?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
gro: Allow tunnel stacking in the case of FOU/GUE
[linux/fpc-iii.git] / fs / btrfs / transaction.c
blob6d43b2ab183b5f6c9156ce025e93524c379fcbac
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/fs.h>
20 #include <linux/slab.h>
21 #include <linux/sched.h>
22 #include <linux/writeback.h>
23 #include <linux/pagemap.h>
24 #include <linux/blkdev.h>
25 #include <linux/uuid.h>
26 #include "ctree.h"
27 #include "disk-io.h"
28 #include "transaction.h"
29 #include "locking.h"
30 #include "tree-log.h"
31 #include "inode-map.h"
32 #include "volumes.h"
33 #include "dev-replace.h"
34 #include "qgroup.h"
35
36 #define BTRFS_ROOT_TRANS_TAG 0
37
38 static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
39 [TRANS_STATE_RUNNING] = 0U,
40 [TRANS_STATE_BLOCKED] = (__TRANS_USERSPACE |
41 __TRANS_START),
42 [TRANS_STATE_COMMIT_START] = (__TRANS_USERSPACE |
43 __TRANS_START |
44 __TRANS_ATTACH),
45 [TRANS_STATE_COMMIT_DOING] = (__TRANS_USERSPACE |
46 __TRANS_START |
47 __TRANS_ATTACH |
48 __TRANS_JOIN),
49 [TRANS_STATE_UNBLOCKED] = (__TRANS_USERSPACE |
50 __TRANS_START |
51 __TRANS_ATTACH |
52 __TRANS_JOIN |
53 __TRANS_JOIN_NOLOCK),
54 [TRANS_STATE_COMPLETED] = (__TRANS_USERSPACE |
55 __TRANS_START |
56 __TRANS_ATTACH |
57 __TRANS_JOIN |
58 __TRANS_JOIN_NOLOCK),
59 };
60
61 void btrfs_put_transaction(struct btrfs_transaction *transaction)
62 {
63 WARN_ON(atomic_read(&transaction->use_count) == 0);
64 if (atomic_dec_and_test(&transaction->use_count)) {
65 BUG_ON(!list_empty(&transaction->list));
66 WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root));
67 if (transaction->delayed_refs.pending_csums)
68 printk(KERN_ERR "pending csums is %llu\n",
69 transaction->delayed_refs.pending_csums);
70 while (!list_empty(&transaction->pending_chunks)) {
71 struct extent_map *em;
72
73 em = list_first_entry(&transaction->pending_chunks,
74 struct extent_map, list);
75 list_del_init(&em->list);
76 free_extent_map(em);
77 }
78 kmem_cache_free(btrfs_transaction_cachep, transaction);
79 }
80 }
81
82 static void clear_btree_io_tree(struct extent_io_tree *tree)
83 {
84 spin_lock(&tree->lock);
85 while (!RB_EMPTY_ROOT(&tree->state)) {
86 struct rb_node *node;
87 struct extent_state *state;
88
89 node = rb_first(&tree->state);
90 state = rb_entry(node, struct extent_state, rb_node);
91 rb_erase(&state->rb_node, &tree->state);
92 RB_CLEAR_NODE(&state->rb_node);
93 /*
94 * btree io trees aren't supposed to have tasks waiting for
95 * changes in the flags of extent states ever.
96 */
97 ASSERT(!waitqueue_active(&state->wq));
98 free_extent_state(state);
99
100 cond_resched_lock(&tree->lock);
101 }
102 spin_unlock(&tree->lock);
103 }
104
105 static noinline void switch_commit_roots(struct btrfs_transaction *trans,
106 struct btrfs_fs_info *fs_info)
107 {
108 struct btrfs_root *root, *tmp;
109
110 down_write(&fs_info->commit_root_sem);
111 list_for_each_entry_safe(root, tmp, &trans->switch_commits,
112 dirty_list) {
113 list_del_init(&root->dirty_list);
114 free_extent_buffer(root->commit_root);
115 root->commit_root = btrfs_root_node(root);
116 if (is_fstree(root->objectid))
117 btrfs_unpin_free_ino(root);
118 clear_btree_io_tree(&root->dirty_log_pages);
119 }
120 up_write(&fs_info->commit_root_sem);
121 }
122
123 static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
124 unsigned int type)
125 {
126 if (type & TRANS_EXTWRITERS)
127 atomic_inc(&trans->num_extwriters);
128 }
129
130 static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
131 unsigned int type)
132 {
133 if (type & TRANS_EXTWRITERS)
134 atomic_dec(&trans->num_extwriters);
135 }
136
137 static inline void extwriter_counter_init(struct btrfs_transaction *trans,
138 unsigned int type)
139 {
140 atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
141 }
142
143 static inline int extwriter_counter_read(struct btrfs_transaction *trans)
144 {
145 return atomic_read(&trans->num_extwriters);
146 }
147
148 /*
149 * either allocate a new transaction or hop into the existing one
150 */
151 static noinline int join_transaction(struct btrfs_root *root, unsigned int type)
152 {
153 struct btrfs_transaction *cur_trans;
154 struct btrfs_fs_info *fs_info = root->fs_info;
155
156 spin_lock(&fs_info->trans_lock);
157 loop:
158 /* The file system has been taken offline. No new transactions. */
159 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
160 spin_unlock(&fs_info->trans_lock);
161 return -EROFS;
162 }
163
164 cur_trans = fs_info->running_transaction;
165 if (cur_trans) {
166 if (cur_trans->aborted) {
167 spin_unlock(&fs_info->trans_lock);
168 return cur_trans->aborted;
169 }
170 if (btrfs_blocked_trans_types[cur_trans->state] & type) {
171 spin_unlock(&fs_info->trans_lock);
172 return -EBUSY;
173 }
174 atomic_inc(&cur_trans->use_count);
175 atomic_inc(&cur_trans->num_writers);
176 extwriter_counter_inc(cur_trans, type);
177 spin_unlock(&fs_info->trans_lock);
178 return 0;
179 }
180 spin_unlock(&fs_info->trans_lock);
181
182 /*
183 * If we are ATTACH, we just want to catch the current transaction,
184 * and commit it. If there is no transaction, just return ENOENT.
185 */
186 if (type == TRANS_ATTACH)
187 return -ENOENT;
188
189 /*
190 * JOIN_NOLOCK only happens during the transaction commit, so
191 * it is impossible that ->running_transaction is NULL
192 */
193 BUG_ON(type == TRANS_JOIN_NOLOCK);
194
195 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
196 if (!cur_trans)
197 return -ENOMEM;
198
199 spin_lock(&fs_info->trans_lock);
200 if (fs_info->running_transaction) {
201 /*
202 * someone started a transaction after we unlocked. Make sure
203 * to redo the checks above
204 */
205 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
206 goto loop;
207 } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
208 spin_unlock(&fs_info->trans_lock);
209 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
210 return -EROFS;
211 }
212
213 atomic_set(&cur_trans->num_writers, 1);
214 extwriter_counter_init(cur_trans, type);
215 init_waitqueue_head(&cur_trans->writer_wait);
216 init_waitqueue_head(&cur_trans->commit_wait);
217 cur_trans->state = TRANS_STATE_RUNNING;
218 /*
219 * One for this trans handle, one so it will live on until we
220 * commit the transaction.
221 */
222 atomic_set(&cur_trans->use_count, 2);
223 cur_trans->have_free_bgs = 0;
224 cur_trans->start_time = get_seconds();
225 cur_trans->dirty_bg_run = 0;
226
227 cur_trans->delayed_refs.href_root = RB_ROOT;
228 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
229 cur_trans->delayed_refs.num_heads_ready = 0;
230 cur_trans->delayed_refs.pending_csums = 0;
231 cur_trans->delayed_refs.num_heads = 0;
232 cur_trans->delayed_refs.flushing = 0;
233 cur_trans->delayed_refs.run_delayed_start = 0;
234
235 /*
236 * although the tree mod log is per file system and not per transaction,
237 * the log must never go across transaction boundaries.
238 */
239 smp_mb();
240 if (!list_empty(&fs_info->tree_mod_seq_list))
241 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when "
242 "creating a fresh transaction\n");
243 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
244 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when "
245 "creating a fresh transaction\n");
246 atomic64_set(&fs_info->tree_mod_seq, 0);
247
248 spin_lock_init(&cur_trans->delayed_refs.lock);
249
250 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
251 INIT_LIST_HEAD(&cur_trans->pending_chunks);
252 INIT_LIST_HEAD(&cur_trans->switch_commits);
253 INIT_LIST_HEAD(&cur_trans->pending_ordered);
254 INIT_LIST_HEAD(&cur_trans->dirty_bgs);
255 INIT_LIST_HEAD(&cur_trans->io_bgs);
256 mutex_init(&cur_trans->cache_write_mutex);
257 cur_trans->num_dirty_bgs = 0;
258 spin_lock_init(&cur_trans->dirty_bgs_lock);
259 list_add_tail(&cur_trans->list, &fs_info->trans_list);
260 extent_io_tree_init(&cur_trans->dirty_pages,
261 fs_info->btree_inode->i_mapping);
262 fs_info->generation++;
263 cur_trans->transid = fs_info->generation;
264 fs_info->running_transaction = cur_trans;
265 cur_trans->aborted = 0;
266 spin_unlock(&fs_info->trans_lock);
267
268 return 0;
269 }
270
271 /*
272 * this does all the record keeping required to make sure that a reference
273 * counted root is properly recorded in a given transaction. This is required
274 * to make sure the old root from before we joined the transaction is deleted
275 * when the transaction commits
276 */
277 static int record_root_in_trans(struct btrfs_trans_handle *trans,
278 struct btrfs_root *root)
279 {
280 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
281 root->last_trans < trans->transid) {
282 WARN_ON(root == root->fs_info->extent_root);
283 WARN_ON(root->commit_root != root->node);
284
285 /*
286 * see below for IN_TRANS_SETUP usage rules
287 * we have the reloc mutex held now, so there
288 * is only one writer in this function
289 */
290 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
291
292 /* make sure readers find IN_TRANS_SETUP before
293 * they find our root->last_trans update
294 */
295 smp_wmb();
296
297 spin_lock(&root->fs_info->fs_roots_radix_lock);
298 if (root->last_trans == trans->transid) {
299 spin_unlock(&root->fs_info->fs_roots_radix_lock);
300 return 0;
301 }
302 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
303 (unsigned long)root->root_key.objectid,
304 BTRFS_ROOT_TRANS_TAG);
305 spin_unlock(&root->fs_info->fs_roots_radix_lock);
306 root->last_trans = trans->transid;
307
308 /* this is pretty tricky. We don't want to
309 * take the relocation lock in btrfs_record_root_in_trans
310 * unless we're really doing the first setup for this root in
311 * this transaction.
312 *
313 * Normally we'd use root->last_trans as a flag to decide
314 * if we want to take the expensive mutex.
315 *
316 * But, we have to set root->last_trans before we
317 * init the relocation root, otherwise, we trip over warnings
318 * in ctree.c. The solution used here is to flag ourselves
319 * with root IN_TRANS_SETUP. When this is 1, we're still
320 * fixing up the reloc trees and everyone must wait.
321 *
322 * When this is zero, they can trust root->last_trans and fly
323 * through btrfs_record_root_in_trans without having to take the
324 * lock. smp_wmb() makes sure that all the writes above are
325 * done before we pop in the zero below
326 */
327 btrfs_init_reloc_root(trans, root);
328 smp_mb__before_atomic();
329 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
330 }
331 return 0;
332 }
333
334
335 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
336 struct btrfs_root *root)
337 {
338 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
339 return 0;
340
341 /*
342 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
343 * and barriers
344 */
345 smp_rmb();
346 if (root->last_trans == trans->transid &&
347 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
348 return 0;
349
350 mutex_lock(&root->fs_info->reloc_mutex);
351 record_root_in_trans(trans, root);
352 mutex_unlock(&root->fs_info->reloc_mutex);
353
354 return 0;
355 }
356
357 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
358 {
359 return (trans->state >= TRANS_STATE_BLOCKED &&
360 trans->state < TRANS_STATE_UNBLOCKED &&
361 !trans->aborted);
362 }
363
364 /* wait for commit against the current transaction to become unblocked
365 * when this is done, it is safe to start a new transaction, but the current
366 * transaction might not be fully on disk.
367 */
368 static void wait_current_trans(struct btrfs_root *root)
369 {
370 struct btrfs_transaction *cur_trans;
371
372 spin_lock(&root->fs_info->trans_lock);
373 cur_trans = root->fs_info->running_transaction;
374 if (cur_trans && is_transaction_blocked(cur_trans)) {
375 atomic_inc(&cur_trans->use_count);
376 spin_unlock(&root->fs_info->trans_lock);
377
378 wait_event(root->fs_info->transaction_wait,
379 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
380 cur_trans->aborted);
381 btrfs_put_transaction(cur_trans);
382 } else {
383 spin_unlock(&root->fs_info->trans_lock);
384 }
385 }
386
387 static int may_wait_transaction(struct btrfs_root *root, int type)
388 {
389 if (root->fs_info->log_root_recovering)
390 return 0;
391
392 if (type == TRANS_USERSPACE)
393 return 1;
394
395 if (type == TRANS_START &&
396 !atomic_read(&root->fs_info->open_ioctl_trans))
397 return 1;
398
399 return 0;
400 }
401
402 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
403 {
404 if (!root->fs_info->reloc_ctl ||
405 !test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
406 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
407 root->reloc_root)
408 return false;
409
410 return true;
411 }
412
413 static struct btrfs_trans_handle *
414 start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type,
415 enum btrfs_reserve_flush_enum flush)
416 {
417 struct btrfs_trans_handle *h;
418 struct btrfs_transaction *cur_trans;
419 u64 num_bytes = 0;
420 u64 qgroup_reserved = 0;
421 bool reloc_reserved = false;
422 int ret;
423
424 /* Send isn't supposed to start transactions. */
425 ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
426
427 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
428 return ERR_PTR(-EROFS);
429
430 if (current->journal_info) {
431 WARN_ON(type & TRANS_EXTWRITERS);
432 h = current->journal_info;
433 h->use_count++;
434 WARN_ON(h->use_count > 2);
435 h->orig_rsv = h->block_rsv;
436 h->block_rsv = NULL;
437 goto got_it;
438 }
439
440 /*
441 * Do the reservation before we join the transaction so we can do all
442 * the appropriate flushing if need be.
443 */
444 if (num_items > 0 && root != root->fs_info->chunk_root) {
445 if (root->fs_info->quota_enabled &&
446 is_fstree(root->root_key.objectid)) {
447 qgroup_reserved = num_items * root->nodesize;
448 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
449 if (ret)
450 return ERR_PTR(ret);
451 }
452
453 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
454 /*
455 * Do the reservation for the relocation root creation
456 */
457 if (need_reserve_reloc_root(root)) {
458 num_bytes += root->nodesize;
459 reloc_reserved = true;
460 }
461
462 ret = btrfs_block_rsv_add(root,
463 &root->fs_info->trans_block_rsv,
464 num_bytes, flush);
465 if (ret)
466 goto reserve_fail;
467 }
468 again:
469 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
470 if (!h) {
471 ret = -ENOMEM;
472 goto alloc_fail;
473 }
474
475 /*
476 * If we are JOIN_NOLOCK we're already committing a transaction and
477 * waiting on this guy, so we don't need to do the sb_start_intwrite
478 * because we're already holding a ref. We need this because we could
479 * have raced in and did an fsync() on a file which can kick a commit
480 * and then we deadlock with somebody doing a freeze.
481 *
482 * If we are ATTACH, it means we just want to catch the current
483 * transaction and commit it, so we needn't do sb_start_intwrite().
484 */
485 if (type & __TRANS_FREEZABLE)
486 sb_start_intwrite(root->fs_info->sb);
487
488 if (may_wait_transaction(root, type))
489 wait_current_trans(root);
490
491 do {
492 ret = join_transaction(root, type);
493 if (ret == -EBUSY) {
494 wait_current_trans(root);
495 if (unlikely(type == TRANS_ATTACH))
496 ret = -ENOENT;
497 }
498 } while (ret == -EBUSY);
499
500 if (ret < 0) {
501 /* We must get the transaction if we are JOIN_NOLOCK. */
502 BUG_ON(type == TRANS_JOIN_NOLOCK);
503 goto join_fail;
504 }
505
506 cur_trans = root->fs_info->running_transaction;
507
508 h->transid = cur_trans->transid;
509 h->transaction = cur_trans;
510 h->bytes_reserved = 0;
511 h->root = root;
512 h->delayed_ref_updates = 0;
513 h->use_count = 1;
514 h->adding_csums = 0;
515 h->block_rsv = NULL;
516 h->orig_rsv = NULL;
517 h->aborted = 0;
518 h->qgroup_reserved = 0;
519 h->delayed_ref_elem.seq = 0;
520 h->type = type;
521 h->allocating_chunk = false;
522 h->reloc_reserved = false;
523 h->sync = false;
524 INIT_LIST_HEAD(&h->qgroup_ref_list);
525 INIT_LIST_HEAD(&h->new_bgs);
526 INIT_LIST_HEAD(&h->ordered);
527
528 smp_mb();
529 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
530 may_wait_transaction(root, type)) {
531 current->journal_info = h;
532 btrfs_commit_transaction(h, root);
533 goto again;
534 }
535
536 if (num_bytes) {
537 trace_btrfs_space_reservation(root->fs_info, "transaction",
538 h->transid, num_bytes, 1);
539 h->block_rsv = &root->fs_info->trans_block_rsv;
540 h->bytes_reserved = num_bytes;
541 h->reloc_reserved = reloc_reserved;
542 }
543 h->qgroup_reserved = qgroup_reserved;
544
545 got_it:
546 btrfs_record_root_in_trans(h, root);
547
548 if (!current->journal_info && type != TRANS_USERSPACE)
549 current->journal_info = h;
550 return h;
551
552 join_fail:
553 if (type & __TRANS_FREEZABLE)
554 sb_end_intwrite(root->fs_info->sb);
555 kmem_cache_free(btrfs_trans_handle_cachep, h);
556 alloc_fail:
557 if (num_bytes)
558 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
559 num_bytes);
560 reserve_fail:
561 if (qgroup_reserved)
562 btrfs_qgroup_free(root, qgroup_reserved);
563 return ERR_PTR(ret);
564 }
565
566 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
567 int num_items)
568 {
569 return start_transaction(root, num_items, TRANS_START,
570 BTRFS_RESERVE_FLUSH_ALL);
571 }
572
573 struct btrfs_trans_handle *btrfs_start_transaction_lflush(
574 struct btrfs_root *root, int num_items)
575 {
576 return start_transaction(root, num_items, TRANS_START,
577 BTRFS_RESERVE_FLUSH_LIMIT);
578 }
579
580 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
581 {
582 return start_transaction(root, 0, TRANS_JOIN, 0);
583 }
584
585 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
586 {
587 return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
588 }
589
590 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
591 {
592 return start_transaction(root, 0, TRANS_USERSPACE, 0);
593 }
594
595 /*
596 * btrfs_attach_transaction() - catch the running transaction
597 *
598 * It is used when we want to commit the current the transaction, but
599 * don't want to start a new one.
600 *
601 * Note: If this function return -ENOENT, it just means there is no
602 * running transaction. But it is possible that the inactive transaction
603 * is still in the memory, not fully on disk. If you hope there is no
604 * inactive transaction in the fs when -ENOENT is returned, you should
605 * invoke
606 * btrfs_attach_transaction_barrier()
607 */
608 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
609 {
610 return start_transaction(root, 0, TRANS_ATTACH, 0);
611 }
612
613 /*
614 * btrfs_attach_transaction_barrier() - catch the running transaction
615 *
616 * It is similar to the above function, the differentia is this one
617 * will wait for all the inactive transactions until they fully
618 * complete.
619 */
620 struct btrfs_trans_handle *
621 btrfs_attach_transaction_barrier(struct btrfs_root *root)
622 {
623 struct btrfs_trans_handle *trans;
624
625 trans = start_transaction(root, 0, TRANS_ATTACH, 0);
626 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
627 btrfs_wait_for_commit(root, 0);
628
629 return trans;
630 }
631
632 /* wait for a transaction commit to be fully complete */
633 static noinline void wait_for_commit(struct btrfs_root *root,
634 struct btrfs_transaction *commit)
635 {
636 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
637 }
638
639 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
640 {
641 struct btrfs_transaction *cur_trans = NULL, *t;
642 int ret = 0;
643
644 if (transid) {
645 if (transid <= root->fs_info->last_trans_committed)
646 goto out;
647
648 /* find specified transaction */
649 spin_lock(&root->fs_info->trans_lock);
650 list_for_each_entry(t, &root->fs_info->trans_list, list) {
651 if (t->transid == transid) {
652 cur_trans = t;
653 atomic_inc(&cur_trans->use_count);
654 ret = 0;
655 break;
656 }
657 if (t->transid > transid) {
658 ret = 0;
659 break;
660 }
661 }
662 spin_unlock(&root->fs_info->trans_lock);
663
664 /*
665 * The specified transaction doesn't exist, or we
666 * raced with btrfs_commit_transaction
667 */
668 if (!cur_trans) {
669 if (transid > root->fs_info->last_trans_committed)
670 ret = -EINVAL;
671 goto out;
672 }
673 } else {
674 /* find newest transaction that is committing | committed */
675 spin_lock(&root->fs_info->trans_lock);
676 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
677 list) {
678 if (t->state >= TRANS_STATE_COMMIT_START) {
679 if (t->state == TRANS_STATE_COMPLETED)
680 break;
681 cur_trans = t;
682 atomic_inc(&cur_trans->use_count);
683 break;
684 }
685 }
686 spin_unlock(&root->fs_info->trans_lock);
687 if (!cur_trans)
688 goto out; /* nothing committing|committed */
689 }
690
691 wait_for_commit(root, cur_trans);
692 btrfs_put_transaction(cur_trans);
693 out:
694 return ret;
695 }
696
697 void btrfs_throttle(struct btrfs_root *root)
698 {
699 if (!atomic_read(&root->fs_info->open_ioctl_trans))
700 wait_current_trans(root);
701 }
702
703 static int should_end_transaction(struct btrfs_trans_handle *trans,
704 struct btrfs_root *root)
705 {
706 if (root->fs_info->global_block_rsv.space_info->full &&
707 btrfs_check_space_for_delayed_refs(trans, root))
708 return 1;
709
710 return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
711 }
712
713 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
714 struct btrfs_root *root)
715 {
716 struct btrfs_transaction *cur_trans = trans->transaction;
717 int updates;
718 int err;
719
720 smp_mb();
721 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
722 cur_trans->delayed_refs.flushing)
723 return 1;
724
725 updates = trans->delayed_ref_updates;
726 trans->delayed_ref_updates = 0;
727 if (updates) {
728 err = btrfs_run_delayed_refs(trans, root, updates * 2);
729 if (err) /* Error code will also eval true */
730 return err;
731 }
732
733 return should_end_transaction(trans, root);
734 }
735
736 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
737 struct btrfs_root *root, int throttle)
738 {
739 struct btrfs_transaction *cur_trans = trans->transaction;
740 struct btrfs_fs_info *info = root->fs_info;
741 unsigned long cur = trans->delayed_ref_updates;
742 int lock = (trans->type != TRANS_JOIN_NOLOCK);
743 int err = 0;
744 int must_run_delayed_refs = 0;
745
746 if (trans->use_count > 1) {
747 trans->use_count--;
748 trans->block_rsv = trans->orig_rsv;
749 return 0;
750 }
751
752 btrfs_trans_release_metadata(trans, root);
753 trans->block_rsv = NULL;
754
755 if (!list_empty(&trans->new_bgs))
756 btrfs_create_pending_block_groups(trans, root);
757
758 if (!list_empty(&trans->ordered)) {
759 spin_lock(&info->trans_lock);
760 list_splice_init(&trans->ordered, &cur_trans->pending_ordered);
761 spin_unlock(&info->trans_lock);
762 }
763
764 trans->delayed_ref_updates = 0;
765 if (!trans->sync) {
766 must_run_delayed_refs =
767 btrfs_should_throttle_delayed_refs(trans, root);
768 cur = max_t(unsigned long, cur, 32);
769
770 /*
771 * don't make the caller wait if they are from a NOLOCK
772 * or ATTACH transaction, it will deadlock with commit
773 */
774 if (must_run_delayed_refs == 1 &&
775 (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH)))
776 must_run_delayed_refs = 2;
777 }
778
779 if (trans->qgroup_reserved) {
780 /*
781 * the same root has to be passed here between start_transaction
782 * and end_transaction. Subvolume quota depends on this.
783 */
784 btrfs_qgroup_free(trans->root, trans->qgroup_reserved);
785 trans->qgroup_reserved = 0;
786 }
787
788 btrfs_trans_release_metadata(trans, root);
789 trans->block_rsv = NULL;
790
791 if (!list_empty(&trans->new_bgs))
792 btrfs_create_pending_block_groups(trans, root);
793
794 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
795 should_end_transaction(trans, root) &&
796 ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
797 spin_lock(&info->trans_lock);
798 if (cur_trans->state == TRANS_STATE_RUNNING)
799 cur_trans->state = TRANS_STATE_BLOCKED;
800 spin_unlock(&info->trans_lock);
801 }
802
803 if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
804 if (throttle)
805 return btrfs_commit_transaction(trans, root);
806 else
807 wake_up_process(info->transaction_kthread);
808 }
809
810 if (trans->type & __TRANS_FREEZABLE)
811 sb_end_intwrite(root->fs_info->sb);
812
813 WARN_ON(cur_trans != info->running_transaction);
814 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
815 atomic_dec(&cur_trans->num_writers);
816 extwriter_counter_dec(cur_trans, trans->type);
817
818 smp_mb();
819 if (waitqueue_active(&cur_trans->writer_wait))
820 wake_up(&cur_trans->writer_wait);
821 btrfs_put_transaction(cur_trans);
822
823 if (current->journal_info == trans)
824 current->journal_info = NULL;
825
826 if (throttle)
827 btrfs_run_delayed_iputs(root);
828
829 if (trans->aborted ||
830 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
831 wake_up_process(info->transaction_kthread);
832 err = -EIO;
833 }
834 assert_qgroups_uptodate(trans);
835
836 kmem_cache_free(btrfs_trans_handle_cachep, trans);
837 if (must_run_delayed_refs) {
838 btrfs_async_run_delayed_refs(root, cur,
839 must_run_delayed_refs == 1);
840 }
841 return err;
842 }
843
844 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
845 struct btrfs_root *root)
846 {
847 return __btrfs_end_transaction(trans, root, 0);
848 }
849
850 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
851 struct btrfs_root *root)
852 {
853 return __btrfs_end_transaction(trans, root, 1);
854 }
855
856 /*
857 * when btree blocks are allocated, they have some corresponding bits set for
858 * them in one of two extent_io trees. This is used to make sure all of
859 * those extents are sent to disk but does not wait on them
860 */
861 int btrfs_write_marked_extents(struct btrfs_root *root,
862 struct extent_io_tree *dirty_pages, int mark)
863 {
864 int err = 0;
865 int werr = 0;
866 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
867 struct extent_state *cached_state = NULL;
868 u64 start = 0;
869 u64 end;
870
871 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
872 mark, &cached_state)) {
873 bool wait_writeback = false;
874
875 err = convert_extent_bit(dirty_pages, start, end,
876 EXTENT_NEED_WAIT,
877 mark, &cached_state, GFP_NOFS);
878 /*
879 * convert_extent_bit can return -ENOMEM, which is most of the
880 * time a temporary error. So when it happens, ignore the error
881 * and wait for writeback of this range to finish - because we
882 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
883 * to btrfs_wait_marked_extents() would not know that writeback
884 * for this range started and therefore wouldn't wait for it to
885 * finish - we don't want to commit a superblock that points to
886 * btree nodes/leafs for which writeback hasn't finished yet
887 * (and without errors).
888 * We cleanup any entries left in the io tree when committing
889 * the transaction (through clear_btree_io_tree()).
890 */
891 if (err == -ENOMEM) {
892 err = 0;
893 wait_writeback = true;
894 }
895 if (!err)
896 err = filemap_fdatawrite_range(mapping, start, end);
897 if (err)
898 werr = err;
899 else if (wait_writeback)
900 werr = filemap_fdatawait_range(mapping, start, end);
901 free_extent_state(cached_state);
902 cached_state = NULL;
903 cond_resched();
904 start = end + 1;
905 }
906 return werr;
907 }
908
909 /*
910 * when btree blocks are allocated, they have some corresponding bits set for
911 * them in one of two extent_io trees. This is used to make sure all of
912 * those extents are on disk for transaction or log commit. We wait
913 * on all the pages and clear them from the dirty pages state tree
914 */
915 int btrfs_wait_marked_extents(struct btrfs_root *root,
916 struct extent_io_tree *dirty_pages, int mark)
917 {
918 int err = 0;
919 int werr = 0;
920 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
921 struct extent_state *cached_state = NULL;
922 u64 start = 0;
923 u64 end;
924 struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
925 bool errors = false;
926
927 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
928 EXTENT_NEED_WAIT, &cached_state)) {
929 /*
930 * Ignore -ENOMEM errors returned by clear_extent_bit().
931 * When committing the transaction, we'll remove any entries
932 * left in the io tree. For a log commit, we don't remove them
933 * after committing the log because the tree can be accessed
934 * concurrently - we do it only at transaction commit time when
935 * it's safe to do it (through clear_btree_io_tree()).
936 */
937 err = clear_extent_bit(dirty_pages, start, end,
938 EXTENT_NEED_WAIT,
939 0, 0, &cached_state, GFP_NOFS);
940 if (err == -ENOMEM)
941 err = 0;
942 if (!err)
943 err = filemap_fdatawait_range(mapping, start, end);
944 if (err)
945 werr = err;
946 free_extent_state(cached_state);
947 cached_state = NULL;
948 cond_resched();
949 start = end + 1;
950 }
951 if (err)
952 werr = err;
953
954 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
955 if ((mark & EXTENT_DIRTY) &&
956 test_and_clear_bit(BTRFS_INODE_BTREE_LOG1_ERR,
957 &btree_ino->runtime_flags))
958 errors = true;
959
960 if ((mark & EXTENT_NEW) &&
961 test_and_clear_bit(BTRFS_INODE_BTREE_LOG2_ERR,
962 &btree_ino->runtime_flags))
963 errors = true;
964 } else {
965 if (test_and_clear_bit(BTRFS_INODE_BTREE_ERR,
966 &btree_ino->runtime_flags))
967 errors = true;
968 }
969
970 if (errors && !werr)
971 werr = -EIO;
972
973 return werr;
974 }
975
976 /*
977 * when btree blocks are allocated, they have some corresponding bits set for
978 * them in one of two extent_io trees. This is used to make sure all of
979 * those extents are on disk for transaction or log commit
980 */
981 static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
982 struct extent_io_tree *dirty_pages, int mark)
983 {
984 int ret;
985 int ret2;
986 struct blk_plug plug;
987
988 blk_start_plug(&plug);
989 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
990 blk_finish_plug(&plug);
991 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
992
993 if (ret)
994 return ret;
995 if (ret2)
996 return ret2;
997 return 0;
998 }
999
1000 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
1001 struct btrfs_root *root)
1002 {
1003 int ret;
1004
1005 ret = btrfs_write_and_wait_marked_extents(root,
1006 &trans->transaction->dirty_pages,
1007 EXTENT_DIRTY);
1008 clear_btree_io_tree(&trans->transaction->dirty_pages);
1009
1010 return ret;
1011 }
1012
1013 /*
1014 * this is used to update the root pointer in the tree of tree roots.
1015 *
1016 * But, in the case of the extent allocation tree, updating the root
1017 * pointer may allocate blocks which may change the root of the extent
1018 * allocation tree.
1019 *
1020 * So, this loops and repeats and makes sure the cowonly root didn't
1021 * change while the root pointer was being updated in the metadata.
1022 */
1023 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1024 struct btrfs_root *root)
1025 {
1026 int ret;
1027 u64 old_root_bytenr;
1028 u64 old_root_used;
1029 struct btrfs_root *tree_root = root->fs_info->tree_root;
1030
1031 old_root_used = btrfs_root_used(&root->root_item);
1032
1033 while (1) {
1034 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1035 if (old_root_bytenr == root->node->start &&
1036 old_root_used == btrfs_root_used(&root->root_item))
1037 break;
1038
1039 btrfs_set_root_node(&root->root_item, root->node);
1040 ret = btrfs_update_root(trans, tree_root,
1041 &root->root_key,
1042 &root->root_item);
1043 if (ret)
1044 return ret;
1045
1046 old_root_used = btrfs_root_used(&root->root_item);
1047 }
1048
1049 return 0;
1050 }
1051
1052 /*
1053 * update all the cowonly tree roots on disk
1054 *
1055 * The error handling in this function may not be obvious. Any of the
1056 * failures will cause the file system to go offline. We still need
1057 * to clean up the delayed refs.
1058 */
1059 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
1060 struct btrfs_root *root)
1061 {
1062 struct btrfs_fs_info *fs_info = root->fs_info;
1063 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1064 struct list_head *io_bgs = &trans->transaction->io_bgs;
1065 struct list_head *next;
1066 struct extent_buffer *eb;
1067 int ret;
1068
1069 eb = btrfs_lock_root_node(fs_info->tree_root);
1070 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1071 0, &eb);
1072 btrfs_tree_unlock(eb);
1073 free_extent_buffer(eb);
1074
1075 if (ret)
1076 return ret;
1077
1078 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1079 if (ret)
1080 return ret;
1081
1082 ret = btrfs_run_dev_stats(trans, root->fs_info);
1083 if (ret)
1084 return ret;
1085 ret = btrfs_run_dev_replace(trans, root->fs_info);
1086 if (ret)
1087 return ret;
1088 ret = btrfs_run_qgroups(trans, root->fs_info);
1089 if (ret)
1090 return ret;
1091
1092 ret = btrfs_setup_space_cache(trans, root);
1093 if (ret)
1094 return ret;
1095
1096 /* run_qgroups might have added some more refs */
1097 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1098 if (ret)
1099 return ret;
1100 again:
1101 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1102 next = fs_info->dirty_cowonly_roots.next;
1103 list_del_init(next);
1104 root = list_entry(next, struct btrfs_root, dirty_list);
1105 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1106
1107 if (root != fs_info->extent_root)
1108 list_add_tail(&root->dirty_list,
1109 &trans->transaction->switch_commits);
1110 ret = update_cowonly_root(trans, root);
1111 if (ret)
1112 return ret;
1113 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1114 if (ret)
1115 return ret;
1116 }
1117
1118 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1119 ret = btrfs_write_dirty_block_groups(trans, root);
1120 if (ret)
1121 return ret;
1122 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1123 if (ret)
1124 return ret;
1125 }
1126
1127 if (!list_empty(&fs_info->dirty_cowonly_roots))
1128 goto again;
1129
1130 list_add_tail(&fs_info->extent_root->dirty_list,
1131 &trans->transaction->switch_commits);
1132 btrfs_after_dev_replace_commit(fs_info);
1133
1134 return 0;
1135 }
1136
1137 /*
1138 * dead roots are old snapshots that need to be deleted. This allocates
1139 * a dirty root struct and adds it into the list of dead roots that need to
1140 * be deleted
1141 */
1142 void btrfs_add_dead_root(struct btrfs_root *root)
1143 {
1144 spin_lock(&root->fs_info->trans_lock);
1145 if (list_empty(&root->root_list))
1146 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
1147 spin_unlock(&root->fs_info->trans_lock);
1148 }
1149
1150 /*
1151 * update all the cowonly tree roots on disk
1152 */
1153 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1154 struct btrfs_root *root)
1155 {
1156 struct btrfs_root *gang[8];
1157 struct btrfs_fs_info *fs_info = root->fs_info;
1158 int i;
1159 int ret;
1160 int err = 0;
1161
1162 spin_lock(&fs_info->fs_roots_radix_lock);
1163 while (1) {
1164 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1165 (void **)gang, 0,
1166 ARRAY_SIZE(gang),
1167 BTRFS_ROOT_TRANS_TAG);
1168 if (ret == 0)
1169 break;
1170 for (i = 0; i < ret; i++) {
1171 root = gang[i];
1172 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1173 (unsigned long)root->root_key.objectid,
1174 BTRFS_ROOT_TRANS_TAG);
1175 spin_unlock(&fs_info->fs_roots_radix_lock);
1176
1177 btrfs_free_log(trans, root);
1178 btrfs_update_reloc_root(trans, root);
1179 btrfs_orphan_commit_root(trans, root);
1180
1181 btrfs_save_ino_cache(root, trans);
1182
1183 /* see comments in should_cow_block() */
1184 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1185 smp_mb__after_atomic();
1186
1187 if (root->commit_root != root->node) {
1188 list_add_tail(&root->dirty_list,
1189 &trans->transaction->switch_commits);
1190 btrfs_set_root_node(&root->root_item,
1191 root->node);
1192 }
1193
1194 err = btrfs_update_root(trans, fs_info->tree_root,
1195 &root->root_key,
1196 &root->root_item);
1197 spin_lock(&fs_info->fs_roots_radix_lock);
1198 if (err)
1199 break;
1200 }
1201 }
1202 spin_unlock(&fs_info->fs_roots_radix_lock);
1203 return err;
1204 }
1205
1206 /*
1207 * defrag a given btree.
1208 * Every leaf in the btree is read and defragged.
1209 */
1210 int btrfs_defrag_root(struct btrfs_root *root)
1211 {
1212 struct btrfs_fs_info *info = root->fs_info;
1213 struct btrfs_trans_handle *trans;
1214 int ret;
1215
1216 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1217 return 0;
1218
1219 while (1) {
1220 trans = btrfs_start_transaction(root, 0);
1221 if (IS_ERR(trans))
1222 return PTR_ERR(trans);
1223
1224 ret = btrfs_defrag_leaves(trans, root);
1225
1226 btrfs_end_transaction(trans, root);
1227 btrfs_btree_balance_dirty(info->tree_root);
1228 cond_resched();
1229
1230 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
1231 break;
1232
1233 if (btrfs_defrag_cancelled(root->fs_info)) {
1234 pr_debug("BTRFS: defrag_root cancelled\n");
1235 ret = -EAGAIN;
1236 break;
1237 }
1238 }
1239 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1240 return ret;
1241 }
1242
1243 /*
1244 * new snapshots need to be created at a very specific time in the
1245 * transaction commit. This does the actual creation.
1246 *
1247 * Note:
1248 * If the error which may affect the commitment of the current transaction
1249 * happens, we should return the error number. If the error which just affect
1250 * the creation of the pending snapshots, just return 0.
1251 */
1252 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1253 struct btrfs_fs_info *fs_info,
1254 struct btrfs_pending_snapshot *pending)
1255 {
1256 struct btrfs_key key;
1257 struct btrfs_root_item *new_root_item;
1258 struct btrfs_root *tree_root = fs_info->tree_root;
1259 struct btrfs_root *root = pending->root;
1260 struct btrfs_root *parent_root;
1261 struct btrfs_block_rsv *rsv;
1262 struct inode *parent_inode;
1263 struct btrfs_path *path;
1264 struct btrfs_dir_item *dir_item;
1265 struct dentry *dentry;
1266 struct extent_buffer *tmp;
1267 struct extent_buffer *old;
1268 struct timespec cur_time = CURRENT_TIME;
1269 int ret = 0;
1270 u64 to_reserve = 0;
1271 u64 index = 0;
1272 u64 objectid;
1273 u64 root_flags;
1274 uuid_le new_uuid;
1275
1276 path = btrfs_alloc_path();
1277 if (!path) {
1278 pending->error = -ENOMEM;
1279 return 0;
1280 }
1281
1282 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
1283 if (!new_root_item) {
1284 pending->error = -ENOMEM;
1285 goto root_item_alloc_fail;
1286 }
1287
1288 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1289 if (pending->error)
1290 goto no_free_objectid;
1291
1292 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
1293
1294 if (to_reserve > 0) {
1295 pending->error = btrfs_block_rsv_add(root,
1296 &pending->block_rsv,
1297 to_reserve,
1298 BTRFS_RESERVE_NO_FLUSH);
1299 if (pending->error)
1300 goto no_free_objectid;
1301 }
1302
1303 key.objectid = objectid;
1304 key.offset = (u64)-1;
1305 key.type = BTRFS_ROOT_ITEM_KEY;
1306
1307 rsv = trans->block_rsv;
1308 trans->block_rsv = &pending->block_rsv;
1309 trans->bytes_reserved = trans->block_rsv->reserved;
1310
1311 dentry = pending->dentry;
1312 parent_inode = pending->dir;
1313 parent_root = BTRFS_I(parent_inode)->root;
1314 record_root_in_trans(trans, parent_root);
1315
1316 /*
1317 * insert the directory item
1318 */
1319 ret = btrfs_set_inode_index(parent_inode, &index);
1320 BUG_ON(ret); /* -ENOMEM */
1321
1322 /* check if there is a file/dir which has the same name. */
1323 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1324 btrfs_ino(parent_inode),
1325 dentry->d_name.name,
1326 dentry->d_name.len, 0);
1327 if (dir_item != NULL && !IS_ERR(dir_item)) {
1328 pending->error = -EEXIST;
1329 goto dir_item_existed;
1330 } else if (IS_ERR(dir_item)) {
1331 ret = PTR_ERR(dir_item);
1332 btrfs_abort_transaction(trans, root, ret);
1333 goto fail;
1334 }
1335 btrfs_release_path(path);
1336
1337 /*
1338 * pull in the delayed directory update
1339 * and the delayed inode item
1340 * otherwise we corrupt the FS during
1341 * snapshot
1342 */
1343 ret = btrfs_run_delayed_items(trans, root);
1344 if (ret) { /* Transaction aborted */
1345 btrfs_abort_transaction(trans, root, ret);
1346 goto fail;
1347 }
1348
1349 record_root_in_trans(trans, root);
1350 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1351 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1352 btrfs_check_and_init_root_item(new_root_item);
1353
1354 root_flags = btrfs_root_flags(new_root_item);
1355 if (pending->readonly)
1356 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1357 else
1358 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1359 btrfs_set_root_flags(new_root_item, root_flags);
1360
1361 btrfs_set_root_generation_v2(new_root_item,
1362 trans->transid);
1363 uuid_le_gen(&new_uuid);
1364 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1365 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1366 BTRFS_UUID_SIZE);
1367 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1368 memset(new_root_item->received_uuid, 0,
1369 sizeof(new_root_item->received_uuid));
1370 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1371 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1372 btrfs_set_root_stransid(new_root_item, 0);
1373 btrfs_set_root_rtransid(new_root_item, 0);
1374 }
1375 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1376 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1377 btrfs_set_root_otransid(new_root_item, trans->transid);
1378
1379 old = btrfs_lock_root_node(root);
1380 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1381 if (ret) {
1382 btrfs_tree_unlock(old);
1383 free_extent_buffer(old);
1384 btrfs_abort_transaction(trans, root, ret);
1385 goto fail;
1386 }
1387
1388 btrfs_set_lock_blocking(old);
1389
1390 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1391 /* clean up in any case */
1392 btrfs_tree_unlock(old);
1393 free_extent_buffer(old);
1394 if (ret) {
1395 btrfs_abort_transaction(trans, root, ret);
1396 goto fail;
1397 }
1398
1399 /*
1400 * We need to flush delayed refs in order to make sure all of our quota
1401 * operations have been done before we call btrfs_qgroup_inherit.
1402 */
1403 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1404 if (ret) {
1405 btrfs_abort_transaction(trans, root, ret);
1406 goto fail;
1407 }
1408
1409 ret = btrfs_qgroup_inherit(trans, fs_info,
1410 root->root_key.objectid,
1411 objectid, pending->inherit);
1412 if (ret) {
1413 btrfs_abort_transaction(trans, root, ret);
1414 goto fail;
1415 }
1416
1417 /* see comments in should_cow_block() */
1418 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1419 smp_wmb();
1420
1421 btrfs_set_root_node(new_root_item, tmp);
1422 /* record when the snapshot was created in key.offset */
1423 key.offset = trans->transid;
1424 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1425 btrfs_tree_unlock(tmp);
1426 free_extent_buffer(tmp);
1427 if (ret) {
1428 btrfs_abort_transaction(trans, root, ret);
1429 goto fail;
1430 }
1431
1432 /*
1433 * insert root back/forward references
1434 */
1435 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1436 parent_root->root_key.objectid,
1437 btrfs_ino(parent_inode), index,
1438 dentry->d_name.name, dentry->d_name.len);
1439 if (ret) {
1440 btrfs_abort_transaction(trans, root, ret);
1441 goto fail;
1442 }
1443
1444 key.offset = (u64)-1;
1445 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1446 if (IS_ERR(pending->snap)) {
1447 ret = PTR_ERR(pending->snap);
1448 btrfs_abort_transaction(trans, root, ret);
1449 goto fail;
1450 }
1451
1452 ret = btrfs_reloc_post_snapshot(trans, pending);
1453 if (ret) {
1454 btrfs_abort_transaction(trans, root, ret);
1455 goto fail;
1456 }
1457
1458 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1459 if (ret) {
1460 btrfs_abort_transaction(trans, root, ret);
1461 goto fail;
1462 }
1463
1464 ret = btrfs_insert_dir_item(trans, parent_root,
1465 dentry->d_name.name, dentry->d_name.len,
1466 parent_inode, &key,
1467 BTRFS_FT_DIR, index);
1468 /* We have check then name at the beginning, so it is impossible. */
1469 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1470 if (ret) {
1471 btrfs_abort_transaction(trans, root, ret);
1472 goto fail;
1473 }
1474
1475 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1476 dentry->d_name.len * 2);
1477 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1478 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
1479 if (ret) {
1480 btrfs_abort_transaction(trans, root, ret);
1481 goto fail;
1482 }
1483 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b,
1484 BTRFS_UUID_KEY_SUBVOL, objectid);
1485 if (ret) {
1486 btrfs_abort_transaction(trans, root, ret);
1487 goto fail;
1488 }
1489 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1490 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
1491 new_root_item->received_uuid,
1492 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1493 objectid);
1494 if (ret && ret != -EEXIST) {
1495 btrfs_abort_transaction(trans, root, ret);
1496 goto fail;
1497 }
1498 }
1499 fail:
1500 pending->error = ret;
1501 dir_item_existed:
1502 trans->block_rsv = rsv;
1503 trans->bytes_reserved = 0;
1504 no_free_objectid:
1505 kfree(new_root_item);
1506 root_item_alloc_fail:
1507 btrfs_free_path(path);
1508 return ret;
1509 }
1510
1511 /*
1512 * create all the snapshots we've scheduled for creation
1513 */
1514 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1515 struct btrfs_fs_info *fs_info)
1516 {
1517 struct btrfs_pending_snapshot *pending, *next;
1518 struct list_head *head = &trans->transaction->pending_snapshots;
1519 int ret = 0;
1520
1521 list_for_each_entry_safe(pending, next, head, list) {
1522 list_del(&pending->list);
1523 ret = create_pending_snapshot(trans, fs_info, pending);
1524 if (ret)
1525 break;
1526 }
1527 return ret;
1528 }
1529
1530 static void update_super_roots(struct btrfs_root *root)
1531 {
1532 struct btrfs_root_item *root_item;
1533 struct btrfs_super_block *super;
1534
1535 super = root->fs_info->super_copy;
1536
1537 root_item = &root->fs_info->chunk_root->root_item;
1538 super->chunk_root = root_item->bytenr;
1539 super->chunk_root_generation = root_item->generation;
1540 super->chunk_root_level = root_item->level;
1541
1542 root_item = &root->fs_info->tree_root->root_item;
1543 super->root = root_item->bytenr;
1544 super->generation = root_item->generation;
1545 super->root_level = root_item->level;
1546 if (btrfs_test_opt(root, SPACE_CACHE))
1547 super->cache_generation = root_item->generation;
1548 if (root->fs_info->update_uuid_tree_gen)
1549 super->uuid_tree_generation = root_item->generation;
1550 }
1551
1552 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1553 {
1554 struct btrfs_transaction *trans;
1555 int ret = 0;
1556
1557 spin_lock(&info->trans_lock);
1558 trans = info->running_transaction;
1559 if (trans)
1560 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1561 spin_unlock(&info->trans_lock);
1562 return ret;
1563 }
1564
1565 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1566 {
1567 struct btrfs_transaction *trans;
1568 int ret = 0;
1569
1570 spin_lock(&info->trans_lock);
1571 trans = info->running_transaction;
1572 if (trans)
1573 ret = is_transaction_blocked(trans);
1574 spin_unlock(&info->trans_lock);
1575 return ret;
1576 }
1577
1578 /*
1579 * wait for the current transaction commit to start and block subsequent
1580 * transaction joins
1581 */
1582 static void wait_current_trans_commit_start(struct btrfs_root *root,
1583 struct btrfs_transaction *trans)
1584 {
1585 wait_event(root->fs_info->transaction_blocked_wait,
1586 trans->state >= TRANS_STATE_COMMIT_START ||
1587 trans->aborted);
1588 }
1589
1590 /*
1591 * wait for the current transaction to start and then become unblocked.
1592 * caller holds ref.
1593 */
1594 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1595 struct btrfs_transaction *trans)
1596 {
1597 wait_event(root->fs_info->transaction_wait,
1598 trans->state >= TRANS_STATE_UNBLOCKED ||
1599 trans->aborted);
1600 }
1601
1602 /*
1603 * commit transactions asynchronously. once btrfs_commit_transaction_async
1604 * returns, any subsequent transaction will not be allowed to join.
1605 */
1606 struct btrfs_async_commit {
1607 struct btrfs_trans_handle *newtrans;
1608 struct btrfs_root *root;
1609 struct work_struct work;
1610 };
1611
1612 static void do_async_commit(struct work_struct *work)
1613 {
1614 struct btrfs_async_commit *ac =
1615 container_of(work, struct btrfs_async_commit, work);
1616
1617 /*
1618 * We've got freeze protection passed with the transaction.
1619 * Tell lockdep about it.
1620 */
1621 if (ac->newtrans->type & __TRANS_FREEZABLE)
1622 rwsem_acquire_read(
1623 &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1624 0, 1, _THIS_IP_);
1625
1626 current->journal_info = ac->newtrans;
1627
1628 btrfs_commit_transaction(ac->newtrans, ac->root);
1629 kfree(ac);
1630 }
1631
1632 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1633 struct btrfs_root *root,
1634 int wait_for_unblock)
1635 {
1636 struct btrfs_async_commit *ac;
1637 struct btrfs_transaction *cur_trans;
1638
1639 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1640 if (!ac)
1641 return -ENOMEM;
1642
1643 INIT_WORK(&ac->work, do_async_commit);
1644 ac->root = root;
1645 ac->newtrans = btrfs_join_transaction(root);
1646 if (IS_ERR(ac->newtrans)) {
1647 int err = PTR_ERR(ac->newtrans);
1648 kfree(ac);
1649 return err;
1650 }
1651
1652 /* take transaction reference */
1653 cur_trans = trans->transaction;
1654 atomic_inc(&cur_trans->use_count);
1655
1656 btrfs_end_transaction(trans, root);
1657
1658 /*
1659 * Tell lockdep we've released the freeze rwsem, since the
1660 * async commit thread will be the one to unlock it.
1661 */
1662 if (ac->newtrans->type & __TRANS_FREEZABLE)
1663 rwsem_release(
1664 &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1665 1, _THIS_IP_);
1666
1667 schedule_work(&ac->work);
1668
1669 /* wait for transaction to start and unblock */
1670 if (wait_for_unblock)
1671 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1672 else
1673 wait_current_trans_commit_start(root, cur_trans);
1674
1675 if (current->journal_info == trans)
1676 current->journal_info = NULL;
1677
1678 btrfs_put_transaction(cur_trans);
1679 return 0;
1680 }
1681
1682
1683 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1684 struct btrfs_root *root, int err)
1685 {
1686 struct btrfs_transaction *cur_trans = trans->transaction;
1687 DEFINE_WAIT(wait);
1688
1689 WARN_ON(trans->use_count > 1);
1690
1691 btrfs_abort_transaction(trans, root, err);
1692
1693 spin_lock(&root->fs_info->trans_lock);
1694
1695 /*
1696 * If the transaction is removed from the list, it means this
1697 * transaction has been committed successfully, so it is impossible
1698 * to call the cleanup function.
1699 */
1700 BUG_ON(list_empty(&cur_trans->list));
1701
1702 list_del_init(&cur_trans->list);
1703 if (cur_trans == root->fs_info->running_transaction) {
1704 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1705 spin_unlock(&root->fs_info->trans_lock);
1706 wait_event(cur_trans->writer_wait,
1707 atomic_read(&cur_trans->num_writers) == 1);
1708
1709 spin_lock(&root->fs_info->trans_lock);
1710 }
1711 spin_unlock(&root->fs_info->trans_lock);
1712
1713 btrfs_cleanup_one_transaction(trans->transaction, root);
1714
1715 spin_lock(&root->fs_info->trans_lock);
1716 if (cur_trans == root->fs_info->running_transaction)
1717 root->fs_info->running_transaction = NULL;
1718 spin_unlock(&root->fs_info->trans_lock);
1719
1720 if (trans->type & __TRANS_FREEZABLE)
1721 sb_end_intwrite(root->fs_info->sb);
1722 btrfs_put_transaction(cur_trans);
1723 btrfs_put_transaction(cur_trans);
1724
1725 trace_btrfs_transaction_commit(root);
1726
1727 if (current->journal_info == trans)
1728 current->journal_info = NULL;
1729 btrfs_scrub_cancel(root->fs_info);
1730
1731 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1732 }
1733
1734 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1735 {
1736 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1737 return btrfs_start_delalloc_roots(fs_info, 1, -1);
1738 return 0;
1739 }
1740
1741 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1742 {
1743 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1744 btrfs_wait_ordered_roots(fs_info, -1);
1745 }
1746
1747 static inline void
1748 btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans,
1749 struct btrfs_fs_info *fs_info)
1750 {
1751 struct btrfs_ordered_extent *ordered;
1752
1753 spin_lock(&fs_info->trans_lock);
1754 while (!list_empty(&cur_trans->pending_ordered)) {
1755 ordered = list_first_entry(&cur_trans->pending_ordered,
1756 struct btrfs_ordered_extent,
1757 trans_list);
1758 list_del_init(&ordered->trans_list);
1759 spin_unlock(&fs_info->trans_lock);
1760
1761 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_COMPLETE,
1762 &ordered->flags));
1763 btrfs_put_ordered_extent(ordered);
1764 spin_lock(&fs_info->trans_lock);
1765 }
1766 spin_unlock(&fs_info->trans_lock);
1767 }
1768
1769 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1770 struct btrfs_root *root)
1771 {
1772 struct btrfs_transaction *cur_trans = trans->transaction;
1773 struct btrfs_transaction *prev_trans = NULL;
1774 struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
1775 int ret;
1776
1777 /* Stop the commit early if ->aborted is set */
1778 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1779 ret = cur_trans->aborted;
1780 btrfs_end_transaction(trans, root);
1781 return ret;
1782 }
1783
1784 /* make a pass through all the delayed refs we have so far
1785 * any runnings procs may add more while we are here
1786 */
1787 ret = btrfs_run_delayed_refs(trans, root, 0);
1788 if (ret) {
1789 btrfs_end_transaction(trans, root);
1790 return ret;
1791 }
1792
1793 btrfs_trans_release_metadata(trans, root);
1794 trans->block_rsv = NULL;
1795 if (trans->qgroup_reserved) {
1796 btrfs_qgroup_free(root, trans->qgroup_reserved);
1797 trans->qgroup_reserved = 0;
1798 }
1799
1800 cur_trans = trans->transaction;
1801
1802 /*
1803 * set the flushing flag so procs in this transaction have to
1804 * start sending their work down.
1805 */
1806 cur_trans->delayed_refs.flushing = 1;
1807 smp_wmb();
1808
1809 if (!list_empty(&trans->new_bgs))
1810 btrfs_create_pending_block_groups(trans, root);
1811
1812 ret = btrfs_run_delayed_refs(trans, root, 0);
1813 if (ret) {
1814 btrfs_end_transaction(trans, root);
1815 return ret;
1816 }
1817
1818 if (!cur_trans->dirty_bg_run) {
1819 int run_it = 0;
1820
1821 /* this mutex is also taken before trying to set
1822 * block groups readonly. We need to make sure
1823 * that nobody has set a block group readonly
1824 * after a extents from that block group have been
1825 * allocated for cache files. btrfs_set_block_group_ro
1826 * will wait for the transaction to commit if it
1827 * finds dirty_bg_run = 1
1828 *
1829 * The dirty_bg_run flag is also used to make sure only
1830 * one process starts all the block group IO. It wouldn't
1831 * hurt to have more than one go through, but there's no
1832 * real advantage to it either.
1833 */
1834 mutex_lock(&root->fs_info->ro_block_group_mutex);
1835 if (!cur_trans->dirty_bg_run) {
1836 run_it = 1;
1837 cur_trans->dirty_bg_run = 1;
1838 }
1839 mutex_unlock(&root->fs_info->ro_block_group_mutex);
1840
1841 if (run_it)
1842 ret = btrfs_start_dirty_block_groups(trans, root);
1843 }
1844 if (ret) {
1845 btrfs_end_transaction(trans, root);
1846 return ret;
1847 }
1848
1849 spin_lock(&root->fs_info->trans_lock);
1850 list_splice_init(&trans->ordered, &cur_trans->pending_ordered);
1851 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1852 spin_unlock(&root->fs_info->trans_lock);
1853 atomic_inc(&cur_trans->use_count);
1854 ret = btrfs_end_transaction(trans, root);
1855
1856 wait_for_commit(root, cur_trans);
1857
1858 if (unlikely(cur_trans->aborted))
1859 ret = cur_trans->aborted;
1860
1861 btrfs_put_transaction(cur_trans);
1862
1863 return ret;
1864 }
1865
1866 cur_trans->state = TRANS_STATE_COMMIT_START;
1867 wake_up(&root->fs_info->transaction_blocked_wait);
1868
1869 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1870 prev_trans = list_entry(cur_trans->list.prev,
1871 struct btrfs_transaction, list);
1872 if (prev_trans->state != TRANS_STATE_COMPLETED) {
1873 atomic_inc(&prev_trans->use_count);
1874 spin_unlock(&root->fs_info->trans_lock);
1875
1876 wait_for_commit(root, prev_trans);
1877 ret = prev_trans->aborted;
1878
1879 btrfs_put_transaction(prev_trans);
1880 if (ret)
1881 goto cleanup_transaction;
1882 } else {
1883 spin_unlock(&root->fs_info->trans_lock);
1884 }
1885 } else {
1886 spin_unlock(&root->fs_info->trans_lock);
1887 }
1888
1889 extwriter_counter_dec(cur_trans, trans->type);
1890
1891 ret = btrfs_start_delalloc_flush(root->fs_info);
1892 if (ret)
1893 goto cleanup_transaction;
1894
1895 ret = btrfs_run_delayed_items(trans, root);
1896 if (ret)
1897 goto cleanup_transaction;
1898
1899 wait_event(cur_trans->writer_wait,
1900 extwriter_counter_read(cur_trans) == 0);
1901
1902 /* some pending stuffs might be added after the previous flush. */
1903 ret = btrfs_run_delayed_items(trans, root);
1904 if (ret)
1905 goto cleanup_transaction;
1906
1907 btrfs_wait_delalloc_flush(root->fs_info);
1908
1909 btrfs_wait_pending_ordered(cur_trans, root->fs_info);
1910
1911 btrfs_scrub_pause(root);
1912 /*
1913 * Ok now we need to make sure to block out any other joins while we
1914 * commit the transaction. We could have started a join before setting
1915 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
1916 */
1917 spin_lock(&root->fs_info->trans_lock);
1918 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1919 spin_unlock(&root->fs_info->trans_lock);
1920 wait_event(cur_trans->writer_wait,
1921 atomic_read(&cur_trans->num_writers) == 1);
1922
1923 /* ->aborted might be set after the previous check, so check it */
1924 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1925 ret = cur_trans->aborted;
1926 goto scrub_continue;
1927 }
1928 /*
1929 * the reloc mutex makes sure that we stop
1930 * the balancing code from coming in and moving
1931 * extents around in the middle of the commit
1932 */
1933 mutex_lock(&root->fs_info->reloc_mutex);
1934
1935 /*
1936 * We needn't worry about the delayed items because we will
1937 * deal with them in create_pending_snapshot(), which is the
1938 * core function of the snapshot creation.
1939 */
1940 ret = create_pending_snapshots(trans, root->fs_info);
1941 if (ret) {
1942 mutex_unlock(&root->fs_info->reloc_mutex);
1943 goto scrub_continue;
1944 }
1945
1946 /*
1947 * We insert the dir indexes of the snapshots and update the inode
1948 * of the snapshots' parents after the snapshot creation, so there
1949 * are some delayed items which are not dealt with. Now deal with
1950 * them.
1951 *
1952 * We needn't worry that this operation will corrupt the snapshots,
1953 * because all the tree which are snapshoted will be forced to COW
1954 * the nodes and leaves.
1955 */
1956 ret = btrfs_run_delayed_items(trans, root);
1957 if (ret) {
1958 mutex_unlock(&root->fs_info->reloc_mutex);
1959 goto scrub_continue;
1960 }
1961
1962 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1963 if (ret) {
1964 mutex_unlock(&root->fs_info->reloc_mutex);
1965 goto scrub_continue;
1966 }
1967
1968 /*
1969 * make sure none of the code above managed to slip in a
1970 * delayed item
1971 */
1972 btrfs_assert_delayed_root_empty(root);
1973
1974 WARN_ON(cur_trans != trans->transaction);
1975
1976 /* btrfs_commit_tree_roots is responsible for getting the
1977 * various roots consistent with each other. Every pointer
1978 * in the tree of tree roots has to point to the most up to date
1979 * root for every subvolume and other tree. So, we have to keep
1980 * the tree logging code from jumping in and changing any
1981 * of the trees.
1982 *
1983 * At this point in the commit, there can't be any tree-log
1984 * writers, but a little lower down we drop the trans mutex
1985 * and let new people in. By holding the tree_log_mutex
1986 * from now until after the super is written, we avoid races
1987 * with the tree-log code.
1988 */
1989 mutex_lock(&root->fs_info->tree_log_mutex);
1990
1991 ret = commit_fs_roots(trans, root);
1992 if (ret) {
1993 mutex_unlock(&root->fs_info->tree_log_mutex);
1994 mutex_unlock(&root->fs_info->reloc_mutex);
1995 goto scrub_continue;
1996 }
1997
1998 /*
1999 * Since the transaction is done, we can apply the pending changes
2000 * before the next transaction.
2001 */
2002 btrfs_apply_pending_changes(root->fs_info);
2003
2004 /* commit_fs_roots gets rid of all the tree log roots, it is now
2005 * safe to free the root of tree log roots
2006 */
2007 btrfs_free_log_root_tree(trans, root->fs_info);
2008
2009 ret = commit_cowonly_roots(trans, root);
2010 if (ret) {
2011 mutex_unlock(&root->fs_info->tree_log_mutex);
2012 mutex_unlock(&root->fs_info->reloc_mutex);
2013 goto scrub_continue;
2014 }
2015
2016 /*
2017 * The tasks which save the space cache and inode cache may also
2018 * update ->aborted, check it.
2019 */
2020 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
2021 ret = cur_trans->aborted;
2022 mutex_unlock(&root->fs_info->tree_log_mutex);
2023 mutex_unlock(&root->fs_info->reloc_mutex);
2024 goto scrub_continue;
2025 }
2026
2027 btrfs_prepare_extent_commit(trans, root);
2028
2029 cur_trans = root->fs_info->running_transaction;
2030
2031 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
2032 root->fs_info->tree_root->node);
2033 list_add_tail(&root->fs_info->tree_root->dirty_list,
2034 &cur_trans->switch_commits);
2035
2036 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
2037 root->fs_info->chunk_root->node);
2038 list_add_tail(&root->fs_info->chunk_root->dirty_list,
2039 &cur_trans->switch_commits);
2040
2041 switch_commit_roots(cur_trans, root->fs_info);
2042
2043 assert_qgroups_uptodate(trans);
2044 ASSERT(list_empty(&cur_trans->dirty_bgs));
2045 ASSERT(list_empty(&cur_trans->io_bgs));
2046 update_super_roots(root);
2047
2048 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
2049 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
2050 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
2051 sizeof(*root->fs_info->super_copy));
2052
2053 btrfs_update_commit_device_size(root->fs_info);
2054 btrfs_update_commit_device_bytes_used(root, cur_trans);
2055
2056 clear_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags);
2057 clear_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags);
2058
2059 spin_lock(&root->fs_info->trans_lock);
2060 cur_trans->state = TRANS_STATE_UNBLOCKED;
2061 root->fs_info->running_transaction = NULL;
2062 spin_unlock(&root->fs_info->trans_lock);
2063 mutex_unlock(&root->fs_info->reloc_mutex);
2064
2065 wake_up(&root->fs_info->transaction_wait);
2066
2067 ret = btrfs_write_and_wait_transaction(trans, root);
2068 if (ret) {
2069 btrfs_error(root->fs_info, ret,
2070 "Error while writing out transaction");
2071 mutex_unlock(&root->fs_info->tree_log_mutex);
2072 goto scrub_continue;
2073 }
2074
2075 ret = write_ctree_super(trans, root, 0);
2076 if (ret) {
2077 mutex_unlock(&root->fs_info->tree_log_mutex);
2078 goto scrub_continue;
2079 }
2080
2081 /*
2082 * the super is written, we can safely allow the tree-loggers
2083 * to go about their business
2084 */
2085 mutex_unlock(&root->fs_info->tree_log_mutex);
2086
2087 btrfs_finish_extent_commit(trans, root);
2088
2089 if (cur_trans->have_free_bgs)
2090 btrfs_clear_space_info_full(root->fs_info);
2091
2092 root->fs_info->last_trans_committed = cur_trans->transid;
2093 /*
2094 * We needn't acquire the lock here because there is no other task
2095 * which can change it.
2096 */
2097 cur_trans->state = TRANS_STATE_COMPLETED;
2098 wake_up(&cur_trans->commit_wait);
2099
2100 spin_lock(&root->fs_info->trans_lock);
2101 list_del_init(&cur_trans->list);
2102 spin_unlock(&root->fs_info->trans_lock);
2103
2104 btrfs_put_transaction(cur_trans);
2105 btrfs_put_transaction(cur_trans);
2106
2107 if (trans->type & __TRANS_FREEZABLE)
2108 sb_end_intwrite(root->fs_info->sb);
2109
2110 trace_btrfs_transaction_commit(root);
2111
2112 btrfs_scrub_continue(root);
2113
2114 if (current->journal_info == trans)
2115 current->journal_info = NULL;
2116
2117 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2118
2119 if (current != root->fs_info->transaction_kthread)
2120 btrfs_run_delayed_iputs(root);
2121
2122 return ret;
2123
2124 scrub_continue:
2125 btrfs_scrub_continue(root);
2126 cleanup_transaction:
2127 btrfs_trans_release_metadata(trans, root);
2128 trans->block_rsv = NULL;
2129 if (trans->qgroup_reserved) {
2130 btrfs_qgroup_free(root, trans->qgroup_reserved);
2131 trans->qgroup_reserved = 0;
2132 }
2133 btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
2134 if (current->journal_info == trans)
2135 current->journal_info = NULL;
2136 cleanup_transaction(trans, root, ret);
2137
2138 return ret;
2139 }
2140
2141 /*
2142 * return < 0 if error
2143 * 0 if there are no more dead_roots at the time of call
2144 * 1 there are more to be processed, call me again
2145 *
2146 * The return value indicates there are certainly more snapshots to delete, but
2147 * if there comes a new one during processing, it may return 0. We don't mind,
2148 * because btrfs_commit_super will poke cleaner thread and it will process it a
2149 * few seconds later.
2150 */
2151 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
2152 {
2153 int ret;
2154 struct btrfs_fs_info *fs_info = root->fs_info;
2155
2156 spin_lock(&fs_info->trans_lock);
2157 if (list_empty(&fs_info->dead_roots)) {
2158 spin_unlock(&fs_info->trans_lock);
2159 return 0;
2160 }
2161 root = list_first_entry(&fs_info->dead_roots,
2162 struct btrfs_root, root_list);
2163 list_del_init(&root->root_list);
2164 spin_unlock(&fs_info->trans_lock);
2165
2166 pr_debug("BTRFS: cleaner removing %llu\n", root->objectid);
2167
2168 btrfs_kill_all_delayed_nodes(root);
2169
2170 if (btrfs_header_backref_rev(root->node) <
2171 BTRFS_MIXED_BACKREF_REV)
2172 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
2173 else
2174 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
2175
2176 return (ret < 0) ? 0 : 1;
2177 }
2178
2179 void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2180 {
2181 unsigned long prev;
2182 unsigned long bit;
2183
2184 prev = xchg(&fs_info->pending_changes, 0);
2185 if (!prev)
2186 return;
2187
2188 bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE;
2189 if (prev & bit)
2190 btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2191 prev &= ~bit;
2192
2193 bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE;
2194 if (prev & bit)
2195 btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2196 prev &= ~bit;
2197
2198 bit = 1 << BTRFS_PENDING_COMMIT;
2199 if (prev & bit)
2200 btrfs_debug(fs_info, "pending commit done");
2201 prev &= ~bit;
2202
2203 if (prev)
2204 btrfs_warn(fs_info,
2205 "unknown pending changes left 0x%lx, ignoring", prev);
2206 }
Linux with added FPC-III support