Merge branch 'misc-4.10' into for-chris-4.10-20161130
[linux-2.6-block.git] / fs / btrfs / transaction.c
CommitLineData
6cbd5570
CM
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
79154b1b 19#include <linux/fs.h>
5a0e3ad6 20#include <linux/slab.h>
34088780 21#include <linux/sched.h>
d3c2fdcf 22#include <linux/writeback.h>
5f39d397 23#include <linux/pagemap.h>
5f2cc086 24#include <linux/blkdev.h>
8ea05e3a 25#include <linux/uuid.h>
79154b1b
CM
26#include "ctree.h"
27#include "disk-io.h"
28#include "transaction.h"
925baedd 29#include "locking.h"
e02119d5 30#include "tree-log.h"
581bb050 31#include "inode-map.h"
733f4fbb 32#include "volumes.h"
8dabb742 33#include "dev-replace.h"
fcebe456 34#include "qgroup.h"
79154b1b 35
0f7d52f4
CM
36#define BTRFS_ROOT_TRANS_TAG 0
37
e8c9f186 38static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
4a9d8bde
MX
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
724e2315 61void btrfs_put_transaction(struct btrfs_transaction *transaction)
79154b1b 62{
13c5a93e
JB
63 WARN_ON(atomic_read(&transaction->use_count) == 0);
64 if (atomic_dec_and_test(&transaction->use_count)) {
a4abeea4 65 BUG_ON(!list_empty(&transaction->list));
c46effa6 66 WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root));
1262133b 67 if (transaction->delayed_refs.pending_csums)
ab8d0fc4
JM
68 btrfs_err(transaction->fs_info,
69 "pending csums is %llu",
70 transaction->delayed_refs.pending_csums);
6df9a95e
JB
71 while (!list_empty(&transaction->pending_chunks)) {
72 struct extent_map *em;
73
74 em = list_first_entry(&transaction->pending_chunks,
75 struct extent_map, list);
76 list_del_init(&em->list);
77 free_extent_map(em);
78 }
7785a663
FM
79 /*
80 * If any block groups are found in ->deleted_bgs then it's
81 * because the transaction was aborted and a commit did not
82 * happen (things failed before writing the new superblock
83 * and calling btrfs_finish_extent_commit()), so we can not
84 * discard the physical locations of the block groups.
85 */
86 while (!list_empty(&transaction->deleted_bgs)) {
87 struct btrfs_block_group_cache *cache;
88
89 cache = list_first_entry(&transaction->deleted_bgs,
90 struct btrfs_block_group_cache,
91 bg_list);
92 list_del_init(&cache->bg_list);
93 btrfs_put_block_group_trimming(cache);
94 btrfs_put_block_group(cache);
95 }
2c90e5d6 96 kmem_cache_free(btrfs_transaction_cachep, transaction);
78fae27e 97 }
79154b1b
CM
98}
99
663dfbb0
FM
100static void clear_btree_io_tree(struct extent_io_tree *tree)
101{
102 spin_lock(&tree->lock);
b666a9cd
DS
103 /*
104 * Do a single barrier for the waitqueue_active check here, the state
105 * of the waitqueue should not change once clear_btree_io_tree is
106 * called.
107 */
108 smp_mb();
663dfbb0
FM
109 while (!RB_EMPTY_ROOT(&tree->state)) {
110 struct rb_node *node;
111 struct extent_state *state;
112
113 node = rb_first(&tree->state);
114 state = rb_entry(node, struct extent_state, rb_node);
115 rb_erase(&state->rb_node, &tree->state);
116 RB_CLEAR_NODE(&state->rb_node);
117 /*
118 * btree io trees aren't supposed to have tasks waiting for
119 * changes in the flags of extent states ever.
120 */
121 ASSERT(!waitqueue_active(&state->wq));
122 free_extent_state(state);
351810c1
DS
123
124 cond_resched_lock(&tree->lock);
663dfbb0
FM
125 }
126 spin_unlock(&tree->lock);
127}
128
9e351cc8
JB
129static noinline void switch_commit_roots(struct btrfs_transaction *trans,
130 struct btrfs_fs_info *fs_info)
817d52f8 131{
9e351cc8
JB
132 struct btrfs_root *root, *tmp;
133
134 down_write(&fs_info->commit_root_sem);
135 list_for_each_entry_safe(root, tmp, &trans->switch_commits,
136 dirty_list) {
137 list_del_init(&root->dirty_list);
138 free_extent_buffer(root->commit_root);
139 root->commit_root = btrfs_root_node(root);
140 if (is_fstree(root->objectid))
141 btrfs_unpin_free_ino(root);
663dfbb0 142 clear_btree_io_tree(&root->dirty_log_pages);
9e351cc8 143 }
2b9dbef2
JB
144
145 /* We can free old roots now. */
146 spin_lock(&trans->dropped_roots_lock);
147 while (!list_empty(&trans->dropped_roots)) {
148 root = list_first_entry(&trans->dropped_roots,
149 struct btrfs_root, root_list);
150 list_del_init(&root->root_list);
151 spin_unlock(&trans->dropped_roots_lock);
152 btrfs_drop_and_free_fs_root(fs_info, root);
153 spin_lock(&trans->dropped_roots_lock);
154 }
155 spin_unlock(&trans->dropped_roots_lock);
9e351cc8 156 up_write(&fs_info->commit_root_sem);
817d52f8
JB
157}
158
0860adfd
MX
159static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
160 unsigned int type)
161{
162 if (type & TRANS_EXTWRITERS)
163 atomic_inc(&trans->num_extwriters);
164}
165
166static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
167 unsigned int type)
168{
169 if (type & TRANS_EXTWRITERS)
170 atomic_dec(&trans->num_extwriters);
171}
172
173static inline void extwriter_counter_init(struct btrfs_transaction *trans,
174 unsigned int type)
175{
176 atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
177}
178
179static inline int extwriter_counter_read(struct btrfs_transaction *trans)
180{
181 return atomic_read(&trans->num_extwriters);
178260b2
MX
182}
183
d352ac68
CM
184/*
185 * either allocate a new transaction or hop into the existing one
186 */
0860adfd 187static noinline int join_transaction(struct btrfs_root *root, unsigned int type)
79154b1b
CM
188{
189 struct btrfs_transaction *cur_trans;
19ae4e81 190 struct btrfs_fs_info *fs_info = root->fs_info;
a4abeea4 191
19ae4e81 192 spin_lock(&fs_info->trans_lock);
d43317dc 193loop:
49b25e05 194 /* The file system has been taken offline. No new transactions. */
87533c47 195 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
19ae4e81 196 spin_unlock(&fs_info->trans_lock);
49b25e05
JM
197 return -EROFS;
198 }
199
19ae4e81 200 cur_trans = fs_info->running_transaction;
a4abeea4 201 if (cur_trans) {
871383be 202 if (cur_trans->aborted) {
19ae4e81 203 spin_unlock(&fs_info->trans_lock);
49b25e05 204 return cur_trans->aborted;
871383be 205 }
4a9d8bde 206 if (btrfs_blocked_trans_types[cur_trans->state] & type) {
178260b2
MX
207 spin_unlock(&fs_info->trans_lock);
208 return -EBUSY;
209 }
a4abeea4 210 atomic_inc(&cur_trans->use_count);
13c5a93e 211 atomic_inc(&cur_trans->num_writers);
0860adfd 212 extwriter_counter_inc(cur_trans, type);
19ae4e81 213 spin_unlock(&fs_info->trans_lock);
a4abeea4 214 return 0;
79154b1b 215 }
19ae4e81 216 spin_unlock(&fs_info->trans_lock);
a4abeea4 217
354aa0fb
MX
218 /*
219 * If we are ATTACH, we just want to catch the current transaction,
220 * and commit it. If there is no transaction, just return ENOENT.
221 */
222 if (type == TRANS_ATTACH)
223 return -ENOENT;
224
4a9d8bde
MX
225 /*
226 * JOIN_NOLOCK only happens during the transaction commit, so
227 * it is impossible that ->running_transaction is NULL
228 */
229 BUG_ON(type == TRANS_JOIN_NOLOCK);
230
a4abeea4
JB
231 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
232 if (!cur_trans)
233 return -ENOMEM;
d43317dc 234
19ae4e81
JS
235 spin_lock(&fs_info->trans_lock);
236 if (fs_info->running_transaction) {
d43317dc
CM
237 /*
238 * someone started a transaction after we unlocked. Make sure
4a9d8bde 239 * to redo the checks above
d43317dc 240 */
a4abeea4 241 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
d43317dc 242 goto loop;
87533c47 243 } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
e4b50e14 244 spin_unlock(&fs_info->trans_lock);
7b8b92af
JB
245 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
246 return -EROFS;
79154b1b 247 }
d43317dc 248
ab8d0fc4 249 cur_trans->fs_info = fs_info;
a4abeea4 250 atomic_set(&cur_trans->num_writers, 1);
0860adfd 251 extwriter_counter_init(cur_trans, type);
a4abeea4
JB
252 init_waitqueue_head(&cur_trans->writer_wait);
253 init_waitqueue_head(&cur_trans->commit_wait);
161c3549 254 init_waitqueue_head(&cur_trans->pending_wait);
4a9d8bde 255 cur_trans->state = TRANS_STATE_RUNNING;
a4abeea4
JB
256 /*
257 * One for this trans handle, one so it will live on until we
258 * commit the transaction.
259 */
260 atomic_set(&cur_trans->use_count, 2);
161c3549 261 atomic_set(&cur_trans->pending_ordered, 0);
3204d33c 262 cur_trans->flags = 0;
a4abeea4
JB
263 cur_trans->start_time = get_seconds();
264
a099d0fd
AM
265 memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));
266
c46effa6 267 cur_trans->delayed_refs.href_root = RB_ROOT;
3368d001 268 cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
d7df2c79 269 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
20b297d6
JS
270
271 /*
272 * although the tree mod log is per file system and not per transaction,
273 * the log must never go across transaction boundaries.
274 */
275 smp_mb();
31b1a2bd 276 if (!list_empty(&fs_info->tree_mod_seq_list))
5d163e0e 277 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");
31b1a2bd 278 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
5d163e0e 279 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");
fc36ed7e 280 atomic64_set(&fs_info->tree_mod_seq, 0);
20b297d6 281
a4abeea4
JB
282 spin_lock_init(&cur_trans->delayed_refs.lock);
283
284 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
6df9a95e 285 INIT_LIST_HEAD(&cur_trans->pending_chunks);
9e351cc8 286 INIT_LIST_HEAD(&cur_trans->switch_commits);
ce93ec54 287 INIT_LIST_HEAD(&cur_trans->dirty_bgs);
1bbc621e 288 INIT_LIST_HEAD(&cur_trans->io_bgs);
2b9dbef2 289 INIT_LIST_HEAD(&cur_trans->dropped_roots);
1bbc621e 290 mutex_init(&cur_trans->cache_write_mutex);
cb723e49 291 cur_trans->num_dirty_bgs = 0;
ce93ec54 292 spin_lock_init(&cur_trans->dirty_bgs_lock);
e33e17ee 293 INIT_LIST_HEAD(&cur_trans->deleted_bgs);
2b9dbef2 294 spin_lock_init(&cur_trans->dropped_roots_lock);
19ae4e81 295 list_add_tail(&cur_trans->list, &fs_info->trans_list);
a4abeea4 296 extent_io_tree_init(&cur_trans->dirty_pages,
19ae4e81
JS
297 fs_info->btree_inode->i_mapping);
298 fs_info->generation++;
299 cur_trans->transid = fs_info->generation;
300 fs_info->running_transaction = cur_trans;
49b25e05 301 cur_trans->aborted = 0;
19ae4e81 302 spin_unlock(&fs_info->trans_lock);
15ee9bc7 303
79154b1b
CM
304 return 0;
305}
306
d352ac68 307/*
d397712b
CM
308 * this does all the record keeping required to make sure that a reference
309 * counted root is properly recorded in a given transaction. This is required
310 * to make sure the old root from before we joined the transaction is deleted
311 * when the transaction commits
d352ac68 312 */
7585717f 313static int record_root_in_trans(struct btrfs_trans_handle *trans,
6426c7ad
QW
314 struct btrfs_root *root,
315 int force)
6702ed49 316{
6426c7ad
QW
317 if ((test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
318 root->last_trans < trans->transid) || force) {
6702ed49 319 WARN_ON(root == root->fs_info->extent_root);
5d4f98a2
YZ
320 WARN_ON(root->commit_root != root->node);
321
7585717f 322 /*
27cdeb70 323 * see below for IN_TRANS_SETUP usage rules
7585717f
CM
324 * we have the reloc mutex held now, so there
325 * is only one writer in this function
326 */
27cdeb70 327 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
7585717f 328
27cdeb70 329 /* make sure readers find IN_TRANS_SETUP before
7585717f
CM
330 * they find our root->last_trans update
331 */
332 smp_wmb();
333
a4abeea4 334 spin_lock(&root->fs_info->fs_roots_radix_lock);
6426c7ad 335 if (root->last_trans == trans->transid && !force) {
a4abeea4
JB
336 spin_unlock(&root->fs_info->fs_roots_radix_lock);
337 return 0;
338 }
5d4f98a2
YZ
339 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
340 (unsigned long)root->root_key.objectid,
341 BTRFS_ROOT_TRANS_TAG);
a4abeea4 342 spin_unlock(&root->fs_info->fs_roots_radix_lock);
7585717f
CM
343 root->last_trans = trans->transid;
344
345 /* this is pretty tricky. We don't want to
346 * take the relocation lock in btrfs_record_root_in_trans
347 * unless we're really doing the first setup for this root in
348 * this transaction.
349 *
350 * Normally we'd use root->last_trans as a flag to decide
351 * if we want to take the expensive mutex.
352 *
353 * But, we have to set root->last_trans before we
354 * init the relocation root, otherwise, we trip over warnings
355 * in ctree.c. The solution used here is to flag ourselves
27cdeb70 356 * with root IN_TRANS_SETUP. When this is 1, we're still
7585717f
CM
357 * fixing up the reloc trees and everyone must wait.
358 *
359 * When this is zero, they can trust root->last_trans and fly
360 * through btrfs_record_root_in_trans without having to take the
361 * lock. smp_wmb() makes sure that all the writes above are
362 * done before we pop in the zero below
363 */
5d4f98a2 364 btrfs_init_reloc_root(trans, root);
c7548af6 365 smp_mb__before_atomic();
27cdeb70 366 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
5d4f98a2
YZ
367 }
368 return 0;
369}
bcc63abb 370
7585717f 371
2b9dbef2
JB
372void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
373 struct btrfs_root *root)
374{
375 struct btrfs_transaction *cur_trans = trans->transaction;
376
377 /* Add ourselves to the transaction dropped list */
378 spin_lock(&cur_trans->dropped_roots_lock);
379 list_add_tail(&root->root_list, &cur_trans->dropped_roots);
380 spin_unlock(&cur_trans->dropped_roots_lock);
381
382 /* Make sure we don't try to update the root at commit time */
383 spin_lock(&root->fs_info->fs_roots_radix_lock);
384 radix_tree_tag_clear(&root->fs_info->fs_roots_radix,
385 (unsigned long)root->root_key.objectid,
386 BTRFS_ROOT_TRANS_TAG);
387 spin_unlock(&root->fs_info->fs_roots_radix_lock);
388}
389
7585717f
CM
390int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
391 struct btrfs_root *root)
392{
27cdeb70 393 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
7585717f
CM
394 return 0;
395
396 /*
27cdeb70 397 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
7585717f
CM
398 * and barriers
399 */
400 smp_rmb();
401 if (root->last_trans == trans->transid &&
27cdeb70 402 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
7585717f
CM
403 return 0;
404
405 mutex_lock(&root->fs_info->reloc_mutex);
6426c7ad 406 record_root_in_trans(trans, root, 0);
7585717f
CM
407 mutex_unlock(&root->fs_info->reloc_mutex);
408
409 return 0;
410}
411
4a9d8bde
MX
412static inline int is_transaction_blocked(struct btrfs_transaction *trans)
413{
414 return (trans->state >= TRANS_STATE_BLOCKED &&
501407aa
JB
415 trans->state < TRANS_STATE_UNBLOCKED &&
416 !trans->aborted);
4a9d8bde
MX
417}
418
d352ac68
CM
419/* wait for commit against the current transaction to become unblocked
420 * when this is done, it is safe to start a new transaction, but the current
421 * transaction might not be fully on disk.
422 */
37d1aeee 423static void wait_current_trans(struct btrfs_root *root)
79154b1b 424{
f9295749 425 struct btrfs_transaction *cur_trans;
79154b1b 426
a4abeea4 427 spin_lock(&root->fs_info->trans_lock);
f9295749 428 cur_trans = root->fs_info->running_transaction;
4a9d8bde 429 if (cur_trans && is_transaction_blocked(cur_trans)) {
13c5a93e 430 atomic_inc(&cur_trans->use_count);
a4abeea4 431 spin_unlock(&root->fs_info->trans_lock);
72d63ed6
LZ
432
433 wait_event(root->fs_info->transaction_wait,
501407aa
JB
434 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
435 cur_trans->aborted);
724e2315 436 btrfs_put_transaction(cur_trans);
a4abeea4
JB
437 } else {
438 spin_unlock(&root->fs_info->trans_lock);
f9295749 439 }
37d1aeee
CM
440}
441
a22285a6
YZ
442static int may_wait_transaction(struct btrfs_root *root, int type)
443{
afcdd129 444 if (test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags))
a4abeea4
JB
445 return 0;
446
447 if (type == TRANS_USERSPACE)
448 return 1;
449
450 if (type == TRANS_START &&
451 !atomic_read(&root->fs_info->open_ioctl_trans))
a22285a6 452 return 1;
a4abeea4 453
a22285a6
YZ
454 return 0;
455}
456
20dd2cbf
MX
457static inline bool need_reserve_reloc_root(struct btrfs_root *root)
458{
459 if (!root->fs_info->reloc_ctl ||
27cdeb70 460 !test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
20dd2cbf
MX
461 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
462 root->reloc_root)
463 return false;
464
465 return true;
466}
467
08e007d2 468static struct btrfs_trans_handle *
5aed1dd8
AM
469start_transaction(struct btrfs_root *root, unsigned int num_items,
470 unsigned int type, enum btrfs_reserve_flush_enum flush)
37d1aeee 471{
a22285a6
YZ
472 struct btrfs_trans_handle *h;
473 struct btrfs_transaction *cur_trans;
b5009945 474 u64 num_bytes = 0;
c5567237 475 u64 qgroup_reserved = 0;
20dd2cbf
MX
476 bool reloc_reserved = false;
477 int ret;
acce952b 478
46c4e71e 479 /* Send isn't supposed to start transactions. */
2755a0de 480 ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
46c4e71e 481
87533c47 482 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
acce952b 483 return ERR_PTR(-EROFS);
2a1eb461 484
46c4e71e 485 if (current->journal_info) {
0860adfd 486 WARN_ON(type & TRANS_EXTWRITERS);
2a1eb461
JB
487 h = current->journal_info;
488 h->use_count++;
b7d5b0a8 489 WARN_ON(h->use_count > 2);
2a1eb461
JB
490 h->orig_rsv = h->block_rsv;
491 h->block_rsv = NULL;
492 goto got_it;
493 }
b5009945
JB
494
495 /*
496 * Do the reservation before we join the transaction so we can do all
497 * the appropriate flushing if need be.
498 */
499 if (num_items > 0 && root != root->fs_info->chunk_root) {
7174109c
QW
500 qgroup_reserved = num_items * root->nodesize;
501 ret = btrfs_qgroup_reserve_meta(root, qgroup_reserved);
502 if (ret)
503 return ERR_PTR(ret);
c5567237 504
b5009945 505 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
20dd2cbf
MX
506 /*
507 * Do the reservation for the relocation root creation
508 */
ee39b432 509 if (need_reserve_reloc_root(root)) {
20dd2cbf
MX
510 num_bytes += root->nodesize;
511 reloc_reserved = true;
512 }
513
08e007d2
MX
514 ret = btrfs_block_rsv_add(root,
515 &root->fs_info->trans_block_rsv,
516 num_bytes, flush);
b5009945 517 if (ret)
843fcf35 518 goto reserve_fail;
b5009945 519 }
a22285a6 520again:
f2f767e7 521 h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
843fcf35
MX
522 if (!h) {
523 ret = -ENOMEM;
524 goto alloc_fail;
525 }
37d1aeee 526
98114659
JB
527 /*
528 * If we are JOIN_NOLOCK we're already committing a transaction and
529 * waiting on this guy, so we don't need to do the sb_start_intwrite
530 * because we're already holding a ref. We need this because we could
531 * have raced in and did an fsync() on a file which can kick a commit
532 * and then we deadlock with somebody doing a freeze.
354aa0fb
MX
533 *
534 * If we are ATTACH, it means we just want to catch the current
535 * transaction and commit it, so we needn't do sb_start_intwrite().
98114659 536 */
0860adfd 537 if (type & __TRANS_FREEZABLE)
60376ce4 538 sb_start_intwrite(root->fs_info->sb);
b2b5ef5c 539
a22285a6 540 if (may_wait_transaction(root, type))
37d1aeee 541 wait_current_trans(root);
a22285a6 542
a4abeea4 543 do {
354aa0fb 544 ret = join_transaction(root, type);
178260b2 545 if (ret == -EBUSY) {
a4abeea4 546 wait_current_trans(root);
178260b2
MX
547 if (unlikely(type == TRANS_ATTACH))
548 ret = -ENOENT;
549 }
a4abeea4
JB
550 } while (ret == -EBUSY);
551
a43f7f82 552 if (ret < 0)
843fcf35 553 goto join_fail;
0f7d52f4 554
a22285a6 555 cur_trans = root->fs_info->running_transaction;
a22285a6
YZ
556
557 h->transid = cur_trans->transid;
558 h->transaction = cur_trans;
d13603ef 559 h->root = root;
2a1eb461 560 h->use_count = 1;
64b63580 561 h->fs_info = root->fs_info;
7174109c 562
a698d075 563 h->type = type;
d9a0540a 564 h->can_flush_pending_bgs = true;
bed92eae 565 INIT_LIST_HEAD(&h->qgroup_ref_list);
ea658bad 566 INIT_LIST_HEAD(&h->new_bgs);
b7ec40d7 567
a22285a6 568 smp_mb();
4a9d8bde
MX
569 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
570 may_wait_transaction(root, type)) {
abdd2e80 571 current->journal_info = h;
a22285a6
YZ
572 btrfs_commit_transaction(h, root);
573 goto again;
574 }
575
b5009945 576 if (num_bytes) {
8c2a3ca2 577 trace_btrfs_space_reservation(root->fs_info, "transaction",
2bcc0328 578 h->transid, num_bytes, 1);
b5009945
JB
579 h->block_rsv = &root->fs_info->trans_block_rsv;
580 h->bytes_reserved = num_bytes;
20dd2cbf 581 h->reloc_reserved = reloc_reserved;
a22285a6 582 }
9ed74f2d 583
2a1eb461 584got_it:
a4abeea4 585 btrfs_record_root_in_trans(h, root);
a22285a6
YZ
586
587 if (!current->journal_info && type != TRANS_USERSPACE)
588 current->journal_info = h;
79154b1b 589 return h;
843fcf35
MX
590
591join_fail:
0860adfd 592 if (type & __TRANS_FREEZABLE)
843fcf35
MX
593 sb_end_intwrite(root->fs_info->sb);
594 kmem_cache_free(btrfs_trans_handle_cachep, h);
595alloc_fail:
596 if (num_bytes)
597 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
598 num_bytes);
599reserve_fail:
7174109c 600 btrfs_qgroup_free_meta(root, qgroup_reserved);
843fcf35 601 return ERR_PTR(ret);
79154b1b
CM
602}
603
f9295749 604struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
5aed1dd8 605 unsigned int num_items)
f9295749 606{
08e007d2
MX
607 return start_transaction(root, num_items, TRANS_START,
608 BTRFS_RESERVE_FLUSH_ALL);
f9295749 609}
8eab77ff
FM
610struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
611 struct btrfs_root *root,
612 unsigned int num_items,
613 int min_factor)
614{
615 struct btrfs_trans_handle *trans;
616 u64 num_bytes;
617 int ret;
618
619 trans = btrfs_start_transaction(root, num_items);
620 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
621 return trans;
622
623 trans = btrfs_start_transaction(root, 0);
624 if (IS_ERR(trans))
625 return trans;
626
627 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
628 ret = btrfs_cond_migrate_bytes(root->fs_info,
629 &root->fs_info->trans_block_rsv,
630 num_bytes,
631 min_factor);
632 if (ret) {
633 btrfs_end_transaction(trans, root);
634 return ERR_PTR(ret);
635 }
636
637 trans->block_rsv = &root->fs_info->trans_block_rsv;
638 trans->bytes_reserved = num_bytes;
88d3a5aa
JB
639 trace_btrfs_space_reservation(root->fs_info, "transaction",
640 trans->transid, num_bytes, 1);
8eab77ff
FM
641
642 return trans;
643}
8407aa46 644
08e007d2 645struct btrfs_trans_handle *btrfs_start_transaction_lflush(
5aed1dd8
AM
646 struct btrfs_root *root,
647 unsigned int num_items)
8407aa46 648{
08e007d2
MX
649 return start_transaction(root, num_items, TRANS_START,
650 BTRFS_RESERVE_FLUSH_LIMIT);
8407aa46
MX
651}
652
7a7eaa40 653struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
f9295749 654{
575a75d6
AM
655 return start_transaction(root, 0, TRANS_JOIN,
656 BTRFS_RESERVE_NO_FLUSH);
f9295749
CM
657}
658
7a7eaa40 659struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
0af3d00b 660{
575a75d6
AM
661 return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
662 BTRFS_RESERVE_NO_FLUSH);
0af3d00b
JB
663}
664
7a7eaa40 665struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
9ca9ee09 666{
575a75d6
AM
667 return start_transaction(root, 0, TRANS_USERSPACE,
668 BTRFS_RESERVE_NO_FLUSH);
9ca9ee09
SW
669}
670
d4edf39b
MX
671/*
672 * btrfs_attach_transaction() - catch the running transaction
673 *
674 * It is used when we want to commit the current the transaction, but
675 * don't want to start a new one.
676 *
677 * Note: If this function return -ENOENT, it just means there is no
678 * running transaction. But it is possible that the inactive transaction
679 * is still in the memory, not fully on disk. If you hope there is no
680 * inactive transaction in the fs when -ENOENT is returned, you should
681 * invoke
682 * btrfs_attach_transaction_barrier()
683 */
354aa0fb 684struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
60376ce4 685{
575a75d6
AM
686 return start_transaction(root, 0, TRANS_ATTACH,
687 BTRFS_RESERVE_NO_FLUSH);
60376ce4
JB
688}
689
d4edf39b 690/*
90b6d283 691 * btrfs_attach_transaction_barrier() - catch the running transaction
d4edf39b
MX
692 *
693 * It is similar to the above function, the differentia is this one
694 * will wait for all the inactive transactions until they fully
695 * complete.
696 */
697struct btrfs_trans_handle *
698btrfs_attach_transaction_barrier(struct btrfs_root *root)
699{
700 struct btrfs_trans_handle *trans;
701
575a75d6
AM
702 trans = start_transaction(root, 0, TRANS_ATTACH,
703 BTRFS_RESERVE_NO_FLUSH);
d4edf39b
MX
704 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
705 btrfs_wait_for_commit(root, 0);
706
707 return trans;
708}
709
d352ac68 710/* wait for a transaction commit to be fully complete */
b9c8300c 711static noinline void wait_for_commit(struct btrfs_root *root,
89ce8a63
CM
712 struct btrfs_transaction *commit)
713{
4a9d8bde 714 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
89ce8a63
CM
715}
716
46204592
SW
717int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
718{
719 struct btrfs_transaction *cur_trans = NULL, *t;
8cd2807f 720 int ret = 0;
46204592 721
46204592
SW
722 if (transid) {
723 if (transid <= root->fs_info->last_trans_committed)
a4abeea4 724 goto out;
46204592
SW
725
726 /* find specified transaction */
a4abeea4 727 spin_lock(&root->fs_info->trans_lock);
46204592
SW
728 list_for_each_entry(t, &root->fs_info->trans_list, list) {
729 if (t->transid == transid) {
730 cur_trans = t;
a4abeea4 731 atomic_inc(&cur_trans->use_count);
8cd2807f 732 ret = 0;
46204592
SW
733 break;
734 }
8cd2807f
MX
735 if (t->transid > transid) {
736 ret = 0;
46204592 737 break;
8cd2807f 738 }
46204592 739 }
a4abeea4 740 spin_unlock(&root->fs_info->trans_lock);
42383020
SW
741
742 /*
743 * The specified transaction doesn't exist, or we
744 * raced with btrfs_commit_transaction
745 */
746 if (!cur_trans) {
747 if (transid > root->fs_info->last_trans_committed)
748 ret = -EINVAL;
8cd2807f 749 goto out;
42383020 750 }
46204592
SW
751 } else {
752 /* find newest transaction that is committing | committed */
a4abeea4 753 spin_lock(&root->fs_info->trans_lock);
46204592
SW
754 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
755 list) {
4a9d8bde
MX
756 if (t->state >= TRANS_STATE_COMMIT_START) {
757 if (t->state == TRANS_STATE_COMPLETED)
3473f3c0 758 break;
46204592 759 cur_trans = t;
a4abeea4 760 atomic_inc(&cur_trans->use_count);
46204592
SW
761 break;
762 }
763 }
a4abeea4 764 spin_unlock(&root->fs_info->trans_lock);
46204592 765 if (!cur_trans)
a4abeea4 766 goto out; /* nothing committing|committed */
46204592
SW
767 }
768
46204592 769 wait_for_commit(root, cur_trans);
724e2315 770 btrfs_put_transaction(cur_trans);
a4abeea4 771out:
46204592
SW
772 return ret;
773}
774
37d1aeee
CM
775void btrfs_throttle(struct btrfs_root *root)
776{
a4abeea4 777 if (!atomic_read(&root->fs_info->open_ioctl_trans))
9ca9ee09 778 wait_current_trans(root);
37d1aeee
CM
779}
780
8929ecfa
YZ
781static int should_end_transaction(struct btrfs_trans_handle *trans,
782 struct btrfs_root *root)
783{
1be41b78 784 if (root->fs_info->global_block_rsv.space_info->full &&
0a2b2a84 785 btrfs_check_space_for_delayed_refs(trans, root))
1be41b78 786 return 1;
36ba022a 787
1be41b78 788 return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
8929ecfa
YZ
789}
790
791int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
792 struct btrfs_root *root)
793{
794 struct btrfs_transaction *cur_trans = trans->transaction;
795 int updates;
49b25e05 796 int err;
8929ecfa 797
a4abeea4 798 smp_mb();
4a9d8bde
MX
799 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
800 cur_trans->delayed_refs.flushing)
8929ecfa
YZ
801 return 1;
802
803 updates = trans->delayed_ref_updates;
804 trans->delayed_ref_updates = 0;
49b25e05 805 if (updates) {
28ed1345 806 err = btrfs_run_delayed_refs(trans, root, updates * 2);
49b25e05
JM
807 if (err) /* Error code will also eval true */
808 return err;
809 }
8929ecfa
YZ
810
811 return should_end_transaction(trans, root);
812}
813
89ce8a63 814static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
a698d075 815 struct btrfs_root *root, int throttle)
79154b1b 816{
8929ecfa 817 struct btrfs_transaction *cur_trans = trans->transaction;
ab78c84d 818 struct btrfs_fs_info *info = root->fs_info;
31b9655f 819 u64 transid = trans->transid;
1be41b78 820 unsigned long cur = trans->delayed_ref_updates;
a698d075 821 int lock = (trans->type != TRANS_JOIN_NOLOCK);
4edc2ca3 822 int err = 0;
a79b7d4b 823 int must_run_delayed_refs = 0;
c3e69d58 824
3bbb24b2
JB
825 if (trans->use_count > 1) {
826 trans->use_count--;
2a1eb461
JB
827 trans->block_rsv = trans->orig_rsv;
828 return 0;
829 }
830
b24e03db 831 btrfs_trans_release_metadata(trans, root);
4c13d758 832 trans->block_rsv = NULL;
c5567237 833
ea658bad
JB
834 if (!list_empty(&trans->new_bgs))
835 btrfs_create_pending_block_groups(trans, root);
836
1be41b78 837 trans->delayed_ref_updates = 0;
a79b7d4b
CM
838 if (!trans->sync) {
839 must_run_delayed_refs =
840 btrfs_should_throttle_delayed_refs(trans, root);
0a2b2a84 841 cur = max_t(unsigned long, cur, 32);
a79b7d4b
CM
842
843 /*
844 * don't make the caller wait if they are from a NOLOCK
845 * or ATTACH transaction, it will deadlock with commit
846 */
847 if (must_run_delayed_refs == 1 &&
848 (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH)))
849 must_run_delayed_refs = 2;
56bec294 850 }
bb721703 851
0e721106
JB
852 btrfs_trans_release_metadata(trans, root);
853 trans->block_rsv = NULL;
56bec294 854
ea658bad
JB
855 if (!list_empty(&trans->new_bgs))
856 btrfs_create_pending_block_groups(trans, root);
857
4fbcdf66
FM
858 btrfs_trans_release_chunk_metadata(trans);
859
a4abeea4 860 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
4a9d8bde
MX
861 should_end_transaction(trans, root) &&
862 ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
863 spin_lock(&info->trans_lock);
864 if (cur_trans->state == TRANS_STATE_RUNNING)
865 cur_trans->state = TRANS_STATE_BLOCKED;
866 spin_unlock(&info->trans_lock);
a4abeea4 867 }
8929ecfa 868
4a9d8bde 869 if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
3bbb24b2 870 if (throttle)
8929ecfa 871 return btrfs_commit_transaction(trans, root);
3bbb24b2 872 else
8929ecfa
YZ
873 wake_up_process(info->transaction_kthread);
874 }
875
0860adfd 876 if (trans->type & __TRANS_FREEZABLE)
98114659 877 sb_end_intwrite(root->fs_info->sb);
6df7881a 878
8929ecfa 879 WARN_ON(cur_trans != info->running_transaction);
13c5a93e
JB
880 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
881 atomic_dec(&cur_trans->num_writers);
0860adfd 882 extwriter_counter_dec(cur_trans, trans->type);
89ce8a63 883
a83342aa
DS
884 /*
885 * Make sure counter is updated before we wake up waiters.
886 */
99d16cbc 887 smp_mb();
79154b1b
CM
888 if (waitqueue_active(&cur_trans->writer_wait))
889 wake_up(&cur_trans->writer_wait);
724e2315 890 btrfs_put_transaction(cur_trans);
9ed74f2d
JB
891
892 if (current->journal_info == trans)
893 current->journal_info = NULL;
ab78c84d 894
24bbcf04
YZ
895 if (throttle)
896 btrfs_run_delayed_iputs(root);
897
49b25e05 898 if (trans->aborted ||
4e121c06
JB
899 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
900 wake_up_process(info->transaction_kthread);
4edc2ca3 901 err = -EIO;
4e121c06 902 }
edf39272 903 assert_qgroups_uptodate(trans);
49b25e05 904
4edc2ca3 905 kmem_cache_free(btrfs_trans_handle_cachep, trans);
a79b7d4b 906 if (must_run_delayed_refs) {
31b9655f 907 btrfs_async_run_delayed_refs(root, cur, transid,
a79b7d4b
CM
908 must_run_delayed_refs == 1);
909 }
4edc2ca3 910 return err;
79154b1b
CM
911}
912
89ce8a63
CM
913int btrfs_end_transaction(struct btrfs_trans_handle *trans,
914 struct btrfs_root *root)
915{
98ad43be 916 return __btrfs_end_transaction(trans, root, 0);
89ce8a63
CM
917}
918
919int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
920 struct btrfs_root *root)
921{
98ad43be 922 return __btrfs_end_transaction(trans, root, 1);
16cdcec7
MX
923}
924
d352ac68
CM
925/*
926 * when btree blocks are allocated, they have some corresponding bits set for
927 * them in one of two extent_io trees. This is used to make sure all of
690587d1 928 * those extents are sent to disk but does not wait on them
d352ac68 929 */
690587d1 930int btrfs_write_marked_extents(struct btrfs_root *root,
8cef4e16 931 struct extent_io_tree *dirty_pages, int mark)
79154b1b 932{
777e6bd7 933 int err = 0;
7c4452b9 934 int werr = 0;
1728366e 935 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
e6138876 936 struct extent_state *cached_state = NULL;
777e6bd7 937 u64 start = 0;
5f39d397 938 u64 end;
7c4452b9 939
1728366e 940 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
e6138876 941 mark, &cached_state)) {
663dfbb0
FM
942 bool wait_writeback = false;
943
944 err = convert_extent_bit(dirty_pages, start, end,
945 EXTENT_NEED_WAIT,
210aa277 946 mark, &cached_state);
663dfbb0
FM
947 /*
948 * convert_extent_bit can return -ENOMEM, which is most of the
949 * time a temporary error. So when it happens, ignore the error
950 * and wait for writeback of this range to finish - because we
951 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
952 * to btrfs_wait_marked_extents() would not know that writeback
953 * for this range started and therefore wouldn't wait for it to
954 * finish - we don't want to commit a superblock that points to
955 * btree nodes/leafs for which writeback hasn't finished yet
956 * (and without errors).
957 * We cleanup any entries left in the io tree when committing
958 * the transaction (through clear_btree_io_tree()).
959 */
960 if (err == -ENOMEM) {
961 err = 0;
962 wait_writeback = true;
963 }
964 if (!err)
965 err = filemap_fdatawrite_range(mapping, start, end);
1728366e
JB
966 if (err)
967 werr = err;
663dfbb0
FM
968 else if (wait_writeback)
969 werr = filemap_fdatawait_range(mapping, start, end);
e38e2ed7 970 free_extent_state(cached_state);
663dfbb0 971 cached_state = NULL;
1728366e
JB
972 cond_resched();
973 start = end + 1;
7c4452b9 974 }
690587d1
CM
975 return werr;
976}
977
978/*
979 * when btree blocks are allocated, they have some corresponding bits set for
980 * them in one of two extent_io trees. This is used to make sure all of
981 * those extents are on disk for transaction or log commit. We wait
982 * on all the pages and clear them from the dirty pages state tree
983 */
984int btrfs_wait_marked_extents(struct btrfs_root *root,
8cef4e16 985 struct extent_io_tree *dirty_pages, int mark)
690587d1 986{
690587d1
CM
987 int err = 0;
988 int werr = 0;
1728366e 989 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
e6138876 990 struct extent_state *cached_state = NULL;
690587d1
CM
991 u64 start = 0;
992 u64 end;
656f30db 993 bool errors = false;
777e6bd7 994
1728366e 995 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
e6138876 996 EXTENT_NEED_WAIT, &cached_state)) {
663dfbb0
FM
997 /*
998 * Ignore -ENOMEM errors returned by clear_extent_bit().
999 * When committing the transaction, we'll remove any entries
1000 * left in the io tree. For a log commit, we don't remove them
1001 * after committing the log because the tree can be accessed
1002 * concurrently - we do it only at transaction commit time when
1003 * it's safe to do it (through clear_btree_io_tree()).
1004 */
1005 err = clear_extent_bit(dirty_pages, start, end,
1006 EXTENT_NEED_WAIT,
1007 0, 0, &cached_state, GFP_NOFS);
1008 if (err == -ENOMEM)
1009 err = 0;
1010 if (!err)
1011 err = filemap_fdatawait_range(mapping, start, end);
1728366e
JB
1012 if (err)
1013 werr = err;
e38e2ed7
FM
1014 free_extent_state(cached_state);
1015 cached_state = NULL;
1728366e
JB
1016 cond_resched();
1017 start = end + 1;
777e6bd7 1018 }
7c4452b9
CM
1019 if (err)
1020 werr = err;
656f30db
FM
1021
1022 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1023 if ((mark & EXTENT_DIRTY) &&
afcdd129
JB
1024 test_and_clear_bit(BTRFS_FS_LOG1_ERR,
1025 &root->fs_info->flags))
656f30db
FM
1026 errors = true;
1027
1028 if ((mark & EXTENT_NEW) &&
afcdd129
JB
1029 test_and_clear_bit(BTRFS_FS_LOG2_ERR,
1030 &root->fs_info->flags))
656f30db
FM
1031 errors = true;
1032 } else {
afcdd129
JB
1033 if (test_and_clear_bit(BTRFS_FS_BTREE_ERR,
1034 &root->fs_info->flags))
656f30db
FM
1035 errors = true;
1036 }
1037
1038 if (errors && !werr)
1039 werr = -EIO;
1040
7c4452b9 1041 return werr;
79154b1b
CM
1042}
1043
690587d1
CM
1044/*
1045 * when btree blocks are allocated, they have some corresponding bits set for
1046 * them in one of two extent_io trees. This is used to make sure all of
1047 * those extents are on disk for transaction or log commit
1048 */
171170c1 1049static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
8cef4e16 1050 struct extent_io_tree *dirty_pages, int mark)
690587d1
CM
1051{
1052 int ret;
1053 int ret2;
c6adc9cc 1054 struct blk_plug plug;
690587d1 1055
c6adc9cc 1056 blk_start_plug(&plug);
8cef4e16 1057 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
c6adc9cc 1058 blk_finish_plug(&plug);
8cef4e16 1059 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
bf0da8c1
CM
1060
1061 if (ret)
1062 return ret;
1063 if (ret2)
1064 return ret2;
1065 return 0;
690587d1
CM
1066}
1067
663dfbb0 1068static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
d0c803c4
CM
1069 struct btrfs_root *root)
1070{
663dfbb0
FM
1071 int ret;
1072
1073 ret = btrfs_write_and_wait_marked_extents(root,
8cef4e16
YZ
1074 &trans->transaction->dirty_pages,
1075 EXTENT_DIRTY);
663dfbb0
FM
1076 clear_btree_io_tree(&trans->transaction->dirty_pages);
1077
1078 return ret;
d0c803c4
CM
1079}
1080
d352ac68
CM
1081/*
1082 * this is used to update the root pointer in the tree of tree roots.
1083 *
1084 * But, in the case of the extent allocation tree, updating the root
1085 * pointer may allocate blocks which may change the root of the extent
1086 * allocation tree.
1087 *
1088 * So, this loops and repeats and makes sure the cowonly root didn't
1089 * change while the root pointer was being updated in the metadata.
1090 */
0b86a832
CM
1091static int update_cowonly_root(struct btrfs_trans_handle *trans,
1092 struct btrfs_root *root)
79154b1b
CM
1093{
1094 int ret;
0b86a832 1095 u64 old_root_bytenr;
86b9f2ec 1096 u64 old_root_used;
0b86a832 1097 struct btrfs_root *tree_root = root->fs_info->tree_root;
79154b1b 1098
86b9f2ec 1099 old_root_used = btrfs_root_used(&root->root_item);
56bec294 1100
d397712b 1101 while (1) {
0b86a832 1102 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
86b9f2ec 1103 if (old_root_bytenr == root->node->start &&
ea526d18 1104 old_root_used == btrfs_root_used(&root->root_item))
79154b1b 1105 break;
87ef2bb4 1106
5d4f98a2 1107 btrfs_set_root_node(&root->root_item, root->node);
79154b1b 1108 ret = btrfs_update_root(trans, tree_root,
0b86a832
CM
1109 &root->root_key,
1110 &root->root_item);
49b25e05
JM
1111 if (ret)
1112 return ret;
56bec294 1113
86b9f2ec 1114 old_root_used = btrfs_root_used(&root->root_item);
0b86a832 1115 }
276e680d 1116
0b86a832
CM
1117 return 0;
1118}
1119
d352ac68
CM
1120/*
1121 * update all the cowonly tree roots on disk
49b25e05
JM
1122 *
1123 * The error handling in this function may not be obvious. Any of the
1124 * failures will cause the file system to go offline. We still need
1125 * to clean up the delayed refs.
d352ac68 1126 */
5d4f98a2
YZ
1127static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
1128 struct btrfs_root *root)
0b86a832
CM
1129{
1130 struct btrfs_fs_info *fs_info = root->fs_info;
ea526d18 1131 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1bbc621e 1132 struct list_head *io_bgs = &trans->transaction->io_bgs;
0b86a832 1133 struct list_head *next;
84234f3a 1134 struct extent_buffer *eb;
56bec294 1135 int ret;
84234f3a
YZ
1136
1137 eb = btrfs_lock_root_node(fs_info->tree_root);
49b25e05
JM
1138 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1139 0, &eb);
84234f3a
YZ
1140 btrfs_tree_unlock(eb);
1141 free_extent_buffer(eb);
0b86a832 1142
49b25e05
JM
1143 if (ret)
1144 return ret;
1145
56bec294 1146 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
49b25e05
JM
1147 if (ret)
1148 return ret;
87ef2bb4 1149
733f4fbb 1150 ret = btrfs_run_dev_stats(trans, root->fs_info);
c16ce190
JB
1151 if (ret)
1152 return ret;
8dabb742 1153 ret = btrfs_run_dev_replace(trans, root->fs_info);
c16ce190
JB
1154 if (ret)
1155 return ret;
546adb0d 1156 ret = btrfs_run_qgroups(trans, root->fs_info);
c16ce190
JB
1157 if (ret)
1158 return ret;
546adb0d 1159
dcdf7f6d
JB
1160 ret = btrfs_setup_space_cache(trans, root);
1161 if (ret)
1162 return ret;
1163
546adb0d
JS
1164 /* run_qgroups might have added some more refs */
1165 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
c16ce190
JB
1166 if (ret)
1167 return ret;
ea526d18 1168again:
d397712b 1169 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
0b86a832
CM
1170 next = fs_info->dirty_cowonly_roots.next;
1171 list_del_init(next);
1172 root = list_entry(next, struct btrfs_root, dirty_list);
e7070be1 1173 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
87ef2bb4 1174
9e351cc8
JB
1175 if (root != fs_info->extent_root)
1176 list_add_tail(&root->dirty_list,
1177 &trans->transaction->switch_commits);
49b25e05
JM
1178 ret = update_cowonly_root(trans, root);
1179 if (ret)
1180 return ret;
ea526d18
JB
1181 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1182 if (ret)
1183 return ret;
79154b1b 1184 }
276e680d 1185
1bbc621e 1186 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
ea526d18
JB
1187 ret = btrfs_write_dirty_block_groups(trans, root);
1188 if (ret)
1189 return ret;
1190 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1191 if (ret)
1192 return ret;
1193 }
1194
1195 if (!list_empty(&fs_info->dirty_cowonly_roots))
1196 goto again;
1197
9e351cc8
JB
1198 list_add_tail(&fs_info->extent_root->dirty_list,
1199 &trans->transaction->switch_commits);
8dabb742
SB
1200 btrfs_after_dev_replace_commit(fs_info);
1201
79154b1b
CM
1202 return 0;
1203}
1204
d352ac68
CM
1205/*
1206 * dead roots are old snapshots that need to be deleted. This allocates
1207 * a dirty root struct and adds it into the list of dead roots that need to
1208 * be deleted
1209 */
cfad392b 1210void btrfs_add_dead_root(struct btrfs_root *root)
5eda7b5e 1211{
a4abeea4 1212 spin_lock(&root->fs_info->trans_lock);
cfad392b
JB
1213 if (list_empty(&root->root_list))
1214 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
a4abeea4 1215 spin_unlock(&root->fs_info->trans_lock);
5eda7b5e
CM
1216}
1217
d352ac68 1218/*
5d4f98a2 1219 * update all the cowonly tree roots on disk
d352ac68 1220 */
5d4f98a2
YZ
1221static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1222 struct btrfs_root *root)
0f7d52f4 1223{
0f7d52f4 1224 struct btrfs_root *gang[8];
5d4f98a2 1225 struct btrfs_fs_info *fs_info = root->fs_info;
0f7d52f4
CM
1226 int i;
1227 int ret;
54aa1f4d
CM
1228 int err = 0;
1229
a4abeea4 1230 spin_lock(&fs_info->fs_roots_radix_lock);
d397712b 1231 while (1) {
5d4f98a2
YZ
1232 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1233 (void **)gang, 0,
0f7d52f4
CM
1234 ARRAY_SIZE(gang),
1235 BTRFS_ROOT_TRANS_TAG);
1236 if (ret == 0)
1237 break;
1238 for (i = 0; i < ret; i++) {
1239 root = gang[i];
5d4f98a2
YZ
1240 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1241 (unsigned long)root->root_key.objectid,
1242 BTRFS_ROOT_TRANS_TAG);
a4abeea4 1243 spin_unlock(&fs_info->fs_roots_radix_lock);
31153d81 1244
e02119d5 1245 btrfs_free_log(trans, root);
5d4f98a2 1246 btrfs_update_reloc_root(trans, root);
d68fc57b 1247 btrfs_orphan_commit_root(trans, root);
bcc63abb 1248
82d5902d
LZ
1249 btrfs_save_ino_cache(root, trans);
1250
f1ebcc74 1251 /* see comments in should_cow_block() */
27cdeb70 1252 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
c7548af6 1253 smp_mb__after_atomic();
f1ebcc74 1254
978d910d 1255 if (root->commit_root != root->node) {
9e351cc8
JB
1256 list_add_tail(&root->dirty_list,
1257 &trans->transaction->switch_commits);
978d910d
YZ
1258 btrfs_set_root_node(&root->root_item,
1259 root->node);
1260 }
5d4f98a2 1261
5d4f98a2 1262 err = btrfs_update_root(trans, fs_info->tree_root,
0f7d52f4
CM
1263 &root->root_key,
1264 &root->root_item);
a4abeea4 1265 spin_lock(&fs_info->fs_roots_radix_lock);
54aa1f4d
CM
1266 if (err)
1267 break;
7174109c 1268 btrfs_qgroup_free_meta_all(root);
0f7d52f4
CM
1269 }
1270 }
a4abeea4 1271 spin_unlock(&fs_info->fs_roots_radix_lock);
54aa1f4d 1272 return err;
0f7d52f4
CM
1273}
1274
d352ac68 1275/*
de78b51a
ES
1276 * defrag a given btree.
1277 * Every leaf in the btree is read and defragged.
d352ac68 1278 */
de78b51a 1279int btrfs_defrag_root(struct btrfs_root *root)
e9d0b13b
CM
1280{
1281 struct btrfs_fs_info *info = root->fs_info;
e9d0b13b 1282 struct btrfs_trans_handle *trans;
8929ecfa 1283 int ret;
e9d0b13b 1284
27cdeb70 1285 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
e9d0b13b 1286 return 0;
8929ecfa 1287
6b80053d 1288 while (1) {
8929ecfa
YZ
1289 trans = btrfs_start_transaction(root, 0);
1290 if (IS_ERR(trans))
1291 return PTR_ERR(trans);
1292
de78b51a 1293 ret = btrfs_defrag_leaves(trans, root);
8929ecfa 1294
e9d0b13b 1295 btrfs_end_transaction(trans, root);
b53d3f5d 1296 btrfs_btree_balance_dirty(info->tree_root);
e9d0b13b
CM
1297 cond_resched();
1298
ab8d0fc4 1299 if (btrfs_fs_closing(info) || ret != -EAGAIN)
e9d0b13b 1300 break;
210549eb 1301
ab8d0fc4
JM
1302 if (btrfs_defrag_cancelled(info)) {
1303 btrfs_debug(info, "defrag_root cancelled");
210549eb
DS
1304 ret = -EAGAIN;
1305 break;
1306 }
e9d0b13b 1307 }
27cdeb70 1308 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
8929ecfa 1309 return ret;
e9d0b13b
CM
1310}
1311
6426c7ad
QW
1312/*
1313 * Do all special snapshot related qgroup dirty hack.
1314 *
1315 * Will do all needed qgroup inherit and dirty hack like switch commit
1316 * roots inside one transaction and write all btree into disk, to make
1317 * qgroup works.
1318 */
1319static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1320 struct btrfs_root *src,
1321 struct btrfs_root *parent,
1322 struct btrfs_qgroup_inherit *inherit,
1323 u64 dst_objectid)
1324{
1325 struct btrfs_fs_info *fs_info = src->fs_info;
1326 int ret;
1327
1328 /*
1329 * Save some performance in the case that qgroups are not
1330 * enabled. If this check races with the ioctl, rescan will
1331 * kick in anyway.
1332 */
1333 mutex_lock(&fs_info->qgroup_ioctl_lock);
afcdd129 1334 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
6426c7ad
QW
1335 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1336 return 0;
1337 }
1338 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1339
1340 /*
1341 * We are going to commit transaction, see btrfs_commit_transaction()
1342 * comment for reason locking tree_log_mutex
1343 */
1344 mutex_lock(&fs_info->tree_log_mutex);
1345
1346 ret = commit_fs_roots(trans, src);
1347 if (ret)
1348 goto out;
1349 ret = btrfs_qgroup_prepare_account_extents(trans, fs_info);
1350 if (ret < 0)
1351 goto out;
1352 ret = btrfs_qgroup_account_extents(trans, fs_info);
1353 if (ret < 0)
1354 goto out;
1355
1356 /* Now qgroup are all updated, we can inherit it to new qgroups */
1357 ret = btrfs_qgroup_inherit(trans, fs_info,
1358 src->root_key.objectid, dst_objectid,
1359 inherit);
1360 if (ret < 0)
1361 goto out;
1362
1363 /*
1364 * Now we do a simplified commit transaction, which will:
1365 * 1) commit all subvolume and extent tree
1366 * To ensure all subvolume and extent tree have a valid
1367 * commit_root to accounting later insert_dir_item()
1368 * 2) write all btree blocks onto disk
1369 * This is to make sure later btree modification will be cowed
1370 * Or commit_root can be populated and cause wrong qgroup numbers
1371 * In this simplified commit, we don't really care about other trees
1372 * like chunk and root tree, as they won't affect qgroup.
1373 * And we don't write super to avoid half committed status.
1374 */
1375 ret = commit_cowonly_roots(trans, src);
1376 if (ret)
1377 goto out;
1378 switch_commit_roots(trans->transaction, fs_info);
1379 ret = btrfs_write_and_wait_transaction(trans, src);
1380 if (ret)
f7af3934 1381 btrfs_handle_fs_error(fs_info, ret,
6426c7ad
QW
1382 "Error while writing out transaction for qgroup");
1383
1384out:
1385 mutex_unlock(&fs_info->tree_log_mutex);
1386
1387 /*
1388 * Force parent root to be updated, as we recorded it before so its
1389 * last_trans == cur_transid.
1390 * Or it won't be committed again onto disk after later
1391 * insert_dir_item()
1392 */
1393 if (!ret)
1394 record_root_in_trans(trans, parent, 1);
1395 return ret;
1396}
1397
d352ac68
CM
1398/*
1399 * new snapshots need to be created at a very specific time in the
aec8030a
MX
1400 * transaction commit. This does the actual creation.
1401 *
1402 * Note:
1403 * If the error which may affect the commitment of the current transaction
1404 * happens, we should return the error number. If the error which just affect
1405 * the creation of the pending snapshots, just return 0.
d352ac68 1406 */
80b6794d 1407static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
3063d29f
CM
1408 struct btrfs_fs_info *fs_info,
1409 struct btrfs_pending_snapshot *pending)
1410{
1411 struct btrfs_key key;
80b6794d 1412 struct btrfs_root_item *new_root_item;
3063d29f
CM
1413 struct btrfs_root *tree_root = fs_info->tree_root;
1414 struct btrfs_root *root = pending->root;
6bdb72de 1415 struct btrfs_root *parent_root;
98c9942a 1416 struct btrfs_block_rsv *rsv;
6bdb72de 1417 struct inode *parent_inode;
42874b3d
MX
1418 struct btrfs_path *path;
1419 struct btrfs_dir_item *dir_item;
a22285a6 1420 struct dentry *dentry;
3063d29f 1421 struct extent_buffer *tmp;
925baedd 1422 struct extent_buffer *old;
04b285f3 1423 struct timespec cur_time;
aec8030a 1424 int ret = 0;
d68fc57b 1425 u64 to_reserve = 0;
6bdb72de 1426 u64 index = 0;
a22285a6 1427 u64 objectid;
b83cc969 1428 u64 root_flags;
8ea05e3a 1429 uuid_le new_uuid;
3063d29f 1430
8546b570
DS
1431 ASSERT(pending->path);
1432 path = pending->path;
42874b3d 1433
b0c0ea63
DS
1434 ASSERT(pending->root_item);
1435 new_root_item = pending->root_item;
a22285a6 1436
aec8030a
MX
1437 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1438 if (pending->error)
6fa9700e 1439 goto no_free_objectid;
3063d29f 1440
d6726335
QW
1441 /*
1442 * Make qgroup to skip current new snapshot's qgroupid, as it is
1443 * accounted by later btrfs_qgroup_inherit().
1444 */
1445 btrfs_set_skip_qgroup(trans, objectid);
1446
147d256e 1447 btrfs_reloc_pre_snapshot(pending, &to_reserve);
d68fc57b
YZ
1448
1449 if (to_reserve > 0) {
aec8030a
MX
1450 pending->error = btrfs_block_rsv_add(root,
1451 &pending->block_rsv,
1452 to_reserve,
1453 BTRFS_RESERVE_NO_FLUSH);
1454 if (pending->error)
d6726335 1455 goto clear_skip_qgroup;
d68fc57b
YZ
1456 }
1457
3063d29f 1458 key.objectid = objectid;
a22285a6
YZ
1459 key.offset = (u64)-1;
1460 key.type = BTRFS_ROOT_ITEM_KEY;
3063d29f 1461
6fa9700e 1462 rsv = trans->block_rsv;
a22285a6 1463 trans->block_rsv = &pending->block_rsv;
2382c5cc 1464 trans->bytes_reserved = trans->block_rsv->reserved;
88d3a5aa
JB
1465 trace_btrfs_space_reservation(root->fs_info, "transaction",
1466 trans->transid,
1467 trans->bytes_reserved, 1);
a22285a6 1468 dentry = pending->dentry;
e9662f70 1469 parent_inode = pending->dir;
a22285a6 1470 parent_root = BTRFS_I(parent_inode)->root;
6426c7ad 1471 record_root_in_trans(trans, parent_root, 0);
a22285a6 1472
c2050a45 1473 cur_time = current_time(parent_inode);
04b285f3 1474
3063d29f
CM
1475 /*
1476 * insert the directory item
1477 */
3de4586c 1478 ret = btrfs_set_inode_index(parent_inode, &index);
49b25e05 1479 BUG_ON(ret); /* -ENOMEM */
42874b3d
MX
1480
1481 /* check if there is a file/dir which has the same name. */
1482 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1483 btrfs_ino(parent_inode),
1484 dentry->d_name.name,
1485 dentry->d_name.len, 0);
1486 if (dir_item != NULL && !IS_ERR(dir_item)) {
fe66a05a 1487 pending->error = -EEXIST;
aec8030a 1488 goto dir_item_existed;
42874b3d
MX
1489 } else if (IS_ERR(dir_item)) {
1490 ret = PTR_ERR(dir_item);
66642832 1491 btrfs_abort_transaction(trans, ret);
8732d44f 1492 goto fail;
79787eaa 1493 }
42874b3d 1494 btrfs_release_path(path);
52c26179 1495
e999376f
CM
1496 /*
1497 * pull in the delayed directory update
1498 * and the delayed inode item
1499 * otherwise we corrupt the FS during
1500 * snapshot
1501 */
1502 ret = btrfs_run_delayed_items(trans, root);
8732d44f 1503 if (ret) { /* Transaction aborted */
66642832 1504 btrfs_abort_transaction(trans, ret);
8732d44f
MX
1505 goto fail;
1506 }
e999376f 1507
6426c7ad 1508 record_root_in_trans(trans, root, 0);
6bdb72de
SW
1509 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1510 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
08fe4db1 1511 btrfs_check_and_init_root_item(new_root_item);
6bdb72de 1512
b83cc969
LZ
1513 root_flags = btrfs_root_flags(new_root_item);
1514 if (pending->readonly)
1515 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1516 else
1517 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1518 btrfs_set_root_flags(new_root_item, root_flags);
1519
8ea05e3a
AB
1520 btrfs_set_root_generation_v2(new_root_item,
1521 trans->transid);
1522 uuid_le_gen(&new_uuid);
1523 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1524 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1525 BTRFS_UUID_SIZE);
70023da2
SB
1526 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1527 memset(new_root_item->received_uuid, 0,
1528 sizeof(new_root_item->received_uuid));
1529 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1530 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1531 btrfs_set_root_stransid(new_root_item, 0);
1532 btrfs_set_root_rtransid(new_root_item, 0);
1533 }
3cae210f
QW
1534 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1535 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
8ea05e3a 1536 btrfs_set_root_otransid(new_root_item, trans->transid);
8ea05e3a 1537
6bdb72de 1538 old = btrfs_lock_root_node(root);
49b25e05 1539 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
79787eaa
JM
1540 if (ret) {
1541 btrfs_tree_unlock(old);
1542 free_extent_buffer(old);
66642832 1543 btrfs_abort_transaction(trans, ret);
8732d44f 1544 goto fail;
79787eaa 1545 }
49b25e05 1546
6bdb72de
SW
1547 btrfs_set_lock_blocking(old);
1548
49b25e05 1549 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
79787eaa 1550 /* clean up in any case */
6bdb72de
SW
1551 btrfs_tree_unlock(old);
1552 free_extent_buffer(old);
8732d44f 1553 if (ret) {
66642832 1554 btrfs_abort_transaction(trans, ret);
8732d44f
MX
1555 goto fail;
1556 }
f1ebcc74 1557 /* see comments in should_cow_block() */
27cdeb70 1558 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
f1ebcc74
LB
1559 smp_wmb();
1560
6bdb72de 1561 btrfs_set_root_node(new_root_item, tmp);
a22285a6
YZ
1562 /* record when the snapshot was created in key.offset */
1563 key.offset = trans->transid;
1564 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
6bdb72de
SW
1565 btrfs_tree_unlock(tmp);
1566 free_extent_buffer(tmp);
8732d44f 1567 if (ret) {
66642832 1568 btrfs_abort_transaction(trans, ret);
8732d44f
MX
1569 goto fail;
1570 }
6bdb72de 1571
a22285a6
YZ
1572 /*
1573 * insert root back/forward references
1574 */
1575 ret = btrfs_add_root_ref(trans, tree_root, objectid,
0660b5af 1576 parent_root->root_key.objectid,
33345d01 1577 btrfs_ino(parent_inode), index,
a22285a6 1578 dentry->d_name.name, dentry->d_name.len);
8732d44f 1579 if (ret) {
66642832 1580 btrfs_abort_transaction(trans, ret);
8732d44f
MX
1581 goto fail;
1582 }
0660b5af 1583
a22285a6
YZ
1584 key.offset = (u64)-1;
1585 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
79787eaa
JM
1586 if (IS_ERR(pending->snap)) {
1587 ret = PTR_ERR(pending->snap);
66642832 1588 btrfs_abort_transaction(trans, ret);
8732d44f 1589 goto fail;
79787eaa 1590 }
d68fc57b 1591
49b25e05 1592 ret = btrfs_reloc_post_snapshot(trans, pending);
8732d44f 1593 if (ret) {
66642832 1594 btrfs_abort_transaction(trans, ret);
8732d44f
MX
1595 goto fail;
1596 }
361048f5
MX
1597
1598 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
8732d44f 1599 if (ret) {
66642832 1600 btrfs_abort_transaction(trans, ret);
8732d44f
MX
1601 goto fail;
1602 }
42874b3d 1603
6426c7ad
QW
1604 /*
1605 * Do special qgroup accounting for snapshot, as we do some qgroup
1606 * snapshot hack to do fast snapshot.
1607 * To co-operate with that hack, we do hack again.
1608 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1609 */
1610 ret = qgroup_account_snapshot(trans, root, parent_root,
1611 pending->inherit, objectid);
1612 if (ret < 0)
1613 goto fail;
1614
42874b3d
MX
1615 ret = btrfs_insert_dir_item(trans, parent_root,
1616 dentry->d_name.name, dentry->d_name.len,
1617 parent_inode, &key,
1618 BTRFS_FT_DIR, index);
1619 /* We have check then name at the beginning, so it is impossible. */
9c52057c 1620 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
8732d44f 1621 if (ret) {
66642832 1622 btrfs_abort_transaction(trans, ret);
8732d44f
MX
1623 goto fail;
1624 }
42874b3d
MX
1625
1626 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1627 dentry->d_name.len * 2);
04b285f3 1628 parent_inode->i_mtime = parent_inode->i_ctime =
c2050a45 1629 current_time(parent_inode);
be6aef60 1630 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
dd5f9615 1631 if (ret) {
66642832 1632 btrfs_abort_transaction(trans, ret);
dd5f9615
SB
1633 goto fail;
1634 }
1635 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b,
1636 BTRFS_UUID_KEY_SUBVOL, objectid);
1637 if (ret) {
66642832 1638 btrfs_abort_transaction(trans, ret);
dd5f9615
SB
1639 goto fail;
1640 }
1641 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1642 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
1643 new_root_item->received_uuid,
1644 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1645 objectid);
1646 if (ret && ret != -EEXIST) {
66642832 1647 btrfs_abort_transaction(trans, ret);
dd5f9615
SB
1648 goto fail;
1649 }
1650 }
d6726335
QW
1651
1652 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1653 if (ret) {
66642832 1654 btrfs_abort_transaction(trans, ret);
d6726335
QW
1655 goto fail;
1656 }
1657
3063d29f 1658fail:
aec8030a
MX
1659 pending->error = ret;
1660dir_item_existed:
98c9942a 1661 trans->block_rsv = rsv;
2382c5cc 1662 trans->bytes_reserved = 0;
d6726335
QW
1663clear_skip_qgroup:
1664 btrfs_clear_skip_qgroup(trans);
6fa9700e
MX
1665no_free_objectid:
1666 kfree(new_root_item);
b0c0ea63 1667 pending->root_item = NULL;
42874b3d 1668 btrfs_free_path(path);
8546b570
DS
1669 pending->path = NULL;
1670
49b25e05 1671 return ret;
3063d29f
CM
1672}
1673
d352ac68
CM
1674/*
1675 * create all the snapshots we've scheduled for creation
1676 */
80b6794d
CM
1677static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1678 struct btrfs_fs_info *fs_info)
3de4586c 1679{
aec8030a 1680 struct btrfs_pending_snapshot *pending, *next;
3de4586c 1681 struct list_head *head = &trans->transaction->pending_snapshots;
aec8030a 1682 int ret = 0;
3de4586c 1683
aec8030a
MX
1684 list_for_each_entry_safe(pending, next, head, list) {
1685 list_del(&pending->list);
1686 ret = create_pending_snapshot(trans, fs_info, pending);
1687 if (ret)
1688 break;
1689 }
1690 return ret;
3de4586c
CM
1691}
1692
5d4f98a2
YZ
1693static void update_super_roots(struct btrfs_root *root)
1694{
1695 struct btrfs_root_item *root_item;
1696 struct btrfs_super_block *super;
1697
6c41761f 1698 super = root->fs_info->super_copy;
5d4f98a2
YZ
1699
1700 root_item = &root->fs_info->chunk_root->root_item;
1701 super->chunk_root = root_item->bytenr;
1702 super->chunk_root_generation = root_item->generation;
1703 super->chunk_root_level = root_item->level;
1704
1705 root_item = &root->fs_info->tree_root->root_item;
1706 super->root = root_item->bytenr;
1707 super->generation = root_item->generation;
1708 super->root_level = root_item->level;
3cdde224 1709 if (btrfs_test_opt(root->fs_info, SPACE_CACHE))
0af3d00b 1710 super->cache_generation = root_item->generation;
afcdd129 1711 if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &root->fs_info->flags))
70f80175 1712 super->uuid_tree_generation = root_item->generation;
5d4f98a2
YZ
1713}
1714
f36f3042
CM
1715int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1716{
4a9d8bde 1717 struct btrfs_transaction *trans;
f36f3042 1718 int ret = 0;
4a9d8bde 1719
a4abeea4 1720 spin_lock(&info->trans_lock);
4a9d8bde
MX
1721 trans = info->running_transaction;
1722 if (trans)
1723 ret = (trans->state >= TRANS_STATE_COMMIT_START);
a4abeea4 1724 spin_unlock(&info->trans_lock);
f36f3042
CM
1725 return ret;
1726}
1727
8929ecfa
YZ
1728int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1729{
4a9d8bde 1730 struct btrfs_transaction *trans;
8929ecfa 1731 int ret = 0;
4a9d8bde 1732
a4abeea4 1733 spin_lock(&info->trans_lock);
4a9d8bde
MX
1734 trans = info->running_transaction;
1735 if (trans)
1736 ret = is_transaction_blocked(trans);
a4abeea4 1737 spin_unlock(&info->trans_lock);
8929ecfa
YZ
1738 return ret;
1739}
1740
bb9c12c9
SW
1741/*
1742 * wait for the current transaction commit to start and block subsequent
1743 * transaction joins
1744 */
1745static void wait_current_trans_commit_start(struct btrfs_root *root,
1746 struct btrfs_transaction *trans)
1747{
4a9d8bde 1748 wait_event(root->fs_info->transaction_blocked_wait,
501407aa
JB
1749 trans->state >= TRANS_STATE_COMMIT_START ||
1750 trans->aborted);
bb9c12c9
SW
1751}
1752
1753/*
1754 * wait for the current transaction to start and then become unblocked.
1755 * caller holds ref.
1756 */
1757static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1758 struct btrfs_transaction *trans)
1759{
72d63ed6 1760 wait_event(root->fs_info->transaction_wait,
501407aa
JB
1761 trans->state >= TRANS_STATE_UNBLOCKED ||
1762 trans->aborted);
bb9c12c9
SW
1763}
1764
1765/*
1766 * commit transactions asynchronously. once btrfs_commit_transaction_async
1767 * returns, any subsequent transaction will not be allowed to join.
1768 */
1769struct btrfs_async_commit {
1770 struct btrfs_trans_handle *newtrans;
1771 struct btrfs_root *root;
7892b5af 1772 struct work_struct work;
bb9c12c9
SW
1773};
1774
1775static void do_async_commit(struct work_struct *work)
1776{
1777 struct btrfs_async_commit *ac =
7892b5af 1778 container_of(work, struct btrfs_async_commit, work);
bb9c12c9 1779
6fc4e354
SW
1780 /*
1781 * We've got freeze protection passed with the transaction.
1782 * Tell lockdep about it.
1783 */
b1a06a4b 1784 if (ac->newtrans->type & __TRANS_FREEZABLE)
bee9182d 1785 __sb_writers_acquired(ac->root->fs_info->sb, SB_FREEZE_FS);
6fc4e354 1786
e209db7a
SW
1787 current->journal_info = ac->newtrans;
1788
bb9c12c9
SW
1789 btrfs_commit_transaction(ac->newtrans, ac->root);
1790 kfree(ac);
1791}
1792
1793int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1794 struct btrfs_root *root,
1795 int wait_for_unblock)
1796{
1797 struct btrfs_async_commit *ac;
1798 struct btrfs_transaction *cur_trans;
1799
1800 ac = kmalloc(sizeof(*ac), GFP_NOFS);
db5b493a
TI
1801 if (!ac)
1802 return -ENOMEM;
bb9c12c9 1803
7892b5af 1804 INIT_WORK(&ac->work, do_async_commit);
bb9c12c9 1805 ac->root = root;
7a7eaa40 1806 ac->newtrans = btrfs_join_transaction(root);
3612b495
TI
1807 if (IS_ERR(ac->newtrans)) {
1808 int err = PTR_ERR(ac->newtrans);
1809 kfree(ac);
1810 return err;
1811 }
bb9c12c9
SW
1812
1813 /* take transaction reference */
bb9c12c9 1814 cur_trans = trans->transaction;
13c5a93e 1815 atomic_inc(&cur_trans->use_count);
bb9c12c9
SW
1816
1817 btrfs_end_transaction(trans, root);
6fc4e354
SW
1818
1819 /*
1820 * Tell lockdep we've released the freeze rwsem, since the
1821 * async commit thread will be the one to unlock it.
1822 */
b1a06a4b 1823 if (ac->newtrans->type & __TRANS_FREEZABLE)
bee9182d 1824 __sb_writers_release(root->fs_info->sb, SB_FREEZE_FS);
6fc4e354 1825
7892b5af 1826 schedule_work(&ac->work);
bb9c12c9
SW
1827
1828 /* wait for transaction to start and unblock */
bb9c12c9
SW
1829 if (wait_for_unblock)
1830 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1831 else
1832 wait_current_trans_commit_start(root, cur_trans);
bb9c12c9 1833
38e88054
SW
1834 if (current->journal_info == trans)
1835 current->journal_info = NULL;
1836
724e2315 1837 btrfs_put_transaction(cur_trans);
bb9c12c9
SW
1838 return 0;
1839}
1840
49b25e05
JM
1841
1842static void cleanup_transaction(struct btrfs_trans_handle *trans,
7b8b92af 1843 struct btrfs_root *root, int err)
49b25e05
JM
1844{
1845 struct btrfs_transaction *cur_trans = trans->transaction;
f094ac32 1846 DEFINE_WAIT(wait);
49b25e05
JM
1847
1848 WARN_ON(trans->use_count > 1);
1849
66642832 1850 btrfs_abort_transaction(trans, err);
7b8b92af 1851
49b25e05 1852 spin_lock(&root->fs_info->trans_lock);
66b6135b 1853
25d8c284
MX
1854 /*
1855 * If the transaction is removed from the list, it means this
1856 * transaction has been committed successfully, so it is impossible
1857 * to call the cleanup function.
1858 */
1859 BUG_ON(list_empty(&cur_trans->list));
66b6135b 1860
49b25e05 1861 list_del_init(&cur_trans->list);
d7096fc3 1862 if (cur_trans == root->fs_info->running_transaction) {
4a9d8bde 1863 cur_trans->state = TRANS_STATE_COMMIT_DOING;
f094ac32
LB
1864 spin_unlock(&root->fs_info->trans_lock);
1865 wait_event(cur_trans->writer_wait,
1866 atomic_read(&cur_trans->num_writers) == 1);
1867
1868 spin_lock(&root->fs_info->trans_lock);
d7096fc3 1869 }
49b25e05
JM
1870 spin_unlock(&root->fs_info->trans_lock);
1871
1872 btrfs_cleanup_one_transaction(trans->transaction, root);
1873
4a9d8bde
MX
1874 spin_lock(&root->fs_info->trans_lock);
1875 if (cur_trans == root->fs_info->running_transaction)
1876 root->fs_info->running_transaction = NULL;
1877 spin_unlock(&root->fs_info->trans_lock);
1878
e0228285
JB
1879 if (trans->type & __TRANS_FREEZABLE)
1880 sb_end_intwrite(root->fs_info->sb);
724e2315
JB
1881 btrfs_put_transaction(cur_trans);
1882 btrfs_put_transaction(cur_trans);
49b25e05
JM
1883
1884 trace_btrfs_transaction_commit(root);
1885
49b25e05
JM
1886 if (current->journal_info == trans)
1887 current->journal_info = NULL;
c0af8f0b 1888 btrfs_scrub_cancel(root->fs_info);
49b25e05
JM
1889
1890 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1891}
1892
82436617
MX
1893static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1894{
3cdde224 1895 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
6c255e67 1896 return btrfs_start_delalloc_roots(fs_info, 1, -1);
82436617
MX
1897 return 0;
1898}
1899
1900static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1901{
3cdde224 1902 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
578def7c 1903 btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
82436617
MX
1904}
1905
50d9aa99 1906static inline void
161c3549 1907btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans)
50d9aa99 1908{
161c3549
JB
1909 wait_event(cur_trans->pending_wait,
1910 atomic_read(&cur_trans->pending_ordered) == 0);
50d9aa99
JB
1911}
1912
79154b1b
CM
1913int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1914 struct btrfs_root *root)
1915{
49b25e05 1916 struct btrfs_transaction *cur_trans = trans->transaction;
8fd17795 1917 struct btrfs_transaction *prev_trans = NULL;
25287e0a 1918 int ret;
79154b1b 1919
8d25a086
MX
1920 /* Stop the commit early if ->aborted is set */
1921 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
25287e0a 1922 ret = cur_trans->aborted;
e4a2bcac
JB
1923 btrfs_end_transaction(trans, root);
1924 return ret;
25287e0a 1925 }
49b25e05 1926
56bec294
CM
1927 /* make a pass through all the delayed refs we have so far
1928 * any runnings procs may add more while we are here
1929 */
1930 ret = btrfs_run_delayed_refs(trans, root, 0);
e4a2bcac
JB
1931 if (ret) {
1932 btrfs_end_transaction(trans, root);
1933 return ret;
1934 }
56bec294 1935
0e721106
JB
1936 btrfs_trans_release_metadata(trans, root);
1937 trans->block_rsv = NULL;
1938
b7ec40d7 1939 cur_trans = trans->transaction;
49b25e05 1940
56bec294
CM
1941 /*
1942 * set the flushing flag so procs in this transaction have to
1943 * start sending their work down.
1944 */
b7ec40d7 1945 cur_trans->delayed_refs.flushing = 1;
1be41b78 1946 smp_wmb();
56bec294 1947
ea658bad
JB
1948 if (!list_empty(&trans->new_bgs))
1949 btrfs_create_pending_block_groups(trans, root);
1950
c3e69d58 1951 ret = btrfs_run_delayed_refs(trans, root, 0);
e4a2bcac
JB
1952 if (ret) {
1953 btrfs_end_transaction(trans, root);
1954 return ret;
1955 }
56bec294 1956
3204d33c 1957 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
1bbc621e
CM
1958 int run_it = 0;
1959
1960 /* this mutex is also taken before trying to set
1961 * block groups readonly. We need to make sure
1962 * that nobody has set a block group readonly
1963 * after a extents from that block group have been
1964 * allocated for cache files. btrfs_set_block_group_ro
1965 * will wait for the transaction to commit if it
3204d33c 1966 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
1bbc621e 1967 *
3204d33c
JB
1968 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
1969 * only one process starts all the block group IO. It wouldn't
1bbc621e
CM
1970 * hurt to have more than one go through, but there's no
1971 * real advantage to it either.
1972 */
1973 mutex_lock(&root->fs_info->ro_block_group_mutex);
3204d33c
JB
1974 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
1975 &cur_trans->flags))
1bbc621e 1976 run_it = 1;
1bbc621e
CM
1977 mutex_unlock(&root->fs_info->ro_block_group_mutex);
1978
1979 if (run_it)
1980 ret = btrfs_start_dirty_block_groups(trans, root);
1981 }
1982 if (ret) {
1983 btrfs_end_transaction(trans, root);
1984 return ret;
1985 }
1986
4a9d8bde
MX
1987 spin_lock(&root->fs_info->trans_lock);
1988 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1989 spin_unlock(&root->fs_info->trans_lock);
13c5a93e 1990 atomic_inc(&cur_trans->use_count);
49b25e05 1991 ret = btrfs_end_transaction(trans, root);
ccd467d6 1992
b9c8300c 1993 wait_for_commit(root, cur_trans);
15ee9bc7 1994
b4924a0f
LB
1995 if (unlikely(cur_trans->aborted))
1996 ret = cur_trans->aborted;
1997
724e2315 1998 btrfs_put_transaction(cur_trans);
15ee9bc7 1999
49b25e05 2000 return ret;
79154b1b 2001 }
4313b399 2002
4a9d8bde 2003 cur_trans->state = TRANS_STATE_COMMIT_START;
bb9c12c9
SW
2004 wake_up(&root->fs_info->transaction_blocked_wait);
2005
ccd467d6
CM
2006 if (cur_trans->list.prev != &root->fs_info->trans_list) {
2007 prev_trans = list_entry(cur_trans->list.prev,
2008 struct btrfs_transaction, list);
4a9d8bde 2009 if (prev_trans->state != TRANS_STATE_COMPLETED) {
13c5a93e 2010 atomic_inc(&prev_trans->use_count);
a4abeea4 2011 spin_unlock(&root->fs_info->trans_lock);
ccd467d6
CM
2012
2013 wait_for_commit(root, prev_trans);
1f9b8c8f 2014 ret = prev_trans->aborted;
ccd467d6 2015
724e2315 2016 btrfs_put_transaction(prev_trans);
1f9b8c8f
FM
2017 if (ret)
2018 goto cleanup_transaction;
a4abeea4
JB
2019 } else {
2020 spin_unlock(&root->fs_info->trans_lock);
ccd467d6 2021 }
a4abeea4
JB
2022 } else {
2023 spin_unlock(&root->fs_info->trans_lock);
ccd467d6 2024 }
15ee9bc7 2025
0860adfd
MX
2026 extwriter_counter_dec(cur_trans, trans->type);
2027
82436617
MX
2028 ret = btrfs_start_delalloc_flush(root->fs_info);
2029 if (ret)
2030 goto cleanup_transaction;
2031
8d875f95 2032 ret = btrfs_run_delayed_items(trans, root);
581227d0
MX
2033 if (ret)
2034 goto cleanup_transaction;
15ee9bc7 2035
581227d0
MX
2036 wait_event(cur_trans->writer_wait,
2037 extwriter_counter_read(cur_trans) == 0);
15ee9bc7 2038
581227d0 2039 /* some pending stuffs might be added after the previous flush. */
8d875f95 2040 ret = btrfs_run_delayed_items(trans, root);
ca469637
MX
2041 if (ret)
2042 goto cleanup_transaction;
2043
82436617 2044 btrfs_wait_delalloc_flush(root->fs_info);
cb7ab021 2045
161c3549 2046 btrfs_wait_pending_ordered(cur_trans);
50d9aa99 2047
cb7ab021 2048 btrfs_scrub_pause(root);
ed0ca140
JB
2049 /*
2050 * Ok now we need to make sure to block out any other joins while we
2051 * commit the transaction. We could have started a join before setting
4a9d8bde 2052 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
ed0ca140
JB
2053 */
2054 spin_lock(&root->fs_info->trans_lock);
4a9d8bde 2055 cur_trans->state = TRANS_STATE_COMMIT_DOING;
ed0ca140
JB
2056 spin_unlock(&root->fs_info->trans_lock);
2057 wait_event(cur_trans->writer_wait,
2058 atomic_read(&cur_trans->num_writers) == 1);
2059
2cba30f1
MX
2060 /* ->aborted might be set after the previous check, so check it */
2061 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
2062 ret = cur_trans->aborted;
6cf7f77e 2063 goto scrub_continue;
2cba30f1 2064 }
7585717f
CM
2065 /*
2066 * the reloc mutex makes sure that we stop
2067 * the balancing code from coming in and moving
2068 * extents around in the middle of the commit
2069 */
2070 mutex_lock(&root->fs_info->reloc_mutex);
2071
42874b3d
MX
2072 /*
2073 * We needn't worry about the delayed items because we will
2074 * deal with them in create_pending_snapshot(), which is the
2075 * core function of the snapshot creation.
2076 */
2077 ret = create_pending_snapshots(trans, root->fs_info);
49b25e05
JM
2078 if (ret) {
2079 mutex_unlock(&root->fs_info->reloc_mutex);
6cf7f77e 2080 goto scrub_continue;
49b25e05 2081 }
3063d29f 2082
42874b3d
MX
2083 /*
2084 * We insert the dir indexes of the snapshots and update the inode
2085 * of the snapshots' parents after the snapshot creation, so there
2086 * are some delayed items which are not dealt with. Now deal with
2087 * them.
2088 *
2089 * We needn't worry that this operation will corrupt the snapshots,
2090 * because all the tree which are snapshoted will be forced to COW
2091 * the nodes and leaves.
2092 */
2093 ret = btrfs_run_delayed_items(trans, root);
49b25e05
JM
2094 if (ret) {
2095 mutex_unlock(&root->fs_info->reloc_mutex);
6cf7f77e 2096 goto scrub_continue;
49b25e05 2097 }
16cdcec7 2098
56bec294 2099 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
49b25e05
JM
2100 if (ret) {
2101 mutex_unlock(&root->fs_info->reloc_mutex);
6cf7f77e 2102 goto scrub_continue;
49b25e05 2103 }
56bec294 2104
0ed4792a
QW
2105 /* Reocrd old roots for later qgroup accounting */
2106 ret = btrfs_qgroup_prepare_account_extents(trans, root->fs_info);
2107 if (ret) {
2108 mutex_unlock(&root->fs_info->reloc_mutex);
2109 goto scrub_continue;
2110 }
2111
e999376f
CM
2112 /*
2113 * make sure none of the code above managed to slip in a
2114 * delayed item
2115 */
2116 btrfs_assert_delayed_root_empty(root);
2117
2c90e5d6 2118 WARN_ON(cur_trans != trans->transaction);
dc17ff8f 2119
e02119d5
CM
2120 /* btrfs_commit_tree_roots is responsible for getting the
2121 * various roots consistent with each other. Every pointer
2122 * in the tree of tree roots has to point to the most up to date
2123 * root for every subvolume and other tree. So, we have to keep
2124 * the tree logging code from jumping in and changing any
2125 * of the trees.
2126 *
2127 * At this point in the commit, there can't be any tree-log
2128 * writers, but a little lower down we drop the trans mutex
2129 * and let new people in. By holding the tree_log_mutex
2130 * from now until after the super is written, we avoid races
2131 * with the tree-log code.
2132 */
2133 mutex_lock(&root->fs_info->tree_log_mutex);
2134
5d4f98a2 2135 ret = commit_fs_roots(trans, root);
49b25e05
JM
2136 if (ret) {
2137 mutex_unlock(&root->fs_info->tree_log_mutex);
871383be 2138 mutex_unlock(&root->fs_info->reloc_mutex);
6cf7f77e 2139 goto scrub_continue;
49b25e05 2140 }
54aa1f4d 2141
3818aea2 2142 /*
7e1876ac
DS
2143 * Since the transaction is done, we can apply the pending changes
2144 * before the next transaction.
3818aea2 2145 */
572d9ab7 2146 btrfs_apply_pending_changes(root->fs_info);
3818aea2 2147
5d4f98a2 2148 /* commit_fs_roots gets rid of all the tree log roots, it is now
e02119d5
CM
2149 * safe to free the root of tree log roots
2150 */
2151 btrfs_free_log_root_tree(trans, root->fs_info);
2152
0ed4792a
QW
2153 /*
2154 * Since fs roots are all committed, we can get a quite accurate
2155 * new_roots. So let's do quota accounting.
2156 */
2157 ret = btrfs_qgroup_account_extents(trans, root->fs_info);
2158 if (ret < 0) {
2159 mutex_unlock(&root->fs_info->tree_log_mutex);
2160 mutex_unlock(&root->fs_info->reloc_mutex);
2161 goto scrub_continue;
2162 }
2163
5d4f98a2 2164 ret = commit_cowonly_roots(trans, root);
49b25e05
JM
2165 if (ret) {
2166 mutex_unlock(&root->fs_info->tree_log_mutex);
871383be 2167 mutex_unlock(&root->fs_info->reloc_mutex);
6cf7f77e 2168 goto scrub_continue;
49b25e05 2169 }
54aa1f4d 2170
2cba30f1
MX
2171 /*
2172 * The tasks which save the space cache and inode cache may also
2173 * update ->aborted, check it.
2174 */
2175 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
2176 ret = cur_trans->aborted;
2177 mutex_unlock(&root->fs_info->tree_log_mutex);
2178 mutex_unlock(&root->fs_info->reloc_mutex);
6cf7f77e 2179 goto scrub_continue;
2cba30f1
MX
2180 }
2181
11833d66
YZ
2182 btrfs_prepare_extent_commit(trans, root);
2183
78fae27e 2184 cur_trans = root->fs_info->running_transaction;
5d4f98a2
YZ
2185
2186 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
2187 root->fs_info->tree_root->node);
9e351cc8
JB
2188 list_add_tail(&root->fs_info->tree_root->dirty_list,
2189 &cur_trans->switch_commits);
5d4f98a2
YZ
2190
2191 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
2192 root->fs_info->chunk_root->node);
9e351cc8
JB
2193 list_add_tail(&root->fs_info->chunk_root->dirty_list,
2194 &cur_trans->switch_commits);
2195
2196 switch_commit_roots(cur_trans, root->fs_info);
5d4f98a2 2197
edf39272 2198 assert_qgroups_uptodate(trans);
ce93ec54 2199 ASSERT(list_empty(&cur_trans->dirty_bgs));
1bbc621e 2200 ASSERT(list_empty(&cur_trans->io_bgs));
5d4f98a2 2201 update_super_roots(root);
e02119d5 2202
60e7cd3a
JB
2203 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
2204 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
6c41761f
DS
2205 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
2206 sizeof(*root->fs_info->super_copy));
ccd467d6 2207
935e5cc9 2208 btrfs_update_commit_device_size(root->fs_info);
ce7213c7 2209 btrfs_update_commit_device_bytes_used(root, cur_trans);
935e5cc9 2210
afcdd129
JB
2211 clear_bit(BTRFS_FS_LOG1_ERR, &root->fs_info->flags);
2212 clear_bit(BTRFS_FS_LOG2_ERR, &root->fs_info->flags);
656f30db 2213
4fbcdf66
FM
2214 btrfs_trans_release_chunk_metadata(trans);
2215
a4abeea4 2216 spin_lock(&root->fs_info->trans_lock);
4a9d8bde 2217 cur_trans->state = TRANS_STATE_UNBLOCKED;
a4abeea4 2218 root->fs_info->running_transaction = NULL;
a4abeea4 2219 spin_unlock(&root->fs_info->trans_lock);
7585717f 2220 mutex_unlock(&root->fs_info->reloc_mutex);
b7ec40d7 2221
f9295749 2222 wake_up(&root->fs_info->transaction_wait);
e6dcd2dc 2223
79154b1b 2224 ret = btrfs_write_and_wait_transaction(trans, root);
49b25e05 2225 if (ret) {
34d97007 2226 btrfs_handle_fs_error(root->fs_info, ret,
08748810 2227 "Error while writing out transaction");
49b25e05 2228 mutex_unlock(&root->fs_info->tree_log_mutex);
6cf7f77e 2229 goto scrub_continue;
49b25e05
JM
2230 }
2231
2232 ret = write_ctree_super(trans, root, 0);
2233 if (ret) {
2234 mutex_unlock(&root->fs_info->tree_log_mutex);
6cf7f77e 2235 goto scrub_continue;
49b25e05 2236 }
4313b399 2237
e02119d5
CM
2238 /*
2239 * the super is written, we can safely allow the tree-loggers
2240 * to go about their business
2241 */
2242 mutex_unlock(&root->fs_info->tree_log_mutex);
2243
11833d66 2244 btrfs_finish_extent_commit(trans, root);
4313b399 2245
3204d33c 2246 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
13212b54
ZL
2247 btrfs_clear_space_info_full(root->fs_info);
2248
15ee9bc7 2249 root->fs_info->last_trans_committed = cur_trans->transid;
4a9d8bde
MX
2250 /*
2251 * We needn't acquire the lock here because there is no other task
2252 * which can change it.
2253 */
2254 cur_trans->state = TRANS_STATE_COMPLETED;
2c90e5d6 2255 wake_up(&cur_trans->commit_wait);
3de4586c 2256
a4abeea4 2257 spin_lock(&root->fs_info->trans_lock);
13c5a93e 2258 list_del_init(&cur_trans->list);
a4abeea4
JB
2259 spin_unlock(&root->fs_info->trans_lock);
2260
724e2315
JB
2261 btrfs_put_transaction(cur_trans);
2262 btrfs_put_transaction(cur_trans);
58176a96 2263
0860adfd 2264 if (trans->type & __TRANS_FREEZABLE)
354aa0fb 2265 sb_end_intwrite(root->fs_info->sb);
b2b5ef5c 2266
1abe9b8a 2267 trace_btrfs_transaction_commit(root);
2268
a2de733c
AJ
2269 btrfs_scrub_continue(root);
2270
9ed74f2d
JB
2271 if (current->journal_info == trans)
2272 current->journal_info = NULL;
2273
2c90e5d6 2274 kmem_cache_free(btrfs_trans_handle_cachep, trans);
24bbcf04 2275
9e7cc91a
WX
2276 /*
2277 * If fs has been frozen, we can not handle delayed iputs, otherwise
2278 * it'll result in deadlock about SB_FREEZE_FS.
2279 */
8a733013 2280 if (current != root->fs_info->transaction_kthread &&
9e7cc91a
WX
2281 current != root->fs_info->cleaner_kthread &&
2282 !root->fs_info->fs_frozen)
24bbcf04
YZ
2283 btrfs_run_delayed_iputs(root);
2284
79154b1b 2285 return ret;
49b25e05 2286
6cf7f77e
WS
2287scrub_continue:
2288 btrfs_scrub_continue(root);
49b25e05 2289cleanup_transaction:
0e721106 2290 btrfs_trans_release_metadata(trans, root);
4fbcdf66 2291 btrfs_trans_release_chunk_metadata(trans);
0e721106 2292 trans->block_rsv = NULL;
c2cf52eb 2293 btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
49b25e05
JM
2294 if (current->journal_info == trans)
2295 current->journal_info = NULL;
7b8b92af 2296 cleanup_transaction(trans, root, ret);
49b25e05
JM
2297
2298 return ret;
79154b1b
CM
2299}
2300
d352ac68 2301/*
9d1a2a3a
DS
2302 * return < 0 if error
2303 * 0 if there are no more dead_roots at the time of call
2304 * 1 there are more to be processed, call me again
2305 *
2306 * The return value indicates there are certainly more snapshots to delete, but
2307 * if there comes a new one during processing, it may return 0. We don't mind,
2308 * because btrfs_commit_super will poke cleaner thread and it will process it a
2309 * few seconds later.
d352ac68 2310 */
9d1a2a3a 2311int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
e9d0b13b 2312{
9d1a2a3a 2313 int ret;
5d4f98a2
YZ
2314 struct btrfs_fs_info *fs_info = root->fs_info;
2315
a4abeea4 2316 spin_lock(&fs_info->trans_lock);
9d1a2a3a
DS
2317 if (list_empty(&fs_info->dead_roots)) {
2318 spin_unlock(&fs_info->trans_lock);
2319 return 0;
2320 }
2321 root = list_first_entry(&fs_info->dead_roots,
2322 struct btrfs_root, root_list);
cfad392b 2323 list_del_init(&root->root_list);
a4abeea4 2324 spin_unlock(&fs_info->trans_lock);
e9d0b13b 2325
ab8d0fc4 2326 btrfs_debug(fs_info, "cleaner removing %llu", root->objectid);
76dda93c 2327
9d1a2a3a 2328 btrfs_kill_all_delayed_nodes(root);
16cdcec7 2329
9d1a2a3a
DS
2330 if (btrfs_header_backref_rev(root->node) <
2331 BTRFS_MIXED_BACKREF_REV)
2332 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
2333 else
2334 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
32471dc2 2335
6596a928 2336 return (ret < 0) ? 0 : 1;
e9d0b13b 2337}
572d9ab7
DS
2338
2339void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2340{
2341 unsigned long prev;
2342 unsigned long bit;
2343
6c9fe14f 2344 prev = xchg(&fs_info->pending_changes, 0);
572d9ab7
DS
2345 if (!prev)
2346 return;
2347
7e1876ac
DS
2348 bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE;
2349 if (prev & bit)
2350 btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2351 prev &= ~bit;
2352
2353 bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE;
2354 if (prev & bit)
2355 btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2356 prev &= ~bit;
2357
d51033d0
DS
2358 bit = 1 << BTRFS_PENDING_COMMIT;
2359 if (prev & bit)
2360 btrfs_debug(fs_info, "pending commit done");
2361 prev &= ~bit;
2362
572d9ab7
DS
2363 if (prev)
2364 btrfs_warn(fs_info,
2365 "unknown pending changes left 0x%lx, ignoring", prev);
2366}