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> |
ab3c5c18 | 9 | #include <linux/sched/mm.h> |
d3c2fdcf | 10 | #include <linux/writeback.h> |
5f39d397 | 11 | #include <linux/pagemap.h> |
5f2cc086 | 12 | #include <linux/blkdev.h> |
8ea05e3a | 13 | #include <linux/uuid.h> |
e55958c8 | 14 | #include <linux/timekeeping.h> |
602cbe91 | 15 | #include "misc.h" |
79154b1b CM |
16 | #include "ctree.h" |
17 | #include "disk-io.h" | |
18 | #include "transaction.h" | |
925baedd | 19 | #include "locking.h" |
e02119d5 | 20 | #include "tree-log.h" |
733f4fbb | 21 | #include "volumes.h" |
8dabb742 | 22 | #include "dev-replace.h" |
fcebe456 | 23 | #include "qgroup.h" |
aac0023c | 24 | #include "block-group.h" |
9c343784 | 25 | #include "space-info.h" |
d3575156 | 26 | #include "zoned.h" |
c7f13d42 | 27 | #include "fs.h" |
07e81dc9 | 28 | #include "accessors.h" |
a0231804 | 29 | #include "extent-tree.h" |
45c40c8f | 30 | #include "root-tree.h" |
59b818e0 | 31 | #include "defrag.h" |
f2b39277 | 32 | #include "dir-item.h" |
c7a03b52 | 33 | #include "uuid-tree.h" |
7572dec8 | 34 | #include "ioctl.h" |
67707479 | 35 | #include "relocation.h" |
2fc6822c | 36 | #include "scrub.h" |
79154b1b | 37 | |
956504a3 JB |
38 | static struct kmem_cache *btrfs_trans_handle_cachep; |
39 | ||
61c047b5 QW |
40 | /* |
41 | * Transaction states and transitions | |
42 | * | |
43 | * No running transaction (fs tree blocks are not modified) | |
44 | * | | |
45 | * | To next stage: | |
46 | * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). | |
47 | * V | |
48 | * Transaction N [[TRANS_STATE_RUNNING]] | |
49 | * | | |
50 | * | New trans handles can be attached to transaction N by calling all | |
51 | * | start_transaction() variants. | |
52 | * | | |
53 | * | To next stage: | |
54 | * | Call btrfs_commit_transaction() on any trans handle attached to | |
55 | * | transaction N | |
56 | * V | |
77d20c68 JB |
57 | * Transaction N [[TRANS_STATE_COMMIT_PREP]] |
58 | * | | |
59 | * | If there are simultaneous calls to btrfs_commit_transaction() one will win | |
60 | * | the race and the rest will wait for the winner to commit the transaction. | |
61 | * | | |
62 | * | The winner will wait for previous running transaction to completely finish | |
63 | * | if there is one. | |
61c047b5 | 64 | * | |
77d20c68 | 65 | * Transaction N [[TRANS_STATE_COMMIT_START]] |
61c047b5 | 66 | * | |
77d20c68 | 67 | * | Then one of the following happens: |
61c047b5 QW |
68 | * | - Wait for all other trans handle holders to release. |
69 | * | The btrfs_commit_transaction() caller will do the commit work. | |
70 | * | - Wait for current transaction to be committed by others. | |
71 | * | Other btrfs_commit_transaction() caller will do the commit work. | |
72 | * | | |
73 | * | At this stage, only btrfs_join_transaction*() variants can attach | |
74 | * | to this running transaction. | |
75 | * | All other variants will wait for current one to finish and attach to | |
76 | * | transaction N+1. | |
77 | * | | |
78 | * | To next stage: | |
79 | * | Caller is chosen to commit transaction N, and all other trans handle | |
80 | * | haven been released. | |
81 | * V | |
82 | * Transaction N [[TRANS_STATE_COMMIT_DOING]] | |
83 | * | | |
84 | * | The heavy lifting transaction work is started. | |
85 | * | From running delayed refs (modifying extent tree) to creating pending | |
86 | * | snapshots, running qgroups. | |
87 | * | In short, modify supporting trees to reflect modifications of subvolume | |
88 | * | trees. | |
89 | * | | |
90 | * | At this stage, all start_transaction() calls will wait for this | |
91 | * | transaction to finish and attach to transaction N+1. | |
92 | * | | |
93 | * | To next stage: | |
94 | * | Until all supporting trees are updated. | |
95 | * V | |
96 | * Transaction N [[TRANS_STATE_UNBLOCKED]] | |
97 | * | Transaction N+1 | |
98 | * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] | |
99 | * | need to write them back to disk and update | | |
100 | * | super blocks. | | |
101 | * | | | |
102 | * | At this stage, new transaction is allowed to | | |
103 | * | start. | | |
104 | * | All new start_transaction() calls will be | | |
105 | * | attached to transid N+1. | | |
106 | * | | | |
107 | * | To next stage: | | |
108 | * | Until all tree blocks are super blocks are | | |
109 | * | written to block devices | | |
110 | * V | | |
111 | * Transaction N [[TRANS_STATE_COMPLETED]] V | |
112 | * All tree blocks and super blocks are written. Transaction N+1 | |
113 | * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] | |
114 | * data structures will be cleaned up. | Life goes on | |
115 | */ | |
e8c9f186 | 116 | static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { |
4a9d8bde | 117 | [TRANS_STATE_RUNNING] = 0U, |
77d20c68 | 118 | [TRANS_STATE_COMMIT_PREP] = 0U, |
bcf3a3e7 NB |
119 | [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), |
120 | [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | | |
4a9d8bde | 121 | __TRANS_ATTACH | |
a6d155d2 FM |
122 | __TRANS_JOIN | |
123 | __TRANS_JOIN_NOSTART), | |
bcf3a3e7 | 124 | [TRANS_STATE_UNBLOCKED] = (__TRANS_START | |
4a9d8bde MX |
125 | __TRANS_ATTACH | |
126 | __TRANS_JOIN | | |
a6d155d2 FM |
127 | __TRANS_JOIN_NOLOCK | |
128 | __TRANS_JOIN_NOSTART), | |
d0c2f4fa FM |
129 | [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START | |
130 | __TRANS_ATTACH | | |
131 | __TRANS_JOIN | | |
132 | __TRANS_JOIN_NOLOCK | | |
133 | __TRANS_JOIN_NOSTART), | |
bcf3a3e7 | 134 | [TRANS_STATE_COMPLETED] = (__TRANS_START | |
4a9d8bde MX |
135 | __TRANS_ATTACH | |
136 | __TRANS_JOIN | | |
a6d155d2 FM |
137 | __TRANS_JOIN_NOLOCK | |
138 | __TRANS_JOIN_NOSTART), | |
4a9d8bde MX |
139 | }; |
140 | ||
724e2315 | 141 | void btrfs_put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 142 | { |
9b64f57d ER |
143 | WARN_ON(refcount_read(&transaction->use_count) == 0); |
144 | if (refcount_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 145 | BUG_ON(!list_empty(&transaction->list)); |
5c9d028b LB |
146 | WARN_ON(!RB_EMPTY_ROOT( |
147 | &transaction->delayed_refs.href_root.rb_root)); | |
81f7eb00 JM |
148 | WARN_ON(!RB_EMPTY_ROOT( |
149 | &transaction->delayed_refs.dirty_extent_root)); | |
1262133b | 150 | if (transaction->delayed_refs.pending_csums) |
ab8d0fc4 JM |
151 | btrfs_err(transaction->fs_info, |
152 | "pending csums is %llu", | |
153 | transaction->delayed_refs.pending_csums); | |
7785a663 FM |
154 | /* |
155 | * If any block groups are found in ->deleted_bgs then it's | |
156 | * because the transaction was aborted and a commit did not | |
157 | * happen (things failed before writing the new superblock | |
158 | * and calling btrfs_finish_extent_commit()), so we can not | |
159 | * discard the physical locations of the block groups. | |
160 | */ | |
161 | while (!list_empty(&transaction->deleted_bgs)) { | |
32da5386 | 162 | struct btrfs_block_group *cache; |
7785a663 FM |
163 | |
164 | cache = list_first_entry(&transaction->deleted_bgs, | |
32da5386 | 165 | struct btrfs_block_group, |
7785a663 FM |
166 | bg_list); |
167 | list_del_init(&cache->bg_list); | |
6b7304af | 168 | btrfs_unfreeze_block_group(cache); |
7785a663 FM |
169 | btrfs_put_block_group(cache); |
170 | } | |
bbbf7243 | 171 | WARN_ON(!list_empty(&transaction->dev_update_list)); |
4b5faeac | 172 | kfree(transaction); |
78fae27e | 173 | } |
79154b1b CM |
174 | } |
175 | ||
889bfa39 | 176 | static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) |
817d52f8 | 177 | { |
889bfa39 | 178 | struct btrfs_transaction *cur_trans = trans->transaction; |
16916a88 | 179 | struct btrfs_fs_info *fs_info = trans->fs_info; |
9e351cc8 JB |
180 | struct btrfs_root *root, *tmp; |
181 | ||
dfba78dc FM |
182 | /* |
183 | * At this point no one can be using this transaction to modify any tree | |
184 | * and no one can start another transaction to modify any tree either. | |
185 | */ | |
186 | ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING); | |
187 | ||
9e351cc8 | 188 | down_write(&fs_info->commit_root_sem); |
d96b3424 FM |
189 | |
190 | if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) | |
191 | fs_info->last_reloc_trans = trans->transid; | |
192 | ||
889bfa39 | 193 | list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, |
9e351cc8 JB |
194 | dirty_list) { |
195 | list_del_init(&root->dirty_list); | |
196 | free_extent_buffer(root->commit_root); | |
197 | root->commit_root = btrfs_root_node(root); | |
41e7acd3 | 198 | extent_io_tree_release(&root->dirty_log_pages); |
370a11b8 | 199 | btrfs_qgroup_clean_swapped_blocks(root); |
9e351cc8 | 200 | } |
2b9dbef2 JB |
201 | |
202 | /* We can free old roots now. */ | |
889bfa39 JB |
203 | spin_lock(&cur_trans->dropped_roots_lock); |
204 | while (!list_empty(&cur_trans->dropped_roots)) { | |
205 | root = list_first_entry(&cur_trans->dropped_roots, | |
2b9dbef2 JB |
206 | struct btrfs_root, root_list); |
207 | list_del_init(&root->root_list); | |
889bfa39 JB |
208 | spin_unlock(&cur_trans->dropped_roots_lock); |
209 | btrfs_free_log(trans, root); | |
2b9dbef2 | 210 | btrfs_drop_and_free_fs_root(fs_info, root); |
889bfa39 | 211 | spin_lock(&cur_trans->dropped_roots_lock); |
2b9dbef2 | 212 | } |
889bfa39 | 213 | spin_unlock(&cur_trans->dropped_roots_lock); |
27d56e62 | 214 | |
9e351cc8 | 215 | up_write(&fs_info->commit_root_sem); |
817d52f8 JB |
216 | } |
217 | ||
0860adfd MX |
218 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
219 | unsigned int type) | |
220 | { | |
221 | if (type & TRANS_EXTWRITERS) | |
222 | atomic_inc(&trans->num_extwriters); | |
223 | } | |
224 | ||
225 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
226 | unsigned int type) | |
227 | { | |
228 | if (type & TRANS_EXTWRITERS) | |
229 | atomic_dec(&trans->num_extwriters); | |
230 | } | |
231 | ||
232 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
233 | unsigned int type) | |
234 | { | |
235 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
236 | } | |
237 | ||
238 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
239 | { | |
240 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
241 | } |
242 | ||
fb6dea26 | 243 | /* |
79bd3712 FM |
244 | * To be called after doing the chunk btree updates right after allocating a new |
245 | * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a | |
246 | * chunk after all chunk btree updates and after finishing the second phase of | |
247 | * chunk allocation (btrfs_create_pending_block_groups()) in case some block | |
248 | * group had its chunk item insertion delayed to the second phase. | |
fb6dea26 JB |
249 | */ |
250 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
251 | { | |
252 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
253 | ||
254 | if (!trans->chunk_bytes_reserved) | |
255 | return; | |
256 | ||
fb6dea26 | 257 | btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, |
63f018be | 258 | trans->chunk_bytes_reserved, NULL); |
fb6dea26 JB |
259 | trans->chunk_bytes_reserved = 0; |
260 | } | |
261 | ||
d352ac68 CM |
262 | /* |
263 | * either allocate a new transaction or hop into the existing one | |
264 | */ | |
2ff7e61e JM |
265 | static noinline int join_transaction(struct btrfs_fs_info *fs_info, |
266 | unsigned int type) | |
79154b1b CM |
267 | { |
268 | struct btrfs_transaction *cur_trans; | |
a4abeea4 | 269 | |
19ae4e81 | 270 | spin_lock(&fs_info->trans_lock); |
d43317dc | 271 | loop: |
49b25e05 | 272 | /* The file system has been taken offline. No new transactions. */ |
84961539 | 273 | if (BTRFS_FS_ERROR(fs_info)) { |
19ae4e81 | 274 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
275 | return -EROFS; |
276 | } | |
277 | ||
19ae4e81 | 278 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 279 | if (cur_trans) { |
bf31f87f | 280 | if (TRANS_ABORTED(cur_trans)) { |
19ae4e81 | 281 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 282 | return cur_trans->aborted; |
871383be | 283 | } |
4a9d8bde | 284 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
285 | spin_unlock(&fs_info->trans_lock); |
286 | return -EBUSY; | |
287 | } | |
9b64f57d | 288 | refcount_inc(&cur_trans->use_count); |
13c5a93e | 289 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 290 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 291 | spin_unlock(&fs_info->trans_lock); |
e1489b4f | 292 | btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); |
5a9ba670 | 293 | btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); |
a4abeea4 | 294 | return 0; |
79154b1b | 295 | } |
19ae4e81 | 296 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 297 | |
354aa0fb | 298 | /* |
4490e803 FM |
299 | * If we are ATTACH or TRANS_JOIN_NOSTART, we just want to catch the |
300 | * current transaction, and commit it. If there is no transaction, just | |
301 | * return ENOENT. | |
354aa0fb | 302 | */ |
4490e803 | 303 | if (type == TRANS_ATTACH || type == TRANS_JOIN_NOSTART) |
354aa0fb MX |
304 | return -ENOENT; |
305 | ||
4a9d8bde MX |
306 | /* |
307 | * JOIN_NOLOCK only happens during the transaction commit, so | |
308 | * it is impossible that ->running_transaction is NULL | |
309 | */ | |
310 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
311 | ||
4b5faeac | 312 | cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); |
a4abeea4 JB |
313 | if (!cur_trans) |
314 | return -ENOMEM; | |
d43317dc | 315 | |
e1489b4f | 316 | btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); |
5a9ba670 | 317 | btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); |
e1489b4f | 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 | */ |
5a9ba670 | 325 | btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); |
e1489b4f | 326 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); |
4b5faeac | 327 | kfree(cur_trans); |
d43317dc | 328 | goto loop; |
84961539 | 329 | } else if (BTRFS_FS_ERROR(fs_info)) { |
e4b50e14 | 330 | spin_unlock(&fs_info->trans_lock); |
5a9ba670 | 331 | btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); |
e1489b4f | 332 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); |
4b5faeac | 333 | kfree(cur_trans); |
7b8b92af | 334 | return -EROFS; |
79154b1b | 335 | } |
d43317dc | 336 | |
ab8d0fc4 | 337 | cur_trans->fs_info = fs_info; |
48778179 FM |
338 | atomic_set(&cur_trans->pending_ordered, 0); |
339 | init_waitqueue_head(&cur_trans->pending_wait); | |
a4abeea4 | 340 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 341 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
342 | init_waitqueue_head(&cur_trans->writer_wait); |
343 | init_waitqueue_head(&cur_trans->commit_wait); | |
4a9d8bde | 344 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
345 | /* |
346 | * One for this trans handle, one so it will live on until we | |
347 | * commit the transaction. | |
348 | */ | |
9b64f57d | 349 | refcount_set(&cur_trans->use_count, 2); |
3204d33c | 350 | cur_trans->flags = 0; |
afd48513 | 351 | cur_trans->start_time = ktime_get_seconds(); |
a4abeea4 | 352 | |
a099d0fd AM |
353 | memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); |
354 | ||
5c9d028b | 355 | cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; |
3368d001 | 356 | cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; |
d7df2c79 | 357 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
20b297d6 JS |
358 | |
359 | /* | |
360 | * although the tree mod log is per file system and not per transaction, | |
361 | * the log must never go across transaction boundaries. | |
362 | */ | |
363 | smp_mb(); | |
31b1a2bd | 364 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
5d163e0e | 365 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); |
31b1a2bd | 366 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
5d163e0e | 367 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); |
fc36ed7e | 368 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 369 | |
a4abeea4 JB |
370 | spin_lock_init(&cur_trans->delayed_refs.lock); |
371 | ||
372 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
bbbf7243 | 373 | INIT_LIST_HEAD(&cur_trans->dev_update_list); |
9e351cc8 | 374 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
ce93ec54 | 375 | INIT_LIST_HEAD(&cur_trans->dirty_bgs); |
1bbc621e | 376 | INIT_LIST_HEAD(&cur_trans->io_bgs); |
2b9dbef2 | 377 | INIT_LIST_HEAD(&cur_trans->dropped_roots); |
1bbc621e | 378 | mutex_init(&cur_trans->cache_write_mutex); |
ce93ec54 | 379 | spin_lock_init(&cur_trans->dirty_bgs_lock); |
e33e17ee | 380 | INIT_LIST_HEAD(&cur_trans->deleted_bgs); |
2b9dbef2 | 381 | spin_lock_init(&cur_trans->dropped_roots_lock); |
19ae4e81 | 382 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
c258d6e3 | 383 | extent_io_tree_init(fs_info, &cur_trans->dirty_pages, |
35da5a7e | 384 | IO_TREE_TRANS_DIRTY_PAGES); |
fe119a6e | 385 | extent_io_tree_init(fs_info, &cur_trans->pinned_extents, |
35da5a7e | 386 | IO_TREE_FS_PINNED_EXTENTS); |
4a4f8fe2 | 387 | btrfs_set_fs_generation(fs_info, fs_info->generation + 1); |
19ae4e81 JS |
388 | cur_trans->transid = fs_info->generation; |
389 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 390 | cur_trans->aborted = 0; |
19ae4e81 | 391 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 392 | |
79154b1b CM |
393 | return 0; |
394 | } | |
395 | ||
d352ac68 | 396 | /* |
92a7cc42 QW |
397 | * This does all the record keeping required to make sure that a shareable root |
398 | * is properly recorded in a given transaction. This is required to make sure | |
399 | * the old root from before we joined the transaction is deleted when the | |
400 | * transaction commits. | |
d352ac68 | 401 | */ |
7585717f | 402 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
6426c7ad QW |
403 | struct btrfs_root *root, |
404 | int force) | |
6702ed49 | 405 | { |
0b246afa | 406 | struct btrfs_fs_info *fs_info = root->fs_info; |
03a7e111 | 407 | int ret = 0; |
0b246afa | 408 | |
92a7cc42 | 409 | if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
6426c7ad | 410 | root->last_trans < trans->transid) || force) { |
4d31778a | 411 | WARN_ON(!force && root->commit_root != root->node); |
5d4f98a2 | 412 | |
7585717f | 413 | /* |
27cdeb70 | 414 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
415 | * we have the reloc mutex held now, so there |
416 | * is only one writer in this function | |
417 | */ | |
27cdeb70 | 418 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 419 | |
27cdeb70 | 420 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
421 | * they find our root->last_trans update |
422 | */ | |
423 | smp_wmb(); | |
424 | ||
fc7cbcd4 | 425 | spin_lock(&fs_info->fs_roots_radix_lock); |
6426c7ad | 426 | if (root->last_trans == trans->transid && !force) { |
fc7cbcd4 | 427 | spin_unlock(&fs_info->fs_roots_radix_lock); |
a4abeea4 JB |
428 | return 0; |
429 | } | |
fc7cbcd4 DS |
430 | radix_tree_tag_set(&fs_info->fs_roots_radix, |
431 | (unsigned long)root->root_key.objectid, | |
432 | BTRFS_ROOT_TRANS_TAG); | |
433 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
7585717f CM |
434 | root->last_trans = trans->transid; |
435 | ||
436 | /* this is pretty tricky. We don't want to | |
437 | * take the relocation lock in btrfs_record_root_in_trans | |
438 | * unless we're really doing the first setup for this root in | |
439 | * this transaction. | |
440 | * | |
441 | * Normally we'd use root->last_trans as a flag to decide | |
442 | * if we want to take the expensive mutex. | |
443 | * | |
444 | * But, we have to set root->last_trans before we | |
445 | * init the relocation root, otherwise, we trip over warnings | |
446 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 447 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
448 | * fixing up the reloc trees and everyone must wait. |
449 | * | |
450 | * When this is zero, they can trust root->last_trans and fly | |
451 | * through btrfs_record_root_in_trans without having to take the | |
452 | * lock. smp_wmb() makes sure that all the writes above are | |
453 | * done before we pop in the zero below | |
454 | */ | |
03a7e111 | 455 | ret = btrfs_init_reloc_root(trans, root); |
c7548af6 | 456 | smp_mb__before_atomic(); |
27cdeb70 | 457 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 | 458 | } |
03a7e111 | 459 | return ret; |
5d4f98a2 | 460 | } |
bcc63abb | 461 | |
7585717f | 462 | |
2b9dbef2 JB |
463 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
464 | struct btrfs_root *root) | |
465 | { | |
0b246afa | 466 | struct btrfs_fs_info *fs_info = root->fs_info; |
2b9dbef2 JB |
467 | struct btrfs_transaction *cur_trans = trans->transaction; |
468 | ||
469 | /* Add ourselves to the transaction dropped list */ | |
470 | spin_lock(&cur_trans->dropped_roots_lock); | |
471 | list_add_tail(&root->root_list, &cur_trans->dropped_roots); | |
472 | spin_unlock(&cur_trans->dropped_roots_lock); | |
473 | ||
474 | /* Make sure we don't try to update the root at commit time */ | |
fc7cbcd4 DS |
475 | spin_lock(&fs_info->fs_roots_radix_lock); |
476 | radix_tree_tag_clear(&fs_info->fs_roots_radix, | |
477 | (unsigned long)root->root_key.objectid, | |
478 | BTRFS_ROOT_TRANS_TAG); | |
479 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
2b9dbef2 JB |
480 | } |
481 | ||
7585717f CM |
482 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
483 | struct btrfs_root *root) | |
484 | { | |
0b246afa | 485 | struct btrfs_fs_info *fs_info = root->fs_info; |
1409e6cc | 486 | int ret; |
0b246afa | 487 | |
92a7cc42 | 488 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
7585717f CM |
489 | return 0; |
490 | ||
491 | /* | |
27cdeb70 | 492 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
493 | * and barriers |
494 | */ | |
495 | smp_rmb(); | |
496 | if (root->last_trans == trans->transid && | |
27cdeb70 | 497 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
498 | return 0; |
499 | ||
0b246afa | 500 | mutex_lock(&fs_info->reloc_mutex); |
1409e6cc | 501 | ret = record_root_in_trans(trans, root, 0); |
0b246afa | 502 | mutex_unlock(&fs_info->reloc_mutex); |
7585717f | 503 | |
1409e6cc | 504 | return ret; |
7585717f CM |
505 | } |
506 | ||
4a9d8bde MX |
507 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
508 | { | |
3296bf56 | 509 | return (trans->state >= TRANS_STATE_COMMIT_START && |
501407aa | 510 | trans->state < TRANS_STATE_UNBLOCKED && |
bf31f87f | 511 | !TRANS_ABORTED(trans)); |
4a9d8bde MX |
512 | } |
513 | ||
d352ac68 CM |
514 | /* wait for commit against the current transaction to become unblocked |
515 | * when this is done, it is safe to start a new transaction, but the current | |
516 | * transaction might not be fully on disk. | |
517 | */ | |
2ff7e61e | 518 | static void wait_current_trans(struct btrfs_fs_info *fs_info) |
79154b1b | 519 | { |
f9295749 | 520 | struct btrfs_transaction *cur_trans; |
79154b1b | 521 | |
0b246afa JM |
522 | spin_lock(&fs_info->trans_lock); |
523 | cur_trans = fs_info->running_transaction; | |
4a9d8bde | 524 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
9b64f57d | 525 | refcount_inc(&cur_trans->use_count); |
0b246afa | 526 | spin_unlock(&fs_info->trans_lock); |
72d63ed6 | 527 | |
3e738c53 | 528 | btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); |
0b246afa | 529 | wait_event(fs_info->transaction_wait, |
501407aa | 530 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
bf31f87f | 531 | TRANS_ABORTED(cur_trans)); |
724e2315 | 532 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 533 | } else { |
0b246afa | 534 | spin_unlock(&fs_info->trans_lock); |
f9295749 | 535 | } |
37d1aeee CM |
536 | } |
537 | ||
2ff7e61e | 538 | static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) |
a22285a6 | 539 | { |
0b246afa | 540 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
a4abeea4 JB |
541 | return 0; |
542 | ||
92e2f7e3 | 543 | if (type == TRANS_START) |
a22285a6 | 544 | return 1; |
a4abeea4 | 545 | |
a22285a6 YZ |
546 | return 0; |
547 | } | |
548 | ||
20dd2cbf MX |
549 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
550 | { | |
0b246afa JM |
551 | struct btrfs_fs_info *fs_info = root->fs_info; |
552 | ||
553 | if (!fs_info->reloc_ctl || | |
92a7cc42 | 554 | !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || |
20dd2cbf MX |
555 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
556 | root->reloc_root) | |
557 | return false; | |
558 | ||
559 | return true; | |
560 | } | |
561 | ||
28270e25 FM |
562 | static int btrfs_reserve_trans_metadata(struct btrfs_fs_info *fs_info, |
563 | enum btrfs_reserve_flush_enum flush, | |
564 | u64 num_bytes, | |
565 | u64 *delayed_refs_bytes) | |
566 | { | |
567 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; | |
568 | struct btrfs_space_info *si = fs_info->trans_block_rsv.space_info; | |
569 | u64 extra_delayed_refs_bytes = 0; | |
570 | u64 bytes; | |
571 | int ret; | |
572 | ||
573 | /* | |
574 | * If there's a gap between the size of the delayed refs reserve and | |
575 | * its reserved space, than some tasks have added delayed refs or bumped | |
576 | * its size otherwise (due to block group creation or removal, or block | |
577 | * group item update). Also try to allocate that gap in order to prevent | |
578 | * using (and possibly abusing) the global reserve when committing the | |
579 | * transaction. | |
580 | */ | |
581 | if (flush == BTRFS_RESERVE_FLUSH_ALL && | |
582 | !btrfs_block_rsv_full(delayed_refs_rsv)) { | |
583 | spin_lock(&delayed_refs_rsv->lock); | |
584 | if (delayed_refs_rsv->size > delayed_refs_rsv->reserved) | |
585 | extra_delayed_refs_bytes = delayed_refs_rsv->size - | |
586 | delayed_refs_rsv->reserved; | |
587 | spin_unlock(&delayed_refs_rsv->lock); | |
588 | } | |
589 | ||
590 | bytes = num_bytes + *delayed_refs_bytes + extra_delayed_refs_bytes; | |
591 | ||
592 | /* | |
593 | * We want to reserve all the bytes we may need all at once, so we only | |
594 | * do 1 enospc flushing cycle per transaction start. | |
595 | */ | |
596 | ret = btrfs_reserve_metadata_bytes(fs_info, si, bytes, flush); | |
597 | if (ret == 0) { | |
598 | if (extra_delayed_refs_bytes > 0) | |
599 | btrfs_migrate_to_delayed_refs_rsv(fs_info, | |
600 | extra_delayed_refs_bytes); | |
601 | return 0; | |
602 | } | |
603 | ||
604 | if (extra_delayed_refs_bytes > 0) { | |
605 | bytes -= extra_delayed_refs_bytes; | |
606 | ret = btrfs_reserve_metadata_bytes(fs_info, si, bytes, flush); | |
607 | if (ret == 0) | |
608 | return 0; | |
609 | } | |
610 | ||
611 | /* | |
612 | * If we are an emergency flush, which can steal from the global block | |
613 | * reserve, then attempt to not reserve space for the delayed refs, as | |
614 | * we will consume space for them from the global block reserve. | |
615 | */ | |
616 | if (flush == BTRFS_RESERVE_FLUSH_ALL_STEAL) { | |
617 | bytes -= *delayed_refs_bytes; | |
618 | *delayed_refs_bytes = 0; | |
619 | ret = btrfs_reserve_metadata_bytes(fs_info, si, bytes, flush); | |
620 | } | |
621 | ||
622 | return ret; | |
623 | } | |
624 | ||
08e007d2 | 625 | static struct btrfs_trans_handle * |
5aed1dd8 | 626 | start_transaction(struct btrfs_root *root, unsigned int num_items, |
003d7c59 JM |
627 | unsigned int type, enum btrfs_reserve_flush_enum flush, |
628 | bool enforce_qgroups) | |
37d1aeee | 629 | { |
0b246afa | 630 | struct btrfs_fs_info *fs_info = root->fs_info; |
ba2c4d4e | 631 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
28270e25 | 632 | struct btrfs_block_rsv *trans_rsv = &fs_info->trans_block_rsv; |
a22285a6 YZ |
633 | struct btrfs_trans_handle *h; |
634 | struct btrfs_transaction *cur_trans; | |
b5009945 | 635 | u64 num_bytes = 0; |
c5567237 | 636 | u64 qgroup_reserved = 0; |
28270e25 | 637 | u64 delayed_refs_bytes = 0; |
20dd2cbf | 638 | bool reloc_reserved = false; |
9c343784 | 639 | bool do_chunk_alloc = false; |
20dd2cbf | 640 | int ret; |
acce952b | 641 | |
84961539 | 642 | if (BTRFS_FS_ERROR(fs_info)) |
acce952b | 643 | return ERR_PTR(-EROFS); |
2a1eb461 | 644 | |
46c4e71e | 645 | if (current->journal_info) { |
0860adfd | 646 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 | 647 | h = current->journal_info; |
b50fff81 DS |
648 | refcount_inc(&h->use_count); |
649 | WARN_ON(refcount_read(&h->use_count) > 2); | |
2a1eb461 JB |
650 | h->orig_rsv = h->block_rsv; |
651 | h->block_rsv = NULL; | |
652 | goto got_it; | |
653 | } | |
b5009945 JB |
654 | |
655 | /* | |
656 | * Do the reservation before we join the transaction so we can do all | |
657 | * the appropriate flushing if need be. | |
658 | */ | |
003d7c59 | 659 | if (num_items && root != fs_info->chunk_root) { |
0b246afa | 660 | qgroup_reserved = num_items * fs_info->nodesize; |
a6496849 BB |
661 | /* |
662 | * Use prealloc for now, as there might be a currently running | |
663 | * transaction that could free this reserved space prematurely | |
664 | * by committing. | |
665 | */ | |
666 | ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_reserved, | |
667 | enforce_qgroups, false); | |
7174109c QW |
668 | if (ret) |
669 | return ERR_PTR(ret); | |
c5567237 | 670 | |
28270e25 | 671 | num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); |
ba2c4d4e | 672 | /* |
28270e25 FM |
673 | * If we plan to insert/update/delete "num_items" from a btree, |
674 | * we will also generate delayed refs for extent buffers in the | |
675 | * respective btree paths, so reserve space for the delayed refs | |
676 | * that will be generated by the caller as it modifies btrees. | |
677 | * Try to reserve them to avoid excessive use of the global | |
678 | * block reserve. | |
ba2c4d4e | 679 | */ |
28270e25 | 680 | delayed_refs_bytes = btrfs_calc_delayed_ref_bytes(fs_info, num_items); |
ba2c4d4e | 681 | |
20dd2cbf MX |
682 | /* |
683 | * Do the reservation for the relocation root creation | |
684 | */ | |
ee39b432 | 685 | if (need_reserve_reloc_root(root)) { |
0b246afa | 686 | num_bytes += fs_info->nodesize; |
20dd2cbf MX |
687 | reloc_reserved = true; |
688 | } | |
689 | ||
28270e25 FM |
690 | ret = btrfs_reserve_trans_metadata(fs_info, flush, num_bytes, |
691 | &delayed_refs_bytes); | |
ba2c4d4e JB |
692 | if (ret) |
693 | goto reserve_fail; | |
9c343784 | 694 | |
28270e25 FM |
695 | btrfs_block_rsv_add_bytes(trans_rsv, num_bytes, true); |
696 | ||
697 | if (trans_rsv->space_info->force_alloc) | |
9c343784 | 698 | do_chunk_alloc = true; |
ba2c4d4e | 699 | } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && |
748f553c | 700 | !btrfs_block_rsv_full(delayed_refs_rsv)) { |
ba2c4d4e JB |
701 | /* |
702 | * Some people call with btrfs_start_transaction(root, 0) | |
703 | * because they can be throttled, but have some other mechanism | |
704 | * for reserving space. We still want these guys to refill the | |
705 | * delayed block_rsv so just add 1 items worth of reservation | |
706 | * here. | |
707 | */ | |
708 | ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); | |
b5009945 | 709 | if (ret) |
843fcf35 | 710 | goto reserve_fail; |
b5009945 | 711 | } |
a22285a6 | 712 | again: |
f2f767e7 | 713 | h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); |
843fcf35 MX |
714 | if (!h) { |
715 | ret = -ENOMEM; | |
716 | goto alloc_fail; | |
717 | } | |
37d1aeee | 718 | |
98114659 JB |
719 | /* |
720 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
721 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
722 | * because we're already holding a ref. We need this because we could | |
723 | * have raced in and did an fsync() on a file which can kick a commit | |
724 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
725 | * |
726 | * If we are ATTACH, it means we just want to catch the current | |
727 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 728 | */ |
0860adfd | 729 | if (type & __TRANS_FREEZABLE) |
0b246afa | 730 | sb_start_intwrite(fs_info->sb); |
b2b5ef5c | 731 | |
2ff7e61e JM |
732 | if (may_wait_transaction(fs_info, type)) |
733 | wait_current_trans(fs_info); | |
a22285a6 | 734 | |
a4abeea4 | 735 | do { |
2ff7e61e | 736 | ret = join_transaction(fs_info, type); |
178260b2 | 737 | if (ret == -EBUSY) { |
2ff7e61e | 738 | wait_current_trans(fs_info); |
a6d155d2 FM |
739 | if (unlikely(type == TRANS_ATTACH || |
740 | type == TRANS_JOIN_NOSTART)) | |
178260b2 MX |
741 | ret = -ENOENT; |
742 | } | |
a4abeea4 JB |
743 | } while (ret == -EBUSY); |
744 | ||
a43f7f82 | 745 | if (ret < 0) |
843fcf35 | 746 | goto join_fail; |
0f7d52f4 | 747 | |
0b246afa | 748 | cur_trans = fs_info->running_transaction; |
a22285a6 YZ |
749 | |
750 | h->transid = cur_trans->transid; | |
751 | h->transaction = cur_trans; | |
b50fff81 | 752 | refcount_set(&h->use_count, 1); |
64b63580 | 753 | h->fs_info = root->fs_info; |
7174109c | 754 | |
a698d075 | 755 | h->type = type; |
ea658bad | 756 | INIT_LIST_HEAD(&h->new_bgs); |
28270e25 | 757 | btrfs_init_metadata_block_rsv(fs_info, &h->delayed_rsv, BTRFS_BLOCK_RSV_DELOPS); |
b7ec40d7 | 758 | |
a22285a6 | 759 | smp_mb(); |
3296bf56 | 760 | if (cur_trans->state >= TRANS_STATE_COMMIT_START && |
2ff7e61e | 761 | may_wait_transaction(fs_info, type)) { |
abdd2e80 | 762 | current->journal_info = h; |
3a45bb20 | 763 | btrfs_commit_transaction(h); |
a22285a6 YZ |
764 | goto again; |
765 | } | |
766 | ||
b5009945 | 767 | if (num_bytes) { |
0b246afa | 768 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 769 | h->transid, num_bytes, 1); |
28270e25 | 770 | h->block_rsv = trans_rsv; |
b5009945 | 771 | h->bytes_reserved = num_bytes; |
28270e25 FM |
772 | if (delayed_refs_bytes > 0) { |
773 | trace_btrfs_space_reservation(fs_info, | |
774 | "local_delayed_refs_rsv", | |
775 | h->transid, | |
776 | delayed_refs_bytes, 1); | |
777 | h->delayed_refs_bytes_reserved = delayed_refs_bytes; | |
778 | btrfs_block_rsv_add_bytes(&h->delayed_rsv, delayed_refs_bytes, true); | |
779 | delayed_refs_bytes = 0; | |
780 | } | |
20dd2cbf | 781 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 782 | } |
9ed74f2d | 783 | |
a6496849 BB |
784 | /* |
785 | * Now that we have found a transaction to be a part of, convert the | |
786 | * qgroup reservation from prealloc to pertrans. A different transaction | |
787 | * can't race in and free our pertrans out from under us. | |
788 | */ | |
789 | if (qgroup_reserved) | |
790 | btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved); | |
791 | ||
2a1eb461 | 792 | got_it: |
bcf3a3e7 | 793 | if (!current->journal_info) |
a22285a6 | 794 | current->journal_info = h; |
fcc99734 | 795 | |
9c343784 JB |
796 | /* |
797 | * If the space_info is marked ALLOC_FORCE then we'll get upgraded to | |
798 | * ALLOC_FORCE the first run through, and then we won't allocate for | |
799 | * anybody else who races in later. We don't care about the return | |
800 | * value here. | |
801 | */ | |
802 | if (do_chunk_alloc && num_bytes) { | |
803 | u64 flags = h->block_rsv->space_info->flags; | |
804 | ||
805 | btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), | |
806 | CHUNK_ALLOC_NO_FORCE); | |
807 | } | |
808 | ||
fcc99734 QW |
809 | /* |
810 | * btrfs_record_root_in_trans() needs to alloc new extents, and may | |
811 | * call btrfs_join_transaction() while we're also starting a | |
812 | * transaction. | |
813 | * | |
814 | * Thus it need to be called after current->journal_info initialized, | |
815 | * or we can deadlock. | |
816 | */ | |
68075ea8 JB |
817 | ret = btrfs_record_root_in_trans(h, root); |
818 | if (ret) { | |
819 | /* | |
820 | * The transaction handle is fully initialized and linked with | |
821 | * other structures so it needs to be ended in case of errors, | |
822 | * not just freed. | |
823 | */ | |
824 | btrfs_end_transaction(h); | |
825 | return ERR_PTR(ret); | |
826 | } | |
fcc99734 | 827 | |
79154b1b | 828 | return h; |
843fcf35 MX |
829 | |
830 | join_fail: | |
0860adfd | 831 | if (type & __TRANS_FREEZABLE) |
0b246afa | 832 | sb_end_intwrite(fs_info->sb); |
843fcf35 MX |
833 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
834 | alloc_fail: | |
835 | if (num_bytes) | |
28270e25 FM |
836 | btrfs_block_rsv_release(fs_info, trans_rsv, num_bytes, NULL); |
837 | if (delayed_refs_bytes) | |
838 | btrfs_space_info_free_bytes_may_use(fs_info, trans_rsv->space_info, | |
839 | delayed_refs_bytes); | |
843fcf35 | 840 | reserve_fail: |
a6496849 | 841 | btrfs_qgroup_free_meta_prealloc(root, qgroup_reserved); |
843fcf35 | 842 | return ERR_PTR(ret); |
79154b1b CM |
843 | } |
844 | ||
f9295749 | 845 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
5aed1dd8 | 846 | unsigned int num_items) |
f9295749 | 847 | { |
08e007d2 | 848 | return start_transaction(root, num_items, TRANS_START, |
003d7c59 | 849 | BTRFS_RESERVE_FLUSH_ALL, true); |
f9295749 | 850 | } |
003d7c59 | 851 | |
8eab77ff FM |
852 | struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( |
853 | struct btrfs_root *root, | |
7f9fe614 | 854 | unsigned int num_items) |
8eab77ff | 855 | { |
7f9fe614 JB |
856 | return start_transaction(root, num_items, TRANS_START, |
857 | BTRFS_RESERVE_FLUSH_ALL_STEAL, false); | |
8eab77ff | 858 | } |
8407aa46 | 859 | |
7a7eaa40 | 860 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 861 | { |
003d7c59 JM |
862 | return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, |
863 | true); | |
f9295749 CM |
864 | } |
865 | ||
8d510121 | 866 | struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) |
0af3d00b | 867 | { |
575a75d6 | 868 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, |
003d7c59 | 869 | BTRFS_RESERVE_NO_FLUSH, true); |
0af3d00b JB |
870 | } |
871 | ||
a6d155d2 FM |
872 | /* |
873 | * Similar to regular join but it never starts a transaction when none is | |
19288951 FM |
874 | * running or when there's a running one at a state >= TRANS_STATE_UNBLOCKED. |
875 | * This is similar to btrfs_attach_transaction() but it allows the join to | |
876 | * happen if the transaction commit already started but it's not yet in the | |
877 | * "doing" phase (the state is < TRANS_STATE_COMMIT_DOING). | |
a6d155d2 FM |
878 | */ |
879 | struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) | |
880 | { | |
881 | return start_transaction(root, 0, TRANS_JOIN_NOSTART, | |
882 | BTRFS_RESERVE_NO_FLUSH, true); | |
883 | } | |
884 | ||
d4edf39b | 885 | /* |
9580503b | 886 | * Catch the running transaction. |
d4edf39b MX |
887 | * |
888 | * It is used when we want to commit the current the transaction, but | |
889 | * don't want to start a new one. | |
890 | * | |
891 | * Note: If this function return -ENOENT, it just means there is no | |
892 | * running transaction. But it is possible that the inactive transaction | |
893 | * is still in the memory, not fully on disk. If you hope there is no | |
894 | * inactive transaction in the fs when -ENOENT is returned, you should | |
895 | * invoke | |
896 | * btrfs_attach_transaction_barrier() | |
897 | */ | |
354aa0fb | 898 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 899 | { |
575a75d6 | 900 | return start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 901 | BTRFS_RESERVE_NO_FLUSH, true); |
60376ce4 JB |
902 | } |
903 | ||
d4edf39b | 904 | /* |
9580503b | 905 | * Catch the running transaction. |
d4edf39b | 906 | * |
52042d8e | 907 | * It is similar to the above function, the difference is this one |
d4edf39b MX |
908 | * will wait for all the inactive transactions until they fully |
909 | * complete. | |
910 | */ | |
911 | struct btrfs_trans_handle * | |
912 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
913 | { | |
914 | struct btrfs_trans_handle *trans; | |
915 | ||
575a75d6 | 916 | trans = start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 917 | BTRFS_RESERVE_NO_FLUSH, true); |
b28ff3a7 FM |
918 | if (trans == ERR_PTR(-ENOENT)) { |
919 | int ret; | |
920 | ||
921 | ret = btrfs_wait_for_commit(root->fs_info, 0); | |
922 | if (ret) | |
923 | return ERR_PTR(ret); | |
924 | } | |
d4edf39b MX |
925 | |
926 | return trans; | |
927 | } | |
928 | ||
d0c2f4fa FM |
929 | /* Wait for a transaction commit to reach at least the given state. */ |
930 | static noinline void wait_for_commit(struct btrfs_transaction *commit, | |
931 | const enum btrfs_trans_state min_state) | |
89ce8a63 | 932 | { |
5fd76bf3 OS |
933 | struct btrfs_fs_info *fs_info = commit->fs_info; |
934 | u64 transid = commit->transid; | |
935 | bool put = false; | |
936 | ||
3e738c53 IA |
937 | /* |
938 | * At the moment this function is called with min_state either being | |
939 | * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED. | |
940 | */ | |
941 | if (min_state == TRANS_STATE_COMPLETED) | |
942 | btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); | |
943 | else | |
944 | btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); | |
945 | ||
5fd76bf3 OS |
946 | while (1) { |
947 | wait_event(commit->commit_wait, commit->state >= min_state); | |
948 | if (put) | |
949 | btrfs_put_transaction(commit); | |
950 | ||
951 | if (min_state < TRANS_STATE_COMPLETED) | |
952 | break; | |
953 | ||
954 | /* | |
955 | * A transaction isn't really completed until all of the | |
956 | * previous transactions are completed, but with fsync we can | |
957 | * end up with SUPER_COMMITTED transactions before a COMPLETED | |
958 | * transaction. Wait for those. | |
959 | */ | |
960 | ||
961 | spin_lock(&fs_info->trans_lock); | |
962 | commit = list_first_entry_or_null(&fs_info->trans_list, | |
963 | struct btrfs_transaction, | |
964 | list); | |
965 | if (!commit || commit->transid > transid) { | |
966 | spin_unlock(&fs_info->trans_lock); | |
967 | break; | |
968 | } | |
969 | refcount_inc(&commit->use_count); | |
970 | put = true; | |
971 | spin_unlock(&fs_info->trans_lock); | |
972 | } | |
89ce8a63 CM |
973 | } |
974 | ||
2ff7e61e | 975 | int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) |
46204592 SW |
976 | { |
977 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 978 | int ret = 0; |
46204592 | 979 | |
46204592 | 980 | if (transid) { |
0124855f | 981 | if (transid <= btrfs_get_last_trans_committed(fs_info)) |
a4abeea4 | 982 | goto out; |
46204592 SW |
983 | |
984 | /* find specified transaction */ | |
0b246afa JM |
985 | spin_lock(&fs_info->trans_lock); |
986 | list_for_each_entry(t, &fs_info->trans_list, list) { | |
46204592 SW |
987 | if (t->transid == transid) { |
988 | cur_trans = t; | |
9b64f57d | 989 | refcount_inc(&cur_trans->use_count); |
8cd2807f | 990 | ret = 0; |
46204592 SW |
991 | break; |
992 | } | |
8cd2807f MX |
993 | if (t->transid > transid) { |
994 | ret = 0; | |
46204592 | 995 | break; |
8cd2807f | 996 | } |
46204592 | 997 | } |
0b246afa | 998 | spin_unlock(&fs_info->trans_lock); |
42383020 SW |
999 | |
1000 | /* | |
1001 | * The specified transaction doesn't exist, or we | |
1002 | * raced with btrfs_commit_transaction | |
1003 | */ | |
1004 | if (!cur_trans) { | |
0124855f | 1005 | if (transid > btrfs_get_last_trans_committed(fs_info)) |
42383020 | 1006 | ret = -EINVAL; |
8cd2807f | 1007 | goto out; |
42383020 | 1008 | } |
46204592 SW |
1009 | } else { |
1010 | /* find newest transaction that is committing | committed */ | |
0b246afa JM |
1011 | spin_lock(&fs_info->trans_lock); |
1012 | list_for_each_entry_reverse(t, &fs_info->trans_list, | |
46204592 | 1013 | list) { |
4a9d8bde MX |
1014 | if (t->state >= TRANS_STATE_COMMIT_START) { |
1015 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 1016 | break; |
46204592 | 1017 | cur_trans = t; |
9b64f57d | 1018 | refcount_inc(&cur_trans->use_count); |
46204592 SW |
1019 | break; |
1020 | } | |
1021 | } | |
0b246afa | 1022 | spin_unlock(&fs_info->trans_lock); |
46204592 | 1023 | if (!cur_trans) |
a4abeea4 | 1024 | goto out; /* nothing committing|committed */ |
46204592 SW |
1025 | } |
1026 | ||
d0c2f4fa | 1027 | wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); |
bf7ecbe9 | 1028 | ret = cur_trans->aborted; |
724e2315 | 1029 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 1030 | out: |
46204592 SW |
1031 | return ret; |
1032 | } | |
1033 | ||
2ff7e61e | 1034 | void btrfs_throttle(struct btrfs_fs_info *fs_info) |
37d1aeee | 1035 | { |
92e2f7e3 | 1036 | wait_current_trans(fs_info); |
37d1aeee CM |
1037 | } |
1038 | ||
a2633b6a | 1039 | bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa YZ |
1040 | { |
1041 | struct btrfs_transaction *cur_trans = trans->transaction; | |
8929ecfa | 1042 | |
3296bf56 | 1043 | if (cur_trans->state >= TRANS_STATE_COMMIT_START || |
e19eb11f | 1044 | test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) |
a2633b6a | 1045 | return true; |
8929ecfa | 1046 | |
04fb3285 FM |
1047 | if (btrfs_check_space_for_delayed_refs(trans->fs_info)) |
1048 | return true; | |
1049 | ||
1050 | return !!btrfs_block_rsv_check(&trans->fs_info->global_block_rsv, 50); | |
8929ecfa YZ |
1051 | } |
1052 | ||
dc60c525 NB |
1053 | static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) |
1054 | ||
0e34693f | 1055 | { |
dc60c525 NB |
1056 | struct btrfs_fs_info *fs_info = trans->fs_info; |
1057 | ||
0e34693f NB |
1058 | if (!trans->block_rsv) { |
1059 | ASSERT(!trans->bytes_reserved); | |
28270e25 | 1060 | ASSERT(!trans->delayed_refs_bytes_reserved); |
0e34693f NB |
1061 | return; |
1062 | } | |
1063 | ||
28270e25 FM |
1064 | if (!trans->bytes_reserved) { |
1065 | ASSERT(!trans->delayed_refs_bytes_reserved); | |
0e34693f | 1066 | return; |
28270e25 | 1067 | } |
0e34693f NB |
1068 | |
1069 | ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); | |
1070 | trace_btrfs_space_reservation(fs_info, "transaction", | |
1071 | trans->transid, trans->bytes_reserved, 0); | |
1072 | btrfs_block_rsv_release(fs_info, trans->block_rsv, | |
63f018be | 1073 | trans->bytes_reserved, NULL); |
0e34693f | 1074 | trans->bytes_reserved = 0; |
28270e25 FM |
1075 | |
1076 | if (!trans->delayed_refs_bytes_reserved) | |
1077 | return; | |
1078 | ||
1079 | trace_btrfs_space_reservation(fs_info, "local_delayed_refs_rsv", | |
1080 | trans->transid, | |
1081 | trans->delayed_refs_bytes_reserved, 0); | |
1082 | btrfs_block_rsv_release(fs_info, &trans->delayed_rsv, | |
1083 | trans->delayed_refs_bytes_reserved, NULL); | |
1084 | trans->delayed_refs_bytes_reserved = 0; | |
0e34693f NB |
1085 | } |
1086 | ||
89ce8a63 | 1087 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
3a45bb20 | 1088 | int throttle) |
79154b1b | 1089 | { |
3a45bb20 | 1090 | struct btrfs_fs_info *info = trans->fs_info; |
8929ecfa | 1091 | struct btrfs_transaction *cur_trans = trans->transaction; |
4edc2ca3 | 1092 | int err = 0; |
c3e69d58 | 1093 | |
b50fff81 DS |
1094 | if (refcount_read(&trans->use_count) > 1) { |
1095 | refcount_dec(&trans->use_count); | |
2a1eb461 JB |
1096 | trans->block_rsv = trans->orig_rsv; |
1097 | return 0; | |
1098 | } | |
1099 | ||
dc60c525 | 1100 | btrfs_trans_release_metadata(trans); |
4c13d758 | 1101 | trans->block_rsv = NULL; |
c5567237 | 1102 | |
119e80df | 1103 | btrfs_create_pending_block_groups(trans); |
ea658bad | 1104 | |
4fbcdf66 FM |
1105 | btrfs_trans_release_chunk_metadata(trans); |
1106 | ||
0860adfd | 1107 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 1108 | sb_end_intwrite(info->sb); |
6df7881a | 1109 | |
8929ecfa | 1110 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
1111 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
1112 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 1113 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 1114 | |
093258e6 | 1115 | cond_wake_up(&cur_trans->writer_wait); |
e1489b4f | 1116 | |
5a9ba670 | 1117 | btrfs_lockdep_release(info, btrfs_trans_num_extwriters); |
e1489b4f IA |
1118 | btrfs_lockdep_release(info, btrfs_trans_num_writers); |
1119 | ||
724e2315 | 1120 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
1121 | |
1122 | if (current->journal_info == trans) | |
1123 | current->journal_info = NULL; | |
ab78c84d | 1124 | |
24bbcf04 | 1125 | if (throttle) |
2ff7e61e | 1126 | btrfs_run_delayed_iputs(info); |
24bbcf04 | 1127 | |
84961539 | 1128 | if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) { |
4e121c06 | 1129 | wake_up_process(info->transaction_kthread); |
fbabd4a3 JB |
1130 | if (TRANS_ABORTED(trans)) |
1131 | err = trans->aborted; | |
1132 | else | |
1133 | err = -EROFS; | |
4e121c06 | 1134 | } |
49b25e05 | 1135 | |
4edc2ca3 DJ |
1136 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
1137 | return err; | |
79154b1b CM |
1138 | } |
1139 | ||
3a45bb20 | 1140 | int btrfs_end_transaction(struct btrfs_trans_handle *trans) |
89ce8a63 | 1141 | { |
3a45bb20 | 1142 | return __btrfs_end_transaction(trans, 0); |
89ce8a63 CM |
1143 | } |
1144 | ||
3a45bb20 | 1145 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) |
89ce8a63 | 1146 | { |
3a45bb20 | 1147 | return __btrfs_end_transaction(trans, 1); |
16cdcec7 MX |
1148 | } |
1149 | ||
d352ac68 CM |
1150 | /* |
1151 | * when btree blocks are allocated, they have some corresponding bits set for | |
1152 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 1153 | * those extents are sent to disk but does not wait on them |
d352ac68 | 1154 | */ |
2ff7e61e | 1155 | int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, |
8cef4e16 | 1156 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 1157 | { |
777e6bd7 | 1158 | int err = 0; |
7c4452b9 | 1159 | int werr = 0; |
0b246afa | 1160 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 1161 | struct extent_state *cached_state = NULL; |
777e6bd7 | 1162 | u64 start = 0; |
5f39d397 | 1163 | u64 end; |
7c4452b9 | 1164 | |
e5860f82 FM |
1165 | while (find_first_extent_bit(dirty_pages, start, &start, &end, |
1166 | mark, &cached_state)) { | |
663dfbb0 FM |
1167 | bool wait_writeback = false; |
1168 | ||
1169 | err = convert_extent_bit(dirty_pages, start, end, | |
1170 | EXTENT_NEED_WAIT, | |
210aa277 | 1171 | mark, &cached_state); |
663dfbb0 FM |
1172 | /* |
1173 | * convert_extent_bit can return -ENOMEM, which is most of the | |
1174 | * time a temporary error. So when it happens, ignore the error | |
1175 | * and wait for writeback of this range to finish - because we | |
1176 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
bf89d38f JM |
1177 | * to __btrfs_wait_marked_extents() would not know that |
1178 | * writeback for this range started and therefore wouldn't | |
1179 | * wait for it to finish - we don't want to commit a | |
1180 | * superblock that points to btree nodes/leafs for which | |
1181 | * writeback hasn't finished yet (and without errors). | |
663dfbb0 | 1182 | * We cleanup any entries left in the io tree when committing |
41e7acd3 | 1183 | * the transaction (through extent_io_tree_release()). |
663dfbb0 FM |
1184 | */ |
1185 | if (err == -ENOMEM) { | |
1186 | err = 0; | |
1187 | wait_writeback = true; | |
1188 | } | |
1189 | if (!err) | |
1190 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
1191 | if (err) |
1192 | werr = err; | |
663dfbb0 FM |
1193 | else if (wait_writeback) |
1194 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 1195 | free_extent_state(cached_state); |
663dfbb0 | 1196 | cached_state = NULL; |
1728366e JB |
1197 | cond_resched(); |
1198 | start = end + 1; | |
7c4452b9 | 1199 | } |
690587d1 CM |
1200 | return werr; |
1201 | } | |
1202 | ||
1203 | /* | |
1204 | * when btree blocks are allocated, they have some corresponding bits set for | |
1205 | * them in one of two extent_io trees. This is used to make sure all of | |
1206 | * those extents are on disk for transaction or log commit. We wait | |
1207 | * on all the pages and clear them from the dirty pages state tree | |
1208 | */ | |
bf89d38f JM |
1209 | static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, |
1210 | struct extent_io_tree *dirty_pages) | |
690587d1 | 1211 | { |
690587d1 CM |
1212 | int err = 0; |
1213 | int werr = 0; | |
0b246afa | 1214 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 1215 | struct extent_state *cached_state = NULL; |
690587d1 CM |
1216 | u64 start = 0; |
1217 | u64 end; | |
777e6bd7 | 1218 | |
e5860f82 FM |
1219 | while (find_first_extent_bit(dirty_pages, start, &start, &end, |
1220 | EXTENT_NEED_WAIT, &cached_state)) { | |
663dfbb0 FM |
1221 | /* |
1222 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
1223 | * When committing the transaction, we'll remove any entries | |
1224 | * left in the io tree. For a log commit, we don't remove them | |
1225 | * after committing the log because the tree can be accessed | |
1226 | * concurrently - we do it only at transaction commit time when | |
41e7acd3 | 1227 | * it's safe to do it (through extent_io_tree_release()). |
663dfbb0 FM |
1228 | */ |
1229 | err = clear_extent_bit(dirty_pages, start, end, | |
bd015294 | 1230 | EXTENT_NEED_WAIT, &cached_state); |
663dfbb0 FM |
1231 | if (err == -ENOMEM) |
1232 | err = 0; | |
1233 | if (!err) | |
1234 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
1235 | if (err) |
1236 | werr = err; | |
e38e2ed7 FM |
1237 | free_extent_state(cached_state); |
1238 | cached_state = NULL; | |
1728366e JB |
1239 | cond_resched(); |
1240 | start = end + 1; | |
777e6bd7 | 1241 | } |
7c4452b9 CM |
1242 | if (err) |
1243 | werr = err; | |
bf89d38f JM |
1244 | return werr; |
1245 | } | |
656f30db | 1246 | |
b9fae2eb | 1247 | static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, |
bf89d38f JM |
1248 | struct extent_io_tree *dirty_pages) |
1249 | { | |
1250 | bool errors = false; | |
1251 | int err; | |
656f30db | 1252 | |
bf89d38f JM |
1253 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); |
1254 | if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) | |
1255 | errors = true; | |
1256 | ||
1257 | if (errors && !err) | |
1258 | err = -EIO; | |
1259 | return err; | |
1260 | } | |
656f30db | 1261 | |
bf89d38f JM |
1262 | int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) |
1263 | { | |
1264 | struct btrfs_fs_info *fs_info = log_root->fs_info; | |
1265 | struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; | |
1266 | bool errors = false; | |
1267 | int err; | |
656f30db | 1268 | |
bf89d38f JM |
1269 | ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
1270 | ||
1271 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); | |
1272 | if ((mark & EXTENT_DIRTY) && | |
1273 | test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) | |
1274 | errors = true; | |
1275 | ||
1276 | if ((mark & EXTENT_NEW) && | |
1277 | test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) | |
1278 | errors = true; | |
1279 | ||
1280 | if (errors && !err) | |
1281 | err = -EIO; | |
1282 | return err; | |
79154b1b CM |
1283 | } |
1284 | ||
690587d1 | 1285 | /* |
c9b577c0 NB |
1286 | * When btree blocks are allocated the corresponding extents are marked dirty. |
1287 | * This function ensures such extents are persisted on disk for transaction or | |
1288 | * log commit. | |
1289 | * | |
1290 | * @trans: transaction whose dirty pages we'd like to write | |
690587d1 | 1291 | */ |
70458a58 | 1292 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) |
690587d1 CM |
1293 | { |
1294 | int ret; | |
1295 | int ret2; | |
c9b577c0 | 1296 | struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; |
70458a58 | 1297 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c6adc9cc | 1298 | struct blk_plug plug; |
690587d1 | 1299 | |
c6adc9cc | 1300 | blk_start_plug(&plug); |
c9b577c0 | 1301 | ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); |
c6adc9cc | 1302 | blk_finish_plug(&plug); |
bf89d38f | 1303 | ret2 = btrfs_wait_extents(fs_info, dirty_pages); |
bf0da8c1 | 1304 | |
41e7acd3 | 1305 | extent_io_tree_release(&trans->transaction->dirty_pages); |
c9b577c0 | 1306 | |
bf0da8c1 CM |
1307 | if (ret) |
1308 | return ret; | |
c9b577c0 | 1309 | else if (ret2) |
bf0da8c1 | 1310 | return ret2; |
c9b577c0 NB |
1311 | else |
1312 | return 0; | |
d0c803c4 CM |
1313 | } |
1314 | ||
d352ac68 CM |
1315 | /* |
1316 | * this is used to update the root pointer in the tree of tree roots. | |
1317 | * | |
1318 | * But, in the case of the extent allocation tree, updating the root | |
1319 | * pointer may allocate blocks which may change the root of the extent | |
1320 | * allocation tree. | |
1321 | * | |
1322 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1323 | * change while the root pointer was being updated in the metadata. | |
1324 | */ | |
0b86a832 CM |
1325 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1326 | struct btrfs_root *root) | |
79154b1b CM |
1327 | { |
1328 | int ret; | |
0b86a832 | 1329 | u64 old_root_bytenr; |
86b9f2ec | 1330 | u64 old_root_used; |
0b246afa JM |
1331 | struct btrfs_fs_info *fs_info = root->fs_info; |
1332 | struct btrfs_root *tree_root = fs_info->tree_root; | |
79154b1b | 1333 | |
86b9f2ec | 1334 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1335 | |
d397712b | 1336 | while (1) { |
0b86a832 | 1337 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1338 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1339 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1340 | break; |
87ef2bb4 | 1341 | |
5d4f98a2 | 1342 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1343 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1344 | &root->root_key, |
1345 | &root->root_item); | |
49b25e05 JM |
1346 | if (ret) |
1347 | return ret; | |
56bec294 | 1348 | |
86b9f2ec | 1349 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1350 | } |
276e680d | 1351 | |
0b86a832 CM |
1352 | return 0; |
1353 | } | |
1354 | ||
d352ac68 CM |
1355 | /* |
1356 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1357 | * |
1358 | * The error handling in this function may not be obvious. Any of the | |
1359 | * failures will cause the file system to go offline. We still need | |
1360 | * to clean up the delayed refs. | |
d352ac68 | 1361 | */ |
9386d8bc | 1362 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) |
0b86a832 | 1363 | { |
9386d8bc | 1364 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ea526d18 | 1365 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1366 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1367 | struct list_head *next; |
84234f3a | 1368 | struct extent_buffer *eb; |
56bec294 | 1369 | int ret; |
84234f3a | 1370 | |
dfba78dc FM |
1371 | /* |
1372 | * At this point no one can be using this transaction to modify any tree | |
1373 | * and no one can start another transaction to modify any tree either. | |
1374 | */ | |
1375 | ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); | |
1376 | ||
84234f3a | 1377 | eb = btrfs_lock_root_node(fs_info->tree_root); |
49b25e05 | 1378 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
9631e4cc | 1379 | 0, &eb, BTRFS_NESTING_COW); |
84234f3a YZ |
1380 | btrfs_tree_unlock(eb); |
1381 | free_extent_buffer(eb); | |
0b86a832 | 1382 | |
49b25e05 JM |
1383 | if (ret) |
1384 | return ret; | |
87ef2bb4 | 1385 | |
196c9d8d | 1386 | ret = btrfs_run_dev_stats(trans); |
c16ce190 JB |
1387 | if (ret) |
1388 | return ret; | |
2b584c68 | 1389 | ret = btrfs_run_dev_replace(trans); |
c16ce190 JB |
1390 | if (ret) |
1391 | return ret; | |
280f8bd2 | 1392 | ret = btrfs_run_qgroups(trans); |
c16ce190 JB |
1393 | if (ret) |
1394 | return ret; | |
546adb0d | 1395 | |
bbebb3e0 | 1396 | ret = btrfs_setup_space_cache(trans); |
dcdf7f6d JB |
1397 | if (ret) |
1398 | return ret; | |
1399 | ||
ea526d18 | 1400 | again: |
d397712b | 1401 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
2ff7e61e | 1402 | struct btrfs_root *root; |
0b86a832 CM |
1403 | next = fs_info->dirty_cowonly_roots.next; |
1404 | list_del_init(next); | |
1405 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1406 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1407 | |
826582ca JB |
1408 | list_add_tail(&root->dirty_list, |
1409 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1410 | ret = update_cowonly_root(trans, root); |
1411 | if (ret) | |
1412 | return ret; | |
79154b1b | 1413 | } |
276e680d | 1414 | |
488bc2a2 | 1415 | /* Now flush any delayed refs generated by updating all of the roots */ |
8a526c44 | 1416 | ret = btrfs_run_delayed_refs(trans, U64_MAX); |
488bc2a2 JB |
1417 | if (ret) |
1418 | return ret; | |
1419 | ||
1bbc621e | 1420 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
5742d15f | 1421 | ret = btrfs_write_dirty_block_groups(trans); |
ea526d18 JB |
1422 | if (ret) |
1423 | return ret; | |
488bc2a2 JB |
1424 | |
1425 | /* | |
1426 | * We're writing the dirty block groups, which could generate | |
1427 | * delayed refs, which could generate more dirty block groups, | |
1428 | * so we want to keep this flushing in this loop to make sure | |
1429 | * everything gets run. | |
1430 | */ | |
8a526c44 | 1431 | ret = btrfs_run_delayed_refs(trans, U64_MAX); |
ea526d18 JB |
1432 | if (ret) |
1433 | return ret; | |
1434 | } | |
1435 | ||
1436 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1437 | goto again; | |
1438 | ||
9f6cbcbb DS |
1439 | /* Update dev-replace pointer once everything is committed */ |
1440 | fs_info->dev_replace.committed_cursor_left = | |
1441 | fs_info->dev_replace.cursor_left_last_write_of_item; | |
8dabb742 | 1442 | |
79154b1b CM |
1443 | return 0; |
1444 | } | |
1445 | ||
b4be6aef JB |
1446 | /* |
1447 | * If we had a pending drop we need to see if there are any others left in our | |
1448 | * dead roots list, and if not clear our bit and wake any waiters. | |
1449 | */ | |
1450 | void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info) | |
1451 | { | |
1452 | /* | |
1453 | * We put the drop in progress roots at the front of the list, so if the | |
1454 | * first entry doesn't have UNFINISHED_DROP set we can wake everybody | |
1455 | * up. | |
1456 | */ | |
1457 | spin_lock(&fs_info->trans_lock); | |
1458 | if (!list_empty(&fs_info->dead_roots)) { | |
1459 | struct btrfs_root *root = list_first_entry(&fs_info->dead_roots, | |
1460 | struct btrfs_root, | |
1461 | root_list); | |
1462 | if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) { | |
1463 | spin_unlock(&fs_info->trans_lock); | |
1464 | return; | |
1465 | } | |
1466 | } | |
1467 | spin_unlock(&fs_info->trans_lock); | |
1468 | ||
1469 | btrfs_wake_unfinished_drop(fs_info); | |
1470 | } | |
1471 | ||
d352ac68 CM |
1472 | /* |
1473 | * dead roots are old snapshots that need to be deleted. This allocates | |
1474 | * a dirty root struct and adds it into the list of dead roots that need to | |
1475 | * be deleted | |
1476 | */ | |
cfad392b | 1477 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1478 | { |
0b246afa JM |
1479 | struct btrfs_fs_info *fs_info = root->fs_info; |
1480 | ||
1481 | spin_lock(&fs_info->trans_lock); | |
dc9492c1 JB |
1482 | if (list_empty(&root->root_list)) { |
1483 | btrfs_grab_root(root); | |
b4be6aef JB |
1484 | |
1485 | /* We want to process the partially complete drops first. */ | |
1486 | if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) | |
1487 | list_add(&root->root_list, &fs_info->dead_roots); | |
1488 | else | |
1489 | list_add_tail(&root->root_list, &fs_info->dead_roots); | |
dc9492c1 | 1490 | } |
0b246afa | 1491 | spin_unlock(&fs_info->trans_lock); |
5eda7b5e CM |
1492 | } |
1493 | ||
d352ac68 | 1494 | /* |
dfba78dc FM |
1495 | * Update each subvolume root and its relocation root, if it exists, in the tree |
1496 | * of tree roots. Also free log roots if they exist. | |
d352ac68 | 1497 | */ |
7e4443d9 | 1498 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) |
0f7d52f4 | 1499 | { |
7e4443d9 | 1500 | struct btrfs_fs_info *fs_info = trans->fs_info; |
fc7cbcd4 DS |
1501 | struct btrfs_root *gang[8]; |
1502 | int i; | |
1503 | int ret; | |
54aa1f4d | 1504 | |
dfba78dc FM |
1505 | /* |
1506 | * At this point no one can be using this transaction to modify any tree | |
1507 | * and no one can start another transaction to modify any tree either. | |
1508 | */ | |
1509 | ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); | |
1510 | ||
fc7cbcd4 DS |
1511 | spin_lock(&fs_info->fs_roots_radix_lock); |
1512 | while (1) { | |
1513 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, | |
1514 | (void **)gang, 0, | |
1515 | ARRAY_SIZE(gang), | |
1516 | BTRFS_ROOT_TRANS_TAG); | |
1517 | if (ret == 0) | |
1518 | break; | |
1519 | for (i = 0; i < ret; i++) { | |
1520 | struct btrfs_root *root = gang[i]; | |
1521 | int ret2; | |
1522 | ||
1523 | /* | |
1524 | * At this point we can neither have tasks logging inodes | |
1525 | * from a root nor trying to commit a log tree. | |
1526 | */ | |
1527 | ASSERT(atomic_read(&root->log_writers) == 0); | |
1528 | ASSERT(atomic_read(&root->log_commit[0]) == 0); | |
1529 | ASSERT(atomic_read(&root->log_commit[1]) == 0); | |
1530 | ||
1531 | radix_tree_tag_clear(&fs_info->fs_roots_radix, | |
1532 | (unsigned long)root->root_key.objectid, | |
1533 | BTRFS_ROOT_TRANS_TAG); | |
1534 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
1535 | ||
1536 | btrfs_free_log(trans, root); | |
1537 | ret2 = btrfs_update_reloc_root(trans, root); | |
1538 | if (ret2) | |
1539 | return ret2; | |
1540 | ||
1541 | /* see comments in should_cow_block() */ | |
1542 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); | |
1543 | smp_mb__after_atomic(); | |
1544 | ||
1545 | if (root->commit_root != root->node) { | |
1546 | list_add_tail(&root->dirty_list, | |
1547 | &trans->transaction->switch_commits); | |
1548 | btrfs_set_root_node(&root->root_item, | |
1549 | root->node); | |
1550 | } | |
48b36a60 | 1551 | |
fc7cbcd4 DS |
1552 | ret2 = btrfs_update_root(trans, fs_info->tree_root, |
1553 | &root->root_key, | |
1554 | &root->root_item); | |
1555 | if (ret2) | |
1556 | return ret2; | |
1557 | spin_lock(&fs_info->fs_roots_radix_lock); | |
1558 | btrfs_qgroup_free_meta_all_pertrans(root); | |
0f7d52f4 CM |
1559 | } |
1560 | } | |
fc7cbcd4 | 1561 | spin_unlock(&fs_info->fs_roots_radix_lock); |
4f4317c1 | 1562 | return 0; |
0f7d52f4 CM |
1563 | } |
1564 | ||
6426c7ad QW |
1565 | /* |
1566 | * Do all special snapshot related qgroup dirty hack. | |
1567 | * | |
1568 | * Will do all needed qgroup inherit and dirty hack like switch commit | |
1569 | * roots inside one transaction and write all btree into disk, to make | |
1570 | * qgroup works. | |
1571 | */ | |
1572 | static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, | |
1573 | struct btrfs_root *src, | |
1574 | struct btrfs_root *parent, | |
1575 | struct btrfs_qgroup_inherit *inherit, | |
1576 | u64 dst_objectid) | |
1577 | { | |
1578 | struct btrfs_fs_info *fs_info = src->fs_info; | |
1579 | int ret; | |
1580 | ||
1581 | /* | |
5343cd93 BB |
1582 | * Save some performance in the case that qgroups are not enabled. If |
1583 | * this check races with the ioctl, rescan will kick in anyway. | |
6426c7ad | 1584 | */ |
182940f4 | 1585 | if (!btrfs_qgroup_full_accounting(fs_info)) |
6426c7ad | 1586 | return 0; |
6426c7ad | 1587 | |
4d31778a | 1588 | /* |
52042d8e | 1589 | * Ensure dirty @src will be committed. Or, after coming |
4d31778a QW |
1590 | * commit_fs_roots() and switch_commit_roots(), any dirty but not |
1591 | * recorded root will never be updated again, causing an outdated root | |
1592 | * item. | |
1593 | */ | |
1c442d22 JB |
1594 | ret = record_root_in_trans(trans, src, 1); |
1595 | if (ret) | |
1596 | return ret; | |
4d31778a | 1597 | |
2a4d84c1 JB |
1598 | /* |
1599 | * btrfs_qgroup_inherit relies on a consistent view of the usage for the | |
1600 | * src root, so we must run the delayed refs here. | |
1601 | * | |
1602 | * However this isn't particularly fool proof, because there's no | |
1603 | * synchronization keeping us from changing the tree after this point | |
1604 | * before we do the qgroup_inherit, or even from making changes while | |
1605 | * we're doing the qgroup_inherit. But that's a problem for the future, | |
1606 | * for now flush the delayed refs to narrow the race window where the | |
1607 | * qgroup counters could end up wrong. | |
1608 | */ | |
8a526c44 | 1609 | ret = btrfs_run_delayed_refs(trans, U64_MAX); |
2a4d84c1 JB |
1610 | if (ret) { |
1611 | btrfs_abort_transaction(trans, ret); | |
44365827 | 1612 | return ret; |
2a4d84c1 JB |
1613 | } |
1614 | ||
7e4443d9 | 1615 | ret = commit_fs_roots(trans); |
6426c7ad QW |
1616 | if (ret) |
1617 | goto out; | |
460fb20a | 1618 | ret = btrfs_qgroup_account_extents(trans); |
6426c7ad QW |
1619 | if (ret < 0) |
1620 | goto out; | |
1621 | ||
1622 | /* Now qgroup are all updated, we can inherit it to new qgroups */ | |
a9377422 | 1623 | ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, |
5343cd93 | 1624 | parent->root_key.objectid, inherit); |
6426c7ad QW |
1625 | if (ret < 0) |
1626 | goto out; | |
1627 | ||
1628 | /* | |
1629 | * Now we do a simplified commit transaction, which will: | |
1630 | * 1) commit all subvolume and extent tree | |
1631 | * To ensure all subvolume and extent tree have a valid | |
1632 | * commit_root to accounting later insert_dir_item() | |
1633 | * 2) write all btree blocks onto disk | |
1634 | * This is to make sure later btree modification will be cowed | |
1635 | * Or commit_root can be populated and cause wrong qgroup numbers | |
1636 | * In this simplified commit, we don't really care about other trees | |
1637 | * like chunk and root tree, as they won't affect qgroup. | |
1638 | * And we don't write super to avoid half committed status. | |
1639 | */ | |
9386d8bc | 1640 | ret = commit_cowonly_roots(trans); |
6426c7ad QW |
1641 | if (ret) |
1642 | goto out; | |
889bfa39 | 1643 | switch_commit_roots(trans); |
70458a58 | 1644 | ret = btrfs_write_and_wait_transaction(trans); |
6426c7ad | 1645 | if (ret) |
f7af3934 | 1646 | btrfs_handle_fs_error(fs_info, ret, |
6426c7ad QW |
1647 | "Error while writing out transaction for qgroup"); |
1648 | ||
1649 | out: | |
6426c7ad QW |
1650 | /* |
1651 | * Force parent root to be updated, as we recorded it before so its | |
1652 | * last_trans == cur_transid. | |
1653 | * Or it won't be committed again onto disk after later | |
1654 | * insert_dir_item() | |
1655 | */ | |
1656 | if (!ret) | |
1c442d22 | 1657 | ret = record_root_in_trans(trans, parent, 1); |
6426c7ad QW |
1658 | return ret; |
1659 | } | |
1660 | ||
d352ac68 CM |
1661 | /* |
1662 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1663 | * transaction commit. This does the actual creation. |
1664 | * | |
1665 | * Note: | |
1666 | * If the error which may affect the commitment of the current transaction | |
1667 | * happens, we should return the error number. If the error which just affect | |
1668 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1669 | */ |
80b6794d | 1670 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1671 | struct btrfs_pending_snapshot *pending) |
1672 | { | |
08d50ca3 NB |
1673 | |
1674 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3063d29f | 1675 | struct btrfs_key key; |
80b6794d | 1676 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1677 | struct btrfs_root *tree_root = fs_info->tree_root; |
1678 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1679 | struct btrfs_root *parent_root; |
98c9942a | 1680 | struct btrfs_block_rsv *rsv; |
ab3c5c18 | 1681 | struct inode *parent_inode = pending->dir; |
42874b3d MX |
1682 | struct btrfs_path *path; |
1683 | struct btrfs_dir_item *dir_item; | |
3063d29f | 1684 | struct extent_buffer *tmp; |
925baedd | 1685 | struct extent_buffer *old; |
95582b00 | 1686 | struct timespec64 cur_time; |
aec8030a | 1687 | int ret = 0; |
d68fc57b | 1688 | u64 to_reserve = 0; |
6bdb72de | 1689 | u64 index = 0; |
a22285a6 | 1690 | u64 objectid; |
b83cc969 | 1691 | u64 root_flags; |
ab3c5c18 STD |
1692 | unsigned int nofs_flags; |
1693 | struct fscrypt_name fname; | |
3063d29f | 1694 | |
8546b570 DS |
1695 | ASSERT(pending->path); |
1696 | path = pending->path; | |
42874b3d | 1697 | |
b0c0ea63 DS |
1698 | ASSERT(pending->root_item); |
1699 | new_root_item = pending->root_item; | |
a22285a6 | 1700 | |
ab3c5c18 STD |
1701 | /* |
1702 | * We're inside a transaction and must make sure that any potential | |
1703 | * allocations with GFP_KERNEL in fscrypt won't recurse back to | |
1704 | * filesystem. | |
1705 | */ | |
1706 | nofs_flags = memalloc_nofs_save(); | |
1707 | pending->error = fscrypt_setup_filename(parent_inode, | |
1708 | &pending->dentry->d_name, 0, | |
1709 | &fname); | |
1710 | memalloc_nofs_restore(nofs_flags); | |
1711 | if (pending->error) | |
1712 | goto free_pending; | |
ab3c5c18 | 1713 | |
543068a2 | 1714 | pending->error = btrfs_get_free_objectid(tree_root, &objectid); |
aec8030a | 1715 | if (pending->error) |
ab3c5c18 | 1716 | goto free_fname; |
3063d29f | 1717 | |
d6726335 QW |
1718 | /* |
1719 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1720 | * accounted by later btrfs_qgroup_inherit(). | |
1721 | */ | |
1722 | btrfs_set_skip_qgroup(trans, objectid); | |
1723 | ||
147d256e | 1724 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
d68fc57b YZ |
1725 | |
1726 | if (to_reserve > 0) { | |
9270501c | 1727 | pending->error = btrfs_block_rsv_add(fs_info, |
aec8030a MX |
1728 | &pending->block_rsv, |
1729 | to_reserve, | |
1730 | BTRFS_RESERVE_NO_FLUSH); | |
1731 | if (pending->error) | |
d6726335 | 1732 | goto clear_skip_qgroup; |
d68fc57b YZ |
1733 | } |
1734 | ||
3063d29f | 1735 | key.objectid = objectid; |
a22285a6 YZ |
1736 | key.offset = (u64)-1; |
1737 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1738 | |
6fa9700e | 1739 | rsv = trans->block_rsv; |
a22285a6 | 1740 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1741 | trans->bytes_reserved = trans->block_rsv->reserved; |
0b246afa | 1742 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa JB |
1743 | trans->transid, |
1744 | trans->bytes_reserved, 1); | |
a22285a6 | 1745 | parent_root = BTRFS_I(parent_inode)->root; |
f0118cb6 JB |
1746 | ret = record_root_in_trans(trans, parent_root, 0); |
1747 | if (ret) | |
1748 | goto fail; | |
c2050a45 | 1749 | cur_time = current_time(parent_inode); |
04b285f3 | 1750 | |
3063d29f CM |
1751 | /* |
1752 | * insert the directory item | |
1753 | */ | |
877574e2 | 1754 | ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); |
df9f2782 FM |
1755 | if (ret) { |
1756 | btrfs_abort_transaction(trans, ret); | |
1757 | goto fail; | |
1758 | } | |
42874b3d MX |
1759 | |
1760 | /* check if there is a file/dir which has the same name. */ | |
1761 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
4a0cc7ca | 1762 | btrfs_ino(BTRFS_I(parent_inode)), |
6db75318 | 1763 | &fname.disk_name, 0); |
42874b3d | 1764 | if (dir_item != NULL && !IS_ERR(dir_item)) { |
fe66a05a | 1765 | pending->error = -EEXIST; |
aec8030a | 1766 | goto dir_item_existed; |
42874b3d MX |
1767 | } else if (IS_ERR(dir_item)) { |
1768 | ret = PTR_ERR(dir_item); | |
66642832 | 1769 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1770 | goto fail; |
79787eaa | 1771 | } |
42874b3d | 1772 | btrfs_release_path(path); |
52c26179 | 1773 | |
6ed05643 | 1774 | ret = btrfs_create_qgroup(trans, objectid); |
8049ba5d | 1775 | if (ret && ret != -EEXIST) { |
6ed05643 BB |
1776 | btrfs_abort_transaction(trans, ret); |
1777 | goto fail; | |
1778 | } | |
1779 | ||
e999376f CM |
1780 | /* |
1781 | * pull in the delayed directory update | |
1782 | * and the delayed inode item | |
1783 | * otherwise we corrupt the FS during | |
1784 | * snapshot | |
1785 | */ | |
e5c304e6 | 1786 | ret = btrfs_run_delayed_items(trans); |
8732d44f | 1787 | if (ret) { /* Transaction aborted */ |
66642832 | 1788 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1789 | goto fail; |
1790 | } | |
e999376f | 1791 | |
f0118cb6 JB |
1792 | ret = record_root_in_trans(trans, root, 0); |
1793 | if (ret) { | |
1794 | btrfs_abort_transaction(trans, ret); | |
1795 | goto fail; | |
1796 | } | |
6bdb72de SW |
1797 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1798 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1799 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1800 | |
b83cc969 LZ |
1801 | root_flags = btrfs_root_flags(new_root_item); |
1802 | if (pending->readonly) | |
1803 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1804 | else | |
1805 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1806 | btrfs_set_root_flags(new_root_item, root_flags); | |
1807 | ||
8ea05e3a AB |
1808 | btrfs_set_root_generation_v2(new_root_item, |
1809 | trans->transid); | |
807fc790 | 1810 | generate_random_guid(new_root_item->uuid); |
8ea05e3a AB |
1811 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, |
1812 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1813 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1814 | memset(new_root_item->received_uuid, 0, | |
1815 | sizeof(new_root_item->received_uuid)); | |
1816 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1817 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1818 | btrfs_set_root_stransid(new_root_item, 0); | |
1819 | btrfs_set_root_rtransid(new_root_item, 0); | |
1820 | } | |
3cae210f QW |
1821 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1822 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1823 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1824 | |
6bdb72de | 1825 | old = btrfs_lock_root_node(root); |
9631e4cc JB |
1826 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, |
1827 | BTRFS_NESTING_COW); | |
79787eaa JM |
1828 | if (ret) { |
1829 | btrfs_tree_unlock(old); | |
1830 | free_extent_buffer(old); | |
66642832 | 1831 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1832 | goto fail; |
79787eaa | 1833 | } |
49b25e05 | 1834 | |
49b25e05 | 1835 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1836 | /* clean up in any case */ |
6bdb72de SW |
1837 | btrfs_tree_unlock(old); |
1838 | free_extent_buffer(old); | |
8732d44f | 1839 | if (ret) { |
66642832 | 1840 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1841 | goto fail; |
1842 | } | |
f1ebcc74 | 1843 | /* see comments in should_cow_block() */ |
27cdeb70 | 1844 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1845 | smp_wmb(); |
1846 | ||
6bdb72de | 1847 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1848 | /* record when the snapshot was created in key.offset */ |
1849 | key.offset = trans->transid; | |
1850 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1851 | btrfs_tree_unlock(tmp); |
1852 | free_extent_buffer(tmp); | |
8732d44f | 1853 | if (ret) { |
66642832 | 1854 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1855 | goto fail; |
1856 | } | |
6bdb72de | 1857 | |
a22285a6 YZ |
1858 | /* |
1859 | * insert root back/forward references | |
1860 | */ | |
6025c19f | 1861 | ret = btrfs_add_root_ref(trans, objectid, |
0660b5af | 1862 | parent_root->root_key.objectid, |
4a0cc7ca | 1863 | btrfs_ino(BTRFS_I(parent_inode)), index, |
6db75318 | 1864 | &fname.disk_name); |
8732d44f | 1865 | if (ret) { |
66642832 | 1866 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1867 | goto fail; |
1868 | } | |
0660b5af | 1869 | |
a22285a6 | 1870 | key.offset = (u64)-1; |
2dfb1e43 | 1871 | pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); |
79787eaa JM |
1872 | if (IS_ERR(pending->snap)) { |
1873 | ret = PTR_ERR(pending->snap); | |
2d892ccd | 1874 | pending->snap = NULL; |
66642832 | 1875 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1876 | goto fail; |
79787eaa | 1877 | } |
d68fc57b | 1878 | |
49b25e05 | 1879 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f | 1880 | if (ret) { |
66642832 | 1881 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1882 | goto fail; |
1883 | } | |
361048f5 | 1884 | |
6426c7ad QW |
1885 | /* |
1886 | * Do special qgroup accounting for snapshot, as we do some qgroup | |
1887 | * snapshot hack to do fast snapshot. | |
1888 | * To co-operate with that hack, we do hack again. | |
1889 | * Or snapshot will be greatly slowed down by a subtree qgroup rescan | |
1890 | */ | |
5343cd93 BB |
1891 | if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_FULL) |
1892 | ret = qgroup_account_snapshot(trans, root, parent_root, | |
1893 | pending->inherit, objectid); | |
1894 | else if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE) | |
1895 | ret = btrfs_qgroup_inherit(trans, root->root_key.objectid, objectid, | |
1896 | parent_root->root_key.objectid, pending->inherit); | |
6426c7ad QW |
1897 | if (ret < 0) |
1898 | goto fail; | |
1899 | ||
6db75318 STD |
1900 | ret = btrfs_insert_dir_item(trans, &fname.disk_name, |
1901 | BTRFS_I(parent_inode), &key, BTRFS_FT_DIR, | |
1902 | index); | |
42874b3d | 1903 | /* We have check then name at the beginning, so it is impossible. */ |
9c52057c | 1904 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f | 1905 | if (ret) { |
66642832 | 1906 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1907 | goto fail; |
1908 | } | |
42874b3d | 1909 | |
6ef06d27 | 1910 | btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + |
6db75318 | 1911 | fname.disk_name.len * 2); |
b1c38a13 JL |
1912 | inode_set_mtime_to_ts(parent_inode, |
1913 | inode_set_ctime_current(parent_inode)); | |
0a5d0dc5 | 1914 | ret = btrfs_update_inode_fallback(trans, BTRFS_I(parent_inode)); |
dd5f9615 | 1915 | if (ret) { |
66642832 | 1916 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1917 | goto fail; |
1918 | } | |
807fc790 AS |
1919 | ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, |
1920 | BTRFS_UUID_KEY_SUBVOL, | |
cdb345a8 | 1921 | objectid); |
dd5f9615 | 1922 | if (ret) { |
66642832 | 1923 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1924 | goto fail; |
1925 | } | |
1926 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
cdb345a8 | 1927 | ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, |
dd5f9615 SB |
1928 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, |
1929 | objectid); | |
1930 | if (ret && ret != -EEXIST) { | |
66642832 | 1931 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1932 | goto fail; |
1933 | } | |
1934 | } | |
d6726335 | 1935 | |
3063d29f | 1936 | fail: |
aec8030a MX |
1937 | pending->error = ret; |
1938 | dir_item_existed: | |
98c9942a | 1939 | trans->block_rsv = rsv; |
2382c5cc | 1940 | trans->bytes_reserved = 0; |
d6726335 QW |
1941 | clear_skip_qgroup: |
1942 | btrfs_clear_skip_qgroup(trans); | |
ab3c5c18 STD |
1943 | free_fname: |
1944 | fscrypt_free_filename(&fname); | |
1945 | free_pending: | |
6fa9700e | 1946 | kfree(new_root_item); |
b0c0ea63 | 1947 | pending->root_item = NULL; |
42874b3d | 1948 | btrfs_free_path(path); |
8546b570 DS |
1949 | pending->path = NULL; |
1950 | ||
49b25e05 | 1951 | return ret; |
3063d29f CM |
1952 | } |
1953 | ||
d352ac68 CM |
1954 | /* |
1955 | * create all the snapshots we've scheduled for creation | |
1956 | */ | |
08d50ca3 | 1957 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) |
3de4586c | 1958 | { |
aec8030a | 1959 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1960 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1961 | int ret = 0; |
3de4586c | 1962 | |
aec8030a MX |
1963 | list_for_each_entry_safe(pending, next, head, list) { |
1964 | list_del(&pending->list); | |
08d50ca3 | 1965 | ret = create_pending_snapshot(trans, pending); |
aec8030a MX |
1966 | if (ret) |
1967 | break; | |
1968 | } | |
1969 | return ret; | |
3de4586c CM |
1970 | } |
1971 | ||
2ff7e61e | 1972 | static void update_super_roots(struct btrfs_fs_info *fs_info) |
5d4f98a2 YZ |
1973 | { |
1974 | struct btrfs_root_item *root_item; | |
1975 | struct btrfs_super_block *super; | |
1976 | ||
0b246afa | 1977 | super = fs_info->super_copy; |
5d4f98a2 | 1978 | |
0b246afa | 1979 | root_item = &fs_info->chunk_root->root_item; |
093e037c DS |
1980 | super->chunk_root = root_item->bytenr; |
1981 | super->chunk_root_generation = root_item->generation; | |
1982 | super->chunk_root_level = root_item->level; | |
5d4f98a2 | 1983 | |
0b246afa | 1984 | root_item = &fs_info->tree_root->root_item; |
093e037c DS |
1985 | super->root = root_item->bytenr; |
1986 | super->generation = root_item->generation; | |
1987 | super->root_level = root_item->level; | |
0b246afa | 1988 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) |
093e037c | 1989 | super->cache_generation = root_item->generation; |
94846229 BB |
1990 | else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) |
1991 | super->cache_generation = 0; | |
0b246afa | 1992 | if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) |
093e037c | 1993 | super->uuid_tree_generation = root_item->generation; |
5d4f98a2 YZ |
1994 | } |
1995 | ||
f36f3042 CM |
1996 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1997 | { | |
4a9d8bde | 1998 | struct btrfs_transaction *trans; |
f36f3042 | 1999 | int ret = 0; |
4a9d8bde | 2000 | |
a4abeea4 | 2001 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
2002 | trans = info->running_transaction; |
2003 | if (trans) | |
2004 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 2005 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
2006 | return ret; |
2007 | } | |
2008 | ||
8929ecfa YZ |
2009 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
2010 | { | |
4a9d8bde | 2011 | struct btrfs_transaction *trans; |
8929ecfa | 2012 | int ret = 0; |
4a9d8bde | 2013 | |
a4abeea4 | 2014 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
2015 | trans = info->running_transaction; |
2016 | if (trans) | |
2017 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 2018 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
2019 | return ret; |
2020 | } | |
2021 | ||
fdfbf020 | 2022 | void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans) |
bb9c12c9 | 2023 | { |
3a45bb20 | 2024 | struct btrfs_fs_info *fs_info = trans->fs_info; |
bb9c12c9 SW |
2025 | struct btrfs_transaction *cur_trans; |
2026 | ||
fdfbf020 JB |
2027 | /* Kick the transaction kthread. */ |
2028 | set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); | |
2029 | wake_up_process(fs_info->transaction_kthread); | |
bb9c12c9 SW |
2030 | |
2031 | /* take transaction reference */ | |
bb9c12c9 | 2032 | cur_trans = trans->transaction; |
9b64f57d | 2033 | refcount_inc(&cur_trans->use_count); |
bb9c12c9 | 2034 | |
3a45bb20 | 2035 | btrfs_end_transaction(trans); |
6fc4e354 | 2036 | |
ae5d29d4 DS |
2037 | /* |
2038 | * Wait for the current transaction commit to start and block | |
2039 | * subsequent transaction joins | |
2040 | */ | |
77d20c68 | 2041 | btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP); |
ae5d29d4 DS |
2042 | wait_event(fs_info->transaction_blocked_wait, |
2043 | cur_trans->state >= TRANS_STATE_COMMIT_START || | |
2044 | TRANS_ABORTED(cur_trans)); | |
724e2315 | 2045 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
2046 | } |
2047 | ||
97cb39bb | 2048 | static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) |
49b25e05 | 2049 | { |
97cb39bb | 2050 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 JM |
2051 | struct btrfs_transaction *cur_trans = trans->transaction; |
2052 | ||
b50fff81 | 2053 | WARN_ON(refcount_read(&trans->use_count) > 1); |
49b25e05 | 2054 | |
66642832 | 2055 | btrfs_abort_transaction(trans, err); |
7b8b92af | 2056 | |
0b246afa | 2057 | spin_lock(&fs_info->trans_lock); |
66b6135b | 2058 | |
25d8c284 MX |
2059 | /* |
2060 | * If the transaction is removed from the list, it means this | |
2061 | * transaction has been committed successfully, so it is impossible | |
2062 | * to call the cleanup function. | |
2063 | */ | |
2064 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 2065 | |
0b246afa | 2066 | if (cur_trans == fs_info->running_transaction) { |
4a9d8bde | 2067 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 2068 | spin_unlock(&fs_info->trans_lock); |
e1489b4f IA |
2069 | |
2070 | /* | |
2071 | * The thread has already released the lockdep map as reader | |
2072 | * already in btrfs_commit_transaction(). | |
2073 | */ | |
2074 | btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); | |
f094ac32 LB |
2075 | wait_event(cur_trans->writer_wait, |
2076 | atomic_read(&cur_trans->num_writers) == 1); | |
2077 | ||
0b246afa | 2078 | spin_lock(&fs_info->trans_lock); |
d7096fc3 | 2079 | } |
061dde82 FM |
2080 | |
2081 | /* | |
2082 | * Now that we know no one else is still using the transaction we can | |
2083 | * remove the transaction from the list of transactions. This avoids | |
2084 | * the transaction kthread from cleaning up the transaction while some | |
2085 | * other task is still using it, which could result in a use-after-free | |
2086 | * on things like log trees, as it forces the transaction kthread to | |
2087 | * wait for this transaction to be cleaned up by us. | |
2088 | */ | |
2089 | list_del_init(&cur_trans->list); | |
2090 | ||
0b246afa | 2091 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 2092 | |
2ff7e61e | 2093 | btrfs_cleanup_one_transaction(trans->transaction, fs_info); |
49b25e05 | 2094 | |
0b246afa JM |
2095 | spin_lock(&fs_info->trans_lock); |
2096 | if (cur_trans == fs_info->running_transaction) | |
2097 | fs_info->running_transaction = NULL; | |
2098 | spin_unlock(&fs_info->trans_lock); | |
4a9d8bde | 2099 | |
e0228285 | 2100 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2101 | sb_end_intwrite(fs_info->sb); |
724e2315 JB |
2102 | btrfs_put_transaction(cur_trans); |
2103 | btrfs_put_transaction(cur_trans); | |
49b25e05 | 2104 | |
2e4e97ab | 2105 | trace_btrfs_transaction_commit(fs_info); |
49b25e05 | 2106 | |
49b25e05 JM |
2107 | if (current->journal_info == trans) |
2108 | current->journal_info = NULL; | |
2d82a40a FM |
2109 | |
2110 | /* | |
2111 | * If relocation is running, we can't cancel scrub because that will | |
2112 | * result in a deadlock. Before relocating a block group, relocation | |
2113 | * pauses scrub, then starts and commits a transaction before unpausing | |
2114 | * scrub. If the transaction commit is being done by the relocation | |
2115 | * task or triggered by another task and the relocation task is waiting | |
2116 | * for the commit, and we end up here due to an error in the commit | |
2117 | * path, then calling btrfs_scrub_cancel() will deadlock, as we are | |
2118 | * asking for scrub to stop while having it asked to be paused higher | |
2119 | * above in relocation code. | |
2120 | */ | |
2121 | if (!test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) | |
2122 | btrfs_scrub_cancel(fs_info); | |
49b25e05 JM |
2123 | |
2124 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
2125 | } | |
2126 | ||
c7cc64a9 DS |
2127 | /* |
2128 | * Release reserved delayed ref space of all pending block groups of the | |
2129 | * transaction and remove them from the list | |
2130 | */ | |
2131 | static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) | |
2132 | { | |
2133 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2134 | struct btrfs_block_group *block_group, *tmp; |
c7cc64a9 DS |
2135 | |
2136 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { | |
9ef17228 | 2137 | btrfs_dec_delayed_refs_rsv_bg_inserts(fs_info); |
c7cc64a9 DS |
2138 | list_del_init(&block_group->bg_list); |
2139 | } | |
2140 | } | |
2141 | ||
88090ad3 | 2142 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
82436617 | 2143 | { |
ce8ea7cc | 2144 | /* |
a0f0cf83 | 2145 | * We use try_to_writeback_inodes_sb() here because if we used |
ce8ea7cc JB |
2146 | * btrfs_start_delalloc_roots we would deadlock with fs freeze. |
2147 | * Currently are holding the fs freeze lock, if we do an async flush | |
2148 | * we'll do btrfs_join_transaction() and deadlock because we need to | |
2149 | * wait for the fs freeze lock. Using the direct flushing we benefit | |
2150 | * from already being in a transaction and our join_transaction doesn't | |
2151 | * have to re-take the fs freeze lock. | |
a0f0cf83 FM |
2152 | * |
2153 | * Note that try_to_writeback_inodes_sb() will only trigger writeback | |
2154 | * if it can read lock sb->s_umount. It will always be able to lock it, | |
2155 | * except when the filesystem is being unmounted or being frozen, but in | |
2156 | * those cases sync_filesystem() is called, which results in calling | |
2157 | * writeback_inodes_sb() while holding a write lock on sb->s_umount. | |
2158 | * Note that we don't call writeback_inodes_sb() directly, because it | |
2159 | * will emit a warning if sb->s_umount is not locked. | |
ce8ea7cc | 2160 | */ |
88090ad3 | 2161 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
a0f0cf83 | 2162 | try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); |
82436617 MX |
2163 | return 0; |
2164 | } | |
2165 | ||
88090ad3 | 2166 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) |
82436617 | 2167 | { |
88090ad3 | 2168 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
6374e57a | 2169 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); |
82436617 MX |
2170 | } |
2171 | ||
28b21c55 FM |
2172 | /* |
2173 | * Add a pending snapshot associated with the given transaction handle to the | |
2174 | * respective handle. This must be called after the transaction commit started | |
2175 | * and while holding fs_info->trans_lock. | |
2176 | * This serves to guarantee a caller of btrfs_commit_transaction() that it can | |
2177 | * safely free the pending snapshot pointer in case btrfs_commit_transaction() | |
2178 | * returns an error. | |
2179 | */ | |
2180 | static void add_pending_snapshot(struct btrfs_trans_handle *trans) | |
2181 | { | |
2182 | struct btrfs_transaction *cur_trans = trans->transaction; | |
2183 | ||
2184 | if (!trans->pending_snapshot) | |
2185 | return; | |
2186 | ||
2187 | lockdep_assert_held(&trans->fs_info->trans_lock); | |
77d20c68 | 2188 | ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_PREP); |
28b21c55 FM |
2189 | |
2190 | list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots); | |
2191 | } | |
2192 | ||
e55958c8 IA |
2193 | static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval) |
2194 | { | |
2195 | fs_info->commit_stats.commit_count++; | |
2196 | fs_info->commit_stats.last_commit_dur = interval; | |
2197 | fs_info->commit_stats.max_commit_dur = | |
2198 | max_t(u64, fs_info->commit_stats.max_commit_dur, interval); | |
2199 | fs_info->commit_stats.total_commit_dur += interval; | |
2200 | } | |
2201 | ||
3a45bb20 | 2202 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans) |
79154b1b | 2203 | { |
3a45bb20 | 2204 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 | 2205 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 2206 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 2207 | int ret; |
e55958c8 IA |
2208 | ktime_t start_time; |
2209 | ktime_t interval; | |
79154b1b | 2210 | |
35b814f3 | 2211 | ASSERT(refcount_read(&trans->use_count) == 1); |
77d20c68 | 2212 | btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP); |
35b814f3 | 2213 | |
c52cc7b7 JB |
2214 | clear_bit(BTRFS_FS_NEED_TRANS_COMMIT, &fs_info->flags); |
2215 | ||
8d25a086 | 2216 | /* Stop the commit early if ->aborted is set */ |
bf31f87f | 2217 | if (TRANS_ABORTED(cur_trans)) { |
25287e0a | 2218 | ret = cur_trans->aborted; |
3e738c53 | 2219 | goto lockdep_trans_commit_start_release; |
25287e0a | 2220 | } |
49b25e05 | 2221 | |
f45c752b JB |
2222 | btrfs_trans_release_metadata(trans); |
2223 | trans->block_rsv = NULL; | |
2224 | ||
56bec294 | 2225 | /* |
e19eb11f JB |
2226 | * We only want one transaction commit doing the flushing so we do not |
2227 | * waste a bunch of time on lock contention on the extent root node. | |
56bec294 | 2228 | */ |
e19eb11f JB |
2229 | if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, |
2230 | &cur_trans->delayed_refs.flags)) { | |
2231 | /* | |
2232 | * Make a pass through all the delayed refs we have so far. | |
2233 | * Any running threads may add more while we are here. | |
2234 | */ | |
2235 | ret = btrfs_run_delayed_refs(trans, 0); | |
3e738c53 IA |
2236 | if (ret) |
2237 | goto lockdep_trans_commit_start_release; | |
e19eb11f | 2238 | } |
56bec294 | 2239 | |
119e80df | 2240 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2241 | |
3204d33c | 2242 | if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { |
1bbc621e CM |
2243 | int run_it = 0; |
2244 | ||
2245 | /* this mutex is also taken before trying to set | |
2246 | * block groups readonly. We need to make sure | |
2247 | * that nobody has set a block group readonly | |
2248 | * after a extents from that block group have been | |
2249 | * allocated for cache files. btrfs_set_block_group_ro | |
2250 | * will wait for the transaction to commit if it | |
3204d33c | 2251 | * finds BTRFS_TRANS_DIRTY_BG_RUN set. |
1bbc621e | 2252 | * |
3204d33c JB |
2253 | * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure |
2254 | * only one process starts all the block group IO. It wouldn't | |
1bbc621e CM |
2255 | * hurt to have more than one go through, but there's no |
2256 | * real advantage to it either. | |
2257 | */ | |
0b246afa | 2258 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c JB |
2259 | if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, |
2260 | &cur_trans->flags)) | |
1bbc621e | 2261 | run_it = 1; |
0b246afa | 2262 | mutex_unlock(&fs_info->ro_block_group_mutex); |
1bbc621e | 2263 | |
f9cacae3 | 2264 | if (run_it) { |
21217054 | 2265 | ret = btrfs_start_dirty_block_groups(trans); |
3e738c53 IA |
2266 | if (ret) |
2267 | goto lockdep_trans_commit_start_release; | |
f9cacae3 | 2268 | } |
1bbc621e CM |
2269 | } |
2270 | ||
0b246afa | 2271 | spin_lock(&fs_info->trans_lock); |
77d20c68 | 2272 | if (cur_trans->state >= TRANS_STATE_COMMIT_PREP) { |
d0c2f4fa FM |
2273 | enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; |
2274 | ||
28b21c55 FM |
2275 | add_pending_snapshot(trans); |
2276 | ||
0b246afa | 2277 | spin_unlock(&fs_info->trans_lock); |
9b64f57d | 2278 | refcount_inc(&cur_trans->use_count); |
ccd467d6 | 2279 | |
d0c2f4fa FM |
2280 | if (trans->in_fsync) |
2281 | want_state = TRANS_STATE_SUPER_COMMITTED; | |
3e738c53 IA |
2282 | |
2283 | btrfs_trans_state_lockdep_release(fs_info, | |
77d20c68 | 2284 | BTRFS_LOCKDEP_TRANS_COMMIT_PREP); |
d0c2f4fa FM |
2285 | ret = btrfs_end_transaction(trans); |
2286 | wait_for_commit(cur_trans, want_state); | |
15ee9bc7 | 2287 | |
bf31f87f | 2288 | if (TRANS_ABORTED(cur_trans)) |
b4924a0f LB |
2289 | ret = cur_trans->aborted; |
2290 | ||
724e2315 | 2291 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 2292 | |
49b25e05 | 2293 | return ret; |
79154b1b | 2294 | } |
4313b399 | 2295 | |
77d20c68 | 2296 | cur_trans->state = TRANS_STATE_COMMIT_PREP; |
0b246afa | 2297 | wake_up(&fs_info->transaction_blocked_wait); |
77d20c68 | 2298 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP); |
bb9c12c9 | 2299 | |
0b246afa | 2300 | if (cur_trans->list.prev != &fs_info->trans_list) { |
d0c2f4fa FM |
2301 | enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; |
2302 | ||
2303 | if (trans->in_fsync) | |
2304 | want_state = TRANS_STATE_SUPER_COMMITTED; | |
2305 | ||
ccd467d6 CM |
2306 | prev_trans = list_entry(cur_trans->list.prev, |
2307 | struct btrfs_transaction, list); | |
d0c2f4fa | 2308 | if (prev_trans->state < want_state) { |
9b64f57d | 2309 | refcount_inc(&prev_trans->use_count); |
0b246afa | 2310 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2311 | |
d0c2f4fa FM |
2312 | wait_for_commit(prev_trans, want_state); |
2313 | ||
bf31f87f | 2314 | ret = READ_ONCE(prev_trans->aborted); |
ccd467d6 | 2315 | |
724e2315 | 2316 | btrfs_put_transaction(prev_trans); |
1f9b8c8f | 2317 | if (ret) |
e1489b4f | 2318 | goto lockdep_release; |
77d20c68 | 2319 | spin_lock(&fs_info->trans_lock); |
ccd467d6 | 2320 | } |
a4abeea4 | 2321 | } else { |
cb2d3dad FM |
2322 | /* |
2323 | * The previous transaction was aborted and was already removed | |
2324 | * from the list of transactions at fs_info->trans_list. So we | |
2325 | * abort to prevent writing a new superblock that reflects a | |
2326 | * corrupt state (pointing to trees with unwritten nodes/leafs). | |
2327 | */ | |
84961539 | 2328 | if (BTRFS_FS_ERROR(fs_info)) { |
77d20c68 | 2329 | spin_unlock(&fs_info->trans_lock); |
cb2d3dad | 2330 | ret = -EROFS; |
e1489b4f | 2331 | goto lockdep_release; |
cb2d3dad | 2332 | } |
ccd467d6 | 2333 | } |
15ee9bc7 | 2334 | |
77d20c68 JB |
2335 | cur_trans->state = TRANS_STATE_COMMIT_START; |
2336 | wake_up(&fs_info->transaction_blocked_wait); | |
2337 | spin_unlock(&fs_info->trans_lock); | |
2338 | ||
e55958c8 IA |
2339 | /* |
2340 | * Get the time spent on the work done by the commit thread and not | |
2341 | * the time spent waiting on a previous commit | |
2342 | */ | |
2343 | start_time = ktime_get_ns(); | |
2344 | ||
0860adfd MX |
2345 | extwriter_counter_dec(cur_trans, trans->type); |
2346 | ||
88090ad3 | 2347 | ret = btrfs_start_delalloc_flush(fs_info); |
82436617 | 2348 | if (ret) |
e1489b4f | 2349 | goto lockdep_release; |
82436617 | 2350 | |
e5c304e6 | 2351 | ret = btrfs_run_delayed_items(trans); |
581227d0 | 2352 | if (ret) |
e1489b4f | 2353 | goto lockdep_release; |
15ee9bc7 | 2354 | |
5a9ba670 IA |
2355 | /* |
2356 | * The thread has started/joined the transaction thus it holds the | |
2357 | * lockdep map as a reader. It has to release it before acquiring the | |
2358 | * lockdep map as a writer. | |
2359 | */ | |
2360 | btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); | |
2361 | btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters); | |
581227d0 MX |
2362 | wait_event(cur_trans->writer_wait, |
2363 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 2364 | |
581227d0 | 2365 | /* some pending stuffs might be added after the previous flush. */ |
e5c304e6 | 2366 | ret = btrfs_run_delayed_items(trans); |
e1489b4f IA |
2367 | if (ret) { |
2368 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); | |
ca469637 | 2369 | goto cleanup_transaction; |
e1489b4f | 2370 | } |
ca469637 | 2371 | |
88090ad3 | 2372 | btrfs_wait_delalloc_flush(fs_info); |
cb7ab021 | 2373 | |
48778179 FM |
2374 | /* |
2375 | * Wait for all ordered extents started by a fast fsync that joined this | |
2376 | * transaction. Otherwise if this transaction commits before the ordered | |
2377 | * extents complete we lose logged data after a power failure. | |
2378 | */ | |
8b53779e | 2379 | btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered); |
48778179 FM |
2380 | wait_event(cur_trans->pending_wait, |
2381 | atomic_read(&cur_trans->pending_ordered) == 0); | |
2382 | ||
2ff7e61e | 2383 | btrfs_scrub_pause(fs_info); |
ed0ca140 JB |
2384 | /* |
2385 | * Ok now we need to make sure to block out any other joins while we | |
2386 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 2387 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 | 2388 | */ |
0b246afa | 2389 | spin_lock(&fs_info->trans_lock); |
28b21c55 | 2390 | add_pending_snapshot(trans); |
4a9d8bde | 2391 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 2392 | spin_unlock(&fs_info->trans_lock); |
e1489b4f IA |
2393 | |
2394 | /* | |
2395 | * The thread has started/joined the transaction thus it holds the | |
2396 | * lockdep map as a reader. It has to release it before acquiring the | |
2397 | * lockdep map as a writer. | |
2398 | */ | |
2399 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); | |
2400 | btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); | |
ed0ca140 JB |
2401 | wait_event(cur_trans->writer_wait, |
2402 | atomic_read(&cur_trans->num_writers) == 1); | |
2403 | ||
3e738c53 IA |
2404 | /* |
2405 | * Make lockdep happy by acquiring the state locks after | |
2406 | * btrfs_trans_num_writers is released. If we acquired the state locks | |
2407 | * before releasing the btrfs_trans_num_writers lock then lockdep would | |
2408 | * complain because we did not follow the reverse order unlocking rule. | |
2409 | */ | |
2410 | btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); | |
2411 | btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); | |
2412 | btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); | |
2413 | ||
fdfbf020 JB |
2414 | /* |
2415 | * We've started the commit, clear the flag in case we were triggered to | |
2416 | * do an async commit but somebody else started before the transaction | |
2417 | * kthread could do the work. | |
2418 | */ | |
2419 | clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); | |
2420 | ||
bf31f87f | 2421 | if (TRANS_ABORTED(cur_trans)) { |
2cba30f1 | 2422 | ret = cur_trans->aborted; |
3e738c53 | 2423 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); |
6cf7f77e | 2424 | goto scrub_continue; |
2cba30f1 | 2425 | } |
7585717f CM |
2426 | /* |
2427 | * the reloc mutex makes sure that we stop | |
2428 | * the balancing code from coming in and moving | |
2429 | * extents around in the middle of the commit | |
2430 | */ | |
0b246afa | 2431 | mutex_lock(&fs_info->reloc_mutex); |
7585717f | 2432 | |
42874b3d MX |
2433 | /* |
2434 | * We needn't worry about the delayed items because we will | |
2435 | * deal with them in create_pending_snapshot(), which is the | |
2436 | * core function of the snapshot creation. | |
2437 | */ | |
08d50ca3 | 2438 | ret = create_pending_snapshots(trans); |
56e9f6ea DS |
2439 | if (ret) |
2440 | goto unlock_reloc; | |
3063d29f | 2441 | |
42874b3d MX |
2442 | /* |
2443 | * We insert the dir indexes of the snapshots and update the inode | |
2444 | * of the snapshots' parents after the snapshot creation, so there | |
2445 | * are some delayed items which are not dealt with. Now deal with | |
2446 | * them. | |
2447 | * | |
2448 | * We needn't worry that this operation will corrupt the snapshots, | |
2449 | * because all the tree which are snapshoted will be forced to COW | |
2450 | * the nodes and leaves. | |
2451 | */ | |
e5c304e6 | 2452 | ret = btrfs_run_delayed_items(trans); |
56e9f6ea DS |
2453 | if (ret) |
2454 | goto unlock_reloc; | |
16cdcec7 | 2455 | |
8a526c44 | 2456 | ret = btrfs_run_delayed_refs(trans, U64_MAX); |
56e9f6ea DS |
2457 | if (ret) |
2458 | goto unlock_reloc; | |
56bec294 | 2459 | |
e999376f CM |
2460 | /* |
2461 | * make sure none of the code above managed to slip in a | |
2462 | * delayed item | |
2463 | */ | |
ccdf9b30 | 2464 | btrfs_assert_delayed_root_empty(fs_info); |
e999376f | 2465 | |
2c90e5d6 | 2466 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 2467 | |
7e4443d9 | 2468 | ret = commit_fs_roots(trans); |
56e9f6ea | 2469 | if (ret) |
dfba78dc | 2470 | goto unlock_reloc; |
54aa1f4d | 2471 | |
5d4f98a2 | 2472 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2473 | * safe to free the root of tree log roots |
2474 | */ | |
0b246afa | 2475 | btrfs_free_log_root_tree(trans, fs_info); |
e02119d5 | 2476 | |
0ed4792a QW |
2477 | /* |
2478 | * Since fs roots are all committed, we can get a quite accurate | |
2479 | * new_roots. So let's do quota accounting. | |
2480 | */ | |
460fb20a | 2481 | ret = btrfs_qgroup_account_extents(trans); |
56e9f6ea | 2482 | if (ret < 0) |
dfba78dc | 2483 | goto unlock_reloc; |
0ed4792a | 2484 | |
9386d8bc | 2485 | ret = commit_cowonly_roots(trans); |
56e9f6ea | 2486 | if (ret) |
dfba78dc | 2487 | goto unlock_reloc; |
54aa1f4d | 2488 | |
2cba30f1 MX |
2489 | /* |
2490 | * The tasks which save the space cache and inode cache may also | |
2491 | * update ->aborted, check it. | |
2492 | */ | |
bf31f87f | 2493 | if (TRANS_ABORTED(cur_trans)) { |
2cba30f1 | 2494 | ret = cur_trans->aborted; |
dfba78dc | 2495 | goto unlock_reloc; |
2cba30f1 MX |
2496 | } |
2497 | ||
0b246afa | 2498 | cur_trans = fs_info->running_transaction; |
5d4f98a2 | 2499 | |
0b246afa JM |
2500 | btrfs_set_root_node(&fs_info->tree_root->root_item, |
2501 | fs_info->tree_root->node); | |
2502 | list_add_tail(&fs_info->tree_root->dirty_list, | |
9e351cc8 | 2503 | &cur_trans->switch_commits); |
5d4f98a2 | 2504 | |
0b246afa JM |
2505 | btrfs_set_root_node(&fs_info->chunk_root->root_item, |
2506 | fs_info->chunk_root->node); | |
2507 | list_add_tail(&fs_info->chunk_root->dirty_list, | |
9e351cc8 JB |
2508 | &cur_trans->switch_commits); |
2509 | ||
f7238e50 JB |
2510 | if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { |
2511 | btrfs_set_root_node(&fs_info->block_group_root->root_item, | |
2512 | fs_info->block_group_root->node); | |
2513 | list_add_tail(&fs_info->block_group_root->dirty_list, | |
2514 | &cur_trans->switch_commits); | |
2515 | } | |
2516 | ||
889bfa39 | 2517 | switch_commit_roots(trans); |
5d4f98a2 | 2518 | |
ce93ec54 | 2519 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2520 | ASSERT(list_empty(&cur_trans->io_bgs)); |
2ff7e61e | 2521 | update_super_roots(fs_info); |
e02119d5 | 2522 | |
0b246afa JM |
2523 | btrfs_set_super_log_root(fs_info->super_copy, 0); |
2524 | btrfs_set_super_log_root_level(fs_info->super_copy, 0); | |
2525 | memcpy(fs_info->super_for_commit, fs_info->super_copy, | |
2526 | sizeof(*fs_info->super_copy)); | |
ccd467d6 | 2527 | |
bbbf7243 | 2528 | btrfs_commit_device_sizes(cur_trans); |
935e5cc9 | 2529 | |
0b246afa JM |
2530 | clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
2531 | clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); | |
656f30db | 2532 | |
4fbcdf66 FM |
2533 | btrfs_trans_release_chunk_metadata(trans); |
2534 | ||
dfba78dc FM |
2535 | /* |
2536 | * Before changing the transaction state to TRANS_STATE_UNBLOCKED and | |
2537 | * setting fs_info->running_transaction to NULL, lock tree_log_mutex to | |
2538 | * make sure that before we commit our superblock, no other task can | |
2539 | * start a new transaction and commit a log tree before we commit our | |
2540 | * superblock. Anyone trying to commit a log tree locks this mutex before | |
2541 | * writing its superblock. | |
2542 | */ | |
2543 | mutex_lock(&fs_info->tree_log_mutex); | |
2544 | ||
0b246afa | 2545 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2546 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
0b246afa JM |
2547 | fs_info->running_transaction = NULL; |
2548 | spin_unlock(&fs_info->trans_lock); | |
2549 | mutex_unlock(&fs_info->reloc_mutex); | |
b7ec40d7 | 2550 | |
0b246afa | 2551 | wake_up(&fs_info->transaction_wait); |
3e738c53 | 2552 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); |
e6dcd2dc | 2553 | |
b7625f46 QW |
2554 | /* If we have features changed, wake up the cleaner to update sysfs. */ |
2555 | if (test_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags) && | |
2556 | fs_info->cleaner_kthread) | |
2557 | wake_up_process(fs_info->cleaner_kthread); | |
2558 | ||
70458a58 | 2559 | ret = btrfs_write_and_wait_transaction(trans); |
49b25e05 | 2560 | if (ret) { |
0b246afa JM |
2561 | btrfs_handle_fs_error(fs_info, ret, |
2562 | "Error while writing out transaction"); | |
2563 | mutex_unlock(&fs_info->tree_log_mutex); | |
6cf7f77e | 2564 | goto scrub_continue; |
49b25e05 JM |
2565 | } |
2566 | ||
eece6a9c | 2567 | ret = write_all_supers(fs_info, 0); |
e02119d5 CM |
2568 | /* |
2569 | * the super is written, we can safely allow the tree-loggers | |
2570 | * to go about their business | |
2571 | */ | |
0b246afa | 2572 | mutex_unlock(&fs_info->tree_log_mutex); |
c1f32b7c AJ |
2573 | if (ret) |
2574 | goto scrub_continue; | |
e02119d5 | 2575 | |
d0c2f4fa FM |
2576 | /* |
2577 | * We needn't acquire the lock here because there is no other task | |
2578 | * which can change it. | |
2579 | */ | |
2580 | cur_trans->state = TRANS_STATE_SUPER_COMMITTED; | |
2581 | wake_up(&cur_trans->commit_wait); | |
3e738c53 | 2582 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); |
d0c2f4fa | 2583 | |
5ead2dd0 | 2584 | btrfs_finish_extent_commit(trans); |
4313b399 | 2585 | |
3204d33c | 2586 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) |
0b246afa | 2587 | btrfs_clear_space_info_full(fs_info); |
13212b54 | 2588 | |
0124855f | 2589 | btrfs_set_last_trans_committed(fs_info, cur_trans->transid); |
4a9d8bde MX |
2590 | /* |
2591 | * We needn't acquire the lock here because there is no other task | |
2592 | * which can change it. | |
2593 | */ | |
2594 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2595 | wake_up(&cur_trans->commit_wait); |
3e738c53 | 2596 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); |
3de4586c | 2597 | |
0b246afa | 2598 | spin_lock(&fs_info->trans_lock); |
13c5a93e | 2599 | list_del_init(&cur_trans->list); |
0b246afa | 2600 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 2601 | |
724e2315 JB |
2602 | btrfs_put_transaction(cur_trans); |
2603 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2604 | |
0860adfd | 2605 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2606 | sb_end_intwrite(fs_info->sb); |
b2b5ef5c | 2607 | |
2e4e97ab | 2608 | trace_btrfs_transaction_commit(fs_info); |
1abe9b8a | 2609 | |
e55958c8 IA |
2610 | interval = ktime_get_ns() - start_time; |
2611 | ||
2ff7e61e | 2612 | btrfs_scrub_continue(fs_info); |
a2de733c | 2613 | |
9ed74f2d JB |
2614 | if (current->journal_info == trans) |
2615 | current->journal_info = NULL; | |
2616 | ||
2c90e5d6 | 2617 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2618 | |
e55958c8 IA |
2619 | update_commit_stats(fs_info, interval); |
2620 | ||
79154b1b | 2621 | return ret; |
49b25e05 | 2622 | |
56e9f6ea DS |
2623 | unlock_reloc: |
2624 | mutex_unlock(&fs_info->reloc_mutex); | |
3e738c53 | 2625 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); |
6cf7f77e | 2626 | scrub_continue: |
3e738c53 IA |
2627 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); |
2628 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); | |
2ff7e61e | 2629 | btrfs_scrub_continue(fs_info); |
49b25e05 | 2630 | cleanup_transaction: |
dc60c525 | 2631 | btrfs_trans_release_metadata(trans); |
c7cc64a9 | 2632 | btrfs_cleanup_pending_block_groups(trans); |
4fbcdf66 | 2633 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2634 | trans->block_rsv = NULL; |
0b246afa | 2635 | btrfs_warn(fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2636 | if (current->journal_info == trans) |
2637 | current->journal_info = NULL; | |
97cb39bb | 2638 | cleanup_transaction(trans, ret); |
49b25e05 JM |
2639 | |
2640 | return ret; | |
e1489b4f IA |
2641 | |
2642 | lockdep_release: | |
5a9ba670 | 2643 | btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); |
e1489b4f IA |
2644 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); |
2645 | goto cleanup_transaction; | |
3e738c53 IA |
2646 | |
2647 | lockdep_trans_commit_start_release: | |
77d20c68 | 2648 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP); |
3e738c53 IA |
2649 | btrfs_end_transaction(trans); |
2650 | return ret; | |
79154b1b CM |
2651 | } |
2652 | ||
d352ac68 | 2653 | /* |
9d1a2a3a DS |
2654 | * return < 0 if error |
2655 | * 0 if there are no more dead_roots at the time of call | |
2656 | * 1 there are more to be processed, call me again | |
2657 | * | |
2658 | * The return value indicates there are certainly more snapshots to delete, but | |
2659 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2660 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2661 | * few seconds later. | |
d352ac68 | 2662 | */ |
33c44184 | 2663 | int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info) |
e9d0b13b | 2664 | { |
33c44184 | 2665 | struct btrfs_root *root; |
9d1a2a3a | 2666 | int ret; |
5d4f98a2 | 2667 | |
a4abeea4 | 2668 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2669 | if (list_empty(&fs_info->dead_roots)) { |
2670 | spin_unlock(&fs_info->trans_lock); | |
2671 | return 0; | |
2672 | } | |
2673 | root = list_first_entry(&fs_info->dead_roots, | |
2674 | struct btrfs_root, root_list); | |
cfad392b | 2675 | list_del_init(&root->root_list); |
a4abeea4 | 2676 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2677 | |
4fd786e6 | 2678 | btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); |
76dda93c | 2679 | |
9d1a2a3a | 2680 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2681 | |
9d1a2a3a DS |
2682 | if (btrfs_header_backref_rev(root->node) < |
2683 | BTRFS_MIXED_BACKREF_REV) | |
0078a9f9 | 2684 | ret = btrfs_drop_snapshot(root, 0, 0); |
9d1a2a3a | 2685 | else |
0078a9f9 | 2686 | ret = btrfs_drop_snapshot(root, 1, 0); |
32471dc2 | 2687 | |
dc9492c1 | 2688 | btrfs_put_root(root); |
6596a928 | 2689 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2690 | } |
572d9ab7 | 2691 | |
fccf0c84 JB |
2692 | /* |
2693 | * We only mark the transaction aborted and then set the file system read-only. | |
2694 | * This will prevent new transactions from starting or trying to join this | |
2695 | * one. | |
2696 | * | |
2697 | * This means that error recovery at the call site is limited to freeing | |
2698 | * any local memory allocations and passing the error code up without | |
2699 | * further cleanup. The transaction should complete as it normally would | |
2700 | * in the call path but will return -EIO. | |
2701 | * | |
2702 | * We'll complete the cleanup in btrfs_end_transaction and | |
2703 | * btrfs_commit_transaction. | |
2704 | */ | |
2705 | void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans, | |
2706 | const char *function, | |
ed164802 | 2707 | unsigned int line, int error, bool first_hit) |
fccf0c84 JB |
2708 | { |
2709 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2710 | ||
ed164802 DS |
2711 | WRITE_ONCE(trans->aborted, error); |
2712 | WRITE_ONCE(trans->transaction->aborted, error); | |
2713 | if (first_hit && error == -ENOSPC) | |
fccf0c84 JB |
2714 | btrfs_dump_space_info_for_trans_abort(fs_info); |
2715 | /* Wake up anybody who may be waiting on this transaction */ | |
2716 | wake_up(&fs_info->transaction_wait); | |
2717 | wake_up(&fs_info->transaction_blocked_wait); | |
ed164802 | 2718 | __btrfs_handle_fs_error(fs_info, function, line, error, NULL); |
fccf0c84 JB |
2719 | } |
2720 | ||
956504a3 JB |
2721 | int __init btrfs_transaction_init(void) |
2722 | { | |
2723 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", | |
2724 | sizeof(struct btrfs_trans_handle), 0, | |
2725 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); | |
2726 | if (!btrfs_trans_handle_cachep) | |
2727 | return -ENOMEM; | |
2728 | return 0; | |
2729 | } | |
2730 | ||
2731 | void __cold btrfs_transaction_exit(void) | |
2732 | { | |
2733 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
2734 | } |