Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 CM |
4 | */ |
5 | ||
79154b1b | 6 | #include <linux/fs.h> |
5a0e3ad6 | 7 | #include <linux/slab.h> |
34088780 | 8 | #include <linux/sched.h> |
d3c2fdcf | 9 | #include <linux/writeback.h> |
5f39d397 | 10 | #include <linux/pagemap.h> |
5f2cc086 | 11 | #include <linux/blkdev.h> |
8ea05e3a | 12 | #include <linux/uuid.h> |
602cbe91 | 13 | #include "misc.h" |
79154b1b CM |
14 | #include "ctree.h" |
15 | #include "disk-io.h" | |
16 | #include "transaction.h" | |
925baedd | 17 | #include "locking.h" |
e02119d5 | 18 | #include "tree-log.h" |
733f4fbb | 19 | #include "volumes.h" |
8dabb742 | 20 | #include "dev-replace.h" |
fcebe456 | 21 | #include "qgroup.h" |
aac0023c | 22 | #include "block-group.h" |
9c343784 | 23 | #include "space-info.h" |
79154b1b | 24 | |
0f7d52f4 CM |
25 | #define BTRFS_ROOT_TRANS_TAG 0 |
26 | ||
61c047b5 QW |
27 | /* |
28 | * Transaction states and transitions | |
29 | * | |
30 | * No running transaction (fs tree blocks are not modified) | |
31 | * | | |
32 | * | To next stage: | |
33 | * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). | |
34 | * V | |
35 | * Transaction N [[TRANS_STATE_RUNNING]] | |
36 | * | | |
37 | * | New trans handles can be attached to transaction N by calling all | |
38 | * | start_transaction() variants. | |
39 | * | | |
40 | * | To next stage: | |
41 | * | Call btrfs_commit_transaction() on any trans handle attached to | |
42 | * | transaction N | |
43 | * V | |
44 | * Transaction N [[TRANS_STATE_COMMIT_START]] | |
45 | * | | |
46 | * | Will wait for previous running transaction to completely finish if there | |
47 | * | is one | |
48 | * | | |
49 | * | Then one of the following happes: | |
50 | * | - Wait for all other trans handle holders to release. | |
51 | * | The btrfs_commit_transaction() caller will do the commit work. | |
52 | * | - Wait for current transaction to be committed by others. | |
53 | * | Other btrfs_commit_transaction() caller will do the commit work. | |
54 | * | | |
55 | * | At this stage, only btrfs_join_transaction*() variants can attach | |
56 | * | to this running transaction. | |
57 | * | All other variants will wait for current one to finish and attach to | |
58 | * | transaction N+1. | |
59 | * | | |
60 | * | To next stage: | |
61 | * | Caller is chosen to commit transaction N, and all other trans handle | |
62 | * | haven been released. | |
63 | * V | |
64 | * Transaction N [[TRANS_STATE_COMMIT_DOING]] | |
65 | * | | |
66 | * | The heavy lifting transaction work is started. | |
67 | * | From running delayed refs (modifying extent tree) to creating pending | |
68 | * | snapshots, running qgroups. | |
69 | * | In short, modify supporting trees to reflect modifications of subvolume | |
70 | * | trees. | |
71 | * | | |
72 | * | At this stage, all start_transaction() calls will wait for this | |
73 | * | transaction to finish and attach to transaction N+1. | |
74 | * | | |
75 | * | To next stage: | |
76 | * | Until all supporting trees are updated. | |
77 | * V | |
78 | * Transaction N [[TRANS_STATE_UNBLOCKED]] | |
79 | * | Transaction N+1 | |
80 | * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] | |
81 | * | need to write them back to disk and update | | |
82 | * | super blocks. | | |
83 | * | | | |
84 | * | At this stage, new transaction is allowed to | | |
85 | * | start. | | |
86 | * | All new start_transaction() calls will be | | |
87 | * | attached to transid N+1. | | |
88 | * | | | |
89 | * | To next stage: | | |
90 | * | Until all tree blocks are super blocks are | | |
91 | * | written to block devices | | |
92 | * V | | |
93 | * Transaction N [[TRANS_STATE_COMPLETED]] V | |
94 | * All tree blocks and super blocks are written. Transaction N+1 | |
95 | * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] | |
96 | * data structures will be cleaned up. | Life goes on | |
97 | */ | |
e8c9f186 | 98 | static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { |
4a9d8bde | 99 | [TRANS_STATE_RUNNING] = 0U, |
bcf3a3e7 NB |
100 | [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), |
101 | [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | | |
4a9d8bde | 102 | __TRANS_ATTACH | |
a6d155d2 FM |
103 | __TRANS_JOIN | |
104 | __TRANS_JOIN_NOSTART), | |
bcf3a3e7 | 105 | [TRANS_STATE_UNBLOCKED] = (__TRANS_START | |
4a9d8bde MX |
106 | __TRANS_ATTACH | |
107 | __TRANS_JOIN | | |
a6d155d2 FM |
108 | __TRANS_JOIN_NOLOCK | |
109 | __TRANS_JOIN_NOSTART), | |
bcf3a3e7 | 110 | [TRANS_STATE_COMPLETED] = (__TRANS_START | |
4a9d8bde MX |
111 | __TRANS_ATTACH | |
112 | __TRANS_JOIN | | |
a6d155d2 FM |
113 | __TRANS_JOIN_NOLOCK | |
114 | __TRANS_JOIN_NOSTART), | |
4a9d8bde MX |
115 | }; |
116 | ||
724e2315 | 117 | void btrfs_put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 118 | { |
9b64f57d ER |
119 | WARN_ON(refcount_read(&transaction->use_count) == 0); |
120 | if (refcount_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 121 | BUG_ON(!list_empty(&transaction->list)); |
5c9d028b LB |
122 | WARN_ON(!RB_EMPTY_ROOT( |
123 | &transaction->delayed_refs.href_root.rb_root)); | |
81f7eb00 JM |
124 | WARN_ON(!RB_EMPTY_ROOT( |
125 | &transaction->delayed_refs.dirty_extent_root)); | |
1262133b | 126 | if (transaction->delayed_refs.pending_csums) |
ab8d0fc4 JM |
127 | btrfs_err(transaction->fs_info, |
128 | "pending csums is %llu", | |
129 | transaction->delayed_refs.pending_csums); | |
7785a663 FM |
130 | /* |
131 | * If any block groups are found in ->deleted_bgs then it's | |
132 | * because the transaction was aborted and a commit did not | |
133 | * happen (things failed before writing the new superblock | |
134 | * and calling btrfs_finish_extent_commit()), so we can not | |
135 | * discard the physical locations of the block groups. | |
136 | */ | |
137 | while (!list_empty(&transaction->deleted_bgs)) { | |
32da5386 | 138 | struct btrfs_block_group *cache; |
7785a663 FM |
139 | |
140 | cache = list_first_entry(&transaction->deleted_bgs, | |
32da5386 | 141 | struct btrfs_block_group, |
7785a663 FM |
142 | bg_list); |
143 | list_del_init(&cache->bg_list); | |
6b7304af | 144 | btrfs_unfreeze_block_group(cache); |
7785a663 FM |
145 | btrfs_put_block_group(cache); |
146 | } | |
bbbf7243 | 147 | WARN_ON(!list_empty(&transaction->dev_update_list)); |
4b5faeac | 148 | kfree(transaction); |
78fae27e | 149 | } |
79154b1b CM |
150 | } |
151 | ||
889bfa39 | 152 | static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) |
817d52f8 | 153 | { |
889bfa39 | 154 | struct btrfs_transaction *cur_trans = trans->transaction; |
16916a88 | 155 | struct btrfs_fs_info *fs_info = trans->fs_info; |
9e351cc8 | 156 | struct btrfs_root *root, *tmp; |
27d56e62 | 157 | struct btrfs_caching_control *caching_ctl, *next; |
9e351cc8 JB |
158 | |
159 | down_write(&fs_info->commit_root_sem); | |
889bfa39 | 160 | list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, |
9e351cc8 JB |
161 | dirty_list) { |
162 | list_del_init(&root->dirty_list); | |
163 | free_extent_buffer(root->commit_root); | |
164 | root->commit_root = btrfs_root_node(root); | |
41e7acd3 | 165 | extent_io_tree_release(&root->dirty_log_pages); |
370a11b8 | 166 | btrfs_qgroup_clean_swapped_blocks(root); |
9e351cc8 | 167 | } |
2b9dbef2 JB |
168 | |
169 | /* We can free old roots now. */ | |
889bfa39 JB |
170 | spin_lock(&cur_trans->dropped_roots_lock); |
171 | while (!list_empty(&cur_trans->dropped_roots)) { | |
172 | root = list_first_entry(&cur_trans->dropped_roots, | |
2b9dbef2 JB |
173 | struct btrfs_root, root_list); |
174 | list_del_init(&root->root_list); | |
889bfa39 JB |
175 | spin_unlock(&cur_trans->dropped_roots_lock); |
176 | btrfs_free_log(trans, root); | |
2b9dbef2 | 177 | btrfs_drop_and_free_fs_root(fs_info, root); |
889bfa39 | 178 | spin_lock(&cur_trans->dropped_roots_lock); |
2b9dbef2 | 179 | } |
889bfa39 | 180 | spin_unlock(&cur_trans->dropped_roots_lock); |
27d56e62 JB |
181 | |
182 | /* | |
183 | * We have to update the last_byte_to_unpin under the commit_root_sem, | |
184 | * at the same time we swap out the commit roots. | |
185 | * | |
186 | * This is because we must have a real view of the last spot the caching | |
187 | * kthreads were while caching. Consider the following views of the | |
188 | * extent tree for a block group | |
189 | * | |
190 | * commit root | |
191 | * +----+----+----+----+----+----+----+ | |
192 | * |\\\\| |\\\\|\\\\| |\\\\|\\\\| | |
193 | * +----+----+----+----+----+----+----+ | |
194 | * 0 1 2 3 4 5 6 7 | |
195 | * | |
196 | * new commit root | |
197 | * +----+----+----+----+----+----+----+ | |
198 | * | | | |\\\\| | |\\\\| | |
199 | * +----+----+----+----+----+----+----+ | |
200 | * 0 1 2 3 4 5 6 7 | |
201 | * | |
202 | * If the cache_ctl->progress was at 3, then we are only allowed to | |
203 | * unpin [0,1) and [2,3], because the caching thread has already | |
204 | * processed those extents. We are not allowed to unpin [5,6), because | |
205 | * the caching thread will re-start it's search from 3, and thus find | |
206 | * the hole from [4,6) to add to the free space cache. | |
207 | */ | |
bbb86a37 | 208 | spin_lock(&fs_info->block_group_cache_lock); |
27d56e62 JB |
209 | list_for_each_entry_safe(caching_ctl, next, |
210 | &fs_info->caching_block_groups, list) { | |
211 | struct btrfs_block_group *cache = caching_ctl->block_group; | |
212 | ||
213 | if (btrfs_block_group_done(cache)) { | |
214 | cache->last_byte_to_unpin = (u64)-1; | |
215 | list_del_init(&caching_ctl->list); | |
216 | btrfs_put_caching_control(caching_ctl); | |
217 | } else { | |
218 | cache->last_byte_to_unpin = caching_ctl->progress; | |
219 | } | |
220 | } | |
bbb86a37 | 221 | spin_unlock(&fs_info->block_group_cache_lock); |
9e351cc8 | 222 | up_write(&fs_info->commit_root_sem); |
817d52f8 JB |
223 | } |
224 | ||
0860adfd MX |
225 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
226 | unsigned int type) | |
227 | { | |
228 | if (type & TRANS_EXTWRITERS) | |
229 | atomic_inc(&trans->num_extwriters); | |
230 | } | |
231 | ||
232 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
233 | unsigned int type) | |
234 | { | |
235 | if (type & TRANS_EXTWRITERS) | |
236 | atomic_dec(&trans->num_extwriters); | |
237 | } | |
238 | ||
239 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
240 | unsigned int type) | |
241 | { | |
242 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
243 | } | |
244 | ||
245 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
246 | { | |
247 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
248 | } |
249 | ||
fb6dea26 JB |
250 | /* |
251 | * To be called after all the new block groups attached to the transaction | |
252 | * handle have been created (btrfs_create_pending_block_groups()). | |
253 | */ | |
254 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
255 | { | |
256 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
257 | ||
258 | if (!trans->chunk_bytes_reserved) | |
259 | return; | |
260 | ||
261 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
262 | ||
263 | btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, | |
63f018be | 264 | trans->chunk_bytes_reserved, NULL); |
fb6dea26 JB |
265 | trans->chunk_bytes_reserved = 0; |
266 | } | |
267 | ||
d352ac68 CM |
268 | /* |
269 | * either allocate a new transaction or hop into the existing one | |
270 | */ | |
2ff7e61e JM |
271 | static noinline int join_transaction(struct btrfs_fs_info *fs_info, |
272 | unsigned int type) | |
79154b1b CM |
273 | { |
274 | struct btrfs_transaction *cur_trans; | |
a4abeea4 | 275 | |
19ae4e81 | 276 | spin_lock(&fs_info->trans_lock); |
d43317dc | 277 | loop: |
49b25e05 | 278 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 279 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 280 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
281 | return -EROFS; |
282 | } | |
283 | ||
19ae4e81 | 284 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 285 | if (cur_trans) { |
bf31f87f | 286 | if (TRANS_ABORTED(cur_trans)) { |
19ae4e81 | 287 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 288 | return cur_trans->aborted; |
871383be | 289 | } |
4a9d8bde | 290 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
291 | spin_unlock(&fs_info->trans_lock); |
292 | return -EBUSY; | |
293 | } | |
9b64f57d | 294 | refcount_inc(&cur_trans->use_count); |
13c5a93e | 295 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 296 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 297 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 298 | return 0; |
79154b1b | 299 | } |
19ae4e81 | 300 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 301 | |
354aa0fb MX |
302 | /* |
303 | * If we are ATTACH, we just want to catch the current transaction, | |
304 | * and commit it. If there is no transaction, just return ENOENT. | |
305 | */ | |
306 | if (type == TRANS_ATTACH) | |
307 | return -ENOENT; | |
308 | ||
4a9d8bde MX |
309 | /* |
310 | * JOIN_NOLOCK only happens during the transaction commit, so | |
311 | * it is impossible that ->running_transaction is NULL | |
312 | */ | |
313 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
314 | ||
4b5faeac | 315 | cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); |
a4abeea4 JB |
316 | if (!cur_trans) |
317 | return -ENOMEM; | |
d43317dc | 318 | |
19ae4e81 JS |
319 | spin_lock(&fs_info->trans_lock); |
320 | if (fs_info->running_transaction) { | |
d43317dc CM |
321 | /* |
322 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 323 | * to redo the checks above |
d43317dc | 324 | */ |
4b5faeac | 325 | kfree(cur_trans); |
d43317dc | 326 | goto loop; |
87533c47 | 327 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 328 | spin_unlock(&fs_info->trans_lock); |
4b5faeac | 329 | kfree(cur_trans); |
7b8b92af | 330 | return -EROFS; |
79154b1b | 331 | } |
d43317dc | 332 | |
ab8d0fc4 | 333 | cur_trans->fs_info = fs_info; |
48778179 FM |
334 | atomic_set(&cur_trans->pending_ordered, 0); |
335 | init_waitqueue_head(&cur_trans->pending_wait); | |
a4abeea4 | 336 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 337 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
338 | init_waitqueue_head(&cur_trans->writer_wait); |
339 | init_waitqueue_head(&cur_trans->commit_wait); | |
4a9d8bde | 340 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
341 | /* |
342 | * One for this trans handle, one so it will live on until we | |
343 | * commit the transaction. | |
344 | */ | |
9b64f57d | 345 | refcount_set(&cur_trans->use_count, 2); |
3204d33c | 346 | cur_trans->flags = 0; |
afd48513 | 347 | cur_trans->start_time = ktime_get_seconds(); |
a4abeea4 | 348 | |
a099d0fd AM |
349 | memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); |
350 | ||
5c9d028b | 351 | cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; |
3368d001 | 352 | cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; |
d7df2c79 | 353 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
20b297d6 JS |
354 | |
355 | /* | |
356 | * although the tree mod log is per file system and not per transaction, | |
357 | * the log must never go across transaction boundaries. | |
358 | */ | |
359 | smp_mb(); | |
31b1a2bd | 360 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
5d163e0e | 361 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); |
31b1a2bd | 362 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
5d163e0e | 363 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); |
fc36ed7e | 364 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 365 | |
a4abeea4 JB |
366 | spin_lock_init(&cur_trans->delayed_refs.lock); |
367 | ||
368 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
bbbf7243 | 369 | INIT_LIST_HEAD(&cur_trans->dev_update_list); |
9e351cc8 | 370 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
ce93ec54 | 371 | INIT_LIST_HEAD(&cur_trans->dirty_bgs); |
1bbc621e | 372 | INIT_LIST_HEAD(&cur_trans->io_bgs); |
2b9dbef2 | 373 | INIT_LIST_HEAD(&cur_trans->dropped_roots); |
1bbc621e | 374 | mutex_init(&cur_trans->cache_write_mutex); |
ce93ec54 | 375 | spin_lock_init(&cur_trans->dirty_bgs_lock); |
e33e17ee | 376 | INIT_LIST_HEAD(&cur_trans->deleted_bgs); |
2b9dbef2 | 377 | spin_lock_init(&cur_trans->dropped_roots_lock); |
19ae4e81 | 378 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
c258d6e3 | 379 | extent_io_tree_init(fs_info, &cur_trans->dirty_pages, |
43eb5f29 | 380 | IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode); |
fe119a6e NB |
381 | extent_io_tree_init(fs_info, &cur_trans->pinned_extents, |
382 | IO_TREE_FS_PINNED_EXTENTS, NULL); | |
19ae4e81 JS |
383 | fs_info->generation++; |
384 | cur_trans->transid = fs_info->generation; | |
385 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 386 | cur_trans->aborted = 0; |
19ae4e81 | 387 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 388 | |
79154b1b CM |
389 | return 0; |
390 | } | |
391 | ||
d352ac68 | 392 | /* |
92a7cc42 QW |
393 | * This does all the record keeping required to make sure that a shareable root |
394 | * is properly recorded in a given transaction. This is required to make sure | |
395 | * the old root from before we joined the transaction is deleted when the | |
396 | * transaction commits. | |
d352ac68 | 397 | */ |
7585717f | 398 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
6426c7ad QW |
399 | struct btrfs_root *root, |
400 | int force) | |
6702ed49 | 401 | { |
0b246afa JM |
402 | struct btrfs_fs_info *fs_info = root->fs_info; |
403 | ||
92a7cc42 | 404 | if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
6426c7ad | 405 | root->last_trans < trans->transid) || force) { |
0b246afa | 406 | WARN_ON(root == fs_info->extent_root); |
4d31778a | 407 | WARN_ON(!force && root->commit_root != root->node); |
5d4f98a2 | 408 | |
7585717f | 409 | /* |
27cdeb70 | 410 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
411 | * we have the reloc mutex held now, so there |
412 | * is only one writer in this function | |
413 | */ | |
27cdeb70 | 414 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 415 | |
27cdeb70 | 416 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
417 | * they find our root->last_trans update |
418 | */ | |
419 | smp_wmb(); | |
420 | ||
0b246afa | 421 | spin_lock(&fs_info->fs_roots_radix_lock); |
6426c7ad | 422 | if (root->last_trans == trans->transid && !force) { |
0b246afa | 423 | spin_unlock(&fs_info->fs_roots_radix_lock); |
a4abeea4 JB |
424 | return 0; |
425 | } | |
0b246afa JM |
426 | radix_tree_tag_set(&fs_info->fs_roots_radix, |
427 | (unsigned long)root->root_key.objectid, | |
428 | BTRFS_ROOT_TRANS_TAG); | |
429 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
7585717f CM |
430 | root->last_trans = trans->transid; |
431 | ||
432 | /* this is pretty tricky. We don't want to | |
433 | * take the relocation lock in btrfs_record_root_in_trans | |
434 | * unless we're really doing the first setup for this root in | |
435 | * this transaction. | |
436 | * | |
437 | * Normally we'd use root->last_trans as a flag to decide | |
438 | * if we want to take the expensive mutex. | |
439 | * | |
440 | * But, we have to set root->last_trans before we | |
441 | * init the relocation root, otherwise, we trip over warnings | |
442 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 443 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
444 | * fixing up the reloc trees and everyone must wait. |
445 | * | |
446 | * When this is zero, they can trust root->last_trans and fly | |
447 | * through btrfs_record_root_in_trans without having to take the | |
448 | * lock. smp_wmb() makes sure that all the writes above are | |
449 | * done before we pop in the zero below | |
450 | */ | |
5d4f98a2 | 451 | btrfs_init_reloc_root(trans, root); |
c7548af6 | 452 | smp_mb__before_atomic(); |
27cdeb70 | 453 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 YZ |
454 | } |
455 | return 0; | |
456 | } | |
bcc63abb | 457 | |
7585717f | 458 | |
2b9dbef2 JB |
459 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
460 | struct btrfs_root *root) | |
461 | { | |
0b246afa | 462 | struct btrfs_fs_info *fs_info = root->fs_info; |
2b9dbef2 JB |
463 | struct btrfs_transaction *cur_trans = trans->transaction; |
464 | ||
465 | /* Add ourselves to the transaction dropped list */ | |
466 | spin_lock(&cur_trans->dropped_roots_lock); | |
467 | list_add_tail(&root->root_list, &cur_trans->dropped_roots); | |
468 | spin_unlock(&cur_trans->dropped_roots_lock); | |
469 | ||
470 | /* Make sure we don't try to update the root at commit time */ | |
0b246afa JM |
471 | spin_lock(&fs_info->fs_roots_radix_lock); |
472 | radix_tree_tag_clear(&fs_info->fs_roots_radix, | |
2b9dbef2 JB |
473 | (unsigned long)root->root_key.objectid, |
474 | BTRFS_ROOT_TRANS_TAG); | |
0b246afa | 475 | spin_unlock(&fs_info->fs_roots_radix_lock); |
2b9dbef2 JB |
476 | } |
477 | ||
7585717f CM |
478 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
479 | struct btrfs_root *root) | |
480 | { | |
0b246afa JM |
481 | struct btrfs_fs_info *fs_info = root->fs_info; |
482 | ||
92a7cc42 | 483 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
7585717f CM |
484 | return 0; |
485 | ||
486 | /* | |
27cdeb70 | 487 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
488 | * and barriers |
489 | */ | |
490 | smp_rmb(); | |
491 | if (root->last_trans == trans->transid && | |
27cdeb70 | 492 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
493 | return 0; |
494 | ||
0b246afa | 495 | mutex_lock(&fs_info->reloc_mutex); |
6426c7ad | 496 | record_root_in_trans(trans, root, 0); |
0b246afa | 497 | mutex_unlock(&fs_info->reloc_mutex); |
7585717f CM |
498 | |
499 | return 0; | |
500 | } | |
501 | ||
4a9d8bde MX |
502 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
503 | { | |
3296bf56 | 504 | return (trans->state >= TRANS_STATE_COMMIT_START && |
501407aa | 505 | trans->state < TRANS_STATE_UNBLOCKED && |
bf31f87f | 506 | !TRANS_ABORTED(trans)); |
4a9d8bde MX |
507 | } |
508 | ||
d352ac68 CM |
509 | /* wait for commit against the current transaction to become unblocked |
510 | * when this is done, it is safe to start a new transaction, but the current | |
511 | * transaction might not be fully on disk. | |
512 | */ | |
2ff7e61e | 513 | static void wait_current_trans(struct btrfs_fs_info *fs_info) |
79154b1b | 514 | { |
f9295749 | 515 | struct btrfs_transaction *cur_trans; |
79154b1b | 516 | |
0b246afa JM |
517 | spin_lock(&fs_info->trans_lock); |
518 | cur_trans = fs_info->running_transaction; | |
4a9d8bde | 519 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
9b64f57d | 520 | refcount_inc(&cur_trans->use_count); |
0b246afa | 521 | spin_unlock(&fs_info->trans_lock); |
72d63ed6 | 522 | |
0b246afa | 523 | wait_event(fs_info->transaction_wait, |
501407aa | 524 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
bf31f87f | 525 | TRANS_ABORTED(cur_trans)); |
724e2315 | 526 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 527 | } else { |
0b246afa | 528 | spin_unlock(&fs_info->trans_lock); |
f9295749 | 529 | } |
37d1aeee CM |
530 | } |
531 | ||
2ff7e61e | 532 | static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) |
a22285a6 | 533 | { |
0b246afa | 534 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
a4abeea4 JB |
535 | return 0; |
536 | ||
92e2f7e3 | 537 | if (type == TRANS_START) |
a22285a6 | 538 | return 1; |
a4abeea4 | 539 | |
a22285a6 YZ |
540 | return 0; |
541 | } | |
542 | ||
20dd2cbf MX |
543 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
544 | { | |
0b246afa JM |
545 | struct btrfs_fs_info *fs_info = root->fs_info; |
546 | ||
547 | if (!fs_info->reloc_ctl || | |
92a7cc42 | 548 | !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || |
20dd2cbf MX |
549 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
550 | root->reloc_root) | |
551 | return false; | |
552 | ||
553 | return true; | |
554 | } | |
555 | ||
08e007d2 | 556 | static struct btrfs_trans_handle * |
5aed1dd8 | 557 | start_transaction(struct btrfs_root *root, unsigned int num_items, |
003d7c59 JM |
558 | unsigned int type, enum btrfs_reserve_flush_enum flush, |
559 | bool enforce_qgroups) | |
37d1aeee | 560 | { |
0b246afa | 561 | struct btrfs_fs_info *fs_info = root->fs_info; |
ba2c4d4e | 562 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
a22285a6 YZ |
563 | struct btrfs_trans_handle *h; |
564 | struct btrfs_transaction *cur_trans; | |
b5009945 | 565 | u64 num_bytes = 0; |
c5567237 | 566 | u64 qgroup_reserved = 0; |
20dd2cbf | 567 | bool reloc_reserved = false; |
9c343784 | 568 | bool do_chunk_alloc = false; |
20dd2cbf | 569 | int ret; |
acce952b | 570 | |
46c4e71e | 571 | /* Send isn't supposed to start transactions. */ |
2755a0de | 572 | ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); |
46c4e71e | 573 | |
0b246afa | 574 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
acce952b | 575 | return ERR_PTR(-EROFS); |
2a1eb461 | 576 | |
46c4e71e | 577 | if (current->journal_info) { |
0860adfd | 578 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 | 579 | h = current->journal_info; |
b50fff81 DS |
580 | refcount_inc(&h->use_count); |
581 | WARN_ON(refcount_read(&h->use_count) > 2); | |
2a1eb461 JB |
582 | h->orig_rsv = h->block_rsv; |
583 | h->block_rsv = NULL; | |
584 | goto got_it; | |
585 | } | |
b5009945 JB |
586 | |
587 | /* | |
588 | * Do the reservation before we join the transaction so we can do all | |
589 | * the appropriate flushing if need be. | |
590 | */ | |
003d7c59 | 591 | if (num_items && root != fs_info->chunk_root) { |
ba2c4d4e JB |
592 | struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; |
593 | u64 delayed_refs_bytes = 0; | |
594 | ||
0b246afa | 595 | qgroup_reserved = num_items * fs_info->nodesize; |
733e03a0 QW |
596 | ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, |
597 | enforce_qgroups); | |
7174109c QW |
598 | if (ret) |
599 | return ERR_PTR(ret); | |
c5567237 | 600 | |
ba2c4d4e JB |
601 | /* |
602 | * We want to reserve all the bytes we may need all at once, so | |
603 | * we only do 1 enospc flushing cycle per transaction start. We | |
604 | * accomplish this by simply assuming we'll do 2 x num_items | |
605 | * worth of delayed refs updates in this trans handle, and | |
606 | * refill that amount for whatever is missing in the reserve. | |
607 | */ | |
2bd36e7b | 608 | num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); |
7f9fe614 JB |
609 | if (flush == BTRFS_RESERVE_FLUSH_ALL && |
610 | delayed_refs_rsv->full == 0) { | |
ba2c4d4e JB |
611 | delayed_refs_bytes = num_bytes; |
612 | num_bytes <<= 1; | |
613 | } | |
614 | ||
20dd2cbf MX |
615 | /* |
616 | * Do the reservation for the relocation root creation | |
617 | */ | |
ee39b432 | 618 | if (need_reserve_reloc_root(root)) { |
0b246afa | 619 | num_bytes += fs_info->nodesize; |
20dd2cbf MX |
620 | reloc_reserved = true; |
621 | } | |
622 | ||
ba2c4d4e JB |
623 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush); |
624 | if (ret) | |
625 | goto reserve_fail; | |
626 | if (delayed_refs_bytes) { | |
627 | btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, | |
628 | delayed_refs_bytes); | |
629 | num_bytes -= delayed_refs_bytes; | |
630 | } | |
9c343784 JB |
631 | |
632 | if (rsv->space_info->force_alloc) | |
633 | do_chunk_alloc = true; | |
ba2c4d4e JB |
634 | } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && |
635 | !delayed_refs_rsv->full) { | |
636 | /* | |
637 | * Some people call with btrfs_start_transaction(root, 0) | |
638 | * because they can be throttled, but have some other mechanism | |
639 | * for reserving space. We still want these guys to refill the | |
640 | * delayed block_rsv so just add 1 items worth of reservation | |
641 | * here. | |
642 | */ | |
643 | ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); | |
b5009945 | 644 | if (ret) |
843fcf35 | 645 | goto reserve_fail; |
b5009945 | 646 | } |
a22285a6 | 647 | again: |
f2f767e7 | 648 | h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); |
843fcf35 MX |
649 | if (!h) { |
650 | ret = -ENOMEM; | |
651 | goto alloc_fail; | |
652 | } | |
37d1aeee | 653 | |
98114659 JB |
654 | /* |
655 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
656 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
657 | * because we're already holding a ref. We need this because we could | |
658 | * have raced in and did an fsync() on a file which can kick a commit | |
659 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
660 | * |
661 | * If we are ATTACH, it means we just want to catch the current | |
662 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 663 | */ |
0860adfd | 664 | if (type & __TRANS_FREEZABLE) |
0b246afa | 665 | sb_start_intwrite(fs_info->sb); |
b2b5ef5c | 666 | |
2ff7e61e JM |
667 | if (may_wait_transaction(fs_info, type)) |
668 | wait_current_trans(fs_info); | |
a22285a6 | 669 | |
a4abeea4 | 670 | do { |
2ff7e61e | 671 | ret = join_transaction(fs_info, type); |
178260b2 | 672 | if (ret == -EBUSY) { |
2ff7e61e | 673 | wait_current_trans(fs_info); |
a6d155d2 FM |
674 | if (unlikely(type == TRANS_ATTACH || |
675 | type == TRANS_JOIN_NOSTART)) | |
178260b2 MX |
676 | ret = -ENOENT; |
677 | } | |
a4abeea4 JB |
678 | } while (ret == -EBUSY); |
679 | ||
a43f7f82 | 680 | if (ret < 0) |
843fcf35 | 681 | goto join_fail; |
0f7d52f4 | 682 | |
0b246afa | 683 | cur_trans = fs_info->running_transaction; |
a22285a6 YZ |
684 | |
685 | h->transid = cur_trans->transid; | |
686 | h->transaction = cur_trans; | |
d13603ef | 687 | h->root = root; |
b50fff81 | 688 | refcount_set(&h->use_count, 1); |
64b63580 | 689 | h->fs_info = root->fs_info; |
7174109c | 690 | |
a698d075 | 691 | h->type = type; |
d9a0540a | 692 | h->can_flush_pending_bgs = true; |
ea658bad | 693 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 694 | |
a22285a6 | 695 | smp_mb(); |
3296bf56 | 696 | if (cur_trans->state >= TRANS_STATE_COMMIT_START && |
2ff7e61e | 697 | may_wait_transaction(fs_info, type)) { |
abdd2e80 | 698 | current->journal_info = h; |
3a45bb20 | 699 | btrfs_commit_transaction(h); |
a22285a6 YZ |
700 | goto again; |
701 | } | |
702 | ||
b5009945 | 703 | if (num_bytes) { |
0b246afa | 704 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 705 | h->transid, num_bytes, 1); |
0b246afa | 706 | h->block_rsv = &fs_info->trans_block_rsv; |
b5009945 | 707 | h->bytes_reserved = num_bytes; |
20dd2cbf | 708 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 709 | } |
9ed74f2d | 710 | |
2a1eb461 | 711 | got_it: |
bcf3a3e7 | 712 | if (!current->journal_info) |
a22285a6 | 713 | current->journal_info = h; |
fcc99734 | 714 | |
9c343784 JB |
715 | /* |
716 | * If the space_info is marked ALLOC_FORCE then we'll get upgraded to | |
717 | * ALLOC_FORCE the first run through, and then we won't allocate for | |
718 | * anybody else who races in later. We don't care about the return | |
719 | * value here. | |
720 | */ | |
721 | if (do_chunk_alloc && num_bytes) { | |
722 | u64 flags = h->block_rsv->space_info->flags; | |
723 | ||
724 | btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), | |
725 | CHUNK_ALLOC_NO_FORCE); | |
726 | } | |
727 | ||
fcc99734 QW |
728 | /* |
729 | * btrfs_record_root_in_trans() needs to alloc new extents, and may | |
730 | * call btrfs_join_transaction() while we're also starting a | |
731 | * transaction. | |
732 | * | |
733 | * Thus it need to be called after current->journal_info initialized, | |
734 | * or we can deadlock. | |
735 | */ | |
736 | btrfs_record_root_in_trans(h, root); | |
737 | ||
79154b1b | 738 | return h; |
843fcf35 MX |
739 | |
740 | join_fail: | |
0860adfd | 741 | if (type & __TRANS_FREEZABLE) |
0b246afa | 742 | sb_end_intwrite(fs_info->sb); |
843fcf35 MX |
743 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
744 | alloc_fail: | |
745 | if (num_bytes) | |
2ff7e61e | 746 | btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, |
63f018be | 747 | num_bytes, NULL); |
843fcf35 | 748 | reserve_fail: |
733e03a0 | 749 | btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); |
843fcf35 | 750 | return ERR_PTR(ret); |
79154b1b CM |
751 | } |
752 | ||
f9295749 | 753 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
5aed1dd8 | 754 | unsigned int num_items) |
f9295749 | 755 | { |
08e007d2 | 756 | return start_transaction(root, num_items, TRANS_START, |
003d7c59 | 757 | BTRFS_RESERVE_FLUSH_ALL, true); |
f9295749 | 758 | } |
003d7c59 | 759 | |
8eab77ff FM |
760 | struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( |
761 | struct btrfs_root *root, | |
7f9fe614 | 762 | unsigned int num_items) |
8eab77ff | 763 | { |
7f9fe614 JB |
764 | return start_transaction(root, num_items, TRANS_START, |
765 | BTRFS_RESERVE_FLUSH_ALL_STEAL, false); | |
8eab77ff | 766 | } |
8407aa46 | 767 | |
7a7eaa40 | 768 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 769 | { |
003d7c59 JM |
770 | return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, |
771 | true); | |
f9295749 CM |
772 | } |
773 | ||
8d510121 | 774 | struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) |
0af3d00b | 775 | { |
575a75d6 | 776 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, |
003d7c59 | 777 | BTRFS_RESERVE_NO_FLUSH, true); |
0af3d00b JB |
778 | } |
779 | ||
a6d155d2 FM |
780 | /* |
781 | * Similar to regular join but it never starts a transaction when none is | |
782 | * running or after waiting for the current one to finish. | |
783 | */ | |
784 | struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) | |
785 | { | |
786 | return start_transaction(root, 0, TRANS_JOIN_NOSTART, | |
787 | BTRFS_RESERVE_NO_FLUSH, true); | |
788 | } | |
789 | ||
d4edf39b MX |
790 | /* |
791 | * btrfs_attach_transaction() - catch the running transaction | |
792 | * | |
793 | * It is used when we want to commit the current the transaction, but | |
794 | * don't want to start a new one. | |
795 | * | |
796 | * Note: If this function return -ENOENT, it just means there is no | |
797 | * running transaction. But it is possible that the inactive transaction | |
798 | * is still in the memory, not fully on disk. If you hope there is no | |
799 | * inactive transaction in the fs when -ENOENT is returned, you should | |
800 | * invoke | |
801 | * btrfs_attach_transaction_barrier() | |
802 | */ | |
354aa0fb | 803 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 804 | { |
575a75d6 | 805 | return start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 806 | BTRFS_RESERVE_NO_FLUSH, true); |
60376ce4 JB |
807 | } |
808 | ||
d4edf39b | 809 | /* |
90b6d283 | 810 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b | 811 | * |
52042d8e | 812 | * It is similar to the above function, the difference is this one |
d4edf39b MX |
813 | * will wait for all the inactive transactions until they fully |
814 | * complete. | |
815 | */ | |
816 | struct btrfs_trans_handle * | |
817 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
818 | { | |
819 | struct btrfs_trans_handle *trans; | |
820 | ||
575a75d6 | 821 | trans = start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 822 | BTRFS_RESERVE_NO_FLUSH, true); |
8d9e220c | 823 | if (trans == ERR_PTR(-ENOENT)) |
2ff7e61e | 824 | btrfs_wait_for_commit(root->fs_info, 0); |
d4edf39b MX |
825 | |
826 | return trans; | |
827 | } | |
828 | ||
d352ac68 | 829 | /* wait for a transaction commit to be fully complete */ |
2ff7e61e | 830 | static noinline void wait_for_commit(struct btrfs_transaction *commit) |
89ce8a63 | 831 | { |
4a9d8bde | 832 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
833 | } |
834 | ||
2ff7e61e | 835 | int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) |
46204592 SW |
836 | { |
837 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 838 | int ret = 0; |
46204592 | 839 | |
46204592 | 840 | if (transid) { |
0b246afa | 841 | if (transid <= fs_info->last_trans_committed) |
a4abeea4 | 842 | goto out; |
46204592 SW |
843 | |
844 | /* find specified transaction */ | |
0b246afa JM |
845 | spin_lock(&fs_info->trans_lock); |
846 | list_for_each_entry(t, &fs_info->trans_list, list) { | |
46204592 SW |
847 | if (t->transid == transid) { |
848 | cur_trans = t; | |
9b64f57d | 849 | refcount_inc(&cur_trans->use_count); |
8cd2807f | 850 | ret = 0; |
46204592 SW |
851 | break; |
852 | } | |
8cd2807f MX |
853 | if (t->transid > transid) { |
854 | ret = 0; | |
46204592 | 855 | break; |
8cd2807f | 856 | } |
46204592 | 857 | } |
0b246afa | 858 | spin_unlock(&fs_info->trans_lock); |
42383020 SW |
859 | |
860 | /* | |
861 | * The specified transaction doesn't exist, or we | |
862 | * raced with btrfs_commit_transaction | |
863 | */ | |
864 | if (!cur_trans) { | |
0b246afa | 865 | if (transid > fs_info->last_trans_committed) |
42383020 | 866 | ret = -EINVAL; |
8cd2807f | 867 | goto out; |
42383020 | 868 | } |
46204592 SW |
869 | } else { |
870 | /* find newest transaction that is committing | committed */ | |
0b246afa JM |
871 | spin_lock(&fs_info->trans_lock); |
872 | list_for_each_entry_reverse(t, &fs_info->trans_list, | |
46204592 | 873 | list) { |
4a9d8bde MX |
874 | if (t->state >= TRANS_STATE_COMMIT_START) { |
875 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 876 | break; |
46204592 | 877 | cur_trans = t; |
9b64f57d | 878 | refcount_inc(&cur_trans->use_count); |
46204592 SW |
879 | break; |
880 | } | |
881 | } | |
0b246afa | 882 | spin_unlock(&fs_info->trans_lock); |
46204592 | 883 | if (!cur_trans) |
a4abeea4 | 884 | goto out; /* nothing committing|committed */ |
46204592 SW |
885 | } |
886 | ||
2ff7e61e | 887 | wait_for_commit(cur_trans); |
724e2315 | 888 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 889 | out: |
46204592 SW |
890 | return ret; |
891 | } | |
892 | ||
2ff7e61e | 893 | void btrfs_throttle(struct btrfs_fs_info *fs_info) |
37d1aeee | 894 | { |
92e2f7e3 | 895 | wait_current_trans(fs_info); |
37d1aeee CM |
896 | } |
897 | ||
8a8f4dea | 898 | static bool should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa | 899 | { |
2ff7e61e | 900 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0b246afa | 901 | |
64403612 | 902 | if (btrfs_check_space_for_delayed_refs(fs_info)) |
8a8f4dea | 903 | return true; |
36ba022a | 904 | |
2ff7e61e | 905 | return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); |
8929ecfa YZ |
906 | } |
907 | ||
a2633b6a | 908 | bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa YZ |
909 | { |
910 | struct btrfs_transaction *cur_trans = trans->transaction; | |
8929ecfa | 911 | |
3296bf56 | 912 | if (cur_trans->state >= TRANS_STATE_COMMIT_START || |
e19eb11f | 913 | test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) |
a2633b6a | 914 | return true; |
8929ecfa | 915 | |
2ff7e61e | 916 | return should_end_transaction(trans); |
8929ecfa YZ |
917 | } |
918 | ||
dc60c525 NB |
919 | static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) |
920 | ||
0e34693f | 921 | { |
dc60c525 NB |
922 | struct btrfs_fs_info *fs_info = trans->fs_info; |
923 | ||
0e34693f NB |
924 | if (!trans->block_rsv) { |
925 | ASSERT(!trans->bytes_reserved); | |
926 | return; | |
927 | } | |
928 | ||
929 | if (!trans->bytes_reserved) | |
930 | return; | |
931 | ||
932 | ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); | |
933 | trace_btrfs_space_reservation(fs_info, "transaction", | |
934 | trans->transid, trans->bytes_reserved, 0); | |
935 | btrfs_block_rsv_release(fs_info, trans->block_rsv, | |
63f018be | 936 | trans->bytes_reserved, NULL); |
0e34693f NB |
937 | trans->bytes_reserved = 0; |
938 | } | |
939 | ||
89ce8a63 | 940 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
3a45bb20 | 941 | int throttle) |
79154b1b | 942 | { |
3a45bb20 | 943 | struct btrfs_fs_info *info = trans->fs_info; |
8929ecfa | 944 | struct btrfs_transaction *cur_trans = trans->transaction; |
4edc2ca3 | 945 | int err = 0; |
c3e69d58 | 946 | |
b50fff81 DS |
947 | if (refcount_read(&trans->use_count) > 1) { |
948 | refcount_dec(&trans->use_count); | |
2a1eb461 JB |
949 | trans->block_rsv = trans->orig_rsv; |
950 | return 0; | |
951 | } | |
952 | ||
dc60c525 | 953 | btrfs_trans_release_metadata(trans); |
4c13d758 | 954 | trans->block_rsv = NULL; |
c5567237 | 955 | |
119e80df | 956 | btrfs_create_pending_block_groups(trans); |
ea658bad | 957 | |
4fbcdf66 FM |
958 | btrfs_trans_release_chunk_metadata(trans); |
959 | ||
0860adfd | 960 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 961 | sb_end_intwrite(info->sb); |
6df7881a | 962 | |
8929ecfa | 963 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
964 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
965 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 966 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 967 | |
093258e6 | 968 | cond_wake_up(&cur_trans->writer_wait); |
724e2315 | 969 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
970 | |
971 | if (current->journal_info == trans) | |
972 | current->journal_info = NULL; | |
ab78c84d | 973 | |
24bbcf04 | 974 | if (throttle) |
2ff7e61e | 975 | btrfs_run_delayed_iputs(info); |
24bbcf04 | 976 | |
bf31f87f | 977 | if (TRANS_ABORTED(trans) || |
0b246afa | 978 | test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) { |
4e121c06 | 979 | wake_up_process(info->transaction_kthread); |
fbabd4a3 JB |
980 | if (TRANS_ABORTED(trans)) |
981 | err = trans->aborted; | |
982 | else | |
983 | err = -EROFS; | |
4e121c06 | 984 | } |
49b25e05 | 985 | |
4edc2ca3 DJ |
986 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
987 | return err; | |
79154b1b CM |
988 | } |
989 | ||
3a45bb20 | 990 | int btrfs_end_transaction(struct btrfs_trans_handle *trans) |
89ce8a63 | 991 | { |
3a45bb20 | 992 | return __btrfs_end_transaction(trans, 0); |
89ce8a63 CM |
993 | } |
994 | ||
3a45bb20 | 995 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) |
89ce8a63 | 996 | { |
3a45bb20 | 997 | return __btrfs_end_transaction(trans, 1); |
16cdcec7 MX |
998 | } |
999 | ||
d352ac68 CM |
1000 | /* |
1001 | * when btree blocks are allocated, they have some corresponding bits set for | |
1002 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 1003 | * those extents are sent to disk but does not wait on them |
d352ac68 | 1004 | */ |
2ff7e61e | 1005 | int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, |
8cef4e16 | 1006 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 1007 | { |
777e6bd7 | 1008 | int err = 0; |
7c4452b9 | 1009 | int werr = 0; |
0b246afa | 1010 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 1011 | struct extent_state *cached_state = NULL; |
777e6bd7 | 1012 | u64 start = 0; |
5f39d397 | 1013 | u64 end; |
7c4452b9 | 1014 | |
6300463b | 1015 | atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
1728366e | 1016 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 1017 | mark, &cached_state)) { |
663dfbb0 FM |
1018 | bool wait_writeback = false; |
1019 | ||
1020 | err = convert_extent_bit(dirty_pages, start, end, | |
1021 | EXTENT_NEED_WAIT, | |
210aa277 | 1022 | mark, &cached_state); |
663dfbb0 FM |
1023 | /* |
1024 | * convert_extent_bit can return -ENOMEM, which is most of the | |
1025 | * time a temporary error. So when it happens, ignore the error | |
1026 | * and wait for writeback of this range to finish - because we | |
1027 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
bf89d38f JM |
1028 | * to __btrfs_wait_marked_extents() would not know that |
1029 | * writeback for this range started and therefore wouldn't | |
1030 | * wait for it to finish - we don't want to commit a | |
1031 | * superblock that points to btree nodes/leafs for which | |
1032 | * writeback hasn't finished yet (and without errors). | |
663dfbb0 | 1033 | * We cleanup any entries left in the io tree when committing |
41e7acd3 | 1034 | * the transaction (through extent_io_tree_release()). |
663dfbb0 FM |
1035 | */ |
1036 | if (err == -ENOMEM) { | |
1037 | err = 0; | |
1038 | wait_writeback = true; | |
1039 | } | |
1040 | if (!err) | |
1041 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
1042 | if (err) |
1043 | werr = err; | |
663dfbb0 FM |
1044 | else if (wait_writeback) |
1045 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 1046 | free_extent_state(cached_state); |
663dfbb0 | 1047 | cached_state = NULL; |
1728366e JB |
1048 | cond_resched(); |
1049 | start = end + 1; | |
7c4452b9 | 1050 | } |
6300463b | 1051 | atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
690587d1 CM |
1052 | return werr; |
1053 | } | |
1054 | ||
1055 | /* | |
1056 | * when btree blocks are allocated, they have some corresponding bits set for | |
1057 | * them in one of two extent_io trees. This is used to make sure all of | |
1058 | * those extents are on disk for transaction or log commit. We wait | |
1059 | * on all the pages and clear them from the dirty pages state tree | |
1060 | */ | |
bf89d38f JM |
1061 | static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, |
1062 | struct extent_io_tree *dirty_pages) | |
690587d1 | 1063 | { |
690587d1 CM |
1064 | int err = 0; |
1065 | int werr = 0; | |
0b246afa | 1066 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 1067 | struct extent_state *cached_state = NULL; |
690587d1 CM |
1068 | u64 start = 0; |
1069 | u64 end; | |
777e6bd7 | 1070 | |
1728366e | 1071 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 1072 | EXTENT_NEED_WAIT, &cached_state)) { |
663dfbb0 FM |
1073 | /* |
1074 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
1075 | * When committing the transaction, we'll remove any entries | |
1076 | * left in the io tree. For a log commit, we don't remove them | |
1077 | * after committing the log because the tree can be accessed | |
1078 | * concurrently - we do it only at transaction commit time when | |
41e7acd3 | 1079 | * it's safe to do it (through extent_io_tree_release()). |
663dfbb0 FM |
1080 | */ |
1081 | err = clear_extent_bit(dirty_pages, start, end, | |
ae0f1625 | 1082 | EXTENT_NEED_WAIT, 0, 0, &cached_state); |
663dfbb0 FM |
1083 | if (err == -ENOMEM) |
1084 | err = 0; | |
1085 | if (!err) | |
1086 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
1087 | if (err) |
1088 | werr = err; | |
e38e2ed7 FM |
1089 | free_extent_state(cached_state); |
1090 | cached_state = NULL; | |
1728366e JB |
1091 | cond_resched(); |
1092 | start = end + 1; | |
777e6bd7 | 1093 | } |
7c4452b9 CM |
1094 | if (err) |
1095 | werr = err; | |
bf89d38f JM |
1096 | return werr; |
1097 | } | |
656f30db | 1098 | |
b9fae2eb | 1099 | static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, |
bf89d38f JM |
1100 | struct extent_io_tree *dirty_pages) |
1101 | { | |
1102 | bool errors = false; | |
1103 | int err; | |
656f30db | 1104 | |
bf89d38f JM |
1105 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); |
1106 | if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) | |
1107 | errors = true; | |
1108 | ||
1109 | if (errors && !err) | |
1110 | err = -EIO; | |
1111 | return err; | |
1112 | } | |
656f30db | 1113 | |
bf89d38f JM |
1114 | int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) |
1115 | { | |
1116 | struct btrfs_fs_info *fs_info = log_root->fs_info; | |
1117 | struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; | |
1118 | bool errors = false; | |
1119 | int err; | |
656f30db | 1120 | |
bf89d38f JM |
1121 | ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
1122 | ||
1123 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); | |
1124 | if ((mark & EXTENT_DIRTY) && | |
1125 | test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) | |
1126 | errors = true; | |
1127 | ||
1128 | if ((mark & EXTENT_NEW) && | |
1129 | test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) | |
1130 | errors = true; | |
1131 | ||
1132 | if (errors && !err) | |
1133 | err = -EIO; | |
1134 | return err; | |
79154b1b CM |
1135 | } |
1136 | ||
690587d1 | 1137 | /* |
c9b577c0 NB |
1138 | * When btree blocks are allocated the corresponding extents are marked dirty. |
1139 | * This function ensures such extents are persisted on disk for transaction or | |
1140 | * log commit. | |
1141 | * | |
1142 | * @trans: transaction whose dirty pages we'd like to write | |
690587d1 | 1143 | */ |
70458a58 | 1144 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) |
690587d1 CM |
1145 | { |
1146 | int ret; | |
1147 | int ret2; | |
c9b577c0 | 1148 | struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; |
70458a58 | 1149 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c6adc9cc | 1150 | struct blk_plug plug; |
690587d1 | 1151 | |
c6adc9cc | 1152 | blk_start_plug(&plug); |
c9b577c0 | 1153 | ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); |
c6adc9cc | 1154 | blk_finish_plug(&plug); |
bf89d38f | 1155 | ret2 = btrfs_wait_extents(fs_info, dirty_pages); |
bf0da8c1 | 1156 | |
41e7acd3 | 1157 | extent_io_tree_release(&trans->transaction->dirty_pages); |
c9b577c0 | 1158 | |
bf0da8c1 CM |
1159 | if (ret) |
1160 | return ret; | |
c9b577c0 | 1161 | else if (ret2) |
bf0da8c1 | 1162 | return ret2; |
c9b577c0 NB |
1163 | else |
1164 | return 0; | |
d0c803c4 CM |
1165 | } |
1166 | ||
d352ac68 CM |
1167 | /* |
1168 | * this is used to update the root pointer in the tree of tree roots. | |
1169 | * | |
1170 | * But, in the case of the extent allocation tree, updating the root | |
1171 | * pointer may allocate blocks which may change the root of the extent | |
1172 | * allocation tree. | |
1173 | * | |
1174 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1175 | * change while the root pointer was being updated in the metadata. | |
1176 | */ | |
0b86a832 CM |
1177 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1178 | struct btrfs_root *root) | |
79154b1b CM |
1179 | { |
1180 | int ret; | |
0b86a832 | 1181 | u64 old_root_bytenr; |
86b9f2ec | 1182 | u64 old_root_used; |
0b246afa JM |
1183 | struct btrfs_fs_info *fs_info = root->fs_info; |
1184 | struct btrfs_root *tree_root = fs_info->tree_root; | |
79154b1b | 1185 | |
86b9f2ec | 1186 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1187 | |
d397712b | 1188 | while (1) { |
0b86a832 | 1189 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1190 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1191 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1192 | break; |
87ef2bb4 | 1193 | |
5d4f98a2 | 1194 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1195 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1196 | &root->root_key, |
1197 | &root->root_item); | |
49b25e05 JM |
1198 | if (ret) |
1199 | return ret; | |
56bec294 | 1200 | |
86b9f2ec | 1201 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1202 | } |
276e680d | 1203 | |
0b86a832 CM |
1204 | return 0; |
1205 | } | |
1206 | ||
d352ac68 CM |
1207 | /* |
1208 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1209 | * |
1210 | * The error handling in this function may not be obvious. Any of the | |
1211 | * failures will cause the file system to go offline. We still need | |
1212 | * to clean up the delayed refs. | |
d352ac68 | 1213 | */ |
9386d8bc | 1214 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) |
0b86a832 | 1215 | { |
9386d8bc | 1216 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ea526d18 | 1217 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1218 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1219 | struct list_head *next; |
84234f3a | 1220 | struct extent_buffer *eb; |
56bec294 | 1221 | int ret; |
84234f3a YZ |
1222 | |
1223 | eb = btrfs_lock_root_node(fs_info->tree_root); | |
49b25e05 | 1224 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
9631e4cc | 1225 | 0, &eb, BTRFS_NESTING_COW); |
84234f3a YZ |
1226 | btrfs_tree_unlock(eb); |
1227 | free_extent_buffer(eb); | |
0b86a832 | 1228 | |
49b25e05 JM |
1229 | if (ret) |
1230 | return ret; | |
87ef2bb4 | 1231 | |
196c9d8d | 1232 | ret = btrfs_run_dev_stats(trans); |
c16ce190 JB |
1233 | if (ret) |
1234 | return ret; | |
2b584c68 | 1235 | ret = btrfs_run_dev_replace(trans); |
c16ce190 JB |
1236 | if (ret) |
1237 | return ret; | |
280f8bd2 | 1238 | ret = btrfs_run_qgroups(trans); |
c16ce190 JB |
1239 | if (ret) |
1240 | return ret; | |
546adb0d | 1241 | |
bbebb3e0 | 1242 | ret = btrfs_setup_space_cache(trans); |
dcdf7f6d JB |
1243 | if (ret) |
1244 | return ret; | |
1245 | ||
ea526d18 | 1246 | again: |
d397712b | 1247 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
2ff7e61e | 1248 | struct btrfs_root *root; |
0b86a832 CM |
1249 | next = fs_info->dirty_cowonly_roots.next; |
1250 | list_del_init(next); | |
1251 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1252 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1253 | |
9e351cc8 JB |
1254 | if (root != fs_info->extent_root) |
1255 | list_add_tail(&root->dirty_list, | |
1256 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1257 | ret = update_cowonly_root(trans, root); |
1258 | if (ret) | |
1259 | return ret; | |
79154b1b | 1260 | } |
276e680d | 1261 | |
488bc2a2 JB |
1262 | /* Now flush any delayed refs generated by updating all of the roots */ |
1263 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); | |
1264 | if (ret) | |
1265 | return ret; | |
1266 | ||
1bbc621e | 1267 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
5742d15f | 1268 | ret = btrfs_write_dirty_block_groups(trans); |
ea526d18 JB |
1269 | if (ret) |
1270 | return ret; | |
488bc2a2 JB |
1271 | |
1272 | /* | |
1273 | * We're writing the dirty block groups, which could generate | |
1274 | * delayed refs, which could generate more dirty block groups, | |
1275 | * so we want to keep this flushing in this loop to make sure | |
1276 | * everything gets run. | |
1277 | */ | |
c79a70b1 | 1278 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
ea526d18 JB |
1279 | if (ret) |
1280 | return ret; | |
1281 | } | |
1282 | ||
1283 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1284 | goto again; | |
1285 | ||
9e351cc8 JB |
1286 | list_add_tail(&fs_info->extent_root->dirty_list, |
1287 | &trans->transaction->switch_commits); | |
9f6cbcbb DS |
1288 | |
1289 | /* Update dev-replace pointer once everything is committed */ | |
1290 | fs_info->dev_replace.committed_cursor_left = | |
1291 | fs_info->dev_replace.cursor_left_last_write_of_item; | |
8dabb742 | 1292 | |
79154b1b CM |
1293 | return 0; |
1294 | } | |
1295 | ||
d352ac68 CM |
1296 | /* |
1297 | * dead roots are old snapshots that need to be deleted. This allocates | |
1298 | * a dirty root struct and adds it into the list of dead roots that need to | |
1299 | * be deleted | |
1300 | */ | |
cfad392b | 1301 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1302 | { |
0b246afa JM |
1303 | struct btrfs_fs_info *fs_info = root->fs_info; |
1304 | ||
1305 | spin_lock(&fs_info->trans_lock); | |
dc9492c1 JB |
1306 | if (list_empty(&root->root_list)) { |
1307 | btrfs_grab_root(root); | |
0b246afa | 1308 | list_add_tail(&root->root_list, &fs_info->dead_roots); |
dc9492c1 | 1309 | } |
0b246afa | 1310 | spin_unlock(&fs_info->trans_lock); |
5eda7b5e CM |
1311 | } |
1312 | ||
d352ac68 | 1313 | /* |
5d4f98a2 | 1314 | * update all the cowonly tree roots on disk |
d352ac68 | 1315 | */ |
7e4443d9 | 1316 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) |
0f7d52f4 | 1317 | { |
7e4443d9 | 1318 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0f7d52f4 | 1319 | struct btrfs_root *gang[8]; |
0f7d52f4 CM |
1320 | int i; |
1321 | int ret; | |
54aa1f4d | 1322 | |
a4abeea4 | 1323 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1324 | while (1) { |
5d4f98a2 YZ |
1325 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1326 | (void **)gang, 0, | |
0f7d52f4 CM |
1327 | ARRAY_SIZE(gang), |
1328 | BTRFS_ROOT_TRANS_TAG); | |
1329 | if (ret == 0) | |
1330 | break; | |
1331 | for (i = 0; i < ret; i++) { | |
5b4aacef | 1332 | struct btrfs_root *root = gang[i]; |
4f4317c1 JB |
1333 | int ret2; |
1334 | ||
5d4f98a2 YZ |
1335 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1336 | (unsigned long)root->root_key.objectid, | |
1337 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1338 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1339 | |
e02119d5 | 1340 | btrfs_free_log(trans, root); |
5d4f98a2 | 1341 | btrfs_update_reloc_root(trans, root); |
bcc63abb | 1342 | |
f1ebcc74 | 1343 | /* see comments in should_cow_block() */ |
27cdeb70 | 1344 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
c7548af6 | 1345 | smp_mb__after_atomic(); |
f1ebcc74 | 1346 | |
978d910d | 1347 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1348 | list_add_tail(&root->dirty_list, |
1349 | &trans->transaction->switch_commits); | |
978d910d YZ |
1350 | btrfs_set_root_node(&root->root_item, |
1351 | root->node); | |
1352 | } | |
5d4f98a2 | 1353 | |
4f4317c1 | 1354 | ret2 = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1355 | &root->root_key, |
1356 | &root->root_item); | |
4f4317c1 JB |
1357 | if (ret2) |
1358 | return ret2; | |
a4abeea4 | 1359 | spin_lock(&fs_info->fs_roots_radix_lock); |
733e03a0 | 1360 | btrfs_qgroup_free_meta_all_pertrans(root); |
0f7d52f4 CM |
1361 | } |
1362 | } | |
a4abeea4 | 1363 | spin_unlock(&fs_info->fs_roots_radix_lock); |
4f4317c1 | 1364 | return 0; |
0f7d52f4 CM |
1365 | } |
1366 | ||
d352ac68 | 1367 | /* |
de78b51a ES |
1368 | * defrag a given btree. |
1369 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1370 | */ |
de78b51a | 1371 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1372 | { |
1373 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1374 | struct btrfs_trans_handle *trans; |
8929ecfa | 1375 | int ret; |
e9d0b13b | 1376 | |
27cdeb70 | 1377 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1378 | return 0; |
8929ecfa | 1379 | |
6b80053d | 1380 | while (1) { |
8929ecfa YZ |
1381 | trans = btrfs_start_transaction(root, 0); |
1382 | if (IS_ERR(trans)) | |
1383 | return PTR_ERR(trans); | |
1384 | ||
de78b51a | 1385 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1386 | |
3a45bb20 | 1387 | btrfs_end_transaction(trans); |
2ff7e61e | 1388 | btrfs_btree_balance_dirty(info); |
e9d0b13b CM |
1389 | cond_resched(); |
1390 | ||
ab8d0fc4 | 1391 | if (btrfs_fs_closing(info) || ret != -EAGAIN) |
e9d0b13b | 1392 | break; |
210549eb | 1393 | |
ab8d0fc4 JM |
1394 | if (btrfs_defrag_cancelled(info)) { |
1395 | btrfs_debug(info, "defrag_root cancelled"); | |
210549eb DS |
1396 | ret = -EAGAIN; |
1397 | break; | |
1398 | } | |
e9d0b13b | 1399 | } |
27cdeb70 | 1400 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1401 | return ret; |
e9d0b13b CM |
1402 | } |
1403 | ||
6426c7ad QW |
1404 | /* |
1405 | * Do all special snapshot related qgroup dirty hack. | |
1406 | * | |
1407 | * Will do all needed qgroup inherit and dirty hack like switch commit | |
1408 | * roots inside one transaction and write all btree into disk, to make | |
1409 | * qgroup works. | |
1410 | */ | |
1411 | static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, | |
1412 | struct btrfs_root *src, | |
1413 | struct btrfs_root *parent, | |
1414 | struct btrfs_qgroup_inherit *inherit, | |
1415 | u64 dst_objectid) | |
1416 | { | |
1417 | struct btrfs_fs_info *fs_info = src->fs_info; | |
1418 | int ret; | |
1419 | ||
1420 | /* | |
1421 | * Save some performance in the case that qgroups are not | |
1422 | * enabled. If this check races with the ioctl, rescan will | |
1423 | * kick in anyway. | |
1424 | */ | |
9ea6e2b5 | 1425 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
6426c7ad | 1426 | return 0; |
6426c7ad | 1427 | |
4d31778a | 1428 | /* |
52042d8e | 1429 | * Ensure dirty @src will be committed. Or, after coming |
4d31778a QW |
1430 | * commit_fs_roots() and switch_commit_roots(), any dirty but not |
1431 | * recorded root will never be updated again, causing an outdated root | |
1432 | * item. | |
1433 | */ | |
1434 | record_root_in_trans(trans, src, 1); | |
1435 | ||
2a4d84c1 JB |
1436 | /* |
1437 | * btrfs_qgroup_inherit relies on a consistent view of the usage for the | |
1438 | * src root, so we must run the delayed refs here. | |
1439 | * | |
1440 | * However this isn't particularly fool proof, because there's no | |
1441 | * synchronization keeping us from changing the tree after this point | |
1442 | * before we do the qgroup_inherit, or even from making changes while | |
1443 | * we're doing the qgroup_inherit. But that's a problem for the future, | |
1444 | * for now flush the delayed refs to narrow the race window where the | |
1445 | * qgroup counters could end up wrong. | |
1446 | */ | |
1447 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); | |
1448 | if (ret) { | |
1449 | btrfs_abort_transaction(trans, ret); | |
1450 | goto out; | |
1451 | } | |
1452 | ||
6426c7ad QW |
1453 | /* |
1454 | * We are going to commit transaction, see btrfs_commit_transaction() | |
1455 | * comment for reason locking tree_log_mutex | |
1456 | */ | |
1457 | mutex_lock(&fs_info->tree_log_mutex); | |
1458 | ||
7e4443d9 | 1459 | ret = commit_fs_roots(trans); |
6426c7ad QW |
1460 | if (ret) |
1461 | goto out; | |
460fb20a | 1462 | ret = btrfs_qgroup_account_extents(trans); |
6426c7ad QW |
1463 | if (ret < 0) |
1464 | goto out; | |
1465 | ||
1466 | /* Now qgroup are all updated, we can inherit it to new qgroups */ | |
a9377422 | 1467 | ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, |
6426c7ad QW |
1468 | inherit); |
1469 | if (ret < 0) | |
1470 | goto out; | |
1471 | ||
1472 | /* | |
1473 | * Now we do a simplified commit transaction, which will: | |
1474 | * 1) commit all subvolume and extent tree | |
1475 | * To ensure all subvolume and extent tree have a valid | |
1476 | * commit_root to accounting later insert_dir_item() | |
1477 | * 2) write all btree blocks onto disk | |
1478 | * This is to make sure later btree modification will be cowed | |
1479 | * Or commit_root can be populated and cause wrong qgroup numbers | |
1480 | * In this simplified commit, we don't really care about other trees | |
1481 | * like chunk and root tree, as they won't affect qgroup. | |
1482 | * And we don't write super to avoid half committed status. | |
1483 | */ | |
9386d8bc | 1484 | ret = commit_cowonly_roots(trans); |
6426c7ad QW |
1485 | if (ret) |
1486 | goto out; | |
889bfa39 | 1487 | switch_commit_roots(trans); |
70458a58 | 1488 | ret = btrfs_write_and_wait_transaction(trans); |
6426c7ad | 1489 | if (ret) |
f7af3934 | 1490 | btrfs_handle_fs_error(fs_info, ret, |
6426c7ad QW |
1491 | "Error while writing out transaction for qgroup"); |
1492 | ||
1493 | out: | |
1494 | mutex_unlock(&fs_info->tree_log_mutex); | |
1495 | ||
1496 | /* | |
1497 | * Force parent root to be updated, as we recorded it before so its | |
1498 | * last_trans == cur_transid. | |
1499 | * Or it won't be committed again onto disk after later | |
1500 | * insert_dir_item() | |
1501 | */ | |
1502 | if (!ret) | |
1503 | record_root_in_trans(trans, parent, 1); | |
1504 | return ret; | |
1505 | } | |
1506 | ||
d352ac68 CM |
1507 | /* |
1508 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1509 | * transaction commit. This does the actual creation. |
1510 | * | |
1511 | * Note: | |
1512 | * If the error which may affect the commitment of the current transaction | |
1513 | * happens, we should return the error number. If the error which just affect | |
1514 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1515 | */ |
80b6794d | 1516 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1517 | struct btrfs_pending_snapshot *pending) |
1518 | { | |
08d50ca3 NB |
1519 | |
1520 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3063d29f | 1521 | struct btrfs_key key; |
80b6794d | 1522 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1523 | struct btrfs_root *tree_root = fs_info->tree_root; |
1524 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1525 | struct btrfs_root *parent_root; |
98c9942a | 1526 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1527 | struct inode *parent_inode; |
42874b3d MX |
1528 | struct btrfs_path *path; |
1529 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1530 | struct dentry *dentry; |
3063d29f | 1531 | struct extent_buffer *tmp; |
925baedd | 1532 | struct extent_buffer *old; |
95582b00 | 1533 | struct timespec64 cur_time; |
aec8030a | 1534 | int ret = 0; |
d68fc57b | 1535 | u64 to_reserve = 0; |
6bdb72de | 1536 | u64 index = 0; |
a22285a6 | 1537 | u64 objectid; |
b83cc969 | 1538 | u64 root_flags; |
3063d29f | 1539 | |
8546b570 DS |
1540 | ASSERT(pending->path); |
1541 | path = pending->path; | |
42874b3d | 1542 | |
b0c0ea63 DS |
1543 | ASSERT(pending->root_item); |
1544 | new_root_item = pending->root_item; | |
a22285a6 | 1545 | |
543068a2 | 1546 | pending->error = btrfs_get_free_objectid(tree_root, &objectid); |
aec8030a | 1547 | if (pending->error) |
6fa9700e | 1548 | goto no_free_objectid; |
3063d29f | 1549 | |
d6726335 QW |
1550 | /* |
1551 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1552 | * accounted by later btrfs_qgroup_inherit(). | |
1553 | */ | |
1554 | btrfs_set_skip_qgroup(trans, objectid); | |
1555 | ||
147d256e | 1556 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
d68fc57b YZ |
1557 | |
1558 | if (to_reserve > 0) { | |
aec8030a MX |
1559 | pending->error = btrfs_block_rsv_add(root, |
1560 | &pending->block_rsv, | |
1561 | to_reserve, | |
1562 | BTRFS_RESERVE_NO_FLUSH); | |
1563 | if (pending->error) | |
d6726335 | 1564 | goto clear_skip_qgroup; |
d68fc57b YZ |
1565 | } |
1566 | ||
3063d29f | 1567 | key.objectid = objectid; |
a22285a6 YZ |
1568 | key.offset = (u64)-1; |
1569 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1570 | |
6fa9700e | 1571 | rsv = trans->block_rsv; |
a22285a6 | 1572 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1573 | trans->bytes_reserved = trans->block_rsv->reserved; |
0b246afa | 1574 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa JB |
1575 | trans->transid, |
1576 | trans->bytes_reserved, 1); | |
a22285a6 | 1577 | dentry = pending->dentry; |
e9662f70 | 1578 | parent_inode = pending->dir; |
a22285a6 | 1579 | parent_root = BTRFS_I(parent_inode)->root; |
6426c7ad | 1580 | record_root_in_trans(trans, parent_root, 0); |
a22285a6 | 1581 | |
c2050a45 | 1582 | cur_time = current_time(parent_inode); |
04b285f3 | 1583 | |
3063d29f CM |
1584 | /* |
1585 | * insert the directory item | |
1586 | */ | |
877574e2 | 1587 | ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); |
49b25e05 | 1588 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1589 | |
1590 | /* check if there is a file/dir which has the same name. */ | |
1591 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
4a0cc7ca | 1592 | btrfs_ino(BTRFS_I(parent_inode)), |
42874b3d MX |
1593 | dentry->d_name.name, |
1594 | dentry->d_name.len, 0); | |
1595 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1596 | pending->error = -EEXIST; |
aec8030a | 1597 | goto dir_item_existed; |
42874b3d MX |
1598 | } else if (IS_ERR(dir_item)) { |
1599 | ret = PTR_ERR(dir_item); | |
66642832 | 1600 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1601 | goto fail; |
79787eaa | 1602 | } |
42874b3d | 1603 | btrfs_release_path(path); |
52c26179 | 1604 | |
e999376f CM |
1605 | /* |
1606 | * pull in the delayed directory update | |
1607 | * and the delayed inode item | |
1608 | * otherwise we corrupt the FS during | |
1609 | * snapshot | |
1610 | */ | |
e5c304e6 | 1611 | ret = btrfs_run_delayed_items(trans); |
8732d44f | 1612 | if (ret) { /* Transaction aborted */ |
66642832 | 1613 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1614 | goto fail; |
1615 | } | |
e999376f | 1616 | |
6426c7ad | 1617 | record_root_in_trans(trans, root, 0); |
6bdb72de SW |
1618 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1619 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1620 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1621 | |
b83cc969 LZ |
1622 | root_flags = btrfs_root_flags(new_root_item); |
1623 | if (pending->readonly) | |
1624 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1625 | else | |
1626 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1627 | btrfs_set_root_flags(new_root_item, root_flags); | |
1628 | ||
8ea05e3a AB |
1629 | btrfs_set_root_generation_v2(new_root_item, |
1630 | trans->transid); | |
807fc790 | 1631 | generate_random_guid(new_root_item->uuid); |
8ea05e3a AB |
1632 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, |
1633 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1634 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1635 | memset(new_root_item->received_uuid, 0, | |
1636 | sizeof(new_root_item->received_uuid)); | |
1637 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1638 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1639 | btrfs_set_root_stransid(new_root_item, 0); | |
1640 | btrfs_set_root_rtransid(new_root_item, 0); | |
1641 | } | |
3cae210f QW |
1642 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1643 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1644 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1645 | |
6bdb72de | 1646 | old = btrfs_lock_root_node(root); |
9631e4cc JB |
1647 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, |
1648 | BTRFS_NESTING_COW); | |
79787eaa JM |
1649 | if (ret) { |
1650 | btrfs_tree_unlock(old); | |
1651 | free_extent_buffer(old); | |
66642832 | 1652 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1653 | goto fail; |
79787eaa | 1654 | } |
49b25e05 | 1655 | |
49b25e05 | 1656 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1657 | /* clean up in any case */ |
6bdb72de SW |
1658 | btrfs_tree_unlock(old); |
1659 | free_extent_buffer(old); | |
8732d44f | 1660 | if (ret) { |
66642832 | 1661 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1662 | goto fail; |
1663 | } | |
f1ebcc74 | 1664 | /* see comments in should_cow_block() */ |
27cdeb70 | 1665 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1666 | smp_wmb(); |
1667 | ||
6bdb72de | 1668 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1669 | /* record when the snapshot was created in key.offset */ |
1670 | key.offset = trans->transid; | |
1671 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1672 | btrfs_tree_unlock(tmp); |
1673 | free_extent_buffer(tmp); | |
8732d44f | 1674 | if (ret) { |
66642832 | 1675 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1676 | goto fail; |
1677 | } | |
6bdb72de | 1678 | |
a22285a6 YZ |
1679 | /* |
1680 | * insert root back/forward references | |
1681 | */ | |
6025c19f | 1682 | ret = btrfs_add_root_ref(trans, objectid, |
0660b5af | 1683 | parent_root->root_key.objectid, |
4a0cc7ca | 1684 | btrfs_ino(BTRFS_I(parent_inode)), index, |
a22285a6 | 1685 | dentry->d_name.name, dentry->d_name.len); |
8732d44f | 1686 | if (ret) { |
66642832 | 1687 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1688 | goto fail; |
1689 | } | |
0660b5af | 1690 | |
a22285a6 | 1691 | key.offset = (u64)-1; |
2dfb1e43 | 1692 | pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); |
79787eaa JM |
1693 | if (IS_ERR(pending->snap)) { |
1694 | ret = PTR_ERR(pending->snap); | |
2d892ccd | 1695 | pending->snap = NULL; |
66642832 | 1696 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1697 | goto fail; |
79787eaa | 1698 | } |
d68fc57b | 1699 | |
49b25e05 | 1700 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f | 1701 | if (ret) { |
66642832 | 1702 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1703 | goto fail; |
1704 | } | |
361048f5 | 1705 | |
6426c7ad QW |
1706 | /* |
1707 | * Do special qgroup accounting for snapshot, as we do some qgroup | |
1708 | * snapshot hack to do fast snapshot. | |
1709 | * To co-operate with that hack, we do hack again. | |
1710 | * Or snapshot will be greatly slowed down by a subtree qgroup rescan | |
1711 | */ | |
1712 | ret = qgroup_account_snapshot(trans, root, parent_root, | |
1713 | pending->inherit, objectid); | |
1714 | if (ret < 0) | |
1715 | goto fail; | |
1716 | ||
684572df LF |
1717 | ret = btrfs_insert_dir_item(trans, dentry->d_name.name, |
1718 | dentry->d_name.len, BTRFS_I(parent_inode), | |
1719 | &key, BTRFS_FT_DIR, index); | |
42874b3d | 1720 | /* We have check then name at the beginning, so it is impossible. */ |
9c52057c | 1721 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f | 1722 | if (ret) { |
66642832 | 1723 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1724 | goto fail; |
1725 | } | |
42874b3d | 1726 | |
6ef06d27 | 1727 | btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + |
42874b3d | 1728 | dentry->d_name.len * 2); |
04b285f3 | 1729 | parent_inode->i_mtime = parent_inode->i_ctime = |
c2050a45 | 1730 | current_time(parent_inode); |
729f7961 | 1731 | ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); |
dd5f9615 | 1732 | if (ret) { |
66642832 | 1733 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1734 | goto fail; |
1735 | } | |
807fc790 AS |
1736 | ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, |
1737 | BTRFS_UUID_KEY_SUBVOL, | |
cdb345a8 | 1738 | objectid); |
dd5f9615 | 1739 | if (ret) { |
66642832 | 1740 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1741 | goto fail; |
1742 | } | |
1743 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
cdb345a8 | 1744 | ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, |
dd5f9615 SB |
1745 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, |
1746 | objectid); | |
1747 | if (ret && ret != -EEXIST) { | |
66642832 | 1748 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1749 | goto fail; |
1750 | } | |
1751 | } | |
d6726335 | 1752 | |
3063d29f | 1753 | fail: |
aec8030a MX |
1754 | pending->error = ret; |
1755 | dir_item_existed: | |
98c9942a | 1756 | trans->block_rsv = rsv; |
2382c5cc | 1757 | trans->bytes_reserved = 0; |
d6726335 QW |
1758 | clear_skip_qgroup: |
1759 | btrfs_clear_skip_qgroup(trans); | |
6fa9700e MX |
1760 | no_free_objectid: |
1761 | kfree(new_root_item); | |
b0c0ea63 | 1762 | pending->root_item = NULL; |
42874b3d | 1763 | btrfs_free_path(path); |
8546b570 DS |
1764 | pending->path = NULL; |
1765 | ||
49b25e05 | 1766 | return ret; |
3063d29f CM |
1767 | } |
1768 | ||
d352ac68 CM |
1769 | /* |
1770 | * create all the snapshots we've scheduled for creation | |
1771 | */ | |
08d50ca3 | 1772 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) |
3de4586c | 1773 | { |
aec8030a | 1774 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1775 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1776 | int ret = 0; |
3de4586c | 1777 | |
aec8030a MX |
1778 | list_for_each_entry_safe(pending, next, head, list) { |
1779 | list_del(&pending->list); | |
08d50ca3 | 1780 | ret = create_pending_snapshot(trans, pending); |
aec8030a MX |
1781 | if (ret) |
1782 | break; | |
1783 | } | |
1784 | return ret; | |
3de4586c CM |
1785 | } |
1786 | ||
2ff7e61e | 1787 | static void update_super_roots(struct btrfs_fs_info *fs_info) |
5d4f98a2 YZ |
1788 | { |
1789 | struct btrfs_root_item *root_item; | |
1790 | struct btrfs_super_block *super; | |
1791 | ||
0b246afa | 1792 | super = fs_info->super_copy; |
5d4f98a2 | 1793 | |
0b246afa | 1794 | root_item = &fs_info->chunk_root->root_item; |
093e037c DS |
1795 | super->chunk_root = root_item->bytenr; |
1796 | super->chunk_root_generation = root_item->generation; | |
1797 | super->chunk_root_level = root_item->level; | |
5d4f98a2 | 1798 | |
0b246afa | 1799 | root_item = &fs_info->tree_root->root_item; |
093e037c DS |
1800 | super->root = root_item->bytenr; |
1801 | super->generation = root_item->generation; | |
1802 | super->root_level = root_item->level; | |
0b246afa | 1803 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) |
093e037c | 1804 | super->cache_generation = root_item->generation; |
94846229 BB |
1805 | else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) |
1806 | super->cache_generation = 0; | |
0b246afa | 1807 | if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) |
093e037c | 1808 | super->uuid_tree_generation = root_item->generation; |
5d4f98a2 YZ |
1809 | } |
1810 | ||
f36f3042 CM |
1811 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1812 | { | |
4a9d8bde | 1813 | struct btrfs_transaction *trans; |
f36f3042 | 1814 | int ret = 0; |
4a9d8bde | 1815 | |
a4abeea4 | 1816 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1817 | trans = info->running_transaction; |
1818 | if (trans) | |
1819 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1820 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1821 | return ret; |
1822 | } | |
1823 | ||
8929ecfa YZ |
1824 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1825 | { | |
4a9d8bde | 1826 | struct btrfs_transaction *trans; |
8929ecfa | 1827 | int ret = 0; |
4a9d8bde | 1828 | |
a4abeea4 | 1829 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1830 | trans = info->running_transaction; |
1831 | if (trans) | |
1832 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1833 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1834 | return ret; |
1835 | } | |
1836 | ||
bb9c12c9 SW |
1837 | /* |
1838 | * wait for the current transaction commit to start and block subsequent | |
1839 | * transaction joins | |
1840 | */ | |
2ff7e61e | 1841 | static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info, |
bb9c12c9 SW |
1842 | struct btrfs_transaction *trans) |
1843 | { | |
2ff7e61e | 1844 | wait_event(fs_info->transaction_blocked_wait, |
bf31f87f DS |
1845 | trans->state >= TRANS_STATE_COMMIT_START || |
1846 | TRANS_ABORTED(trans)); | |
bb9c12c9 SW |
1847 | } |
1848 | ||
1849 | /* | |
1850 | * wait for the current transaction to start and then become unblocked. | |
1851 | * caller holds ref. | |
1852 | */ | |
2ff7e61e JM |
1853 | static void wait_current_trans_commit_start_and_unblock( |
1854 | struct btrfs_fs_info *fs_info, | |
1855 | struct btrfs_transaction *trans) | |
bb9c12c9 | 1856 | { |
2ff7e61e | 1857 | wait_event(fs_info->transaction_wait, |
bf31f87f DS |
1858 | trans->state >= TRANS_STATE_UNBLOCKED || |
1859 | TRANS_ABORTED(trans)); | |
bb9c12c9 SW |
1860 | } |
1861 | ||
1862 | /* | |
1863 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1864 | * returns, any subsequent transaction will not be allowed to join. | |
1865 | */ | |
1866 | struct btrfs_async_commit { | |
1867 | struct btrfs_trans_handle *newtrans; | |
7892b5af | 1868 | struct work_struct work; |
bb9c12c9 SW |
1869 | }; |
1870 | ||
1871 | static void do_async_commit(struct work_struct *work) | |
1872 | { | |
1873 | struct btrfs_async_commit *ac = | |
7892b5af | 1874 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1875 | |
6fc4e354 SW |
1876 | /* |
1877 | * We've got freeze protection passed with the transaction. | |
1878 | * Tell lockdep about it. | |
1879 | */ | |
b1a06a4b | 1880 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
3a45bb20 | 1881 | __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1882 | |
e209db7a SW |
1883 | current->journal_info = ac->newtrans; |
1884 | ||
3a45bb20 | 1885 | btrfs_commit_transaction(ac->newtrans); |
bb9c12c9 SW |
1886 | kfree(ac); |
1887 | } | |
1888 | ||
1889 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
bb9c12c9 SW |
1890 | int wait_for_unblock) |
1891 | { | |
3a45bb20 | 1892 | struct btrfs_fs_info *fs_info = trans->fs_info; |
bb9c12c9 SW |
1893 | struct btrfs_async_commit *ac; |
1894 | struct btrfs_transaction *cur_trans; | |
1895 | ||
1896 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1897 | if (!ac) |
1898 | return -ENOMEM; | |
bb9c12c9 | 1899 | |
7892b5af | 1900 | INIT_WORK(&ac->work, do_async_commit); |
3a45bb20 | 1901 | ac->newtrans = btrfs_join_transaction(trans->root); |
3612b495 TI |
1902 | if (IS_ERR(ac->newtrans)) { |
1903 | int err = PTR_ERR(ac->newtrans); | |
1904 | kfree(ac); | |
1905 | return err; | |
1906 | } | |
bb9c12c9 SW |
1907 | |
1908 | /* take transaction reference */ | |
bb9c12c9 | 1909 | cur_trans = trans->transaction; |
9b64f57d | 1910 | refcount_inc(&cur_trans->use_count); |
bb9c12c9 | 1911 | |
3a45bb20 | 1912 | btrfs_end_transaction(trans); |
6fc4e354 SW |
1913 | |
1914 | /* | |
1915 | * Tell lockdep we've released the freeze rwsem, since the | |
1916 | * async commit thread will be the one to unlock it. | |
1917 | */ | |
b1a06a4b | 1918 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
0b246afa | 1919 | __sb_writers_release(fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1920 | |
7892b5af | 1921 | schedule_work(&ac->work); |
bb9c12c9 SW |
1922 | |
1923 | /* wait for transaction to start and unblock */ | |
bb9c12c9 | 1924 | if (wait_for_unblock) |
2ff7e61e | 1925 | wait_current_trans_commit_start_and_unblock(fs_info, cur_trans); |
bb9c12c9 | 1926 | else |
2ff7e61e | 1927 | wait_current_trans_commit_start(fs_info, cur_trans); |
bb9c12c9 | 1928 | |
38e88054 SW |
1929 | if (current->journal_info == trans) |
1930 | current->journal_info = NULL; | |
1931 | ||
724e2315 | 1932 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1933 | return 0; |
1934 | } | |
1935 | ||
49b25e05 | 1936 | |
97cb39bb | 1937 | static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) |
49b25e05 | 1938 | { |
97cb39bb | 1939 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 JM |
1940 | struct btrfs_transaction *cur_trans = trans->transaction; |
1941 | ||
b50fff81 | 1942 | WARN_ON(refcount_read(&trans->use_count) > 1); |
49b25e05 | 1943 | |
66642832 | 1944 | btrfs_abort_transaction(trans, err); |
7b8b92af | 1945 | |
0b246afa | 1946 | spin_lock(&fs_info->trans_lock); |
66b6135b | 1947 | |
25d8c284 MX |
1948 | /* |
1949 | * If the transaction is removed from the list, it means this | |
1950 | * transaction has been committed successfully, so it is impossible | |
1951 | * to call the cleanup function. | |
1952 | */ | |
1953 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1954 | |
49b25e05 | 1955 | list_del_init(&cur_trans->list); |
0b246afa | 1956 | if (cur_trans == fs_info->running_transaction) { |
4a9d8bde | 1957 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 1958 | spin_unlock(&fs_info->trans_lock); |
f094ac32 LB |
1959 | wait_event(cur_trans->writer_wait, |
1960 | atomic_read(&cur_trans->num_writers) == 1); | |
1961 | ||
0b246afa | 1962 | spin_lock(&fs_info->trans_lock); |
d7096fc3 | 1963 | } |
0b246afa | 1964 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 1965 | |
2ff7e61e | 1966 | btrfs_cleanup_one_transaction(trans->transaction, fs_info); |
49b25e05 | 1967 | |
0b246afa JM |
1968 | spin_lock(&fs_info->trans_lock); |
1969 | if (cur_trans == fs_info->running_transaction) | |
1970 | fs_info->running_transaction = NULL; | |
1971 | spin_unlock(&fs_info->trans_lock); | |
4a9d8bde | 1972 | |
e0228285 | 1973 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 1974 | sb_end_intwrite(fs_info->sb); |
724e2315 JB |
1975 | btrfs_put_transaction(cur_trans); |
1976 | btrfs_put_transaction(cur_trans); | |
49b25e05 | 1977 | |
97cb39bb | 1978 | trace_btrfs_transaction_commit(trans->root); |
49b25e05 | 1979 | |
49b25e05 JM |
1980 | if (current->journal_info == trans) |
1981 | current->journal_info = NULL; | |
0b246afa | 1982 | btrfs_scrub_cancel(fs_info); |
49b25e05 JM |
1983 | |
1984 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1985 | } | |
1986 | ||
c7cc64a9 DS |
1987 | /* |
1988 | * Release reserved delayed ref space of all pending block groups of the | |
1989 | * transaction and remove them from the list | |
1990 | */ | |
1991 | static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) | |
1992 | { | |
1993 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 1994 | struct btrfs_block_group *block_group, *tmp; |
c7cc64a9 DS |
1995 | |
1996 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { | |
1997 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1998 | list_del_init(&block_group->bg_list); | |
1999 | } | |
2000 | } | |
2001 | ||
88090ad3 | 2002 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
82436617 | 2003 | { |
ce8ea7cc JB |
2004 | /* |
2005 | * We use writeback_inodes_sb here because if we used | |
2006 | * btrfs_start_delalloc_roots we would deadlock with fs freeze. | |
2007 | * Currently are holding the fs freeze lock, if we do an async flush | |
2008 | * we'll do btrfs_join_transaction() and deadlock because we need to | |
2009 | * wait for the fs freeze lock. Using the direct flushing we benefit | |
2010 | * from already being in a transaction and our join_transaction doesn't | |
2011 | * have to re-take the fs freeze lock. | |
2012 | */ | |
88090ad3 | 2013 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
ce8ea7cc | 2014 | writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); |
82436617 MX |
2015 | return 0; |
2016 | } | |
2017 | ||
88090ad3 | 2018 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) |
82436617 | 2019 | { |
88090ad3 | 2020 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
6374e57a | 2021 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); |
82436617 MX |
2022 | } |
2023 | ||
3a45bb20 | 2024 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans) |
79154b1b | 2025 | { |
3a45bb20 | 2026 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 | 2027 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 2028 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 2029 | int ret; |
79154b1b | 2030 | |
35b814f3 NB |
2031 | ASSERT(refcount_read(&trans->use_count) == 1); |
2032 | ||
d62b23c9 JB |
2033 | /* |
2034 | * Some places just start a transaction to commit it. We need to make | |
2035 | * sure that if this commit fails that the abort code actually marks the | |
2036 | * transaction as failed, so set trans->dirty to make the abort code do | |
2037 | * the right thing. | |
2038 | */ | |
2039 | trans->dirty = true; | |
2040 | ||
8d25a086 | 2041 | /* Stop the commit early if ->aborted is set */ |
bf31f87f | 2042 | if (TRANS_ABORTED(cur_trans)) { |
25287e0a | 2043 | ret = cur_trans->aborted; |
3a45bb20 | 2044 | btrfs_end_transaction(trans); |
e4a2bcac | 2045 | return ret; |
25287e0a | 2046 | } |
49b25e05 | 2047 | |
f45c752b JB |
2048 | btrfs_trans_release_metadata(trans); |
2049 | trans->block_rsv = NULL; | |
2050 | ||
56bec294 | 2051 | /* |
e19eb11f JB |
2052 | * We only want one transaction commit doing the flushing so we do not |
2053 | * waste a bunch of time on lock contention on the extent root node. | |
56bec294 | 2054 | */ |
e19eb11f JB |
2055 | if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, |
2056 | &cur_trans->delayed_refs.flags)) { | |
2057 | /* | |
2058 | * Make a pass through all the delayed refs we have so far. | |
2059 | * Any running threads may add more while we are here. | |
2060 | */ | |
2061 | ret = btrfs_run_delayed_refs(trans, 0); | |
2062 | if (ret) { | |
2063 | btrfs_end_transaction(trans); | |
2064 | return ret; | |
2065 | } | |
2066 | } | |
56bec294 | 2067 | |
119e80df | 2068 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2069 | |
3204d33c | 2070 | if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { |
1bbc621e CM |
2071 | int run_it = 0; |
2072 | ||
2073 | /* this mutex is also taken before trying to set | |
2074 | * block groups readonly. We need to make sure | |
2075 | * that nobody has set a block group readonly | |
2076 | * after a extents from that block group have been | |
2077 | * allocated for cache files. btrfs_set_block_group_ro | |
2078 | * will wait for the transaction to commit if it | |
3204d33c | 2079 | * finds BTRFS_TRANS_DIRTY_BG_RUN set. |
1bbc621e | 2080 | * |
3204d33c JB |
2081 | * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure |
2082 | * only one process starts all the block group IO. It wouldn't | |
1bbc621e CM |
2083 | * hurt to have more than one go through, but there's no |
2084 | * real advantage to it either. | |
2085 | */ | |
0b246afa | 2086 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c JB |
2087 | if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, |
2088 | &cur_trans->flags)) | |
1bbc621e | 2089 | run_it = 1; |
0b246afa | 2090 | mutex_unlock(&fs_info->ro_block_group_mutex); |
1bbc621e | 2091 | |
f9cacae3 | 2092 | if (run_it) { |
21217054 | 2093 | ret = btrfs_start_dirty_block_groups(trans); |
f9cacae3 NB |
2094 | if (ret) { |
2095 | btrfs_end_transaction(trans); | |
2096 | return ret; | |
2097 | } | |
2098 | } | |
1bbc621e CM |
2099 | } |
2100 | ||
0b246afa | 2101 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2102 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { |
0b246afa | 2103 | spin_unlock(&fs_info->trans_lock); |
9b64f57d | 2104 | refcount_inc(&cur_trans->use_count); |
3a45bb20 | 2105 | ret = btrfs_end_transaction(trans); |
ccd467d6 | 2106 | |
2ff7e61e | 2107 | wait_for_commit(cur_trans); |
15ee9bc7 | 2108 | |
bf31f87f | 2109 | if (TRANS_ABORTED(cur_trans)) |
b4924a0f LB |
2110 | ret = cur_trans->aborted; |
2111 | ||
724e2315 | 2112 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 2113 | |
49b25e05 | 2114 | return ret; |
79154b1b | 2115 | } |
4313b399 | 2116 | |
4a9d8bde | 2117 | cur_trans->state = TRANS_STATE_COMMIT_START; |
0b246afa | 2118 | wake_up(&fs_info->transaction_blocked_wait); |
bb9c12c9 | 2119 | |
0b246afa | 2120 | if (cur_trans->list.prev != &fs_info->trans_list) { |
ccd467d6 CM |
2121 | prev_trans = list_entry(cur_trans->list.prev, |
2122 | struct btrfs_transaction, list); | |
4a9d8bde | 2123 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
9b64f57d | 2124 | refcount_inc(&prev_trans->use_count); |
0b246afa | 2125 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2126 | |
2ff7e61e | 2127 | wait_for_commit(prev_trans); |
bf31f87f | 2128 | ret = READ_ONCE(prev_trans->aborted); |
ccd467d6 | 2129 | |
724e2315 | 2130 | btrfs_put_transaction(prev_trans); |
1f9b8c8f FM |
2131 | if (ret) |
2132 | goto cleanup_transaction; | |
a4abeea4 | 2133 | } else { |
0b246afa | 2134 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2135 | } |
a4abeea4 | 2136 | } else { |
0b246afa | 2137 | spin_unlock(&fs_info->trans_lock); |
cb2d3dad FM |
2138 | /* |
2139 | * The previous transaction was aborted and was already removed | |
2140 | * from the list of transactions at fs_info->trans_list. So we | |
2141 | * abort to prevent writing a new superblock that reflects a | |
2142 | * corrupt state (pointing to trees with unwritten nodes/leafs). | |
2143 | */ | |
2144 | if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) { | |
2145 | ret = -EROFS; | |
2146 | goto cleanup_transaction; | |
2147 | } | |
ccd467d6 | 2148 | } |
15ee9bc7 | 2149 | |
0860adfd MX |
2150 | extwriter_counter_dec(cur_trans, trans->type); |
2151 | ||
88090ad3 | 2152 | ret = btrfs_start_delalloc_flush(fs_info); |
82436617 MX |
2153 | if (ret) |
2154 | goto cleanup_transaction; | |
2155 | ||
e5c304e6 | 2156 | ret = btrfs_run_delayed_items(trans); |
581227d0 MX |
2157 | if (ret) |
2158 | goto cleanup_transaction; | |
15ee9bc7 | 2159 | |
581227d0 MX |
2160 | wait_event(cur_trans->writer_wait, |
2161 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 2162 | |
581227d0 | 2163 | /* some pending stuffs might be added after the previous flush. */ |
e5c304e6 | 2164 | ret = btrfs_run_delayed_items(trans); |
ca469637 MX |
2165 | if (ret) |
2166 | goto cleanup_transaction; | |
2167 | ||
88090ad3 | 2168 | btrfs_wait_delalloc_flush(fs_info); |
cb7ab021 | 2169 | |
48778179 FM |
2170 | /* |
2171 | * Wait for all ordered extents started by a fast fsync that joined this | |
2172 | * transaction. Otherwise if this transaction commits before the ordered | |
2173 | * extents complete we lose logged data after a power failure. | |
2174 | */ | |
2175 | wait_event(cur_trans->pending_wait, | |
2176 | atomic_read(&cur_trans->pending_ordered) == 0); | |
2177 | ||
2ff7e61e | 2178 | btrfs_scrub_pause(fs_info); |
ed0ca140 JB |
2179 | /* |
2180 | * Ok now we need to make sure to block out any other joins while we | |
2181 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 2182 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 | 2183 | */ |
0b246afa | 2184 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2185 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 2186 | spin_unlock(&fs_info->trans_lock); |
ed0ca140 JB |
2187 | wait_event(cur_trans->writer_wait, |
2188 | atomic_read(&cur_trans->num_writers) == 1); | |
2189 | ||
bf31f87f | 2190 | if (TRANS_ABORTED(cur_trans)) { |
2cba30f1 | 2191 | ret = cur_trans->aborted; |
6cf7f77e | 2192 | goto scrub_continue; |
2cba30f1 | 2193 | } |
7585717f CM |
2194 | /* |
2195 | * the reloc mutex makes sure that we stop | |
2196 | * the balancing code from coming in and moving | |
2197 | * extents around in the middle of the commit | |
2198 | */ | |
0b246afa | 2199 | mutex_lock(&fs_info->reloc_mutex); |
7585717f | 2200 | |
42874b3d MX |
2201 | /* |
2202 | * We needn't worry about the delayed items because we will | |
2203 | * deal with them in create_pending_snapshot(), which is the | |
2204 | * core function of the snapshot creation. | |
2205 | */ | |
08d50ca3 | 2206 | ret = create_pending_snapshots(trans); |
56e9f6ea DS |
2207 | if (ret) |
2208 | goto unlock_reloc; | |
3063d29f | 2209 | |
42874b3d MX |
2210 | /* |
2211 | * We insert the dir indexes of the snapshots and update the inode | |
2212 | * of the snapshots' parents after the snapshot creation, so there | |
2213 | * are some delayed items which are not dealt with. Now deal with | |
2214 | * them. | |
2215 | * | |
2216 | * We needn't worry that this operation will corrupt the snapshots, | |
2217 | * because all the tree which are snapshoted will be forced to COW | |
2218 | * the nodes and leaves. | |
2219 | */ | |
e5c304e6 | 2220 | ret = btrfs_run_delayed_items(trans); |
56e9f6ea DS |
2221 | if (ret) |
2222 | goto unlock_reloc; | |
16cdcec7 | 2223 | |
c79a70b1 | 2224 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
56e9f6ea DS |
2225 | if (ret) |
2226 | goto unlock_reloc; | |
56bec294 | 2227 | |
e999376f CM |
2228 | /* |
2229 | * make sure none of the code above managed to slip in a | |
2230 | * delayed item | |
2231 | */ | |
ccdf9b30 | 2232 | btrfs_assert_delayed_root_empty(fs_info); |
e999376f | 2233 | |
2c90e5d6 | 2234 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 2235 | |
e02119d5 CM |
2236 | /* btrfs_commit_tree_roots is responsible for getting the |
2237 | * various roots consistent with each other. Every pointer | |
2238 | * in the tree of tree roots has to point to the most up to date | |
2239 | * root for every subvolume and other tree. So, we have to keep | |
2240 | * the tree logging code from jumping in and changing any | |
2241 | * of the trees. | |
2242 | * | |
2243 | * At this point in the commit, there can't be any tree-log | |
2244 | * writers, but a little lower down we drop the trans mutex | |
2245 | * and let new people in. By holding the tree_log_mutex | |
2246 | * from now until after the super is written, we avoid races | |
2247 | * with the tree-log code. | |
2248 | */ | |
0b246afa | 2249 | mutex_lock(&fs_info->tree_log_mutex); |
e02119d5 | 2250 | |
7e4443d9 | 2251 | ret = commit_fs_roots(trans); |
56e9f6ea DS |
2252 | if (ret) |
2253 | goto unlock_tree_log; | |
54aa1f4d | 2254 | |
3818aea2 | 2255 | /* |
7e1876ac DS |
2256 | * Since the transaction is done, we can apply the pending changes |
2257 | * before the next transaction. | |
3818aea2 | 2258 | */ |
0b246afa | 2259 | btrfs_apply_pending_changes(fs_info); |
3818aea2 | 2260 | |
5d4f98a2 | 2261 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2262 | * safe to free the root of tree log roots |
2263 | */ | |
0b246afa | 2264 | btrfs_free_log_root_tree(trans, fs_info); |
e02119d5 | 2265 | |
0ed4792a QW |
2266 | /* |
2267 | * Since fs roots are all committed, we can get a quite accurate | |
2268 | * new_roots. So let's do quota accounting. | |
2269 | */ | |
460fb20a | 2270 | ret = btrfs_qgroup_account_extents(trans); |
56e9f6ea DS |
2271 | if (ret < 0) |
2272 | goto unlock_tree_log; | |
0ed4792a | 2273 | |
9386d8bc | 2274 | ret = commit_cowonly_roots(trans); |
56e9f6ea DS |
2275 | if (ret) |
2276 | goto unlock_tree_log; | |
54aa1f4d | 2277 | |
2cba30f1 MX |
2278 | /* |
2279 | * The tasks which save the space cache and inode cache may also | |
2280 | * update ->aborted, check it. | |
2281 | */ | |
bf31f87f | 2282 | if (TRANS_ABORTED(cur_trans)) { |
2cba30f1 | 2283 | ret = cur_trans->aborted; |
56e9f6ea | 2284 | goto unlock_tree_log; |
2cba30f1 MX |
2285 | } |
2286 | ||
0b246afa | 2287 | cur_trans = fs_info->running_transaction; |
5d4f98a2 | 2288 | |
0b246afa JM |
2289 | btrfs_set_root_node(&fs_info->tree_root->root_item, |
2290 | fs_info->tree_root->node); | |
2291 | list_add_tail(&fs_info->tree_root->dirty_list, | |
9e351cc8 | 2292 | &cur_trans->switch_commits); |
5d4f98a2 | 2293 | |
0b246afa JM |
2294 | btrfs_set_root_node(&fs_info->chunk_root->root_item, |
2295 | fs_info->chunk_root->node); | |
2296 | list_add_tail(&fs_info->chunk_root->dirty_list, | |
9e351cc8 JB |
2297 | &cur_trans->switch_commits); |
2298 | ||
889bfa39 | 2299 | switch_commit_roots(trans); |
5d4f98a2 | 2300 | |
ce93ec54 | 2301 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2302 | ASSERT(list_empty(&cur_trans->io_bgs)); |
2ff7e61e | 2303 | update_super_roots(fs_info); |
e02119d5 | 2304 | |
0b246afa JM |
2305 | btrfs_set_super_log_root(fs_info->super_copy, 0); |
2306 | btrfs_set_super_log_root_level(fs_info->super_copy, 0); | |
2307 | memcpy(fs_info->super_for_commit, fs_info->super_copy, | |
2308 | sizeof(*fs_info->super_copy)); | |
ccd467d6 | 2309 | |
bbbf7243 | 2310 | btrfs_commit_device_sizes(cur_trans); |
935e5cc9 | 2311 | |
0b246afa JM |
2312 | clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
2313 | clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); | |
656f30db | 2314 | |
4fbcdf66 FM |
2315 | btrfs_trans_release_chunk_metadata(trans); |
2316 | ||
0b246afa | 2317 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2318 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
0b246afa JM |
2319 | fs_info->running_transaction = NULL; |
2320 | spin_unlock(&fs_info->trans_lock); | |
2321 | mutex_unlock(&fs_info->reloc_mutex); | |
b7ec40d7 | 2322 | |
0b246afa | 2323 | wake_up(&fs_info->transaction_wait); |
e6dcd2dc | 2324 | |
70458a58 | 2325 | ret = btrfs_write_and_wait_transaction(trans); |
49b25e05 | 2326 | if (ret) { |
0b246afa JM |
2327 | btrfs_handle_fs_error(fs_info, ret, |
2328 | "Error while writing out transaction"); | |
56e9f6ea DS |
2329 | /* |
2330 | * reloc_mutex has been unlocked, tree_log_mutex is still held | |
2331 | * but we can't jump to unlock_tree_log causing double unlock | |
2332 | */ | |
0b246afa | 2333 | mutex_unlock(&fs_info->tree_log_mutex); |
6cf7f77e | 2334 | goto scrub_continue; |
49b25e05 JM |
2335 | } |
2336 | ||
eece6a9c | 2337 | ret = write_all_supers(fs_info, 0); |
e02119d5 CM |
2338 | /* |
2339 | * the super is written, we can safely allow the tree-loggers | |
2340 | * to go about their business | |
2341 | */ | |
0b246afa | 2342 | mutex_unlock(&fs_info->tree_log_mutex); |
c1f32b7c AJ |
2343 | if (ret) |
2344 | goto scrub_continue; | |
e02119d5 | 2345 | |
5ead2dd0 | 2346 | btrfs_finish_extent_commit(trans); |
4313b399 | 2347 | |
3204d33c | 2348 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) |
0b246afa | 2349 | btrfs_clear_space_info_full(fs_info); |
13212b54 | 2350 | |
0b246afa | 2351 | fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
2352 | /* |
2353 | * We needn't acquire the lock here because there is no other task | |
2354 | * which can change it. | |
2355 | */ | |
2356 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2357 | wake_up(&cur_trans->commit_wait); |
3de4586c | 2358 | |
0b246afa | 2359 | spin_lock(&fs_info->trans_lock); |
13c5a93e | 2360 | list_del_init(&cur_trans->list); |
0b246afa | 2361 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 2362 | |
724e2315 JB |
2363 | btrfs_put_transaction(cur_trans); |
2364 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2365 | |
0860adfd | 2366 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2367 | sb_end_intwrite(fs_info->sb); |
b2b5ef5c | 2368 | |
3a45bb20 | 2369 | trace_btrfs_transaction_commit(trans->root); |
1abe9b8a | 2370 | |
2ff7e61e | 2371 | btrfs_scrub_continue(fs_info); |
a2de733c | 2372 | |
9ed74f2d JB |
2373 | if (current->journal_info == trans) |
2374 | current->journal_info = NULL; | |
2375 | ||
2c90e5d6 | 2376 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2377 | |
79154b1b | 2378 | return ret; |
49b25e05 | 2379 | |
56e9f6ea DS |
2380 | unlock_tree_log: |
2381 | mutex_unlock(&fs_info->tree_log_mutex); | |
2382 | unlock_reloc: | |
2383 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2384 | scrub_continue: |
2ff7e61e | 2385 | btrfs_scrub_continue(fs_info); |
49b25e05 | 2386 | cleanup_transaction: |
dc60c525 | 2387 | btrfs_trans_release_metadata(trans); |
c7cc64a9 | 2388 | btrfs_cleanup_pending_block_groups(trans); |
4fbcdf66 | 2389 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2390 | trans->block_rsv = NULL; |
0b246afa | 2391 | btrfs_warn(fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2392 | if (current->journal_info == trans) |
2393 | current->journal_info = NULL; | |
97cb39bb | 2394 | cleanup_transaction(trans, ret); |
49b25e05 JM |
2395 | |
2396 | return ret; | |
79154b1b CM |
2397 | } |
2398 | ||
d352ac68 | 2399 | /* |
9d1a2a3a DS |
2400 | * return < 0 if error |
2401 | * 0 if there are no more dead_roots at the time of call | |
2402 | * 1 there are more to be processed, call me again | |
2403 | * | |
2404 | * The return value indicates there are certainly more snapshots to delete, but | |
2405 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2406 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2407 | * few seconds later. | |
d352ac68 | 2408 | */ |
9d1a2a3a | 2409 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 2410 | { |
9d1a2a3a | 2411 | int ret; |
5d4f98a2 YZ |
2412 | struct btrfs_fs_info *fs_info = root->fs_info; |
2413 | ||
a4abeea4 | 2414 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2415 | if (list_empty(&fs_info->dead_roots)) { |
2416 | spin_unlock(&fs_info->trans_lock); | |
2417 | return 0; | |
2418 | } | |
2419 | root = list_first_entry(&fs_info->dead_roots, | |
2420 | struct btrfs_root, root_list); | |
cfad392b | 2421 | list_del_init(&root->root_list); |
a4abeea4 | 2422 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2423 | |
4fd786e6 | 2424 | btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); |
76dda93c | 2425 | |
9d1a2a3a | 2426 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2427 | |
9d1a2a3a DS |
2428 | if (btrfs_header_backref_rev(root->node) < |
2429 | BTRFS_MIXED_BACKREF_REV) | |
0078a9f9 | 2430 | ret = btrfs_drop_snapshot(root, 0, 0); |
9d1a2a3a | 2431 | else |
0078a9f9 | 2432 | ret = btrfs_drop_snapshot(root, 1, 0); |
32471dc2 | 2433 | |
dc9492c1 | 2434 | btrfs_put_root(root); |
6596a928 | 2435 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2436 | } |
572d9ab7 DS |
2437 | |
2438 | void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) | |
2439 | { | |
2440 | unsigned long prev; | |
2441 | unsigned long bit; | |
2442 | ||
6c9fe14f | 2443 | prev = xchg(&fs_info->pending_changes, 0); |
572d9ab7 DS |
2444 | if (!prev) |
2445 | return; | |
2446 | ||
d51033d0 DS |
2447 | bit = 1 << BTRFS_PENDING_COMMIT; |
2448 | if (prev & bit) | |
2449 | btrfs_debug(fs_info, "pending commit done"); | |
2450 | prev &= ~bit; | |
2451 | ||
572d9ab7 DS |
2452 | if (prev) |
2453 | btrfs_warn(fs_info, | |
2454 | "unknown pending changes left 0x%lx, ignoring", prev); | |
2455 | } |