapshots;
1910 	int ret = 0;
1911 
1912 	list_for_each_entry_safe(pending, next, head, list) {
1913 		list_del(&pending->list);
1914 		ret = create_pending_snapshot(trans, pending);
1915 		if (ret)
1916 			break;
1917 	}
1918 	return ret;
1919 }
1920 
update_super_roots(struct btrfs_fs_info * fs_info)1921 static void update_super_roots(struct btrfs_fs_info *fs_info)
1922 {
1923 	struct btrfs_root_item *root_item;
1924 	struct btrfs_super_block *super;
1925 
1926 	super = fs_info->super_copy;
1927 
1928 	root_item = &fs_info->chunk_root->root_item;
1929 	super->chunk_root = root_item->bytenr;
1930 	super->chunk_root_generation = root_item->generation;
1931 	super->chunk_root_level = root_item->level;
1932 
1933 	root_item = &fs_info->tree_root->root_item;
1934 	super->root = root_item->bytenr;
1935 	super->generation = root_item->generation;
1936 	super->root_level = root_item->level;
1937 	if (btrfs_test_opt(fs_info, SPACE_CACHE))
1938 		super->cache_generation = root_item->generation;
1939 	else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags))
1940 		super->cache_generation = 0;
1941 	if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags))
1942 		super->uuid_tree_generation = root_item->generation;
1943 }
1944 
btrfs_transaction_in_commit(struct btrfs_fs_info * info)1945 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1946 {
1947 	struct btrfs_transaction *trans;
1948 	int ret = 0;
1949 
1950 	spin_lock(&info->trans_lock);
1951 	trans = info->running_transaction;
1952 	if (trans)
1953 		ret = (trans->state >= TRANS_STATE_COMMIT_START);
1954 	spin_unlock(&info->trans_lock);
1955 	return ret;
1956 }
1957 
btrfs_transaction_blocked(struct btrfs_fs_info * info)1958 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1959 {
1960 	struct btrfs_transaction *trans;
1961 	int ret = 0;
1962 
1963 	spin_lock(&info->trans_lock);
1964 	trans = info->running_transaction;
1965 	if (trans)
1966 		ret = is_transaction_blocked(trans);
1967 	spin_unlock(&info->trans_lock);
1968 	return ret;
1969 }
1970 
btrfs_commit_transaction_async(struct btrfs_trans_handle * trans)1971 void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans)
1972 {
1973 	struct btrfs_fs_info *fs_info = trans->fs_info;
1974 	struct btrfs_transaction *cur_trans;
1975 
1976 	/* Kick the transaction kthread. */
1977 	set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags);
1978 	wake_up_process(fs_info->transaction_kthread);
1979 
1980 	/* take transaction reference */
1981 	cur_trans = trans->transaction;
1982 	refcount_inc(&cur_trans->use_count);
1983 
1984 	btrfs_end_transaction(trans);
1985 
1986 	/*
1987 	 * Wait for the current transaction commit to start and block
1988 	 * subsequent transaction joins
1989 	 */
1990 	btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
1991 	wait_event(fs_info->transaction_blocked_wait,
1992 		   cur_trans->state >= TRANS_STATE_COMMIT_START ||
1993 		   TRANS_ABORTED(cur_trans));
1994 	btrfs_put_transaction(cur_trans);
1995 }
1996 
cleanup_transaction(struct btrfs_trans_handle * trans,int err)1997 static void cleanup_transaction(struct btrfs_trans_handle *trans, int err)
1998 {
1999 	struct btrfs_fs_info *fs_info = trans->fs_info;
2000 	struct btrfs_transaction *cur_trans = trans->transaction;
2001 
2002 	WARN_ON(refcount_read(&trans->use_count) > 1);
2003 
2004 	btrfs_abort_transaction(trans, err);
2005 
2006 	spin_lock(&fs_info->trans_lock);
2007 
2008 	/*
2009 	 * If the transaction is removed from the list, it means this
2010 	 * transaction has been committed successfully, so it is impossible
2011 	 * to call the cleanup function.
2012 	 */
2013 	BUG_ON(list_empty(&cur_trans->list));
2014 
2015 	if (cur_trans == fs_info->running_transaction) {
2016 		cur_trans->state = TRANS_STATE_COMMIT_DOING;
2017 		spin_unlock(&fs_info->trans_lock);
2018 
2019 		/*
2020 		 * The thread has already released the lockdep map as reader
2021 		 * already in btrfs_commit_transaction().
2022 		 */
2023 		btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers);
2024 		wait_event(cur_trans->writer_wait,
2025 			   atomic_read(&cur_trans->num_writers) == 1);
2026 
2027 		spin_lock(&fs_info->trans_lock);
2028 	}
2029 
2030 	/*
2031 	 * Now that we know no one else is still using the transaction we can
2032 	 * remove the transaction from the list of transactions. This avoids
2033 	 * the transaction kthread from cleaning up the transaction while some
2034 	 * other task is still using it, which could result in a use-after-free
2035 	 * on things like log trees, as it forces the transaction kthread to
2036 	 * wait for this transaction to be cleaned up by us.
2037 	 */
2038 	list_del_init(&cur_trans->list);
2039 
2040 	spin_unlock(&fs_info->trans_lock);
2041 
2042 	btrfs_cleanup_one_transaction(trans->transaction, fs_info);
2043 
2044 	spin_lock(&fs_info->trans_lock);
2045 	if (cur_trans == fs_info->running_transaction)
2046 		fs_info->running_transaction = NULL;
2047 	spin_unlock(&fs_info->trans_lock);
2048 
2049 	if (trans->type & __TRANS_FREEZABLE)
2050 		sb_end_intwrite(fs_info->sb);
2051 	btrfs_put_transaction(cur_trans);
2052 	btrfs_put_transaction(cur_trans);
2053 
2054 	trace_btrfs_transaction_commit(fs_info);
2055 
2056 	if (current->journal_info == trans)
2057 		current->journal_info = NULL;
2058 
2059 	/*
2060 	 * If relocation is running, we can't cancel scrub because that will
2061 	 * result in a deadlock. Before relocating a block group, relocation
2062 	 * pauses scrub, then starts and commits a transaction before unpausing
2063 	 * scrub. If the transaction commit is being done by the relocation
2064 	 * task or triggered by another task and the relocation task is waiting
2065 	 * for the commit, and we end up here due to an error in the commit
2066 	 * path, then calling btrfs_scrub_cancel() will deadlock, as we are
2067 	 * asking for scrub to stop while having it asked to be paused higher
2068 	 * above in relocation code.
2069 	 */
2070 	if (!test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
2071 		btrfs_scrub_cancel(fs_info);
2072 
2073 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
2074 }
2075 
2076 /*
2077  * Release reserved delayed ref space of all pending block groups of the
2078  * transaction and remove them from the list
2079  */
btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle * trans)2080 static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans)
2081 {
2082        struct btrfs_fs_info *fs_info = trans->fs_info;
2083        struct btrfs_block_group *block_group, *tmp;
2084 
2085        list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
2086                btrfs_delayed_refs_rsv_release(fs_info, 1);
2087                list_del_init(&block_group->bg_list);
2088        }
2089 }
2090 
btrfs_start_delalloc_flush(struct btrfs_fs_info * fs_info)2091 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
2092 {
2093 	/*
2094 	 * We use try_to_writeback_inodes_sb() here because if we used
2095 	 * btrfs_start_delalloc_roots we would deadlock with fs freeze.
2096 	 * Currently are holding the fs freeze lock, if we do an async flush
2097 	 * we'll do btrfs_join_transaction() and deadlock because we need to
2098 	 * wait for the fs freeze lock.  Using the direct flushing we benefit
2099 	 * from already being in a transaction and our join_transaction doesn't
2100 	 * have to re-take the fs freeze lock.
2101 	 *
2102 	 * Note that try_to_writeback_inodes_sb() will only trigger writeback
2103 	 * if it can read lock sb->s_umount. It will always be able to lock it,
2104 	 * except when the filesystem is being unmounted or being frozen, but in
2105 	 * those cases sync_filesystem() is called, which results in calling
2106 	 * writeback_inodes_sb() while holding a write lock on sb->s_umount.
2107 	 * Note that we don't call writeback_inodes_sb() directly, because it
2108 	 * will emit a warning if sb->s_umount is not locked.
2109 	 */
2110 	if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2111 		try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC);
2112 	return 0;
2113 }
2114 
btrfs_wait_delalloc_flush(struct btrfs_fs_info * fs_info)2115 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
2116 {
2117 	if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2118 		btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
2119 }
2120 
2121 /*
2122  * Add a pending snapshot associated with the given transaction handle to the
2123  * respective handle. This must be called after the transaction commit started
2124  * and while holding fs_info->trans_lock.
2125  * This serves to guarantee a caller of btrfs_commit_transaction() that it can
2126  * safely free the pending snapshot pointer in case btrfs_commit_transaction()
2127  * returns an error.
2128  */
add_pending_snapshot(struct btrfs_trans_handle * trans)2129 static void add_pending_snapshot(struct btrfs_trans_handle *trans)
2130 {
2131 	struct btrfs_transaction *cur_trans = trans->transaction;
2132 
2133 	if (!trans->pending_snapshot)
2134 		return;
2135 
2136 	lockdep_assert_held(&trans->fs_info->trans_lock);
2137 	ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_PREP);
2138 
2139 	list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots);
2140 }
2141 
update_commit_stats(struct btrfs_fs_info * fs_info,ktime_t interval)2142 static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval)
2143 {
2144 	fs_info->commit_stats.commit_count++;
2145 	fs_info->commit_stats.last_commit_dur = interval;
2146 	fs_info->commit_stats.max_commit_dur =
2147 			max_t(u64, fs_info->commit_stats.max_commit_dur, interval);
2148 	fs_info->commit_stats.total_commit_dur += interval;
2149 }
2150 
btrfs_commit_transaction(struct btrfs_trans_handle * trans)2151 int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
2152 {
2153 	struct btrfs_fs_info *fs_info = trans->fs_info;
2154 	struct btrfs_transaction *cur_trans = trans->transaction;
2155 	struct btrfs_transaction *prev_trans = NULL;
2156 	int ret;
2157 	ktime_t start_time;
2158 	ktime_t interval;
2159 
2160 	ASSERT(refcount_read(&trans->use_count) == 1);
2161 	btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2162 
2163 	clear_bit(BTRFS_FS_NEED_TRANS_COMMIT, &fs_info->flags);
2164 
2165 	/* Stop the commit early if ->aborted is set */
2166 	if (TRANS_ABORTED(cur_trans)) {
2167 		ret = cur_trans->aborted;
2168 		goto lockdep_trans_commit_start_release;
2169 	}
2170 
2171 	btrfs_trans_release_metadata(trans);
2172 	trans->block_rsv = NULL;
2173 
2174 	/*
2175 	 * We only want one transaction commit doing the flushing so we do not
2176 	 * waste a bunch of time on lock contention on the extent root node.
2177 	 */
2178 	if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING,
2179 			      &cur_trans->delayed_refs.flags)) {
2180 		/*
2181 		 * Make a pass through all the delayed refs we have so far.
2182 		 * Any running threads may add more while we are here.
2183 		 */
2184 		ret = btrfs_run_delayed_refs(trans, 0);
2185 		if (ret)
2186 			goto lockdep_trans_commit_start_release;
2187 	}
2188 
2189 	btrfs_create_pending_block_groups(trans);
2190 
2191 	if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
2192 		int run_it = 0;
2193 
2194 		/* this mutex is also taken before trying to set
2195 		 * block groups readonly.  We need to make sure
2196 		 * that nobody has set a block group readonly
2197 		 * after a extents from that block group have been
2198 		 * allocated for cache files.  btrfs_set_block_group_ro
2199 		 * will wait for the transaction to commit if it
2200 		 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
2201 		 *
2202 		 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
2203 		 * only one process starts all the block group IO.  It wouldn't
2204 		 * hurt to have more than one go through, but there's no
2205 		 * real advantage to it either.
2206 		 */
2207 		mutex_lock(&fs_info->ro_block_group_mutex);
2208 		if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
2209 				      &cur_trans->flags))
2210 			run_it = 1;
2211 		mutex_unlock(&fs_info->ro_block_group_mutex);
2212 
2213 		if (run_it) {
2214 			ret = btrfs_start_dirty_block_groups(trans);
2215 			if (ret)
2216 				goto lockdep_trans_commit_start_release;
2217 		}
2218 	}
2219 
2220 	spin_lock(&fs_info->trans_lock);
2221 	if (cur_trans->state >= TRANS_STATE_COMMIT_PREP) {
2222 		enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2223 
2224 		add_pending_snapshot(trans);
2225 
2226 		spin_unlock(&fs_info->trans_lock);
2227 		refcount_inc(&cur_trans->use_count);
2228 
2229 		if (trans->in_fsync)
2230 			want_state = TRANS_STATE_SUPER_COMMITTED;
2231 
2232 		btrfs_trans_state_lockdep_release(fs_info,
2233 						  BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2234 		ret = btrfs_end_transaction(trans);
2235 		wait_for_commit(cur_trans, want_state);
2236 
2237 		if (TRANS_ABORTED(cur_trans))
2238 			ret = cur_trans->aborted;
2239 
2240 		btrfs_put_transaction(cur_trans);
2241 
2242 		return ret;
2243 	}
2244 
2245 	cur_trans->state = TRANS_STATE_COMMIT_PREP;
2246 	wake_up(&fs_info->transaction_blocked_wait);
2247 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2248 
2249 	if (cur_trans->list.prev != &fs_info->trans_list) {
2250 		enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2251 
2252 		if (trans->in_fsync)
2253 			want_state = TRANS_STATE_SUPER_COMMITTED;
2254 
2255 		prev_trans = list_entry(cur_trans->list.prev,
2256 					struct btrfs_transaction, list);
2257 		if (prev_trans->state < want_state) {
2258 			refcount_inc(&prev_trans->use_count);
2259 			spin_unlock(&fs_info->trans_lock);
2260 
2261 			wait_for_commit(prev_trans, want_state);
2262 
2263 			ret = READ_ONCE(prev_trans->aborted);
2264 
2265 			btrfs_put_transaction(prev_trans);
2266 			if (ret)
2267 				goto lockdep_release;
2268 			spin_lock(&fs_info->trans_lock);
2269 		}
2270 	} else {
2271 		/*
2272 		 * The previous transaction was aborted and was already removed
2273 		 * from the list of transactions at fs_info->trans_list. So we
2274 		 * abort to prevent writing a new superblock that reflects a
2275 		 * corrupt state (pointing to trees with unwritten nodes/leafs).
2276 		 */
2277 		if (BTRFS_FS_ERROR(fs_info)) {
2278 			spin_unlock(&fs_info->trans_lock);
2279 			ret = -EROFS;
2280 			goto lockdep_release;
2281 		}
2282 	}
2283 
2284 	cur_trans->state = TRANS_STATE_COMMIT_START;
2285 	wake_up(&fs_info->transaction_blocked_wait);
2286 	spin_unlock(&fs_info->trans_lock);
2287 
2288 	/*
2289 	 * Get the time spent on the work done by the commit thread and not
2290 	 * the time spent waiting on a previous commit
2291 	 */
2292 	start_time = ktime_get_ns();
2293 
2294 	extwriter_counter_dec(cur_trans, trans->type);
2295 
2296 	ret = btrfs_start_delalloc_flush(fs_info);
2297 	if (ret)
2298 		goto lockdep_release;
2299 
2300 	ret = btrfs_run_delayed_items(trans);
2301 	if (ret)
2302 		goto lockdep_release;
2303 
2304 	/*
2305 	 * The thread has started/joined the transaction thus it holds the
2306 	 * lockdep map as a reader. It has to release it before acquiring the
2307 	 * lockdep map as a writer.
2308 	 */
2309 	btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
2310 	btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters);
2311 	wait_event(cur_trans->writer_wait,
2312 		   extwriter_counter_read(cur_trans) == 0);
2313 
2314 	/* some pending stuffs might be added after the previous flush. */
2315 	ret = btrfs_run_delayed_items(trans);
2316 	if (ret) {
2317 		btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2318 		goto cleanup_transaction;
2319 	}
2320 
2321 	btrfs_wait_delalloc_flush(fs_info);
2322 
2323 	/*
2324 	 * Wait for all ordered extents started by a fast fsync that joined this
2325 	 * transaction. Otherwise if this transaction commits before the ordered
2326 	 * extents complete we lose logged data after a power failure.
2327 	 */
2328 	btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered);
2329 	wait_event(cur_trans->pending_wait,
2330 		   atomic_read(&cur_trans->pending_ordered) == 0);
2331 
2332 	btrfs_scrub_pause(fs_info);
2333 	/*
2334 	 * Ok now we need to make sure to block out any other joins while we
2335 	 * commit the transaction.  We could have started a join before setting
2336 	 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2337 	 */
2338 	spin_lock(&fs_info->trans_lock);
2339 	add_pending_snapshot(trans);
2340 	cur_trans->state = TRANS_STATE_COMMIT_DOING;
2341 	spin_unlock(&fs_info->trans_lock);
2342 
2343 	/*
2344 	 * The thread has started/joined the transaction thus it holds the
2345 	 * lockdep map as a reader. It has to release it before acquiring the
2346 	 * lockdep map as a writer.
2347 	 */
2348 	btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2349 	btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers);
2350 	wait_event(cur_trans->writer_wait,
2351 		   atomic_read(&cur_trans->num_writers) == 1);
2352 
2353 	/*
2354 	 * Make lockdep happy by acquiring the state locks after
2355 	 * btrfs_trans_num_writers is released. If we acquired the state locks
2356 	 * before releasing the btrfs_trans_num_writers lock then lockdep would
2357 	 * complain because we did not follow the reverse order unlocking rule.
2358 	 */
2359 	btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2360 	btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2361 	btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2362 
2363 	/*
2364 	 * We've started the commit, clear the flag in case we were triggered to
2365 	 * do an async commit but somebody else started before the transaction
2366 	 * kthread could do the work.
2367 	 */
2368 	clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags);
2369 
2370 	if (TRANS_ABORTED(cur_trans)) {
2371 		ret = cur_trans->aborted;
2372 		btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2373 		goto scrub_continue;
2374 	}
2375 	/*
2376 	 * the reloc mutex makes sure that we stop
2377 	 * the balancing code from coming in and moving
2378 	 * extents around in the middle of the commit
2379 	 */
2380 	mutex_lock(&fs_info->reloc_mutex);
2381 
2382 	/*
2383 	 * We needn't worry about the delayed items because we will
2384 	 * deal with them in create_pending_snapshot(), which is the
2385 	 * core function of the snapshot creation.
2386 	 */
2387 	ret = create_pending_snapshots(trans);
2388 	if (ret)
2389 		goto unlock_reloc;
2390 
2391 	/*
2392 	 * We insert the dir indexes of the snapshots and update the inode
2393 	 * of the snapshots' parents after the snapshot creation, so there
2394 	 * are some delayed items which are not dealt with. Now deal with
2395 	 * them.
2396 	 *
2397 	 * We needn't worry that this operation will corrupt the snapshots,
2398 	 * because all the tree which are snapshoted will be forced to COW
2399 	 * the nodes and leaves.
2400 	 */
2401 	ret = btrfs_run_delayed_items(trans);
2402 	if (ret)
2403 		goto unlock_reloc;
2404 
2405 	ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
2406 	if (ret)
2407 		goto unlock_reloc;
2408 
2409 	/*
2410 	 * make sure none of the code above managed to slip in a
2411 	 * delayed item
2412 	 */
2413 	btrfs_assert_delayed_root_empty(fs_info);
2414 
2415 	WARN_ON(cur_trans != trans->transaction);
2416 
2417 	ret = commit_fs_roots(trans);
2418 	if (ret)
2419 		goto unlock_reloc;
2420 
2421 	/* commit_fs_roots gets rid of all the tree log roots, it is now
2422 	 * safe to free the root of tree log roots
2423 	 */
2424 	btrfs_free_log_root_tree(trans, fs_info);
2425 
2426 	/*
2427 	 * Since fs roots are all committed, we can get a quite accurate
2428 	 * new_roots. So let's do quota accounting.
2429 	 */
2430 	ret = btrfs_qgroup_account_extents(trans);
2431 	if (ret < 0)
2432 		goto unlock_reloc;
2433 
2434 	ret = commit_cowonly_roots(trans);
2435 	if (ret)
2436 		goto unlock_reloc;
2437 
2438 	/*
2439 	 * The tasks which save the space cache and inode cache may also
2440 	 * update ->aborted, check it.
2441 	 */
2442 	if (TRANS_ABORTED(cur_trans)) {
2443 		ret = cur_trans->aborted;
2444 		goto unlock_reloc;
2445 	}
2446 
2447 	cur_trans = fs_info->running_transaction;
2448 
2449 	btrfs_set_root_node(&fs_info->tree_root->root_item,
2450 			    fs_info->tree_root->node);
2451 	list_add_tail(&fs_info->tree_root->dirty_list,
2452 		      &cur_trans->switch_commits);
2453 
2454 	btrfs_set_root_node(&fs_info->chunk_root->root_item,
2455 			    fs_info->chunk_root->node);
2456 	list_add_tail(&fs_info->chunk_root->dirty_list,
2457 		      &cur_trans->switch_commits);
2458 
2459 	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2460 		btrfs_set_root_node(&fs_info->block_group_root->root_item,
2461 				    fs_info->block_group_root->node);
2462 		list_add_tail(&fs_info->block_group_root->dirty_list,
2463 			      &cur_trans->switch_commits);
2464 	}
2465 
2466 	switch_commit_roots(trans);
2467 
2468 	ASSERT(list_empty(&cur_trans->dirty_bgs));
2469 	ASSERT(list_empty(&cur_trans->io_bgs));
2470 	update_super_roots(fs_info);
2471 
2472 	btrfs_set_super_log_root(fs_info->super_copy, 0);
2473 	btrfs_set_super_log_root_level(fs_info->super_copy, 0);
2474 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
2475 	       sizeof(*fs_info->super_copy));
2476 
2477 	btrfs_commit_device_sizes(cur_trans);
2478 
2479 	clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
2480 	clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
2481 
2482 	btrfs_trans_release_chunk_metadata(trans);
2483 
2484 	/*
2485 	 * Before changing the transaction state to TRANS_STATE_UNBLOCKED and
2486 	 * setting fs_info->running_transaction to NULL, lock tree_log_mutex to
2487 	 * make sure that before we commit our superblock, no other task can
2488 	 * start a new transaction and commit a log tree before we commit our
2489 	 * superblock. Anyone trying to commit a log tree locks this mutex before
2490 	 * writing its superblock.
2491 	 */
2492 	mutex_lock(&fs_info->tree_log_mutex);
2493 
2494 	spin_lock(&fs_info->trans_lock);
2495 	cur_trans->state = TRANS_STATE_UNBLOCKED;
2496 	fs_info->running_transaction = NULL;
2497 	spin_unlock(&fs_info->trans_lock);
2498 	mutex_unlock(&fs_info->reloc_mutex);
2499 
2500 	wake_up(&fs_info->transaction_wait);
2501 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2502 
2503 	/* If we have features changed, wake up the cleaner to update sysfs. */
2504 	if (test_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags) &&
2505 	    fs_info->cleaner_kthread)
2506 		wake_up_process(fs_info->cleaner_kthread);
2507 
2508 	ret = btrfs_write_and_wait_transaction(trans);
2509 	if (ret) {
2510 		btrfs_handle_fs_error(fs_info, ret,
2511 				      "Error while writing out transaction");
2512 		mutex_unlock(&fs_info->tree_log_mutex);
2513 		goto scrub_continue;
2514 	}
2515 
2516 	ret = write_all_supers(fs_info, 0);
2517 	/*
2518 	 * the super is written, we can safely allow the tree-loggers
2519 	 * to go about their business
2520 	 */
2521 	mutex_unlock(&fs_info->tree_log_mutex);
2522 	if (ret)
2523 		goto scrub_continue;
2524 
2525 	/*
2526 	 * We needn't acquire the lock here because there is no other task
2527 	 * which can change it.
2528 	 */
2529 	cur_trans->state = TRANS_STATE_SUPER_COMMITTED;
2530 	wake_up(&cur_trans->commit_wait);
2531 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2532 
2533 	btrfs_finish_extent_commit(trans);
2534 
2535 	if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
2536 		btrfs_clear_space_info_full(fs_info);
2537 
2538 	fs_info->last_trans_committed = cur_trans->transid;
2539 	/*
2540 	 * We needn't acquire the lock here because there is no other task
2541 	 * which can change it.
2542 	 */
2543 	cur_trans->state = TRANS_STATE_COMPLETED;
2544 	wake_up(&cur_trans->commit_wait);
2545 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2546 
2547 	spin_lock(&fs_info->trans_lock);
2548 	list_del_init(&cur_trans->list);
2549 	spin_unlock(&fs_info->trans_lock);
2550 
2551 	btrfs_put_transaction(cur_trans);
2552 	btrfs_put_transaction(cur_trans);
2553 
2554 	if (trans->type & __TRANS_FREEZABLE)
2555 		sb_end_intwrite(fs_info->sb);
2556 
2557 	trace_btrfs_transaction_commit(fs_info);
2558 
2559 	interval = ktime_get_ns() - start_time;
2560 
2561 	btrfs_scrub_continue(fs_info);
2562 
2563 	if (current->journal_info == trans)
2564 		current->journal_info = NULL;
2565 
2566 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
2567 
2568 	update_commit_stats(fs_info, interval);
2569 
2570 	return ret;
2571 
2572 unlock_reloc:
2573 	mutex_unlock(&fs_info->reloc_mutex);
2574 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2575 scrub_continue:
2576 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2577 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2578 	btrfs_scrub_continue(fs_info);
2579 cleanup_transaction:
2580 	btrfs_trans_release_metadata(trans);
2581 	btrfs_cleanup_pending_block_groups(trans);
2582 	btrfs_trans_release_chunk_metadata(trans);
2583 	trans->block_rsv = NULL;
2584 	btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
2585 	if (current->journal_info == trans)
2586 		current->journal_info = NULL;
2587 	cleanup_transaction(trans, ret);
2588 
2589 	return ret;
2590 
2591 lockdep_release:
2592 	btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
2593 	btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2594 	goto cleanup_transaction;
2595 
2596 lockdep_trans_commit_start_release:
2597 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2598 	btrfs_end_transaction(trans);
2599 	return ret;
2600 }
2601 
2602 /*
2603  * return < 0 if error
2604  * 0 if there are no more dead_roots at the time of call
2605  * 1 there are more to be processed, call me again
2606  *
2607  * The return value indicates there are certainly more snapshots to delete, but
2608  * if there comes a new one during processing, it may return 0. We don't mind,
2609  * because btrfs_commit_super will poke cleaner thread and it will process it a
2610  * few seconds later.
2611  */
btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info * fs_info)2612 int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info)
2613 {
2614 	struct btrfs_root *root;
2615 	int ret;
2616 
2617 	spin_lock(&fs_info->trans_lock);
2618 	if (list_empty(&fs_info->dead_roots)) {
2619 		spin_unlock(&fs_info->trans_lock);
2620 		return 0;
2621 	}
2622 	root = list_first_entry(&fs_info->dead_roots,
2623 			struct btrfs_root, root_list);
2624 	list_del_init(&root->root_list);
2625 	spin_unlock(&fs_info->trans_lock);
2626 
2627 	btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid);
2628 
2629 	btrfs_kill_all_delayed_nodes(root);
2630 
2631 	if (btrfs_header_backref_rev(root->node) <
2632 			BTRFS_MIXED_BACKREF_REV)
2633 		ret = btrfs_drop_snapshot(root, 0, 0);
2634 	else
2635 		ret = btrfs_drop_snapshot(root, 1, 0);
2636 
2637 	btrfs_put_root(root);
2638 	return (ret < 0) ? 0 : 1;
2639 }
2640 
2641 /*
2642  * We only mark the transaction aborted and then set the file system read-only.
2643  * This will prevent new transactions from starting or trying to join this
2644  * one.
2645  *
2646  * This means that error recovery at the call site is limited to freeing
2647  * any local memory allocations and passing the error code up without
2648  * further cleanup. The transaction should complete as it normally would
2649  * in the call path but will return -EIO.
2650  *
2651  * We'll complete the cleanup in btrfs_end_transaction and
2652  * btrfs_commit_transaction.
2653  */
__btrfs_abort_transaction(struct btrfs_trans_handle * trans,const char * function,unsigned int line,int errno,bool first_hit)2654 void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
2655 				      const char *function,
2656 				      unsigned int line, int errno, bool first_hit)
2657 {
2658 	struct btrfs_fs_info *fs_info = trans->fs_info;
2659 
2660 	WRITE_ONCE(trans->aborted, errno);
2661 	WRITE_ONCE(trans->transaction->aborted, errno);
2662 	if (first_hit && errno == -ENOSPC)
2663 		btrfs_dump_space_info_for_trans_abort(fs_info);
2664 	/* Wake up anybody who may be waiting on this transaction */
2665 	wake_up(&fs_info->transaction_wait);
2666 	wake_up(&fs_info->transaction_blocked_wait);
2667 	__btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
2668 }
2669 
btrfs_transaction_init(void)2670 int __init btrfs_transaction_init(void)
2671 {
2672 	btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
2673 			sizeof(struct btrfs_trans_handle), 0,
2674 			SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
2675 	if (!btrfs_trans_handle_cachep)
2676 		return -ENOMEM;
2677 	return 0;
2678 }
2679 
btrfs_transaction_exit(void)2680 void __cold btrfs_transaction_exit(void)
2681 {
2682 	kmem_cache_destroy(btrfs_trans_handle_cachep);
2683 }
2684