]>
Commit | Line | Data |
---|---|---|
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 | 38 | static 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 | 61 | void 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 |
100 | static 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 |
129 | static 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 |
159 | static 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 | ||
166 | static 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 | ||
173 | static 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 | ||
179 | static 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 | 187 | static 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 | 193 | loop: |
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 | 313 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
6426c7ad QW |
314 | struct btrfs_root *root, |
315 | int force) | |
6702ed49 | 316 | { |
0b246afa JM |
317 | struct btrfs_fs_info *fs_info = root->fs_info; |
318 | ||
6426c7ad QW |
319 | if ((test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
320 | root->last_trans < trans->transid) || force) { | |
0b246afa | 321 | WARN_ON(root == fs_info->extent_root); |
5d4f98a2 YZ |
322 | WARN_ON(root->commit_root != root->node); |
323 | ||
7585717f | 324 | /* |
27cdeb70 | 325 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
326 | * we have the reloc mutex held now, so there |
327 | * is only one writer in this function | |
328 | */ | |
27cdeb70 | 329 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 330 | |
27cdeb70 | 331 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
332 | * they find our root->last_trans update |
333 | */ | |
334 | smp_wmb(); | |
335 | ||
0b246afa | 336 | spin_lock(&fs_info->fs_roots_radix_lock); |
6426c7ad | 337 | if (root->last_trans == trans->transid && !force) { |
0b246afa | 338 | spin_unlock(&fs_info->fs_roots_radix_lock); |
a4abeea4 JB |
339 | return 0; |
340 | } | |
0b246afa JM |
341 | radix_tree_tag_set(&fs_info->fs_roots_radix, |
342 | (unsigned long)root->root_key.objectid, | |
343 | BTRFS_ROOT_TRANS_TAG); | |
344 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
7585717f CM |
345 | root->last_trans = trans->transid; |
346 | ||
347 | /* this is pretty tricky. We don't want to | |
348 | * take the relocation lock in btrfs_record_root_in_trans | |
349 | * unless we're really doing the first setup for this root in | |
350 | * this transaction. | |
351 | * | |
352 | * Normally we'd use root->last_trans as a flag to decide | |
353 | * if we want to take the expensive mutex. | |
354 | * | |
355 | * But, we have to set root->last_trans before we | |
356 | * init the relocation root, otherwise, we trip over warnings | |
357 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 358 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
359 | * fixing up the reloc trees and everyone must wait. |
360 | * | |
361 | * When this is zero, they can trust root->last_trans and fly | |
362 | * through btrfs_record_root_in_trans without having to take the | |
363 | * lock. smp_wmb() makes sure that all the writes above are | |
364 | * done before we pop in the zero below | |
365 | */ | |
5d4f98a2 | 366 | btrfs_init_reloc_root(trans, root); |
c7548af6 | 367 | smp_mb__before_atomic(); |
27cdeb70 | 368 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 YZ |
369 | } |
370 | return 0; | |
371 | } | |
bcc63abb | 372 | |
7585717f | 373 | |
2b9dbef2 JB |
374 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
375 | struct btrfs_root *root) | |
376 | { | |
0b246afa | 377 | struct btrfs_fs_info *fs_info = root->fs_info; |
2b9dbef2 JB |
378 | struct btrfs_transaction *cur_trans = trans->transaction; |
379 | ||
380 | /* Add ourselves to the transaction dropped list */ | |
381 | spin_lock(&cur_trans->dropped_roots_lock); | |
382 | list_add_tail(&root->root_list, &cur_trans->dropped_roots); | |
383 | spin_unlock(&cur_trans->dropped_roots_lock); | |
384 | ||
385 | /* Make sure we don't try to update the root at commit time */ | |
0b246afa JM |
386 | spin_lock(&fs_info->fs_roots_radix_lock); |
387 | radix_tree_tag_clear(&fs_info->fs_roots_radix, | |
2b9dbef2 JB |
388 | (unsigned long)root->root_key.objectid, |
389 | BTRFS_ROOT_TRANS_TAG); | |
0b246afa | 390 | spin_unlock(&fs_info->fs_roots_radix_lock); |
2b9dbef2 JB |
391 | } |
392 | ||
7585717f CM |
393 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
394 | struct btrfs_root *root) | |
395 | { | |
0b246afa JM |
396 | struct btrfs_fs_info *fs_info = root->fs_info; |
397 | ||
27cdeb70 | 398 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
7585717f CM |
399 | return 0; |
400 | ||
401 | /* | |
27cdeb70 | 402 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
403 | * and barriers |
404 | */ | |
405 | smp_rmb(); | |
406 | if (root->last_trans == trans->transid && | |
27cdeb70 | 407 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
408 | return 0; |
409 | ||
0b246afa | 410 | mutex_lock(&fs_info->reloc_mutex); |
6426c7ad | 411 | record_root_in_trans(trans, root, 0); |
0b246afa | 412 | mutex_unlock(&fs_info->reloc_mutex); |
7585717f CM |
413 | |
414 | return 0; | |
415 | } | |
416 | ||
4a9d8bde MX |
417 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
418 | { | |
419 | return (trans->state >= TRANS_STATE_BLOCKED && | |
501407aa JB |
420 | trans->state < TRANS_STATE_UNBLOCKED && |
421 | !trans->aborted); | |
4a9d8bde MX |
422 | } |
423 | ||
d352ac68 CM |
424 | /* wait for commit against the current transaction to become unblocked |
425 | * when this is done, it is safe to start a new transaction, but the current | |
426 | * transaction might not be fully on disk. | |
427 | */ | |
37d1aeee | 428 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 429 | { |
0b246afa | 430 | struct btrfs_fs_info *fs_info = root->fs_info; |
f9295749 | 431 | struct btrfs_transaction *cur_trans; |
79154b1b | 432 | |
0b246afa JM |
433 | spin_lock(&fs_info->trans_lock); |
434 | cur_trans = fs_info->running_transaction; | |
4a9d8bde | 435 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
13c5a93e | 436 | atomic_inc(&cur_trans->use_count); |
0b246afa | 437 | spin_unlock(&fs_info->trans_lock); |
72d63ed6 | 438 | |
0b246afa | 439 | wait_event(fs_info->transaction_wait, |
501407aa JB |
440 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
441 | cur_trans->aborted); | |
724e2315 | 442 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 443 | } else { |
0b246afa | 444 | spin_unlock(&fs_info->trans_lock); |
f9295749 | 445 | } |
37d1aeee CM |
446 | } |
447 | ||
a22285a6 YZ |
448 | static int may_wait_transaction(struct btrfs_root *root, int type) |
449 | { | |
0b246afa JM |
450 | struct btrfs_fs_info *fs_info = root->fs_info; |
451 | ||
452 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) | |
a4abeea4 JB |
453 | return 0; |
454 | ||
455 | if (type == TRANS_USERSPACE) | |
456 | return 1; | |
457 | ||
458 | if (type == TRANS_START && | |
0b246afa | 459 | !atomic_read(&fs_info->open_ioctl_trans)) |
a22285a6 | 460 | return 1; |
a4abeea4 | 461 | |
a22285a6 YZ |
462 | return 0; |
463 | } | |
464 | ||
20dd2cbf MX |
465 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
466 | { | |
0b246afa JM |
467 | struct btrfs_fs_info *fs_info = root->fs_info; |
468 | ||
469 | if (!fs_info->reloc_ctl || | |
27cdeb70 | 470 | !test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
20dd2cbf MX |
471 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
472 | root->reloc_root) | |
473 | return false; | |
474 | ||
475 | return true; | |
476 | } | |
477 | ||
08e007d2 | 478 | static struct btrfs_trans_handle * |
5aed1dd8 AM |
479 | start_transaction(struct btrfs_root *root, unsigned int num_items, |
480 | unsigned int type, enum btrfs_reserve_flush_enum flush) | |
37d1aeee | 481 | { |
0b246afa JM |
482 | struct btrfs_fs_info *fs_info = root->fs_info; |
483 | ||
a22285a6 YZ |
484 | struct btrfs_trans_handle *h; |
485 | struct btrfs_transaction *cur_trans; | |
b5009945 | 486 | u64 num_bytes = 0; |
c5567237 | 487 | u64 qgroup_reserved = 0; |
20dd2cbf MX |
488 | bool reloc_reserved = false; |
489 | int ret; | |
acce952b | 490 | |
46c4e71e | 491 | /* Send isn't supposed to start transactions. */ |
2755a0de | 492 | ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); |
46c4e71e | 493 | |
0b246afa | 494 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
acce952b | 495 | return ERR_PTR(-EROFS); |
2a1eb461 | 496 | |
46c4e71e | 497 | if (current->journal_info) { |
0860adfd | 498 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 JB |
499 | h = current->journal_info; |
500 | h->use_count++; | |
b7d5b0a8 | 501 | WARN_ON(h->use_count > 2); |
2a1eb461 JB |
502 | h->orig_rsv = h->block_rsv; |
503 | h->block_rsv = NULL; | |
504 | goto got_it; | |
505 | } | |
b5009945 JB |
506 | |
507 | /* | |
508 | * Do the reservation before we join the transaction so we can do all | |
509 | * the appropriate flushing if need be. | |
510 | */ | |
0b246afa JM |
511 | if (num_items > 0 && root != fs_info->chunk_root) { |
512 | qgroup_reserved = num_items * fs_info->nodesize; | |
7174109c QW |
513 | ret = btrfs_qgroup_reserve_meta(root, qgroup_reserved); |
514 | if (ret) | |
515 | return ERR_PTR(ret); | |
c5567237 | 516 | |
0b246afa | 517 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, num_items); |
20dd2cbf MX |
518 | /* |
519 | * Do the reservation for the relocation root creation | |
520 | */ | |
ee39b432 | 521 | if (need_reserve_reloc_root(root)) { |
0b246afa | 522 | num_bytes += fs_info->nodesize; |
20dd2cbf MX |
523 | reloc_reserved = true; |
524 | } | |
525 | ||
0b246afa | 526 | ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv, |
08e007d2 | 527 | num_bytes, flush); |
b5009945 | 528 | if (ret) |
843fcf35 | 529 | goto reserve_fail; |
b5009945 | 530 | } |
a22285a6 | 531 | again: |
f2f767e7 | 532 | h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); |
843fcf35 MX |
533 | if (!h) { |
534 | ret = -ENOMEM; | |
535 | goto alloc_fail; | |
536 | } | |
37d1aeee | 537 | |
98114659 JB |
538 | /* |
539 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
540 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
541 | * because we're already holding a ref. We need this because we could | |
542 | * have raced in and did an fsync() on a file which can kick a commit | |
543 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
544 | * |
545 | * If we are ATTACH, it means we just want to catch the current | |
546 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 547 | */ |
0860adfd | 548 | if (type & __TRANS_FREEZABLE) |
0b246afa | 549 | sb_start_intwrite(fs_info->sb); |
b2b5ef5c | 550 | |
a22285a6 | 551 | if (may_wait_transaction(root, type)) |
37d1aeee | 552 | wait_current_trans(root); |
a22285a6 | 553 | |
a4abeea4 | 554 | do { |
354aa0fb | 555 | ret = join_transaction(root, type); |
178260b2 | 556 | if (ret == -EBUSY) { |
a4abeea4 | 557 | wait_current_trans(root); |
178260b2 MX |
558 | if (unlikely(type == TRANS_ATTACH)) |
559 | ret = -ENOENT; | |
560 | } | |
a4abeea4 JB |
561 | } while (ret == -EBUSY); |
562 | ||
a43f7f82 | 563 | if (ret < 0) |
843fcf35 | 564 | goto join_fail; |
0f7d52f4 | 565 | |
0b246afa | 566 | cur_trans = fs_info->running_transaction; |
a22285a6 YZ |
567 | |
568 | h->transid = cur_trans->transid; | |
569 | h->transaction = cur_trans; | |
d13603ef | 570 | h->root = root; |
2a1eb461 | 571 | h->use_count = 1; |
64b63580 | 572 | h->fs_info = root->fs_info; |
7174109c | 573 | |
a698d075 | 574 | h->type = type; |
d9a0540a | 575 | h->can_flush_pending_bgs = true; |
bed92eae | 576 | INIT_LIST_HEAD(&h->qgroup_ref_list); |
ea658bad | 577 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 578 | |
a22285a6 | 579 | smp_mb(); |
4a9d8bde MX |
580 | if (cur_trans->state >= TRANS_STATE_BLOCKED && |
581 | may_wait_transaction(root, type)) { | |
abdd2e80 | 582 | current->journal_info = h; |
a22285a6 YZ |
583 | btrfs_commit_transaction(h, root); |
584 | goto again; | |
585 | } | |
586 | ||
b5009945 | 587 | if (num_bytes) { |
0b246afa | 588 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 589 | h->transid, num_bytes, 1); |
0b246afa | 590 | h->block_rsv = &fs_info->trans_block_rsv; |
b5009945 | 591 | h->bytes_reserved = num_bytes; |
20dd2cbf | 592 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 593 | } |
9ed74f2d | 594 | |
2a1eb461 | 595 | got_it: |
a4abeea4 | 596 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
597 | |
598 | if (!current->journal_info && type != TRANS_USERSPACE) | |
599 | current->journal_info = h; | |
79154b1b | 600 | return h; |
843fcf35 MX |
601 | |
602 | join_fail: | |
0860adfd | 603 | if (type & __TRANS_FREEZABLE) |
0b246afa | 604 | sb_end_intwrite(fs_info->sb); |
843fcf35 MX |
605 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
606 | alloc_fail: | |
607 | if (num_bytes) | |
0b246afa | 608 | btrfs_block_rsv_release(root, &fs_info->trans_block_rsv, |
843fcf35 MX |
609 | num_bytes); |
610 | reserve_fail: | |
7174109c | 611 | btrfs_qgroup_free_meta(root, qgroup_reserved); |
843fcf35 | 612 | return ERR_PTR(ret); |
79154b1b CM |
613 | } |
614 | ||
f9295749 | 615 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
5aed1dd8 | 616 | unsigned int num_items) |
f9295749 | 617 | { |
08e007d2 MX |
618 | return start_transaction(root, num_items, TRANS_START, |
619 | BTRFS_RESERVE_FLUSH_ALL); | |
f9295749 | 620 | } |
8eab77ff FM |
621 | struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( |
622 | struct btrfs_root *root, | |
623 | unsigned int num_items, | |
624 | int min_factor) | |
625 | { | |
0b246afa | 626 | struct btrfs_fs_info *fs_info = root->fs_info; |
8eab77ff FM |
627 | struct btrfs_trans_handle *trans; |
628 | u64 num_bytes; | |
629 | int ret; | |
630 | ||
631 | trans = btrfs_start_transaction(root, num_items); | |
632 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) | |
633 | return trans; | |
634 | ||
635 | trans = btrfs_start_transaction(root, 0); | |
636 | if (IS_ERR(trans)) | |
637 | return trans; | |
638 | ||
0b246afa JM |
639 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, num_items); |
640 | ret = btrfs_cond_migrate_bytes(fs_info, &fs_info->trans_block_rsv, | |
641 | num_bytes, min_factor); | |
8eab77ff FM |
642 | if (ret) { |
643 | btrfs_end_transaction(trans, root); | |
644 | return ERR_PTR(ret); | |
645 | } | |
646 | ||
0b246afa | 647 | trans->block_rsv = &fs_info->trans_block_rsv; |
8eab77ff | 648 | trans->bytes_reserved = num_bytes; |
0b246afa | 649 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa | 650 | trans->transid, num_bytes, 1); |
8eab77ff FM |
651 | |
652 | return trans; | |
653 | } | |
8407aa46 | 654 | |
08e007d2 | 655 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
5aed1dd8 AM |
656 | struct btrfs_root *root, |
657 | unsigned int num_items) | |
8407aa46 | 658 | { |
08e007d2 MX |
659 | return start_transaction(root, num_items, TRANS_START, |
660 | BTRFS_RESERVE_FLUSH_LIMIT); | |
8407aa46 MX |
661 | } |
662 | ||
7a7eaa40 | 663 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 664 | { |
575a75d6 AM |
665 | return start_transaction(root, 0, TRANS_JOIN, |
666 | BTRFS_RESERVE_NO_FLUSH); | |
f9295749 CM |
667 | } |
668 | ||
7a7eaa40 | 669 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 670 | { |
575a75d6 AM |
671 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, |
672 | BTRFS_RESERVE_NO_FLUSH); | |
0af3d00b JB |
673 | } |
674 | ||
7a7eaa40 | 675 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 676 | { |
575a75d6 AM |
677 | return start_transaction(root, 0, TRANS_USERSPACE, |
678 | BTRFS_RESERVE_NO_FLUSH); | |
9ca9ee09 SW |
679 | } |
680 | ||
d4edf39b MX |
681 | /* |
682 | * btrfs_attach_transaction() - catch the running transaction | |
683 | * | |
684 | * It is used when we want to commit the current the transaction, but | |
685 | * don't want to start a new one. | |
686 | * | |
687 | * Note: If this function return -ENOENT, it just means there is no | |
688 | * running transaction. But it is possible that the inactive transaction | |
689 | * is still in the memory, not fully on disk. If you hope there is no | |
690 | * inactive transaction in the fs when -ENOENT is returned, you should | |
691 | * invoke | |
692 | * btrfs_attach_transaction_barrier() | |
693 | */ | |
354aa0fb | 694 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 695 | { |
575a75d6 AM |
696 | return start_transaction(root, 0, TRANS_ATTACH, |
697 | BTRFS_RESERVE_NO_FLUSH); | |
60376ce4 JB |
698 | } |
699 | ||
d4edf39b | 700 | /* |
90b6d283 | 701 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b MX |
702 | * |
703 | * It is similar to the above function, the differentia is this one | |
704 | * will wait for all the inactive transactions until they fully | |
705 | * complete. | |
706 | */ | |
707 | struct btrfs_trans_handle * | |
708 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
709 | { | |
710 | struct btrfs_trans_handle *trans; | |
711 | ||
575a75d6 AM |
712 | trans = start_transaction(root, 0, TRANS_ATTACH, |
713 | BTRFS_RESERVE_NO_FLUSH); | |
d4edf39b MX |
714 | if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT) |
715 | btrfs_wait_for_commit(root, 0); | |
716 | ||
717 | return trans; | |
718 | } | |
719 | ||
d352ac68 | 720 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 721 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
722 | struct btrfs_transaction *commit) |
723 | { | |
4a9d8bde | 724 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
725 | } |
726 | ||
46204592 SW |
727 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
728 | { | |
0b246afa | 729 | struct btrfs_fs_info *fs_info = root->fs_info; |
46204592 | 730 | struct btrfs_transaction *cur_trans = NULL, *t; |
8cd2807f | 731 | int ret = 0; |
46204592 | 732 | |
46204592 | 733 | if (transid) { |
0b246afa | 734 | if (transid <= fs_info->last_trans_committed) |
a4abeea4 | 735 | goto out; |
46204592 SW |
736 | |
737 | /* find specified transaction */ | |
0b246afa JM |
738 | spin_lock(&fs_info->trans_lock); |
739 | list_for_each_entry(t, &fs_info->trans_list, list) { | |
46204592 SW |
740 | if (t->transid == transid) { |
741 | cur_trans = t; | |
a4abeea4 | 742 | atomic_inc(&cur_trans->use_count); |
8cd2807f | 743 | ret = 0; |
46204592 SW |
744 | break; |
745 | } | |
8cd2807f MX |
746 | if (t->transid > transid) { |
747 | ret = 0; | |
46204592 | 748 | break; |
8cd2807f | 749 | } |
46204592 | 750 | } |
0b246afa | 751 | spin_unlock(&fs_info->trans_lock); |
42383020 SW |
752 | |
753 | /* | |
754 | * The specified transaction doesn't exist, or we | |
755 | * raced with btrfs_commit_transaction | |
756 | */ | |
757 | if (!cur_trans) { | |
0b246afa | 758 | if (transid > fs_info->last_trans_committed) |
42383020 | 759 | ret = -EINVAL; |
8cd2807f | 760 | goto out; |
42383020 | 761 | } |
46204592 SW |
762 | } else { |
763 | /* find newest transaction that is committing | committed */ | |
0b246afa JM |
764 | spin_lock(&fs_info->trans_lock); |
765 | list_for_each_entry_reverse(t, &fs_info->trans_list, | |
46204592 | 766 | list) { |
4a9d8bde MX |
767 | if (t->state >= TRANS_STATE_COMMIT_START) { |
768 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 769 | break; |
46204592 | 770 | cur_trans = t; |
a4abeea4 | 771 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
772 | break; |
773 | } | |
774 | } | |
0b246afa | 775 | spin_unlock(&fs_info->trans_lock); |
46204592 | 776 | if (!cur_trans) |
a4abeea4 | 777 | goto out; /* nothing committing|committed */ |
46204592 SW |
778 | } |
779 | ||
46204592 | 780 | wait_for_commit(root, cur_trans); |
724e2315 | 781 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 782 | out: |
46204592 SW |
783 | return ret; |
784 | } | |
785 | ||
37d1aeee CM |
786 | void btrfs_throttle(struct btrfs_root *root) |
787 | { | |
0b246afa JM |
788 | struct btrfs_fs_info *fs_info = root->fs_info; |
789 | ||
790 | if (!atomic_read(&fs_info->open_ioctl_trans)) | |
9ca9ee09 | 791 | wait_current_trans(root); |
37d1aeee CM |
792 | } |
793 | ||
8929ecfa YZ |
794 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
795 | struct btrfs_root *root) | |
796 | { | |
0b246afa JM |
797 | struct btrfs_fs_info *fs_info = root->fs_info; |
798 | ||
799 | if (fs_info->global_block_rsv.space_info->full && | |
0a2b2a84 | 800 | btrfs_check_space_for_delayed_refs(trans, root)) |
1be41b78 | 801 | return 1; |
36ba022a | 802 | |
0b246afa | 803 | return !!btrfs_block_rsv_check(root, &fs_info->global_block_rsv, 5); |
8929ecfa YZ |
804 | } |
805 | ||
806 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
807 | struct btrfs_root *root) | |
808 | { | |
809 | struct btrfs_transaction *cur_trans = trans->transaction; | |
810 | int updates; | |
49b25e05 | 811 | int err; |
8929ecfa | 812 | |
a4abeea4 | 813 | smp_mb(); |
4a9d8bde MX |
814 | if (cur_trans->state >= TRANS_STATE_BLOCKED || |
815 | cur_trans->delayed_refs.flushing) | |
8929ecfa YZ |
816 | return 1; |
817 | ||
818 | updates = trans->delayed_ref_updates; | |
819 | trans->delayed_ref_updates = 0; | |
49b25e05 | 820 | if (updates) { |
28ed1345 | 821 | err = btrfs_run_delayed_refs(trans, root, updates * 2); |
49b25e05 JM |
822 | if (err) /* Error code will also eval true */ |
823 | return err; | |
824 | } | |
8929ecfa YZ |
825 | |
826 | return should_end_transaction(trans, root); | |
827 | } | |
828 | ||
89ce8a63 | 829 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
a698d075 | 830 | struct btrfs_root *root, int throttle) |
79154b1b | 831 | { |
8929ecfa | 832 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 833 | struct btrfs_fs_info *info = root->fs_info; |
31b9655f | 834 | u64 transid = trans->transid; |
1be41b78 | 835 | unsigned long cur = trans->delayed_ref_updates; |
a698d075 | 836 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 837 | int err = 0; |
a79b7d4b | 838 | int must_run_delayed_refs = 0; |
c3e69d58 | 839 | |
3bbb24b2 JB |
840 | if (trans->use_count > 1) { |
841 | trans->use_count--; | |
2a1eb461 JB |
842 | trans->block_rsv = trans->orig_rsv; |
843 | return 0; | |
844 | } | |
845 | ||
b24e03db | 846 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 847 | trans->block_rsv = NULL; |
c5567237 | 848 | |
ea658bad JB |
849 | if (!list_empty(&trans->new_bgs)) |
850 | btrfs_create_pending_block_groups(trans, root); | |
851 | ||
1be41b78 | 852 | trans->delayed_ref_updates = 0; |
a79b7d4b CM |
853 | if (!trans->sync) { |
854 | must_run_delayed_refs = | |
855 | btrfs_should_throttle_delayed_refs(trans, root); | |
0a2b2a84 | 856 | cur = max_t(unsigned long, cur, 32); |
a79b7d4b CM |
857 | |
858 | /* | |
859 | * don't make the caller wait if they are from a NOLOCK | |
860 | * or ATTACH transaction, it will deadlock with commit | |
861 | */ | |
862 | if (must_run_delayed_refs == 1 && | |
863 | (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH))) | |
864 | must_run_delayed_refs = 2; | |
56bec294 | 865 | } |
bb721703 | 866 | |
0e721106 JB |
867 | btrfs_trans_release_metadata(trans, root); |
868 | trans->block_rsv = NULL; | |
56bec294 | 869 | |
ea658bad JB |
870 | if (!list_empty(&trans->new_bgs)) |
871 | btrfs_create_pending_block_groups(trans, root); | |
872 | ||
4fbcdf66 FM |
873 | btrfs_trans_release_chunk_metadata(trans); |
874 | ||
0b246afa | 875 | if (lock && !atomic_read(&info->open_ioctl_trans) && |
4a9d8bde MX |
876 | should_end_transaction(trans, root) && |
877 | ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) { | |
878 | spin_lock(&info->trans_lock); | |
879 | if (cur_trans->state == TRANS_STATE_RUNNING) | |
880 | cur_trans->state = TRANS_STATE_BLOCKED; | |
881 | spin_unlock(&info->trans_lock); | |
a4abeea4 | 882 | } |
8929ecfa | 883 | |
4a9d8bde | 884 | if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
3bbb24b2 | 885 | if (throttle) |
8929ecfa | 886 | return btrfs_commit_transaction(trans, root); |
3bbb24b2 | 887 | else |
8929ecfa YZ |
888 | wake_up_process(info->transaction_kthread); |
889 | } | |
890 | ||
0860adfd | 891 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 892 | sb_end_intwrite(info->sb); |
6df7881a | 893 | |
8929ecfa | 894 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
895 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
896 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 897 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 898 | |
a83342aa DS |
899 | /* |
900 | * Make sure counter is updated before we wake up waiters. | |
901 | */ | |
99d16cbc | 902 | smp_mb(); |
79154b1b CM |
903 | if (waitqueue_active(&cur_trans->writer_wait)) |
904 | wake_up(&cur_trans->writer_wait); | |
724e2315 | 905 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
906 | |
907 | if (current->journal_info == trans) | |
908 | current->journal_info = NULL; | |
ab78c84d | 909 | |
24bbcf04 YZ |
910 | if (throttle) |
911 | btrfs_run_delayed_iputs(root); | |
912 | ||
49b25e05 | 913 | if (trans->aborted || |
0b246afa | 914 | test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) { |
4e121c06 | 915 | wake_up_process(info->transaction_kthread); |
4edc2ca3 | 916 | err = -EIO; |
4e121c06 | 917 | } |
edf39272 | 918 | assert_qgroups_uptodate(trans); |
49b25e05 | 919 | |
4edc2ca3 | 920 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
a79b7d4b | 921 | if (must_run_delayed_refs) { |
31b9655f | 922 | btrfs_async_run_delayed_refs(root, cur, transid, |
a79b7d4b CM |
923 | must_run_delayed_refs == 1); |
924 | } | |
4edc2ca3 | 925 | return err; |
79154b1b CM |
926 | } |
927 | ||
89ce8a63 CM |
928 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
929 | struct btrfs_root *root) | |
930 | { | |
98ad43be | 931 | return __btrfs_end_transaction(trans, root, 0); |
89ce8a63 CM |
932 | } |
933 | ||
934 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
935 | struct btrfs_root *root) | |
936 | { | |
98ad43be | 937 | return __btrfs_end_transaction(trans, root, 1); |
16cdcec7 MX |
938 | } |
939 | ||
d352ac68 CM |
940 | /* |
941 | * when btree blocks are allocated, they have some corresponding bits set for | |
942 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 943 | * those extents are sent to disk but does not wait on them |
d352ac68 | 944 | */ |
690587d1 | 945 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 946 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 947 | { |
777e6bd7 | 948 | int err = 0; |
7c4452b9 | 949 | int werr = 0; |
0b246afa JM |
950 | struct btrfs_fs_info *fs_info = root->fs_info; |
951 | struct address_space *mapping = fs_info->btree_inode->i_mapping; | |
e6138876 | 952 | struct extent_state *cached_state = NULL; |
777e6bd7 | 953 | u64 start = 0; |
5f39d397 | 954 | u64 end; |
7c4452b9 | 955 | |
1728366e | 956 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 957 | mark, &cached_state)) { |
663dfbb0 FM |
958 | bool wait_writeback = false; |
959 | ||
960 | err = convert_extent_bit(dirty_pages, start, end, | |
961 | EXTENT_NEED_WAIT, | |
210aa277 | 962 | mark, &cached_state); |
663dfbb0 FM |
963 | /* |
964 | * convert_extent_bit can return -ENOMEM, which is most of the | |
965 | * time a temporary error. So when it happens, ignore the error | |
966 | * and wait for writeback of this range to finish - because we | |
967 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
968 | * to btrfs_wait_marked_extents() would not know that writeback | |
969 | * for this range started and therefore wouldn't wait for it to | |
970 | * finish - we don't want to commit a superblock that points to | |
971 | * btree nodes/leafs for which writeback hasn't finished yet | |
972 | * (and without errors). | |
973 | * We cleanup any entries left in the io tree when committing | |
974 | * the transaction (through clear_btree_io_tree()). | |
975 | */ | |
976 | if (err == -ENOMEM) { | |
977 | err = 0; | |
978 | wait_writeback = true; | |
979 | } | |
980 | if (!err) | |
981 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
982 | if (err) |
983 | werr = err; | |
663dfbb0 FM |
984 | else if (wait_writeback) |
985 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 986 | free_extent_state(cached_state); |
663dfbb0 | 987 | cached_state = NULL; |
1728366e JB |
988 | cond_resched(); |
989 | start = end + 1; | |
7c4452b9 | 990 | } |
690587d1 CM |
991 | return werr; |
992 | } | |
993 | ||
994 | /* | |
995 | * when btree blocks are allocated, they have some corresponding bits set for | |
996 | * them in one of two extent_io trees. This is used to make sure all of | |
997 | * those extents are on disk for transaction or log commit. We wait | |
998 | * on all the pages and clear them from the dirty pages state tree | |
999 | */ | |
1000 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 1001 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 1002 | { |
690587d1 CM |
1003 | int err = 0; |
1004 | int werr = 0; | |
0b246afa JM |
1005 | struct btrfs_fs_info *fs_info = root->fs_info; |
1006 | struct address_space *mapping = fs_info->btree_inode->i_mapping; | |
e6138876 | 1007 | struct extent_state *cached_state = NULL; |
690587d1 CM |
1008 | u64 start = 0; |
1009 | u64 end; | |
656f30db | 1010 | bool errors = false; |
777e6bd7 | 1011 | |
1728366e | 1012 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 1013 | EXTENT_NEED_WAIT, &cached_state)) { |
663dfbb0 FM |
1014 | /* |
1015 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
1016 | * When committing the transaction, we'll remove any entries | |
1017 | * left in the io tree. For a log commit, we don't remove them | |
1018 | * after committing the log because the tree can be accessed | |
1019 | * concurrently - we do it only at transaction commit time when | |
1020 | * it's safe to do it (through clear_btree_io_tree()). | |
1021 | */ | |
1022 | err = clear_extent_bit(dirty_pages, start, end, | |
1023 | EXTENT_NEED_WAIT, | |
1024 | 0, 0, &cached_state, GFP_NOFS); | |
1025 | if (err == -ENOMEM) | |
1026 | err = 0; | |
1027 | if (!err) | |
1028 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
1029 | if (err) |
1030 | werr = err; | |
e38e2ed7 FM |
1031 | free_extent_state(cached_state); |
1032 | cached_state = NULL; | |
1728366e JB |
1033 | cond_resched(); |
1034 | start = end + 1; | |
777e6bd7 | 1035 | } |
7c4452b9 CM |
1036 | if (err) |
1037 | werr = err; | |
656f30db FM |
1038 | |
1039 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { | |
1040 | if ((mark & EXTENT_DIRTY) && | |
0b246afa | 1041 | test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) |
656f30db FM |
1042 | errors = true; |
1043 | ||
1044 | if ((mark & EXTENT_NEW) && | |
0b246afa | 1045 | test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) |
656f30db FM |
1046 | errors = true; |
1047 | } else { | |
0b246afa | 1048 | if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) |
656f30db FM |
1049 | errors = true; |
1050 | } | |
1051 | ||
1052 | if (errors && !werr) | |
1053 | werr = -EIO; | |
1054 | ||
7c4452b9 | 1055 | return werr; |
79154b1b CM |
1056 | } |
1057 | ||
690587d1 CM |
1058 | /* |
1059 | * when btree blocks are allocated, they have some corresponding bits set for | |
1060 | * them in one of two extent_io trees. This is used to make sure all of | |
1061 | * those extents are on disk for transaction or log commit | |
1062 | */ | |
171170c1 | 1063 | static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, |
8cef4e16 | 1064 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
1065 | { |
1066 | int ret; | |
1067 | int ret2; | |
c6adc9cc | 1068 | struct blk_plug plug; |
690587d1 | 1069 | |
c6adc9cc | 1070 | blk_start_plug(&plug); |
8cef4e16 | 1071 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
c6adc9cc | 1072 | blk_finish_plug(&plug); |
8cef4e16 | 1073 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); |
bf0da8c1 CM |
1074 | |
1075 | if (ret) | |
1076 | return ret; | |
1077 | if (ret2) | |
1078 | return ret2; | |
1079 | return 0; | |
690587d1 CM |
1080 | } |
1081 | ||
663dfbb0 | 1082 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
d0c803c4 CM |
1083 | struct btrfs_root *root) |
1084 | { | |
663dfbb0 FM |
1085 | int ret; |
1086 | ||
1087 | ret = btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
1088 | &trans->transaction->dirty_pages, |
1089 | EXTENT_DIRTY); | |
663dfbb0 FM |
1090 | clear_btree_io_tree(&trans->transaction->dirty_pages); |
1091 | ||
1092 | return ret; | |
d0c803c4 CM |
1093 | } |
1094 | ||
d352ac68 CM |
1095 | /* |
1096 | * this is used to update the root pointer in the tree of tree roots. | |
1097 | * | |
1098 | * But, in the case of the extent allocation tree, updating the root | |
1099 | * pointer may allocate blocks which may change the root of the extent | |
1100 | * allocation tree. | |
1101 | * | |
1102 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1103 | * change while the root pointer was being updated in the metadata. | |
1104 | */ | |
0b86a832 CM |
1105 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1106 | struct btrfs_root *root) | |
79154b1b CM |
1107 | { |
1108 | int ret; | |
0b86a832 | 1109 | u64 old_root_bytenr; |
86b9f2ec | 1110 | u64 old_root_used; |
0b246afa JM |
1111 | struct btrfs_fs_info *fs_info = root->fs_info; |
1112 | struct btrfs_root *tree_root = fs_info->tree_root; | |
79154b1b | 1113 | |
86b9f2ec | 1114 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1115 | |
d397712b | 1116 | while (1) { |
0b86a832 | 1117 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1118 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1119 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1120 | break; |
87ef2bb4 | 1121 | |
5d4f98a2 | 1122 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1123 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1124 | &root->root_key, |
1125 | &root->root_item); | |
49b25e05 JM |
1126 | if (ret) |
1127 | return ret; | |
56bec294 | 1128 | |
86b9f2ec | 1129 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1130 | } |
276e680d | 1131 | |
0b86a832 CM |
1132 | return 0; |
1133 | } | |
1134 | ||
d352ac68 CM |
1135 | /* |
1136 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1137 | * |
1138 | * The error handling in this function may not be obvious. Any of the | |
1139 | * failures will cause the file system to go offline. We still need | |
1140 | * to clean up the delayed refs. | |
d352ac68 | 1141 | */ |
5d4f98a2 YZ |
1142 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
1143 | struct btrfs_root *root) | |
0b86a832 CM |
1144 | { |
1145 | struct btrfs_fs_info *fs_info = root->fs_info; | |
ea526d18 | 1146 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1147 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1148 | struct list_head *next; |
84234f3a | 1149 | struct extent_buffer *eb; |
56bec294 | 1150 | int ret; |
84234f3a YZ |
1151 | |
1152 | eb = btrfs_lock_root_node(fs_info->tree_root); | |
49b25e05 JM |
1153 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
1154 | 0, &eb); | |
84234f3a YZ |
1155 | btrfs_tree_unlock(eb); |
1156 | free_extent_buffer(eb); | |
0b86a832 | 1157 | |
49b25e05 JM |
1158 | if (ret) |
1159 | return ret; | |
1160 | ||
56bec294 | 1161 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1162 | if (ret) |
1163 | return ret; | |
87ef2bb4 | 1164 | |
0b246afa | 1165 | ret = btrfs_run_dev_stats(trans, fs_info); |
c16ce190 JB |
1166 | if (ret) |
1167 | return ret; | |
0b246afa | 1168 | ret = btrfs_run_dev_replace(trans, fs_info); |
c16ce190 JB |
1169 | if (ret) |
1170 | return ret; | |
0b246afa | 1171 | ret = btrfs_run_qgroups(trans, fs_info); |
c16ce190 JB |
1172 | if (ret) |
1173 | return ret; | |
546adb0d | 1174 | |
dcdf7f6d JB |
1175 | ret = btrfs_setup_space_cache(trans, root); |
1176 | if (ret) | |
1177 | return ret; | |
1178 | ||
546adb0d JS |
1179 | /* run_qgroups might have added some more refs */ |
1180 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
c16ce190 JB |
1181 | if (ret) |
1182 | return ret; | |
ea526d18 | 1183 | again: |
d397712b | 1184 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
1185 | next = fs_info->dirty_cowonly_roots.next; |
1186 | list_del_init(next); | |
1187 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1188 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1189 | |
9e351cc8 JB |
1190 | if (root != fs_info->extent_root) |
1191 | list_add_tail(&root->dirty_list, | |
1192 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1193 | ret = update_cowonly_root(trans, root); |
1194 | if (ret) | |
1195 | return ret; | |
ea526d18 JB |
1196 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1197 | if (ret) | |
1198 | return ret; | |
79154b1b | 1199 | } |
276e680d | 1200 | |
1bbc621e | 1201 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
ea526d18 JB |
1202 | ret = btrfs_write_dirty_block_groups(trans, root); |
1203 | if (ret) | |
1204 | return ret; | |
1205 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
1206 | if (ret) | |
1207 | return ret; | |
1208 | } | |
1209 | ||
1210 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1211 | goto again; | |
1212 | ||
9e351cc8 JB |
1213 | list_add_tail(&fs_info->extent_root->dirty_list, |
1214 | &trans->transaction->switch_commits); | |
8dabb742 SB |
1215 | btrfs_after_dev_replace_commit(fs_info); |
1216 | ||
79154b1b CM |
1217 | return 0; |
1218 | } | |
1219 | ||
d352ac68 CM |
1220 | /* |
1221 | * dead roots are old snapshots that need to be deleted. This allocates | |
1222 | * a dirty root struct and adds it into the list of dead roots that need to | |
1223 | * be deleted | |
1224 | */ | |
cfad392b | 1225 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1226 | { |
0b246afa JM |
1227 | struct btrfs_fs_info *fs_info = root->fs_info; |
1228 | ||
1229 | spin_lock(&fs_info->trans_lock); | |
cfad392b | 1230 | if (list_empty(&root->root_list)) |
0b246afa JM |
1231 | list_add_tail(&root->root_list, &fs_info->dead_roots); |
1232 | spin_unlock(&fs_info->trans_lock); | |
5eda7b5e CM |
1233 | } |
1234 | ||
d352ac68 | 1235 | /* |
5d4f98a2 | 1236 | * update all the cowonly tree roots on disk |
d352ac68 | 1237 | */ |
5d4f98a2 | 1238 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
5b4aacef | 1239 | struct btrfs_fs_info *fs_info) |
0f7d52f4 | 1240 | { |
0f7d52f4 | 1241 | struct btrfs_root *gang[8]; |
0f7d52f4 CM |
1242 | int i; |
1243 | int ret; | |
54aa1f4d CM |
1244 | int err = 0; |
1245 | ||
a4abeea4 | 1246 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1247 | while (1) { |
5d4f98a2 YZ |
1248 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1249 | (void **)gang, 0, | |
0f7d52f4 CM |
1250 | ARRAY_SIZE(gang), |
1251 | BTRFS_ROOT_TRANS_TAG); | |
1252 | if (ret == 0) | |
1253 | break; | |
1254 | for (i = 0; i < ret; i++) { | |
5b4aacef | 1255 | struct btrfs_root *root = gang[i]; |
5d4f98a2 YZ |
1256 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1257 | (unsigned long)root->root_key.objectid, | |
1258 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1259 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1260 | |
e02119d5 | 1261 | btrfs_free_log(trans, root); |
5d4f98a2 | 1262 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 1263 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 1264 | |
82d5902d LZ |
1265 | btrfs_save_ino_cache(root, trans); |
1266 | ||
f1ebcc74 | 1267 | /* see comments in should_cow_block() */ |
27cdeb70 | 1268 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
c7548af6 | 1269 | smp_mb__after_atomic(); |
f1ebcc74 | 1270 | |
978d910d | 1271 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1272 | list_add_tail(&root->dirty_list, |
1273 | &trans->transaction->switch_commits); | |
978d910d YZ |
1274 | btrfs_set_root_node(&root->root_item, |
1275 | root->node); | |
1276 | } | |
5d4f98a2 | 1277 | |
5d4f98a2 | 1278 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1279 | &root->root_key, |
1280 | &root->root_item); | |
a4abeea4 | 1281 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1282 | if (err) |
1283 | break; | |
7174109c | 1284 | btrfs_qgroup_free_meta_all(root); |
0f7d52f4 CM |
1285 | } |
1286 | } | |
a4abeea4 | 1287 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1288 | return err; |
0f7d52f4 CM |
1289 | } |
1290 | ||
d352ac68 | 1291 | /* |
de78b51a ES |
1292 | * defrag a given btree. |
1293 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1294 | */ |
de78b51a | 1295 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1296 | { |
1297 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1298 | struct btrfs_trans_handle *trans; |
8929ecfa | 1299 | int ret; |
e9d0b13b | 1300 | |
27cdeb70 | 1301 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1302 | return 0; |
8929ecfa | 1303 | |
6b80053d | 1304 | while (1) { |
8929ecfa YZ |
1305 | trans = btrfs_start_transaction(root, 0); |
1306 | if (IS_ERR(trans)) | |
1307 | return PTR_ERR(trans); | |
1308 | ||
de78b51a | 1309 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1310 | |
e9d0b13b | 1311 | btrfs_end_transaction(trans, root); |
b53d3f5d | 1312 | btrfs_btree_balance_dirty(info->tree_root); |
e9d0b13b CM |
1313 | cond_resched(); |
1314 | ||
ab8d0fc4 | 1315 | if (btrfs_fs_closing(info) || ret != -EAGAIN) |
e9d0b13b | 1316 | break; |
210549eb | 1317 | |
ab8d0fc4 JM |
1318 | if (btrfs_defrag_cancelled(info)) { |
1319 | btrfs_debug(info, "defrag_root cancelled"); | |
210549eb DS |
1320 | ret = -EAGAIN; |
1321 | break; | |
1322 | } | |
e9d0b13b | 1323 | } |
27cdeb70 | 1324 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1325 | return ret; |
e9d0b13b CM |
1326 | } |
1327 | ||
6426c7ad QW |
1328 | /* |
1329 | * Do all special snapshot related qgroup dirty hack. | |
1330 | * | |
1331 | * Will do all needed qgroup inherit and dirty hack like switch commit | |
1332 | * roots inside one transaction and write all btree into disk, to make | |
1333 | * qgroup works. | |
1334 | */ | |
1335 | static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, | |
1336 | struct btrfs_root *src, | |
1337 | struct btrfs_root *parent, | |
1338 | struct btrfs_qgroup_inherit *inherit, | |
1339 | u64 dst_objectid) | |
1340 | { | |
1341 | struct btrfs_fs_info *fs_info = src->fs_info; | |
1342 | int ret; | |
1343 | ||
1344 | /* | |
1345 | * Save some performance in the case that qgroups are not | |
1346 | * enabled. If this check races with the ioctl, rescan will | |
1347 | * kick in anyway. | |
1348 | */ | |
1349 | mutex_lock(&fs_info->qgroup_ioctl_lock); | |
afcdd129 | 1350 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
6426c7ad QW |
1351 | mutex_unlock(&fs_info->qgroup_ioctl_lock); |
1352 | return 0; | |
1353 | } | |
1354 | mutex_unlock(&fs_info->qgroup_ioctl_lock); | |
1355 | ||
1356 | /* | |
1357 | * We are going to commit transaction, see btrfs_commit_transaction() | |
1358 | * comment for reason locking tree_log_mutex | |
1359 | */ | |
1360 | mutex_lock(&fs_info->tree_log_mutex); | |
1361 | ||
5b4aacef | 1362 | ret = commit_fs_roots(trans, fs_info); |
6426c7ad QW |
1363 | if (ret) |
1364 | goto out; | |
1365 | ret = btrfs_qgroup_prepare_account_extents(trans, fs_info); | |
1366 | if (ret < 0) | |
1367 | goto out; | |
1368 | ret = btrfs_qgroup_account_extents(trans, fs_info); | |
1369 | if (ret < 0) | |
1370 | goto out; | |
1371 | ||
1372 | /* Now qgroup are all updated, we can inherit it to new qgroups */ | |
1373 | ret = btrfs_qgroup_inherit(trans, fs_info, | |
1374 | src->root_key.objectid, dst_objectid, | |
1375 | inherit); | |
1376 | if (ret < 0) | |
1377 | goto out; | |
1378 | ||
1379 | /* | |
1380 | * Now we do a simplified commit transaction, which will: | |
1381 | * 1) commit all subvolume and extent tree | |
1382 | * To ensure all subvolume and extent tree have a valid | |
1383 | * commit_root to accounting later insert_dir_item() | |
1384 | * 2) write all btree blocks onto disk | |
1385 | * This is to make sure later btree modification will be cowed | |
1386 | * Or commit_root can be populated and cause wrong qgroup numbers | |
1387 | * In this simplified commit, we don't really care about other trees | |
1388 | * like chunk and root tree, as they won't affect qgroup. | |
1389 | * And we don't write super to avoid half committed status. | |
1390 | */ | |
1391 | ret = commit_cowonly_roots(trans, src); | |
1392 | if (ret) | |
1393 | goto out; | |
1394 | switch_commit_roots(trans->transaction, fs_info); | |
1395 | ret = btrfs_write_and_wait_transaction(trans, src); | |
1396 | if (ret) | |
f7af3934 | 1397 | btrfs_handle_fs_error(fs_info, ret, |
6426c7ad QW |
1398 | "Error while writing out transaction for qgroup"); |
1399 | ||
1400 | out: | |
1401 | mutex_unlock(&fs_info->tree_log_mutex); | |
1402 | ||
1403 | /* | |
1404 | * Force parent root to be updated, as we recorded it before so its | |
1405 | * last_trans == cur_transid. | |
1406 | * Or it won't be committed again onto disk after later | |
1407 | * insert_dir_item() | |
1408 | */ | |
1409 | if (!ret) | |
1410 | record_root_in_trans(trans, parent, 1); | |
1411 | return ret; | |
1412 | } | |
1413 | ||
d352ac68 CM |
1414 | /* |
1415 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1416 | * transaction commit. This does the actual creation. |
1417 | * | |
1418 | * Note: | |
1419 | * If the error which may affect the commitment of the current transaction | |
1420 | * happens, we should return the error number. If the error which just affect | |
1421 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1422 | */ |
80b6794d | 1423 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1424 | struct btrfs_fs_info *fs_info, |
1425 | struct btrfs_pending_snapshot *pending) | |
1426 | { | |
1427 | struct btrfs_key key; | |
80b6794d | 1428 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1429 | struct btrfs_root *tree_root = fs_info->tree_root; |
1430 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1431 | struct btrfs_root *parent_root; |
98c9942a | 1432 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1433 | struct inode *parent_inode; |
42874b3d MX |
1434 | struct btrfs_path *path; |
1435 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1436 | struct dentry *dentry; |
3063d29f | 1437 | struct extent_buffer *tmp; |
925baedd | 1438 | struct extent_buffer *old; |
04b285f3 | 1439 | struct timespec cur_time; |
aec8030a | 1440 | int ret = 0; |
d68fc57b | 1441 | u64 to_reserve = 0; |
6bdb72de | 1442 | u64 index = 0; |
a22285a6 | 1443 | u64 objectid; |
b83cc969 | 1444 | u64 root_flags; |
8ea05e3a | 1445 | uuid_le new_uuid; |
3063d29f | 1446 | |
8546b570 DS |
1447 | ASSERT(pending->path); |
1448 | path = pending->path; | |
42874b3d | 1449 | |
b0c0ea63 DS |
1450 | ASSERT(pending->root_item); |
1451 | new_root_item = pending->root_item; | |
a22285a6 | 1452 | |
aec8030a MX |
1453 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1454 | if (pending->error) | |
6fa9700e | 1455 | goto no_free_objectid; |
3063d29f | 1456 | |
d6726335 QW |
1457 | /* |
1458 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1459 | * accounted by later btrfs_qgroup_inherit(). | |
1460 | */ | |
1461 | btrfs_set_skip_qgroup(trans, objectid); | |
1462 | ||
147d256e | 1463 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
d68fc57b YZ |
1464 | |
1465 | if (to_reserve > 0) { | |
aec8030a MX |
1466 | pending->error = btrfs_block_rsv_add(root, |
1467 | &pending->block_rsv, | |
1468 | to_reserve, | |
1469 | BTRFS_RESERVE_NO_FLUSH); | |
1470 | if (pending->error) | |
d6726335 | 1471 | goto clear_skip_qgroup; |
d68fc57b YZ |
1472 | } |
1473 | ||
3063d29f | 1474 | key.objectid = objectid; |
a22285a6 YZ |
1475 | key.offset = (u64)-1; |
1476 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1477 | |
6fa9700e | 1478 | rsv = trans->block_rsv; |
a22285a6 | 1479 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1480 | trans->bytes_reserved = trans->block_rsv->reserved; |
0b246afa | 1481 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa JB |
1482 | trans->transid, |
1483 | trans->bytes_reserved, 1); | |
a22285a6 | 1484 | dentry = pending->dentry; |
e9662f70 | 1485 | parent_inode = pending->dir; |
a22285a6 | 1486 | parent_root = BTRFS_I(parent_inode)->root; |
6426c7ad | 1487 | record_root_in_trans(trans, parent_root, 0); |
a22285a6 | 1488 | |
c2050a45 | 1489 | cur_time = current_time(parent_inode); |
04b285f3 | 1490 | |
3063d29f CM |
1491 | /* |
1492 | * insert the directory item | |
1493 | */ | |
3de4586c | 1494 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1495 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1496 | |
1497 | /* check if there is a file/dir which has the same name. */ | |
1498 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
1499 | btrfs_ino(parent_inode), | |
1500 | dentry->d_name.name, | |
1501 | dentry->d_name.len, 0); | |
1502 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1503 | pending->error = -EEXIST; |
aec8030a | 1504 | goto dir_item_existed; |
42874b3d MX |
1505 | } else if (IS_ERR(dir_item)) { |
1506 | ret = PTR_ERR(dir_item); | |
66642832 | 1507 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1508 | goto fail; |
79787eaa | 1509 | } |
42874b3d | 1510 | btrfs_release_path(path); |
52c26179 | 1511 | |
e999376f CM |
1512 | /* |
1513 | * pull in the delayed directory update | |
1514 | * and the delayed inode item | |
1515 | * otherwise we corrupt the FS during | |
1516 | * snapshot | |
1517 | */ | |
1518 | ret = btrfs_run_delayed_items(trans, root); | |
8732d44f | 1519 | if (ret) { /* Transaction aborted */ |
66642832 | 1520 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1521 | goto fail; |
1522 | } | |
e999376f | 1523 | |
6426c7ad | 1524 | record_root_in_trans(trans, root, 0); |
6bdb72de SW |
1525 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1526 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1527 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1528 | |
b83cc969 LZ |
1529 | root_flags = btrfs_root_flags(new_root_item); |
1530 | if (pending->readonly) | |
1531 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1532 | else | |
1533 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1534 | btrfs_set_root_flags(new_root_item, root_flags); | |
1535 | ||
8ea05e3a AB |
1536 | btrfs_set_root_generation_v2(new_root_item, |
1537 | trans->transid); | |
1538 | uuid_le_gen(&new_uuid); | |
1539 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1540 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1541 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1542 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1543 | memset(new_root_item->received_uuid, 0, | |
1544 | sizeof(new_root_item->received_uuid)); | |
1545 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1546 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1547 | btrfs_set_root_stransid(new_root_item, 0); | |
1548 | btrfs_set_root_rtransid(new_root_item, 0); | |
1549 | } | |
3cae210f QW |
1550 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1551 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1552 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1553 | |
6bdb72de | 1554 | old = btrfs_lock_root_node(root); |
49b25e05 | 1555 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1556 | if (ret) { |
1557 | btrfs_tree_unlock(old); | |
1558 | free_extent_buffer(old); | |
66642832 | 1559 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1560 | goto fail; |
79787eaa | 1561 | } |
49b25e05 | 1562 | |
6bdb72de SW |
1563 | btrfs_set_lock_blocking(old); |
1564 | ||
49b25e05 | 1565 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1566 | /* clean up in any case */ |
6bdb72de SW |
1567 | btrfs_tree_unlock(old); |
1568 | free_extent_buffer(old); | |
8732d44f | 1569 | if (ret) { |
66642832 | 1570 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1571 | goto fail; |
1572 | } | |
f1ebcc74 | 1573 | /* see comments in should_cow_block() */ |
27cdeb70 | 1574 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1575 | smp_wmb(); |
1576 | ||
6bdb72de | 1577 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1578 | /* record when the snapshot was created in key.offset */ |
1579 | key.offset = trans->transid; | |
1580 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1581 | btrfs_tree_unlock(tmp); |
1582 | free_extent_buffer(tmp); | |
8732d44f | 1583 | if (ret) { |
66642832 | 1584 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1585 | goto fail; |
1586 | } | |
6bdb72de | 1587 | |
a22285a6 YZ |
1588 | /* |
1589 | * insert root back/forward references | |
1590 | */ | |
6bccf3ab | 1591 | ret = btrfs_add_root_ref(trans, fs_info, objectid, |
0660b5af | 1592 | parent_root->root_key.objectid, |
33345d01 | 1593 | btrfs_ino(parent_inode), index, |
a22285a6 | 1594 | dentry->d_name.name, dentry->d_name.len); |
8732d44f | 1595 | if (ret) { |
66642832 | 1596 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1597 | goto fail; |
1598 | } | |
0660b5af | 1599 | |
a22285a6 | 1600 | key.offset = (u64)-1; |
0b246afa | 1601 | pending->snap = btrfs_read_fs_root_no_name(fs_info, &key); |
79787eaa JM |
1602 | if (IS_ERR(pending->snap)) { |
1603 | ret = PTR_ERR(pending->snap); | |
66642832 | 1604 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1605 | goto fail; |
79787eaa | 1606 | } |
d68fc57b | 1607 | |
49b25e05 | 1608 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f | 1609 | if (ret) { |
66642832 | 1610 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1611 | goto fail; |
1612 | } | |
361048f5 MX |
1613 | |
1614 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
8732d44f | 1615 | if (ret) { |
66642832 | 1616 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1617 | goto fail; |
1618 | } | |
42874b3d | 1619 | |
6426c7ad QW |
1620 | /* |
1621 | * Do special qgroup accounting for snapshot, as we do some qgroup | |
1622 | * snapshot hack to do fast snapshot. | |
1623 | * To co-operate with that hack, we do hack again. | |
1624 | * Or snapshot will be greatly slowed down by a subtree qgroup rescan | |
1625 | */ | |
1626 | ret = qgroup_account_snapshot(trans, root, parent_root, | |
1627 | pending->inherit, objectid); | |
1628 | if (ret < 0) | |
1629 | goto fail; | |
1630 | ||
42874b3d MX |
1631 | ret = btrfs_insert_dir_item(trans, parent_root, |
1632 | dentry->d_name.name, dentry->d_name.len, | |
1633 | parent_inode, &key, | |
1634 | BTRFS_FT_DIR, index); | |
1635 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1636 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f | 1637 | if (ret) { |
66642832 | 1638 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1639 | goto fail; |
1640 | } | |
42874b3d MX |
1641 | |
1642 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
1643 | dentry->d_name.len * 2); | |
04b285f3 | 1644 | parent_inode->i_mtime = parent_inode->i_ctime = |
c2050a45 | 1645 | current_time(parent_inode); |
be6aef60 | 1646 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
dd5f9615 | 1647 | if (ret) { |
66642832 | 1648 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1649 | goto fail; |
1650 | } | |
6bccf3ab | 1651 | ret = btrfs_uuid_tree_add(trans, fs_info, new_uuid.b, |
dd5f9615 SB |
1652 | BTRFS_UUID_KEY_SUBVOL, objectid); |
1653 | if (ret) { | |
66642832 | 1654 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1655 | goto fail; |
1656 | } | |
1657 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
6bccf3ab | 1658 | ret = btrfs_uuid_tree_add(trans, fs_info, |
dd5f9615 SB |
1659 | new_root_item->received_uuid, |
1660 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
1661 | objectid); | |
1662 | if (ret && ret != -EEXIST) { | |
66642832 | 1663 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1664 | goto fail; |
1665 | } | |
1666 | } | |
d6726335 QW |
1667 | |
1668 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
1669 | if (ret) { | |
66642832 | 1670 | btrfs_abort_transaction(trans, ret); |
d6726335 QW |
1671 | goto fail; |
1672 | } | |
1673 | ||
3063d29f | 1674 | fail: |
aec8030a MX |
1675 | pending->error = ret; |
1676 | dir_item_existed: | |
98c9942a | 1677 | trans->block_rsv = rsv; |
2382c5cc | 1678 | trans->bytes_reserved = 0; |
d6726335 QW |
1679 | clear_skip_qgroup: |
1680 | btrfs_clear_skip_qgroup(trans); | |
6fa9700e MX |
1681 | no_free_objectid: |
1682 | kfree(new_root_item); | |
b0c0ea63 | 1683 | pending->root_item = NULL; |
42874b3d | 1684 | btrfs_free_path(path); |
8546b570 DS |
1685 | pending->path = NULL; |
1686 | ||
49b25e05 | 1687 | return ret; |
3063d29f CM |
1688 | } |
1689 | ||
d352ac68 CM |
1690 | /* |
1691 | * create all the snapshots we've scheduled for creation | |
1692 | */ | |
80b6794d CM |
1693 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1694 | struct btrfs_fs_info *fs_info) | |
3de4586c | 1695 | { |
aec8030a | 1696 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1697 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1698 | int ret = 0; |
3de4586c | 1699 | |
aec8030a MX |
1700 | list_for_each_entry_safe(pending, next, head, list) { |
1701 | list_del(&pending->list); | |
1702 | ret = create_pending_snapshot(trans, fs_info, pending); | |
1703 | if (ret) | |
1704 | break; | |
1705 | } | |
1706 | return ret; | |
3de4586c CM |
1707 | } |
1708 | ||
5d4f98a2 YZ |
1709 | static void update_super_roots(struct btrfs_root *root) |
1710 | { | |
0b246afa | 1711 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
1712 | struct btrfs_root_item *root_item; |
1713 | struct btrfs_super_block *super; | |
1714 | ||
0b246afa | 1715 | super = fs_info->super_copy; |
5d4f98a2 | 1716 | |
0b246afa | 1717 | root_item = &fs_info->chunk_root->root_item; |
5d4f98a2 YZ |
1718 | super->chunk_root = root_item->bytenr; |
1719 | super->chunk_root_generation = root_item->generation; | |
1720 | super->chunk_root_level = root_item->level; | |
1721 | ||
0b246afa | 1722 | root_item = &fs_info->tree_root->root_item; |
5d4f98a2 YZ |
1723 | super->root = root_item->bytenr; |
1724 | super->generation = root_item->generation; | |
1725 | super->root_level = root_item->level; | |
0b246afa | 1726 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) |
0af3d00b | 1727 | super->cache_generation = root_item->generation; |
0b246afa | 1728 | if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) |
70f80175 | 1729 | super->uuid_tree_generation = root_item->generation; |
5d4f98a2 YZ |
1730 | } |
1731 | ||
f36f3042 CM |
1732 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1733 | { | |
4a9d8bde | 1734 | struct btrfs_transaction *trans; |
f36f3042 | 1735 | int ret = 0; |
4a9d8bde | 1736 | |
a4abeea4 | 1737 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1738 | trans = info->running_transaction; |
1739 | if (trans) | |
1740 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1741 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1742 | return ret; |
1743 | } | |
1744 | ||
8929ecfa YZ |
1745 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1746 | { | |
4a9d8bde | 1747 | struct btrfs_transaction *trans; |
8929ecfa | 1748 | int ret = 0; |
4a9d8bde | 1749 | |
a4abeea4 | 1750 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1751 | trans = info->running_transaction; |
1752 | if (trans) | |
1753 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1754 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1755 | return ret; |
1756 | } | |
1757 | ||
bb9c12c9 SW |
1758 | /* |
1759 | * wait for the current transaction commit to start and block subsequent | |
1760 | * transaction joins | |
1761 | */ | |
1762 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1763 | struct btrfs_transaction *trans) | |
1764 | { | |
4a9d8bde | 1765 | wait_event(root->fs_info->transaction_blocked_wait, |
501407aa JB |
1766 | trans->state >= TRANS_STATE_COMMIT_START || |
1767 | trans->aborted); | |
bb9c12c9 SW |
1768 | } |
1769 | ||
1770 | /* | |
1771 | * wait for the current transaction to start and then become unblocked. | |
1772 | * caller holds ref. | |
1773 | */ | |
1774 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1775 | struct btrfs_transaction *trans) | |
1776 | { | |
72d63ed6 | 1777 | wait_event(root->fs_info->transaction_wait, |
501407aa JB |
1778 | trans->state >= TRANS_STATE_UNBLOCKED || |
1779 | trans->aborted); | |
bb9c12c9 SW |
1780 | } |
1781 | ||
1782 | /* | |
1783 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1784 | * returns, any subsequent transaction will not be allowed to join. | |
1785 | */ | |
1786 | struct btrfs_async_commit { | |
1787 | struct btrfs_trans_handle *newtrans; | |
1788 | struct btrfs_root *root; | |
7892b5af | 1789 | struct work_struct work; |
bb9c12c9 SW |
1790 | }; |
1791 | ||
1792 | static void do_async_commit(struct work_struct *work) | |
1793 | { | |
1794 | struct btrfs_async_commit *ac = | |
7892b5af | 1795 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1796 | |
6fc4e354 SW |
1797 | /* |
1798 | * We've got freeze protection passed with the transaction. | |
1799 | * Tell lockdep about it. | |
1800 | */ | |
b1a06a4b | 1801 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
bee9182d | 1802 | __sb_writers_acquired(ac->root->fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1803 | |
e209db7a SW |
1804 | current->journal_info = ac->newtrans; |
1805 | ||
bb9c12c9 SW |
1806 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1807 | kfree(ac); | |
1808 | } | |
1809 | ||
1810 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1811 | struct btrfs_root *root, | |
1812 | int wait_for_unblock) | |
1813 | { | |
0b246afa | 1814 | struct btrfs_fs_info *fs_info = root->fs_info; |
bb9c12c9 SW |
1815 | struct btrfs_async_commit *ac; |
1816 | struct btrfs_transaction *cur_trans; | |
1817 | ||
1818 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1819 | if (!ac) |
1820 | return -ENOMEM; | |
bb9c12c9 | 1821 | |
7892b5af | 1822 | INIT_WORK(&ac->work, do_async_commit); |
bb9c12c9 | 1823 | ac->root = root; |
7a7eaa40 | 1824 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1825 | if (IS_ERR(ac->newtrans)) { |
1826 | int err = PTR_ERR(ac->newtrans); | |
1827 | kfree(ac); | |
1828 | return err; | |
1829 | } | |
bb9c12c9 SW |
1830 | |
1831 | /* take transaction reference */ | |
bb9c12c9 | 1832 | cur_trans = trans->transaction; |
13c5a93e | 1833 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1834 | |
1835 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1836 | |
1837 | /* | |
1838 | * Tell lockdep we've released the freeze rwsem, since the | |
1839 | * async commit thread will be the one to unlock it. | |
1840 | */ | |
b1a06a4b | 1841 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
0b246afa | 1842 | __sb_writers_release(fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1843 | |
7892b5af | 1844 | schedule_work(&ac->work); |
bb9c12c9 SW |
1845 | |
1846 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1847 | if (wait_for_unblock) |
1848 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1849 | else | |
1850 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1851 | |
38e88054 SW |
1852 | if (current->journal_info == trans) |
1853 | current->journal_info = NULL; | |
1854 | ||
724e2315 | 1855 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1856 | return 0; |
1857 | } | |
1858 | ||
49b25e05 JM |
1859 | |
1860 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1861 | struct btrfs_root *root, int err) |
49b25e05 | 1862 | { |
0b246afa | 1863 | struct btrfs_fs_info *fs_info = root->fs_info; |
49b25e05 | 1864 | struct btrfs_transaction *cur_trans = trans->transaction; |
f094ac32 | 1865 | DEFINE_WAIT(wait); |
49b25e05 JM |
1866 | |
1867 | WARN_ON(trans->use_count > 1); | |
1868 | ||
66642832 | 1869 | btrfs_abort_transaction(trans, err); |
7b8b92af | 1870 | |
0b246afa | 1871 | spin_lock(&fs_info->trans_lock); |
66b6135b | 1872 | |
25d8c284 MX |
1873 | /* |
1874 | * If the transaction is removed from the list, it means this | |
1875 | * transaction has been committed successfully, so it is impossible | |
1876 | * to call the cleanup function. | |
1877 | */ | |
1878 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1879 | |
49b25e05 | 1880 | list_del_init(&cur_trans->list); |
0b246afa | 1881 | if (cur_trans == fs_info->running_transaction) { |
4a9d8bde | 1882 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 1883 | spin_unlock(&fs_info->trans_lock); |
f094ac32 LB |
1884 | wait_event(cur_trans->writer_wait, |
1885 | atomic_read(&cur_trans->num_writers) == 1); | |
1886 | ||
0b246afa | 1887 | spin_lock(&fs_info->trans_lock); |
d7096fc3 | 1888 | } |
0b246afa | 1889 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
1890 | |
1891 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1892 | ||
0b246afa JM |
1893 | spin_lock(&fs_info->trans_lock); |
1894 | if (cur_trans == fs_info->running_transaction) | |
1895 | fs_info->running_transaction = NULL; | |
1896 | spin_unlock(&fs_info->trans_lock); | |
4a9d8bde | 1897 | |
e0228285 | 1898 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 1899 | sb_end_intwrite(fs_info->sb); |
724e2315 JB |
1900 | btrfs_put_transaction(cur_trans); |
1901 | btrfs_put_transaction(cur_trans); | |
49b25e05 JM |
1902 | |
1903 | trace_btrfs_transaction_commit(root); | |
1904 | ||
49b25e05 JM |
1905 | if (current->journal_info == trans) |
1906 | current->journal_info = NULL; | |
0b246afa | 1907 | btrfs_scrub_cancel(fs_info); |
49b25e05 JM |
1908 | |
1909 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1910 | } | |
1911 | ||
82436617 MX |
1912 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
1913 | { | |
3cdde224 | 1914 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
6c255e67 | 1915 | return btrfs_start_delalloc_roots(fs_info, 1, -1); |
82436617 MX |
1916 | return 0; |
1917 | } | |
1918 | ||
1919 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) | |
1920 | { | |
3cdde224 | 1921 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
578def7c | 1922 | btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1); |
82436617 MX |
1923 | } |
1924 | ||
50d9aa99 | 1925 | static inline void |
161c3549 | 1926 | btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans) |
50d9aa99 | 1927 | { |
161c3549 JB |
1928 | wait_event(cur_trans->pending_wait, |
1929 | atomic_read(&cur_trans->pending_ordered) == 0); | |
50d9aa99 JB |
1930 | } |
1931 | ||
79154b1b CM |
1932 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1933 | struct btrfs_root *root) | |
1934 | { | |
0b246afa | 1935 | struct btrfs_fs_info *fs_info = root->fs_info; |
49b25e05 | 1936 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1937 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 1938 | int ret; |
79154b1b | 1939 | |
8d25a086 MX |
1940 | /* Stop the commit early if ->aborted is set */ |
1941 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
25287e0a | 1942 | ret = cur_trans->aborted; |
e4a2bcac JB |
1943 | btrfs_end_transaction(trans, root); |
1944 | return ret; | |
25287e0a | 1945 | } |
49b25e05 | 1946 | |
56bec294 CM |
1947 | /* make a pass through all the delayed refs we have so far |
1948 | * any runnings procs may add more while we are here | |
1949 | */ | |
1950 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
e4a2bcac JB |
1951 | if (ret) { |
1952 | btrfs_end_transaction(trans, root); | |
1953 | return ret; | |
1954 | } | |
56bec294 | 1955 | |
0e721106 JB |
1956 | btrfs_trans_release_metadata(trans, root); |
1957 | trans->block_rsv = NULL; | |
1958 | ||
b7ec40d7 | 1959 | cur_trans = trans->transaction; |
49b25e05 | 1960 | |
56bec294 CM |
1961 | /* |
1962 | * set the flushing flag so procs in this transaction have to | |
1963 | * start sending their work down. | |
1964 | */ | |
b7ec40d7 | 1965 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 1966 | smp_wmb(); |
56bec294 | 1967 | |
ea658bad JB |
1968 | if (!list_empty(&trans->new_bgs)) |
1969 | btrfs_create_pending_block_groups(trans, root); | |
1970 | ||
c3e69d58 | 1971 | ret = btrfs_run_delayed_refs(trans, root, 0); |
e4a2bcac JB |
1972 | if (ret) { |
1973 | btrfs_end_transaction(trans, root); | |
1974 | return ret; | |
1975 | } | |
56bec294 | 1976 | |
3204d33c | 1977 | if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { |
1bbc621e CM |
1978 | int run_it = 0; |
1979 | ||
1980 | /* this mutex is also taken before trying to set | |
1981 | * block groups readonly. We need to make sure | |
1982 | * that nobody has set a block group readonly | |
1983 | * after a extents from that block group have been | |
1984 | * allocated for cache files. btrfs_set_block_group_ro | |
1985 | * will wait for the transaction to commit if it | |
3204d33c | 1986 | * finds BTRFS_TRANS_DIRTY_BG_RUN set. |
1bbc621e | 1987 | * |
3204d33c JB |
1988 | * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure |
1989 | * only one process starts all the block group IO. It wouldn't | |
1bbc621e CM |
1990 | * hurt to have more than one go through, but there's no |
1991 | * real advantage to it either. | |
1992 | */ | |
0b246afa | 1993 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c JB |
1994 | if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, |
1995 | &cur_trans->flags)) | |
1bbc621e | 1996 | run_it = 1; |
0b246afa | 1997 | mutex_unlock(&fs_info->ro_block_group_mutex); |
1bbc621e CM |
1998 | |
1999 | if (run_it) | |
2000 | ret = btrfs_start_dirty_block_groups(trans, root); | |
2001 | } | |
2002 | if (ret) { | |
2003 | btrfs_end_transaction(trans, root); | |
2004 | return ret; | |
2005 | } | |
2006 | ||
0b246afa | 2007 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2008 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { |
0b246afa | 2009 | spin_unlock(&fs_info->trans_lock); |
13c5a93e | 2010 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 2011 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 2012 | |
b9c8300c | 2013 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 2014 | |
b4924a0f LB |
2015 | if (unlikely(cur_trans->aborted)) |
2016 | ret = cur_trans->aborted; | |
2017 | ||
724e2315 | 2018 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 2019 | |
49b25e05 | 2020 | return ret; |
79154b1b | 2021 | } |
4313b399 | 2022 | |
4a9d8bde | 2023 | cur_trans->state = TRANS_STATE_COMMIT_START; |
0b246afa | 2024 | wake_up(&fs_info->transaction_blocked_wait); |
bb9c12c9 | 2025 | |
0b246afa | 2026 | if (cur_trans->list.prev != &fs_info->trans_list) { |
ccd467d6 CM |
2027 | prev_trans = list_entry(cur_trans->list.prev, |
2028 | struct btrfs_transaction, list); | |
4a9d8bde | 2029 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
13c5a93e | 2030 | atomic_inc(&prev_trans->use_count); |
0b246afa | 2031 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 CM |
2032 | |
2033 | wait_for_commit(root, prev_trans); | |
1f9b8c8f | 2034 | ret = prev_trans->aborted; |
ccd467d6 | 2035 | |
724e2315 | 2036 | btrfs_put_transaction(prev_trans); |
1f9b8c8f FM |
2037 | if (ret) |
2038 | goto cleanup_transaction; | |
a4abeea4 | 2039 | } else { |
0b246afa | 2040 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2041 | } |
a4abeea4 | 2042 | } else { |
0b246afa | 2043 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2044 | } |
15ee9bc7 | 2045 | |
0860adfd MX |
2046 | extwriter_counter_dec(cur_trans, trans->type); |
2047 | ||
0b246afa | 2048 | ret = btrfs_start_delalloc_flush(fs_info); |
82436617 MX |
2049 | if (ret) |
2050 | goto cleanup_transaction; | |
2051 | ||
8d875f95 | 2052 | ret = btrfs_run_delayed_items(trans, root); |
581227d0 MX |
2053 | if (ret) |
2054 | goto cleanup_transaction; | |
15ee9bc7 | 2055 | |
581227d0 MX |
2056 | wait_event(cur_trans->writer_wait, |
2057 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 2058 | |
581227d0 | 2059 | /* some pending stuffs might be added after the previous flush. */ |
8d875f95 | 2060 | ret = btrfs_run_delayed_items(trans, root); |
ca469637 MX |
2061 | if (ret) |
2062 | goto cleanup_transaction; | |
2063 | ||
0b246afa | 2064 | btrfs_wait_delalloc_flush(fs_info); |
cb7ab021 | 2065 | |
161c3549 | 2066 | btrfs_wait_pending_ordered(cur_trans); |
50d9aa99 | 2067 | |
cb7ab021 | 2068 | btrfs_scrub_pause(root); |
ed0ca140 JB |
2069 | /* |
2070 | * Ok now we need to make sure to block out any other joins while we | |
2071 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 2072 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 | 2073 | */ |
0b246afa | 2074 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2075 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 2076 | spin_unlock(&fs_info->trans_lock); |
ed0ca140 JB |
2077 | wait_event(cur_trans->writer_wait, |
2078 | atomic_read(&cur_trans->num_writers) == 1); | |
2079 | ||
2cba30f1 MX |
2080 | /* ->aborted might be set after the previous check, so check it */ |
2081 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
2082 | ret = cur_trans->aborted; | |
6cf7f77e | 2083 | goto scrub_continue; |
2cba30f1 | 2084 | } |
7585717f CM |
2085 | /* |
2086 | * the reloc mutex makes sure that we stop | |
2087 | * the balancing code from coming in and moving | |
2088 | * extents around in the middle of the commit | |
2089 | */ | |
0b246afa | 2090 | mutex_lock(&fs_info->reloc_mutex); |
7585717f | 2091 | |
42874b3d MX |
2092 | /* |
2093 | * We needn't worry about the delayed items because we will | |
2094 | * deal with them in create_pending_snapshot(), which is the | |
2095 | * core function of the snapshot creation. | |
2096 | */ | |
0b246afa | 2097 | ret = create_pending_snapshots(trans, fs_info); |
49b25e05 | 2098 | if (ret) { |
0b246afa | 2099 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2100 | goto scrub_continue; |
49b25e05 | 2101 | } |
3063d29f | 2102 | |
42874b3d MX |
2103 | /* |
2104 | * We insert the dir indexes of the snapshots and update the inode | |
2105 | * of the snapshots' parents after the snapshot creation, so there | |
2106 | * are some delayed items which are not dealt with. Now deal with | |
2107 | * them. | |
2108 | * | |
2109 | * We needn't worry that this operation will corrupt the snapshots, | |
2110 | * because all the tree which are snapshoted will be forced to COW | |
2111 | * the nodes and leaves. | |
2112 | */ | |
2113 | ret = btrfs_run_delayed_items(trans, root); | |
49b25e05 | 2114 | if (ret) { |
0b246afa | 2115 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2116 | goto scrub_continue; |
49b25e05 | 2117 | } |
16cdcec7 | 2118 | |
56bec294 | 2119 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 | 2120 | if (ret) { |
0b246afa | 2121 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2122 | goto scrub_continue; |
49b25e05 | 2123 | } |
56bec294 | 2124 | |
0ed4792a | 2125 | /* Reocrd old roots for later qgroup accounting */ |
0b246afa | 2126 | ret = btrfs_qgroup_prepare_account_extents(trans, fs_info); |
0ed4792a | 2127 | if (ret) { |
0b246afa | 2128 | mutex_unlock(&fs_info->reloc_mutex); |
0ed4792a QW |
2129 | goto scrub_continue; |
2130 | } | |
2131 | ||
e999376f CM |
2132 | /* |
2133 | * make sure none of the code above managed to slip in a | |
2134 | * delayed item | |
2135 | */ | |
ccdf9b30 | 2136 | btrfs_assert_delayed_root_empty(fs_info); |
e999376f | 2137 | |
2c90e5d6 | 2138 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 2139 | |
e02119d5 CM |
2140 | /* btrfs_commit_tree_roots is responsible for getting the |
2141 | * various roots consistent with each other. Every pointer | |
2142 | * in the tree of tree roots has to point to the most up to date | |
2143 | * root for every subvolume and other tree. So, we have to keep | |
2144 | * the tree logging code from jumping in and changing any | |
2145 | * of the trees. | |
2146 | * | |
2147 | * At this point in the commit, there can't be any tree-log | |
2148 | * writers, but a little lower down we drop the trans mutex | |
2149 | * and let new people in. By holding the tree_log_mutex | |
2150 | * from now until after the super is written, we avoid races | |
2151 | * with the tree-log code. | |
2152 | */ | |
0b246afa | 2153 | mutex_lock(&fs_info->tree_log_mutex); |
e02119d5 | 2154 | |
0b246afa | 2155 | ret = commit_fs_roots(trans, fs_info); |
49b25e05 | 2156 | if (ret) { |
0b246afa JM |
2157 | mutex_unlock(&fs_info->tree_log_mutex); |
2158 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2159 | goto scrub_continue; |
49b25e05 | 2160 | } |
54aa1f4d | 2161 | |
3818aea2 | 2162 | /* |
7e1876ac DS |
2163 | * Since the transaction is done, we can apply the pending changes |
2164 | * before the next transaction. | |
3818aea2 | 2165 | */ |
0b246afa | 2166 | btrfs_apply_pending_changes(fs_info); |
3818aea2 | 2167 | |
5d4f98a2 | 2168 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2169 | * safe to free the root of tree log roots |
2170 | */ | |
0b246afa | 2171 | btrfs_free_log_root_tree(trans, fs_info); |
e02119d5 | 2172 | |
0ed4792a QW |
2173 | /* |
2174 | * Since fs roots are all committed, we can get a quite accurate | |
2175 | * new_roots. So let's do quota accounting. | |
2176 | */ | |
0b246afa | 2177 | ret = btrfs_qgroup_account_extents(trans, fs_info); |
0ed4792a | 2178 | if (ret < 0) { |
0b246afa JM |
2179 | mutex_unlock(&fs_info->tree_log_mutex); |
2180 | mutex_unlock(&fs_info->reloc_mutex); | |
0ed4792a QW |
2181 | goto scrub_continue; |
2182 | } | |
2183 | ||
5d4f98a2 | 2184 | ret = commit_cowonly_roots(trans, root); |
49b25e05 | 2185 | if (ret) { |
0b246afa JM |
2186 | mutex_unlock(&fs_info->tree_log_mutex); |
2187 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2188 | goto scrub_continue; |
49b25e05 | 2189 | } |
54aa1f4d | 2190 | |
2cba30f1 MX |
2191 | /* |
2192 | * The tasks which save the space cache and inode cache may also | |
2193 | * update ->aborted, check it. | |
2194 | */ | |
2195 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
2196 | ret = cur_trans->aborted; | |
0b246afa JM |
2197 | mutex_unlock(&fs_info->tree_log_mutex); |
2198 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2199 | goto scrub_continue; |
2cba30f1 MX |
2200 | } |
2201 | ||
11833d66 YZ |
2202 | btrfs_prepare_extent_commit(trans, root); |
2203 | ||
0b246afa | 2204 | cur_trans = fs_info->running_transaction; |
5d4f98a2 | 2205 | |
0b246afa JM |
2206 | btrfs_set_root_node(&fs_info->tree_root->root_item, |
2207 | fs_info->tree_root->node); | |
2208 | list_add_tail(&fs_info->tree_root->dirty_list, | |
9e351cc8 | 2209 | &cur_trans->switch_commits); |
5d4f98a2 | 2210 | |
0b246afa JM |
2211 | btrfs_set_root_node(&fs_info->chunk_root->root_item, |
2212 | fs_info->chunk_root->node); | |
2213 | list_add_tail(&fs_info->chunk_root->dirty_list, | |
9e351cc8 JB |
2214 | &cur_trans->switch_commits); |
2215 | ||
0b246afa | 2216 | switch_commit_roots(cur_trans, fs_info); |
5d4f98a2 | 2217 | |
edf39272 | 2218 | assert_qgroups_uptodate(trans); |
ce93ec54 | 2219 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2220 | ASSERT(list_empty(&cur_trans->io_bgs)); |
5d4f98a2 | 2221 | update_super_roots(root); |
e02119d5 | 2222 | |
0b246afa JM |
2223 | btrfs_set_super_log_root(fs_info->super_copy, 0); |
2224 | btrfs_set_super_log_root_level(fs_info->super_copy, 0); | |
2225 | memcpy(fs_info->super_for_commit, fs_info->super_copy, | |
2226 | sizeof(*fs_info->super_copy)); | |
ccd467d6 | 2227 | |
0b246afa | 2228 | btrfs_update_commit_device_size(fs_info); |
ce7213c7 | 2229 | btrfs_update_commit_device_bytes_used(root, cur_trans); |
935e5cc9 | 2230 | |
0b246afa JM |
2231 | clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
2232 | clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); | |
656f30db | 2233 | |
4fbcdf66 FM |
2234 | btrfs_trans_release_chunk_metadata(trans); |
2235 | ||
0b246afa | 2236 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2237 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
0b246afa JM |
2238 | fs_info->running_transaction = NULL; |
2239 | spin_unlock(&fs_info->trans_lock); | |
2240 | mutex_unlock(&fs_info->reloc_mutex); | |
b7ec40d7 | 2241 | |
0b246afa | 2242 | wake_up(&fs_info->transaction_wait); |
e6dcd2dc | 2243 | |
79154b1b | 2244 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 | 2245 | if (ret) { |
0b246afa JM |
2246 | btrfs_handle_fs_error(fs_info, ret, |
2247 | "Error while writing out transaction"); | |
2248 | mutex_unlock(&fs_info->tree_log_mutex); | |
6cf7f77e | 2249 | goto scrub_continue; |
49b25e05 JM |
2250 | } |
2251 | ||
2252 | ret = write_ctree_super(trans, root, 0); | |
2253 | if (ret) { | |
0b246afa | 2254 | mutex_unlock(&fs_info->tree_log_mutex); |
6cf7f77e | 2255 | goto scrub_continue; |
49b25e05 | 2256 | } |
4313b399 | 2257 | |
e02119d5 CM |
2258 | /* |
2259 | * the super is written, we can safely allow the tree-loggers | |
2260 | * to go about their business | |
2261 | */ | |
0b246afa | 2262 | mutex_unlock(&fs_info->tree_log_mutex); |
e02119d5 | 2263 | |
11833d66 | 2264 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 2265 | |
3204d33c | 2266 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) |
0b246afa | 2267 | btrfs_clear_space_info_full(fs_info); |
13212b54 | 2268 | |
0b246afa | 2269 | fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
2270 | /* |
2271 | * We needn't acquire the lock here because there is no other task | |
2272 | * which can change it. | |
2273 | */ | |
2274 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2275 | wake_up(&cur_trans->commit_wait); |
3de4586c | 2276 | |
0b246afa | 2277 | spin_lock(&fs_info->trans_lock); |
13c5a93e | 2278 | list_del_init(&cur_trans->list); |
0b246afa | 2279 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 2280 | |
724e2315 JB |
2281 | btrfs_put_transaction(cur_trans); |
2282 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2283 | |
0860adfd | 2284 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2285 | sb_end_intwrite(fs_info->sb); |
b2b5ef5c | 2286 | |
1abe9b8a | 2287 | trace_btrfs_transaction_commit(root); |
2288 | ||
a2de733c AJ |
2289 | btrfs_scrub_continue(root); |
2290 | ||
9ed74f2d JB |
2291 | if (current->journal_info == trans) |
2292 | current->journal_info = NULL; | |
2293 | ||
2c90e5d6 | 2294 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2295 | |
9e7cc91a WX |
2296 | /* |
2297 | * If fs has been frozen, we can not handle delayed iputs, otherwise | |
2298 | * it'll result in deadlock about SB_FREEZE_FS. | |
2299 | */ | |
0b246afa JM |
2300 | if (current != fs_info->transaction_kthread && |
2301 | current != fs_info->cleaner_kthread && !fs_info->fs_frozen) | |
24bbcf04 YZ |
2302 | btrfs_run_delayed_iputs(root); |
2303 | ||
79154b1b | 2304 | return ret; |
49b25e05 | 2305 | |
6cf7f77e WS |
2306 | scrub_continue: |
2307 | btrfs_scrub_continue(root); | |
49b25e05 | 2308 | cleanup_transaction: |
0e721106 | 2309 | btrfs_trans_release_metadata(trans, root); |
4fbcdf66 | 2310 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2311 | trans->block_rsv = NULL; |
0b246afa | 2312 | btrfs_warn(fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2313 | if (current->journal_info == trans) |
2314 | current->journal_info = NULL; | |
7b8b92af | 2315 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
2316 | |
2317 | return ret; | |
79154b1b CM |
2318 | } |
2319 | ||
d352ac68 | 2320 | /* |
9d1a2a3a DS |
2321 | * return < 0 if error |
2322 | * 0 if there are no more dead_roots at the time of call | |
2323 | * 1 there are more to be processed, call me again | |
2324 | * | |
2325 | * The return value indicates there are certainly more snapshots to delete, but | |
2326 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2327 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2328 | * few seconds later. | |
d352ac68 | 2329 | */ |
9d1a2a3a | 2330 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 2331 | { |
9d1a2a3a | 2332 | int ret; |
5d4f98a2 YZ |
2333 | struct btrfs_fs_info *fs_info = root->fs_info; |
2334 | ||
a4abeea4 | 2335 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2336 | if (list_empty(&fs_info->dead_roots)) { |
2337 | spin_unlock(&fs_info->trans_lock); | |
2338 | return 0; | |
2339 | } | |
2340 | root = list_first_entry(&fs_info->dead_roots, | |
2341 | struct btrfs_root, root_list); | |
cfad392b | 2342 | list_del_init(&root->root_list); |
a4abeea4 | 2343 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2344 | |
ab8d0fc4 | 2345 | btrfs_debug(fs_info, "cleaner removing %llu", root->objectid); |
76dda93c | 2346 | |
9d1a2a3a | 2347 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2348 | |
9d1a2a3a DS |
2349 | if (btrfs_header_backref_rev(root->node) < |
2350 | BTRFS_MIXED_BACKREF_REV) | |
2351 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
2352 | else | |
2353 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
32471dc2 | 2354 | |
6596a928 | 2355 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2356 | } |
572d9ab7 DS |
2357 | |
2358 | void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) | |
2359 | { | |
2360 | unsigned long prev; | |
2361 | unsigned long bit; | |
2362 | ||
6c9fe14f | 2363 | prev = xchg(&fs_info->pending_changes, 0); |
572d9ab7 DS |
2364 | if (!prev) |
2365 | return; | |
2366 | ||
7e1876ac DS |
2367 | bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE; |
2368 | if (prev & bit) | |
2369 | btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2370 | prev &= ~bit; | |
2371 | ||
2372 | bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE; | |
2373 | if (prev & bit) | |
2374 | btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2375 | prev &= ~bit; | |
2376 | ||
d51033d0 DS |
2377 | bit = 1 << BTRFS_PENDING_COMMIT; |
2378 | if (prev & bit) | |
2379 | btrfs_debug(fs_info, "pending commit done"); | |
2380 | prev &= ~bit; | |
2381 | ||
572d9ab7 DS |
2382 | if (prev) |
2383 | btrfs_warn(fs_info, | |
2384 | "unknown pending changes left 0x%lx, ignoring", prev); | |
2385 | } |