Line data Source code
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 : #define BTRFS_ROOT_TRANS_TAG 0
41 :
42 : /*
43 : * Transaction states and transitions
44 : *
45 : * No running transaction (fs tree blocks are not modified)
46 : * |
47 : * | To next stage:
48 : * | Call start_transaction() variants. Except btrfs_join_transaction_nostart().
49 : * V
50 : * Transaction N [[TRANS_STATE_RUNNING]]
51 : * |
52 : * | New trans handles can be attached to transaction N by calling all
53 : * | start_transaction() variants.
54 : * |
55 : * | To next stage:
56 : * | Call btrfs_commit_transaction() on any trans handle attached to
57 : * | transaction N
58 : * V
59 : * Transaction N [[TRANS_STATE_COMMIT_START]]
60 : * |
61 : * | Will wait for previous running transaction to completely finish if there
62 : * | is one
63 : * |
64 : * | Then one of the following happes:
65 : * | - Wait for all other trans handle holders to release.
66 : * | The btrfs_commit_transaction() caller will do the commit work.
67 : * | - Wait for current transaction to be committed by others.
68 : * | Other btrfs_commit_transaction() caller will do the commit work.
69 : * |
70 : * | At this stage, only btrfs_join_transaction*() variants can attach
71 : * | to this running transaction.
72 : * | All other variants will wait for current one to finish and attach to
73 : * | transaction N+1.
74 : * |
75 : * | To next stage:
76 : * | Caller is chosen to commit transaction N, and all other trans handle
77 : * | haven been released.
78 : * V
79 : * Transaction N [[TRANS_STATE_COMMIT_DOING]]
80 : * |
81 : * | The heavy lifting transaction work is started.
82 : * | From running delayed refs (modifying extent tree) to creating pending
83 : * | snapshots, running qgroups.
84 : * | In short, modify supporting trees to reflect modifications of subvolume
85 : * | trees.
86 : * |
87 : * | At this stage, all start_transaction() calls will wait for this
88 : * | transaction to finish and attach to transaction N+1.
89 : * |
90 : * | To next stage:
91 : * | Until all supporting trees are updated.
92 : * V
93 : * Transaction N [[TRANS_STATE_UNBLOCKED]]
94 : * | Transaction N+1
95 : * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]]
96 : * | need to write them back to disk and update |
97 : * | super blocks. |
98 : * | |
99 : * | At this stage, new transaction is allowed to |
100 : * | start. |
101 : * | All new start_transaction() calls will be |
102 : * | attached to transid N+1. |
103 : * | |
104 : * | To next stage: |
105 : * | Until all tree blocks are super blocks are |
106 : * | written to block devices |
107 : * V |
108 : * Transaction N [[TRANS_STATE_COMPLETED]] V
109 : * All tree blocks and super blocks are written. Transaction N+1
110 : * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]]
111 : * data structures will be cleaned up. | Life goes on
112 : */
113 : static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
114 : [TRANS_STATE_RUNNING] = 0U,
115 : [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH),
116 : [TRANS_STATE_COMMIT_DOING] = (__TRANS_START |
117 : __TRANS_ATTACH |
118 : __TRANS_JOIN |
119 : __TRANS_JOIN_NOSTART),
120 : [TRANS_STATE_UNBLOCKED] = (__TRANS_START |
121 : __TRANS_ATTACH |
122 : __TRANS_JOIN |
123 : __TRANS_JOIN_NOLOCK |
124 : __TRANS_JOIN_NOSTART),
125 : [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START |
126 : __TRANS_ATTACH |
127 : __TRANS_JOIN |
128 : __TRANS_JOIN_NOLOCK |
129 : __TRANS_JOIN_NOSTART),
130 : [TRANS_STATE_COMPLETED] = (__TRANS_START |
131 : __TRANS_ATTACH |
132 : __TRANS_JOIN |
133 : __TRANS_JOIN_NOLOCK |
134 : __TRANS_JOIN_NOSTART),
135 : };
136 :
137 91287263 : void btrfs_put_transaction(struct btrfs_transaction *transaction)
138 : {
139 91287263 : WARN_ON(refcount_read(&transaction->use_count) == 0);
140 91287263 : if (refcount_dec_and_test(&transaction->use_count)) {
141 203273 : BUG_ON(!list_empty(&transaction->list));
142 203273 : WARN_ON(!RB_EMPTY_ROOT(
143 : &transaction->delayed_refs.href_root.rb_root));
144 203273 : WARN_ON(!RB_EMPTY_ROOT(
145 : &transaction->delayed_refs.dirty_extent_root));
146 203273 : if (transaction->delayed_refs.pending_csums)
147 0 : btrfs_err(transaction->fs_info,
148 : "pending csums is %llu",
149 : transaction->delayed_refs.pending_csums);
150 : /*
151 : * If any block groups are found in ->deleted_bgs then it's
152 : * because the transaction was aborted and a commit did not
153 : * happen (things failed before writing the new superblock
154 : * and calling btrfs_finish_extent_commit()), so we can not
155 : * discard the physical locations of the block groups.
156 : */
157 203273 : while (!list_empty(&transaction->deleted_bgs)) {
158 0 : struct btrfs_block_group *cache;
159 :
160 0 : cache = list_first_entry(&transaction->deleted_bgs,
161 : struct btrfs_block_group,
162 : bg_list);
163 0 : list_del_init(&cache->bg_list);
164 0 : btrfs_unfreeze_block_group(cache);
165 0 : btrfs_put_block_group(cache);
166 : }
167 203273 : WARN_ON(!list_empty(&transaction->dev_update_list));
168 203273 : kfree(transaction);
169 : }
170 91177731 : }
171 :
172 203434 : static noinline void switch_commit_roots(struct btrfs_trans_handle *trans)
173 : {
174 203434 : struct btrfs_transaction *cur_trans = trans->transaction;
175 203434 : struct btrfs_fs_info *fs_info = trans->fs_info;
176 203434 : struct btrfs_root *root, *tmp;
177 :
178 : /*
179 : * At this point no one can be using this transaction to modify any tree
180 : * and no one can start another transaction to modify any tree either.
181 : */
182 203434 : ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING);
183 :
184 203434 : down_write(&fs_info->commit_root_sem);
185 :
186 406868 : if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
187 3430 : fs_info->last_reloc_trans = trans->transid;
188 :
189 1201608 : list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits,
190 : dirty_list) {
191 998174 : list_del_init(&root->dirty_list);
192 998174 : free_extent_buffer(root->commit_root);
193 998174 : root->commit_root = btrfs_root_node(root);
194 998174 : extent_io_tree_release(&root->dirty_log_pages);
195 998174 : btrfs_qgroup_clean_swapped_blocks(root);
196 : }
197 :
198 : /* We can free old roots now. */
199 203434 : spin_lock(&cur_trans->dropped_roots_lock);
200 203552 : while (!list_empty(&cur_trans->dropped_roots)) {
201 118 : root = list_first_entry(&cur_trans->dropped_roots,
202 : struct btrfs_root, root_list);
203 118 : list_del_init(&root->root_list);
204 118 : spin_unlock(&cur_trans->dropped_roots_lock);
205 118 : btrfs_free_log(trans, root);
206 118 : btrfs_drop_and_free_fs_root(fs_info, root);
207 118 : spin_lock(&cur_trans->dropped_roots_lock);
208 : }
209 203434 : spin_unlock(&cur_trans->dropped_roots_lock);
210 :
211 203434 : up_write(&fs_info->commit_root_sem);
212 203434 : }
213 :
214 54001239 : static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
215 : unsigned int type)
216 : {
217 54001239 : if (type & TRANS_EXTWRITERS)
218 28646515 : atomic_inc(&trans->num_extwriters);
219 54001239 : }
220 :
221 54193426 : static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
222 : unsigned int type)
223 : {
224 54193426 : if (type & TRANS_EXTWRITERS)
225 28710024 : atomic_dec(&trans->num_extwriters);
226 54193456 : }
227 :
228 : static inline void extwriter_counter_init(struct btrfs_transaction *trans,
229 : unsigned int type)
230 : {
231 203272 : atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
232 : }
233 :
234 : static inline int extwriter_counter_read(struct btrfs_transaction *trans)
235 : {
236 218688 : return atomic_read(&trans->num_extwriters);
237 : }
238 :
239 : /*
240 : * To be called after doing the chunk btree updates right after allocating a new
241 : * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a
242 : * chunk after all chunk btree updates and after finishing the second phase of
243 : * chunk allocation (btrfs_create_pending_block_groups()) in case some block
244 : * group had its chunk item insertion delayed to the second phase.
245 : */
246 111496314 : void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
247 : {
248 111496314 : struct btrfs_fs_info *fs_info = trans->fs_info;
249 :
250 111496314 : if (!trans->chunk_bytes_reserved)
251 : return;
252 :
253 2052 : btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv,
254 : trans->chunk_bytes_reserved, NULL);
255 2052 : trans->chunk_bytes_reserved = 0;
256 : }
257 :
258 : /*
259 : * either allocate a new transaction or hop into the existing one
260 : */
261 54198647 : static noinline int join_transaction(struct btrfs_fs_info *fs_info,
262 : unsigned int type)
263 : {
264 54198647 : struct btrfs_transaction *cur_trans;
265 :
266 54198647 : spin_lock(&fs_info->trans_lock);
267 54229375 : loop:
268 : /* The file system has been taken offline. No new transactions. */
269 54229375 : if (BTRFS_FS_ERROR(fs_info)) {
270 0 : spin_unlock(&fs_info->trans_lock);
271 0 : return -EROFS;
272 : }
273 :
274 54229375 : cur_trans = fs_info->running_transaction;
275 54229375 : if (cur_trans) {
276 54012671 : if (TRANS_ABORTED(cur_trans)) {
277 0 : spin_unlock(&fs_info->trans_lock);
278 0 : return cur_trans->aborted;
279 : }
280 54012671 : if (btrfs_blocked_trans_types[cur_trans->state] & type) {
281 11440 : spin_unlock(&fs_info->trans_lock);
282 11440 : return -EBUSY;
283 : }
284 54001231 : refcount_inc(&cur_trans->use_count);
285 54001238 : atomic_inc(&cur_trans->num_writers);
286 54001239 : extwriter_counter_inc(cur_trans, type);
287 54001240 : spin_unlock(&fs_info->trans_lock);
288 54001240 : btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers);
289 54001240 : btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters);
290 54001240 : return 0;
291 : }
292 216704 : spin_unlock(&fs_info->trans_lock);
293 :
294 : /*
295 : * If we are ATTACH, we just want to catch the current transaction,
296 : * and commit it. If there is no transaction, just return ENOENT.
297 : */
298 216672 : if (type == TRANS_ATTACH)
299 : return -ENOENT;
300 :
301 : /*
302 : * JOIN_NOLOCK only happens during the transaction commit, so
303 : * it is impossible that ->running_transaction is NULL
304 : */
305 204481 : BUG_ON(type == TRANS_JOIN_NOLOCK);
306 :
307 204481 : cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS);
308 204463 : if (!cur_trans)
309 : return -ENOMEM;
310 :
311 204463 : btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers);
312 204463 : btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters);
313 :
314 204463 : spin_lock(&fs_info->trans_lock);
315 204507 : if (fs_info->running_transaction) {
316 : /*
317 : * someone started a transaction after we unlocked. Make sure
318 : * to redo the checks above
319 : */
320 1235 : btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
321 1235 : btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
322 1235 : kfree(cur_trans);
323 1235 : goto loop;
324 203272 : } else if (BTRFS_FS_ERROR(fs_info)) {
325 0 : spin_unlock(&fs_info->trans_lock);
326 0 : btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
327 0 : btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
328 0 : kfree(cur_trans);
329 0 : return -EROFS;
330 : }
331 :
332 203272 : cur_trans->fs_info = fs_info;
333 203272 : atomic_set(&cur_trans->pending_ordered, 0);
334 203272 : init_waitqueue_head(&cur_trans->pending_wait);
335 203272 : atomic_set(&cur_trans->num_writers, 1);
336 203272 : extwriter_counter_init(cur_trans, type);
337 203272 : init_waitqueue_head(&cur_trans->writer_wait);
338 203272 : init_waitqueue_head(&cur_trans->commit_wait);
339 203272 : cur_trans->state = TRANS_STATE_RUNNING;
340 : /*
341 : * One for this trans handle, one so it will live on until we
342 : * commit the transaction.
343 : */
344 203272 : refcount_set(&cur_trans->use_count, 2);
345 203272 : cur_trans->flags = 0;
346 203272 : cur_trans->start_time = ktime_get_seconds();
347 :
348 203272 : memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));
349 :
350 203272 : cur_trans->delayed_refs.href_root = RB_ROOT_CACHED;
351 203272 : cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
352 203272 : atomic_set(&cur_trans->delayed_refs.num_entries, 0);
353 :
354 : /*
355 : * although the tree mod log is per file system and not per transaction,
356 : * the log must never go across transaction boundaries.
357 : */
358 203272 : smp_mb();
359 203272 : if (!list_empty(&fs_info->tree_mod_seq_list))
360 0 : WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");
361 203272 : if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
362 0 : WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");
363 203272 : atomic64_set(&fs_info->tree_mod_seq, 0);
364 :
365 203272 : spin_lock_init(&cur_trans->delayed_refs.lock);
366 :
367 203272 : INIT_LIST_HEAD(&cur_trans->pending_snapshots);
368 203272 : INIT_LIST_HEAD(&cur_trans->dev_update_list);
369 203272 : INIT_LIST_HEAD(&cur_trans->switch_commits);
370 203272 : INIT_LIST_HEAD(&cur_trans->dirty_bgs);
371 203272 : INIT_LIST_HEAD(&cur_trans->io_bgs);
372 203272 : INIT_LIST_HEAD(&cur_trans->dropped_roots);
373 203272 : mutex_init(&cur_trans->cache_write_mutex);
374 203272 : spin_lock_init(&cur_trans->dirty_bgs_lock);
375 203272 : INIT_LIST_HEAD(&cur_trans->deleted_bgs);
376 203272 : spin_lock_init(&cur_trans->dropped_roots_lock);
377 203272 : list_add_tail(&cur_trans->list, &fs_info->trans_list);
378 203272 : extent_io_tree_init(fs_info, &cur_trans->dirty_pages,
379 : IO_TREE_TRANS_DIRTY_PAGES);
380 203272 : extent_io_tree_init(fs_info, &cur_trans->pinned_extents,
381 : IO_TREE_FS_PINNED_EXTENTS);
382 203272 : fs_info->generation++;
383 203272 : cur_trans->transid = fs_info->generation;
384 203272 : fs_info->running_transaction = cur_trans;
385 203272 : cur_trans->aborted = 0;
386 203272 : spin_unlock(&fs_info->trans_lock);
387 :
388 203272 : return 0;
389 : }
390 :
391 : /*
392 : * This does all the record keeping required to make sure that a shareable root
393 : * is properly recorded in a given transaction. This is required to make sure
394 : * the old root from before we joined the transaction is deleted when the
395 : * transaction commits.
396 : */
397 154078 : static int record_root_in_trans(struct btrfs_trans_handle *trans,
398 : struct btrfs_root *root,
399 : int force)
400 : {
401 154078 : struct btrfs_fs_info *fs_info = root->fs_info;
402 154078 : int ret = 0;
403 :
404 154078 : if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
405 154078 : root->last_trans < trans->transid) || force) {
406 301216 : WARN_ON(!force && root->commit_root != root->node);
407 :
408 : /*
409 : * see below for IN_TRANS_SETUP usage rules
410 : * we have the reloc mutex held now, so there
411 : * is only one writer in this function
412 : */
413 150608 : set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
414 :
415 : /* make sure readers find IN_TRANS_SETUP before
416 : * they find our root->last_trans update
417 : */
418 150608 : smp_wmb();
419 :
420 150608 : spin_lock(&fs_info->fs_roots_radix_lock);
421 150608 : if (root->last_trans == trans->transid && !force) {
422 0 : spin_unlock(&fs_info->fs_roots_radix_lock);
423 0 : return 0;
424 : }
425 150608 : radix_tree_tag_set(&fs_info->fs_roots_radix,
426 150608 : (unsigned long)root->root_key.objectid,
427 : BTRFS_ROOT_TRANS_TAG);
428 150608 : spin_unlock(&fs_info->fs_roots_radix_lock);
429 150608 : root->last_trans = trans->transid;
430 :
431 : /* this is pretty tricky. We don't want to
432 : * take the relocation lock in btrfs_record_root_in_trans
433 : * unless we're really doing the first setup for this root in
434 : * this transaction.
435 : *
436 : * Normally we'd use root->last_trans as a flag to decide
437 : * if we want to take the expensive mutex.
438 : *
439 : * But, we have to set root->last_trans before we
440 : * init the relocation root, otherwise, we trip over warnings
441 : * in ctree.c. The solution used here is to flag ourselves
442 : * with root IN_TRANS_SETUP. When this is 1, we're still
443 : * fixing up the reloc trees and everyone must wait.
444 : *
445 : * When this is zero, they can trust root->last_trans and fly
446 : * through btrfs_record_root_in_trans without having to take the
447 : * lock. smp_wmb() makes sure that all the writes above are
448 : * done before we pop in the zero below
449 : */
450 150608 : ret = btrfs_init_reloc_root(trans, root);
451 150608 : smp_mb__before_atomic();
452 150608 : clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
453 : }
454 : return ret;
455 : }
456 :
457 :
458 118 : void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
459 : struct btrfs_root *root)
460 : {
461 118 : struct btrfs_fs_info *fs_info = root->fs_info;
462 118 : struct btrfs_transaction *cur_trans = trans->transaction;
463 :
464 : /* Add ourselves to the transaction dropped list */
465 118 : spin_lock(&cur_trans->dropped_roots_lock);
466 118 : list_add_tail(&root->root_list, &cur_trans->dropped_roots);
467 118 : spin_unlock(&cur_trans->dropped_roots_lock);
468 :
469 : /* Make sure we don't try to update the root at commit time */
470 118 : spin_lock(&fs_info->fs_roots_radix_lock);
471 118 : radix_tree_tag_clear(&fs_info->fs_roots_radix,
472 118 : (unsigned long)root->root_key.objectid,
473 : BTRFS_ROOT_TRANS_TAG);
474 118 : spin_unlock(&fs_info->fs_roots_radix_lock);
475 118 : }
476 :
477 54222371 : int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
478 : struct btrfs_root *root)
479 : {
480 54222371 : struct btrfs_fs_info *fs_info = root->fs_info;
481 54222371 : int ret;
482 :
483 54222371 : if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
484 : return 0;
485 :
486 : /*
487 : * see record_root_in_trans for comments about IN_TRANS_SETUP usage
488 : * and barriers
489 : */
490 46919871 : smp_rmb();
491 46919699 : if (root->last_trans == trans->transid &&
492 0 : !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
493 : return 0;
494 :
495 151789 : mutex_lock(&fs_info->reloc_mutex);
496 151674 : ret = record_root_in_trans(trans, root, 0);
497 151674 : mutex_unlock(&fs_info->reloc_mutex);
498 :
499 151674 : return ret;
500 : }
501 :
502 : static inline int is_transaction_blocked(struct btrfs_transaction *trans)
503 : {
504 29850779 : return (trans->state >= TRANS_STATE_COMMIT_START &&
505 29850475 : trans->state < TRANS_STATE_UNBLOCKED &&
506 88738 : !TRANS_ABORTED(trans));
507 : }
508 :
509 : /* wait for commit against the current transaction to become unblocked
510 : * when this is done, it is safe to start a new transaction, but the current
511 : * transaction might not be fully on disk.
512 : */
513 29890578 : static void wait_current_trans(struct btrfs_fs_info *fs_info)
514 : {
515 29890578 : struct btrfs_transaction *cur_trans;
516 :
517 29890578 : spin_lock(&fs_info->trans_lock);
518 29892461 : cur_trans = fs_info->running_transaction;
519 29892461 : if (cur_trans && is_transaction_blocked(cur_trans)) {
520 88737 : refcount_inc(&cur_trans->use_count);
521 88737 : spin_unlock(&fs_info->trans_lock);
522 :
523 88737 : btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
524 177306 : wait_event(fs_info->transaction_wait,
525 : cur_trans->state >= TRANS_STATE_UNBLOCKED ||
526 : TRANS_ABORTED(cur_trans));
527 88733 : btrfs_put_transaction(cur_trans);
528 : } else {
529 29803724 : spin_unlock(&fs_info->trans_lock);
530 : }
531 29892182 : }
532 :
533 54310700 : static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type)
534 : {
535 108621400 : if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
536 : return 0;
537 :
538 54310420 : if (type == TRANS_START)
539 28691435 : return 1;
540 :
541 : return 0;
542 : }
543 :
544 26444247 : static inline bool need_reserve_reloc_root(struct btrfs_root *root)
545 : {
546 26444247 : struct btrfs_fs_info *fs_info = root->fs_info;
547 :
548 26483344 : if (!fs_info->reloc_ctl ||
549 39097 : !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
550 16435 : root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
551 16435 : root->reloc_root)
552 26442641 : return false;
553 :
554 : return true;
555 : }
556 :
557 : static struct btrfs_trans_handle *
558 54184804 : start_transaction(struct btrfs_root *root, unsigned int num_items,
559 : unsigned int type, enum btrfs_reserve_flush_enum flush,
560 : bool enforce_qgroups)
561 : {
562 54184804 : struct btrfs_fs_info *fs_info = root->fs_info;
563 54184804 : struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
564 54184804 : struct btrfs_trans_handle *h;
565 54184804 : struct btrfs_transaction *cur_trans;
566 54184804 : u64 num_bytes = 0;
567 54184804 : u64 qgroup_reserved = 0;
568 54184804 : bool reloc_reserved = false;
569 54184804 : bool do_chunk_alloc = false;
570 54184804 : int ret;
571 :
572 54184804 : if (BTRFS_FS_ERROR(fs_info))
573 : return ERR_PTR(-EROFS);
574 :
575 54184804 : if (current->journal_info) {
576 0 : WARN_ON(type & TRANS_EXTWRITERS);
577 0 : h = current->journal_info;
578 0 : refcount_inc(&h->use_count);
579 0 : WARN_ON(refcount_read(&h->use_count) > 2);
580 0 : h->orig_rsv = h->block_rsv;
581 0 : h->block_rsv = NULL;
582 0 : goto got_it;
583 : }
584 :
585 : /*
586 : * Do the reservation before we join the transaction so we can do all
587 : * the appropriate flushing if need be.
588 : */
589 54184804 : if (num_items && root != fs_info->chunk_root) {
590 26441621 : struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv;
591 26441621 : u64 delayed_refs_bytes = 0;
592 :
593 26441621 : qgroup_reserved = num_items * fs_info->nodesize;
594 26441621 : ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved,
595 : enforce_qgroups);
596 26445185 : if (ret)
597 6063 : return ERR_PTR(ret);
598 :
599 : /*
600 : * We want to reserve all the bytes we may need all at once, so
601 : * we only do 1 enospc flushing cycle per transaction start. We
602 : * accomplish this by simply assuming we'll do num_items worth
603 : * of delayed refs updates in this trans handle, and refill that
604 : * amount for whatever is missing in the reserve.
605 : */
606 26439122 : num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items);
607 26439122 : if (flush == BTRFS_RESERVE_FLUSH_ALL &&
608 : !btrfs_block_rsv_full(delayed_refs_rsv)) {
609 1405237 : delayed_refs_bytes = btrfs_calc_delayed_ref_bytes(fs_info,
610 : num_items);
611 1405237 : num_bytes += delayed_refs_bytes;
612 : }
613 :
614 : /*
615 : * Do the reservation for the relocation root creation
616 : */
617 26439122 : if (need_reserve_reloc_root(root)) {
618 1606 : num_bytes += fs_info->nodesize;
619 1606 : reloc_reserved = true;
620 : }
621 :
622 26427465 : ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes, flush);
623 26480531 : if (ret)
624 23885 : goto reserve_fail;
625 26456646 : if (delayed_refs_bytes) {
626 1384432 : btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv,
627 : delayed_refs_bytes);
628 1384437 : num_bytes -= delayed_refs_bytes;
629 : }
630 :
631 26456651 : if (rsv->space_info->force_alloc)
632 0 : do_chunk_alloc = true;
633 27743183 : } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL &&
634 : !btrfs_block_rsv_full(delayed_refs_rsv)) {
635 : /*
636 : * Some people call with btrfs_start_transaction(root, 0)
637 : * because they can be throttled, but have some other mechanism
638 : * for reserving space. We still want these guys to refill the
639 : * delayed block_rsv so just add 1 items worth of reservation
640 : * here.
641 : */
642 1998379 : ret = btrfs_delayed_refs_rsv_refill(fs_info, flush);
643 1998381 : if (ret)
644 0 : goto reserve_fail;
645 : }
646 54199836 : again:
647 54199845 : h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
648 54194840 : if (!h) {
649 0 : ret = -ENOMEM;
650 0 : goto alloc_fail;
651 : }
652 :
653 : /*
654 : * If we are JOIN_NOLOCK we're already committing a transaction and
655 : * waiting on this guy, so we don't need to do the sb_start_intwrite
656 : * because we're already holding a ref. We need this because we could
657 : * have raced in and did an fsync() on a file which can kick a commit
658 : * and then we deadlock with somebody doing a freeze.
659 : *
660 : * If we are ATTACH, it means we just want to catch the current
661 : * transaction and commit it, so we needn't do sb_start_intwrite().
662 : */
663 54194840 : if (type & __TRANS_FREEZABLE)
664 52720547 : sb_start_intwrite(fs_info->sb);
665 :
666 54192473 : if (may_wait_transaction(fs_info, type))
667 28691181 : wait_current_trans(fs_info);
668 :
669 54202314 : do {
670 54202314 : ret = join_transaction(fs_info, type);
671 54228087 : if (ret == -EBUSY) {
672 11440 : wait_current_trans(fs_info);
673 11434 : if (unlikely(type == TRANS_ATTACH ||
674 : type == TRANS_JOIN_NOSTART))
675 : ret = -ENOENT;
676 : }
677 54226867 : } while (ret == -EBUSY);
678 :
679 54217859 : if (ret < 0)
680 13403 : goto join_fail;
681 :
682 54204456 : cur_trans = fs_info->running_transaction;
683 :
684 54204456 : h->transid = cur_trans->transid;
685 54204456 : h->transaction = cur_trans;
686 54204456 : refcount_set(&h->use_count, 1);
687 54204456 : h->fs_info = root->fs_info;
688 :
689 54204456 : h->type = type;
690 54204456 : INIT_LIST_HEAD(&h->new_bgs);
691 :
692 54204456 : smp_mb();
693 54323298 : if (cur_trans->state >= TRANS_STATE_COMMIT_START &&
694 119032 : may_wait_transaction(fs_info, type)) {
695 9 : current->journal_info = h;
696 9 : btrfs_commit_transaction(h);
697 9 : goto again;
698 : }
699 :
700 54204257 : if (num_bytes) {
701 26456809 : trace_btrfs_space_reservation(fs_info, "transaction",
702 : h->transid, num_bytes, 1);
703 26456769 : h->block_rsv = &fs_info->trans_block_rsv;
704 26456769 : h->bytes_reserved = num_bytes;
705 26456769 : h->reloc_reserved = reloc_reserved;
706 : }
707 :
708 27747448 : got_it:
709 54204217 : if (!current->journal_info)
710 54204107 : current->journal_info = h;
711 :
712 : /*
713 : * If the space_info is marked ALLOC_FORCE then we'll get upgraded to
714 : * ALLOC_FORCE the first run through, and then we won't allocate for
715 : * anybody else who races in later. We don't care about the return
716 : * value here.
717 : */
718 54204217 : if (do_chunk_alloc && num_bytes) {
719 0 : u64 flags = h->block_rsv->space_info->flags;
720 :
721 0 : btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags),
722 : CHUNK_ALLOC_NO_FORCE);
723 : }
724 :
725 : /*
726 : * btrfs_record_root_in_trans() needs to alloc new extents, and may
727 : * call btrfs_join_transaction() while we're also starting a
728 : * transaction.
729 : *
730 : * Thus it need to be called after current->journal_info initialized,
731 : * or we can deadlock.
732 : */
733 54204217 : ret = btrfs_record_root_in_trans(h, root);
734 54202703 : if (ret) {
735 : /*
736 : * The transaction handle is fully initialized and linked with
737 : * other structures so it needs to be ended in case of errors,
738 : * not just freed.
739 : */
740 0 : btrfs_end_transaction(h);
741 0 : return ERR_PTR(ret);
742 : }
743 :
744 : return h;
745 :
746 : join_fail:
747 13403 : if (type & __TRANS_FREEZABLE)
748 0 : sb_end_intwrite(fs_info->sb);
749 13403 : kmem_cache_free(btrfs_trans_handle_cachep, h);
750 13402 : alloc_fail:
751 13402 : if (num_bytes)
752 0 : btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv,
753 : num_bytes, NULL);
754 13402 : reserve_fail:
755 37287 : btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved);
756 37286 : return ERR_PTR(ret);
757 : }
758 :
759 27304485 : struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
760 : unsigned int num_items)
761 : {
762 27304485 : return start_transaction(root, num_items, TRANS_START,
763 : BTRFS_RESERVE_FLUSH_ALL, true);
764 : }
765 :
766 1385718 : struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
767 : struct btrfs_root *root,
768 : unsigned int num_items)
769 : {
770 1385718 : return start_transaction(root, num_items, TRANS_START,
771 : BTRFS_RESERVE_FLUSH_ALL_STEAL, false);
772 : }
773 :
774 24032240 : struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
775 : {
776 24032240 : return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH,
777 : true);
778 : }
779 :
780 41 : struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root)
781 : {
782 41 : return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
783 : BTRFS_RESERVE_NO_FLUSH, true);
784 : }
785 :
786 : /*
787 : * Similar to regular join but it never starts a transaction when none is
788 : * running or after waiting for the current one to finish.
789 : */
790 1443864 : struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root)
791 : {
792 1443864 : return start_transaction(root, 0, TRANS_JOIN_NOSTART,
793 : BTRFS_RESERVE_NO_FLUSH, true);
794 : }
795 :
796 : /*
797 : * btrfs_attach_transaction() - catch the running transaction
798 : *
799 : * It is used when we want to commit the current the transaction, but
800 : * don't want to start a new one.
801 : *
802 : * Note: If this function return -ENOENT, it just means there is no
803 : * running transaction. But it is possible that the inactive transaction
804 : * is still in the memory, not fully on disk. If you hope there is no
805 : * inactive transaction in the fs when -ENOENT is returned, you should
806 : * invoke
807 : * btrfs_attach_transaction_barrier()
808 : */
809 1176 : struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
810 : {
811 1176 : return start_transaction(root, 0, TRANS_ATTACH,
812 : BTRFS_RESERVE_NO_FLUSH, true);
813 : }
814 :
815 : /*
816 : * btrfs_attach_transaction_barrier() - catch the running transaction
817 : *
818 : * It is similar to the above function, the difference is this one
819 : * will wait for all the inactive transactions until they fully
820 : * complete.
821 : */
822 : struct btrfs_trans_handle *
823 29564 : btrfs_attach_transaction_barrier(struct btrfs_root *root)
824 : {
825 29564 : struct btrfs_trans_handle *trans;
826 :
827 29564 : trans = start_transaction(root, 0, TRANS_ATTACH,
828 : BTRFS_RESERVE_NO_FLUSH, true);
829 29561 : if (trans == ERR_PTR(-ENOENT))
830 13238 : btrfs_wait_for_commit(root->fs_info, 0);
831 :
832 29563 : return trans;
833 : }
834 :
835 : /* Wait for a transaction commit to reach at least the given state. */
836 7391 : static noinline void wait_for_commit(struct btrfs_transaction *commit,
837 : const enum btrfs_trans_state min_state)
838 : {
839 7391 : struct btrfs_fs_info *fs_info = commit->fs_info;
840 7391 : u64 transid = commit->transid;
841 7391 : bool put = false;
842 :
843 : /*
844 : * At the moment this function is called with min_state either being
845 : * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED.
846 : */
847 7391 : if (min_state == TRANS_STATE_COMPLETED)
848 : btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
849 : else
850 36438349 : btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
851 :
852 36438349 : while (1) {
853 36450668 : wait_event(commit->commit_wait, commit->state >= min_state);
854 36436010 : if (put)
855 36430231 : btrfs_put_transaction(commit);
856 :
857 36324390 : if (min_state < TRANS_STATE_COMPLETED)
858 : break;
859 :
860 : /*
861 : * A transaction isn't really completed until all of the
862 : * previous transactions are completed, but with fsync we can
863 : * end up with SUPER_COMMITTED transactions before a COMPLETED
864 : * transaction. Wait for those.
865 : */
866 :
867 36324254 : spin_lock(&fs_info->trans_lock);
868 36438430 : commit = list_first_entry_or_null(&fs_info->trans_list,
869 : struct btrfs_transaction,
870 : list);
871 36438049 : if (!commit || commit->transid > transid) {
872 7258 : spin_unlock(&fs_info->trans_lock);
873 : break;
874 : }
875 36431172 : refcount_inc(&commit->use_count);
876 36431172 : put = true;
877 36431172 : spin_unlock(&fs_info->trans_lock);
878 : }
879 7392 : }
880 :
881 13240 : int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid)
882 : {
883 13240 : struct btrfs_transaction *cur_trans = NULL, *t;
884 13240 : int ret = 0;
885 :
886 13240 : if (transid) {
887 5 : if (transid <= fs_info->last_trans_committed)
888 0 : goto out;
889 :
890 : /* find specified transaction */
891 5 : spin_lock(&fs_info->trans_lock);
892 5 : list_for_each_entry(t, &fs_info->trans_list, list) {
893 5 : if (t->transid == transid) {
894 5 : cur_trans = t;
895 5 : refcount_inc(&cur_trans->use_count);
896 5 : ret = 0;
897 5 : break;
898 : }
899 0 : if (t->transid > transid) {
900 : ret = 0;
901 : break;
902 : }
903 : }
904 5 : spin_unlock(&fs_info->trans_lock);
905 :
906 : /*
907 : * The specified transaction doesn't exist, or we
908 : * raced with btrfs_commit_transaction
909 : */
910 5 : if (!cur_trans) {
911 0 : if (transid > fs_info->last_trans_committed)
912 0 : ret = -EINVAL;
913 0 : goto out;
914 : }
915 : } else {
916 : /* find newest transaction that is committing | committed */
917 13235 : spin_lock(&fs_info->trans_lock);
918 13906 : list_for_each_entry_reverse(t, &fs_info->trans_list,
919 : list) {
920 2131 : if (t->state >= TRANS_STATE_COMMIT_START) {
921 1468 : if (t->state == TRANS_STATE_COMPLETED)
922 : break;
923 1468 : cur_trans = t;
924 1468 : refcount_inc(&cur_trans->use_count);
925 : break;
926 : }
927 : }
928 13243 : spin_unlock(&fs_info->trans_lock);
929 13242 : if (!cur_trans)
930 11774 : goto out; /* nothing committing|committed */
931 : }
932 :
933 1473 : wait_for_commit(cur_trans, TRANS_STATE_COMPLETED);
934 1470 : btrfs_put_transaction(cur_trans);
935 13247 : out:
936 13247 : return ret;
937 : }
938 :
939 1188068 : void btrfs_throttle(struct btrfs_fs_info *fs_info)
940 : {
941 1188068 : wait_current_trans(fs_info);
942 1188068 : }
943 :
944 1853452 : bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
945 : {
946 1853452 : struct btrfs_transaction *cur_trans = trans->transaction;
947 :
948 3706875 : if (cur_trans->state >= TRANS_STATE_COMMIT_START ||
949 1853423 : test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags))
950 : return true;
951 :
952 1853077 : if (btrfs_check_space_for_delayed_refs(trans->fs_info))
953 : return true;
954 :
955 2675 : return !!btrfs_block_rsv_check(&trans->fs_info->global_block_rsv, 50);
956 : }
957 :
958 54197107 : static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans)
959 :
960 : {
961 54197107 : struct btrfs_fs_info *fs_info = trans->fs_info;
962 :
963 54197107 : if (!trans->block_rsv) {
964 : ASSERT(!trans->bytes_reserved);
965 : return;
966 : }
967 :
968 36452261 : if (!trans->bytes_reserved)
969 : return;
970 :
971 30022182 : ASSERT(trans->block_rsv == &fs_info->trans_block_rsv);
972 30022182 : trace_btrfs_space_reservation(fs_info, "transaction",
973 : trans->transid, trans->bytes_reserved, 0);
974 30021940 : btrfs_block_rsv_release(fs_info, trans->block_rsv,
975 : trans->bytes_reserved, NULL);
976 30025461 : trans->bytes_reserved = 0;
977 : }
978 :
979 53990084 : static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
980 : int throttle)
981 : {
982 53990084 : struct btrfs_fs_info *info = trans->fs_info;
983 53990084 : struct btrfs_transaction *cur_trans = trans->transaction;
984 53990084 : int err = 0;
985 :
986 53990084 : if (refcount_read(&trans->use_count) > 1) {
987 0 : refcount_dec(&trans->use_count);
988 0 : trans->block_rsv = trans->orig_rsv;
989 0 : return 0;
990 : }
991 :
992 53990084 : btrfs_trans_release_metadata(trans);
993 53990209 : trans->block_rsv = NULL;
994 :
995 53990209 : btrfs_create_pending_block_groups(trans);
996 :
997 53986004 : btrfs_trans_release_chunk_metadata(trans);
998 :
999 53985440 : if (trans->type & __TRANS_FREEZABLE)
1000 52539876 : sb_end_intwrite(info->sb);
1001 :
1002 53986887 : WARN_ON(cur_trans != info->running_transaction);
1003 53986887 : WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
1004 53986887 : atomic_dec(&cur_trans->num_writers);
1005 53990578 : extwriter_counter_dec(cur_trans, trans->type);
1006 :
1007 53990279 : cond_wake_up(&cur_trans->writer_wait);
1008 :
1009 53995298 : btrfs_lockdep_release(info, btrfs_trans_num_extwriters);
1010 53995298 : btrfs_lockdep_release(info, btrfs_trans_num_writers);
1011 :
1012 53995298 : btrfs_put_transaction(cur_trans);
1013 :
1014 53988012 : if (current->journal_info == trans)
1015 53987822 : current->journal_info = NULL;
1016 :
1017 53988012 : if (throttle)
1018 3824864 : btrfs_run_delayed_iputs(info);
1019 :
1020 53988012 : if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) {
1021 1 : wake_up_process(info->transaction_kthread);
1022 1 : if (TRANS_ABORTED(trans))
1023 1 : err = trans->aborted;
1024 : else
1025 : err = -EROFS;
1026 : }
1027 :
1028 53988012 : kmem_cache_free(btrfs_trans_handle_cachep, trans);
1029 53988012 : return err;
1030 : }
1031 :
1032 50166240 : int btrfs_end_transaction(struct btrfs_trans_handle *trans)
1033 : {
1034 50166240 : return __btrfs_end_transaction(trans, 0);
1035 : }
1036 :
1037 3824864 : int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans)
1038 : {
1039 3824864 : return __btrfs_end_transaction(trans, 1);
1040 : }
1041 :
1042 : /*
1043 : * when btree blocks are allocated, they have some corresponding bits set for
1044 : * them in one of two extent_io trees. This is used to make sure all of
1045 : * those extents are sent to disk but does not wait on them
1046 : */
1047 691932 : int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
1048 : struct extent_io_tree *dirty_pages, int mark)
1049 : {
1050 691932 : int err = 0;
1051 691932 : int werr = 0;
1052 691932 : struct address_space *mapping = fs_info->btree_inode->i_mapping;
1053 691932 : struct extent_state *cached_state = NULL;
1054 691932 : u64 start = 0;
1055 691932 : u64 end;
1056 :
1057 2907655 : while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1058 : mark, &cached_state)) {
1059 2215702 : bool wait_writeback = false;
1060 :
1061 2215702 : err = convert_extent_bit(dirty_pages, start, end,
1062 : EXTENT_NEED_WAIT,
1063 : mark, &cached_state);
1064 : /*
1065 : * convert_extent_bit can return -ENOMEM, which is most of the
1066 : * time a temporary error. So when it happens, ignore the error
1067 : * and wait for writeback of this range to finish - because we
1068 : * failed to set the bit EXTENT_NEED_WAIT for the range, a call
1069 : * to __btrfs_wait_marked_extents() would not know that
1070 : * writeback for this range started and therefore wouldn't
1071 : * wait for it to finish - we don't want to commit a
1072 : * superblock that points to btree nodes/leafs for which
1073 : * writeback hasn't finished yet (and without errors).
1074 : * We cleanup any entries left in the io tree when committing
1075 : * the transaction (through extent_io_tree_release()).
1076 : */
1077 2215702 : if (err == -ENOMEM) {
1078 : err = 0;
1079 : wait_writeback = true;
1080 : }
1081 2215702 : if (!err)
1082 2215702 : err = filemap_fdatawrite_range(mapping, start, end);
1083 2215723 : if (err)
1084 : werr = err;
1085 2215723 : else if (wait_writeback)
1086 0 : werr = filemap_fdatawait_range(mapping, start, end);
1087 2215723 : free_extent_state(cached_state);
1088 2215723 : cached_state = NULL;
1089 2215723 : cond_resched();
1090 2215723 : start = end + 1;
1091 : }
1092 691969 : return werr;
1093 : }
1094 :
1095 : /*
1096 : * when btree blocks are allocated, they have some corresponding bits set for
1097 : * them in one of two extent_io trees. This is used to make sure all of
1098 : * those extents are on disk for transaction or log commit. We wait
1099 : * on all the pages and clear them from the dirty pages state tree
1100 : */
1101 691969 : static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info,
1102 : struct extent_io_tree *dirty_pages)
1103 : {
1104 691969 : int err = 0;
1105 691969 : int werr = 0;
1106 691969 : struct address_space *mapping = fs_info->btree_inode->i_mapping;
1107 691969 : struct extent_state *cached_state = NULL;
1108 691969 : u64 start = 0;
1109 691969 : u64 end;
1110 :
1111 2907691 : while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1112 : EXTENT_NEED_WAIT, &cached_state)) {
1113 : /*
1114 : * Ignore -ENOMEM errors returned by clear_extent_bit().
1115 : * When committing the transaction, we'll remove any entries
1116 : * left in the io tree. For a log commit, we don't remove them
1117 : * after committing the log because the tree can be accessed
1118 : * concurrently - we do it only at transaction commit time when
1119 : * it's safe to do it (through extent_io_tree_release()).
1120 : */
1121 2215722 : err = clear_extent_bit(dirty_pages, start, end,
1122 : EXTENT_NEED_WAIT, &cached_state);
1123 2215722 : if (err == -ENOMEM)
1124 : err = 0;
1125 2215722 : if (!err)
1126 2215722 : err = filemap_fdatawait_range(mapping, start, end);
1127 2215720 : if (err)
1128 6 : werr = err;
1129 2215720 : free_extent_state(cached_state);
1130 2215722 : cached_state = NULL;
1131 2215722 : cond_resched();
1132 2215722 : start = end + 1;
1133 : }
1134 691969 : if (err)
1135 2 : werr = err;
1136 691969 : return werr;
1137 : }
1138 :
1139 203434 : static int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
1140 : struct extent_io_tree *dirty_pages)
1141 : {
1142 203434 : bool errors = false;
1143 203434 : int err;
1144 :
1145 203434 : err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1146 203434 : if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags))
1147 4 : errors = true;
1148 :
1149 203434 : if (errors && !err)
1150 0 : err = -EIO;
1151 203434 : return err;
1152 : }
1153 :
1154 488535 : int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark)
1155 : {
1156 488535 : struct btrfs_fs_info *fs_info = log_root->fs_info;
1157 488535 : struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages;
1158 488535 : bool errors = false;
1159 488535 : int err;
1160 :
1161 488535 : ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
1162 :
1163 488535 : err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1164 855811 : if ((mark & EXTENT_DIRTY) &&
1165 367276 : test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags))
1166 1 : errors = true;
1167 :
1168 853329 : if ((mark & EXTENT_NEW) &&
1169 364794 : test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags))
1170 0 : errors = true;
1171 :
1172 488535 : if (errors && !err)
1173 0 : err = -EIO;
1174 488535 : return err;
1175 : }
1176 :
1177 : /*
1178 : * When btree blocks are allocated the corresponding extents are marked dirty.
1179 : * This function ensures such extents are persisted on disk for transaction or
1180 : * log commit.
1181 : *
1182 : * @trans: transaction whose dirty pages we'd like to write
1183 : */
1184 203434 : static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans)
1185 : {
1186 203434 : int ret;
1187 203434 : int ret2;
1188 203434 : struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages;
1189 203434 : struct btrfs_fs_info *fs_info = trans->fs_info;
1190 203434 : struct blk_plug plug;
1191 :
1192 203434 : blk_start_plug(&plug);
1193 203434 : ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY);
1194 203434 : blk_finish_plug(&plug);
1195 203434 : ret2 = btrfs_wait_extents(fs_info, dirty_pages);
1196 :
1197 203434 : extent_io_tree_release(&trans->transaction->dirty_pages);
1198 :
1199 203434 : if (ret)
1200 : return ret;
1201 203434 : else if (ret2)
1202 : return ret2;
1203 : else
1204 203430 : return 0;
1205 : }
1206 :
1207 : /*
1208 : * this is used to update the root pointer in the tree of tree roots.
1209 : *
1210 : * But, in the case of the extent allocation tree, updating the root
1211 : * pointer may allocate blocks which may change the root of the extent
1212 : * allocation tree.
1213 : *
1214 : * So, this loops and repeats and makes sure the cowonly root didn't
1215 : * change while the root pointer was being updated in the metadata.
1216 : */
1217 446371 : static int update_cowonly_root(struct btrfs_trans_handle *trans,
1218 : struct btrfs_root *root)
1219 : {
1220 446371 : int ret;
1221 446371 : u64 old_root_bytenr;
1222 446371 : u64 old_root_used;
1223 446371 : struct btrfs_fs_info *fs_info = root->fs_info;
1224 446371 : struct btrfs_root *tree_root = fs_info->tree_root;
1225 :
1226 446371 : old_root_used = btrfs_root_used(&root->root_item);
1227 :
1228 1339113 : while (1) {
1229 892742 : old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1230 892742 : if (old_root_bytenr == root->node->start &&
1231 : old_root_used == btrfs_root_used(&root->root_item))
1232 : break;
1233 :
1234 446371 : btrfs_set_root_node(&root->root_item, root->node);
1235 446371 : ret = btrfs_update_root(trans, tree_root,
1236 : &root->root_key,
1237 : &root->root_item);
1238 446371 : if (ret)
1239 0 : return ret;
1240 :
1241 446371 : old_root_used = btrfs_root_used(&root->root_item);
1242 : }
1243 :
1244 : return 0;
1245 : }
1246 :
1247 : /*
1248 : * update all the cowonly tree roots on disk
1249 : *
1250 : * The error handling in this function may not be obvious. Any of the
1251 : * failures will cause the file system to go offline. We still need
1252 : * to clean up the delayed refs.
1253 : */
1254 203435 : static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans)
1255 : {
1256 203435 : struct btrfs_fs_info *fs_info = trans->fs_info;
1257 203435 : struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1258 203435 : struct list_head *io_bgs = &trans->transaction->io_bgs;
1259 203435 : struct list_head *next;
1260 203435 : struct extent_buffer *eb;
1261 203435 : int ret;
1262 :
1263 : /*
1264 : * At this point no one can be using this transaction to modify any tree
1265 : * and no one can start another transaction to modify any tree either.
1266 : */
1267 203435 : ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);
1268 :
1269 203435 : eb = btrfs_lock_root_node(fs_info->tree_root);
1270 203435 : ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1271 : 0, &eb, BTRFS_NESTING_COW);
1272 203435 : btrfs_tree_unlock(eb);
1273 203435 : free_extent_buffer(eb);
1274 :
1275 203435 : if (ret)
1276 : return ret;
1277 :
1278 203435 : ret = btrfs_run_dev_stats(trans);
1279 203435 : if (ret)
1280 : return ret;
1281 203435 : ret = btrfs_run_dev_replace(trans);
1282 203435 : if (ret)
1283 : return ret;
1284 203435 : ret = btrfs_run_qgroups(trans);
1285 203435 : if (ret)
1286 : return ret;
1287 :
1288 203435 : ret = btrfs_setup_space_cache(trans);
1289 203435 : if (ret)
1290 : return ret;
1291 :
1292 203435 : again:
1293 711656 : while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1294 446371 : struct btrfs_root *root;
1295 446371 : next = fs_info->dirty_cowonly_roots.next;
1296 446371 : list_del_init(next);
1297 446371 : root = list_entry(next, struct btrfs_root, dirty_list);
1298 446371 : clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1299 :
1300 446371 : list_add_tail(&root->dirty_list,
1301 446371 : &trans->transaction->switch_commits);
1302 446371 : ret = update_cowonly_root(trans, root);
1303 446371 : if (ret)
1304 0 : return ret;
1305 : }
1306 :
1307 : /* Now flush any delayed refs generated by updating all of the roots */
1308 265285 : ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1309 265285 : if (ret)
1310 0 : return ret;
1311 :
1312 468858 : while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1313 203573 : ret = btrfs_write_dirty_block_groups(trans);
1314 203573 : if (ret)
1315 0 : return ret;
1316 :
1317 : /*
1318 : * We're writing the dirty block groups, which could generate
1319 : * delayed refs, which could generate more dirty block groups,
1320 : * so we want to keep this flushing in this loop to make sure
1321 : * everything gets run.
1322 : */
1323 203573 : ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1324 203573 : if (ret)
1325 0 : return ret;
1326 : }
1327 :
1328 265285 : if (!list_empty(&fs_info->dirty_cowonly_roots))
1329 61850 : goto again;
1330 :
1331 : /* Update dev-replace pointer once everything is committed */
1332 203435 : fs_info->dev_replace.committed_cursor_left =
1333 203435 : fs_info->dev_replace.cursor_left_last_write_of_item;
1334 :
1335 203435 : return 0;
1336 : }
1337 :
1338 : /*
1339 : * If we had a pending drop we need to see if there are any others left in our
1340 : * dead roots list, and if not clear our bit and wake any waiters.
1341 : */
1342 0 : void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info)
1343 : {
1344 : /*
1345 : * We put the drop in progress roots at the front of the list, so if the
1346 : * first entry doesn't have UNFINISHED_DROP set we can wake everybody
1347 : * up.
1348 : */
1349 0 : spin_lock(&fs_info->trans_lock);
1350 0 : if (!list_empty(&fs_info->dead_roots)) {
1351 0 : struct btrfs_root *root = list_first_entry(&fs_info->dead_roots,
1352 : struct btrfs_root,
1353 : root_list);
1354 0 : if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) {
1355 0 : spin_unlock(&fs_info->trans_lock);
1356 0 : return;
1357 : }
1358 : }
1359 0 : spin_unlock(&fs_info->trans_lock);
1360 :
1361 0 : btrfs_wake_unfinished_drop(fs_info);
1362 : }
1363 :
1364 : /*
1365 : * dead roots are old snapshots that need to be deleted. This allocates
1366 : * a dirty root struct and adds it into the list of dead roots that need to
1367 : * be deleted
1368 : */
1369 339 : void btrfs_add_dead_root(struct btrfs_root *root)
1370 : {
1371 339 : struct btrfs_fs_info *fs_info = root->fs_info;
1372 :
1373 339 : spin_lock(&fs_info->trans_lock);
1374 339 : if (list_empty(&root->root_list)) {
1375 339 : btrfs_grab_root(root);
1376 :
1377 : /* We want to process the partially complete drops first. */
1378 678 : if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state))
1379 0 : list_add(&root->root_list, &fs_info->dead_roots);
1380 : else
1381 339 : list_add_tail(&root->root_list, &fs_info->dead_roots);
1382 : }
1383 339 : spin_unlock(&fs_info->trans_lock);
1384 339 : }
1385 :
1386 : /*
1387 : * Update each subvolume root and its relocation root, if it exists, in the tree
1388 : * of tree roots. Also free log roots if they exist.
1389 : */
1390 203435 : static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
1391 : {
1392 203435 : struct btrfs_fs_info *fs_info = trans->fs_info;
1393 203435 : struct btrfs_root *gang[8];
1394 203435 : int i;
1395 203435 : int ret;
1396 :
1397 : /*
1398 : * At this point no one can be using this transaction to modify any tree
1399 : * and no one can start another transaction to modify any tree either.
1400 : */
1401 203435 : ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);
1402 :
1403 203435 : spin_lock(&fs_info->fs_roots_radix_lock);
1404 347691 : while (1) {
1405 347691 : ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1406 : (void **)gang, 0,
1407 : ARRAY_SIZE(gang),
1408 : BTRFS_ROOT_TRANS_TAG);
1409 347691 : if (ret == 0)
1410 : break;
1411 294676 : for (i = 0; i < ret; i++) {
1412 150420 : struct btrfs_root *root = gang[i];
1413 150420 : int ret2;
1414 :
1415 : /*
1416 : * At this point we can neither have tasks logging inodes
1417 : * from a root nor trying to commit a log tree.
1418 : */
1419 150420 : ASSERT(atomic_read(&root->log_writers) == 0);
1420 150420 : ASSERT(atomic_read(&root->log_commit[0]) == 0);
1421 150420 : ASSERT(atomic_read(&root->log_commit[1]) == 0);
1422 :
1423 150420 : radix_tree_tag_clear(&fs_info->fs_roots_radix,
1424 150420 : (unsigned long)root->root_key.objectid,
1425 : BTRFS_ROOT_TRANS_TAG);
1426 150420 : spin_unlock(&fs_info->fs_roots_radix_lock);
1427 :
1428 150420 : btrfs_free_log(trans, root);
1429 150420 : ret2 = btrfs_update_reloc_root(trans, root);
1430 150420 : if (ret2)
1431 0 : return ret2;
1432 :
1433 : /* see comments in should_cow_block() */
1434 150420 : clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1435 150420 : smp_mb__after_atomic();
1436 :
1437 150420 : if (root->commit_root != root->node) {
1438 145416 : list_add_tail(&root->dirty_list,
1439 145416 : &trans->transaction->switch_commits);
1440 145416 : btrfs_set_root_node(&root->root_item,
1441 : root->node);
1442 : }
1443 :
1444 150420 : ret2 = btrfs_update_root(trans, fs_info->tree_root,
1445 : &root->root_key,
1446 : &root->root_item);
1447 150420 : if (ret2)
1448 0 : return ret2;
1449 150420 : spin_lock(&fs_info->fs_roots_radix_lock);
1450 150420 : btrfs_qgroup_free_meta_all_pertrans(root);
1451 : }
1452 : }
1453 203435 : spin_unlock(&fs_info->fs_roots_radix_lock);
1454 203435 : return 0;
1455 : }
1456 :
1457 : /*
1458 : * defrag a given btree.
1459 : * Every leaf in the btree is read and defragged.
1460 : */
1461 2 : int btrfs_defrag_root(struct btrfs_root *root)
1462 : {
1463 2 : struct btrfs_fs_info *info = root->fs_info;
1464 2 : struct btrfs_trans_handle *trans;
1465 2 : int ret;
1466 :
1467 2 : if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1468 : return 0;
1469 :
1470 2 : while (1) {
1471 2 : trans = btrfs_start_transaction(root, 0);
1472 2 : if (IS_ERR(trans)) {
1473 0 : ret = PTR_ERR(trans);
1474 0 : break;
1475 : }
1476 :
1477 2 : ret = btrfs_defrag_leaves(trans, root);
1478 :
1479 2 : btrfs_end_transaction(trans);
1480 2 : btrfs_btree_balance_dirty(info);
1481 2 : cond_resched();
1482 :
1483 2 : if (btrfs_fs_closing(info) || ret != -EAGAIN)
1484 : break;
1485 :
1486 0 : if (btrfs_defrag_cancelled(info)) {
1487 : btrfs_debug(info, "defrag_root cancelled");
1488 : ret = -EAGAIN;
1489 : break;
1490 : }
1491 : }
1492 2 : clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1493 2 : return ret;
1494 : }
1495 :
1496 : /*
1497 : * Do all special snapshot related qgroup dirty hack.
1498 : *
1499 : * Will do all needed qgroup inherit and dirty hack like switch commit
1500 : * roots inside one transaction and write all btree into disk, to make
1501 : * qgroup works.
1502 : */
1503 1028 : static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1504 : struct btrfs_root *src,
1505 : struct btrfs_root *parent,
1506 : struct btrfs_qgroup_inherit *inherit,
1507 : u64 dst_objectid)
1508 : {
1509 1028 : struct btrfs_fs_info *fs_info = src->fs_info;
1510 1028 : int ret;
1511 :
1512 : /*
1513 : * Save some performance in the case that qgroups are not
1514 : * enabled. If this check races with the ioctl, rescan will
1515 : * kick in anyway.
1516 : */
1517 1028 : if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1518 : return 0;
1519 :
1520 : /*
1521 : * Ensure dirty @src will be committed. Or, after coming
1522 : * commit_fs_roots() and switch_commit_roots(), any dirty but not
1523 : * recorded root will never be updated again, causing an outdated root
1524 : * item.
1525 : */
1526 174 : ret = record_root_in_trans(trans, src, 1);
1527 174 : if (ret)
1528 : return ret;
1529 :
1530 : /*
1531 : * btrfs_qgroup_inherit relies on a consistent view of the usage for the
1532 : * src root, so we must run the delayed refs here.
1533 : *
1534 : * However this isn't particularly fool proof, because there's no
1535 : * synchronization keeping us from changing the tree after this point
1536 : * before we do the qgroup_inherit, or even from making changes while
1537 : * we're doing the qgroup_inherit. But that's a problem for the future,
1538 : * for now flush the delayed refs to narrow the race window where the
1539 : * qgroup counters could end up wrong.
1540 : */
1541 174 : ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1542 174 : if (ret) {
1543 0 : btrfs_abort_transaction(trans, ret);
1544 0 : return ret;
1545 : }
1546 :
1547 174 : ret = commit_fs_roots(trans);
1548 174 : if (ret)
1549 0 : goto out;
1550 174 : ret = btrfs_qgroup_account_extents(trans);
1551 174 : if (ret < 0)
1552 0 : goto out;
1553 :
1554 : /* Now qgroup are all updated, we can inherit it to new qgroups */
1555 174 : ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid,
1556 : inherit);
1557 174 : if (ret < 0)
1558 0 : goto out;
1559 :
1560 : /*
1561 : * Now we do a simplified commit transaction, which will:
1562 : * 1) commit all subvolume and extent tree
1563 : * To ensure all subvolume and extent tree have a valid
1564 : * commit_root to accounting later insert_dir_item()
1565 : * 2) write all btree blocks onto disk
1566 : * This is to make sure later btree modification will be cowed
1567 : * Or commit_root can be populated and cause wrong qgroup numbers
1568 : * In this simplified commit, we don't really care about other trees
1569 : * like chunk and root tree, as they won't affect qgroup.
1570 : * And we don't write super to avoid half committed status.
1571 : */
1572 174 : ret = commit_cowonly_roots(trans);
1573 174 : if (ret)
1574 0 : goto out;
1575 174 : switch_commit_roots(trans);
1576 174 : ret = btrfs_write_and_wait_transaction(trans);
1577 174 : if (ret)
1578 0 : btrfs_handle_fs_error(fs_info, ret,
1579 : "Error while writing out transaction for qgroup");
1580 :
1581 174 : out:
1582 : /*
1583 : * Force parent root to be updated, as we recorded it before so its
1584 : * last_trans == cur_transid.
1585 : * Or it won't be committed again onto disk after later
1586 : * insert_dir_item()
1587 : */
1588 174 : if (!ret)
1589 174 : ret = record_root_in_trans(trans, parent, 1);
1590 : return ret;
1591 : }
1592 :
1593 : /*
1594 : * new snapshots need to be created at a very specific time in the
1595 : * transaction commit. This does the actual creation.
1596 : *
1597 : * Note:
1598 : * If the error which may affect the commitment of the current transaction
1599 : * happens, we should return the error number. If the error which just affect
1600 : * the creation of the pending snapshots, just return 0.
1601 : */
1602 1028 : static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1603 : struct btrfs_pending_snapshot *pending)
1604 : {
1605 :
1606 1028 : struct btrfs_fs_info *fs_info = trans->fs_info;
1607 1028 : struct btrfs_key key;
1608 1028 : struct btrfs_root_item *new_root_item;
1609 1028 : struct btrfs_root *tree_root = fs_info->tree_root;
1610 1028 : struct btrfs_root *root = pending->root;
1611 1028 : struct btrfs_root *parent_root;
1612 1028 : struct btrfs_block_rsv *rsv;
1613 1028 : struct inode *parent_inode = pending->dir;
1614 1028 : struct btrfs_path *path;
1615 1028 : struct btrfs_dir_item *dir_item;
1616 1028 : struct extent_buffer *tmp;
1617 1028 : struct extent_buffer *old;
1618 1028 : struct timespec64 cur_time;
1619 1028 : int ret = 0;
1620 1028 : u64 to_reserve = 0;
1621 1028 : u64 index = 0;
1622 1028 : u64 objectid;
1623 1028 : u64 root_flags;
1624 1028 : unsigned int nofs_flags;
1625 1028 : struct fscrypt_name fname;
1626 :
1627 1028 : ASSERT(pending->path);
1628 1028 : path = pending->path;
1629 :
1630 1028 : ASSERT(pending->root_item);
1631 1028 : new_root_item = pending->root_item;
1632 :
1633 : /*
1634 : * We're inside a transaction and must make sure that any potential
1635 : * allocations with GFP_KERNEL in fscrypt won't recurse back to
1636 : * filesystem.
1637 : */
1638 1028 : nofs_flags = memalloc_nofs_save();
1639 2056 : pending->error = fscrypt_setup_filename(parent_inode,
1640 1028 : &pending->dentry->d_name, 0,
1641 : &fname);
1642 1028 : memalloc_nofs_restore(nofs_flags);
1643 1028 : if (pending->error)
1644 0 : goto free_pending;
1645 :
1646 1028 : pending->error = btrfs_get_free_objectid(tree_root, &objectid);
1647 1028 : if (pending->error)
1648 0 : goto free_fname;
1649 :
1650 : /*
1651 : * Make qgroup to skip current new snapshot's qgroupid, as it is
1652 : * accounted by later btrfs_qgroup_inherit().
1653 : */
1654 1028 : btrfs_set_skip_qgroup(trans, objectid);
1655 :
1656 1028 : btrfs_reloc_pre_snapshot(pending, &to_reserve);
1657 :
1658 1028 : if (to_reserve > 0) {
1659 0 : pending->error = btrfs_block_rsv_add(fs_info,
1660 : &pending->block_rsv,
1661 : to_reserve,
1662 : BTRFS_RESERVE_NO_FLUSH);
1663 0 : if (pending->error)
1664 0 : goto clear_skip_qgroup;
1665 : }
1666 :
1667 1028 : key.objectid = objectid;
1668 1028 : key.offset = (u64)-1;
1669 1028 : key.type = BTRFS_ROOT_ITEM_KEY;
1670 :
1671 1028 : rsv = trans->block_rsv;
1672 1028 : trans->block_rsv = &pending->block_rsv;
1673 1028 : trans->bytes_reserved = trans->block_rsv->reserved;
1674 1028 : trace_btrfs_space_reservation(fs_info, "transaction",
1675 : trans->transid,
1676 : trans->bytes_reserved, 1);
1677 1028 : parent_root = BTRFS_I(parent_inode)->root;
1678 1028 : ret = record_root_in_trans(trans, parent_root, 0);
1679 1028 : if (ret)
1680 0 : goto fail;
1681 1028 : cur_time = current_time(parent_inode);
1682 :
1683 : /*
1684 : * insert the directory item
1685 : */
1686 1028 : ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index);
1687 1028 : if (ret) {
1688 0 : btrfs_abort_transaction(trans, ret);
1689 0 : goto fail;
1690 : }
1691 :
1692 : /* check if there is a file/dir which has the same name. */
1693 1028 : dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1694 : btrfs_ino(BTRFS_I(parent_inode)),
1695 : &fname.disk_name, 0);
1696 1028 : if (dir_item != NULL && !IS_ERR(dir_item)) {
1697 0 : pending->error = -EEXIST;
1698 0 : goto dir_item_existed;
1699 1028 : } else if (IS_ERR(dir_item)) {
1700 0 : ret = PTR_ERR(dir_item);
1701 0 : btrfs_abort_transaction(trans, ret);
1702 0 : goto fail;
1703 : }
1704 1028 : btrfs_release_path(path);
1705 :
1706 : /*
1707 : * pull in the delayed directory update
1708 : * and the delayed inode item
1709 : * otherwise we corrupt the FS during
1710 : * snapshot
1711 : */
1712 1028 : ret = btrfs_run_delayed_items(trans);
1713 1028 : if (ret) { /* Transaction aborted */
1714 0 : btrfs_abort_transaction(trans, ret);
1715 0 : goto fail;
1716 : }
1717 :
1718 1028 : ret = record_root_in_trans(trans, root, 0);
1719 1028 : if (ret) {
1720 0 : btrfs_abort_transaction(trans, ret);
1721 0 : goto fail;
1722 : }
1723 1028 : btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1724 2056 : memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1725 1028 : btrfs_check_and_init_root_item(new_root_item);
1726 :
1727 1028 : root_flags = btrfs_root_flags(new_root_item);
1728 1028 : if (pending->readonly)
1729 740 : root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1730 : else
1731 288 : root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1732 1028 : btrfs_set_root_flags(new_root_item, root_flags);
1733 :
1734 1028 : btrfs_set_root_generation_v2(new_root_item,
1735 : trans->transid);
1736 1028 : generate_random_guid(new_root_item->uuid);
1737 2056 : memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1738 : BTRFS_UUID_SIZE);
1739 1028 : if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1740 288 : memset(new_root_item->received_uuid, 0,
1741 : sizeof(new_root_item->received_uuid));
1742 288 : memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1743 288 : memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1744 288 : btrfs_set_root_stransid(new_root_item, 0);
1745 288 : btrfs_set_root_rtransid(new_root_item, 0);
1746 : }
1747 1028 : btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1748 1028 : btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1749 1028 : btrfs_set_root_otransid(new_root_item, trans->transid);
1750 :
1751 1028 : old = btrfs_lock_root_node(root);
1752 1028 : ret = btrfs_cow_block(trans, root, old, NULL, 0, &old,
1753 : BTRFS_NESTING_COW);
1754 1028 : if (ret) {
1755 0 : btrfs_tree_unlock(old);
1756 0 : free_extent_buffer(old);
1757 0 : btrfs_abort_transaction(trans, ret);
1758 0 : goto fail;
1759 : }
1760 :
1761 1028 : ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1762 : /* clean up in any case */
1763 1028 : btrfs_tree_unlock(old);
1764 1028 : free_extent_buffer(old);
1765 1028 : if (ret) {
1766 0 : btrfs_abort_transaction(trans, ret);
1767 0 : goto fail;
1768 : }
1769 : /* see comments in should_cow_block() */
1770 1028 : set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1771 1028 : smp_wmb();
1772 :
1773 1028 : btrfs_set_root_node(new_root_item, tmp);
1774 : /* record when the snapshot was created in key.offset */
1775 1028 : key.offset = trans->transid;
1776 1028 : ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1777 1028 : btrfs_tree_unlock(tmp);
1778 1028 : free_extent_buffer(tmp);
1779 1028 : if (ret) {
1780 0 : btrfs_abort_transaction(trans, ret);
1781 0 : goto fail;
1782 : }
1783 :
1784 : /*
1785 : * insert root back/forward references
1786 : */
1787 1028 : ret = btrfs_add_root_ref(trans, objectid,
1788 : parent_root->root_key.objectid,
1789 : btrfs_ino(BTRFS_I(parent_inode)), index,
1790 : &fname.disk_name);
1791 1028 : if (ret) {
1792 0 : btrfs_abort_transaction(trans, ret);
1793 0 : goto fail;
1794 : }
1795 :
1796 1028 : key.offset = (u64)-1;
1797 1028 : pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev);
1798 1028 : if (IS_ERR(pending->snap)) {
1799 0 : ret = PTR_ERR(pending->snap);
1800 0 : pending->snap = NULL;
1801 0 : btrfs_abort_transaction(trans, ret);
1802 0 : goto fail;
1803 : }
1804 :
1805 1028 : ret = btrfs_reloc_post_snapshot(trans, pending);
1806 1028 : if (ret) {
1807 0 : btrfs_abort_transaction(trans, ret);
1808 0 : goto fail;
1809 : }
1810 :
1811 : /*
1812 : * Do special qgroup accounting for snapshot, as we do some qgroup
1813 : * snapshot hack to do fast snapshot.
1814 : * To co-operate with that hack, we do hack again.
1815 : * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1816 : */
1817 1028 : ret = qgroup_account_snapshot(trans, root, parent_root,
1818 : pending->inherit, objectid);
1819 1028 : if (ret < 0)
1820 0 : goto fail;
1821 :
1822 1028 : ret = btrfs_insert_dir_item(trans, &fname.disk_name,
1823 : BTRFS_I(parent_inode), &key, BTRFS_FT_DIR,
1824 : index);
1825 : /* We have check then name at the beginning, so it is impossible. */
1826 1028 : BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1827 1028 : if (ret) {
1828 0 : btrfs_abort_transaction(trans, ret);
1829 0 : goto fail;
1830 : }
1831 :
1832 1028 : btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size +
1833 1028 : fname.disk_name.len * 2);
1834 1028 : parent_inode->i_mtime = current_time(parent_inode);
1835 1028 : parent_inode->i_ctime = parent_inode->i_mtime;
1836 1028 : ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode));
1837 1028 : if (ret) {
1838 0 : btrfs_abort_transaction(trans, ret);
1839 0 : goto fail;
1840 : }
1841 1028 : ret = btrfs_uuid_tree_add(trans, new_root_item->uuid,
1842 : BTRFS_UUID_KEY_SUBVOL,
1843 : objectid);
1844 1028 : if (ret) {
1845 0 : btrfs_abort_transaction(trans, ret);
1846 0 : goto fail;
1847 : }
1848 1028 : if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1849 0 : ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid,
1850 : BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1851 : objectid);
1852 0 : if (ret && ret != -EEXIST) {
1853 0 : btrfs_abort_transaction(trans, ret);
1854 0 : goto fail;
1855 : }
1856 : }
1857 :
1858 1028 : fail:
1859 1028 : pending->error = ret;
1860 1028 : dir_item_existed:
1861 1028 : trans->block_rsv = rsv;
1862 1028 : trans->bytes_reserved = 0;
1863 1028 : clear_skip_qgroup:
1864 1028 : btrfs_clear_skip_qgroup(trans);
1865 : free_fname:
1866 : fscrypt_free_filename(&fname);
1867 1028 : free_pending:
1868 1028 : kfree(new_root_item);
1869 1028 : pending->root_item = NULL;
1870 1028 : btrfs_free_path(path);
1871 1028 : pending->path = NULL;
1872 :
1873 1028 : return ret;
1874 : }
1875 :
1876 : /*
1877 : * create all the snapshots we've scheduled for creation
1878 : */
1879 203261 : static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans)
1880 : {
1881 203261 : struct btrfs_pending_snapshot *pending, *next;
1882 203261 : struct list_head *head = &trans->transaction->pending_snapshots;
1883 203261 : int ret = 0;
1884 :
1885 204289 : list_for_each_entry_safe(pending, next, head, list) {
1886 1028 : list_del(&pending->list);
1887 1028 : ret = create_pending_snapshot(trans, pending);
1888 1028 : if (ret)
1889 : break;
1890 : }
1891 203261 : return ret;
1892 : }
1893 :
1894 203260 : static void update_super_roots(struct btrfs_fs_info *fs_info)
1895 : {
1896 203260 : struct btrfs_root_item *root_item;
1897 203260 : struct btrfs_super_block *super;
1898 :
1899 203260 : super = fs_info->super_copy;
1900 :
1901 203260 : root_item = &fs_info->chunk_root->root_item;
1902 203260 : super->chunk_root = root_item->bytenr;
1903 203260 : super->chunk_root_generation = root_item->generation;
1904 203260 : super->chunk_root_level = root_item->level;
1905 :
1906 203260 : root_item = &fs_info->tree_root->root_item;
1907 203260 : super->root = root_item->bytenr;
1908 203260 : super->generation = root_item->generation;
1909 203260 : super->root_level = root_item->level;
1910 203260 : if (btrfs_test_opt(fs_info, SPACE_CACHE))
1911 24 : super->cache_generation = root_item->generation;
1912 406472 : else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags))
1913 2 : super->cache_generation = 0;
1914 406520 : if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags))
1915 203221 : super->uuid_tree_generation = root_item->generation;
1916 203260 : }
1917 :
1918 40276665 : int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1919 : {
1920 40276665 : struct btrfs_transaction *trans;
1921 40276665 : int ret = 0;
1922 :
1923 40276665 : spin_lock(&info->trans_lock);
1924 40301513 : trans = info->running_transaction;
1925 40301513 : if (trans)
1926 39995674 : ret = (trans->state >= TRANS_STATE_COMMIT_START);
1927 40301513 : spin_unlock(&info->trans_lock);
1928 40293455 : return ret;
1929 : }
1930 :
1931 4461 : int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1932 : {
1933 4461 : struct btrfs_transaction *trans;
1934 4461 : int ret = 0;
1935 :
1936 4461 : spin_lock(&info->trans_lock);
1937 4461 : trans = info->running_transaction;
1938 4461 : if (trans)
1939 304 : ret = is_transaction_blocked(trans);
1940 4461 : spin_unlock(&info->trans_lock);
1941 4461 : return ret;
1942 : }
1943 :
1944 4 : void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans)
1945 : {
1946 4 : struct btrfs_fs_info *fs_info = trans->fs_info;
1947 4 : struct btrfs_transaction *cur_trans;
1948 :
1949 : /* Kick the transaction kthread. */
1950 4 : set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags);
1951 4 : wake_up_process(fs_info->transaction_kthread);
1952 :
1953 : /* take transaction reference */
1954 4 : cur_trans = trans->transaction;
1955 4 : refcount_inc(&cur_trans->use_count);
1956 :
1957 4 : btrfs_end_transaction(trans);
1958 :
1959 : /*
1960 : * Wait for the current transaction commit to start and block
1961 : * subsequent transaction joins
1962 : */
1963 4 : btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START);
1964 6 : wait_event(fs_info->transaction_blocked_wait,
1965 : cur_trans->state >= TRANS_STATE_COMMIT_START ||
1966 : TRANS_ABORTED(cur_trans));
1967 4 : btrfs_put_transaction(cur_trans);
1968 4 : }
1969 :
1970 16 : static void cleanup_transaction(struct btrfs_trans_handle *trans, int err)
1971 : {
1972 16 : struct btrfs_fs_info *fs_info = trans->fs_info;
1973 16 : struct btrfs_transaction *cur_trans = trans->transaction;
1974 :
1975 16 : WARN_ON(refcount_read(&trans->use_count) > 1);
1976 :
1977 16 : btrfs_abort_transaction(trans, err);
1978 :
1979 16 : spin_lock(&fs_info->trans_lock);
1980 :
1981 : /*
1982 : * If the transaction is removed from the list, it means this
1983 : * transaction has been committed successfully, so it is impossible
1984 : * to call the cleanup function.
1985 : */
1986 16 : BUG_ON(list_empty(&cur_trans->list));
1987 :
1988 16 : if (cur_trans == fs_info->running_transaction) {
1989 12 : cur_trans->state = TRANS_STATE_COMMIT_DOING;
1990 12 : spin_unlock(&fs_info->trans_lock);
1991 :
1992 : /*
1993 : * The thread has already released the lockdep map as reader
1994 : * already in btrfs_commit_transaction().
1995 : */
1996 12 : btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers);
1997 12 : wait_event(cur_trans->writer_wait,
1998 : atomic_read(&cur_trans->num_writers) == 1);
1999 :
2000 12 : spin_lock(&fs_info->trans_lock);
2001 : }
2002 :
2003 : /*
2004 : * Now that we know no one else is still using the transaction we can
2005 : * remove the transaction from the list of transactions. This avoids
2006 : * the transaction kthread from cleaning up the transaction while some
2007 : * other task is still using it, which could result in a use-after-free
2008 : * on things like log trees, as it forces the transaction kthread to
2009 : * wait for this transaction to be cleaned up by us.
2010 : */
2011 16 : list_del_init(&cur_trans->list);
2012 :
2013 16 : spin_unlock(&fs_info->trans_lock);
2014 :
2015 16 : btrfs_cleanup_one_transaction(trans->transaction, fs_info);
2016 :
2017 16 : spin_lock(&fs_info->trans_lock);
2018 16 : if (cur_trans == fs_info->running_transaction)
2019 12 : fs_info->running_transaction = NULL;
2020 16 : spin_unlock(&fs_info->trans_lock);
2021 :
2022 16 : if (trans->type & __TRANS_FREEZABLE)
2023 2 : sb_end_intwrite(fs_info->sb);
2024 16 : btrfs_put_transaction(cur_trans);
2025 16 : btrfs_put_transaction(cur_trans);
2026 :
2027 16 : trace_btrfs_transaction_commit(fs_info);
2028 :
2029 16 : if (current->journal_info == trans)
2030 0 : current->journal_info = NULL;
2031 :
2032 : /*
2033 : * If relocation is running, we can't cancel scrub because that will
2034 : * result in a deadlock. Before relocating a block group, relocation
2035 : * pauses scrub, then starts and commits a transaction before unpausing
2036 : * scrub. If the transaction commit is being done by the relocation
2037 : * task or triggered by another task and the relocation task is waiting
2038 : * for the commit, and we end up here due to an error in the commit
2039 : * path, then calling btrfs_scrub_cancel() will deadlock, as we are
2040 : * asking for scrub to stop while having it asked to be paused higher
2041 : * above in relocation code.
2042 : */
2043 16 : if (!test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
2044 16 : btrfs_scrub_cancel(fs_info);
2045 :
2046 16 : kmem_cache_free(btrfs_trans_handle_cachep, trans);
2047 16 : }
2048 :
2049 : /*
2050 : * Release reserved delayed ref space of all pending block groups of the
2051 : * transaction and remove them from the list
2052 : */
2053 16 : static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans)
2054 : {
2055 16 : struct btrfs_fs_info *fs_info = trans->fs_info;
2056 16 : struct btrfs_block_group *block_group, *tmp;
2057 :
2058 16 : list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
2059 0 : btrfs_delayed_refs_rsv_release(fs_info, 1);
2060 0 : list_del_init(&block_group->bg_list);
2061 : }
2062 16 : }
2063 :
2064 203272 : static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
2065 : {
2066 : /*
2067 : * We use try_to_writeback_inodes_sb() here because if we used
2068 : * btrfs_start_delalloc_roots we would deadlock with fs freeze.
2069 : * Currently are holding the fs freeze lock, if we do an async flush
2070 : * we'll do btrfs_join_transaction() and deadlock because we need to
2071 : * wait for the fs freeze lock. Using the direct flushing we benefit
2072 : * from already being in a transaction and our join_transaction doesn't
2073 : * have to re-take the fs freeze lock.
2074 : *
2075 : * Note that try_to_writeback_inodes_sb() will only trigger writeback
2076 : * if it can read lock sb->s_umount. It will always be able to lock it,
2077 : * except when the filesystem is being unmounted or being frozen, but in
2078 : * those cases sync_filesystem() is called, which results in calling
2079 : * writeback_inodes_sb() while holding a write lock on sb->s_umount.
2080 : * Note that we don't call writeback_inodes_sb() directly, because it
2081 : * will emit a warning if sb->s_umount is not locked.
2082 : */
2083 203272 : if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2084 3 : try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC);
2085 203272 : return 0;
2086 : }
2087 :
2088 203261 : static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
2089 : {
2090 203261 : if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2091 3 : btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
2092 203261 : }
2093 :
2094 : /*
2095 : * Add a pending snapshot associated with the given transaction handle to the
2096 : * respective handle. This must be called after the transaction commit started
2097 : * and while holding fs_info->trans_lock.
2098 : * This serves to guarantee a caller of btrfs_commit_transaction() that it can
2099 : * safely free the pending snapshot pointer in case btrfs_commit_transaction()
2100 : * returns an error.
2101 : */
2102 205742 : static void add_pending_snapshot(struct btrfs_trans_handle *trans)
2103 : {
2104 205742 : struct btrfs_transaction *cur_trans = trans->transaction;
2105 :
2106 205742 : if (!trans->pending_snapshot)
2107 : return;
2108 :
2109 1028 : lockdep_assert_held(&trans->fs_info->trans_lock);
2110 1028 : ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_START);
2111 :
2112 1028 : list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots);
2113 : }
2114 :
2115 : static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval)
2116 : {
2117 203256 : fs_info->commit_stats.commit_count++;
2118 203256 : fs_info->commit_stats.last_commit_dur = interval;
2119 203256 : fs_info->commit_stats.max_commit_dur =
2120 203256 : max_t(u64, fs_info->commit_stats.max_commit_dur, interval);
2121 203256 : fs_info->commit_stats.total_commit_dur += interval;
2122 : }
2123 :
2124 205753 : int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
2125 : {
2126 205753 : struct btrfs_fs_info *fs_info = trans->fs_info;
2127 205753 : struct btrfs_transaction *cur_trans = trans->transaction;
2128 205753 : struct btrfs_transaction *prev_trans = NULL;
2129 205753 : int ret;
2130 205753 : ktime_t start_time;
2131 205753 : ktime_t interval;
2132 :
2133 205753 : ASSERT(refcount_read(&trans->use_count) == 1);
2134 205753 : btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START);
2135 :
2136 205753 : clear_bit(BTRFS_FS_NEED_TRANS_COMMIT, &fs_info->flags);
2137 :
2138 : /* Stop the commit early if ->aborted is set */
2139 205752 : if (TRANS_ABORTED(cur_trans)) {
2140 0 : ret = cur_trans->aborted;
2141 0 : goto lockdep_trans_commit_start_release;
2142 : }
2143 :
2144 205752 : btrfs_trans_release_metadata(trans);
2145 205753 : trans->block_rsv = NULL;
2146 :
2147 : /*
2148 : * We only want one transaction commit doing the flushing so we do not
2149 : * waste a bunch of time on lock contention on the extent root node.
2150 : */
2151 205753 : if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING,
2152 205753 : &cur_trans->delayed_refs.flags)) {
2153 : /*
2154 : * Make a pass through all the delayed refs we have so far.
2155 : * Any running threads may add more while we are here.
2156 : */
2157 203273 : ret = btrfs_run_delayed_refs(trans, 0);
2158 203273 : if (ret)
2159 1 : goto lockdep_trans_commit_start_release;
2160 : }
2161 :
2162 205752 : btrfs_create_pending_block_groups(trans);
2163 :
2164 205752 : if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
2165 203273 : int run_it = 0;
2166 :
2167 : /* this mutex is also taken before trying to set
2168 : * block groups readonly. We need to make sure
2169 : * that nobody has set a block group readonly
2170 : * after a extents from that block group have been
2171 : * allocated for cache files. btrfs_set_block_group_ro
2172 : * will wait for the transaction to commit if it
2173 : * finds BTRFS_TRANS_DIRTY_BG_RUN set.
2174 : *
2175 : * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
2176 : * only one process starts all the block group IO. It wouldn't
2177 : * hurt to have more than one go through, but there's no
2178 : * real advantage to it either.
2179 : */
2180 203273 : mutex_lock(&fs_info->ro_block_group_mutex);
2181 203274 : if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
2182 : &cur_trans->flags))
2183 203272 : run_it = 1;
2184 203274 : mutex_unlock(&fs_info->ro_block_group_mutex);
2185 :
2186 203274 : if (run_it) {
2187 203272 : ret = btrfs_start_dirty_block_groups(trans);
2188 203272 : if (ret)
2189 0 : goto lockdep_trans_commit_start_release;
2190 : }
2191 : }
2192 :
2193 205753 : spin_lock(&fs_info->trans_lock);
2194 205753 : if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
2195 2481 : enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2196 :
2197 2481 : add_pending_snapshot(trans);
2198 :
2199 2481 : spin_unlock(&fs_info->trans_lock);
2200 2481 : refcount_inc(&cur_trans->use_count);
2201 :
2202 2481 : if (trans->in_fsync)
2203 95 : want_state = TRANS_STATE_SUPER_COMMITTED;
2204 :
2205 2481 : btrfs_trans_state_lockdep_release(fs_info,
2206 : BTRFS_LOCKDEP_TRANS_COMMIT_START);
2207 2481 : ret = btrfs_end_transaction(trans);
2208 2478 : wait_for_commit(cur_trans, want_state);
2209 :
2210 2480 : if (TRANS_ABORTED(cur_trans))
2211 1 : ret = cur_trans->aborted;
2212 :
2213 2480 : btrfs_put_transaction(cur_trans);
2214 :
2215 2480 : return ret;
2216 : }
2217 :
2218 203272 : cur_trans->state = TRANS_STATE_COMMIT_START;
2219 203272 : wake_up(&fs_info->transaction_blocked_wait);
2220 203272 : btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START);
2221 :
2222 203272 : if (cur_trans->list.prev != &fs_info->trans_list) {
2223 3465 : enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2224 :
2225 3465 : if (trans->in_fsync)
2226 41 : want_state = TRANS_STATE_SUPER_COMMITTED;
2227 :
2228 3465 : prev_trans = list_entry(cur_trans->list.prev,
2229 : struct btrfs_transaction, list);
2230 3465 : if (prev_trans->state < want_state) {
2231 3440 : refcount_inc(&prev_trans->use_count);
2232 3440 : spin_unlock(&fs_info->trans_lock);
2233 :
2234 3440 : wait_for_commit(prev_trans, want_state);
2235 :
2236 3440 : ret = READ_ONCE(prev_trans->aborted);
2237 :
2238 3440 : btrfs_put_transaction(prev_trans);
2239 3440 : if (ret)
2240 0 : goto lockdep_release;
2241 : } else {
2242 25 : spin_unlock(&fs_info->trans_lock);
2243 : }
2244 : } else {
2245 199807 : spin_unlock(&fs_info->trans_lock);
2246 : /*
2247 : * The previous transaction was aborted and was already removed
2248 : * from the list of transactions at fs_info->trans_list. So we
2249 : * abort to prevent writing a new superblock that reflects a
2250 : * corrupt state (pointing to trees with unwritten nodes/leafs).
2251 : */
2252 199807 : if (BTRFS_FS_ERROR(fs_info)) {
2253 0 : ret = -EROFS;
2254 0 : goto lockdep_release;
2255 : }
2256 : }
2257 :
2258 : /*
2259 : * Get the time spent on the work done by the commit thread and not
2260 : * the time spent waiting on a previous commit
2261 : */
2262 203272 : start_time = ktime_get_ns();
2263 :
2264 203272 : extwriter_counter_dec(cur_trans, trans->type);
2265 :
2266 203272 : ret = btrfs_start_delalloc_flush(fs_info);
2267 203272 : if (ret)
2268 0 : goto lockdep_release;
2269 :
2270 203272 : ret = btrfs_run_delayed_items(trans);
2271 203272 : if (ret)
2272 11 : goto lockdep_release;
2273 :
2274 : /*
2275 : * The thread has started/joined the transaction thus it holds the
2276 : * lockdep map as a reader. It has to release it before acquiring the
2277 : * lockdep map as a writer.
2278 : */
2279 203261 : btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
2280 203261 : btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters);
2281 212611 : wait_event(cur_trans->writer_wait,
2282 : extwriter_counter_read(cur_trans) == 0);
2283 :
2284 : /* some pending stuffs might be added after the previous flush. */
2285 203261 : ret = btrfs_run_delayed_items(trans);
2286 203261 : if (ret) {
2287 0 : btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2288 0 : goto cleanup_transaction;
2289 : }
2290 :
2291 203261 : btrfs_wait_delalloc_flush(fs_info);
2292 :
2293 : /*
2294 : * Wait for all ordered extents started by a fast fsync that joined this
2295 : * transaction. Otherwise if this transaction commits before the ordered
2296 : * extents complete we lose logged data after a power failure.
2297 : */
2298 203261 : btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered);
2299 203261 : wait_event(cur_trans->pending_wait,
2300 : atomic_read(&cur_trans->pending_ordered) == 0);
2301 :
2302 203261 : btrfs_scrub_pause(fs_info);
2303 : /*
2304 : * Ok now we need to make sure to block out any other joins while we
2305 : * commit the transaction. We could have started a join before setting
2306 : * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2307 : */
2308 203261 : spin_lock(&fs_info->trans_lock);
2309 203261 : add_pending_snapshot(trans);
2310 203261 : cur_trans->state = TRANS_STATE_COMMIT_DOING;
2311 203261 : spin_unlock(&fs_info->trans_lock);
2312 :
2313 : /*
2314 : * The thread has started/joined the transaction thus it holds the
2315 : * lockdep map as a reader. It has to release it before acquiring the
2316 : * lockdep map as a writer.
2317 : */
2318 203261 : btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2319 203261 : btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers);
2320 204064 : wait_event(cur_trans->writer_wait,
2321 : atomic_read(&cur_trans->num_writers) == 1);
2322 :
2323 : /*
2324 : * Make lockdep happy by acquiring the state locks after
2325 : * btrfs_trans_num_writers is released. If we acquired the state locks
2326 : * before releasing the btrfs_trans_num_writers lock then lockdep would
2327 : * complain because we did not follow the reverse order unlocking rule.
2328 : */
2329 203261 : btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2330 203261 : btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2331 203261 : btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2332 :
2333 : /*
2334 : * We've started the commit, clear the flag in case we were triggered to
2335 : * do an async commit but somebody else started before the transaction
2336 : * kthread could do the work.
2337 : */
2338 203261 : clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags);
2339 :
2340 203261 : if (TRANS_ABORTED(cur_trans)) {
2341 0 : ret = cur_trans->aborted;
2342 0 : btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2343 0 : goto scrub_continue;
2344 : }
2345 : /*
2346 : * the reloc mutex makes sure that we stop
2347 : * the balancing code from coming in and moving
2348 : * extents around in the middle of the commit
2349 : */
2350 203261 : mutex_lock(&fs_info->reloc_mutex);
2351 :
2352 : /*
2353 : * We needn't worry about the delayed items because we will
2354 : * deal with them in create_pending_snapshot(), which is the
2355 : * core function of the snapshot creation.
2356 : */
2357 203261 : ret = create_pending_snapshots(trans);
2358 203261 : if (ret)
2359 0 : goto unlock_reloc;
2360 :
2361 : /*
2362 : * We insert the dir indexes of the snapshots and update the inode
2363 : * of the snapshots' parents after the snapshot creation, so there
2364 : * are some delayed items which are not dealt with. Now deal with
2365 : * them.
2366 : *
2367 : * We needn't worry that this operation will corrupt the snapshots,
2368 : * because all the tree which are snapshoted will be forced to COW
2369 : * the nodes and leaves.
2370 : */
2371 203261 : ret = btrfs_run_delayed_items(trans);
2372 203261 : if (ret)
2373 0 : goto unlock_reloc;
2374 :
2375 203261 : ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
2376 203261 : if (ret)
2377 0 : goto unlock_reloc;
2378 :
2379 : /*
2380 : * make sure none of the code above managed to slip in a
2381 : * delayed item
2382 : */
2383 203261 : btrfs_assert_delayed_root_empty(fs_info);
2384 :
2385 203261 : WARN_ON(cur_trans != trans->transaction);
2386 :
2387 203261 : ret = commit_fs_roots(trans);
2388 203261 : if (ret)
2389 0 : goto unlock_reloc;
2390 :
2391 : /* commit_fs_roots gets rid of all the tree log roots, it is now
2392 : * safe to free the root of tree log roots
2393 : */
2394 203261 : btrfs_free_log_root_tree(trans, fs_info);
2395 :
2396 : /*
2397 : * Since fs roots are all committed, we can get a quite accurate
2398 : * new_roots. So let's do quota accounting.
2399 : */
2400 203261 : ret = btrfs_qgroup_account_extents(trans);
2401 203261 : if (ret < 0)
2402 0 : goto unlock_reloc;
2403 :
2404 203261 : ret = commit_cowonly_roots(trans);
2405 203261 : if (ret)
2406 0 : goto unlock_reloc;
2407 :
2408 : /*
2409 : * The tasks which save the space cache and inode cache may also
2410 : * update ->aborted, check it.
2411 : */
2412 203261 : if (TRANS_ABORTED(cur_trans)) {
2413 1 : ret = cur_trans->aborted;
2414 1 : goto unlock_reloc;
2415 : }
2416 :
2417 203260 : cur_trans = fs_info->running_transaction;
2418 :
2419 203260 : btrfs_set_root_node(&fs_info->tree_root->root_item,
2420 203260 : fs_info->tree_root->node);
2421 203260 : list_add_tail(&fs_info->tree_root->dirty_list,
2422 : &cur_trans->switch_commits);
2423 :
2424 203260 : btrfs_set_root_node(&fs_info->chunk_root->root_item,
2425 203260 : fs_info->chunk_root->node);
2426 203260 : list_add_tail(&fs_info->chunk_root->dirty_list,
2427 : &cur_trans->switch_commits);
2428 :
2429 203260 : if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2430 0 : btrfs_set_root_node(&fs_info->block_group_root->root_item,
2431 0 : fs_info->block_group_root->node);
2432 0 : list_add_tail(&fs_info->block_group_root->dirty_list,
2433 : &cur_trans->switch_commits);
2434 : }
2435 :
2436 203260 : switch_commit_roots(trans);
2437 :
2438 203260 : ASSERT(list_empty(&cur_trans->dirty_bgs));
2439 203260 : ASSERT(list_empty(&cur_trans->io_bgs));
2440 203260 : update_super_roots(fs_info);
2441 :
2442 203260 : btrfs_set_super_log_root(fs_info->super_copy, 0);
2443 203260 : btrfs_set_super_log_root_level(fs_info->super_copy, 0);
2444 406520 : memcpy(fs_info->super_for_commit, fs_info->super_copy,
2445 : sizeof(*fs_info->super_copy));
2446 :
2447 203260 : btrfs_commit_device_sizes(cur_trans);
2448 :
2449 203260 : clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
2450 203260 : clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
2451 :
2452 203260 : btrfs_trans_release_chunk_metadata(trans);
2453 :
2454 : /*
2455 : * Before changing the transaction state to TRANS_STATE_UNBLOCKED and
2456 : * setting fs_info->running_transaction to NULL, lock tree_log_mutex to
2457 : * make sure that before we commit our superblock, no other task can
2458 : * start a new transaction and commit a log tree before we commit our
2459 : * superblock. Anyone trying to commit a log tree locks this mutex before
2460 : * writing its superblock.
2461 : */
2462 203260 : mutex_lock(&fs_info->tree_log_mutex);
2463 :
2464 203260 : spin_lock(&fs_info->trans_lock);
2465 203260 : cur_trans->state = TRANS_STATE_UNBLOCKED;
2466 203260 : fs_info->running_transaction = NULL;
2467 203260 : spin_unlock(&fs_info->trans_lock);
2468 203260 : mutex_unlock(&fs_info->reloc_mutex);
2469 :
2470 203260 : wake_up(&fs_info->transaction_wait);
2471 203260 : btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2472 :
2473 : /* If we have features changed, wake up the cleaner to update sysfs. */
2474 406520 : if (test_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags) &&
2475 37 : fs_info->cleaner_kthread)
2476 37 : wake_up_process(fs_info->cleaner_kthread);
2477 :
2478 203260 : ret = btrfs_write_and_wait_transaction(trans);
2479 203260 : if (ret) {
2480 4 : btrfs_handle_fs_error(fs_info, ret,
2481 : "Error while writing out transaction");
2482 4 : mutex_unlock(&fs_info->tree_log_mutex);
2483 4 : goto scrub_continue;
2484 : }
2485 :
2486 203256 : ret = write_all_supers(fs_info, 0);
2487 : /*
2488 : * the super is written, we can safely allow the tree-loggers
2489 : * to go about their business
2490 : */
2491 203256 : mutex_unlock(&fs_info->tree_log_mutex);
2492 203256 : if (ret)
2493 0 : goto scrub_continue;
2494 :
2495 : /*
2496 : * We needn't acquire the lock here because there is no other task
2497 : * which can change it.
2498 : */
2499 203256 : cur_trans->state = TRANS_STATE_SUPER_COMMITTED;
2500 203256 : wake_up(&cur_trans->commit_wait);
2501 203256 : btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2502 :
2503 203256 : btrfs_finish_extent_commit(trans);
2504 :
2505 406512 : if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
2506 508 : btrfs_clear_space_info_full(fs_info);
2507 :
2508 203256 : fs_info->last_trans_committed = cur_trans->transid;
2509 : /*
2510 : * We needn't acquire the lock here because there is no other task
2511 : * which can change it.
2512 : */
2513 203256 : cur_trans->state = TRANS_STATE_COMPLETED;
2514 203256 : wake_up(&cur_trans->commit_wait);
2515 203256 : btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2516 :
2517 203256 : spin_lock(&fs_info->trans_lock);
2518 203256 : list_del_init(&cur_trans->list);
2519 203256 : spin_unlock(&fs_info->trans_lock);
2520 :
2521 203256 : btrfs_put_transaction(cur_trans);
2522 203256 : btrfs_put_transaction(cur_trans);
2523 :
2524 203256 : if (trans->type & __TRANS_FREEZABLE)
2525 188704 : sb_end_intwrite(fs_info->sb);
2526 :
2527 203256 : trace_btrfs_transaction_commit(fs_info);
2528 :
2529 203256 : interval = ktime_get_ns() - start_time;
2530 :
2531 203256 : btrfs_scrub_continue(fs_info);
2532 :
2533 203256 : if (current->journal_info == trans)
2534 203256 : current->journal_info = NULL;
2535 :
2536 203256 : kmem_cache_free(btrfs_trans_handle_cachep, trans);
2537 :
2538 203256 : update_commit_stats(fs_info, interval);
2539 :
2540 203256 : return ret;
2541 :
2542 1 : unlock_reloc:
2543 1 : mutex_unlock(&fs_info->reloc_mutex);
2544 5 : btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2545 5 : scrub_continue:
2546 5 : btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2547 5 : btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2548 5 : btrfs_scrub_continue(fs_info);
2549 16 : cleanup_transaction:
2550 16 : btrfs_trans_release_metadata(trans);
2551 16 : btrfs_cleanup_pending_block_groups(trans);
2552 16 : btrfs_trans_release_chunk_metadata(trans);
2553 16 : trans->block_rsv = NULL;
2554 16 : btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
2555 16 : if (current->journal_info == trans)
2556 16 : current->journal_info = NULL;
2557 16 : cleanup_transaction(trans, ret);
2558 :
2559 16 : return ret;
2560 :
2561 11 : lockdep_release:
2562 11 : btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
2563 11 : btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2564 11 : goto cleanup_transaction;
2565 :
2566 1 : lockdep_trans_commit_start_release:
2567 1 : btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START);
2568 1 : btrfs_end_transaction(trans);
2569 1 : return ret;
2570 : }
2571 :
2572 : /*
2573 : * return < 0 if error
2574 : * 0 if there are no more dead_roots at the time of call
2575 : * 1 there are more to be processed, call me again
2576 : *
2577 : * The return value indicates there are certainly more snapshots to delete, but
2578 : * if there comes a new one during processing, it may return 0. We don't mind,
2579 : * because btrfs_commit_super will poke cleaner thread and it will process it a
2580 : * few seconds later.
2581 : */
2582 41504 : int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info)
2583 : {
2584 41504 : struct btrfs_root *root;
2585 41504 : int ret;
2586 :
2587 41504 : spin_lock(&fs_info->trans_lock);
2588 41504 : if (list_empty(&fs_info->dead_roots)) {
2589 41386 : spin_unlock(&fs_info->trans_lock);
2590 41386 : return 0;
2591 : }
2592 118 : root = list_first_entry(&fs_info->dead_roots,
2593 : struct btrfs_root, root_list);
2594 118 : list_del_init(&root->root_list);
2595 118 : spin_unlock(&fs_info->trans_lock);
2596 :
2597 118 : btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid);
2598 :
2599 118 : btrfs_kill_all_delayed_nodes(root);
2600 :
2601 118 : if (btrfs_header_backref_rev(root->node) <
2602 : BTRFS_MIXED_BACKREF_REV)
2603 0 : ret = btrfs_drop_snapshot(root, 0, 0);
2604 : else
2605 118 : ret = btrfs_drop_snapshot(root, 1, 0);
2606 :
2607 118 : btrfs_put_root(root);
2608 118 : return (ret < 0) ? 0 : 1;
2609 : }
2610 :
2611 : /*
2612 : * We only mark the transaction aborted and then set the file system read-only.
2613 : * This will prevent new transactions from starting or trying to join this
2614 : * one.
2615 : *
2616 : * This means that error recovery at the call site is limited to freeing
2617 : * any local memory allocations and passing the error code up without
2618 : * further cleanup. The transaction should complete as it normally would
2619 : * in the call path but will return -EIO.
2620 : *
2621 : * We'll complete the cleanup in btrfs_end_transaction and
2622 : * btrfs_commit_transaction.
2623 : */
2624 35 : void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
2625 : const char *function,
2626 : unsigned int line, int errno, bool first_hit)
2627 : {
2628 35 : struct btrfs_fs_info *fs_info = trans->fs_info;
2629 :
2630 35 : WRITE_ONCE(trans->aborted, errno);
2631 35 : WRITE_ONCE(trans->transaction->aborted, errno);
2632 35 : if (first_hit && errno == -ENOSPC)
2633 0 : btrfs_dump_space_info_for_trans_abort(fs_info);
2634 : /* Wake up anybody who may be waiting on this transaction */
2635 35 : wake_up(&fs_info->transaction_wait);
2636 35 : wake_up(&fs_info->transaction_blocked_wait);
2637 35 : __btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
2638 35 : }
2639 :
2640 11 : int __init btrfs_transaction_init(void)
2641 : {
2642 11 : btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
2643 : sizeof(struct btrfs_trans_handle), 0,
2644 : SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
2645 11 : if (!btrfs_trans_handle_cachep)
2646 0 : return -ENOMEM;
2647 : return 0;
2648 : }
2649 :
2650 0 : void __cold btrfs_transaction_exit(void)
2651 : {
2652 0 : kmem_cache_destroy(btrfs_trans_handle_cachep);
2653 0 : }
|
