]>
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 | ||
fc7cbcd4 | 40 | #define BTRFS_ROOT_TRANS_TAG 0 |
0f7d52f4 | 41 | |
61c047b5 QW |
42 | /* |
43 | * Transaction states and transitions | |
44 | * | |
45 | * No running transaction (fs tree blocks are not modified) | |
46 | * | | |
47 | * | To next stage: | |
48 | * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). | |
49 | * V | |
50 | * Transaction N [[TRANS_STATE_RUNNING]] | |
51 | * | | |
52 | * | New trans handles can be attached to transaction N by calling all | |
53 | * | start_transaction() variants. | |
54 | * | | |
55 | * | To next stage: | |
56 | * | Call btrfs_commit_transaction() on any trans handle attached to | |
57 | * | transaction N | |
58 | * V | |
77d20c68 JB |
59 | * Transaction N [[TRANS_STATE_COMMIT_PREP]] |
60 | * | | |
61 | * | If there are simultaneous calls to btrfs_commit_transaction() one will win | |
62 | * | the race and the rest will wait for the winner to commit the transaction. | |
63 | * | | |
64 | * | The winner will wait for previous running transaction to completely finish | |
65 | * | if there is one. | |
61c047b5 | 66 | * | |
77d20c68 | 67 | * Transaction N [[TRANS_STATE_COMMIT_START]] |
61c047b5 | 68 | * | |
77d20c68 | 69 | * | Then one of the following happens: |
61c047b5 QW |
70 | * | - Wait for all other trans handle holders to release. |
71 | * | The btrfs_commit_transaction() caller will do the commit work. | |
72 | * | - Wait for current transaction to be committed by others. | |
73 | * | Other btrfs_commit_transaction() caller will do the commit work. | |
74 | * | | |
75 | * | At this stage, only btrfs_join_transaction*() variants can attach | |
76 | * | to this running transaction. | |
77 | * | All other variants will wait for current one to finish and attach to | |
78 | * | transaction N+1. | |
79 | * | | |
80 | * | To next stage: | |
81 | * | Caller is chosen to commit transaction N, and all other trans handle | |
82 | * | haven been released. | |
83 | * V | |
84 | * Transaction N [[TRANS_STATE_COMMIT_DOING]] | |
85 | * | | |
86 | * | The heavy lifting transaction work is started. | |
87 | * | From running delayed refs (modifying extent tree) to creating pending | |
88 | * | snapshots, running qgroups. | |
89 | * | In short, modify supporting trees to reflect modifications of subvolume | |
90 | * | trees. | |
91 | * | | |
92 | * | At this stage, all start_transaction() calls will wait for this | |
93 | * | transaction to finish and attach to transaction N+1. | |
94 | * | | |
95 | * | To next stage: | |
96 | * | Until all supporting trees are updated. | |
97 | * V | |
98 | * Transaction N [[TRANS_STATE_UNBLOCKED]] | |
99 | * | Transaction N+1 | |
100 | * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] | |
101 | * | need to write them back to disk and update | | |
102 | * | super blocks. | | |
103 | * | | | |
104 | * | At this stage, new transaction is allowed to | | |
105 | * | start. | | |
106 | * | All new start_transaction() calls will be | | |
107 | * | attached to transid N+1. | | |
108 | * | | | |
109 | * | To next stage: | | |
110 | * | Until all tree blocks are super blocks are | | |
111 | * | written to block devices | | |
112 | * V | | |
113 | * Transaction N [[TRANS_STATE_COMPLETED]] V | |
114 | * All tree blocks and super blocks are written. Transaction N+1 | |
115 | * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] | |
116 | * data structures will be cleaned up. | Life goes on | |
117 | */ | |
e8c9f186 | 118 | static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { |
4a9d8bde | 119 | [TRANS_STATE_RUNNING] = 0U, |
77d20c68 | 120 | [TRANS_STATE_COMMIT_PREP] = 0U, |
bcf3a3e7 NB |
121 | [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), |
122 | [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | | |
4a9d8bde | 123 | __TRANS_ATTACH | |
a6d155d2 FM |
124 | __TRANS_JOIN | |
125 | __TRANS_JOIN_NOSTART), | |
bcf3a3e7 | 126 | [TRANS_STATE_UNBLOCKED] = (__TRANS_START | |
4a9d8bde MX |
127 | __TRANS_ATTACH | |
128 | __TRANS_JOIN | | |
a6d155d2 FM |
129 | __TRANS_JOIN_NOLOCK | |
130 | __TRANS_JOIN_NOSTART), | |
d0c2f4fa FM |
131 | [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START | |
132 | __TRANS_ATTACH | | |
133 | __TRANS_JOIN | | |
134 | __TRANS_JOIN_NOLOCK | | |
135 | __TRANS_JOIN_NOSTART), | |
bcf3a3e7 | 136 | [TRANS_STATE_COMPLETED] = (__TRANS_START | |
4a9d8bde MX |
137 | __TRANS_ATTACH | |
138 | __TRANS_JOIN | | |
a6d155d2 FM |
139 | __TRANS_JOIN_NOLOCK | |
140 | __TRANS_JOIN_NOSTART), | |
4a9d8bde MX |
141 | }; |
142 | ||
724e2315 | 143 | void btrfs_put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 144 | { |
9b64f57d ER |
145 | WARN_ON(refcount_read(&transaction->use_count) == 0); |
146 | if (refcount_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 147 | BUG_ON(!list_empty(&transaction->list)); |
5c9d028b LB |
148 | WARN_ON(!RB_EMPTY_ROOT( |
149 | &transaction->delayed_refs.href_root.rb_root)); | |
81f7eb00 JM |
150 | WARN_ON(!RB_EMPTY_ROOT( |
151 | &transaction->delayed_refs.dirty_extent_root)); | |
1262133b | 152 | if (transaction->delayed_refs.pending_csums) |
ab8d0fc4 JM |
153 | btrfs_err(transaction->fs_info, |
154 | "pending csums is %llu", | |
155 | transaction->delayed_refs.pending_csums); | |
7785a663 FM |
156 | /* |
157 | * If any block groups are found in ->deleted_bgs then it's | |
158 | * because the transaction was aborted and a commit did not | |
159 | * happen (things failed before writing the new superblock | |
160 | * and calling btrfs_finish_extent_commit()), so we can not | |
161 | * discard the physical locations of the block groups. | |
162 | */ | |
163 | while (!list_empty(&transaction->deleted_bgs)) { | |
32da5386 | 164 | struct btrfs_block_group *cache; |
7785a663 FM |
165 | |
166 | cache = list_first_entry(&transaction->deleted_bgs, | |
32da5386 | 167 | struct btrfs_block_group, |
7785a663 FM |
168 | bg_list); |
169 | list_del_init(&cache->bg_list); | |
6b7304af | 170 | btrfs_unfreeze_block_group(cache); |
7785a663 FM |
171 | btrfs_put_block_group(cache); |
172 | } | |
bbbf7243 | 173 | WARN_ON(!list_empty(&transaction->dev_update_list)); |
4b5faeac | 174 | kfree(transaction); |
78fae27e | 175 | } |
79154b1b CM |
176 | } |
177 | ||
889bfa39 | 178 | static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) |
817d52f8 | 179 | { |
889bfa39 | 180 | struct btrfs_transaction *cur_trans = trans->transaction; |
16916a88 | 181 | struct btrfs_fs_info *fs_info = trans->fs_info; |
9e351cc8 JB |
182 | struct btrfs_root *root, *tmp; |
183 | ||
dfba78dc FM |
184 | /* |
185 | * At this point no one can be using this transaction to modify any tree | |
186 | * and no one can start another transaction to modify any tree either. | |
187 | */ | |
188 | ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING); | |
189 | ||
9e351cc8 | 190 | down_write(&fs_info->commit_root_sem); |
d96b3424 FM |
191 | |
192 | if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) | |
193 | fs_info->last_reloc_trans = trans->transid; | |
194 | ||
889bfa39 | 195 | list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, |
9e351cc8 JB |
196 | dirty_list) { |
197 | list_del_init(&root->dirty_list); | |
198 | free_extent_buffer(root->commit_root); | |
199 | root->commit_root = btrfs_root_node(root); | |
41e7acd3 | 200 | extent_io_tree_release(&root->dirty_log_pages); |
370a11b8 | 201 | btrfs_qgroup_clean_swapped_blocks(root); |
9e351cc8 | 202 | } |
2b9dbef2 JB |
203 | |
204 | /* We can free old roots now. */ | |
889bfa39 JB |
205 | spin_lock(&cur_trans->dropped_roots_lock); |
206 | while (!list_empty(&cur_trans->dropped_roots)) { | |
207 | root = list_first_entry(&cur_trans->dropped_roots, | |
2b9dbef2 JB |
208 | struct btrfs_root, root_list); |
209 | list_del_init(&root->root_list); | |
889bfa39 JB |
210 | spin_unlock(&cur_trans->dropped_roots_lock); |
211 | btrfs_free_log(trans, root); | |
2b9dbef2 | 212 | btrfs_drop_and_free_fs_root(fs_info, root); |
889bfa39 | 213 | spin_lock(&cur_trans->dropped_roots_lock); |
2b9dbef2 | 214 | } |
889bfa39 | 215 | spin_unlock(&cur_trans->dropped_roots_lock); |
27d56e62 | 216 | |
9e351cc8 | 217 | up_write(&fs_info->commit_root_sem); |
817d52f8 JB |
218 | } |
219 | ||
0860adfd MX |
220 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
221 | unsigned int type) | |
222 | { | |
223 | if (type & TRANS_EXTWRITERS) | |
224 | atomic_inc(&trans->num_extwriters); | |
225 | } | |
226 | ||
227 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
228 | unsigned int type) | |
229 | { | |
230 | if (type & TRANS_EXTWRITERS) | |
231 | atomic_dec(&trans->num_extwriters); | |
232 | } | |
233 | ||
234 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
235 | unsigned int type) | |
236 | { | |
237 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
238 | } | |
239 | ||
240 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
241 | { | |
242 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
243 | } |
244 | ||
fb6dea26 | 245 | /* |
79bd3712 FM |
246 | * To be called after doing the chunk btree updates right after allocating a new |
247 | * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a | |
248 | * chunk after all chunk btree updates and after finishing the second phase of | |
249 | * chunk allocation (btrfs_create_pending_block_groups()) in case some block | |
250 | * group had its chunk item insertion delayed to the second phase. | |
fb6dea26 JB |
251 | */ |
252 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
253 | { | |
254 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
255 | ||
256 | if (!trans->chunk_bytes_reserved) | |
257 | return; | |
258 | ||
fb6dea26 | 259 | btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, |
63f018be | 260 | trans->chunk_bytes_reserved, NULL); |
fb6dea26 JB |
261 | trans->chunk_bytes_reserved = 0; |
262 | } | |
263 | ||
d352ac68 CM |
264 | /* |
265 | * either allocate a new transaction or hop into the existing one | |
266 | */ | |
2ff7e61e JM |
267 | static noinline int join_transaction(struct btrfs_fs_info *fs_info, |
268 | unsigned int type) | |
79154b1b CM |
269 | { |
270 | struct btrfs_transaction *cur_trans; | |
a4abeea4 | 271 | |
19ae4e81 | 272 | spin_lock(&fs_info->trans_lock); |
d43317dc | 273 | loop: |
49b25e05 | 274 | /* The file system has been taken offline. No new transactions. */ |
84961539 | 275 | if (BTRFS_FS_ERROR(fs_info)) { |
19ae4e81 | 276 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
277 | return -EROFS; |
278 | } | |
279 | ||
19ae4e81 | 280 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 281 | if (cur_trans) { |
bf31f87f | 282 | if (TRANS_ABORTED(cur_trans)) { |
19ae4e81 | 283 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 284 | return cur_trans->aborted; |
871383be | 285 | } |
4a9d8bde | 286 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
287 | spin_unlock(&fs_info->trans_lock); |
288 | return -EBUSY; | |
289 | } | |
9b64f57d | 290 | refcount_inc(&cur_trans->use_count); |
13c5a93e | 291 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 292 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 293 | spin_unlock(&fs_info->trans_lock); |
e1489b4f | 294 | btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); |
5a9ba670 | 295 | btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); |
a4abeea4 | 296 | return 0; |
79154b1b | 297 | } |
19ae4e81 | 298 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 299 | |
354aa0fb | 300 | /* |
4490e803 FM |
301 | * If we are ATTACH or TRANS_JOIN_NOSTART, we just want to catch the |
302 | * current transaction, and commit it. If there is no transaction, just | |
303 | * return ENOENT. | |
354aa0fb | 304 | */ |
4490e803 | 305 | if (type == TRANS_ATTACH || type == TRANS_JOIN_NOSTART) |
354aa0fb MX |
306 | return -ENOENT; |
307 | ||
4a9d8bde MX |
308 | /* |
309 | * JOIN_NOLOCK only happens during the transaction commit, so | |
310 | * it is impossible that ->running_transaction is NULL | |
311 | */ | |
312 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
313 | ||
4b5faeac | 314 | cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); |
a4abeea4 JB |
315 | if (!cur_trans) |
316 | return -ENOMEM; | |
d43317dc | 317 | |
e1489b4f | 318 | btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); |
5a9ba670 | 319 | btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); |
e1489b4f | 320 | |
19ae4e81 JS |
321 | spin_lock(&fs_info->trans_lock); |
322 | if (fs_info->running_transaction) { | |
d43317dc CM |
323 | /* |
324 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 325 | * to redo the checks above |
d43317dc | 326 | */ |
5a9ba670 | 327 | btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); |
e1489b4f | 328 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); |
4b5faeac | 329 | kfree(cur_trans); |
d43317dc | 330 | goto loop; |
84961539 | 331 | } else if (BTRFS_FS_ERROR(fs_info)) { |
e4b50e14 | 332 | spin_unlock(&fs_info->trans_lock); |
5a9ba670 | 333 | btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); |
e1489b4f | 334 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); |
4b5faeac | 335 | kfree(cur_trans); |
7b8b92af | 336 | return -EROFS; |
79154b1b | 337 | } |
d43317dc | 338 | |
ab8d0fc4 | 339 | cur_trans->fs_info = fs_info; |
48778179 FM |
340 | atomic_set(&cur_trans->pending_ordered, 0); |
341 | init_waitqueue_head(&cur_trans->pending_wait); | |
a4abeea4 | 342 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 343 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
344 | init_waitqueue_head(&cur_trans->writer_wait); |
345 | init_waitqueue_head(&cur_trans->commit_wait); | |
4a9d8bde | 346 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
347 | /* |
348 | * One for this trans handle, one so it will live on until we | |
349 | * commit the transaction. | |
350 | */ | |
9b64f57d | 351 | refcount_set(&cur_trans->use_count, 2); |
3204d33c | 352 | cur_trans->flags = 0; |
afd48513 | 353 | cur_trans->start_time = ktime_get_seconds(); |
a4abeea4 | 354 | |
a099d0fd AM |
355 | memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); |
356 | ||
5c9d028b | 357 | cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; |
3368d001 | 358 | cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; |
d7df2c79 | 359 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
20b297d6 JS |
360 | |
361 | /* | |
362 | * although the tree mod log is per file system and not per transaction, | |
363 | * the log must never go across transaction boundaries. | |
364 | */ | |
365 | smp_mb(); | |
31b1a2bd | 366 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
5d163e0e | 367 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); |
31b1a2bd | 368 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
5d163e0e | 369 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); |
fc36ed7e | 370 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 371 | |
a4abeea4 JB |
372 | spin_lock_init(&cur_trans->delayed_refs.lock); |
373 | ||
374 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
bbbf7243 | 375 | INIT_LIST_HEAD(&cur_trans->dev_update_list); |
9e351cc8 | 376 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
ce93ec54 | 377 | INIT_LIST_HEAD(&cur_trans->dirty_bgs); |
1bbc621e | 378 | INIT_LIST_HEAD(&cur_trans->io_bgs); |
2b9dbef2 | 379 | INIT_LIST_HEAD(&cur_trans->dropped_roots); |
1bbc621e | 380 | mutex_init(&cur_trans->cache_write_mutex); |
ce93ec54 | 381 | spin_lock_init(&cur_trans->dirty_bgs_lock); |
e33e17ee | 382 | INIT_LIST_HEAD(&cur_trans->deleted_bgs); |
2b9dbef2 | 383 | spin_lock_init(&cur_trans->dropped_roots_lock); |
19ae4e81 | 384 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
c258d6e3 | 385 | extent_io_tree_init(fs_info, &cur_trans->dirty_pages, |
35da5a7e | 386 | IO_TREE_TRANS_DIRTY_PAGES); |
fe119a6e | 387 | extent_io_tree_init(fs_info, &cur_trans->pinned_extents, |
35da5a7e | 388 | IO_TREE_FS_PINNED_EXTENTS); |
4a4f8fe2 | 389 | btrfs_set_fs_generation(fs_info, fs_info->generation + 1); |
19ae4e81 JS |
390 | cur_trans->transid = fs_info->generation; |
391 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 392 | cur_trans->aborted = 0; |
19ae4e81 | 393 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 394 | |
79154b1b CM |
395 | return 0; |
396 | } | |
397 | ||
d352ac68 | 398 | /* |
92a7cc42 QW |
399 | * This does all the record keeping required to make sure that a shareable root |
400 | * is properly recorded in a given transaction. This is required to make sure | |
401 | * the old root from before we joined the transaction is deleted when the | |
402 | * transaction commits. | |
d352ac68 | 403 | */ |
7585717f | 404 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
6426c7ad QW |
405 | struct btrfs_root *root, |
406 | int force) | |
6702ed49 | 407 | { |
0b246afa | 408 | struct btrfs_fs_info *fs_info = root->fs_info; |
03a7e111 | 409 | int ret = 0; |
0b246afa | 410 | |
92a7cc42 | 411 | if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
6426c7ad | 412 | root->last_trans < trans->transid) || force) { |
4d31778a | 413 | WARN_ON(!force && root->commit_root != root->node); |
5d4f98a2 | 414 | |
7585717f | 415 | /* |
27cdeb70 | 416 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
417 | * we have the reloc mutex held now, so there |
418 | * is only one writer in this function | |
419 | */ | |
27cdeb70 | 420 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 421 | |
27cdeb70 | 422 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
423 | * they find our root->last_trans update |
424 | */ | |
425 | smp_wmb(); | |
426 | ||
fc7cbcd4 | 427 | spin_lock(&fs_info->fs_roots_radix_lock); |
6426c7ad | 428 | if (root->last_trans == trans->transid && !force) { |
fc7cbcd4 | 429 | spin_unlock(&fs_info->fs_roots_radix_lock); |
a4abeea4 JB |
430 | return 0; |
431 | } | |
fc7cbcd4 DS |
432 | radix_tree_tag_set(&fs_info->fs_roots_radix, |
433 | (unsigned long)root->root_key.objectid, | |
434 | BTRFS_ROOT_TRANS_TAG); | |
435 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
7585717f CM |
436 | root->last_trans = trans->transid; |
437 | ||
438 | /* this is pretty tricky. We don't want to | |
439 | * take the relocation lock in btrfs_record_root_in_trans | |
440 | * unless we're really doing the first setup for this root in | |
441 | * this transaction. | |
442 | * | |
443 | * Normally we'd use root->last_trans as a flag to decide | |
444 | * if we want to take the expensive mutex. | |
445 | * | |
446 | * But, we have to set root->last_trans before we | |
447 | * init the relocation root, otherwise, we trip over warnings | |
448 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 449 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
450 | * fixing up the reloc trees and everyone must wait. |
451 | * | |
452 | * When this is zero, they can trust root->last_trans and fly | |
453 | * through btrfs_record_root_in_trans without having to take the | |
454 | * lock. smp_wmb() makes sure that all the writes above are | |
455 | * done before we pop in the zero below | |
456 | */ | |
03a7e111 | 457 | ret = btrfs_init_reloc_root(trans, root); |
c7548af6 | 458 | smp_mb__before_atomic(); |
27cdeb70 | 459 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 | 460 | } |
03a7e111 | 461 | return ret; |
5d4f98a2 | 462 | } |
bcc63abb | 463 | |
7585717f | 464 | |
2b9dbef2 JB |
465 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
466 | struct btrfs_root *root) | |
467 | { | |
0b246afa | 468 | struct btrfs_fs_info *fs_info = root->fs_info; |
2b9dbef2 JB |
469 | struct btrfs_transaction *cur_trans = trans->transaction; |
470 | ||
471 | /* Add ourselves to the transaction dropped list */ | |
472 | spin_lock(&cur_trans->dropped_roots_lock); | |
473 | list_add_tail(&root->root_list, &cur_trans->dropped_roots); | |
474 | spin_unlock(&cur_trans->dropped_roots_lock); | |
475 | ||
476 | /* Make sure we don't try to update the root at commit time */ | |
fc7cbcd4 DS |
477 | spin_lock(&fs_info->fs_roots_radix_lock); |
478 | radix_tree_tag_clear(&fs_info->fs_roots_radix, | |
479 | (unsigned long)root->root_key.objectid, | |
480 | BTRFS_ROOT_TRANS_TAG); | |
481 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
2b9dbef2 JB |
482 | } |
483 | ||
7585717f CM |
484 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
485 | struct btrfs_root *root) | |
486 | { | |
0b246afa | 487 | struct btrfs_fs_info *fs_info = root->fs_info; |
1409e6cc | 488 | int ret; |
0b246afa | 489 | |
92a7cc42 | 490 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
7585717f CM |
491 | return 0; |
492 | ||
493 | /* | |
27cdeb70 | 494 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
495 | * and barriers |
496 | */ | |
497 | smp_rmb(); | |
498 | if (root->last_trans == trans->transid && | |
27cdeb70 | 499 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
500 | return 0; |
501 | ||
0b246afa | 502 | mutex_lock(&fs_info->reloc_mutex); |
1409e6cc | 503 | ret = record_root_in_trans(trans, root, 0); |
0b246afa | 504 | mutex_unlock(&fs_info->reloc_mutex); |
7585717f | 505 | |
1409e6cc | 506 | return ret; |
7585717f CM |
507 | } |
508 | ||
4a9d8bde MX |
509 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
510 | { | |
3296bf56 | 511 | return (trans->state >= TRANS_STATE_COMMIT_START && |
501407aa | 512 | trans->state < TRANS_STATE_UNBLOCKED && |
bf31f87f | 513 | !TRANS_ABORTED(trans)); |
4a9d8bde MX |
514 | } |
515 | ||
d352ac68 CM |
516 | /* wait for commit against the current transaction to become unblocked |
517 | * when this is done, it is safe to start a new transaction, but the current | |
518 | * transaction might not be fully on disk. | |
519 | */ | |
2ff7e61e | 520 | static void wait_current_trans(struct btrfs_fs_info *fs_info) |
79154b1b | 521 | { |
f9295749 | 522 | struct btrfs_transaction *cur_trans; |
79154b1b | 523 | |
0b246afa JM |
524 | spin_lock(&fs_info->trans_lock); |
525 | cur_trans = fs_info->running_transaction; | |
4a9d8bde | 526 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
9b64f57d | 527 | refcount_inc(&cur_trans->use_count); |
0b246afa | 528 | spin_unlock(&fs_info->trans_lock); |
72d63ed6 | 529 | |
3e738c53 | 530 | btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); |
0b246afa | 531 | wait_event(fs_info->transaction_wait, |
501407aa | 532 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
bf31f87f | 533 | TRANS_ABORTED(cur_trans)); |
724e2315 | 534 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 535 | } else { |
0b246afa | 536 | spin_unlock(&fs_info->trans_lock); |
f9295749 | 537 | } |
37d1aeee CM |
538 | } |
539 | ||
2ff7e61e | 540 | static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) |
a22285a6 | 541 | { |
0b246afa | 542 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
a4abeea4 JB |
543 | return 0; |
544 | ||
92e2f7e3 | 545 | if (type == TRANS_START) |
a22285a6 | 546 | return 1; |
a4abeea4 | 547 | |
a22285a6 YZ |
548 | return 0; |
549 | } | |
550 | ||
20dd2cbf MX |
551 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
552 | { | |
0b246afa JM |
553 | struct btrfs_fs_info *fs_info = root->fs_info; |
554 | ||
555 | if (!fs_info->reloc_ctl || | |
92a7cc42 | 556 | !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || |
20dd2cbf MX |
557 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
558 | root->reloc_root) | |
559 | return false; | |
560 | ||
561 | return true; | |
562 | } | |
563 | ||
28270e25 FM |
564 | static int btrfs_reserve_trans_metadata(struct btrfs_fs_info *fs_info, |
565 | enum btrfs_reserve_flush_enum flush, | |
566 | u64 num_bytes, | |
567 | u64 *delayed_refs_bytes) | |
568 | { | |
569 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; | |
570 | struct btrfs_space_info *si = fs_info->trans_block_rsv.space_info; | |
571 | u64 extra_delayed_refs_bytes = 0; | |
572 | u64 bytes; | |
573 | int ret; | |
574 | ||
575 | /* | |
576 | * If there's a gap between the size of the delayed refs reserve and | |
577 | * its reserved space, than some tasks have added delayed refs or bumped | |
578 | * its size otherwise (due to block group creation or removal, or block | |
579 | * group item update). Also try to allocate that gap in order to prevent | |
580 | * using (and possibly abusing) the global reserve when committing the | |
581 | * transaction. | |
582 | */ | |
583 | if (flush == BTRFS_RESERVE_FLUSH_ALL && | |
584 | !btrfs_block_rsv_full(delayed_refs_rsv)) { | |
585 | spin_lock(&delayed_refs_rsv->lock); | |
586 | if (delayed_refs_rsv->size > delayed_refs_rsv->reserved) | |
587 | extra_delayed_refs_bytes = delayed_refs_rsv->size - | |
588 | delayed_refs_rsv->reserved; | |
589 | spin_unlock(&delayed_refs_rsv->lock); | |
590 | } | |
591 | ||
592 | bytes = num_bytes + *delayed_refs_bytes + extra_delayed_refs_bytes; | |
593 | ||
594 | /* | |
595 | * We want to reserve all the bytes we may need all at once, so we only | |
596 | * do 1 enospc flushing cycle per transaction start. | |
597 | */ | |
598 | ret = btrfs_reserve_metadata_bytes(fs_info, si, bytes, flush); | |
599 | if (ret == 0) { | |
600 | if (extra_delayed_refs_bytes > 0) | |
601 | btrfs_migrate_to_delayed_refs_rsv(fs_info, | |
602 | extra_delayed_refs_bytes); | |
603 | return 0; | |
604 | } | |
605 | ||
606 | if (extra_delayed_refs_bytes > 0) { | |
607 | bytes -= extra_delayed_refs_bytes; | |
608 | ret = btrfs_reserve_metadata_bytes(fs_info, si, bytes, flush); | |
609 | if (ret == 0) | |
610 | return 0; | |
611 | } | |
612 | ||
613 | /* | |
614 | * If we are an emergency flush, which can steal from the global block | |
615 | * reserve, then attempt to not reserve space for the delayed refs, as | |
616 | * we will consume space for them from the global block reserve. | |
617 | */ | |
618 | if (flush == BTRFS_RESERVE_FLUSH_ALL_STEAL) { | |
619 | bytes -= *delayed_refs_bytes; | |
620 | *delayed_refs_bytes = 0; | |
621 | ret = btrfs_reserve_metadata_bytes(fs_info, si, bytes, flush); | |
622 | } | |
623 | ||
624 | return ret; | |
625 | } | |
626 | ||
08e007d2 | 627 | static struct btrfs_trans_handle * |
5aed1dd8 | 628 | start_transaction(struct btrfs_root *root, unsigned int num_items, |
003d7c59 JM |
629 | unsigned int type, enum btrfs_reserve_flush_enum flush, |
630 | bool enforce_qgroups) | |
37d1aeee | 631 | { |
0b246afa | 632 | struct btrfs_fs_info *fs_info = root->fs_info; |
ba2c4d4e | 633 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
28270e25 | 634 | struct btrfs_block_rsv *trans_rsv = &fs_info->trans_block_rsv; |
a22285a6 YZ |
635 | struct btrfs_trans_handle *h; |
636 | struct btrfs_transaction *cur_trans; | |
b5009945 | 637 | u64 num_bytes = 0; |
c5567237 | 638 | u64 qgroup_reserved = 0; |
28270e25 | 639 | u64 delayed_refs_bytes = 0; |
20dd2cbf | 640 | bool reloc_reserved = false; |
9c343784 | 641 | bool do_chunk_alloc = false; |
20dd2cbf | 642 | int ret; |
acce952b | 643 | |
84961539 | 644 | if (BTRFS_FS_ERROR(fs_info)) |
acce952b | 645 | return ERR_PTR(-EROFS); |
2a1eb461 | 646 | |
46c4e71e | 647 | if (current->journal_info) { |
0860adfd | 648 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 | 649 | h = current->journal_info; |
b50fff81 DS |
650 | refcount_inc(&h->use_count); |
651 | WARN_ON(refcount_read(&h->use_count) > 2); | |
2a1eb461 JB |
652 | h->orig_rsv = h->block_rsv; |
653 | h->block_rsv = NULL; | |
654 | goto got_it; | |
655 | } | |
b5009945 JB |
656 | |
657 | /* | |
658 | * Do the reservation before we join the transaction so we can do all | |
659 | * the appropriate flushing if need be. | |
660 | */ | |
003d7c59 | 661 | if (num_items && root != fs_info->chunk_root) { |
0b246afa | 662 | qgroup_reserved = num_items * fs_info->nodesize; |
a6496849 BB |
663 | /* |
664 | * Use prealloc for now, as there might be a currently running | |
665 | * transaction that could free this reserved space prematurely | |
666 | * by committing. | |
667 | */ | |
668 | ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_reserved, | |
669 | enforce_qgroups, false); | |
7174109c QW |
670 | if (ret) |
671 | return ERR_PTR(ret); | |
c5567237 | 672 | |
28270e25 | 673 | num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); |
ba2c4d4e | 674 | /* |
28270e25 FM |
675 | * If we plan to insert/update/delete "num_items" from a btree, |
676 | * we will also generate delayed refs for extent buffers in the | |
677 | * respective btree paths, so reserve space for the delayed refs | |
678 | * that will be generated by the caller as it modifies btrees. | |
679 | * Try to reserve them to avoid excessive use of the global | |
680 | * block reserve. | |
ba2c4d4e | 681 | */ |
28270e25 | 682 | delayed_refs_bytes = btrfs_calc_delayed_ref_bytes(fs_info, num_items); |
ba2c4d4e | 683 | |
20dd2cbf MX |
684 | /* |
685 | * Do the reservation for the relocation root creation | |
686 | */ | |
ee39b432 | 687 | if (need_reserve_reloc_root(root)) { |
0b246afa | 688 | num_bytes += fs_info->nodesize; |
20dd2cbf MX |
689 | reloc_reserved = true; |
690 | } | |
691 | ||
28270e25 FM |
692 | ret = btrfs_reserve_trans_metadata(fs_info, flush, num_bytes, |
693 | &delayed_refs_bytes); | |
ba2c4d4e JB |
694 | if (ret) |
695 | goto reserve_fail; | |
9c343784 | 696 | |
28270e25 FM |
697 | btrfs_block_rsv_add_bytes(trans_rsv, num_bytes, true); |
698 | ||
699 | if (trans_rsv->space_info->force_alloc) | |
9c343784 | 700 | do_chunk_alloc = true; |
ba2c4d4e | 701 | } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && |
748f553c | 702 | !btrfs_block_rsv_full(delayed_refs_rsv)) { |
ba2c4d4e JB |
703 | /* |
704 | * Some people call with btrfs_start_transaction(root, 0) | |
705 | * because they can be throttled, but have some other mechanism | |
706 | * for reserving space. We still want these guys to refill the | |
707 | * delayed block_rsv so just add 1 items worth of reservation | |
708 | * here. | |
709 | */ | |
710 | ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); | |
b5009945 | 711 | if (ret) |
843fcf35 | 712 | goto reserve_fail; |
b5009945 | 713 | } |
a22285a6 | 714 | again: |
f2f767e7 | 715 | h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); |
843fcf35 MX |
716 | if (!h) { |
717 | ret = -ENOMEM; | |
718 | goto alloc_fail; | |
719 | } | |
37d1aeee | 720 | |
98114659 JB |
721 | /* |
722 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
723 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
724 | * because we're already holding a ref. We need this because we could | |
725 | * have raced in and did an fsync() on a file which can kick a commit | |
726 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
727 | * |
728 | * If we are ATTACH, it means we just want to catch the current | |
729 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 730 | */ |
0860adfd | 731 | if (type & __TRANS_FREEZABLE) |
0b246afa | 732 | sb_start_intwrite(fs_info->sb); |
b2b5ef5c | 733 | |
2ff7e61e JM |
734 | if (may_wait_transaction(fs_info, type)) |
735 | wait_current_trans(fs_info); | |
a22285a6 | 736 | |
a4abeea4 | 737 | do { |
2ff7e61e | 738 | ret = join_transaction(fs_info, type); |
178260b2 | 739 | if (ret == -EBUSY) { |
2ff7e61e | 740 | wait_current_trans(fs_info); |
a6d155d2 FM |
741 | if (unlikely(type == TRANS_ATTACH || |
742 | type == TRANS_JOIN_NOSTART)) | |
178260b2 MX |
743 | ret = -ENOENT; |
744 | } | |
a4abeea4 JB |
745 | } while (ret == -EBUSY); |
746 | ||
a43f7f82 | 747 | if (ret < 0) |
843fcf35 | 748 | goto join_fail; |
0f7d52f4 | 749 | |
0b246afa | 750 | cur_trans = fs_info->running_transaction; |
a22285a6 YZ |
751 | |
752 | h->transid = cur_trans->transid; | |
753 | h->transaction = cur_trans; | |
b50fff81 | 754 | refcount_set(&h->use_count, 1); |
64b63580 | 755 | h->fs_info = root->fs_info; |
7174109c | 756 | |
a698d075 | 757 | h->type = type; |
ea658bad | 758 | INIT_LIST_HEAD(&h->new_bgs); |
28270e25 | 759 | btrfs_init_metadata_block_rsv(fs_info, &h->delayed_rsv, BTRFS_BLOCK_RSV_DELOPS); |
b7ec40d7 | 760 | |
a22285a6 | 761 | smp_mb(); |
3296bf56 | 762 | if (cur_trans->state >= TRANS_STATE_COMMIT_START && |
2ff7e61e | 763 | may_wait_transaction(fs_info, type)) { |
abdd2e80 | 764 | current->journal_info = h; |
3a45bb20 | 765 | btrfs_commit_transaction(h); |
a22285a6 YZ |
766 | goto again; |
767 | } | |
768 | ||
b5009945 | 769 | if (num_bytes) { |
0b246afa | 770 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 771 | h->transid, num_bytes, 1); |
28270e25 | 772 | h->block_rsv = trans_rsv; |
b5009945 | 773 | h->bytes_reserved = num_bytes; |
28270e25 FM |
774 | if (delayed_refs_bytes > 0) { |
775 | trace_btrfs_space_reservation(fs_info, | |
776 | "local_delayed_refs_rsv", | |
777 | h->transid, | |
778 | delayed_refs_bytes, 1); | |
779 | h->delayed_refs_bytes_reserved = delayed_refs_bytes; | |
780 | btrfs_block_rsv_add_bytes(&h->delayed_rsv, delayed_refs_bytes, true); | |
781 | delayed_refs_bytes = 0; | |
782 | } | |
20dd2cbf | 783 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 784 | } |
9ed74f2d | 785 | |
a6496849 BB |
786 | /* |
787 | * Now that we have found a transaction to be a part of, convert the | |
788 | * qgroup reservation from prealloc to pertrans. A different transaction | |
789 | * can't race in and free our pertrans out from under us. | |
790 | */ | |
791 | if (qgroup_reserved) | |
792 | btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved); | |
793 | ||
2a1eb461 | 794 | got_it: |
bcf3a3e7 | 795 | if (!current->journal_info) |
a22285a6 | 796 | current->journal_info = h; |
fcc99734 | 797 | |
9c343784 JB |
798 | /* |
799 | * If the space_info is marked ALLOC_FORCE then we'll get upgraded to | |
800 | * ALLOC_FORCE the first run through, and then we won't allocate for | |
801 | * anybody else who races in later. We don't care about the return | |
802 | * value here. | |
803 | */ | |
804 | if (do_chunk_alloc && num_bytes) { | |
805 | u64 flags = h->block_rsv->space_info->flags; | |
806 | ||
807 | btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), | |
808 | CHUNK_ALLOC_NO_FORCE); | |
809 | } | |
810 | ||
fcc99734 QW |
811 | /* |
812 | * btrfs_record_root_in_trans() needs to alloc new extents, and may | |
813 | * call btrfs_join_transaction() while we're also starting a | |
814 | * transaction. | |
815 | * | |
816 | * Thus it need to be called after current->journal_info initialized, | |
817 | * or we can deadlock. | |
818 | */ | |
68075ea8 JB |
819 | ret = btrfs_record_root_in_trans(h, root); |
820 | if (ret) { | |
821 | /* | |
822 | * The transaction handle is fully initialized and linked with | |
823 | * other structures so it needs to be ended in case of errors, | |
824 | * not just freed. | |
825 | */ | |
826 | btrfs_end_transaction(h); | |
827 | return ERR_PTR(ret); | |
828 | } | |
fcc99734 | 829 | |
79154b1b | 830 | return h; |
843fcf35 MX |
831 | |
832 | join_fail: | |
0860adfd | 833 | if (type & __TRANS_FREEZABLE) |
0b246afa | 834 | sb_end_intwrite(fs_info->sb); |
843fcf35 MX |
835 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
836 | alloc_fail: | |
837 | if (num_bytes) | |
28270e25 FM |
838 | btrfs_block_rsv_release(fs_info, trans_rsv, num_bytes, NULL); |
839 | if (delayed_refs_bytes) | |
840 | btrfs_space_info_free_bytes_may_use(fs_info, trans_rsv->space_info, | |
841 | delayed_refs_bytes); | |
843fcf35 | 842 | reserve_fail: |
a6496849 | 843 | btrfs_qgroup_free_meta_prealloc(root, qgroup_reserved); |
843fcf35 | 844 | return ERR_PTR(ret); |
79154b1b CM |
845 | } |
846 | ||
f9295749 | 847 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
5aed1dd8 | 848 | unsigned int num_items) |
f9295749 | 849 | { |
08e007d2 | 850 | return start_transaction(root, num_items, TRANS_START, |
003d7c59 | 851 | BTRFS_RESERVE_FLUSH_ALL, true); |
f9295749 | 852 | } |
003d7c59 | 853 | |
8eab77ff FM |
854 | struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( |
855 | struct btrfs_root *root, | |
7f9fe614 | 856 | unsigned int num_items) |
8eab77ff | 857 | { |
7f9fe614 JB |
858 | return start_transaction(root, num_items, TRANS_START, |
859 | BTRFS_RESERVE_FLUSH_ALL_STEAL, false); | |
8eab77ff | 860 | } |
8407aa46 | 861 | |
7a7eaa40 | 862 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 863 | { |
003d7c59 JM |
864 | return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, |
865 | true); | |
f9295749 CM |
866 | } |
867 | ||
8d510121 | 868 | struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) |
0af3d00b | 869 | { |
575a75d6 | 870 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, |
003d7c59 | 871 | BTRFS_RESERVE_NO_FLUSH, true); |
0af3d00b JB |
872 | } |
873 | ||
a6d155d2 FM |
874 | /* |
875 | * Similar to regular join but it never starts a transaction when none is | |
19288951 FM |
876 | * running or when there's a running one at a state >= TRANS_STATE_UNBLOCKED. |
877 | * This is similar to btrfs_attach_transaction() but it allows the join to | |
878 | * happen if the transaction commit already started but it's not yet in the | |
879 | * "doing" phase (the state is < TRANS_STATE_COMMIT_DOING). | |
a6d155d2 FM |
880 | */ |
881 | struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) | |
882 | { | |
883 | return start_transaction(root, 0, TRANS_JOIN_NOSTART, | |
884 | BTRFS_RESERVE_NO_FLUSH, true); | |
885 | } | |
886 | ||
d4edf39b | 887 | /* |
9580503b | 888 | * Catch the running transaction. |
d4edf39b MX |
889 | * |
890 | * It is used when we want to commit the current the transaction, but | |
891 | * don't want to start a new one. | |
892 | * | |
893 | * Note: If this function return -ENOENT, it just means there is no | |
894 | * running transaction. But it is possible that the inactive transaction | |
895 | * is still in the memory, not fully on disk. If you hope there is no | |
896 | * inactive transaction in the fs when -ENOENT is returned, you should | |
897 | * invoke | |
898 | * btrfs_attach_transaction_barrier() | |
899 | */ | |
354aa0fb | 900 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 901 | { |
575a75d6 | 902 | return start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 903 | BTRFS_RESERVE_NO_FLUSH, true); |
60376ce4 JB |
904 | } |
905 | ||
d4edf39b | 906 | /* |
9580503b | 907 | * Catch the running transaction. |
d4edf39b | 908 | * |
52042d8e | 909 | * It is similar to the above function, the difference is this one |
d4edf39b MX |
910 | * will wait for all the inactive transactions until they fully |
911 | * complete. | |
912 | */ | |
913 | struct btrfs_trans_handle * | |
914 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
915 | { | |
916 | struct btrfs_trans_handle *trans; | |
917 | ||
575a75d6 | 918 | trans = start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 919 | BTRFS_RESERVE_NO_FLUSH, true); |
b28ff3a7 FM |
920 | if (trans == ERR_PTR(-ENOENT)) { |
921 | int ret; | |
922 | ||
923 | ret = btrfs_wait_for_commit(root->fs_info, 0); | |
924 | if (ret) | |
925 | return ERR_PTR(ret); | |
926 | } | |
d4edf39b MX |
927 | |
928 | return trans; | |
929 | } | |
930 | ||
d0c2f4fa FM |
931 | /* Wait for a transaction commit to reach at least the given state. */ |
932 | static noinline void wait_for_commit(struct btrfs_transaction *commit, | |
933 | const enum btrfs_trans_state min_state) | |
89ce8a63 | 934 | { |
5fd76bf3 OS |
935 | struct btrfs_fs_info *fs_info = commit->fs_info; |
936 | u64 transid = commit->transid; | |
937 | bool put = false; | |
938 | ||
3e738c53 IA |
939 | /* |
940 | * At the moment this function is called with min_state either being | |
941 | * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED. | |
942 | */ | |
943 | if (min_state == TRANS_STATE_COMPLETED) | |
944 | btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); | |
945 | else | |
946 | btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); | |
947 | ||
5fd76bf3 OS |
948 | while (1) { |
949 | wait_event(commit->commit_wait, commit->state >= min_state); | |
950 | if (put) | |
951 | btrfs_put_transaction(commit); | |
952 | ||
953 | if (min_state < TRANS_STATE_COMPLETED) | |
954 | break; | |
955 | ||
956 | /* | |
957 | * A transaction isn't really completed until all of the | |
958 | * previous transactions are completed, but with fsync we can | |
959 | * end up with SUPER_COMMITTED transactions before a COMPLETED | |
960 | * transaction. Wait for those. | |
961 | */ | |
962 | ||
963 | spin_lock(&fs_info->trans_lock); | |
964 | commit = list_first_entry_or_null(&fs_info->trans_list, | |
965 | struct btrfs_transaction, | |
966 | list); | |
967 | if (!commit || commit->transid > transid) { | |
968 | spin_unlock(&fs_info->trans_lock); | |
969 | break; | |
970 | } | |
971 | refcount_inc(&commit->use_count); | |
972 | put = true; | |
973 | spin_unlock(&fs_info->trans_lock); | |
974 | } | |
89ce8a63 CM |
975 | } |
976 | ||
2ff7e61e | 977 | int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) |
46204592 SW |
978 | { |
979 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 980 | int ret = 0; |
46204592 | 981 | |
46204592 | 982 | if (transid) { |
0124855f | 983 | if (transid <= btrfs_get_last_trans_committed(fs_info)) |
a4abeea4 | 984 | goto out; |
46204592 SW |
985 | |
986 | /* find specified transaction */ | |
0b246afa JM |
987 | spin_lock(&fs_info->trans_lock); |
988 | list_for_each_entry(t, &fs_info->trans_list, list) { | |
46204592 SW |
989 | if (t->transid == transid) { |
990 | cur_trans = t; | |
9b64f57d | 991 | refcount_inc(&cur_trans->use_count); |
8cd2807f | 992 | ret = 0; |
46204592 SW |
993 | break; |
994 | } | |
8cd2807f MX |
995 | if (t->transid > transid) { |
996 | ret = 0; | |
46204592 | 997 | break; |
8cd2807f | 998 | } |
46204592 | 999 | } |
0b246afa | 1000 | spin_unlock(&fs_info->trans_lock); |
42383020 SW |
1001 | |
1002 | /* | |
1003 | * The specified transaction doesn't exist, or we | |
1004 | * raced with btrfs_commit_transaction | |
1005 | */ | |
1006 | if (!cur_trans) { | |
0124855f | 1007 | if (transid > btrfs_get_last_trans_committed(fs_info)) |
42383020 | 1008 | ret = -EINVAL; |
8cd2807f | 1009 | goto out; |
42383020 | 1010 | } |
46204592 SW |
1011 | } else { |
1012 | /* find newest transaction that is committing | committed */ | |
0b246afa JM |
1013 | spin_lock(&fs_info->trans_lock); |
1014 | list_for_each_entry_reverse(t, &fs_info->trans_list, | |
46204592 | 1015 | list) { |
4a9d8bde MX |
1016 | if (t->state >= TRANS_STATE_COMMIT_START) { |
1017 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 1018 | break; |
46204592 | 1019 | cur_trans = t; |
9b64f57d | 1020 | refcount_inc(&cur_trans->use_count); |
46204592 SW |
1021 | break; |
1022 | } | |
1023 | } | |
0b246afa | 1024 | spin_unlock(&fs_info->trans_lock); |
46204592 | 1025 | if (!cur_trans) |
a4abeea4 | 1026 | goto out; /* nothing committing|committed */ |
46204592 SW |
1027 | } |
1028 | ||
d0c2f4fa | 1029 | wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); |
bf7ecbe9 | 1030 | ret = cur_trans->aborted; |
724e2315 | 1031 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 1032 | out: |
46204592 SW |
1033 | return ret; |
1034 | } | |
1035 | ||
2ff7e61e | 1036 | void btrfs_throttle(struct btrfs_fs_info *fs_info) |
37d1aeee | 1037 | { |
92e2f7e3 | 1038 | wait_current_trans(fs_info); |
37d1aeee CM |
1039 | } |
1040 | ||
a2633b6a | 1041 | bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa YZ |
1042 | { |
1043 | struct btrfs_transaction *cur_trans = trans->transaction; | |
8929ecfa | 1044 | |
3296bf56 | 1045 | if (cur_trans->state >= TRANS_STATE_COMMIT_START || |
e19eb11f | 1046 | test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) |
a2633b6a | 1047 | return true; |
8929ecfa | 1048 | |
04fb3285 FM |
1049 | if (btrfs_check_space_for_delayed_refs(trans->fs_info)) |
1050 | return true; | |
1051 | ||
1052 | return !!btrfs_block_rsv_check(&trans->fs_info->global_block_rsv, 50); | |
8929ecfa YZ |
1053 | } |
1054 | ||
dc60c525 NB |
1055 | static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) |
1056 | ||
0e34693f | 1057 | { |
dc60c525 NB |
1058 | struct btrfs_fs_info *fs_info = trans->fs_info; |
1059 | ||
0e34693f NB |
1060 | if (!trans->block_rsv) { |
1061 | ASSERT(!trans->bytes_reserved); | |
28270e25 | 1062 | ASSERT(!trans->delayed_refs_bytes_reserved); |
0e34693f NB |
1063 | return; |
1064 | } | |
1065 | ||
28270e25 FM |
1066 | if (!trans->bytes_reserved) { |
1067 | ASSERT(!trans->delayed_refs_bytes_reserved); | |
0e34693f | 1068 | return; |
28270e25 | 1069 | } |
0e34693f NB |
1070 | |
1071 | ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); | |
1072 | trace_btrfs_space_reservation(fs_info, "transaction", | |
1073 | trans->transid, trans->bytes_reserved, 0); | |
1074 | btrfs_block_rsv_release(fs_info, trans->block_rsv, | |
63f018be | 1075 | trans->bytes_reserved, NULL); |
0e34693f | 1076 | trans->bytes_reserved = 0; |
28270e25 FM |
1077 | |
1078 | if (!trans->delayed_refs_bytes_reserved) | |
1079 | return; | |
1080 | ||
1081 | trace_btrfs_space_reservation(fs_info, "local_delayed_refs_rsv", | |
1082 | trans->transid, | |
1083 | trans->delayed_refs_bytes_reserved, 0); | |
1084 | btrfs_block_rsv_release(fs_info, &trans->delayed_rsv, | |
1085 | trans->delayed_refs_bytes_reserved, NULL); | |
1086 | trans->delayed_refs_bytes_reserved = 0; | |
0e34693f NB |
1087 | } |
1088 | ||
89ce8a63 | 1089 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
3a45bb20 | 1090 | int throttle) |
79154b1b | 1091 | { |
3a45bb20 | 1092 | struct btrfs_fs_info *info = trans->fs_info; |
8929ecfa | 1093 | struct btrfs_transaction *cur_trans = trans->transaction; |
4edc2ca3 | 1094 | int err = 0; |
c3e69d58 | 1095 | |
b50fff81 DS |
1096 | if (refcount_read(&trans->use_count) > 1) { |
1097 | refcount_dec(&trans->use_count); | |
2a1eb461 JB |
1098 | trans->block_rsv = trans->orig_rsv; |
1099 | return 0; | |
1100 | } | |
1101 | ||
dc60c525 | 1102 | btrfs_trans_release_metadata(trans); |
4c13d758 | 1103 | trans->block_rsv = NULL; |
c5567237 | 1104 | |
119e80df | 1105 | btrfs_create_pending_block_groups(trans); |
ea658bad | 1106 | |
4fbcdf66 FM |
1107 | btrfs_trans_release_chunk_metadata(trans); |
1108 | ||
0860adfd | 1109 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 1110 | sb_end_intwrite(info->sb); |
6df7881a | 1111 | |
8929ecfa | 1112 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
1113 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
1114 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 1115 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 1116 | |
093258e6 | 1117 | cond_wake_up(&cur_trans->writer_wait); |
e1489b4f | 1118 | |
5a9ba670 | 1119 | btrfs_lockdep_release(info, btrfs_trans_num_extwriters); |
e1489b4f IA |
1120 | btrfs_lockdep_release(info, btrfs_trans_num_writers); |
1121 | ||
724e2315 | 1122 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
1123 | |
1124 | if (current->journal_info == trans) | |
1125 | current->journal_info = NULL; | |
ab78c84d | 1126 | |
24bbcf04 | 1127 | if (throttle) |
2ff7e61e | 1128 | btrfs_run_delayed_iputs(info); |
24bbcf04 | 1129 | |
84961539 | 1130 | if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) { |
4e121c06 | 1131 | wake_up_process(info->transaction_kthread); |
fbabd4a3 JB |
1132 | if (TRANS_ABORTED(trans)) |
1133 | err = trans->aborted; | |
1134 | else | |
1135 | err = -EROFS; | |
4e121c06 | 1136 | } |
49b25e05 | 1137 | |
4edc2ca3 DJ |
1138 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
1139 | return err; | |
79154b1b CM |
1140 | } |
1141 | ||
3a45bb20 | 1142 | int btrfs_end_transaction(struct btrfs_trans_handle *trans) |
89ce8a63 | 1143 | { |
3a45bb20 | 1144 | return __btrfs_end_transaction(trans, 0); |
89ce8a63 CM |
1145 | } |
1146 | ||
3a45bb20 | 1147 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) |
89ce8a63 | 1148 | { |
3a45bb20 | 1149 | return __btrfs_end_transaction(trans, 1); |
16cdcec7 MX |
1150 | } |
1151 | ||
d352ac68 CM |
1152 | /* |
1153 | * when btree blocks are allocated, they have some corresponding bits set for | |
1154 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 1155 | * those extents are sent to disk but does not wait on them |
d352ac68 | 1156 | */ |
2ff7e61e | 1157 | int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, |
8cef4e16 | 1158 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 1159 | { |
777e6bd7 | 1160 | int err = 0; |
7c4452b9 | 1161 | int werr = 0; |
0b246afa | 1162 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 1163 | struct extent_state *cached_state = NULL; |
777e6bd7 | 1164 | u64 start = 0; |
5f39d397 | 1165 | u64 end; |
7c4452b9 | 1166 | |
e5860f82 FM |
1167 | while (find_first_extent_bit(dirty_pages, start, &start, &end, |
1168 | mark, &cached_state)) { | |
663dfbb0 FM |
1169 | bool wait_writeback = false; |
1170 | ||
1171 | err = convert_extent_bit(dirty_pages, start, end, | |
1172 | EXTENT_NEED_WAIT, | |
210aa277 | 1173 | mark, &cached_state); |
663dfbb0 FM |
1174 | /* |
1175 | * convert_extent_bit can return -ENOMEM, which is most of the | |
1176 | * time a temporary error. So when it happens, ignore the error | |
1177 | * and wait for writeback of this range to finish - because we | |
1178 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
bf89d38f JM |
1179 | * to __btrfs_wait_marked_extents() would not know that |
1180 | * writeback for this range started and therefore wouldn't | |
1181 | * wait for it to finish - we don't want to commit a | |
1182 | * superblock that points to btree nodes/leafs for which | |
1183 | * writeback hasn't finished yet (and without errors). | |
663dfbb0 | 1184 | * We cleanup any entries left in the io tree when committing |
41e7acd3 | 1185 | * the transaction (through extent_io_tree_release()). |
663dfbb0 FM |
1186 | */ |
1187 | if (err == -ENOMEM) { | |
1188 | err = 0; | |
1189 | wait_writeback = true; | |
1190 | } | |
1191 | if (!err) | |
1192 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
1193 | if (err) |
1194 | werr = err; | |
663dfbb0 FM |
1195 | else if (wait_writeback) |
1196 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 1197 | free_extent_state(cached_state); |
663dfbb0 | 1198 | cached_state = NULL; |
1728366e JB |
1199 | cond_resched(); |
1200 | start = end + 1; | |
7c4452b9 | 1201 | } |
690587d1 CM |
1202 | return werr; |
1203 | } | |
1204 | ||
1205 | /* | |
1206 | * when btree blocks are allocated, they have some corresponding bits set for | |
1207 | * them in one of two extent_io trees. This is used to make sure all of | |
1208 | * those extents are on disk for transaction or log commit. We wait | |
1209 | * on all the pages and clear them from the dirty pages state tree | |
1210 | */ | |
bf89d38f JM |
1211 | static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, |
1212 | struct extent_io_tree *dirty_pages) | |
690587d1 | 1213 | { |
690587d1 CM |
1214 | int err = 0; |
1215 | int werr = 0; | |
0b246afa | 1216 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 1217 | struct extent_state *cached_state = NULL; |
690587d1 CM |
1218 | u64 start = 0; |
1219 | u64 end; | |
777e6bd7 | 1220 | |
e5860f82 FM |
1221 | while (find_first_extent_bit(dirty_pages, start, &start, &end, |
1222 | EXTENT_NEED_WAIT, &cached_state)) { | |
663dfbb0 FM |
1223 | /* |
1224 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
1225 | * When committing the transaction, we'll remove any entries | |
1226 | * left in the io tree. For a log commit, we don't remove them | |
1227 | * after committing the log because the tree can be accessed | |
1228 | * concurrently - we do it only at transaction commit time when | |
41e7acd3 | 1229 | * it's safe to do it (through extent_io_tree_release()). |
663dfbb0 FM |
1230 | */ |
1231 | err = clear_extent_bit(dirty_pages, start, end, | |
bd015294 | 1232 | EXTENT_NEED_WAIT, &cached_state); |
663dfbb0 FM |
1233 | if (err == -ENOMEM) |
1234 | err = 0; | |
1235 | if (!err) | |
1236 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
1237 | if (err) |
1238 | werr = err; | |
e38e2ed7 FM |
1239 | free_extent_state(cached_state); |
1240 | cached_state = NULL; | |
1728366e JB |
1241 | cond_resched(); |
1242 | start = end + 1; | |
777e6bd7 | 1243 | } |
7c4452b9 CM |
1244 | if (err) |
1245 | werr = err; | |
bf89d38f JM |
1246 | return werr; |
1247 | } | |
656f30db | 1248 | |
b9fae2eb | 1249 | static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, |
bf89d38f JM |
1250 | struct extent_io_tree *dirty_pages) |
1251 | { | |
1252 | bool errors = false; | |
1253 | int err; | |
656f30db | 1254 | |
bf89d38f JM |
1255 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); |
1256 | if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) | |
1257 | errors = true; | |
1258 | ||
1259 | if (errors && !err) | |
1260 | err = -EIO; | |
1261 | return err; | |
1262 | } | |
656f30db | 1263 | |
bf89d38f JM |
1264 | int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) |
1265 | { | |
1266 | struct btrfs_fs_info *fs_info = log_root->fs_info; | |
1267 | struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; | |
1268 | bool errors = false; | |
1269 | int err; | |
656f30db | 1270 | |
bf89d38f JM |
1271 | ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
1272 | ||
1273 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); | |
1274 | if ((mark & EXTENT_DIRTY) && | |
1275 | test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) | |
1276 | errors = true; | |
1277 | ||
1278 | if ((mark & EXTENT_NEW) && | |
1279 | test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) | |
1280 | errors = true; | |
1281 | ||
1282 | if (errors && !err) | |
1283 | err = -EIO; | |
1284 | return err; | |
79154b1b CM |
1285 | } |
1286 | ||
690587d1 | 1287 | /* |
c9b577c0 NB |
1288 | * When btree blocks are allocated the corresponding extents are marked dirty. |
1289 | * This function ensures such extents are persisted on disk for transaction or | |
1290 | * log commit. | |
1291 | * | |
1292 | * @trans: transaction whose dirty pages we'd like to write | |
690587d1 | 1293 | */ |
70458a58 | 1294 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) |
690587d1 CM |
1295 | { |
1296 | int ret; | |
1297 | int ret2; | |
c9b577c0 | 1298 | struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; |
70458a58 | 1299 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c6adc9cc | 1300 | struct blk_plug plug; |
690587d1 | 1301 | |
c6adc9cc | 1302 | blk_start_plug(&plug); |
c9b577c0 | 1303 | ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); |
c6adc9cc | 1304 | blk_finish_plug(&plug); |
bf89d38f | 1305 | ret2 = btrfs_wait_extents(fs_info, dirty_pages); |
bf0da8c1 | 1306 | |
41e7acd3 | 1307 | extent_io_tree_release(&trans->transaction->dirty_pages); |
c9b577c0 | 1308 | |
bf0da8c1 CM |
1309 | if (ret) |
1310 | return ret; | |
c9b577c0 | 1311 | else if (ret2) |
bf0da8c1 | 1312 | return ret2; |
c9b577c0 NB |
1313 | else |
1314 | return 0; | |
d0c803c4 CM |
1315 | } |
1316 | ||
d352ac68 CM |
1317 | /* |
1318 | * this is used to update the root pointer in the tree of tree roots. | |
1319 | * | |
1320 | * But, in the case of the extent allocation tree, updating the root | |
1321 | * pointer may allocate blocks which may change the root of the extent | |
1322 | * allocation tree. | |
1323 | * | |
1324 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1325 | * change while the root pointer was being updated in the metadata. | |
1326 | */ | |
0b86a832 CM |
1327 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1328 | struct btrfs_root *root) | |
79154b1b CM |
1329 | { |
1330 | int ret; | |
0b86a832 | 1331 | u64 old_root_bytenr; |
86b9f2ec | 1332 | u64 old_root_used; |
0b246afa JM |
1333 | struct btrfs_fs_info *fs_info = root->fs_info; |
1334 | struct btrfs_root *tree_root = fs_info->tree_root; | |
79154b1b | 1335 | |
86b9f2ec | 1336 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1337 | |
d397712b | 1338 | while (1) { |
0b86a832 | 1339 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1340 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1341 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1342 | break; |
87ef2bb4 | 1343 | |
5d4f98a2 | 1344 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1345 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1346 | &root->root_key, |
1347 | &root->root_item); | |
49b25e05 JM |
1348 | if (ret) |
1349 | return ret; | |
56bec294 | 1350 | |
86b9f2ec | 1351 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1352 | } |
276e680d | 1353 | |
0b86a832 CM |
1354 | return 0; |
1355 | } | |
1356 | ||
d352ac68 CM |
1357 | /* |
1358 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1359 | * |
1360 | * The error handling in this function may not be obvious. Any of the | |
1361 | * failures will cause the file system to go offline. We still need | |
1362 | * to clean up the delayed refs. | |
d352ac68 | 1363 | */ |
9386d8bc | 1364 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) |
0b86a832 | 1365 | { |
9386d8bc | 1366 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ea526d18 | 1367 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1368 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1369 | struct list_head *next; |
84234f3a | 1370 | struct extent_buffer *eb; |
56bec294 | 1371 | int ret; |
84234f3a | 1372 | |
dfba78dc FM |
1373 | /* |
1374 | * At this point no one can be using this transaction to modify any tree | |
1375 | * and no one can start another transaction to modify any tree either. | |
1376 | */ | |
1377 | ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); | |
1378 | ||
84234f3a | 1379 | eb = btrfs_lock_root_node(fs_info->tree_root); |
49b25e05 | 1380 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
9631e4cc | 1381 | 0, &eb, BTRFS_NESTING_COW); |
84234f3a YZ |
1382 | btrfs_tree_unlock(eb); |
1383 | free_extent_buffer(eb); | |
0b86a832 | 1384 | |
49b25e05 JM |
1385 | if (ret) |
1386 | return ret; | |
87ef2bb4 | 1387 | |
196c9d8d | 1388 | ret = btrfs_run_dev_stats(trans); |
c16ce190 JB |
1389 | if (ret) |
1390 | return ret; | |
2b584c68 | 1391 | ret = btrfs_run_dev_replace(trans); |
c16ce190 JB |
1392 | if (ret) |
1393 | return ret; | |
280f8bd2 | 1394 | ret = btrfs_run_qgroups(trans); |
c16ce190 JB |
1395 | if (ret) |
1396 | return ret; | |
546adb0d | 1397 | |
bbebb3e0 | 1398 | ret = btrfs_setup_space_cache(trans); |
dcdf7f6d JB |
1399 | if (ret) |
1400 | return ret; | |
1401 | ||
ea526d18 | 1402 | again: |
d397712b | 1403 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
2ff7e61e | 1404 | struct btrfs_root *root; |
0b86a832 CM |
1405 | next = fs_info->dirty_cowonly_roots.next; |
1406 | list_del_init(next); | |
1407 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1408 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1409 | |
826582ca JB |
1410 | list_add_tail(&root->dirty_list, |
1411 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1412 | ret = update_cowonly_root(trans, root); |
1413 | if (ret) | |
1414 | return ret; | |
79154b1b | 1415 | } |
276e680d | 1416 | |
488bc2a2 | 1417 | /* Now flush any delayed refs generated by updating all of the roots */ |
8a526c44 | 1418 | ret = btrfs_run_delayed_refs(trans, U64_MAX); |
488bc2a2 JB |
1419 | if (ret) |
1420 | return ret; | |
1421 | ||
1bbc621e | 1422 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
5742d15f | 1423 | ret = btrfs_write_dirty_block_groups(trans); |
ea526d18 JB |
1424 | if (ret) |
1425 | return ret; | |
488bc2a2 JB |
1426 | |
1427 | /* | |
1428 | * We're writing the dirty block groups, which could generate | |
1429 | * delayed refs, which could generate more dirty block groups, | |
1430 | * so we want to keep this flushing in this loop to make sure | |
1431 | * everything gets run. | |
1432 | */ | |
8a526c44 | 1433 | ret = btrfs_run_delayed_refs(trans, U64_MAX); |
ea526d18 JB |
1434 | if (ret) |
1435 | return ret; | |
1436 | } | |
1437 | ||
1438 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1439 | goto again; | |
1440 | ||
9f6cbcbb DS |
1441 | /* Update dev-replace pointer once everything is committed */ |
1442 | fs_info->dev_replace.committed_cursor_left = | |
1443 | fs_info->dev_replace.cursor_left_last_write_of_item; | |
8dabb742 | 1444 | |
79154b1b CM |
1445 | return 0; |
1446 | } | |
1447 | ||
b4be6aef JB |
1448 | /* |
1449 | * If we had a pending drop we need to see if there are any others left in our | |
1450 | * dead roots list, and if not clear our bit and wake any waiters. | |
1451 | */ | |
1452 | void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info) | |
1453 | { | |
1454 | /* | |
1455 | * We put the drop in progress roots at the front of the list, so if the | |
1456 | * first entry doesn't have UNFINISHED_DROP set we can wake everybody | |
1457 | * up. | |
1458 | */ | |
1459 | spin_lock(&fs_info->trans_lock); | |
1460 | if (!list_empty(&fs_info->dead_roots)) { | |
1461 | struct btrfs_root *root = list_first_entry(&fs_info->dead_roots, | |
1462 | struct btrfs_root, | |
1463 | root_list); | |
1464 | if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) { | |
1465 | spin_unlock(&fs_info->trans_lock); | |
1466 | return; | |
1467 | } | |
1468 | } | |
1469 | spin_unlock(&fs_info->trans_lock); | |
1470 | ||
1471 | btrfs_wake_unfinished_drop(fs_info); | |
1472 | } | |
1473 | ||
d352ac68 CM |
1474 | /* |
1475 | * dead roots are old snapshots that need to be deleted. This allocates | |
1476 | * a dirty root struct and adds it into the list of dead roots that need to | |
1477 | * be deleted | |
1478 | */ | |
cfad392b | 1479 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1480 | { |
0b246afa JM |
1481 | struct btrfs_fs_info *fs_info = root->fs_info; |
1482 | ||
1483 | spin_lock(&fs_info->trans_lock); | |
dc9492c1 JB |
1484 | if (list_empty(&root->root_list)) { |
1485 | btrfs_grab_root(root); | |
b4be6aef JB |
1486 | |
1487 | /* We want to process the partially complete drops first. */ | |
1488 | if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) | |
1489 | list_add(&root->root_list, &fs_info->dead_roots); | |
1490 | else | |
1491 | list_add_tail(&root->root_list, &fs_info->dead_roots); | |
dc9492c1 | 1492 | } |
0b246afa | 1493 | spin_unlock(&fs_info->trans_lock); |
5eda7b5e CM |
1494 | } |
1495 | ||
d352ac68 | 1496 | /* |
dfba78dc FM |
1497 | * Update each subvolume root and its relocation root, if it exists, in the tree |
1498 | * of tree roots. Also free log roots if they exist. | |
d352ac68 | 1499 | */ |
7e4443d9 | 1500 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) |
0f7d52f4 | 1501 | { |
7e4443d9 | 1502 | struct btrfs_fs_info *fs_info = trans->fs_info; |
fc7cbcd4 DS |
1503 | struct btrfs_root *gang[8]; |
1504 | int i; | |
1505 | int ret; | |
54aa1f4d | 1506 | |
dfba78dc FM |
1507 | /* |
1508 | * At this point no one can be using this transaction to modify any tree | |
1509 | * and no one can start another transaction to modify any tree either. | |
1510 | */ | |
1511 | ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); | |
1512 | ||
fc7cbcd4 DS |
1513 | spin_lock(&fs_info->fs_roots_radix_lock); |
1514 | while (1) { | |
1515 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, | |
1516 | (void **)gang, 0, | |
1517 | ARRAY_SIZE(gang), | |
1518 | BTRFS_ROOT_TRANS_TAG); | |
1519 | if (ret == 0) | |
1520 | break; | |
1521 | for (i = 0; i < ret; i++) { | |
1522 | struct btrfs_root *root = gang[i]; | |
1523 | int ret2; | |
1524 | ||
1525 | /* | |
1526 | * At this point we can neither have tasks logging inodes | |
1527 | * from a root nor trying to commit a log tree. | |
1528 | */ | |
1529 | ASSERT(atomic_read(&root->log_writers) == 0); | |
1530 | ASSERT(atomic_read(&root->log_commit[0]) == 0); | |
1531 | ASSERT(atomic_read(&root->log_commit[1]) == 0); | |
1532 | ||
1533 | radix_tree_tag_clear(&fs_info->fs_roots_radix, | |
1534 | (unsigned long)root->root_key.objectid, | |
1535 | BTRFS_ROOT_TRANS_TAG); | |
1536 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
1537 | ||
1538 | btrfs_free_log(trans, root); | |
1539 | ret2 = btrfs_update_reloc_root(trans, root); | |
1540 | if (ret2) | |
1541 | return ret2; | |
1542 | ||
1543 | /* see comments in should_cow_block() */ | |
1544 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); | |
1545 | smp_mb__after_atomic(); | |
1546 | ||
1547 | if (root->commit_root != root->node) { | |
1548 | list_add_tail(&root->dirty_list, | |
1549 | &trans->transaction->switch_commits); | |
1550 | btrfs_set_root_node(&root->root_item, | |
1551 | root->node); | |
1552 | } | |
48b36a60 | 1553 | |
fc7cbcd4 DS |
1554 | ret2 = btrfs_update_root(trans, fs_info->tree_root, |
1555 | &root->root_key, | |
1556 | &root->root_item); | |
1557 | if (ret2) | |
1558 | return ret2; | |
1559 | spin_lock(&fs_info->fs_roots_radix_lock); | |
1560 | btrfs_qgroup_free_meta_all_pertrans(root); | |
0f7d52f4 CM |
1561 | } |
1562 | } | |
fc7cbcd4 | 1563 | spin_unlock(&fs_info->fs_roots_radix_lock); |
4f4317c1 | 1564 | return 0; |
0f7d52f4 CM |
1565 | } |
1566 | ||
6426c7ad QW |
1567 | /* |
1568 | * Do all special snapshot related qgroup dirty hack. | |
1569 | * | |
1570 | * Will do all needed qgroup inherit and dirty hack like switch commit | |
1571 | * roots inside one transaction and write all btree into disk, to make | |
1572 | * qgroup works. | |
1573 | */ | |
1574 | static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, | |
1575 | struct btrfs_root *src, | |
1576 | struct btrfs_root *parent, | |
1577 | struct btrfs_qgroup_inherit *inherit, | |
1578 | u64 dst_objectid) | |
1579 | { | |
1580 | struct btrfs_fs_info *fs_info = src->fs_info; | |
1581 | int ret; | |
1582 | ||
1583 | /* | |
5343cd93 BB |
1584 | * Save some performance in the case that qgroups are not enabled. If |
1585 | * this check races with the ioctl, rescan will kick in anyway. | |
6426c7ad | 1586 | */ |
182940f4 | 1587 | if (!btrfs_qgroup_full_accounting(fs_info)) |
6426c7ad | 1588 | return 0; |
6426c7ad | 1589 | |
4d31778a | 1590 | /* |
52042d8e | 1591 | * Ensure dirty @src will be committed. Or, after coming |
4d31778a QW |
1592 | * commit_fs_roots() and switch_commit_roots(), any dirty but not |
1593 | * recorded root will never be updated again, causing an outdated root | |
1594 | * item. | |
1595 | */ | |
1c442d22 JB |
1596 | ret = record_root_in_trans(trans, src, 1); |
1597 | if (ret) | |
1598 | return ret; | |
4d31778a | 1599 | |
2a4d84c1 JB |
1600 | /* |
1601 | * btrfs_qgroup_inherit relies on a consistent view of the usage for the | |
1602 | * src root, so we must run the delayed refs here. | |
1603 | * | |
1604 | * However this isn't particularly fool proof, because there's no | |
1605 | * synchronization keeping us from changing the tree after this point | |
1606 | * before we do the qgroup_inherit, or even from making changes while | |
1607 | * we're doing the qgroup_inherit. But that's a problem for the future, | |
1608 | * for now flush the delayed refs to narrow the race window where the | |
1609 | * qgroup counters could end up wrong. | |
1610 | */ | |
8a526c44 | 1611 | ret = btrfs_run_delayed_refs(trans, U64_MAX); |
2a4d84c1 JB |
1612 | if (ret) { |
1613 | btrfs_abort_transaction(trans, ret); | |
44365827 | 1614 | return ret; |
2a4d84c1 JB |
1615 | } |
1616 | ||
7e4443d9 | 1617 | ret = commit_fs_roots(trans); |
6426c7ad QW |
1618 | if (ret) |
1619 | goto out; | |
460fb20a | 1620 | ret = btrfs_qgroup_account_extents(trans); |
6426c7ad QW |
1621 | if (ret < 0) |
1622 | goto out; | |
1623 | ||
1624 | /* Now qgroup are all updated, we can inherit it to new qgroups */ | |
a9377422 | 1625 | ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, |
5343cd93 | 1626 | parent->root_key.objectid, inherit); |
6426c7ad QW |
1627 | if (ret < 0) |
1628 | goto out; | |
1629 | ||
1630 | /* | |
1631 | * Now we do a simplified commit transaction, which will: | |
1632 | * 1) commit all subvolume and extent tree | |
1633 | * To ensure all subvolume and extent tree have a valid | |
1634 | * commit_root to accounting later insert_dir_item() | |
1635 | * 2) write all btree blocks onto disk | |
1636 | * This is to make sure later btree modification will be cowed | |
1637 | * Or commit_root can be populated and cause wrong qgroup numbers | |
1638 | * In this simplified commit, we don't really care about other trees | |
1639 | * like chunk and root tree, as they won't affect qgroup. | |
1640 | * And we don't write super to avoid half committed status. | |
1641 | */ | |
9386d8bc | 1642 | ret = commit_cowonly_roots(trans); |
6426c7ad QW |
1643 | if (ret) |
1644 | goto out; | |
889bfa39 | 1645 | switch_commit_roots(trans); |
70458a58 | 1646 | ret = btrfs_write_and_wait_transaction(trans); |
6426c7ad | 1647 | if (ret) |
f7af3934 | 1648 | btrfs_handle_fs_error(fs_info, ret, |
6426c7ad QW |
1649 | "Error while writing out transaction for qgroup"); |
1650 | ||
1651 | out: | |
6426c7ad QW |
1652 | /* |
1653 | * Force parent root to be updated, as we recorded it before so its | |
1654 | * last_trans == cur_transid. | |
1655 | * Or it won't be committed again onto disk after later | |
1656 | * insert_dir_item() | |
1657 | */ | |
1658 | if (!ret) | |
1c442d22 | 1659 | ret = record_root_in_trans(trans, parent, 1); |
6426c7ad QW |
1660 | return ret; |
1661 | } | |
1662 | ||
d352ac68 CM |
1663 | /* |
1664 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1665 | * transaction commit. This does the actual creation. |
1666 | * | |
1667 | * Note: | |
1668 | * If the error which may affect the commitment of the current transaction | |
1669 | * happens, we should return the error number. If the error which just affect | |
1670 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1671 | */ |
80b6794d | 1672 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1673 | struct btrfs_pending_snapshot *pending) |
1674 | { | |
08d50ca3 NB |
1675 | |
1676 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3063d29f | 1677 | struct btrfs_key key; |
80b6794d | 1678 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1679 | struct btrfs_root *tree_root = fs_info->tree_root; |
1680 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1681 | struct btrfs_root *parent_root; |
98c9942a | 1682 | struct btrfs_block_rsv *rsv; |
ab3c5c18 | 1683 | struct inode *parent_inode = pending->dir; |
42874b3d MX |
1684 | struct btrfs_path *path; |
1685 | struct btrfs_dir_item *dir_item; | |
3063d29f | 1686 | struct extent_buffer *tmp; |
925baedd | 1687 | struct extent_buffer *old; |
95582b00 | 1688 | struct timespec64 cur_time; |
aec8030a | 1689 | int ret = 0; |
d68fc57b | 1690 | u64 to_reserve = 0; |
6bdb72de | 1691 | u64 index = 0; |
a22285a6 | 1692 | u64 objectid; |
b83cc969 | 1693 | u64 root_flags; |
ab3c5c18 STD |
1694 | unsigned int nofs_flags; |
1695 | struct fscrypt_name fname; | |
3063d29f | 1696 | |
8546b570 DS |
1697 | ASSERT(pending->path); |
1698 | path = pending->path; | |
42874b3d | 1699 | |
b0c0ea63 DS |
1700 | ASSERT(pending->root_item); |
1701 | new_root_item = pending->root_item; | |
a22285a6 | 1702 | |
ab3c5c18 STD |
1703 | /* |
1704 | * We're inside a transaction and must make sure that any potential | |
1705 | * allocations with GFP_KERNEL in fscrypt won't recurse back to | |
1706 | * filesystem. | |
1707 | */ | |
1708 | nofs_flags = memalloc_nofs_save(); | |
1709 | pending->error = fscrypt_setup_filename(parent_inode, | |
1710 | &pending->dentry->d_name, 0, | |
1711 | &fname); | |
1712 | memalloc_nofs_restore(nofs_flags); | |
1713 | if (pending->error) | |
1714 | goto free_pending; | |
ab3c5c18 | 1715 | |
543068a2 | 1716 | pending->error = btrfs_get_free_objectid(tree_root, &objectid); |
aec8030a | 1717 | if (pending->error) |
ab3c5c18 | 1718 | goto free_fname; |
3063d29f | 1719 | |
d6726335 QW |
1720 | /* |
1721 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1722 | * accounted by later btrfs_qgroup_inherit(). | |
1723 | */ | |
1724 | btrfs_set_skip_qgroup(trans, objectid); | |
1725 | ||
147d256e | 1726 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
d68fc57b YZ |
1727 | |
1728 | if (to_reserve > 0) { | |
9270501c | 1729 | pending->error = btrfs_block_rsv_add(fs_info, |
aec8030a MX |
1730 | &pending->block_rsv, |
1731 | to_reserve, | |
1732 | BTRFS_RESERVE_NO_FLUSH); | |
1733 | if (pending->error) | |
d6726335 | 1734 | goto clear_skip_qgroup; |
d68fc57b YZ |
1735 | } |
1736 | ||
3063d29f | 1737 | key.objectid = objectid; |
a22285a6 YZ |
1738 | key.offset = (u64)-1; |
1739 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1740 | |
6fa9700e | 1741 | rsv = trans->block_rsv; |
a22285a6 | 1742 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1743 | trans->bytes_reserved = trans->block_rsv->reserved; |
0b246afa | 1744 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa JB |
1745 | trans->transid, |
1746 | trans->bytes_reserved, 1); | |
a22285a6 | 1747 | parent_root = BTRFS_I(parent_inode)->root; |
f0118cb6 JB |
1748 | ret = record_root_in_trans(trans, parent_root, 0); |
1749 | if (ret) | |
1750 | goto fail; | |
c2050a45 | 1751 | cur_time = current_time(parent_inode); |
04b285f3 | 1752 | |
3063d29f CM |
1753 | /* |
1754 | * insert the directory item | |
1755 | */ | |
877574e2 | 1756 | ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); |
df9f2782 FM |
1757 | if (ret) { |
1758 | btrfs_abort_transaction(trans, ret); | |
1759 | goto fail; | |
1760 | } | |
42874b3d MX |
1761 | |
1762 | /* check if there is a file/dir which has the same name. */ | |
1763 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
4a0cc7ca | 1764 | btrfs_ino(BTRFS_I(parent_inode)), |
6db75318 | 1765 | &fname.disk_name, 0); |
42874b3d | 1766 | if (dir_item != NULL && !IS_ERR(dir_item)) { |
fe66a05a | 1767 | pending->error = -EEXIST; |
aec8030a | 1768 | goto dir_item_existed; |
42874b3d MX |
1769 | } else if (IS_ERR(dir_item)) { |
1770 | ret = PTR_ERR(dir_item); | |
66642832 | 1771 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1772 | goto fail; |
79787eaa | 1773 | } |
42874b3d | 1774 | btrfs_release_path(path); |
52c26179 | 1775 | |
6ed05643 BB |
1776 | ret = btrfs_create_qgroup(trans, objectid); |
1777 | if (ret) { | |
1778 | btrfs_abort_transaction(trans, ret); | |
1779 | goto fail; | |
1780 | } | |
1781 | ||
e999376f CM |
1782 | /* |
1783 | * pull in the delayed directory update | |
1784 | * and the delayed inode item | |
1785 | * otherwise we corrupt the FS during | |
1786 | * snapshot | |
1787 | */ | |
e5c304e6 | 1788 | ret = btrfs_run_delayed_items(trans); |
8732d44f | 1789 | if (ret) { /* Transaction aborted */ |
66642832 | 1790 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1791 | goto fail; |
1792 | } | |
e999376f | 1793 | |
f0118cb6 JB |
1794 | ret = record_root_in_trans(trans, root, 0); |
1795 | if (ret) { | |
1796 | btrfs_abort_transaction(trans, ret); | |
1797 | goto fail; | |
1798 | } | |
6bdb72de SW |
1799 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1800 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1801 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1802 | |
b83cc969 LZ |
1803 | root_flags = btrfs_root_flags(new_root_item); |
1804 | if (pending->readonly) | |
1805 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1806 | else | |
1807 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1808 | btrfs_set_root_flags(new_root_item, root_flags); | |
1809 | ||
8ea05e3a AB |
1810 | btrfs_set_root_generation_v2(new_root_item, |
1811 | trans->transid); | |
807fc790 | 1812 | generate_random_guid(new_root_item->uuid); |
8ea05e3a AB |
1813 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, |
1814 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1815 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1816 | memset(new_root_item->received_uuid, 0, | |
1817 | sizeof(new_root_item->received_uuid)); | |
1818 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1819 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1820 | btrfs_set_root_stransid(new_root_item, 0); | |
1821 | btrfs_set_root_rtransid(new_root_item, 0); | |
1822 | } | |
3cae210f QW |
1823 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1824 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1825 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1826 | |
6bdb72de | 1827 | old = btrfs_lock_root_node(root); |
9631e4cc JB |
1828 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, |
1829 | BTRFS_NESTING_COW); | |
79787eaa JM |
1830 | if (ret) { |
1831 | btrfs_tree_unlock(old); | |
1832 | free_extent_buffer(old); | |
66642832 | 1833 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1834 | goto fail; |
79787eaa | 1835 | } |
49b25e05 | 1836 | |
49b25e05 | 1837 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1838 | /* clean up in any case */ |
6bdb72de SW |
1839 | btrfs_tree_unlock(old); |
1840 | free_extent_buffer(old); | |
8732d44f | 1841 | if (ret) { |
66642832 | 1842 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1843 | goto fail; |
1844 | } | |
f1ebcc74 | 1845 | /* see comments in should_cow_block() */ |
27cdeb70 | 1846 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1847 | smp_wmb(); |
1848 | ||
6bdb72de | 1849 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1850 | /* record when the snapshot was created in key.offset */ |
1851 | key.offset = trans->transid; | |
1852 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1853 | btrfs_tree_unlock(tmp); |
1854 | free_extent_buffer(tmp); | |
8732d44f | 1855 | if (ret) { |
66642832 | 1856 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1857 | goto fail; |
1858 | } | |
6bdb72de | 1859 | |
a22285a6 YZ |
1860 | /* |
1861 | * insert root back/forward references | |
1862 | */ | |
6025c19f | 1863 | ret = btrfs_add_root_ref(trans, objectid, |
0660b5af | 1864 | parent_root->root_key.objectid, |
4a0cc7ca | 1865 | btrfs_ino(BTRFS_I(parent_inode)), index, |
6db75318 | 1866 | &fname.disk_name); |
8732d44f | 1867 | if (ret) { |
66642832 | 1868 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1869 | goto fail; |
1870 | } | |
0660b5af | 1871 | |
a22285a6 | 1872 | key.offset = (u64)-1; |
2dfb1e43 | 1873 | pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); |
79787eaa JM |
1874 | if (IS_ERR(pending->snap)) { |
1875 | ret = PTR_ERR(pending->snap); | |
2d892ccd | 1876 | pending->snap = NULL; |
66642832 | 1877 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1878 | goto fail; |
79787eaa | 1879 | } |
d68fc57b | 1880 | |
49b25e05 | 1881 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f | 1882 | if (ret) { |
66642832 | 1883 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1884 | goto fail; |
1885 | } | |
361048f5 | 1886 | |
6426c7ad QW |
1887 | /* |
1888 | * Do special qgroup accounting for snapshot, as we do some qgroup | |
1889 | * snapshot hack to do fast snapshot. | |
1890 | * To co-operate with that hack, we do hack again. | |
1891 | * Or snapshot will be greatly slowed down by a subtree qgroup rescan | |
1892 | */ | |
5343cd93 BB |
1893 | if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_FULL) |
1894 | ret = qgroup_account_snapshot(trans, root, parent_root, | |
1895 | pending->inherit, objectid); | |
1896 | else if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE) | |
1897 | ret = btrfs_qgroup_inherit(trans, root->root_key.objectid, objectid, | |
1898 | parent_root->root_key.objectid, pending->inherit); | |
6426c7ad QW |
1899 | if (ret < 0) |
1900 | goto fail; | |
1901 | ||
6db75318 STD |
1902 | ret = btrfs_insert_dir_item(trans, &fname.disk_name, |
1903 | BTRFS_I(parent_inode), &key, BTRFS_FT_DIR, | |
1904 | index); | |
42874b3d | 1905 | /* We have check then name at the beginning, so it is impossible. */ |
9c52057c | 1906 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f | 1907 | if (ret) { |
66642832 | 1908 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1909 | goto fail; |
1910 | } | |
42874b3d | 1911 | |
6ef06d27 | 1912 | btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + |
6db75318 | 1913 | fname.disk_name.len * 2); |
2a9462de | 1914 | parent_inode->i_mtime = inode_set_ctime_current(parent_inode); |
0a5d0dc5 | 1915 | ret = btrfs_update_inode_fallback(trans, BTRFS_I(parent_inode)); |
dd5f9615 | 1916 | if (ret) { |
66642832 | 1917 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1918 | goto fail; |
1919 | } | |
807fc790 AS |
1920 | ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, |
1921 | BTRFS_UUID_KEY_SUBVOL, | |
cdb345a8 | 1922 | objectid); |
dd5f9615 | 1923 | if (ret) { |
66642832 | 1924 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1925 | goto fail; |
1926 | } | |
1927 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
cdb345a8 | 1928 | ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, |
dd5f9615 SB |
1929 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, |
1930 | objectid); | |
1931 | if (ret && ret != -EEXIST) { | |
66642832 | 1932 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1933 | goto fail; |
1934 | } | |
1935 | } | |
d6726335 | 1936 | |
3063d29f | 1937 | fail: |
aec8030a MX |
1938 | pending->error = ret; |
1939 | dir_item_existed: | |
98c9942a | 1940 | trans->block_rsv = rsv; |
2382c5cc | 1941 | trans->bytes_reserved = 0; |
d6726335 QW |
1942 | clear_skip_qgroup: |
1943 | btrfs_clear_skip_qgroup(trans); | |
ab3c5c18 STD |
1944 | free_fname: |
1945 | fscrypt_free_filename(&fname); | |
1946 | free_pending: | |
6fa9700e | 1947 | kfree(new_root_item); |
b0c0ea63 | 1948 | pending->root_item = NULL; |
42874b3d | 1949 | btrfs_free_path(path); |
8546b570 DS |
1950 | pending->path = NULL; |
1951 | ||
49b25e05 | 1952 | return ret; |
3063d29f CM |
1953 | } |
1954 | ||
d352ac68 CM |
1955 | /* |
1956 | * create all the snapshots we've scheduled for creation | |
1957 | */ | |
08d50ca3 | 1958 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) |
3de4586c | 1959 | { |
aec8030a | 1960 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1961 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1962 | int ret = 0; |
3de4586c | 1963 | |
aec8030a MX |
1964 | list_for_each_entry_safe(pending, next, head, list) { |
1965 | list_del(&pending->list); | |
08d50ca3 | 1966 | ret = create_pending_snapshot(trans, pending); |
aec8030a MX |
1967 | if (ret) |
1968 | break; | |
1969 | } | |
1970 | return ret; | |
3de4586c CM |
1971 | } |
1972 | ||
2ff7e61e | 1973 | static void update_super_roots(struct btrfs_fs_info *fs_info) |
5d4f98a2 YZ |
1974 | { |
1975 | struct btrfs_root_item *root_item; | |
1976 | struct btrfs_super_block *super; | |
1977 | ||
0b246afa | 1978 | super = fs_info->super_copy; |
5d4f98a2 | 1979 | |
0b246afa | 1980 | root_item = &fs_info->chunk_root->root_item; |
093e037c DS |
1981 | super->chunk_root = root_item->bytenr; |
1982 | super->chunk_root_generation = root_item->generation; | |
1983 | super->chunk_root_level = root_item->level; | |
5d4f98a2 | 1984 | |
0b246afa | 1985 | root_item = &fs_info->tree_root->root_item; |
093e037c DS |
1986 | super->root = root_item->bytenr; |
1987 | super->generation = root_item->generation; | |
1988 | super->root_level = root_item->level; | |
0b246afa | 1989 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) |
093e037c | 1990 | super->cache_generation = root_item->generation; |
94846229 BB |
1991 | else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) |
1992 | super->cache_generation = 0; | |
0b246afa | 1993 | if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) |
093e037c | 1994 | super->uuid_tree_generation = root_item->generation; |
5d4f98a2 YZ |
1995 | } |
1996 | ||
f36f3042 CM |
1997 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1998 | { | |
4a9d8bde | 1999 | struct btrfs_transaction *trans; |
f36f3042 | 2000 | int ret = 0; |
4a9d8bde | 2001 | |
a4abeea4 | 2002 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
2003 | trans = info->running_transaction; |
2004 | if (trans) | |
2005 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 2006 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
2007 | return ret; |
2008 | } | |
2009 | ||
8929ecfa YZ |
2010 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
2011 | { | |
4a9d8bde | 2012 | struct btrfs_transaction *trans; |
8929ecfa | 2013 | int ret = 0; |
4a9d8bde | 2014 | |
a4abeea4 | 2015 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
2016 | trans = info->running_transaction; |
2017 | if (trans) | |
2018 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 2019 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
2020 | return ret; |
2021 | } | |
2022 | ||
fdfbf020 | 2023 | void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans) |
bb9c12c9 | 2024 | { |
3a45bb20 | 2025 | struct btrfs_fs_info *fs_info = trans->fs_info; |
bb9c12c9 SW |
2026 | struct btrfs_transaction *cur_trans; |
2027 | ||
fdfbf020 JB |
2028 | /* Kick the transaction kthread. */ |
2029 | set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); | |
2030 | wake_up_process(fs_info->transaction_kthread); | |
bb9c12c9 SW |
2031 | |
2032 | /* take transaction reference */ | |
bb9c12c9 | 2033 | cur_trans = trans->transaction; |
9b64f57d | 2034 | refcount_inc(&cur_trans->use_count); |
bb9c12c9 | 2035 | |
3a45bb20 | 2036 | btrfs_end_transaction(trans); |
6fc4e354 | 2037 | |
ae5d29d4 DS |
2038 | /* |
2039 | * Wait for the current transaction commit to start and block | |
2040 | * subsequent transaction joins | |
2041 | */ | |
77d20c68 | 2042 | btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP); |
ae5d29d4 DS |
2043 | wait_event(fs_info->transaction_blocked_wait, |
2044 | cur_trans->state >= TRANS_STATE_COMMIT_START || | |
2045 | TRANS_ABORTED(cur_trans)); | |
724e2315 | 2046 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
2047 | } |
2048 | ||
97cb39bb | 2049 | static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) |
49b25e05 | 2050 | { |
97cb39bb | 2051 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 JM |
2052 | struct btrfs_transaction *cur_trans = trans->transaction; |
2053 | ||
b50fff81 | 2054 | WARN_ON(refcount_read(&trans->use_count) > 1); |
49b25e05 | 2055 | |
66642832 | 2056 | btrfs_abort_transaction(trans, err); |
7b8b92af | 2057 | |
0b246afa | 2058 | spin_lock(&fs_info->trans_lock); |
66b6135b | 2059 | |
25d8c284 MX |
2060 | /* |
2061 | * If the transaction is removed from the list, it means this | |
2062 | * transaction has been committed successfully, so it is impossible | |
2063 | * to call the cleanup function. | |
2064 | */ | |
2065 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 2066 | |
0b246afa | 2067 | if (cur_trans == fs_info->running_transaction) { |
4a9d8bde | 2068 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 2069 | spin_unlock(&fs_info->trans_lock); |
e1489b4f IA |
2070 | |
2071 | /* | |
2072 | * The thread has already released the lockdep map as reader | |
2073 | * already in btrfs_commit_transaction(). | |
2074 | */ | |
2075 | btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); | |
f094ac32 LB |
2076 | wait_event(cur_trans->writer_wait, |
2077 | atomic_read(&cur_trans->num_writers) == 1); | |
2078 | ||
0b246afa | 2079 | spin_lock(&fs_info->trans_lock); |
d7096fc3 | 2080 | } |
061dde82 FM |
2081 | |
2082 | /* | |
2083 | * Now that we know no one else is still using the transaction we can | |
2084 | * remove the transaction from the list of transactions. This avoids | |
2085 | * the transaction kthread from cleaning up the transaction while some | |
2086 | * other task is still using it, which could result in a use-after-free | |
2087 | * on things like log trees, as it forces the transaction kthread to | |
2088 | * wait for this transaction to be cleaned up by us. | |
2089 | */ | |
2090 | list_del_init(&cur_trans->list); | |
2091 | ||
0b246afa | 2092 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 2093 | |
2ff7e61e | 2094 | btrfs_cleanup_one_transaction(trans->transaction, fs_info); |
49b25e05 | 2095 | |
0b246afa JM |
2096 | spin_lock(&fs_info->trans_lock); |
2097 | if (cur_trans == fs_info->running_transaction) | |
2098 | fs_info->running_transaction = NULL; | |
2099 | spin_unlock(&fs_info->trans_lock); | |
4a9d8bde | 2100 | |
e0228285 | 2101 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2102 | sb_end_intwrite(fs_info->sb); |
724e2315 JB |
2103 | btrfs_put_transaction(cur_trans); |
2104 | btrfs_put_transaction(cur_trans); | |
49b25e05 | 2105 | |
2e4e97ab | 2106 | trace_btrfs_transaction_commit(fs_info); |
49b25e05 | 2107 | |
49b25e05 JM |
2108 | if (current->journal_info == trans) |
2109 | current->journal_info = NULL; | |
2d82a40a FM |
2110 | |
2111 | /* | |
2112 | * If relocation is running, we can't cancel scrub because that will | |
2113 | * result in a deadlock. Before relocating a block group, relocation | |
2114 | * pauses scrub, then starts and commits a transaction before unpausing | |
2115 | * scrub. If the transaction commit is being done by the relocation | |
2116 | * task or triggered by another task and the relocation task is waiting | |
2117 | * for the commit, and we end up here due to an error in the commit | |
2118 | * path, then calling btrfs_scrub_cancel() will deadlock, as we are | |
2119 | * asking for scrub to stop while having it asked to be paused higher | |
2120 | * above in relocation code. | |
2121 | */ | |
2122 | if (!test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) | |
2123 | btrfs_scrub_cancel(fs_info); | |
49b25e05 JM |
2124 | |
2125 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
2126 | } | |
2127 | ||
c7cc64a9 DS |
2128 | /* |
2129 | * Release reserved delayed ref space of all pending block groups of the | |
2130 | * transaction and remove them from the list | |
2131 | */ | |
2132 | static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) | |
2133 | { | |
2134 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2135 | struct btrfs_block_group *block_group, *tmp; |
c7cc64a9 DS |
2136 | |
2137 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { | |
9ef17228 | 2138 | btrfs_dec_delayed_refs_rsv_bg_inserts(fs_info); |
c7cc64a9 DS |
2139 | list_del_init(&block_group->bg_list); |
2140 | } | |
2141 | } | |
2142 | ||
88090ad3 | 2143 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
82436617 | 2144 | { |
ce8ea7cc | 2145 | /* |
a0f0cf83 | 2146 | * We use try_to_writeback_inodes_sb() here because if we used |
ce8ea7cc JB |
2147 | * btrfs_start_delalloc_roots we would deadlock with fs freeze. |
2148 | * Currently are holding the fs freeze lock, if we do an async flush | |
2149 | * we'll do btrfs_join_transaction() and deadlock because we need to | |
2150 | * wait for the fs freeze lock. Using the direct flushing we benefit | |
2151 | * from already being in a transaction and our join_transaction doesn't | |
2152 | * have to re-take the fs freeze lock. | |
a0f0cf83 FM |
2153 | * |
2154 | * Note that try_to_writeback_inodes_sb() will only trigger writeback | |
2155 | * if it can read lock sb->s_umount. It will always be able to lock it, | |
2156 | * except when the filesystem is being unmounted or being frozen, but in | |
2157 | * those cases sync_filesystem() is called, which results in calling | |
2158 | * writeback_inodes_sb() while holding a write lock on sb->s_umount. | |
2159 | * Note that we don't call writeback_inodes_sb() directly, because it | |
2160 | * will emit a warning if sb->s_umount is not locked. | |
ce8ea7cc | 2161 | */ |
88090ad3 | 2162 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
a0f0cf83 | 2163 | try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); |
82436617 MX |
2164 | return 0; |
2165 | } | |
2166 | ||
88090ad3 | 2167 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) |
82436617 | 2168 | { |
88090ad3 | 2169 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
6374e57a | 2170 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); |
82436617 MX |
2171 | } |
2172 | ||
28b21c55 FM |
2173 | /* |
2174 | * Add a pending snapshot associated with the given transaction handle to the | |
2175 | * respective handle. This must be called after the transaction commit started | |
2176 | * and while holding fs_info->trans_lock. | |
2177 | * This serves to guarantee a caller of btrfs_commit_transaction() that it can | |
2178 | * safely free the pending snapshot pointer in case btrfs_commit_transaction() | |
2179 | * returns an error. | |
2180 | */ | |
2181 | static void add_pending_snapshot(struct btrfs_trans_handle *trans) | |
2182 | { | |
2183 | struct btrfs_transaction *cur_trans = trans->transaction; | |
2184 | ||
2185 | if (!trans->pending_snapshot) | |
2186 | return; | |
2187 | ||
2188 | lockdep_assert_held(&trans->fs_info->trans_lock); | |
77d20c68 | 2189 | ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_PREP); |
28b21c55 FM |
2190 | |
2191 | list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots); | |
2192 | } | |
2193 | ||
e55958c8 IA |
2194 | static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval) |
2195 | { | |
2196 | fs_info->commit_stats.commit_count++; | |
2197 | fs_info->commit_stats.last_commit_dur = interval; | |
2198 | fs_info->commit_stats.max_commit_dur = | |
2199 | max_t(u64, fs_info->commit_stats.max_commit_dur, interval); | |
2200 | fs_info->commit_stats.total_commit_dur += interval; | |
2201 | } | |
2202 | ||
3a45bb20 | 2203 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans) |
79154b1b | 2204 | { |
3a45bb20 | 2205 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 | 2206 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 2207 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 2208 | int ret; |
e55958c8 IA |
2209 | ktime_t start_time; |
2210 | ktime_t interval; | |
79154b1b | 2211 | |
35b814f3 | 2212 | ASSERT(refcount_read(&trans->use_count) == 1); |
77d20c68 | 2213 | btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP); |
35b814f3 | 2214 | |
c52cc7b7 JB |
2215 | clear_bit(BTRFS_FS_NEED_TRANS_COMMIT, &fs_info->flags); |
2216 | ||
8d25a086 | 2217 | /* Stop the commit early if ->aborted is set */ |
bf31f87f | 2218 | if (TRANS_ABORTED(cur_trans)) { |
25287e0a | 2219 | ret = cur_trans->aborted; |
3e738c53 | 2220 | goto lockdep_trans_commit_start_release; |
25287e0a | 2221 | } |
49b25e05 | 2222 | |
f45c752b JB |
2223 | btrfs_trans_release_metadata(trans); |
2224 | trans->block_rsv = NULL; | |
2225 | ||
56bec294 | 2226 | /* |
e19eb11f JB |
2227 | * We only want one transaction commit doing the flushing so we do not |
2228 | * waste a bunch of time on lock contention on the extent root node. | |
56bec294 | 2229 | */ |
e19eb11f JB |
2230 | if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, |
2231 | &cur_trans->delayed_refs.flags)) { | |
2232 | /* | |
2233 | * Make a pass through all the delayed refs we have so far. | |
2234 | * Any running threads may add more while we are here. | |
2235 | */ | |
2236 | ret = btrfs_run_delayed_refs(trans, 0); | |
3e738c53 IA |
2237 | if (ret) |
2238 | goto lockdep_trans_commit_start_release; | |
e19eb11f | 2239 | } |
56bec294 | 2240 | |
119e80df | 2241 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2242 | |
3204d33c | 2243 | if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { |
1bbc621e CM |
2244 | int run_it = 0; |
2245 | ||
2246 | /* this mutex is also taken before trying to set | |
2247 | * block groups readonly. We need to make sure | |
2248 | * that nobody has set a block group readonly | |
2249 | * after a extents from that block group have been | |
2250 | * allocated for cache files. btrfs_set_block_group_ro | |
2251 | * will wait for the transaction to commit if it | |
3204d33c | 2252 | * finds BTRFS_TRANS_DIRTY_BG_RUN set. |
1bbc621e | 2253 | * |
3204d33c JB |
2254 | * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure |
2255 | * only one process starts all the block group IO. It wouldn't | |
1bbc621e CM |
2256 | * hurt to have more than one go through, but there's no |
2257 | * real advantage to it either. | |
2258 | */ | |
0b246afa | 2259 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c JB |
2260 | if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, |
2261 | &cur_trans->flags)) | |
1bbc621e | 2262 | run_it = 1; |
0b246afa | 2263 | mutex_unlock(&fs_info->ro_block_group_mutex); |
1bbc621e | 2264 | |
f9cacae3 | 2265 | if (run_it) { |
21217054 | 2266 | ret = btrfs_start_dirty_block_groups(trans); |
3e738c53 IA |
2267 | if (ret) |
2268 | goto lockdep_trans_commit_start_release; | |
f9cacae3 | 2269 | } |
1bbc621e CM |
2270 | } |
2271 | ||
0b246afa | 2272 | spin_lock(&fs_info->trans_lock); |
77d20c68 | 2273 | if (cur_trans->state >= TRANS_STATE_COMMIT_PREP) { |
d0c2f4fa FM |
2274 | enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; |
2275 | ||
28b21c55 FM |
2276 | add_pending_snapshot(trans); |
2277 | ||
0b246afa | 2278 | spin_unlock(&fs_info->trans_lock); |
9b64f57d | 2279 | refcount_inc(&cur_trans->use_count); |
ccd467d6 | 2280 | |
d0c2f4fa FM |
2281 | if (trans->in_fsync) |
2282 | want_state = TRANS_STATE_SUPER_COMMITTED; | |
3e738c53 IA |
2283 | |
2284 | btrfs_trans_state_lockdep_release(fs_info, | |
77d20c68 | 2285 | BTRFS_LOCKDEP_TRANS_COMMIT_PREP); |
d0c2f4fa FM |
2286 | ret = btrfs_end_transaction(trans); |
2287 | wait_for_commit(cur_trans, want_state); | |
15ee9bc7 | 2288 | |
bf31f87f | 2289 | if (TRANS_ABORTED(cur_trans)) |
b4924a0f LB |
2290 | ret = cur_trans->aborted; |
2291 | ||
724e2315 | 2292 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 2293 | |
49b25e05 | 2294 | return ret; |
79154b1b | 2295 | } |
4313b399 | 2296 | |
77d20c68 | 2297 | cur_trans->state = TRANS_STATE_COMMIT_PREP; |
0b246afa | 2298 | wake_up(&fs_info->transaction_blocked_wait); |
77d20c68 | 2299 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP); |
bb9c12c9 | 2300 | |
0b246afa | 2301 | if (cur_trans->list.prev != &fs_info->trans_list) { |
d0c2f4fa FM |
2302 | enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; |
2303 | ||
2304 | if (trans->in_fsync) | |
2305 | want_state = TRANS_STATE_SUPER_COMMITTED; | |
2306 | ||
ccd467d6 CM |
2307 | prev_trans = list_entry(cur_trans->list.prev, |
2308 | struct btrfs_transaction, list); | |
d0c2f4fa | 2309 | if (prev_trans->state < want_state) { |
9b64f57d | 2310 | refcount_inc(&prev_trans->use_count); |
0b246afa | 2311 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2312 | |
d0c2f4fa FM |
2313 | wait_for_commit(prev_trans, want_state); |
2314 | ||
bf31f87f | 2315 | ret = READ_ONCE(prev_trans->aborted); |
ccd467d6 | 2316 | |
724e2315 | 2317 | btrfs_put_transaction(prev_trans); |
1f9b8c8f | 2318 | if (ret) |
e1489b4f | 2319 | goto lockdep_release; |
77d20c68 | 2320 | spin_lock(&fs_info->trans_lock); |
ccd467d6 | 2321 | } |
a4abeea4 | 2322 | } else { |
cb2d3dad FM |
2323 | /* |
2324 | * The previous transaction was aborted and was already removed | |
2325 | * from the list of transactions at fs_info->trans_list. So we | |
2326 | * abort to prevent writing a new superblock that reflects a | |
2327 | * corrupt state (pointing to trees with unwritten nodes/leafs). | |
2328 | */ | |
84961539 | 2329 | if (BTRFS_FS_ERROR(fs_info)) { |
77d20c68 | 2330 | spin_unlock(&fs_info->trans_lock); |
cb2d3dad | 2331 | ret = -EROFS; |
e1489b4f | 2332 | goto lockdep_release; |
cb2d3dad | 2333 | } |
ccd467d6 | 2334 | } |
15ee9bc7 | 2335 | |
77d20c68 JB |
2336 | cur_trans->state = TRANS_STATE_COMMIT_START; |
2337 | wake_up(&fs_info->transaction_blocked_wait); | |
2338 | spin_unlock(&fs_info->trans_lock); | |
2339 | ||
e55958c8 IA |
2340 | /* |
2341 | * Get the time spent on the work done by the commit thread and not | |
2342 | * the time spent waiting on a previous commit | |
2343 | */ | |
2344 | start_time = ktime_get_ns(); | |
2345 | ||
0860adfd MX |
2346 | extwriter_counter_dec(cur_trans, trans->type); |
2347 | ||
88090ad3 | 2348 | ret = btrfs_start_delalloc_flush(fs_info); |
82436617 | 2349 | if (ret) |
e1489b4f | 2350 | goto lockdep_release; |
82436617 | 2351 | |
e5c304e6 | 2352 | ret = btrfs_run_delayed_items(trans); |
581227d0 | 2353 | if (ret) |
e1489b4f | 2354 | goto lockdep_release; |
15ee9bc7 | 2355 | |
5a9ba670 IA |
2356 | /* |
2357 | * The thread has started/joined the transaction thus it holds the | |
2358 | * lockdep map as a reader. It has to release it before acquiring the | |
2359 | * lockdep map as a writer. | |
2360 | */ | |
2361 | btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); | |
2362 | btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters); | |
581227d0 MX |
2363 | wait_event(cur_trans->writer_wait, |
2364 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 2365 | |
581227d0 | 2366 | /* some pending stuffs might be added after the previous flush. */ |
e5c304e6 | 2367 | ret = btrfs_run_delayed_items(trans); |
e1489b4f IA |
2368 | if (ret) { |
2369 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); | |
ca469637 | 2370 | goto cleanup_transaction; |
e1489b4f | 2371 | } |
ca469637 | 2372 | |
88090ad3 | 2373 | btrfs_wait_delalloc_flush(fs_info); |
cb7ab021 | 2374 | |
48778179 FM |
2375 | /* |
2376 | * Wait for all ordered extents started by a fast fsync that joined this | |
2377 | * transaction. Otherwise if this transaction commits before the ordered | |
2378 | * extents complete we lose logged data after a power failure. | |
2379 | */ | |
8b53779e | 2380 | btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered); |
48778179 FM |
2381 | wait_event(cur_trans->pending_wait, |
2382 | atomic_read(&cur_trans->pending_ordered) == 0); | |
2383 | ||
2ff7e61e | 2384 | btrfs_scrub_pause(fs_info); |
ed0ca140 JB |
2385 | /* |
2386 | * Ok now we need to make sure to block out any other joins while we | |
2387 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 2388 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 | 2389 | */ |
0b246afa | 2390 | spin_lock(&fs_info->trans_lock); |
28b21c55 | 2391 | add_pending_snapshot(trans); |
4a9d8bde | 2392 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 2393 | spin_unlock(&fs_info->trans_lock); |
e1489b4f IA |
2394 | |
2395 | /* | |
2396 | * The thread has started/joined the transaction thus it holds the | |
2397 | * lockdep map as a reader. It has to release it before acquiring the | |
2398 | * lockdep map as a writer. | |
2399 | */ | |
2400 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); | |
2401 | btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); | |
ed0ca140 JB |
2402 | wait_event(cur_trans->writer_wait, |
2403 | atomic_read(&cur_trans->num_writers) == 1); | |
2404 | ||
3e738c53 IA |
2405 | /* |
2406 | * Make lockdep happy by acquiring the state locks after | |
2407 | * btrfs_trans_num_writers is released. If we acquired the state locks | |
2408 | * before releasing the btrfs_trans_num_writers lock then lockdep would | |
2409 | * complain because we did not follow the reverse order unlocking rule. | |
2410 | */ | |
2411 | btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); | |
2412 | btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); | |
2413 | btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); | |
2414 | ||
fdfbf020 JB |
2415 | /* |
2416 | * We've started the commit, clear the flag in case we were triggered to | |
2417 | * do an async commit but somebody else started before the transaction | |
2418 | * kthread could do the work. | |
2419 | */ | |
2420 | clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); | |
2421 | ||
bf31f87f | 2422 | if (TRANS_ABORTED(cur_trans)) { |
2cba30f1 | 2423 | ret = cur_trans->aborted; |
3e738c53 | 2424 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); |
6cf7f77e | 2425 | goto scrub_continue; |
2cba30f1 | 2426 | } |
7585717f CM |
2427 | /* |
2428 | * the reloc mutex makes sure that we stop | |
2429 | * the balancing code from coming in and moving | |
2430 | * extents around in the middle of the commit | |
2431 | */ | |
0b246afa | 2432 | mutex_lock(&fs_info->reloc_mutex); |
7585717f | 2433 | |
42874b3d MX |
2434 | /* |
2435 | * We needn't worry about the delayed items because we will | |
2436 | * deal with them in create_pending_snapshot(), which is the | |
2437 | * core function of the snapshot creation. | |
2438 | */ | |
08d50ca3 | 2439 | ret = create_pending_snapshots(trans); |
56e9f6ea DS |
2440 | if (ret) |
2441 | goto unlock_reloc; | |
3063d29f | 2442 | |
42874b3d MX |
2443 | /* |
2444 | * We insert the dir indexes of the snapshots and update the inode | |
2445 | * of the snapshots' parents after the snapshot creation, so there | |
2446 | * are some delayed items which are not dealt with. Now deal with | |
2447 | * them. | |
2448 | * | |
2449 | * We needn't worry that this operation will corrupt the snapshots, | |
2450 | * because all the tree which are snapshoted will be forced to COW | |
2451 | * the nodes and leaves. | |
2452 | */ | |
e5c304e6 | 2453 | ret = btrfs_run_delayed_items(trans); |
56e9f6ea DS |
2454 | if (ret) |
2455 | goto unlock_reloc; | |
16cdcec7 | 2456 | |
8a526c44 | 2457 | ret = btrfs_run_delayed_refs(trans, U64_MAX); |
56e9f6ea DS |
2458 | if (ret) |
2459 | goto unlock_reloc; | |
56bec294 | 2460 | |
e999376f CM |
2461 | /* |
2462 | * make sure none of the code above managed to slip in a | |
2463 | * delayed item | |
2464 | */ | |
ccdf9b30 | 2465 | btrfs_assert_delayed_root_empty(fs_info); |
e999376f | 2466 | |
2c90e5d6 | 2467 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 2468 | |
7e4443d9 | 2469 | ret = commit_fs_roots(trans); |
56e9f6ea | 2470 | if (ret) |
dfba78dc | 2471 | goto unlock_reloc; |
54aa1f4d | 2472 | |
5d4f98a2 | 2473 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2474 | * safe to free the root of tree log roots |
2475 | */ | |
0b246afa | 2476 | btrfs_free_log_root_tree(trans, fs_info); |
e02119d5 | 2477 | |
0ed4792a QW |
2478 | /* |
2479 | * Since fs roots are all committed, we can get a quite accurate | |
2480 | * new_roots. So let's do quota accounting. | |
2481 | */ | |
460fb20a | 2482 | ret = btrfs_qgroup_account_extents(trans); |
56e9f6ea | 2483 | if (ret < 0) |
dfba78dc | 2484 | goto unlock_reloc; |
0ed4792a | 2485 | |
9386d8bc | 2486 | ret = commit_cowonly_roots(trans); |
56e9f6ea | 2487 | if (ret) |
dfba78dc | 2488 | goto unlock_reloc; |
54aa1f4d | 2489 | |
2cba30f1 MX |
2490 | /* |
2491 | * The tasks which save the space cache and inode cache may also | |
2492 | * update ->aborted, check it. | |
2493 | */ | |
bf31f87f | 2494 | if (TRANS_ABORTED(cur_trans)) { |
2cba30f1 | 2495 | ret = cur_trans->aborted; |
dfba78dc | 2496 | goto unlock_reloc; |
2cba30f1 MX |
2497 | } |
2498 | ||
0b246afa | 2499 | cur_trans = fs_info->running_transaction; |
5d4f98a2 | 2500 | |
0b246afa JM |
2501 | btrfs_set_root_node(&fs_info->tree_root->root_item, |
2502 | fs_info->tree_root->node); | |
2503 | list_add_tail(&fs_info->tree_root->dirty_list, | |
9e351cc8 | 2504 | &cur_trans->switch_commits); |
5d4f98a2 | 2505 | |
0b246afa JM |
2506 | btrfs_set_root_node(&fs_info->chunk_root->root_item, |
2507 | fs_info->chunk_root->node); | |
2508 | list_add_tail(&fs_info->chunk_root->dirty_list, | |
9e351cc8 JB |
2509 | &cur_trans->switch_commits); |
2510 | ||
f7238e50 JB |
2511 | if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { |
2512 | btrfs_set_root_node(&fs_info->block_group_root->root_item, | |
2513 | fs_info->block_group_root->node); | |
2514 | list_add_tail(&fs_info->block_group_root->dirty_list, | |
2515 | &cur_trans->switch_commits); | |
2516 | } | |
2517 | ||
889bfa39 | 2518 | switch_commit_roots(trans); |
5d4f98a2 | 2519 | |
ce93ec54 | 2520 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2521 | ASSERT(list_empty(&cur_trans->io_bgs)); |
2ff7e61e | 2522 | update_super_roots(fs_info); |
e02119d5 | 2523 | |
0b246afa JM |
2524 | btrfs_set_super_log_root(fs_info->super_copy, 0); |
2525 | btrfs_set_super_log_root_level(fs_info->super_copy, 0); | |
2526 | memcpy(fs_info->super_for_commit, fs_info->super_copy, | |
2527 | sizeof(*fs_info->super_copy)); | |
ccd467d6 | 2528 | |
bbbf7243 | 2529 | btrfs_commit_device_sizes(cur_trans); |
935e5cc9 | 2530 | |
0b246afa JM |
2531 | clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
2532 | clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); | |
656f30db | 2533 | |
4fbcdf66 FM |
2534 | btrfs_trans_release_chunk_metadata(trans); |
2535 | ||
dfba78dc FM |
2536 | /* |
2537 | * Before changing the transaction state to TRANS_STATE_UNBLOCKED and | |
2538 | * setting fs_info->running_transaction to NULL, lock tree_log_mutex to | |
2539 | * make sure that before we commit our superblock, no other task can | |
2540 | * start a new transaction and commit a log tree before we commit our | |
2541 | * superblock. Anyone trying to commit a log tree locks this mutex before | |
2542 | * writing its superblock. | |
2543 | */ | |
2544 | mutex_lock(&fs_info->tree_log_mutex); | |
2545 | ||
0b246afa | 2546 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2547 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
0b246afa JM |
2548 | fs_info->running_transaction = NULL; |
2549 | spin_unlock(&fs_info->trans_lock); | |
2550 | mutex_unlock(&fs_info->reloc_mutex); | |
b7ec40d7 | 2551 | |
0b246afa | 2552 | wake_up(&fs_info->transaction_wait); |
3e738c53 | 2553 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); |
e6dcd2dc | 2554 | |
b7625f46 QW |
2555 | /* If we have features changed, wake up the cleaner to update sysfs. */ |
2556 | if (test_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags) && | |
2557 | fs_info->cleaner_kthread) | |
2558 | wake_up_process(fs_info->cleaner_kthread); | |
2559 | ||
70458a58 | 2560 | ret = btrfs_write_and_wait_transaction(trans); |
49b25e05 | 2561 | if (ret) { |
0b246afa JM |
2562 | btrfs_handle_fs_error(fs_info, ret, |
2563 | "Error while writing out transaction"); | |
2564 | mutex_unlock(&fs_info->tree_log_mutex); | |
6cf7f77e | 2565 | goto scrub_continue; |
49b25e05 JM |
2566 | } |
2567 | ||
eece6a9c | 2568 | ret = write_all_supers(fs_info, 0); |
e02119d5 CM |
2569 | /* |
2570 | * the super is written, we can safely allow the tree-loggers | |
2571 | * to go about their business | |
2572 | */ | |
0b246afa | 2573 | mutex_unlock(&fs_info->tree_log_mutex); |
c1f32b7c AJ |
2574 | if (ret) |
2575 | goto scrub_continue; | |
e02119d5 | 2576 | |
d0c2f4fa FM |
2577 | /* |
2578 | * We needn't acquire the lock here because there is no other task | |
2579 | * which can change it. | |
2580 | */ | |
2581 | cur_trans->state = TRANS_STATE_SUPER_COMMITTED; | |
2582 | wake_up(&cur_trans->commit_wait); | |
3e738c53 | 2583 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); |
d0c2f4fa | 2584 | |
5ead2dd0 | 2585 | btrfs_finish_extent_commit(trans); |
4313b399 | 2586 | |
3204d33c | 2587 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) |
0b246afa | 2588 | btrfs_clear_space_info_full(fs_info); |
13212b54 | 2589 | |
0124855f | 2590 | btrfs_set_last_trans_committed(fs_info, cur_trans->transid); |
4a9d8bde MX |
2591 | /* |
2592 | * We needn't acquire the lock here because there is no other task | |
2593 | * which can change it. | |
2594 | */ | |
2595 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2596 | wake_up(&cur_trans->commit_wait); |
3e738c53 | 2597 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); |
3de4586c | 2598 | |
0b246afa | 2599 | spin_lock(&fs_info->trans_lock); |
13c5a93e | 2600 | list_del_init(&cur_trans->list); |
0b246afa | 2601 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 2602 | |
724e2315 JB |
2603 | btrfs_put_transaction(cur_trans); |
2604 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2605 | |
0860adfd | 2606 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2607 | sb_end_intwrite(fs_info->sb); |
b2b5ef5c | 2608 | |
2e4e97ab | 2609 | trace_btrfs_transaction_commit(fs_info); |
1abe9b8a | 2610 | |
e55958c8 IA |
2611 | interval = ktime_get_ns() - start_time; |
2612 | ||
2ff7e61e | 2613 | btrfs_scrub_continue(fs_info); |
a2de733c | 2614 | |
9ed74f2d JB |
2615 | if (current->journal_info == trans) |
2616 | current->journal_info = NULL; | |
2617 | ||
2c90e5d6 | 2618 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2619 | |
e55958c8 IA |
2620 | update_commit_stats(fs_info, interval); |
2621 | ||
79154b1b | 2622 | return ret; |
49b25e05 | 2623 | |
56e9f6ea DS |
2624 | unlock_reloc: |
2625 | mutex_unlock(&fs_info->reloc_mutex); | |
3e738c53 | 2626 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); |
6cf7f77e | 2627 | scrub_continue: |
3e738c53 IA |
2628 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); |
2629 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); | |
2ff7e61e | 2630 | btrfs_scrub_continue(fs_info); |
49b25e05 | 2631 | cleanup_transaction: |
dc60c525 | 2632 | btrfs_trans_release_metadata(trans); |
c7cc64a9 | 2633 | btrfs_cleanup_pending_block_groups(trans); |
4fbcdf66 | 2634 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2635 | trans->block_rsv = NULL; |
0b246afa | 2636 | btrfs_warn(fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2637 | if (current->journal_info == trans) |
2638 | current->journal_info = NULL; | |
97cb39bb | 2639 | cleanup_transaction(trans, ret); |
49b25e05 JM |
2640 | |
2641 | return ret; | |
e1489b4f IA |
2642 | |
2643 | lockdep_release: | |
5a9ba670 | 2644 | btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); |
e1489b4f IA |
2645 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); |
2646 | goto cleanup_transaction; | |
3e738c53 IA |
2647 | |
2648 | lockdep_trans_commit_start_release: | |
77d20c68 | 2649 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP); |
3e738c53 IA |
2650 | btrfs_end_transaction(trans); |
2651 | return ret; | |
79154b1b CM |
2652 | } |
2653 | ||
d352ac68 | 2654 | /* |
9d1a2a3a DS |
2655 | * return < 0 if error |
2656 | * 0 if there are no more dead_roots at the time of call | |
2657 | * 1 there are more to be processed, call me again | |
2658 | * | |
2659 | * The return value indicates there are certainly more snapshots to delete, but | |
2660 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2661 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2662 | * few seconds later. | |
d352ac68 | 2663 | */ |
33c44184 | 2664 | int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info) |
e9d0b13b | 2665 | { |
33c44184 | 2666 | struct btrfs_root *root; |
9d1a2a3a | 2667 | int ret; |
5d4f98a2 | 2668 | |
a4abeea4 | 2669 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2670 | if (list_empty(&fs_info->dead_roots)) { |
2671 | spin_unlock(&fs_info->trans_lock); | |
2672 | return 0; | |
2673 | } | |
2674 | root = list_first_entry(&fs_info->dead_roots, | |
2675 | struct btrfs_root, root_list); | |
cfad392b | 2676 | list_del_init(&root->root_list); |
a4abeea4 | 2677 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2678 | |
4fd786e6 | 2679 | btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); |
76dda93c | 2680 | |
9d1a2a3a | 2681 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2682 | |
9d1a2a3a DS |
2683 | if (btrfs_header_backref_rev(root->node) < |
2684 | BTRFS_MIXED_BACKREF_REV) | |
0078a9f9 | 2685 | ret = btrfs_drop_snapshot(root, 0, 0); |
9d1a2a3a | 2686 | else |
0078a9f9 | 2687 | ret = btrfs_drop_snapshot(root, 1, 0); |
32471dc2 | 2688 | |
dc9492c1 | 2689 | btrfs_put_root(root); |
6596a928 | 2690 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2691 | } |
572d9ab7 | 2692 | |
fccf0c84 JB |
2693 | /* |
2694 | * We only mark the transaction aborted and then set the file system read-only. | |
2695 | * This will prevent new transactions from starting or trying to join this | |
2696 | * one. | |
2697 | * | |
2698 | * This means that error recovery at the call site is limited to freeing | |
2699 | * any local memory allocations and passing the error code up without | |
2700 | * further cleanup. The transaction should complete as it normally would | |
2701 | * in the call path but will return -EIO. | |
2702 | * | |
2703 | * We'll complete the cleanup in btrfs_end_transaction and | |
2704 | * btrfs_commit_transaction. | |
2705 | */ | |
2706 | void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans, | |
2707 | const char *function, | |
ed164802 | 2708 | unsigned int line, int error, bool first_hit) |
fccf0c84 JB |
2709 | { |
2710 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2711 | ||
ed164802 DS |
2712 | WRITE_ONCE(trans->aborted, error); |
2713 | WRITE_ONCE(trans->transaction->aborted, error); | |
2714 | if (first_hit && error == -ENOSPC) | |
fccf0c84 JB |
2715 | btrfs_dump_space_info_for_trans_abort(fs_info); |
2716 | /* Wake up anybody who may be waiting on this transaction */ | |
2717 | wake_up(&fs_info->transaction_wait); | |
2718 | wake_up(&fs_info->transaction_blocked_wait); | |
ed164802 | 2719 | __btrfs_handle_fs_error(fs_info, function, line, error, NULL); |
fccf0c84 JB |
2720 | } |
2721 | ||
956504a3 JB |
2722 | int __init btrfs_transaction_init(void) |
2723 | { | |
2724 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", | |
2725 | sizeof(struct btrfs_trans_handle), 0, | |
2726 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); | |
2727 | if (!btrfs_trans_handle_cachep) | |
2728 | return -ENOMEM; | |
2729 | return 0; | |
2730 | } | |
2731 | ||
2732 | void __cold btrfs_transaction_exit(void) | |
2733 | { | |
2734 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
2735 | } |