Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 CM |
4 | */ |
5 | ||
79154b1b | 6 | #include <linux/fs.h> |
5a0e3ad6 | 7 | #include <linux/slab.h> |
34088780 | 8 | #include <linux/sched.h> |
d3c2fdcf | 9 | #include <linux/writeback.h> |
5f39d397 | 10 | #include <linux/pagemap.h> |
5f2cc086 | 11 | #include <linux/blkdev.h> |
8ea05e3a | 12 | #include <linux/uuid.h> |
e55958c8 | 13 | #include <linux/timekeeping.h> |
602cbe91 | 14 | #include "misc.h" |
79154b1b CM |
15 | #include "ctree.h" |
16 | #include "disk-io.h" | |
17 | #include "transaction.h" | |
925baedd | 18 | #include "locking.h" |
e02119d5 | 19 | #include "tree-log.h" |
733f4fbb | 20 | #include "volumes.h" |
8dabb742 | 21 | #include "dev-replace.h" |
fcebe456 | 22 | #include "qgroup.h" |
aac0023c | 23 | #include "block-group.h" |
9c343784 | 24 | #include "space-info.h" |
d3575156 | 25 | #include "zoned.h" |
79154b1b | 26 | |
fc7cbcd4 | 27 | #define BTRFS_ROOT_TRANS_TAG 0 |
0f7d52f4 | 28 | |
61c047b5 QW |
29 | /* |
30 | * Transaction states and transitions | |
31 | * | |
32 | * No running transaction (fs tree blocks are not modified) | |
33 | * | | |
34 | * | To next stage: | |
35 | * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). | |
36 | * V | |
37 | * Transaction N [[TRANS_STATE_RUNNING]] | |
38 | * | | |
39 | * | New trans handles can be attached to transaction N by calling all | |
40 | * | start_transaction() variants. | |
41 | * | | |
42 | * | To next stage: | |
43 | * | Call btrfs_commit_transaction() on any trans handle attached to | |
44 | * | transaction N | |
45 | * V | |
46 | * Transaction N [[TRANS_STATE_COMMIT_START]] | |
47 | * | | |
48 | * | Will wait for previous running transaction to completely finish if there | |
49 | * | is one | |
50 | * | | |
51 | * | Then one of the following happes: | |
52 | * | - Wait for all other trans handle holders to release. | |
53 | * | The btrfs_commit_transaction() caller will do the commit work. | |
54 | * | - Wait for current transaction to be committed by others. | |
55 | * | Other btrfs_commit_transaction() caller will do the commit work. | |
56 | * | | |
57 | * | At this stage, only btrfs_join_transaction*() variants can attach | |
58 | * | to this running transaction. | |
59 | * | All other variants will wait for current one to finish and attach to | |
60 | * | transaction N+1. | |
61 | * | | |
62 | * | To next stage: | |
63 | * | Caller is chosen to commit transaction N, and all other trans handle | |
64 | * | haven been released. | |
65 | * V | |
66 | * Transaction N [[TRANS_STATE_COMMIT_DOING]] | |
67 | * | | |
68 | * | The heavy lifting transaction work is started. | |
69 | * | From running delayed refs (modifying extent tree) to creating pending | |
70 | * | snapshots, running qgroups. | |
71 | * | In short, modify supporting trees to reflect modifications of subvolume | |
72 | * | trees. | |
73 | * | | |
74 | * | At this stage, all start_transaction() calls will wait for this | |
75 | * | transaction to finish and attach to transaction N+1. | |
76 | * | | |
77 | * | To next stage: | |
78 | * | Until all supporting trees are updated. | |
79 | * V | |
80 | * Transaction N [[TRANS_STATE_UNBLOCKED]] | |
81 | * | Transaction N+1 | |
82 | * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] | |
83 | * | need to write them back to disk and update | | |
84 | * | super blocks. | | |
85 | * | | | |
86 | * | At this stage, new transaction is allowed to | | |
87 | * | start. | | |
88 | * | All new start_transaction() calls will be | | |
89 | * | attached to transid N+1. | | |
90 | * | | | |
91 | * | To next stage: | | |
92 | * | Until all tree blocks are super blocks are | | |
93 | * | written to block devices | | |
94 | * V | | |
95 | * Transaction N [[TRANS_STATE_COMPLETED]] V | |
96 | * All tree blocks and super blocks are written. Transaction N+1 | |
97 | * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] | |
98 | * data structures will be cleaned up. | Life goes on | |
99 | */ | |
e8c9f186 | 100 | static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { |
4a9d8bde | 101 | [TRANS_STATE_RUNNING] = 0U, |
bcf3a3e7 NB |
102 | [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), |
103 | [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | | |
4a9d8bde | 104 | __TRANS_ATTACH | |
a6d155d2 FM |
105 | __TRANS_JOIN | |
106 | __TRANS_JOIN_NOSTART), | |
bcf3a3e7 | 107 | [TRANS_STATE_UNBLOCKED] = (__TRANS_START | |
4a9d8bde MX |
108 | __TRANS_ATTACH | |
109 | __TRANS_JOIN | | |
a6d155d2 FM |
110 | __TRANS_JOIN_NOLOCK | |
111 | __TRANS_JOIN_NOSTART), | |
d0c2f4fa FM |
112 | [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START | |
113 | __TRANS_ATTACH | | |
114 | __TRANS_JOIN | | |
115 | __TRANS_JOIN_NOLOCK | | |
116 | __TRANS_JOIN_NOSTART), | |
bcf3a3e7 | 117 | [TRANS_STATE_COMPLETED] = (__TRANS_START | |
4a9d8bde MX |
118 | __TRANS_ATTACH | |
119 | __TRANS_JOIN | | |
a6d155d2 FM |
120 | __TRANS_JOIN_NOLOCK | |
121 | __TRANS_JOIN_NOSTART), | |
4a9d8bde MX |
122 | }; |
123 | ||
724e2315 | 124 | void btrfs_put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 125 | { |
9b64f57d ER |
126 | WARN_ON(refcount_read(&transaction->use_count) == 0); |
127 | if (refcount_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 128 | BUG_ON(!list_empty(&transaction->list)); |
5c9d028b LB |
129 | WARN_ON(!RB_EMPTY_ROOT( |
130 | &transaction->delayed_refs.href_root.rb_root)); | |
81f7eb00 JM |
131 | WARN_ON(!RB_EMPTY_ROOT( |
132 | &transaction->delayed_refs.dirty_extent_root)); | |
1262133b | 133 | if (transaction->delayed_refs.pending_csums) |
ab8d0fc4 JM |
134 | btrfs_err(transaction->fs_info, |
135 | "pending csums is %llu", | |
136 | transaction->delayed_refs.pending_csums); | |
7785a663 FM |
137 | /* |
138 | * If any block groups are found in ->deleted_bgs then it's | |
139 | * because the transaction was aborted and a commit did not | |
140 | * happen (things failed before writing the new superblock | |
141 | * and calling btrfs_finish_extent_commit()), so we can not | |
142 | * discard the physical locations of the block groups. | |
143 | */ | |
144 | while (!list_empty(&transaction->deleted_bgs)) { | |
32da5386 | 145 | struct btrfs_block_group *cache; |
7785a663 FM |
146 | |
147 | cache = list_first_entry(&transaction->deleted_bgs, | |
32da5386 | 148 | struct btrfs_block_group, |
7785a663 FM |
149 | bg_list); |
150 | list_del_init(&cache->bg_list); | |
6b7304af | 151 | btrfs_unfreeze_block_group(cache); |
7785a663 FM |
152 | btrfs_put_block_group(cache); |
153 | } | |
bbbf7243 | 154 | WARN_ON(!list_empty(&transaction->dev_update_list)); |
4b5faeac | 155 | kfree(transaction); |
78fae27e | 156 | } |
79154b1b CM |
157 | } |
158 | ||
889bfa39 | 159 | static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) |
817d52f8 | 160 | { |
889bfa39 | 161 | struct btrfs_transaction *cur_trans = trans->transaction; |
16916a88 | 162 | struct btrfs_fs_info *fs_info = trans->fs_info; |
9e351cc8 JB |
163 | struct btrfs_root *root, *tmp; |
164 | ||
dfba78dc FM |
165 | /* |
166 | * At this point no one can be using this transaction to modify any tree | |
167 | * and no one can start another transaction to modify any tree either. | |
168 | */ | |
169 | ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING); | |
170 | ||
9e351cc8 | 171 | down_write(&fs_info->commit_root_sem); |
d96b3424 FM |
172 | |
173 | if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) | |
174 | fs_info->last_reloc_trans = trans->transid; | |
175 | ||
889bfa39 | 176 | list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, |
9e351cc8 JB |
177 | dirty_list) { |
178 | list_del_init(&root->dirty_list); | |
179 | free_extent_buffer(root->commit_root); | |
180 | root->commit_root = btrfs_root_node(root); | |
41e7acd3 | 181 | extent_io_tree_release(&root->dirty_log_pages); |
370a11b8 | 182 | btrfs_qgroup_clean_swapped_blocks(root); |
9e351cc8 | 183 | } |
2b9dbef2 JB |
184 | |
185 | /* We can free old roots now. */ | |
889bfa39 JB |
186 | spin_lock(&cur_trans->dropped_roots_lock); |
187 | while (!list_empty(&cur_trans->dropped_roots)) { | |
188 | root = list_first_entry(&cur_trans->dropped_roots, | |
2b9dbef2 JB |
189 | struct btrfs_root, root_list); |
190 | list_del_init(&root->root_list); | |
889bfa39 JB |
191 | spin_unlock(&cur_trans->dropped_roots_lock); |
192 | btrfs_free_log(trans, root); | |
2b9dbef2 | 193 | btrfs_drop_and_free_fs_root(fs_info, root); |
889bfa39 | 194 | spin_lock(&cur_trans->dropped_roots_lock); |
2b9dbef2 | 195 | } |
889bfa39 | 196 | spin_unlock(&cur_trans->dropped_roots_lock); |
27d56e62 | 197 | |
9e351cc8 | 198 | up_write(&fs_info->commit_root_sem); |
817d52f8 JB |
199 | } |
200 | ||
0860adfd MX |
201 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
202 | unsigned int type) | |
203 | { | |
204 | if (type & TRANS_EXTWRITERS) | |
205 | atomic_inc(&trans->num_extwriters); | |
206 | } | |
207 | ||
208 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
209 | unsigned int type) | |
210 | { | |
211 | if (type & TRANS_EXTWRITERS) | |
212 | atomic_dec(&trans->num_extwriters); | |
213 | } | |
214 | ||
215 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
216 | unsigned int type) | |
217 | { | |
218 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
219 | } | |
220 | ||
221 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
222 | { | |
223 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
224 | } |
225 | ||
fb6dea26 | 226 | /* |
79bd3712 FM |
227 | * To be called after doing the chunk btree updates right after allocating a new |
228 | * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a | |
229 | * chunk after all chunk btree updates and after finishing the second phase of | |
230 | * chunk allocation (btrfs_create_pending_block_groups()) in case some block | |
231 | * group had its chunk item insertion delayed to the second phase. | |
fb6dea26 JB |
232 | */ |
233 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
234 | { | |
235 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
236 | ||
237 | if (!trans->chunk_bytes_reserved) | |
238 | return; | |
239 | ||
fb6dea26 | 240 | btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, |
63f018be | 241 | trans->chunk_bytes_reserved, NULL); |
fb6dea26 JB |
242 | trans->chunk_bytes_reserved = 0; |
243 | } | |
244 | ||
d352ac68 CM |
245 | /* |
246 | * either allocate a new transaction or hop into the existing one | |
247 | */ | |
2ff7e61e JM |
248 | static noinline int join_transaction(struct btrfs_fs_info *fs_info, |
249 | unsigned int type) | |
79154b1b CM |
250 | { |
251 | struct btrfs_transaction *cur_trans; | |
a4abeea4 | 252 | |
19ae4e81 | 253 | spin_lock(&fs_info->trans_lock); |
d43317dc | 254 | loop: |
49b25e05 | 255 | /* The file system has been taken offline. No new transactions. */ |
84961539 | 256 | if (BTRFS_FS_ERROR(fs_info)) { |
19ae4e81 | 257 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
258 | return -EROFS; |
259 | } | |
260 | ||
19ae4e81 | 261 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 262 | if (cur_trans) { |
bf31f87f | 263 | if (TRANS_ABORTED(cur_trans)) { |
19ae4e81 | 264 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 265 | return cur_trans->aborted; |
871383be | 266 | } |
4a9d8bde | 267 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
268 | spin_unlock(&fs_info->trans_lock); |
269 | return -EBUSY; | |
270 | } | |
9b64f57d | 271 | refcount_inc(&cur_trans->use_count); |
13c5a93e | 272 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 273 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 274 | spin_unlock(&fs_info->trans_lock); |
e1489b4f | 275 | btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); |
5a9ba670 | 276 | btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); |
a4abeea4 | 277 | return 0; |
79154b1b | 278 | } |
19ae4e81 | 279 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 280 | |
354aa0fb MX |
281 | /* |
282 | * If we are ATTACH, we just want to catch the current transaction, | |
283 | * and commit it. If there is no transaction, just return ENOENT. | |
284 | */ | |
285 | if (type == TRANS_ATTACH) | |
286 | return -ENOENT; | |
287 | ||
4a9d8bde MX |
288 | /* |
289 | * JOIN_NOLOCK only happens during the transaction commit, so | |
290 | * it is impossible that ->running_transaction is NULL | |
291 | */ | |
292 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
293 | ||
4b5faeac | 294 | cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); |
a4abeea4 JB |
295 | if (!cur_trans) |
296 | return -ENOMEM; | |
d43317dc | 297 | |
e1489b4f | 298 | btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); |
5a9ba670 | 299 | btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); |
e1489b4f | 300 | |
19ae4e81 JS |
301 | spin_lock(&fs_info->trans_lock); |
302 | if (fs_info->running_transaction) { | |
d43317dc CM |
303 | /* |
304 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 305 | * to redo the checks above |
d43317dc | 306 | */ |
5a9ba670 | 307 | btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); |
e1489b4f | 308 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); |
4b5faeac | 309 | kfree(cur_trans); |
d43317dc | 310 | goto loop; |
84961539 | 311 | } else if (BTRFS_FS_ERROR(fs_info)) { |
e4b50e14 | 312 | spin_unlock(&fs_info->trans_lock); |
5a9ba670 | 313 | btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); |
e1489b4f | 314 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); |
4b5faeac | 315 | kfree(cur_trans); |
7b8b92af | 316 | return -EROFS; |
79154b1b | 317 | } |
d43317dc | 318 | |
ab8d0fc4 | 319 | cur_trans->fs_info = fs_info; |
48778179 FM |
320 | atomic_set(&cur_trans->pending_ordered, 0); |
321 | init_waitqueue_head(&cur_trans->pending_wait); | |
a4abeea4 | 322 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 323 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
324 | init_waitqueue_head(&cur_trans->writer_wait); |
325 | init_waitqueue_head(&cur_trans->commit_wait); | |
4a9d8bde | 326 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
327 | /* |
328 | * One for this trans handle, one so it will live on until we | |
329 | * commit the transaction. | |
330 | */ | |
9b64f57d | 331 | refcount_set(&cur_trans->use_count, 2); |
3204d33c | 332 | cur_trans->flags = 0; |
afd48513 | 333 | cur_trans->start_time = ktime_get_seconds(); |
a4abeea4 | 334 | |
a099d0fd AM |
335 | memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); |
336 | ||
5c9d028b | 337 | cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; |
3368d001 | 338 | cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; |
d7df2c79 | 339 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
20b297d6 JS |
340 | |
341 | /* | |
342 | * although the tree mod log is per file system and not per transaction, | |
343 | * the log must never go across transaction boundaries. | |
344 | */ | |
345 | smp_mb(); | |
31b1a2bd | 346 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
5d163e0e | 347 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); |
31b1a2bd | 348 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
5d163e0e | 349 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); |
fc36ed7e | 350 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 351 | |
a4abeea4 JB |
352 | spin_lock_init(&cur_trans->delayed_refs.lock); |
353 | ||
354 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
bbbf7243 | 355 | INIT_LIST_HEAD(&cur_trans->dev_update_list); |
9e351cc8 | 356 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
ce93ec54 | 357 | INIT_LIST_HEAD(&cur_trans->dirty_bgs); |
1bbc621e | 358 | INIT_LIST_HEAD(&cur_trans->io_bgs); |
2b9dbef2 | 359 | INIT_LIST_HEAD(&cur_trans->dropped_roots); |
1bbc621e | 360 | mutex_init(&cur_trans->cache_write_mutex); |
ce93ec54 | 361 | spin_lock_init(&cur_trans->dirty_bgs_lock); |
e33e17ee | 362 | INIT_LIST_HEAD(&cur_trans->deleted_bgs); |
2b9dbef2 | 363 | spin_lock_init(&cur_trans->dropped_roots_lock); |
d3575156 NA |
364 | INIT_LIST_HEAD(&cur_trans->releasing_ebs); |
365 | spin_lock_init(&cur_trans->releasing_ebs_lock); | |
19ae4e81 | 366 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
c258d6e3 | 367 | extent_io_tree_init(fs_info, &cur_trans->dirty_pages, |
43eb5f29 | 368 | IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode); |
fe119a6e NB |
369 | extent_io_tree_init(fs_info, &cur_trans->pinned_extents, |
370 | IO_TREE_FS_PINNED_EXTENTS, NULL); | |
19ae4e81 JS |
371 | fs_info->generation++; |
372 | cur_trans->transid = fs_info->generation; | |
373 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 374 | cur_trans->aborted = 0; |
19ae4e81 | 375 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 376 | |
79154b1b CM |
377 | return 0; |
378 | } | |
379 | ||
d352ac68 | 380 | /* |
92a7cc42 QW |
381 | * This does all the record keeping required to make sure that a shareable root |
382 | * is properly recorded in a given transaction. This is required to make sure | |
383 | * the old root from before we joined the transaction is deleted when the | |
384 | * transaction commits. | |
d352ac68 | 385 | */ |
7585717f | 386 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
6426c7ad QW |
387 | struct btrfs_root *root, |
388 | int force) | |
6702ed49 | 389 | { |
0b246afa | 390 | struct btrfs_fs_info *fs_info = root->fs_info; |
03a7e111 | 391 | int ret = 0; |
0b246afa | 392 | |
92a7cc42 | 393 | if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
6426c7ad | 394 | root->last_trans < trans->transid) || force) { |
4d31778a | 395 | WARN_ON(!force && root->commit_root != root->node); |
5d4f98a2 | 396 | |
7585717f | 397 | /* |
27cdeb70 | 398 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
399 | * we have the reloc mutex held now, so there |
400 | * is only one writer in this function | |
401 | */ | |
27cdeb70 | 402 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 403 | |
27cdeb70 | 404 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
405 | * they find our root->last_trans update |
406 | */ | |
407 | smp_wmb(); | |
408 | ||
fc7cbcd4 | 409 | spin_lock(&fs_info->fs_roots_radix_lock); |
6426c7ad | 410 | if (root->last_trans == trans->transid && !force) { |
fc7cbcd4 | 411 | spin_unlock(&fs_info->fs_roots_radix_lock); |
a4abeea4 JB |
412 | return 0; |
413 | } | |
fc7cbcd4 DS |
414 | radix_tree_tag_set(&fs_info->fs_roots_radix, |
415 | (unsigned long)root->root_key.objectid, | |
416 | BTRFS_ROOT_TRANS_TAG); | |
417 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
7585717f CM |
418 | root->last_trans = trans->transid; |
419 | ||
420 | /* this is pretty tricky. We don't want to | |
421 | * take the relocation lock in btrfs_record_root_in_trans | |
422 | * unless we're really doing the first setup for this root in | |
423 | * this transaction. | |
424 | * | |
425 | * Normally we'd use root->last_trans as a flag to decide | |
426 | * if we want to take the expensive mutex. | |
427 | * | |
428 | * But, we have to set root->last_trans before we | |
429 | * init the relocation root, otherwise, we trip over warnings | |
430 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 431 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
432 | * fixing up the reloc trees and everyone must wait. |
433 | * | |
434 | * When this is zero, they can trust root->last_trans and fly | |
435 | * through btrfs_record_root_in_trans without having to take the | |
436 | * lock. smp_wmb() makes sure that all the writes above are | |
437 | * done before we pop in the zero below | |
438 | */ | |
03a7e111 | 439 | ret = btrfs_init_reloc_root(trans, root); |
c7548af6 | 440 | smp_mb__before_atomic(); |
27cdeb70 | 441 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 | 442 | } |
03a7e111 | 443 | return ret; |
5d4f98a2 | 444 | } |
bcc63abb | 445 | |
7585717f | 446 | |
2b9dbef2 JB |
447 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
448 | struct btrfs_root *root) | |
449 | { | |
0b246afa | 450 | struct btrfs_fs_info *fs_info = root->fs_info; |
2b9dbef2 JB |
451 | struct btrfs_transaction *cur_trans = trans->transaction; |
452 | ||
453 | /* Add ourselves to the transaction dropped list */ | |
454 | spin_lock(&cur_trans->dropped_roots_lock); | |
455 | list_add_tail(&root->root_list, &cur_trans->dropped_roots); | |
456 | spin_unlock(&cur_trans->dropped_roots_lock); | |
457 | ||
458 | /* Make sure we don't try to update the root at commit time */ | |
fc7cbcd4 DS |
459 | spin_lock(&fs_info->fs_roots_radix_lock); |
460 | radix_tree_tag_clear(&fs_info->fs_roots_radix, | |
461 | (unsigned long)root->root_key.objectid, | |
462 | BTRFS_ROOT_TRANS_TAG); | |
463 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
2b9dbef2 JB |
464 | } |
465 | ||
7585717f CM |
466 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
467 | struct btrfs_root *root) | |
468 | { | |
0b246afa | 469 | struct btrfs_fs_info *fs_info = root->fs_info; |
1409e6cc | 470 | int ret; |
0b246afa | 471 | |
92a7cc42 | 472 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
7585717f CM |
473 | return 0; |
474 | ||
475 | /* | |
27cdeb70 | 476 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
477 | * and barriers |
478 | */ | |
479 | smp_rmb(); | |
480 | if (root->last_trans == trans->transid && | |
27cdeb70 | 481 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
482 | return 0; |
483 | ||
0b246afa | 484 | mutex_lock(&fs_info->reloc_mutex); |
1409e6cc | 485 | ret = record_root_in_trans(trans, root, 0); |
0b246afa | 486 | mutex_unlock(&fs_info->reloc_mutex); |
7585717f | 487 | |
1409e6cc | 488 | return ret; |
7585717f CM |
489 | } |
490 | ||
4a9d8bde MX |
491 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
492 | { | |
3296bf56 | 493 | return (trans->state >= TRANS_STATE_COMMIT_START && |
501407aa | 494 | trans->state < TRANS_STATE_UNBLOCKED && |
bf31f87f | 495 | !TRANS_ABORTED(trans)); |
4a9d8bde MX |
496 | } |
497 | ||
d352ac68 CM |
498 | /* wait for commit against the current transaction to become unblocked |
499 | * when this is done, it is safe to start a new transaction, but the current | |
500 | * transaction might not be fully on disk. | |
501 | */ | |
2ff7e61e | 502 | static void wait_current_trans(struct btrfs_fs_info *fs_info) |
79154b1b | 503 | { |
f9295749 | 504 | struct btrfs_transaction *cur_trans; |
79154b1b | 505 | |
0b246afa JM |
506 | spin_lock(&fs_info->trans_lock); |
507 | cur_trans = fs_info->running_transaction; | |
4a9d8bde | 508 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
9b64f57d | 509 | refcount_inc(&cur_trans->use_count); |
0b246afa | 510 | spin_unlock(&fs_info->trans_lock); |
72d63ed6 | 511 | |
3e738c53 | 512 | btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); |
0b246afa | 513 | wait_event(fs_info->transaction_wait, |
501407aa | 514 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
bf31f87f | 515 | TRANS_ABORTED(cur_trans)); |
724e2315 | 516 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 517 | } else { |
0b246afa | 518 | spin_unlock(&fs_info->trans_lock); |
f9295749 | 519 | } |
37d1aeee CM |
520 | } |
521 | ||
2ff7e61e | 522 | static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) |
a22285a6 | 523 | { |
0b246afa | 524 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
a4abeea4 JB |
525 | return 0; |
526 | ||
92e2f7e3 | 527 | if (type == TRANS_START) |
a22285a6 | 528 | return 1; |
a4abeea4 | 529 | |
a22285a6 YZ |
530 | return 0; |
531 | } | |
532 | ||
20dd2cbf MX |
533 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
534 | { | |
0b246afa JM |
535 | struct btrfs_fs_info *fs_info = root->fs_info; |
536 | ||
537 | if (!fs_info->reloc_ctl || | |
92a7cc42 | 538 | !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || |
20dd2cbf MX |
539 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
540 | root->reloc_root) | |
541 | return false; | |
542 | ||
543 | return true; | |
544 | } | |
545 | ||
08e007d2 | 546 | static struct btrfs_trans_handle * |
5aed1dd8 | 547 | start_transaction(struct btrfs_root *root, unsigned int num_items, |
003d7c59 JM |
548 | unsigned int type, enum btrfs_reserve_flush_enum flush, |
549 | bool enforce_qgroups) | |
37d1aeee | 550 | { |
0b246afa | 551 | struct btrfs_fs_info *fs_info = root->fs_info; |
ba2c4d4e | 552 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
a22285a6 YZ |
553 | struct btrfs_trans_handle *h; |
554 | struct btrfs_transaction *cur_trans; | |
b5009945 | 555 | u64 num_bytes = 0; |
c5567237 | 556 | u64 qgroup_reserved = 0; |
20dd2cbf | 557 | bool reloc_reserved = false; |
9c343784 | 558 | bool do_chunk_alloc = false; |
20dd2cbf | 559 | int ret; |
acce952b | 560 | |
84961539 | 561 | if (BTRFS_FS_ERROR(fs_info)) |
acce952b | 562 | return ERR_PTR(-EROFS); |
2a1eb461 | 563 | |
46c4e71e | 564 | if (current->journal_info) { |
0860adfd | 565 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 | 566 | h = current->journal_info; |
b50fff81 DS |
567 | refcount_inc(&h->use_count); |
568 | WARN_ON(refcount_read(&h->use_count) > 2); | |
2a1eb461 JB |
569 | h->orig_rsv = h->block_rsv; |
570 | h->block_rsv = NULL; | |
571 | goto got_it; | |
572 | } | |
b5009945 JB |
573 | |
574 | /* | |
575 | * Do the reservation before we join the transaction so we can do all | |
576 | * the appropriate flushing if need be. | |
577 | */ | |
003d7c59 | 578 | if (num_items && root != fs_info->chunk_root) { |
ba2c4d4e JB |
579 | struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; |
580 | u64 delayed_refs_bytes = 0; | |
581 | ||
0b246afa | 582 | qgroup_reserved = num_items * fs_info->nodesize; |
733e03a0 QW |
583 | ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, |
584 | enforce_qgroups); | |
7174109c QW |
585 | if (ret) |
586 | return ERR_PTR(ret); | |
c5567237 | 587 | |
ba2c4d4e JB |
588 | /* |
589 | * We want to reserve all the bytes we may need all at once, so | |
590 | * we only do 1 enospc flushing cycle per transaction start. We | |
591 | * accomplish this by simply assuming we'll do 2 x num_items | |
592 | * worth of delayed refs updates in this trans handle, and | |
593 | * refill that amount for whatever is missing in the reserve. | |
594 | */ | |
2bd36e7b | 595 | num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); |
7f9fe614 | 596 | if (flush == BTRFS_RESERVE_FLUSH_ALL && |
748f553c | 597 | btrfs_block_rsv_full(delayed_refs_rsv) == 0) { |
ba2c4d4e JB |
598 | delayed_refs_bytes = num_bytes; |
599 | num_bytes <<= 1; | |
600 | } | |
601 | ||
20dd2cbf MX |
602 | /* |
603 | * Do the reservation for the relocation root creation | |
604 | */ | |
ee39b432 | 605 | if (need_reserve_reloc_root(root)) { |
0b246afa | 606 | num_bytes += fs_info->nodesize; |
20dd2cbf MX |
607 | reloc_reserved = true; |
608 | } | |
609 | ||
9270501c | 610 | ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes, flush); |
ba2c4d4e JB |
611 | if (ret) |
612 | goto reserve_fail; | |
613 | if (delayed_refs_bytes) { | |
614 | btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, | |
615 | delayed_refs_bytes); | |
616 | num_bytes -= delayed_refs_bytes; | |
617 | } | |
9c343784 JB |
618 | |
619 | if (rsv->space_info->force_alloc) | |
620 | do_chunk_alloc = true; | |
ba2c4d4e | 621 | } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && |
748f553c | 622 | !btrfs_block_rsv_full(delayed_refs_rsv)) { |
ba2c4d4e JB |
623 | /* |
624 | * Some people call with btrfs_start_transaction(root, 0) | |
625 | * because they can be throttled, but have some other mechanism | |
626 | * for reserving space. We still want these guys to refill the | |
627 | * delayed block_rsv so just add 1 items worth of reservation | |
628 | * here. | |
629 | */ | |
630 | ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); | |
b5009945 | 631 | if (ret) |
843fcf35 | 632 | goto reserve_fail; |
b5009945 | 633 | } |
a22285a6 | 634 | again: |
f2f767e7 | 635 | h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); |
843fcf35 MX |
636 | if (!h) { |
637 | ret = -ENOMEM; | |
638 | goto alloc_fail; | |
639 | } | |
37d1aeee | 640 | |
98114659 JB |
641 | /* |
642 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
643 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
644 | * because we're already holding a ref. We need this because we could | |
645 | * have raced in and did an fsync() on a file which can kick a commit | |
646 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
647 | * |
648 | * If we are ATTACH, it means we just want to catch the current | |
649 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 650 | */ |
0860adfd | 651 | if (type & __TRANS_FREEZABLE) |
0b246afa | 652 | sb_start_intwrite(fs_info->sb); |
b2b5ef5c | 653 | |
2ff7e61e JM |
654 | if (may_wait_transaction(fs_info, type)) |
655 | wait_current_trans(fs_info); | |
a22285a6 | 656 | |
a4abeea4 | 657 | do { |
2ff7e61e | 658 | ret = join_transaction(fs_info, type); |
178260b2 | 659 | if (ret == -EBUSY) { |
2ff7e61e | 660 | wait_current_trans(fs_info); |
a6d155d2 FM |
661 | if (unlikely(type == TRANS_ATTACH || |
662 | type == TRANS_JOIN_NOSTART)) | |
178260b2 MX |
663 | ret = -ENOENT; |
664 | } | |
a4abeea4 JB |
665 | } while (ret == -EBUSY); |
666 | ||
a43f7f82 | 667 | if (ret < 0) |
843fcf35 | 668 | goto join_fail; |
0f7d52f4 | 669 | |
0b246afa | 670 | cur_trans = fs_info->running_transaction; |
a22285a6 YZ |
671 | |
672 | h->transid = cur_trans->transid; | |
673 | h->transaction = cur_trans; | |
b50fff81 | 674 | refcount_set(&h->use_count, 1); |
64b63580 | 675 | h->fs_info = root->fs_info; |
7174109c | 676 | |
a698d075 | 677 | h->type = type; |
ea658bad | 678 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 679 | |
a22285a6 | 680 | smp_mb(); |
3296bf56 | 681 | if (cur_trans->state >= TRANS_STATE_COMMIT_START && |
2ff7e61e | 682 | may_wait_transaction(fs_info, type)) { |
abdd2e80 | 683 | current->journal_info = h; |
3a45bb20 | 684 | btrfs_commit_transaction(h); |
a22285a6 YZ |
685 | goto again; |
686 | } | |
687 | ||
b5009945 | 688 | if (num_bytes) { |
0b246afa | 689 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 690 | h->transid, num_bytes, 1); |
0b246afa | 691 | h->block_rsv = &fs_info->trans_block_rsv; |
b5009945 | 692 | h->bytes_reserved = num_bytes; |
20dd2cbf | 693 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 694 | } |
9ed74f2d | 695 | |
2a1eb461 | 696 | got_it: |
bcf3a3e7 | 697 | if (!current->journal_info) |
a22285a6 | 698 | current->journal_info = h; |
fcc99734 | 699 | |
9c343784 JB |
700 | /* |
701 | * If the space_info is marked ALLOC_FORCE then we'll get upgraded to | |
702 | * ALLOC_FORCE the first run through, and then we won't allocate for | |
703 | * anybody else who races in later. We don't care about the return | |
704 | * value here. | |
705 | */ | |
706 | if (do_chunk_alloc && num_bytes) { | |
707 | u64 flags = h->block_rsv->space_info->flags; | |
708 | ||
709 | btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), | |
710 | CHUNK_ALLOC_NO_FORCE); | |
711 | } | |
712 | ||
fcc99734 QW |
713 | /* |
714 | * btrfs_record_root_in_trans() needs to alloc new extents, and may | |
715 | * call btrfs_join_transaction() while we're also starting a | |
716 | * transaction. | |
717 | * | |
718 | * Thus it need to be called after current->journal_info initialized, | |
719 | * or we can deadlock. | |
720 | */ | |
68075ea8 JB |
721 | ret = btrfs_record_root_in_trans(h, root); |
722 | if (ret) { | |
723 | /* | |
724 | * The transaction handle is fully initialized and linked with | |
725 | * other structures so it needs to be ended in case of errors, | |
726 | * not just freed. | |
727 | */ | |
728 | btrfs_end_transaction(h); | |
729 | return ERR_PTR(ret); | |
730 | } | |
fcc99734 | 731 | |
79154b1b | 732 | return h; |
843fcf35 MX |
733 | |
734 | join_fail: | |
0860adfd | 735 | if (type & __TRANS_FREEZABLE) |
0b246afa | 736 | sb_end_intwrite(fs_info->sb); |
843fcf35 MX |
737 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
738 | alloc_fail: | |
739 | if (num_bytes) | |
2ff7e61e | 740 | btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, |
63f018be | 741 | num_bytes, NULL); |
843fcf35 | 742 | reserve_fail: |
733e03a0 | 743 | btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); |
843fcf35 | 744 | return ERR_PTR(ret); |
79154b1b CM |
745 | } |
746 | ||
f9295749 | 747 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
5aed1dd8 | 748 | unsigned int num_items) |
f9295749 | 749 | { |
08e007d2 | 750 | return start_transaction(root, num_items, TRANS_START, |
003d7c59 | 751 | BTRFS_RESERVE_FLUSH_ALL, true); |
f9295749 | 752 | } |
003d7c59 | 753 | |
8eab77ff FM |
754 | struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( |
755 | struct btrfs_root *root, | |
7f9fe614 | 756 | unsigned int num_items) |
8eab77ff | 757 | { |
7f9fe614 JB |
758 | return start_transaction(root, num_items, TRANS_START, |
759 | BTRFS_RESERVE_FLUSH_ALL_STEAL, false); | |
8eab77ff | 760 | } |
8407aa46 | 761 | |
7a7eaa40 | 762 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 763 | { |
003d7c59 JM |
764 | return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, |
765 | true); | |
f9295749 CM |
766 | } |
767 | ||
8d510121 | 768 | struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) |
0af3d00b | 769 | { |
575a75d6 | 770 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, |
003d7c59 | 771 | BTRFS_RESERVE_NO_FLUSH, true); |
0af3d00b JB |
772 | } |
773 | ||
a6d155d2 FM |
774 | /* |
775 | * Similar to regular join but it never starts a transaction when none is | |
776 | * running or after waiting for the current one to finish. | |
777 | */ | |
778 | struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) | |
779 | { | |
780 | return start_transaction(root, 0, TRANS_JOIN_NOSTART, | |
781 | BTRFS_RESERVE_NO_FLUSH, true); | |
782 | } | |
783 | ||
d4edf39b MX |
784 | /* |
785 | * btrfs_attach_transaction() - catch the running transaction | |
786 | * | |
787 | * It is used when we want to commit the current the transaction, but | |
788 | * don't want to start a new one. | |
789 | * | |
790 | * Note: If this function return -ENOENT, it just means there is no | |
791 | * running transaction. But it is possible that the inactive transaction | |
792 | * is still in the memory, not fully on disk. If you hope there is no | |
793 | * inactive transaction in the fs when -ENOENT is returned, you should | |
794 | * invoke | |
795 | * btrfs_attach_transaction_barrier() | |
796 | */ | |
354aa0fb | 797 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 798 | { |
575a75d6 | 799 | return start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 800 | BTRFS_RESERVE_NO_FLUSH, true); |
60376ce4 JB |
801 | } |
802 | ||
d4edf39b | 803 | /* |
90b6d283 | 804 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b | 805 | * |
52042d8e | 806 | * It is similar to the above function, the difference is this one |
d4edf39b MX |
807 | * will wait for all the inactive transactions until they fully |
808 | * complete. | |
809 | */ | |
810 | struct btrfs_trans_handle * | |
811 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
812 | { | |
813 | struct btrfs_trans_handle *trans; | |
814 | ||
575a75d6 | 815 | trans = start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 816 | BTRFS_RESERVE_NO_FLUSH, true); |
8d9e220c | 817 | if (trans == ERR_PTR(-ENOENT)) |
2ff7e61e | 818 | btrfs_wait_for_commit(root->fs_info, 0); |
d4edf39b MX |
819 | |
820 | return trans; | |
821 | } | |
822 | ||
d0c2f4fa FM |
823 | /* Wait for a transaction commit to reach at least the given state. */ |
824 | static noinline void wait_for_commit(struct btrfs_transaction *commit, | |
825 | const enum btrfs_trans_state min_state) | |
89ce8a63 | 826 | { |
5fd76bf3 OS |
827 | struct btrfs_fs_info *fs_info = commit->fs_info; |
828 | u64 transid = commit->transid; | |
829 | bool put = false; | |
830 | ||
3e738c53 IA |
831 | /* |
832 | * At the moment this function is called with min_state either being | |
833 | * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED. | |
834 | */ | |
835 | if (min_state == TRANS_STATE_COMPLETED) | |
836 | btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); | |
837 | else | |
838 | btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); | |
839 | ||
5fd76bf3 OS |
840 | while (1) { |
841 | wait_event(commit->commit_wait, commit->state >= min_state); | |
842 | if (put) | |
843 | btrfs_put_transaction(commit); | |
844 | ||
845 | if (min_state < TRANS_STATE_COMPLETED) | |
846 | break; | |
847 | ||
848 | /* | |
849 | * A transaction isn't really completed until all of the | |
850 | * previous transactions are completed, but with fsync we can | |
851 | * end up with SUPER_COMMITTED transactions before a COMPLETED | |
852 | * transaction. Wait for those. | |
853 | */ | |
854 | ||
855 | spin_lock(&fs_info->trans_lock); | |
856 | commit = list_first_entry_or_null(&fs_info->trans_list, | |
857 | struct btrfs_transaction, | |
858 | list); | |
859 | if (!commit || commit->transid > transid) { | |
860 | spin_unlock(&fs_info->trans_lock); | |
861 | break; | |
862 | } | |
863 | refcount_inc(&commit->use_count); | |
864 | put = true; | |
865 | spin_unlock(&fs_info->trans_lock); | |
866 | } | |
89ce8a63 CM |
867 | } |
868 | ||
2ff7e61e | 869 | int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) |
46204592 SW |
870 | { |
871 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 872 | int ret = 0; |
46204592 | 873 | |
46204592 | 874 | if (transid) { |
0b246afa | 875 | if (transid <= fs_info->last_trans_committed) |
a4abeea4 | 876 | goto out; |
46204592 SW |
877 | |
878 | /* find specified transaction */ | |
0b246afa JM |
879 | spin_lock(&fs_info->trans_lock); |
880 | list_for_each_entry(t, &fs_info->trans_list, list) { | |
46204592 SW |
881 | if (t->transid == transid) { |
882 | cur_trans = t; | |
9b64f57d | 883 | refcount_inc(&cur_trans->use_count); |
8cd2807f | 884 | ret = 0; |
46204592 SW |
885 | break; |
886 | } | |
8cd2807f MX |
887 | if (t->transid > transid) { |
888 | ret = 0; | |
46204592 | 889 | break; |
8cd2807f | 890 | } |
46204592 | 891 | } |
0b246afa | 892 | spin_unlock(&fs_info->trans_lock); |
42383020 SW |
893 | |
894 | /* | |
895 | * The specified transaction doesn't exist, or we | |
896 | * raced with btrfs_commit_transaction | |
897 | */ | |
898 | if (!cur_trans) { | |
0b246afa | 899 | if (transid > fs_info->last_trans_committed) |
42383020 | 900 | ret = -EINVAL; |
8cd2807f | 901 | goto out; |
42383020 | 902 | } |
46204592 SW |
903 | } else { |
904 | /* find newest transaction that is committing | committed */ | |
0b246afa JM |
905 | spin_lock(&fs_info->trans_lock); |
906 | list_for_each_entry_reverse(t, &fs_info->trans_list, | |
46204592 | 907 | list) { |
4a9d8bde MX |
908 | if (t->state >= TRANS_STATE_COMMIT_START) { |
909 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 910 | break; |
46204592 | 911 | cur_trans = t; |
9b64f57d | 912 | refcount_inc(&cur_trans->use_count); |
46204592 SW |
913 | break; |
914 | } | |
915 | } | |
0b246afa | 916 | spin_unlock(&fs_info->trans_lock); |
46204592 | 917 | if (!cur_trans) |
a4abeea4 | 918 | goto out; /* nothing committing|committed */ |
46204592 SW |
919 | } |
920 | ||
d0c2f4fa | 921 | wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); |
724e2315 | 922 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 923 | out: |
46204592 SW |
924 | return ret; |
925 | } | |
926 | ||
2ff7e61e | 927 | void btrfs_throttle(struct btrfs_fs_info *fs_info) |
37d1aeee | 928 | { |
92e2f7e3 | 929 | wait_current_trans(fs_info); |
37d1aeee CM |
930 | } |
931 | ||
8a8f4dea | 932 | static bool should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa | 933 | { |
2ff7e61e | 934 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0b246afa | 935 | |
64403612 | 936 | if (btrfs_check_space_for_delayed_refs(fs_info)) |
8a8f4dea | 937 | return true; |
36ba022a | 938 | |
2ff7e61e | 939 | return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); |
8929ecfa YZ |
940 | } |
941 | ||
a2633b6a | 942 | bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa YZ |
943 | { |
944 | struct btrfs_transaction *cur_trans = trans->transaction; | |
8929ecfa | 945 | |
3296bf56 | 946 | if (cur_trans->state >= TRANS_STATE_COMMIT_START || |
e19eb11f | 947 | test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) |
a2633b6a | 948 | return true; |
8929ecfa | 949 | |
2ff7e61e | 950 | return should_end_transaction(trans); |
8929ecfa YZ |
951 | } |
952 | ||
dc60c525 NB |
953 | static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) |
954 | ||
0e34693f | 955 | { |
dc60c525 NB |
956 | struct btrfs_fs_info *fs_info = trans->fs_info; |
957 | ||
0e34693f NB |
958 | if (!trans->block_rsv) { |
959 | ASSERT(!trans->bytes_reserved); | |
960 | return; | |
961 | } | |
962 | ||
963 | if (!trans->bytes_reserved) | |
964 | return; | |
965 | ||
966 | ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); | |
967 | trace_btrfs_space_reservation(fs_info, "transaction", | |
968 | trans->transid, trans->bytes_reserved, 0); | |
969 | btrfs_block_rsv_release(fs_info, trans->block_rsv, | |
63f018be | 970 | trans->bytes_reserved, NULL); |
0e34693f NB |
971 | trans->bytes_reserved = 0; |
972 | } | |
973 | ||
89ce8a63 | 974 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
3a45bb20 | 975 | int throttle) |
79154b1b | 976 | { |
3a45bb20 | 977 | struct btrfs_fs_info *info = trans->fs_info; |
8929ecfa | 978 | struct btrfs_transaction *cur_trans = trans->transaction; |
4edc2ca3 | 979 | int err = 0; |
c3e69d58 | 980 | |
b50fff81 DS |
981 | if (refcount_read(&trans->use_count) > 1) { |
982 | refcount_dec(&trans->use_count); | |
2a1eb461 JB |
983 | trans->block_rsv = trans->orig_rsv; |
984 | return 0; | |
985 | } | |
986 | ||
dc60c525 | 987 | btrfs_trans_release_metadata(trans); |
4c13d758 | 988 | trans->block_rsv = NULL; |
c5567237 | 989 | |
119e80df | 990 | btrfs_create_pending_block_groups(trans); |
ea658bad | 991 | |
4fbcdf66 FM |
992 | btrfs_trans_release_chunk_metadata(trans); |
993 | ||
0860adfd | 994 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 995 | sb_end_intwrite(info->sb); |
6df7881a | 996 | |
8929ecfa | 997 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
998 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
999 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 1000 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 1001 | |
093258e6 | 1002 | cond_wake_up(&cur_trans->writer_wait); |
e1489b4f | 1003 | |
5a9ba670 | 1004 | btrfs_lockdep_release(info, btrfs_trans_num_extwriters); |
e1489b4f IA |
1005 | btrfs_lockdep_release(info, btrfs_trans_num_writers); |
1006 | ||
724e2315 | 1007 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
1008 | |
1009 | if (current->journal_info == trans) | |
1010 | current->journal_info = NULL; | |
ab78c84d | 1011 | |
24bbcf04 | 1012 | if (throttle) |
2ff7e61e | 1013 | btrfs_run_delayed_iputs(info); |
24bbcf04 | 1014 | |
84961539 | 1015 | if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) { |
4e121c06 | 1016 | wake_up_process(info->transaction_kthread); |
fbabd4a3 JB |
1017 | if (TRANS_ABORTED(trans)) |
1018 | err = trans->aborted; | |
1019 | else | |
1020 | err = -EROFS; | |
4e121c06 | 1021 | } |
49b25e05 | 1022 | |
4edc2ca3 DJ |
1023 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
1024 | return err; | |
79154b1b CM |
1025 | } |
1026 | ||
3a45bb20 | 1027 | int btrfs_end_transaction(struct btrfs_trans_handle *trans) |
89ce8a63 | 1028 | { |
3a45bb20 | 1029 | return __btrfs_end_transaction(trans, 0); |
89ce8a63 CM |
1030 | } |
1031 | ||
3a45bb20 | 1032 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) |
89ce8a63 | 1033 | { |
3a45bb20 | 1034 | return __btrfs_end_transaction(trans, 1); |
16cdcec7 MX |
1035 | } |
1036 | ||
d352ac68 CM |
1037 | /* |
1038 | * when btree blocks are allocated, they have some corresponding bits set for | |
1039 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 1040 | * those extents are sent to disk but does not wait on them |
d352ac68 | 1041 | */ |
2ff7e61e | 1042 | int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, |
8cef4e16 | 1043 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 1044 | { |
777e6bd7 | 1045 | int err = 0; |
7c4452b9 | 1046 | int werr = 0; |
0b246afa | 1047 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 1048 | struct extent_state *cached_state = NULL; |
777e6bd7 | 1049 | u64 start = 0; |
5f39d397 | 1050 | u64 end; |
7c4452b9 | 1051 | |
6300463b | 1052 | atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
1728366e | 1053 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 1054 | mark, &cached_state)) { |
663dfbb0 FM |
1055 | bool wait_writeback = false; |
1056 | ||
1057 | err = convert_extent_bit(dirty_pages, start, end, | |
1058 | EXTENT_NEED_WAIT, | |
210aa277 | 1059 | mark, &cached_state); |
663dfbb0 FM |
1060 | /* |
1061 | * convert_extent_bit can return -ENOMEM, which is most of the | |
1062 | * time a temporary error. So when it happens, ignore the error | |
1063 | * and wait for writeback of this range to finish - because we | |
1064 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
bf89d38f JM |
1065 | * to __btrfs_wait_marked_extents() would not know that |
1066 | * writeback for this range started and therefore wouldn't | |
1067 | * wait for it to finish - we don't want to commit a | |
1068 | * superblock that points to btree nodes/leafs for which | |
1069 | * writeback hasn't finished yet (and without errors). | |
663dfbb0 | 1070 | * We cleanup any entries left in the io tree when committing |
41e7acd3 | 1071 | * the transaction (through extent_io_tree_release()). |
663dfbb0 FM |
1072 | */ |
1073 | if (err == -ENOMEM) { | |
1074 | err = 0; | |
1075 | wait_writeback = true; | |
1076 | } | |
1077 | if (!err) | |
1078 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
1079 | if (err) |
1080 | werr = err; | |
663dfbb0 FM |
1081 | else if (wait_writeback) |
1082 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 1083 | free_extent_state(cached_state); |
663dfbb0 | 1084 | cached_state = NULL; |
1728366e JB |
1085 | cond_resched(); |
1086 | start = end + 1; | |
7c4452b9 | 1087 | } |
6300463b | 1088 | atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
690587d1 CM |
1089 | return werr; |
1090 | } | |
1091 | ||
1092 | /* | |
1093 | * when btree blocks are allocated, they have some corresponding bits set for | |
1094 | * them in one of two extent_io trees. This is used to make sure all of | |
1095 | * those extents are on disk for transaction or log commit. We wait | |
1096 | * on all the pages and clear them from the dirty pages state tree | |
1097 | */ | |
bf89d38f JM |
1098 | static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, |
1099 | struct extent_io_tree *dirty_pages) | |
690587d1 | 1100 | { |
690587d1 CM |
1101 | int err = 0; |
1102 | int werr = 0; | |
0b246afa | 1103 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 1104 | struct extent_state *cached_state = NULL; |
690587d1 CM |
1105 | u64 start = 0; |
1106 | u64 end; | |
777e6bd7 | 1107 | |
1728366e | 1108 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 1109 | EXTENT_NEED_WAIT, &cached_state)) { |
663dfbb0 FM |
1110 | /* |
1111 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
1112 | * When committing the transaction, we'll remove any entries | |
1113 | * left in the io tree. For a log commit, we don't remove them | |
1114 | * after committing the log because the tree can be accessed | |
1115 | * concurrently - we do it only at transaction commit time when | |
41e7acd3 | 1116 | * it's safe to do it (through extent_io_tree_release()). |
663dfbb0 FM |
1117 | */ |
1118 | err = clear_extent_bit(dirty_pages, start, end, | |
dbbf4992 | 1119 | EXTENT_NEED_WAIT, 0, &cached_state); |
663dfbb0 FM |
1120 | if (err == -ENOMEM) |
1121 | err = 0; | |
1122 | if (!err) | |
1123 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
1124 | if (err) |
1125 | werr = err; | |
e38e2ed7 FM |
1126 | free_extent_state(cached_state); |
1127 | cached_state = NULL; | |
1728366e JB |
1128 | cond_resched(); |
1129 | start = end + 1; | |
777e6bd7 | 1130 | } |
7c4452b9 CM |
1131 | if (err) |
1132 | werr = err; | |
bf89d38f JM |
1133 | return werr; |
1134 | } | |
656f30db | 1135 | |
b9fae2eb | 1136 | static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, |
bf89d38f JM |
1137 | struct extent_io_tree *dirty_pages) |
1138 | { | |
1139 | bool errors = false; | |
1140 | int err; | |
656f30db | 1141 | |
bf89d38f JM |
1142 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); |
1143 | if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) | |
1144 | errors = true; | |
1145 | ||
1146 | if (errors && !err) | |
1147 | err = -EIO; | |
1148 | return err; | |
1149 | } | |
656f30db | 1150 | |
bf89d38f JM |
1151 | int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) |
1152 | { | |
1153 | struct btrfs_fs_info *fs_info = log_root->fs_info; | |
1154 | struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; | |
1155 | bool errors = false; | |
1156 | int err; | |
656f30db | 1157 | |
bf89d38f JM |
1158 | ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
1159 | ||
1160 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); | |
1161 | if ((mark & EXTENT_DIRTY) && | |
1162 | test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) | |
1163 | errors = true; | |
1164 | ||
1165 | if ((mark & EXTENT_NEW) && | |
1166 | test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) | |
1167 | errors = true; | |
1168 | ||
1169 | if (errors && !err) | |
1170 | err = -EIO; | |
1171 | return err; | |
79154b1b CM |
1172 | } |
1173 | ||
690587d1 | 1174 | /* |
c9b577c0 NB |
1175 | * When btree blocks are allocated the corresponding extents are marked dirty. |
1176 | * This function ensures such extents are persisted on disk for transaction or | |
1177 | * log commit. | |
1178 | * | |
1179 | * @trans: transaction whose dirty pages we'd like to write | |
690587d1 | 1180 | */ |
70458a58 | 1181 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) |
690587d1 CM |
1182 | { |
1183 | int ret; | |
1184 | int ret2; | |
c9b577c0 | 1185 | struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; |
70458a58 | 1186 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c6adc9cc | 1187 | struct blk_plug plug; |
690587d1 | 1188 | |
c6adc9cc | 1189 | blk_start_plug(&plug); |
c9b577c0 | 1190 | ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); |
c6adc9cc | 1191 | blk_finish_plug(&plug); |
bf89d38f | 1192 | ret2 = btrfs_wait_extents(fs_info, dirty_pages); |
bf0da8c1 | 1193 | |
41e7acd3 | 1194 | extent_io_tree_release(&trans->transaction->dirty_pages); |
c9b577c0 | 1195 | |
bf0da8c1 CM |
1196 | if (ret) |
1197 | return ret; | |
c9b577c0 | 1198 | else if (ret2) |
bf0da8c1 | 1199 | return ret2; |
c9b577c0 NB |
1200 | else |
1201 | return 0; | |
d0c803c4 CM |
1202 | } |
1203 | ||
d352ac68 CM |
1204 | /* |
1205 | * this is used to update the root pointer in the tree of tree roots. | |
1206 | * | |
1207 | * But, in the case of the extent allocation tree, updating the root | |
1208 | * pointer may allocate blocks which may change the root of the extent | |
1209 | * allocation tree. | |
1210 | * | |
1211 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1212 | * change while the root pointer was being updated in the metadata. | |
1213 | */ | |
0b86a832 CM |
1214 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1215 | struct btrfs_root *root) | |
79154b1b CM |
1216 | { |
1217 | int ret; | |
0b86a832 | 1218 | u64 old_root_bytenr; |
86b9f2ec | 1219 | u64 old_root_used; |
0b246afa JM |
1220 | struct btrfs_fs_info *fs_info = root->fs_info; |
1221 | struct btrfs_root *tree_root = fs_info->tree_root; | |
79154b1b | 1222 | |
86b9f2ec | 1223 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1224 | |
d397712b | 1225 | while (1) { |
0b86a832 | 1226 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1227 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1228 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1229 | break; |
87ef2bb4 | 1230 | |
5d4f98a2 | 1231 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1232 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1233 | &root->root_key, |
1234 | &root->root_item); | |
49b25e05 JM |
1235 | if (ret) |
1236 | return ret; | |
56bec294 | 1237 | |
86b9f2ec | 1238 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1239 | } |
276e680d | 1240 | |
0b86a832 CM |
1241 | return 0; |
1242 | } | |
1243 | ||
d352ac68 CM |
1244 | /* |
1245 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1246 | * |
1247 | * The error handling in this function may not be obvious. Any of the | |
1248 | * failures will cause the file system to go offline. We still need | |
1249 | * to clean up the delayed refs. | |
d352ac68 | 1250 | */ |
9386d8bc | 1251 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) |
0b86a832 | 1252 | { |
9386d8bc | 1253 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ea526d18 | 1254 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1255 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1256 | struct list_head *next; |
84234f3a | 1257 | struct extent_buffer *eb; |
56bec294 | 1258 | int ret; |
84234f3a | 1259 | |
dfba78dc FM |
1260 | /* |
1261 | * At this point no one can be using this transaction to modify any tree | |
1262 | * and no one can start another transaction to modify any tree either. | |
1263 | */ | |
1264 | ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); | |
1265 | ||
84234f3a | 1266 | eb = btrfs_lock_root_node(fs_info->tree_root); |
49b25e05 | 1267 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
9631e4cc | 1268 | 0, &eb, BTRFS_NESTING_COW); |
84234f3a YZ |
1269 | btrfs_tree_unlock(eb); |
1270 | free_extent_buffer(eb); | |
0b86a832 | 1271 | |
49b25e05 JM |
1272 | if (ret) |
1273 | return ret; | |
87ef2bb4 | 1274 | |
196c9d8d | 1275 | ret = btrfs_run_dev_stats(trans); |
c16ce190 JB |
1276 | if (ret) |
1277 | return ret; | |
2b584c68 | 1278 | ret = btrfs_run_dev_replace(trans); |
c16ce190 JB |
1279 | if (ret) |
1280 | return ret; | |
280f8bd2 | 1281 | ret = btrfs_run_qgroups(trans); |
c16ce190 JB |
1282 | if (ret) |
1283 | return ret; | |
546adb0d | 1284 | |
bbebb3e0 | 1285 | ret = btrfs_setup_space_cache(trans); |
dcdf7f6d JB |
1286 | if (ret) |
1287 | return ret; | |
1288 | ||
ea526d18 | 1289 | again: |
d397712b | 1290 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
2ff7e61e | 1291 | struct btrfs_root *root; |
0b86a832 CM |
1292 | next = fs_info->dirty_cowonly_roots.next; |
1293 | list_del_init(next); | |
1294 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1295 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1296 | |
826582ca JB |
1297 | list_add_tail(&root->dirty_list, |
1298 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1299 | ret = update_cowonly_root(trans, root); |
1300 | if (ret) | |
1301 | return ret; | |
79154b1b | 1302 | } |
276e680d | 1303 | |
488bc2a2 JB |
1304 | /* Now flush any delayed refs generated by updating all of the roots */ |
1305 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); | |
1306 | if (ret) | |
1307 | return ret; | |
1308 | ||
1bbc621e | 1309 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
5742d15f | 1310 | ret = btrfs_write_dirty_block_groups(trans); |
ea526d18 JB |
1311 | if (ret) |
1312 | return ret; | |
488bc2a2 JB |
1313 | |
1314 | /* | |
1315 | * We're writing the dirty block groups, which could generate | |
1316 | * delayed refs, which could generate more dirty block groups, | |
1317 | * so we want to keep this flushing in this loop to make sure | |
1318 | * everything gets run. | |
1319 | */ | |
c79a70b1 | 1320 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
ea526d18 JB |
1321 | if (ret) |
1322 | return ret; | |
1323 | } | |
1324 | ||
1325 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1326 | goto again; | |
1327 | ||
9f6cbcbb DS |
1328 | /* Update dev-replace pointer once everything is committed */ |
1329 | fs_info->dev_replace.committed_cursor_left = | |
1330 | fs_info->dev_replace.cursor_left_last_write_of_item; | |
8dabb742 | 1331 | |
79154b1b CM |
1332 | return 0; |
1333 | } | |
1334 | ||
b4be6aef JB |
1335 | /* |
1336 | * If we had a pending drop we need to see if there are any others left in our | |
1337 | * dead roots list, and if not clear our bit and wake any waiters. | |
1338 | */ | |
1339 | void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info) | |
1340 | { | |
1341 | /* | |
1342 | * We put the drop in progress roots at the front of the list, so if the | |
1343 | * first entry doesn't have UNFINISHED_DROP set we can wake everybody | |
1344 | * up. | |
1345 | */ | |
1346 | spin_lock(&fs_info->trans_lock); | |
1347 | if (!list_empty(&fs_info->dead_roots)) { | |
1348 | struct btrfs_root *root = list_first_entry(&fs_info->dead_roots, | |
1349 | struct btrfs_root, | |
1350 | root_list); | |
1351 | if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) { | |
1352 | spin_unlock(&fs_info->trans_lock); | |
1353 | return; | |
1354 | } | |
1355 | } | |
1356 | spin_unlock(&fs_info->trans_lock); | |
1357 | ||
1358 | btrfs_wake_unfinished_drop(fs_info); | |
1359 | } | |
1360 | ||
d352ac68 CM |
1361 | /* |
1362 | * dead roots are old snapshots that need to be deleted. This allocates | |
1363 | * a dirty root struct and adds it into the list of dead roots that need to | |
1364 | * be deleted | |
1365 | */ | |
cfad392b | 1366 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1367 | { |
0b246afa JM |
1368 | struct btrfs_fs_info *fs_info = root->fs_info; |
1369 | ||
1370 | spin_lock(&fs_info->trans_lock); | |
dc9492c1 JB |
1371 | if (list_empty(&root->root_list)) { |
1372 | btrfs_grab_root(root); | |
b4be6aef JB |
1373 | |
1374 | /* We want to process the partially complete drops first. */ | |
1375 | if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) | |
1376 | list_add(&root->root_list, &fs_info->dead_roots); | |
1377 | else | |
1378 | list_add_tail(&root->root_list, &fs_info->dead_roots); | |
dc9492c1 | 1379 | } |
0b246afa | 1380 | spin_unlock(&fs_info->trans_lock); |
5eda7b5e CM |
1381 | } |
1382 | ||
d352ac68 | 1383 | /* |
dfba78dc FM |
1384 | * Update each subvolume root and its relocation root, if it exists, in the tree |
1385 | * of tree roots. Also free log roots if they exist. | |
d352ac68 | 1386 | */ |
7e4443d9 | 1387 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) |
0f7d52f4 | 1388 | { |
7e4443d9 | 1389 | struct btrfs_fs_info *fs_info = trans->fs_info; |
fc7cbcd4 DS |
1390 | struct btrfs_root *gang[8]; |
1391 | int i; | |
1392 | int ret; | |
54aa1f4d | 1393 | |
dfba78dc FM |
1394 | /* |
1395 | * At this point no one can be using this transaction to modify any tree | |
1396 | * and no one can start another transaction to modify any tree either. | |
1397 | */ | |
1398 | ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); | |
1399 | ||
fc7cbcd4 DS |
1400 | spin_lock(&fs_info->fs_roots_radix_lock); |
1401 | while (1) { | |
1402 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, | |
1403 | (void **)gang, 0, | |
1404 | ARRAY_SIZE(gang), | |
1405 | BTRFS_ROOT_TRANS_TAG); | |
1406 | if (ret == 0) | |
1407 | break; | |
1408 | for (i = 0; i < ret; i++) { | |
1409 | struct btrfs_root *root = gang[i]; | |
1410 | int ret2; | |
1411 | ||
1412 | /* | |
1413 | * At this point we can neither have tasks logging inodes | |
1414 | * from a root nor trying to commit a log tree. | |
1415 | */ | |
1416 | ASSERT(atomic_read(&root->log_writers) == 0); | |
1417 | ASSERT(atomic_read(&root->log_commit[0]) == 0); | |
1418 | ASSERT(atomic_read(&root->log_commit[1]) == 0); | |
1419 | ||
1420 | radix_tree_tag_clear(&fs_info->fs_roots_radix, | |
1421 | (unsigned long)root->root_key.objectid, | |
1422 | BTRFS_ROOT_TRANS_TAG); | |
1423 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
1424 | ||
1425 | btrfs_free_log(trans, root); | |
1426 | ret2 = btrfs_update_reloc_root(trans, root); | |
1427 | if (ret2) | |
1428 | return ret2; | |
1429 | ||
1430 | /* see comments in should_cow_block() */ | |
1431 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); | |
1432 | smp_mb__after_atomic(); | |
1433 | ||
1434 | if (root->commit_root != root->node) { | |
1435 | list_add_tail(&root->dirty_list, | |
1436 | &trans->transaction->switch_commits); | |
1437 | btrfs_set_root_node(&root->root_item, | |
1438 | root->node); | |
1439 | } | |
48b36a60 | 1440 | |
fc7cbcd4 DS |
1441 | ret2 = btrfs_update_root(trans, fs_info->tree_root, |
1442 | &root->root_key, | |
1443 | &root->root_item); | |
1444 | if (ret2) | |
1445 | return ret2; | |
1446 | spin_lock(&fs_info->fs_roots_radix_lock); | |
1447 | btrfs_qgroup_free_meta_all_pertrans(root); | |
0f7d52f4 CM |
1448 | } |
1449 | } | |
fc7cbcd4 | 1450 | spin_unlock(&fs_info->fs_roots_radix_lock); |
4f4317c1 | 1451 | return 0; |
0f7d52f4 CM |
1452 | } |
1453 | ||
d352ac68 | 1454 | /* |
de78b51a ES |
1455 | * defrag a given btree. |
1456 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1457 | */ |
de78b51a | 1458 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1459 | { |
1460 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1461 | struct btrfs_trans_handle *trans; |
8929ecfa | 1462 | int ret; |
e9d0b13b | 1463 | |
27cdeb70 | 1464 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1465 | return 0; |
8929ecfa | 1466 | |
6b80053d | 1467 | while (1) { |
8929ecfa | 1468 | trans = btrfs_start_transaction(root, 0); |
6819703f DS |
1469 | if (IS_ERR(trans)) { |
1470 | ret = PTR_ERR(trans); | |
1471 | break; | |
1472 | } | |
8929ecfa | 1473 | |
de78b51a | 1474 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1475 | |
3a45bb20 | 1476 | btrfs_end_transaction(trans); |
2ff7e61e | 1477 | btrfs_btree_balance_dirty(info); |
e9d0b13b CM |
1478 | cond_resched(); |
1479 | ||
ab8d0fc4 | 1480 | if (btrfs_fs_closing(info) || ret != -EAGAIN) |
e9d0b13b | 1481 | break; |
210549eb | 1482 | |
ab8d0fc4 JM |
1483 | if (btrfs_defrag_cancelled(info)) { |
1484 | btrfs_debug(info, "defrag_root cancelled"); | |
210549eb DS |
1485 | ret = -EAGAIN; |
1486 | break; | |
1487 | } | |
e9d0b13b | 1488 | } |
27cdeb70 | 1489 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1490 | return ret; |
e9d0b13b CM |
1491 | } |
1492 | ||
6426c7ad QW |
1493 | /* |
1494 | * Do all special snapshot related qgroup dirty hack. | |
1495 | * | |
1496 | * Will do all needed qgroup inherit and dirty hack like switch commit | |
1497 | * roots inside one transaction and write all btree into disk, to make | |
1498 | * qgroup works. | |
1499 | */ | |
1500 | static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, | |
1501 | struct btrfs_root *src, | |
1502 | struct btrfs_root *parent, | |
1503 | struct btrfs_qgroup_inherit *inherit, | |
1504 | u64 dst_objectid) | |
1505 | { | |
1506 | struct btrfs_fs_info *fs_info = src->fs_info; | |
1507 | int ret; | |
1508 | ||
1509 | /* | |
1510 | * Save some performance in the case that qgroups are not | |
1511 | * enabled. If this check races with the ioctl, rescan will | |
1512 | * kick in anyway. | |
1513 | */ | |
9ea6e2b5 | 1514 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
6426c7ad | 1515 | return 0; |
6426c7ad | 1516 | |
4d31778a | 1517 | /* |
52042d8e | 1518 | * Ensure dirty @src will be committed. Or, after coming |
4d31778a QW |
1519 | * commit_fs_roots() and switch_commit_roots(), any dirty but not |
1520 | * recorded root will never be updated again, causing an outdated root | |
1521 | * item. | |
1522 | */ | |
1c442d22 JB |
1523 | ret = record_root_in_trans(trans, src, 1); |
1524 | if (ret) | |
1525 | return ret; | |
4d31778a | 1526 | |
2a4d84c1 JB |
1527 | /* |
1528 | * btrfs_qgroup_inherit relies on a consistent view of the usage for the | |
1529 | * src root, so we must run the delayed refs here. | |
1530 | * | |
1531 | * However this isn't particularly fool proof, because there's no | |
1532 | * synchronization keeping us from changing the tree after this point | |
1533 | * before we do the qgroup_inherit, or even from making changes while | |
1534 | * we're doing the qgroup_inherit. But that's a problem for the future, | |
1535 | * for now flush the delayed refs to narrow the race window where the | |
1536 | * qgroup counters could end up wrong. | |
1537 | */ | |
1538 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); | |
1539 | if (ret) { | |
1540 | btrfs_abort_transaction(trans, ret); | |
44365827 | 1541 | return ret; |
2a4d84c1 JB |
1542 | } |
1543 | ||
7e4443d9 | 1544 | ret = commit_fs_roots(trans); |
6426c7ad QW |
1545 | if (ret) |
1546 | goto out; | |
460fb20a | 1547 | ret = btrfs_qgroup_account_extents(trans); |
6426c7ad QW |
1548 | if (ret < 0) |
1549 | goto out; | |
1550 | ||
1551 | /* Now qgroup are all updated, we can inherit it to new qgroups */ | |
a9377422 | 1552 | ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, |
6426c7ad QW |
1553 | inherit); |
1554 | if (ret < 0) | |
1555 | goto out; | |
1556 | ||
1557 | /* | |
1558 | * Now we do a simplified commit transaction, which will: | |
1559 | * 1) commit all subvolume and extent tree | |
1560 | * To ensure all subvolume and extent tree have a valid | |
1561 | * commit_root to accounting later insert_dir_item() | |
1562 | * 2) write all btree blocks onto disk | |
1563 | * This is to make sure later btree modification will be cowed | |
1564 | * Or commit_root can be populated and cause wrong qgroup numbers | |
1565 | * In this simplified commit, we don't really care about other trees | |
1566 | * like chunk and root tree, as they won't affect qgroup. | |
1567 | * And we don't write super to avoid half committed status. | |
1568 | */ | |
9386d8bc | 1569 | ret = commit_cowonly_roots(trans); |
6426c7ad QW |
1570 | if (ret) |
1571 | goto out; | |
889bfa39 | 1572 | switch_commit_roots(trans); |
70458a58 | 1573 | ret = btrfs_write_and_wait_transaction(trans); |
6426c7ad | 1574 | if (ret) |
f7af3934 | 1575 | btrfs_handle_fs_error(fs_info, ret, |
6426c7ad QW |
1576 | "Error while writing out transaction for qgroup"); |
1577 | ||
1578 | out: | |
6426c7ad QW |
1579 | /* |
1580 | * Force parent root to be updated, as we recorded it before so its | |
1581 | * last_trans == cur_transid. | |
1582 | * Or it won't be committed again onto disk after later | |
1583 | * insert_dir_item() | |
1584 | */ | |
1585 | if (!ret) | |
1c442d22 | 1586 | ret = record_root_in_trans(trans, parent, 1); |
6426c7ad QW |
1587 | return ret; |
1588 | } | |
1589 | ||
d352ac68 CM |
1590 | /* |
1591 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1592 | * transaction commit. This does the actual creation. |
1593 | * | |
1594 | * Note: | |
1595 | * If the error which may affect the commitment of the current transaction | |
1596 | * happens, we should return the error number. If the error which just affect | |
1597 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1598 | */ |
80b6794d | 1599 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1600 | struct btrfs_pending_snapshot *pending) |
1601 | { | |
08d50ca3 NB |
1602 | |
1603 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3063d29f | 1604 | struct btrfs_key key; |
80b6794d | 1605 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1606 | struct btrfs_root *tree_root = fs_info->tree_root; |
1607 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1608 | struct btrfs_root *parent_root; |
98c9942a | 1609 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1610 | struct inode *parent_inode; |
42874b3d MX |
1611 | struct btrfs_path *path; |
1612 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1613 | struct dentry *dentry; |
3063d29f | 1614 | struct extent_buffer *tmp; |
925baedd | 1615 | struct extent_buffer *old; |
95582b00 | 1616 | struct timespec64 cur_time; |
aec8030a | 1617 | int ret = 0; |
d68fc57b | 1618 | u64 to_reserve = 0; |
6bdb72de | 1619 | u64 index = 0; |
a22285a6 | 1620 | u64 objectid; |
b83cc969 | 1621 | u64 root_flags; |
3063d29f | 1622 | |
8546b570 DS |
1623 | ASSERT(pending->path); |
1624 | path = pending->path; | |
42874b3d | 1625 | |
b0c0ea63 DS |
1626 | ASSERT(pending->root_item); |
1627 | new_root_item = pending->root_item; | |
a22285a6 | 1628 | |
543068a2 | 1629 | pending->error = btrfs_get_free_objectid(tree_root, &objectid); |
aec8030a | 1630 | if (pending->error) |
6fa9700e | 1631 | goto no_free_objectid; |
3063d29f | 1632 | |
d6726335 QW |
1633 | /* |
1634 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1635 | * accounted by later btrfs_qgroup_inherit(). | |
1636 | */ | |
1637 | btrfs_set_skip_qgroup(trans, objectid); | |
1638 | ||
147d256e | 1639 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
d68fc57b YZ |
1640 | |
1641 | if (to_reserve > 0) { | |
9270501c | 1642 | pending->error = btrfs_block_rsv_add(fs_info, |
aec8030a MX |
1643 | &pending->block_rsv, |
1644 | to_reserve, | |
1645 | BTRFS_RESERVE_NO_FLUSH); | |
1646 | if (pending->error) | |
d6726335 | 1647 | goto clear_skip_qgroup; |
d68fc57b YZ |
1648 | } |
1649 | ||
3063d29f | 1650 | key.objectid = objectid; |
a22285a6 YZ |
1651 | key.offset = (u64)-1; |
1652 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1653 | |
6fa9700e | 1654 | rsv = trans->block_rsv; |
a22285a6 | 1655 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1656 | trans->bytes_reserved = trans->block_rsv->reserved; |
0b246afa | 1657 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa JB |
1658 | trans->transid, |
1659 | trans->bytes_reserved, 1); | |
a22285a6 | 1660 | dentry = pending->dentry; |
e9662f70 | 1661 | parent_inode = pending->dir; |
a22285a6 | 1662 | parent_root = BTRFS_I(parent_inode)->root; |
f0118cb6 JB |
1663 | ret = record_root_in_trans(trans, parent_root, 0); |
1664 | if (ret) | |
1665 | goto fail; | |
c2050a45 | 1666 | cur_time = current_time(parent_inode); |
04b285f3 | 1667 | |
3063d29f CM |
1668 | /* |
1669 | * insert the directory item | |
1670 | */ | |
877574e2 | 1671 | ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); |
49b25e05 | 1672 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1673 | |
1674 | /* check if there is a file/dir which has the same name. */ | |
1675 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
4a0cc7ca | 1676 | btrfs_ino(BTRFS_I(parent_inode)), |
42874b3d MX |
1677 | dentry->d_name.name, |
1678 | dentry->d_name.len, 0); | |
1679 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1680 | pending->error = -EEXIST; |
aec8030a | 1681 | goto dir_item_existed; |
42874b3d MX |
1682 | } else if (IS_ERR(dir_item)) { |
1683 | ret = PTR_ERR(dir_item); | |
66642832 | 1684 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1685 | goto fail; |
79787eaa | 1686 | } |
42874b3d | 1687 | btrfs_release_path(path); |
52c26179 | 1688 | |
e999376f CM |
1689 | /* |
1690 | * pull in the delayed directory update | |
1691 | * and the delayed inode item | |
1692 | * otherwise we corrupt the FS during | |
1693 | * snapshot | |
1694 | */ | |
e5c304e6 | 1695 | ret = btrfs_run_delayed_items(trans); |
8732d44f | 1696 | if (ret) { /* Transaction aborted */ |
66642832 | 1697 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1698 | goto fail; |
1699 | } | |
e999376f | 1700 | |
f0118cb6 JB |
1701 | ret = record_root_in_trans(trans, root, 0); |
1702 | if (ret) { | |
1703 | btrfs_abort_transaction(trans, ret); | |
1704 | goto fail; | |
1705 | } | |
6bdb72de SW |
1706 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1707 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1708 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1709 | |
b83cc969 LZ |
1710 | root_flags = btrfs_root_flags(new_root_item); |
1711 | if (pending->readonly) | |
1712 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1713 | else | |
1714 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1715 | btrfs_set_root_flags(new_root_item, root_flags); | |
1716 | ||
8ea05e3a AB |
1717 | btrfs_set_root_generation_v2(new_root_item, |
1718 | trans->transid); | |
807fc790 | 1719 | generate_random_guid(new_root_item->uuid); |
8ea05e3a AB |
1720 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, |
1721 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1722 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1723 | memset(new_root_item->received_uuid, 0, | |
1724 | sizeof(new_root_item->received_uuid)); | |
1725 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1726 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1727 | btrfs_set_root_stransid(new_root_item, 0); | |
1728 | btrfs_set_root_rtransid(new_root_item, 0); | |
1729 | } | |
3cae210f QW |
1730 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1731 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1732 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1733 | |
6bdb72de | 1734 | old = btrfs_lock_root_node(root); |
9631e4cc JB |
1735 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, |
1736 | BTRFS_NESTING_COW); | |
79787eaa JM |
1737 | if (ret) { |
1738 | btrfs_tree_unlock(old); | |
1739 | free_extent_buffer(old); | |
66642832 | 1740 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1741 | goto fail; |
79787eaa | 1742 | } |
49b25e05 | 1743 | |
49b25e05 | 1744 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1745 | /* clean up in any case */ |
6bdb72de SW |
1746 | btrfs_tree_unlock(old); |
1747 | free_extent_buffer(old); | |
8732d44f | 1748 | if (ret) { |
66642832 | 1749 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1750 | goto fail; |
1751 | } | |
f1ebcc74 | 1752 | /* see comments in should_cow_block() */ |
27cdeb70 | 1753 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1754 | smp_wmb(); |
1755 | ||
6bdb72de | 1756 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1757 | /* record when the snapshot was created in key.offset */ |
1758 | key.offset = trans->transid; | |
1759 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1760 | btrfs_tree_unlock(tmp); |
1761 | free_extent_buffer(tmp); | |
8732d44f | 1762 | if (ret) { |
66642832 | 1763 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1764 | goto fail; |
1765 | } | |
6bdb72de | 1766 | |
a22285a6 YZ |
1767 | /* |
1768 | * insert root back/forward references | |
1769 | */ | |
6025c19f | 1770 | ret = btrfs_add_root_ref(trans, objectid, |
0660b5af | 1771 | parent_root->root_key.objectid, |
4a0cc7ca | 1772 | btrfs_ino(BTRFS_I(parent_inode)), index, |
a22285a6 | 1773 | dentry->d_name.name, dentry->d_name.len); |
8732d44f | 1774 | if (ret) { |
66642832 | 1775 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1776 | goto fail; |
1777 | } | |
0660b5af | 1778 | |
a22285a6 | 1779 | key.offset = (u64)-1; |
2dfb1e43 | 1780 | pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); |
79787eaa JM |
1781 | if (IS_ERR(pending->snap)) { |
1782 | ret = PTR_ERR(pending->snap); | |
2d892ccd | 1783 | pending->snap = NULL; |
66642832 | 1784 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1785 | goto fail; |
79787eaa | 1786 | } |
d68fc57b | 1787 | |
49b25e05 | 1788 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f | 1789 | if (ret) { |
66642832 | 1790 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1791 | goto fail; |
1792 | } | |
361048f5 | 1793 | |
6426c7ad QW |
1794 | /* |
1795 | * Do special qgroup accounting for snapshot, as we do some qgroup | |
1796 | * snapshot hack to do fast snapshot. | |
1797 | * To co-operate with that hack, we do hack again. | |
1798 | * Or snapshot will be greatly slowed down by a subtree qgroup rescan | |
1799 | */ | |
1800 | ret = qgroup_account_snapshot(trans, root, parent_root, | |
1801 | pending->inherit, objectid); | |
1802 | if (ret < 0) | |
1803 | goto fail; | |
1804 | ||
684572df LF |
1805 | ret = btrfs_insert_dir_item(trans, dentry->d_name.name, |
1806 | dentry->d_name.len, BTRFS_I(parent_inode), | |
1807 | &key, BTRFS_FT_DIR, index); | |
42874b3d | 1808 | /* We have check then name at the beginning, so it is impossible. */ |
9c52057c | 1809 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f | 1810 | if (ret) { |
66642832 | 1811 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1812 | goto fail; |
1813 | } | |
42874b3d | 1814 | |
6ef06d27 | 1815 | btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + |
42874b3d | 1816 | dentry->d_name.len * 2); |
c1867eb3 DS |
1817 | parent_inode->i_mtime = current_time(parent_inode); |
1818 | parent_inode->i_ctime = parent_inode->i_mtime; | |
729f7961 | 1819 | ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); |
dd5f9615 | 1820 | if (ret) { |
66642832 | 1821 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1822 | goto fail; |
1823 | } | |
807fc790 AS |
1824 | ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, |
1825 | BTRFS_UUID_KEY_SUBVOL, | |
cdb345a8 | 1826 | objectid); |
dd5f9615 | 1827 | if (ret) { |
66642832 | 1828 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1829 | goto fail; |
1830 | } | |
1831 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
cdb345a8 | 1832 | ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, |
dd5f9615 SB |
1833 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, |
1834 | objectid); | |
1835 | if (ret && ret != -EEXIST) { | |
66642832 | 1836 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1837 | goto fail; |
1838 | } | |
1839 | } | |
d6726335 | 1840 | |
3063d29f | 1841 | fail: |
aec8030a MX |
1842 | pending->error = ret; |
1843 | dir_item_existed: | |
98c9942a | 1844 | trans->block_rsv = rsv; |
2382c5cc | 1845 | trans->bytes_reserved = 0; |
d6726335 QW |
1846 | clear_skip_qgroup: |
1847 | btrfs_clear_skip_qgroup(trans); | |
6fa9700e MX |
1848 | no_free_objectid: |
1849 | kfree(new_root_item); | |
b0c0ea63 | 1850 | pending->root_item = NULL; |
42874b3d | 1851 | btrfs_free_path(path); |
8546b570 DS |
1852 | pending->path = NULL; |
1853 | ||
49b25e05 | 1854 | return ret; |
3063d29f CM |
1855 | } |
1856 | ||
d352ac68 CM |
1857 | /* |
1858 | * create all the snapshots we've scheduled for creation | |
1859 | */ | |
08d50ca3 | 1860 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) |
3de4586c | 1861 | { |
aec8030a | 1862 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1863 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1864 | int ret = 0; |
3de4586c | 1865 | |
aec8030a MX |
1866 | list_for_each_entry_safe(pending, next, head, list) { |
1867 | list_del(&pending->list); | |
08d50ca3 | 1868 | ret = create_pending_snapshot(trans, pending); |
aec8030a MX |
1869 | if (ret) |
1870 | break; | |
1871 | } | |
1872 | return ret; | |
3de4586c CM |
1873 | } |
1874 | ||
2ff7e61e | 1875 | static void update_super_roots(struct btrfs_fs_info *fs_info) |
5d4f98a2 YZ |
1876 | { |
1877 | struct btrfs_root_item *root_item; | |
1878 | struct btrfs_super_block *super; | |
1879 | ||
0b246afa | 1880 | super = fs_info->super_copy; |
5d4f98a2 | 1881 | |
0b246afa | 1882 | root_item = &fs_info->chunk_root->root_item; |
093e037c DS |
1883 | super->chunk_root = root_item->bytenr; |
1884 | super->chunk_root_generation = root_item->generation; | |
1885 | super->chunk_root_level = root_item->level; | |
5d4f98a2 | 1886 | |
0b246afa | 1887 | root_item = &fs_info->tree_root->root_item; |
093e037c DS |
1888 | super->root = root_item->bytenr; |
1889 | super->generation = root_item->generation; | |
1890 | super->root_level = root_item->level; | |
0b246afa | 1891 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) |
093e037c | 1892 | super->cache_generation = root_item->generation; |
94846229 BB |
1893 | else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) |
1894 | super->cache_generation = 0; | |
0b246afa | 1895 | if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) |
093e037c | 1896 | super->uuid_tree_generation = root_item->generation; |
5d4f98a2 YZ |
1897 | } |
1898 | ||
f36f3042 CM |
1899 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1900 | { | |
4a9d8bde | 1901 | struct btrfs_transaction *trans; |
f36f3042 | 1902 | int ret = 0; |
4a9d8bde | 1903 | |
a4abeea4 | 1904 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1905 | trans = info->running_transaction; |
1906 | if (trans) | |
1907 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1908 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1909 | return ret; |
1910 | } | |
1911 | ||
8929ecfa YZ |
1912 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1913 | { | |
4a9d8bde | 1914 | struct btrfs_transaction *trans; |
8929ecfa | 1915 | int ret = 0; |
4a9d8bde | 1916 | |
a4abeea4 | 1917 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1918 | trans = info->running_transaction; |
1919 | if (trans) | |
1920 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1921 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1922 | return ret; |
1923 | } | |
1924 | ||
fdfbf020 | 1925 | void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans) |
bb9c12c9 | 1926 | { |
3a45bb20 | 1927 | struct btrfs_fs_info *fs_info = trans->fs_info; |
bb9c12c9 SW |
1928 | struct btrfs_transaction *cur_trans; |
1929 | ||
fdfbf020 JB |
1930 | /* Kick the transaction kthread. */ |
1931 | set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); | |
1932 | wake_up_process(fs_info->transaction_kthread); | |
bb9c12c9 SW |
1933 | |
1934 | /* take transaction reference */ | |
bb9c12c9 | 1935 | cur_trans = trans->transaction; |
9b64f57d | 1936 | refcount_inc(&cur_trans->use_count); |
bb9c12c9 | 1937 | |
3a45bb20 | 1938 | btrfs_end_transaction(trans); |
6fc4e354 | 1939 | |
ae5d29d4 DS |
1940 | /* |
1941 | * Wait for the current transaction commit to start and block | |
1942 | * subsequent transaction joins | |
1943 | */ | |
3e738c53 | 1944 | btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); |
ae5d29d4 DS |
1945 | wait_event(fs_info->transaction_blocked_wait, |
1946 | cur_trans->state >= TRANS_STATE_COMMIT_START || | |
1947 | TRANS_ABORTED(cur_trans)); | |
724e2315 | 1948 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1949 | } |
1950 | ||
97cb39bb | 1951 | static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) |
49b25e05 | 1952 | { |
97cb39bb | 1953 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 JM |
1954 | struct btrfs_transaction *cur_trans = trans->transaction; |
1955 | ||
b50fff81 | 1956 | WARN_ON(refcount_read(&trans->use_count) > 1); |
49b25e05 | 1957 | |
66642832 | 1958 | btrfs_abort_transaction(trans, err); |
7b8b92af | 1959 | |
0b246afa | 1960 | spin_lock(&fs_info->trans_lock); |
66b6135b | 1961 | |
25d8c284 MX |
1962 | /* |
1963 | * If the transaction is removed from the list, it means this | |
1964 | * transaction has been committed successfully, so it is impossible | |
1965 | * to call the cleanup function. | |
1966 | */ | |
1967 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1968 | |
0b246afa | 1969 | if (cur_trans == fs_info->running_transaction) { |
4a9d8bde | 1970 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 1971 | spin_unlock(&fs_info->trans_lock); |
e1489b4f IA |
1972 | |
1973 | /* | |
1974 | * The thread has already released the lockdep map as reader | |
1975 | * already in btrfs_commit_transaction(). | |
1976 | */ | |
1977 | btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); | |
f094ac32 LB |
1978 | wait_event(cur_trans->writer_wait, |
1979 | atomic_read(&cur_trans->num_writers) == 1); | |
1980 | ||
0b246afa | 1981 | spin_lock(&fs_info->trans_lock); |
d7096fc3 | 1982 | } |
061dde82 FM |
1983 | |
1984 | /* | |
1985 | * Now that we know no one else is still using the transaction we can | |
1986 | * remove the transaction from the list of transactions. This avoids | |
1987 | * the transaction kthread from cleaning up the transaction while some | |
1988 | * other task is still using it, which could result in a use-after-free | |
1989 | * on things like log trees, as it forces the transaction kthread to | |
1990 | * wait for this transaction to be cleaned up by us. | |
1991 | */ | |
1992 | list_del_init(&cur_trans->list); | |
1993 | ||
0b246afa | 1994 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 1995 | |
2ff7e61e | 1996 | btrfs_cleanup_one_transaction(trans->transaction, fs_info); |
49b25e05 | 1997 | |
0b246afa JM |
1998 | spin_lock(&fs_info->trans_lock); |
1999 | if (cur_trans == fs_info->running_transaction) | |
2000 | fs_info->running_transaction = NULL; | |
2001 | spin_unlock(&fs_info->trans_lock); | |
4a9d8bde | 2002 | |
e0228285 | 2003 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2004 | sb_end_intwrite(fs_info->sb); |
724e2315 JB |
2005 | btrfs_put_transaction(cur_trans); |
2006 | btrfs_put_transaction(cur_trans); | |
49b25e05 | 2007 | |
2e4e97ab | 2008 | trace_btrfs_transaction_commit(fs_info); |
49b25e05 | 2009 | |
49b25e05 JM |
2010 | if (current->journal_info == trans) |
2011 | current->journal_info = NULL; | |
0b246afa | 2012 | btrfs_scrub_cancel(fs_info); |
49b25e05 JM |
2013 | |
2014 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
2015 | } | |
2016 | ||
c7cc64a9 DS |
2017 | /* |
2018 | * Release reserved delayed ref space of all pending block groups of the | |
2019 | * transaction and remove them from the list | |
2020 | */ | |
2021 | static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) | |
2022 | { | |
2023 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2024 | struct btrfs_block_group *block_group, *tmp; |
c7cc64a9 DS |
2025 | |
2026 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { | |
2027 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
2028 | list_del_init(&block_group->bg_list); | |
2029 | } | |
2030 | } | |
2031 | ||
88090ad3 | 2032 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
82436617 | 2033 | { |
ce8ea7cc | 2034 | /* |
a0f0cf83 | 2035 | * We use try_to_writeback_inodes_sb() here because if we used |
ce8ea7cc JB |
2036 | * btrfs_start_delalloc_roots we would deadlock with fs freeze. |
2037 | * Currently are holding the fs freeze lock, if we do an async flush | |
2038 | * we'll do btrfs_join_transaction() and deadlock because we need to | |
2039 | * wait for the fs freeze lock. Using the direct flushing we benefit | |
2040 | * from already being in a transaction and our join_transaction doesn't | |
2041 | * have to re-take the fs freeze lock. | |
a0f0cf83 FM |
2042 | * |
2043 | * Note that try_to_writeback_inodes_sb() will only trigger writeback | |
2044 | * if it can read lock sb->s_umount. It will always be able to lock it, | |
2045 | * except when the filesystem is being unmounted or being frozen, but in | |
2046 | * those cases sync_filesystem() is called, which results in calling | |
2047 | * writeback_inodes_sb() while holding a write lock on sb->s_umount. | |
2048 | * Note that we don't call writeback_inodes_sb() directly, because it | |
2049 | * will emit a warning if sb->s_umount is not locked. | |
ce8ea7cc | 2050 | */ |
88090ad3 | 2051 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
a0f0cf83 | 2052 | try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); |
82436617 MX |
2053 | return 0; |
2054 | } | |
2055 | ||
88090ad3 | 2056 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) |
82436617 | 2057 | { |
88090ad3 | 2058 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
6374e57a | 2059 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); |
82436617 MX |
2060 | } |
2061 | ||
28b21c55 FM |
2062 | /* |
2063 | * Add a pending snapshot associated with the given transaction handle to the | |
2064 | * respective handle. This must be called after the transaction commit started | |
2065 | * and while holding fs_info->trans_lock. | |
2066 | * This serves to guarantee a caller of btrfs_commit_transaction() that it can | |
2067 | * safely free the pending snapshot pointer in case btrfs_commit_transaction() | |
2068 | * returns an error. | |
2069 | */ | |
2070 | static void add_pending_snapshot(struct btrfs_trans_handle *trans) | |
2071 | { | |
2072 | struct btrfs_transaction *cur_trans = trans->transaction; | |
2073 | ||
2074 | if (!trans->pending_snapshot) | |
2075 | return; | |
2076 | ||
2077 | lockdep_assert_held(&trans->fs_info->trans_lock); | |
2078 | ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_START); | |
2079 | ||
2080 | list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots); | |
2081 | } | |
2082 | ||
e55958c8 IA |
2083 | static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval) |
2084 | { | |
2085 | fs_info->commit_stats.commit_count++; | |
2086 | fs_info->commit_stats.last_commit_dur = interval; | |
2087 | fs_info->commit_stats.max_commit_dur = | |
2088 | max_t(u64, fs_info->commit_stats.max_commit_dur, interval); | |
2089 | fs_info->commit_stats.total_commit_dur += interval; | |
2090 | } | |
2091 | ||
3a45bb20 | 2092 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans) |
79154b1b | 2093 | { |
3a45bb20 | 2094 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 | 2095 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 2096 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 2097 | int ret; |
e55958c8 IA |
2098 | ktime_t start_time; |
2099 | ktime_t interval; | |
79154b1b | 2100 | |
35b814f3 | 2101 | ASSERT(refcount_read(&trans->use_count) == 1); |
3e738c53 | 2102 | btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); |
35b814f3 | 2103 | |
8d25a086 | 2104 | /* Stop the commit early if ->aborted is set */ |
bf31f87f | 2105 | if (TRANS_ABORTED(cur_trans)) { |
25287e0a | 2106 | ret = cur_trans->aborted; |
3e738c53 | 2107 | goto lockdep_trans_commit_start_release; |
25287e0a | 2108 | } |
49b25e05 | 2109 | |
f45c752b JB |
2110 | btrfs_trans_release_metadata(trans); |
2111 | trans->block_rsv = NULL; | |
2112 | ||
56bec294 | 2113 | /* |
e19eb11f JB |
2114 | * We only want one transaction commit doing the flushing so we do not |
2115 | * waste a bunch of time on lock contention on the extent root node. | |
56bec294 | 2116 | */ |
e19eb11f JB |
2117 | if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, |
2118 | &cur_trans->delayed_refs.flags)) { | |
2119 | /* | |
2120 | * Make a pass through all the delayed refs we have so far. | |
2121 | * Any running threads may add more while we are here. | |
2122 | */ | |
2123 | ret = btrfs_run_delayed_refs(trans, 0); | |
3e738c53 IA |
2124 | if (ret) |
2125 | goto lockdep_trans_commit_start_release; | |
e19eb11f | 2126 | } |
56bec294 | 2127 | |
119e80df | 2128 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2129 | |
3204d33c | 2130 | if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { |
1bbc621e CM |
2131 | int run_it = 0; |
2132 | ||
2133 | /* this mutex is also taken before trying to set | |
2134 | * block groups readonly. We need to make sure | |
2135 | * that nobody has set a block group readonly | |
2136 | * after a extents from that block group have been | |
2137 | * allocated for cache files. btrfs_set_block_group_ro | |
2138 | * will wait for the transaction to commit if it | |
3204d33c | 2139 | * finds BTRFS_TRANS_DIRTY_BG_RUN set. |
1bbc621e | 2140 | * |
3204d33c JB |
2141 | * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure |
2142 | * only one process starts all the block group IO. It wouldn't | |
1bbc621e CM |
2143 | * hurt to have more than one go through, but there's no |
2144 | * real advantage to it either. | |
2145 | */ | |
0b246afa | 2146 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c JB |
2147 | if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, |
2148 | &cur_trans->flags)) | |
1bbc621e | 2149 | run_it = 1; |
0b246afa | 2150 | mutex_unlock(&fs_info->ro_block_group_mutex); |
1bbc621e | 2151 | |
f9cacae3 | 2152 | if (run_it) { |
21217054 | 2153 | ret = btrfs_start_dirty_block_groups(trans); |
3e738c53 IA |
2154 | if (ret) |
2155 | goto lockdep_trans_commit_start_release; | |
f9cacae3 | 2156 | } |
1bbc621e CM |
2157 | } |
2158 | ||
0b246afa | 2159 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2160 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { |
d0c2f4fa FM |
2161 | enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; |
2162 | ||
28b21c55 FM |
2163 | add_pending_snapshot(trans); |
2164 | ||
0b246afa | 2165 | spin_unlock(&fs_info->trans_lock); |
9b64f57d | 2166 | refcount_inc(&cur_trans->use_count); |
ccd467d6 | 2167 | |
d0c2f4fa FM |
2168 | if (trans->in_fsync) |
2169 | want_state = TRANS_STATE_SUPER_COMMITTED; | |
3e738c53 IA |
2170 | |
2171 | btrfs_trans_state_lockdep_release(fs_info, | |
2172 | BTRFS_LOCKDEP_TRANS_COMMIT_START); | |
d0c2f4fa FM |
2173 | ret = btrfs_end_transaction(trans); |
2174 | wait_for_commit(cur_trans, want_state); | |
15ee9bc7 | 2175 | |
bf31f87f | 2176 | if (TRANS_ABORTED(cur_trans)) |
b4924a0f LB |
2177 | ret = cur_trans->aborted; |
2178 | ||
724e2315 | 2179 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 2180 | |
49b25e05 | 2181 | return ret; |
79154b1b | 2182 | } |
4313b399 | 2183 | |
4a9d8bde | 2184 | cur_trans->state = TRANS_STATE_COMMIT_START; |
0b246afa | 2185 | wake_up(&fs_info->transaction_blocked_wait); |
3e738c53 | 2186 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); |
bb9c12c9 | 2187 | |
0b246afa | 2188 | if (cur_trans->list.prev != &fs_info->trans_list) { |
d0c2f4fa FM |
2189 | enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; |
2190 | ||
2191 | if (trans->in_fsync) | |
2192 | want_state = TRANS_STATE_SUPER_COMMITTED; | |
2193 | ||
ccd467d6 CM |
2194 | prev_trans = list_entry(cur_trans->list.prev, |
2195 | struct btrfs_transaction, list); | |
d0c2f4fa | 2196 | if (prev_trans->state < want_state) { |
9b64f57d | 2197 | refcount_inc(&prev_trans->use_count); |
0b246afa | 2198 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2199 | |
d0c2f4fa FM |
2200 | wait_for_commit(prev_trans, want_state); |
2201 | ||
bf31f87f | 2202 | ret = READ_ONCE(prev_trans->aborted); |
ccd467d6 | 2203 | |
724e2315 | 2204 | btrfs_put_transaction(prev_trans); |
1f9b8c8f | 2205 | if (ret) |
e1489b4f | 2206 | goto lockdep_release; |
a4abeea4 | 2207 | } else { |
0b246afa | 2208 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2209 | } |
a4abeea4 | 2210 | } else { |
0b246afa | 2211 | spin_unlock(&fs_info->trans_lock); |
cb2d3dad FM |
2212 | /* |
2213 | * The previous transaction was aborted and was already removed | |
2214 | * from the list of transactions at fs_info->trans_list. So we | |
2215 | * abort to prevent writing a new superblock that reflects a | |
2216 | * corrupt state (pointing to trees with unwritten nodes/leafs). | |
2217 | */ | |
84961539 | 2218 | if (BTRFS_FS_ERROR(fs_info)) { |
cb2d3dad | 2219 | ret = -EROFS; |
e1489b4f | 2220 | goto lockdep_release; |
cb2d3dad | 2221 | } |
ccd467d6 | 2222 | } |
15ee9bc7 | 2223 | |
e55958c8 IA |
2224 | /* |
2225 | * Get the time spent on the work done by the commit thread and not | |
2226 | * the time spent waiting on a previous commit | |
2227 | */ | |
2228 | start_time = ktime_get_ns(); | |
2229 | ||
0860adfd MX |
2230 | extwriter_counter_dec(cur_trans, trans->type); |
2231 | ||
88090ad3 | 2232 | ret = btrfs_start_delalloc_flush(fs_info); |
82436617 | 2233 | if (ret) |
e1489b4f | 2234 | goto lockdep_release; |
82436617 | 2235 | |
e5c304e6 | 2236 | ret = btrfs_run_delayed_items(trans); |
581227d0 | 2237 | if (ret) |
e1489b4f | 2238 | goto lockdep_release; |
15ee9bc7 | 2239 | |
5a9ba670 IA |
2240 | /* |
2241 | * The thread has started/joined the transaction thus it holds the | |
2242 | * lockdep map as a reader. It has to release it before acquiring the | |
2243 | * lockdep map as a writer. | |
2244 | */ | |
2245 | btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); | |
2246 | btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters); | |
581227d0 MX |
2247 | wait_event(cur_trans->writer_wait, |
2248 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 2249 | |
581227d0 | 2250 | /* some pending stuffs might be added after the previous flush. */ |
e5c304e6 | 2251 | ret = btrfs_run_delayed_items(trans); |
e1489b4f IA |
2252 | if (ret) { |
2253 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); | |
ca469637 | 2254 | goto cleanup_transaction; |
e1489b4f | 2255 | } |
ca469637 | 2256 | |
88090ad3 | 2257 | btrfs_wait_delalloc_flush(fs_info); |
cb7ab021 | 2258 | |
48778179 FM |
2259 | /* |
2260 | * Wait for all ordered extents started by a fast fsync that joined this | |
2261 | * transaction. Otherwise if this transaction commits before the ordered | |
2262 | * extents complete we lose logged data after a power failure. | |
2263 | */ | |
8b53779e | 2264 | btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered); |
48778179 FM |
2265 | wait_event(cur_trans->pending_wait, |
2266 | atomic_read(&cur_trans->pending_ordered) == 0); | |
2267 | ||
2ff7e61e | 2268 | btrfs_scrub_pause(fs_info); |
ed0ca140 JB |
2269 | /* |
2270 | * Ok now we need to make sure to block out any other joins while we | |
2271 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 2272 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 | 2273 | */ |
0b246afa | 2274 | spin_lock(&fs_info->trans_lock); |
28b21c55 | 2275 | add_pending_snapshot(trans); |
4a9d8bde | 2276 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 2277 | spin_unlock(&fs_info->trans_lock); |
e1489b4f IA |
2278 | |
2279 | /* | |
2280 | * The thread has started/joined the transaction thus it holds the | |
2281 | * lockdep map as a reader. It has to release it before acquiring the | |
2282 | * lockdep map as a writer. | |
2283 | */ | |
2284 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); | |
2285 | btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); | |
ed0ca140 JB |
2286 | wait_event(cur_trans->writer_wait, |
2287 | atomic_read(&cur_trans->num_writers) == 1); | |
2288 | ||
3e738c53 IA |
2289 | /* |
2290 | * Make lockdep happy by acquiring the state locks after | |
2291 | * btrfs_trans_num_writers is released. If we acquired the state locks | |
2292 | * before releasing the btrfs_trans_num_writers lock then lockdep would | |
2293 | * complain because we did not follow the reverse order unlocking rule. | |
2294 | */ | |
2295 | btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); | |
2296 | btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); | |
2297 | btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); | |
2298 | ||
fdfbf020 JB |
2299 | /* |
2300 | * We've started the commit, clear the flag in case we were triggered to | |
2301 | * do an async commit but somebody else started before the transaction | |
2302 | * kthread could do the work. | |
2303 | */ | |
2304 | clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); | |
2305 | ||
bf31f87f | 2306 | if (TRANS_ABORTED(cur_trans)) { |
2cba30f1 | 2307 | ret = cur_trans->aborted; |
3e738c53 | 2308 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); |
6cf7f77e | 2309 | goto scrub_continue; |
2cba30f1 | 2310 | } |
7585717f CM |
2311 | /* |
2312 | * the reloc mutex makes sure that we stop | |
2313 | * the balancing code from coming in and moving | |
2314 | * extents around in the middle of the commit | |
2315 | */ | |
0b246afa | 2316 | mutex_lock(&fs_info->reloc_mutex); |
7585717f | 2317 | |
42874b3d MX |
2318 | /* |
2319 | * We needn't worry about the delayed items because we will | |
2320 | * deal with them in create_pending_snapshot(), which is the | |
2321 | * core function of the snapshot creation. | |
2322 | */ | |
08d50ca3 | 2323 | ret = create_pending_snapshots(trans); |
56e9f6ea DS |
2324 | if (ret) |
2325 | goto unlock_reloc; | |
3063d29f | 2326 | |
42874b3d MX |
2327 | /* |
2328 | * We insert the dir indexes of the snapshots and update the inode | |
2329 | * of the snapshots' parents after the snapshot creation, so there | |
2330 | * are some delayed items which are not dealt with. Now deal with | |
2331 | * them. | |
2332 | * | |
2333 | * We needn't worry that this operation will corrupt the snapshots, | |
2334 | * because all the tree which are snapshoted will be forced to COW | |
2335 | * the nodes and leaves. | |
2336 | */ | |
e5c304e6 | 2337 | ret = btrfs_run_delayed_items(trans); |
56e9f6ea DS |
2338 | if (ret) |
2339 | goto unlock_reloc; | |
16cdcec7 | 2340 | |
c79a70b1 | 2341 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
56e9f6ea DS |
2342 | if (ret) |
2343 | goto unlock_reloc; | |
56bec294 | 2344 | |
e999376f CM |
2345 | /* |
2346 | * make sure none of the code above managed to slip in a | |
2347 | * delayed item | |
2348 | */ | |
ccdf9b30 | 2349 | btrfs_assert_delayed_root_empty(fs_info); |
e999376f | 2350 | |
2c90e5d6 | 2351 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 2352 | |
7e4443d9 | 2353 | ret = commit_fs_roots(trans); |
56e9f6ea | 2354 | if (ret) |
dfba78dc | 2355 | goto unlock_reloc; |
54aa1f4d | 2356 | |
3818aea2 | 2357 | /* |
7e1876ac DS |
2358 | * Since the transaction is done, we can apply the pending changes |
2359 | * before the next transaction. | |
3818aea2 | 2360 | */ |
0b246afa | 2361 | btrfs_apply_pending_changes(fs_info); |
3818aea2 | 2362 | |
5d4f98a2 | 2363 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2364 | * safe to free the root of tree log roots |
2365 | */ | |
0b246afa | 2366 | btrfs_free_log_root_tree(trans, fs_info); |
e02119d5 | 2367 | |
0ed4792a QW |
2368 | /* |
2369 | * Since fs roots are all committed, we can get a quite accurate | |
2370 | * new_roots. So let's do quota accounting. | |
2371 | */ | |
460fb20a | 2372 | ret = btrfs_qgroup_account_extents(trans); |
56e9f6ea | 2373 | if (ret < 0) |
dfba78dc | 2374 | goto unlock_reloc; |
0ed4792a | 2375 | |
9386d8bc | 2376 | ret = commit_cowonly_roots(trans); |
56e9f6ea | 2377 | if (ret) |
dfba78dc | 2378 | goto unlock_reloc; |
54aa1f4d | 2379 | |
2cba30f1 MX |
2380 | /* |
2381 | * The tasks which save the space cache and inode cache may also | |
2382 | * update ->aborted, check it. | |
2383 | */ | |
bf31f87f | 2384 | if (TRANS_ABORTED(cur_trans)) { |
2cba30f1 | 2385 | ret = cur_trans->aborted; |
dfba78dc | 2386 | goto unlock_reloc; |
2cba30f1 MX |
2387 | } |
2388 | ||
0b246afa | 2389 | cur_trans = fs_info->running_transaction; |
5d4f98a2 | 2390 | |
0b246afa JM |
2391 | btrfs_set_root_node(&fs_info->tree_root->root_item, |
2392 | fs_info->tree_root->node); | |
2393 | list_add_tail(&fs_info->tree_root->dirty_list, | |
9e351cc8 | 2394 | &cur_trans->switch_commits); |
5d4f98a2 | 2395 | |
0b246afa JM |
2396 | btrfs_set_root_node(&fs_info->chunk_root->root_item, |
2397 | fs_info->chunk_root->node); | |
2398 | list_add_tail(&fs_info->chunk_root->dirty_list, | |
9e351cc8 JB |
2399 | &cur_trans->switch_commits); |
2400 | ||
f7238e50 JB |
2401 | if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { |
2402 | btrfs_set_root_node(&fs_info->block_group_root->root_item, | |
2403 | fs_info->block_group_root->node); | |
2404 | list_add_tail(&fs_info->block_group_root->dirty_list, | |
2405 | &cur_trans->switch_commits); | |
2406 | } | |
2407 | ||
889bfa39 | 2408 | switch_commit_roots(trans); |
5d4f98a2 | 2409 | |
ce93ec54 | 2410 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2411 | ASSERT(list_empty(&cur_trans->io_bgs)); |
2ff7e61e | 2412 | update_super_roots(fs_info); |
e02119d5 | 2413 | |
0b246afa JM |
2414 | btrfs_set_super_log_root(fs_info->super_copy, 0); |
2415 | btrfs_set_super_log_root_level(fs_info->super_copy, 0); | |
2416 | memcpy(fs_info->super_for_commit, fs_info->super_copy, | |
2417 | sizeof(*fs_info->super_copy)); | |
ccd467d6 | 2418 | |
bbbf7243 | 2419 | btrfs_commit_device_sizes(cur_trans); |
935e5cc9 | 2420 | |
0b246afa JM |
2421 | clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
2422 | clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); | |
656f30db | 2423 | |
4fbcdf66 FM |
2424 | btrfs_trans_release_chunk_metadata(trans); |
2425 | ||
dfba78dc FM |
2426 | /* |
2427 | * Before changing the transaction state to TRANS_STATE_UNBLOCKED and | |
2428 | * setting fs_info->running_transaction to NULL, lock tree_log_mutex to | |
2429 | * make sure that before we commit our superblock, no other task can | |
2430 | * start a new transaction and commit a log tree before we commit our | |
2431 | * superblock. Anyone trying to commit a log tree locks this mutex before | |
2432 | * writing its superblock. | |
2433 | */ | |
2434 | mutex_lock(&fs_info->tree_log_mutex); | |
2435 | ||
0b246afa | 2436 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2437 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
0b246afa JM |
2438 | fs_info->running_transaction = NULL; |
2439 | spin_unlock(&fs_info->trans_lock); | |
2440 | mutex_unlock(&fs_info->reloc_mutex); | |
b7ec40d7 | 2441 | |
0b246afa | 2442 | wake_up(&fs_info->transaction_wait); |
3e738c53 | 2443 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); |
e6dcd2dc | 2444 | |
70458a58 | 2445 | ret = btrfs_write_and_wait_transaction(trans); |
49b25e05 | 2446 | if (ret) { |
0b246afa JM |
2447 | btrfs_handle_fs_error(fs_info, ret, |
2448 | "Error while writing out transaction"); | |
2449 | mutex_unlock(&fs_info->tree_log_mutex); | |
6cf7f77e | 2450 | goto scrub_continue; |
49b25e05 JM |
2451 | } |
2452 | ||
d3575156 NA |
2453 | /* |
2454 | * At this point, we should have written all the tree blocks allocated | |
2455 | * in this transaction. So it's now safe to free the redirtyied extent | |
2456 | * buffers. | |
2457 | */ | |
2458 | btrfs_free_redirty_list(cur_trans); | |
2459 | ||
eece6a9c | 2460 | ret = write_all_supers(fs_info, 0); |
e02119d5 CM |
2461 | /* |
2462 | * the super is written, we can safely allow the tree-loggers | |
2463 | * to go about their business | |
2464 | */ | |
0b246afa | 2465 | mutex_unlock(&fs_info->tree_log_mutex); |
c1f32b7c AJ |
2466 | if (ret) |
2467 | goto scrub_continue; | |
e02119d5 | 2468 | |
d0c2f4fa FM |
2469 | /* |
2470 | * We needn't acquire the lock here because there is no other task | |
2471 | * which can change it. | |
2472 | */ | |
2473 | cur_trans->state = TRANS_STATE_SUPER_COMMITTED; | |
2474 | wake_up(&cur_trans->commit_wait); | |
3e738c53 | 2475 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); |
d0c2f4fa | 2476 | |
5ead2dd0 | 2477 | btrfs_finish_extent_commit(trans); |
4313b399 | 2478 | |
3204d33c | 2479 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) |
0b246afa | 2480 | btrfs_clear_space_info_full(fs_info); |
13212b54 | 2481 | |
0b246afa | 2482 | fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
2483 | /* |
2484 | * We needn't acquire the lock here because there is no other task | |
2485 | * which can change it. | |
2486 | */ | |
2487 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2488 | wake_up(&cur_trans->commit_wait); |
3e738c53 | 2489 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); |
3de4586c | 2490 | |
0b246afa | 2491 | spin_lock(&fs_info->trans_lock); |
13c5a93e | 2492 | list_del_init(&cur_trans->list); |
0b246afa | 2493 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 2494 | |
724e2315 JB |
2495 | btrfs_put_transaction(cur_trans); |
2496 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2497 | |
0860adfd | 2498 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2499 | sb_end_intwrite(fs_info->sb); |
b2b5ef5c | 2500 | |
2e4e97ab | 2501 | trace_btrfs_transaction_commit(fs_info); |
1abe9b8a | 2502 | |
e55958c8 IA |
2503 | interval = ktime_get_ns() - start_time; |
2504 | ||
2ff7e61e | 2505 | btrfs_scrub_continue(fs_info); |
a2de733c | 2506 | |
9ed74f2d JB |
2507 | if (current->journal_info == trans) |
2508 | current->journal_info = NULL; | |
2509 | ||
2c90e5d6 | 2510 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2511 | |
e55958c8 IA |
2512 | update_commit_stats(fs_info, interval); |
2513 | ||
79154b1b | 2514 | return ret; |
49b25e05 | 2515 | |
56e9f6ea DS |
2516 | unlock_reloc: |
2517 | mutex_unlock(&fs_info->reloc_mutex); | |
3e738c53 | 2518 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); |
6cf7f77e | 2519 | scrub_continue: |
3e738c53 IA |
2520 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); |
2521 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); | |
2ff7e61e | 2522 | btrfs_scrub_continue(fs_info); |
49b25e05 | 2523 | cleanup_transaction: |
dc60c525 | 2524 | btrfs_trans_release_metadata(trans); |
c7cc64a9 | 2525 | btrfs_cleanup_pending_block_groups(trans); |
4fbcdf66 | 2526 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2527 | trans->block_rsv = NULL; |
0b246afa | 2528 | btrfs_warn(fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2529 | if (current->journal_info == trans) |
2530 | current->journal_info = NULL; | |
97cb39bb | 2531 | cleanup_transaction(trans, ret); |
49b25e05 JM |
2532 | |
2533 | return ret; | |
e1489b4f IA |
2534 | |
2535 | lockdep_release: | |
5a9ba670 | 2536 | btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); |
e1489b4f IA |
2537 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); |
2538 | goto cleanup_transaction; | |
3e738c53 IA |
2539 | |
2540 | lockdep_trans_commit_start_release: | |
2541 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); | |
2542 | btrfs_end_transaction(trans); | |
2543 | return ret; | |
79154b1b CM |
2544 | } |
2545 | ||
d352ac68 | 2546 | /* |
9d1a2a3a DS |
2547 | * return < 0 if error |
2548 | * 0 if there are no more dead_roots at the time of call | |
2549 | * 1 there are more to be processed, call me again | |
2550 | * | |
2551 | * The return value indicates there are certainly more snapshots to delete, but | |
2552 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2553 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2554 | * few seconds later. | |
d352ac68 | 2555 | */ |
33c44184 | 2556 | int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info) |
e9d0b13b | 2557 | { |
33c44184 | 2558 | struct btrfs_root *root; |
9d1a2a3a | 2559 | int ret; |
5d4f98a2 | 2560 | |
a4abeea4 | 2561 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2562 | if (list_empty(&fs_info->dead_roots)) { |
2563 | spin_unlock(&fs_info->trans_lock); | |
2564 | return 0; | |
2565 | } | |
2566 | root = list_first_entry(&fs_info->dead_roots, | |
2567 | struct btrfs_root, root_list); | |
cfad392b | 2568 | list_del_init(&root->root_list); |
a4abeea4 | 2569 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2570 | |
4fd786e6 | 2571 | btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); |
76dda93c | 2572 | |
9d1a2a3a | 2573 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2574 | |
9d1a2a3a DS |
2575 | if (btrfs_header_backref_rev(root->node) < |
2576 | BTRFS_MIXED_BACKREF_REV) | |
0078a9f9 | 2577 | ret = btrfs_drop_snapshot(root, 0, 0); |
9d1a2a3a | 2578 | else |
0078a9f9 | 2579 | ret = btrfs_drop_snapshot(root, 1, 0); |
32471dc2 | 2580 | |
dc9492c1 | 2581 | btrfs_put_root(root); |
6596a928 | 2582 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2583 | } |
572d9ab7 DS |
2584 | |
2585 | void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) | |
2586 | { | |
2587 | unsigned long prev; | |
2588 | unsigned long bit; | |
2589 | ||
6c9fe14f | 2590 | prev = xchg(&fs_info->pending_changes, 0); |
572d9ab7 DS |
2591 | if (!prev) |
2592 | return; | |
2593 | ||
d51033d0 DS |
2594 | bit = 1 << BTRFS_PENDING_COMMIT; |
2595 | if (prev & bit) | |
2596 | btrfs_debug(fs_info, "pending commit done"); | |
2597 | prev &= ~bit; | |
2598 | ||
572d9ab7 DS |
2599 | if (prev) |
2600 | btrfs_warn(fs_info, | |
2601 | "unknown pending changes left 0x%lx, ignoring", prev); | |
2602 | } |