Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 CM |
4 | */ |
5 | ||
79154b1b | 6 | #include <linux/fs.h> |
5a0e3ad6 | 7 | #include <linux/slab.h> |
34088780 | 8 | #include <linux/sched.h> |
d3c2fdcf | 9 | #include <linux/writeback.h> |
5f39d397 | 10 | #include <linux/pagemap.h> |
5f2cc086 | 11 | #include <linux/blkdev.h> |
8ea05e3a | 12 | #include <linux/uuid.h> |
602cbe91 | 13 | #include "misc.h" |
79154b1b CM |
14 | #include "ctree.h" |
15 | #include "disk-io.h" | |
16 | #include "transaction.h" | |
925baedd | 17 | #include "locking.h" |
e02119d5 | 18 | #include "tree-log.h" |
733f4fbb | 19 | #include "volumes.h" |
8dabb742 | 20 | #include "dev-replace.h" |
fcebe456 | 21 | #include "qgroup.h" |
aac0023c | 22 | #include "block-group.h" |
9c343784 | 23 | #include "space-info.h" |
d3575156 | 24 | #include "zoned.h" |
79154b1b | 25 | |
0f7d52f4 CM |
26 | #define BTRFS_ROOT_TRANS_TAG 0 |
27 | ||
61c047b5 QW |
28 | /* |
29 | * Transaction states and transitions | |
30 | * | |
31 | * No running transaction (fs tree blocks are not modified) | |
32 | * | | |
33 | * | To next stage: | |
34 | * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). | |
35 | * V | |
36 | * Transaction N [[TRANS_STATE_RUNNING]] | |
37 | * | | |
38 | * | New trans handles can be attached to transaction N by calling all | |
39 | * | start_transaction() variants. | |
40 | * | | |
41 | * | To next stage: | |
42 | * | Call btrfs_commit_transaction() on any trans handle attached to | |
43 | * | transaction N | |
44 | * V | |
45 | * Transaction N [[TRANS_STATE_COMMIT_START]] | |
46 | * | | |
47 | * | Will wait for previous running transaction to completely finish if there | |
48 | * | is one | |
49 | * | | |
50 | * | Then one of the following happes: | |
51 | * | - Wait for all other trans handle holders to release. | |
52 | * | The btrfs_commit_transaction() caller will do the commit work. | |
53 | * | - Wait for current transaction to be committed by others. | |
54 | * | Other btrfs_commit_transaction() caller will do the commit work. | |
55 | * | | |
56 | * | At this stage, only btrfs_join_transaction*() variants can attach | |
57 | * | to this running transaction. | |
58 | * | All other variants will wait for current one to finish and attach to | |
59 | * | transaction N+1. | |
60 | * | | |
61 | * | To next stage: | |
62 | * | Caller is chosen to commit transaction N, and all other trans handle | |
63 | * | haven been released. | |
64 | * V | |
65 | * Transaction N [[TRANS_STATE_COMMIT_DOING]] | |
66 | * | | |
67 | * | The heavy lifting transaction work is started. | |
68 | * | From running delayed refs (modifying extent tree) to creating pending | |
69 | * | snapshots, running qgroups. | |
70 | * | In short, modify supporting trees to reflect modifications of subvolume | |
71 | * | trees. | |
72 | * | | |
73 | * | At this stage, all start_transaction() calls will wait for this | |
74 | * | transaction to finish and attach to transaction N+1. | |
75 | * | | |
76 | * | To next stage: | |
77 | * | Until all supporting trees are updated. | |
78 | * V | |
79 | * Transaction N [[TRANS_STATE_UNBLOCKED]] | |
80 | * | Transaction N+1 | |
81 | * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] | |
82 | * | need to write them back to disk and update | | |
83 | * | super blocks. | | |
84 | * | | | |
85 | * | At this stage, new transaction is allowed to | | |
86 | * | start. | | |
87 | * | All new start_transaction() calls will be | | |
88 | * | attached to transid N+1. | | |
89 | * | | | |
90 | * | To next stage: | | |
91 | * | Until all tree blocks are super blocks are | | |
92 | * | written to block devices | | |
93 | * V | | |
94 | * Transaction N [[TRANS_STATE_COMPLETED]] V | |
95 | * All tree blocks and super blocks are written. Transaction N+1 | |
96 | * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] | |
97 | * data structures will be cleaned up. | Life goes on | |
98 | */ | |
e8c9f186 | 99 | static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { |
4a9d8bde | 100 | [TRANS_STATE_RUNNING] = 0U, |
bcf3a3e7 NB |
101 | [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), |
102 | [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | | |
4a9d8bde | 103 | __TRANS_ATTACH | |
a6d155d2 FM |
104 | __TRANS_JOIN | |
105 | __TRANS_JOIN_NOSTART), | |
bcf3a3e7 | 106 | [TRANS_STATE_UNBLOCKED] = (__TRANS_START | |
4a9d8bde MX |
107 | __TRANS_ATTACH | |
108 | __TRANS_JOIN | | |
a6d155d2 FM |
109 | __TRANS_JOIN_NOLOCK | |
110 | __TRANS_JOIN_NOSTART), | |
d0c2f4fa FM |
111 | [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START | |
112 | __TRANS_ATTACH | | |
113 | __TRANS_JOIN | | |
114 | __TRANS_JOIN_NOLOCK | | |
115 | __TRANS_JOIN_NOSTART), | |
bcf3a3e7 | 116 | [TRANS_STATE_COMPLETED] = (__TRANS_START | |
4a9d8bde MX |
117 | __TRANS_ATTACH | |
118 | __TRANS_JOIN | | |
a6d155d2 FM |
119 | __TRANS_JOIN_NOLOCK | |
120 | __TRANS_JOIN_NOSTART), | |
4a9d8bde MX |
121 | }; |
122 | ||
724e2315 | 123 | void btrfs_put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 124 | { |
9b64f57d ER |
125 | WARN_ON(refcount_read(&transaction->use_count) == 0); |
126 | if (refcount_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 127 | BUG_ON(!list_empty(&transaction->list)); |
5c9d028b LB |
128 | WARN_ON(!RB_EMPTY_ROOT( |
129 | &transaction->delayed_refs.href_root.rb_root)); | |
81f7eb00 JM |
130 | WARN_ON(!RB_EMPTY_ROOT( |
131 | &transaction->delayed_refs.dirty_extent_root)); | |
1262133b | 132 | if (transaction->delayed_refs.pending_csums) |
ab8d0fc4 JM |
133 | btrfs_err(transaction->fs_info, |
134 | "pending csums is %llu", | |
135 | transaction->delayed_refs.pending_csums); | |
7785a663 FM |
136 | /* |
137 | * If any block groups are found in ->deleted_bgs then it's | |
138 | * because the transaction was aborted and a commit did not | |
139 | * happen (things failed before writing the new superblock | |
140 | * and calling btrfs_finish_extent_commit()), so we can not | |
141 | * discard the physical locations of the block groups. | |
142 | */ | |
143 | while (!list_empty(&transaction->deleted_bgs)) { | |
32da5386 | 144 | struct btrfs_block_group *cache; |
7785a663 FM |
145 | |
146 | cache = list_first_entry(&transaction->deleted_bgs, | |
32da5386 | 147 | struct btrfs_block_group, |
7785a663 FM |
148 | bg_list); |
149 | list_del_init(&cache->bg_list); | |
6b7304af | 150 | btrfs_unfreeze_block_group(cache); |
7785a663 FM |
151 | btrfs_put_block_group(cache); |
152 | } | |
bbbf7243 | 153 | WARN_ON(!list_empty(&transaction->dev_update_list)); |
4b5faeac | 154 | kfree(transaction); |
78fae27e | 155 | } |
79154b1b CM |
156 | } |
157 | ||
889bfa39 | 158 | static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) |
817d52f8 | 159 | { |
889bfa39 | 160 | struct btrfs_transaction *cur_trans = trans->transaction; |
16916a88 | 161 | struct btrfs_fs_info *fs_info = trans->fs_info; |
9e351cc8 | 162 | struct btrfs_root *root, *tmp; |
27d56e62 | 163 | struct btrfs_caching_control *caching_ctl, *next; |
9e351cc8 | 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 JB |
197 | |
198 | /* | |
199 | * We have to update the last_byte_to_unpin under the commit_root_sem, | |
200 | * at the same time we swap out the commit roots. | |
201 | * | |
202 | * This is because we must have a real view of the last spot the caching | |
203 | * kthreads were while caching. Consider the following views of the | |
204 | * extent tree for a block group | |
205 | * | |
206 | * commit root | |
207 | * +----+----+----+----+----+----+----+ | |
208 | * |\\\\| |\\\\|\\\\| |\\\\|\\\\| | |
209 | * +----+----+----+----+----+----+----+ | |
210 | * 0 1 2 3 4 5 6 7 | |
211 | * | |
212 | * new commit root | |
213 | * +----+----+----+----+----+----+----+ | |
214 | * | | | |\\\\| | |\\\\| | |
215 | * +----+----+----+----+----+----+----+ | |
216 | * 0 1 2 3 4 5 6 7 | |
217 | * | |
218 | * If the cache_ctl->progress was at 3, then we are only allowed to | |
219 | * unpin [0,1) and [2,3], because the caching thread has already | |
220 | * processed those extents. We are not allowed to unpin [5,6), because | |
221 | * the caching thread will re-start it's search from 3, and thus find | |
222 | * the hole from [4,6) to add to the free space cache. | |
223 | */ | |
bbb86a37 | 224 | spin_lock(&fs_info->block_group_cache_lock); |
27d56e62 JB |
225 | list_for_each_entry_safe(caching_ctl, next, |
226 | &fs_info->caching_block_groups, list) { | |
227 | struct btrfs_block_group *cache = caching_ctl->block_group; | |
228 | ||
229 | if (btrfs_block_group_done(cache)) { | |
230 | cache->last_byte_to_unpin = (u64)-1; | |
231 | list_del_init(&caching_ctl->list); | |
232 | btrfs_put_caching_control(caching_ctl); | |
233 | } else { | |
234 | cache->last_byte_to_unpin = caching_ctl->progress; | |
235 | } | |
236 | } | |
bbb86a37 | 237 | spin_unlock(&fs_info->block_group_cache_lock); |
9e351cc8 | 238 | up_write(&fs_info->commit_root_sem); |
817d52f8 JB |
239 | } |
240 | ||
0860adfd MX |
241 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
242 | unsigned int type) | |
243 | { | |
244 | if (type & TRANS_EXTWRITERS) | |
245 | atomic_inc(&trans->num_extwriters); | |
246 | } | |
247 | ||
248 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
249 | unsigned int type) | |
250 | { | |
251 | if (type & TRANS_EXTWRITERS) | |
252 | atomic_dec(&trans->num_extwriters); | |
253 | } | |
254 | ||
255 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
256 | unsigned int type) | |
257 | { | |
258 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
259 | } | |
260 | ||
261 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
262 | { | |
263 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
264 | } |
265 | ||
fb6dea26 | 266 | /* |
79bd3712 FM |
267 | * To be called after doing the chunk btree updates right after allocating a new |
268 | * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a | |
269 | * chunk after all chunk btree updates and after finishing the second phase of | |
270 | * chunk allocation (btrfs_create_pending_block_groups()) in case some block | |
271 | * group had its chunk item insertion delayed to the second phase. | |
fb6dea26 JB |
272 | */ |
273 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
274 | { | |
275 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
276 | ||
277 | if (!trans->chunk_bytes_reserved) | |
278 | return; | |
279 | ||
fb6dea26 | 280 | btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, |
63f018be | 281 | trans->chunk_bytes_reserved, NULL); |
fb6dea26 JB |
282 | trans->chunk_bytes_reserved = 0; |
283 | } | |
284 | ||
d352ac68 CM |
285 | /* |
286 | * either allocate a new transaction or hop into the existing one | |
287 | */ | |
2ff7e61e JM |
288 | static noinline int join_transaction(struct btrfs_fs_info *fs_info, |
289 | unsigned int type) | |
79154b1b CM |
290 | { |
291 | struct btrfs_transaction *cur_trans; | |
a4abeea4 | 292 | |
19ae4e81 | 293 | spin_lock(&fs_info->trans_lock); |
d43317dc | 294 | loop: |
49b25e05 | 295 | /* The file system has been taken offline. No new transactions. */ |
84961539 | 296 | if (BTRFS_FS_ERROR(fs_info)) { |
19ae4e81 | 297 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
298 | return -EROFS; |
299 | } | |
300 | ||
19ae4e81 | 301 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 302 | if (cur_trans) { |
bf31f87f | 303 | if (TRANS_ABORTED(cur_trans)) { |
19ae4e81 | 304 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 305 | return cur_trans->aborted; |
871383be | 306 | } |
4a9d8bde | 307 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
308 | spin_unlock(&fs_info->trans_lock); |
309 | return -EBUSY; | |
310 | } | |
9b64f57d | 311 | refcount_inc(&cur_trans->use_count); |
13c5a93e | 312 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 313 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 314 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 315 | return 0; |
79154b1b | 316 | } |
19ae4e81 | 317 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 318 | |
354aa0fb MX |
319 | /* |
320 | * If we are ATTACH, we just want to catch the current transaction, | |
321 | * and commit it. If there is no transaction, just return ENOENT. | |
322 | */ | |
323 | if (type == TRANS_ATTACH) | |
324 | return -ENOENT; | |
325 | ||
4a9d8bde MX |
326 | /* |
327 | * JOIN_NOLOCK only happens during the transaction commit, so | |
328 | * it is impossible that ->running_transaction is NULL | |
329 | */ | |
330 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
331 | ||
4b5faeac | 332 | cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); |
a4abeea4 JB |
333 | if (!cur_trans) |
334 | return -ENOMEM; | |
d43317dc | 335 | |
19ae4e81 JS |
336 | spin_lock(&fs_info->trans_lock); |
337 | if (fs_info->running_transaction) { | |
d43317dc CM |
338 | /* |
339 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 340 | * to redo the checks above |
d43317dc | 341 | */ |
4b5faeac | 342 | kfree(cur_trans); |
d43317dc | 343 | goto loop; |
84961539 | 344 | } else if (BTRFS_FS_ERROR(fs_info)) { |
e4b50e14 | 345 | spin_unlock(&fs_info->trans_lock); |
4b5faeac | 346 | kfree(cur_trans); |
7b8b92af | 347 | return -EROFS; |
79154b1b | 348 | } |
d43317dc | 349 | |
ab8d0fc4 | 350 | cur_trans->fs_info = fs_info; |
48778179 FM |
351 | atomic_set(&cur_trans->pending_ordered, 0); |
352 | init_waitqueue_head(&cur_trans->pending_wait); | |
a4abeea4 | 353 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 354 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
355 | init_waitqueue_head(&cur_trans->writer_wait); |
356 | init_waitqueue_head(&cur_trans->commit_wait); | |
4a9d8bde | 357 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
358 | /* |
359 | * One for this trans handle, one so it will live on until we | |
360 | * commit the transaction. | |
361 | */ | |
9b64f57d | 362 | refcount_set(&cur_trans->use_count, 2); |
3204d33c | 363 | cur_trans->flags = 0; |
afd48513 | 364 | cur_trans->start_time = ktime_get_seconds(); |
a4abeea4 | 365 | |
a099d0fd AM |
366 | memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); |
367 | ||
5c9d028b | 368 | cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; |
3368d001 | 369 | cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; |
d7df2c79 | 370 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
20b297d6 JS |
371 | |
372 | /* | |
373 | * although the tree mod log is per file system and not per transaction, | |
374 | * the log must never go across transaction boundaries. | |
375 | */ | |
376 | smp_mb(); | |
31b1a2bd | 377 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
5d163e0e | 378 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); |
31b1a2bd | 379 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
5d163e0e | 380 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); |
fc36ed7e | 381 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 382 | |
a4abeea4 JB |
383 | spin_lock_init(&cur_trans->delayed_refs.lock); |
384 | ||
385 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
bbbf7243 | 386 | INIT_LIST_HEAD(&cur_trans->dev_update_list); |
9e351cc8 | 387 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
ce93ec54 | 388 | INIT_LIST_HEAD(&cur_trans->dirty_bgs); |
1bbc621e | 389 | INIT_LIST_HEAD(&cur_trans->io_bgs); |
2b9dbef2 | 390 | INIT_LIST_HEAD(&cur_trans->dropped_roots); |
1bbc621e | 391 | mutex_init(&cur_trans->cache_write_mutex); |
ce93ec54 | 392 | spin_lock_init(&cur_trans->dirty_bgs_lock); |
e33e17ee | 393 | INIT_LIST_HEAD(&cur_trans->deleted_bgs); |
2b9dbef2 | 394 | spin_lock_init(&cur_trans->dropped_roots_lock); |
d3575156 NA |
395 | INIT_LIST_HEAD(&cur_trans->releasing_ebs); |
396 | spin_lock_init(&cur_trans->releasing_ebs_lock); | |
19ae4e81 | 397 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
c258d6e3 | 398 | extent_io_tree_init(fs_info, &cur_trans->dirty_pages, |
43eb5f29 | 399 | IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode); |
fe119a6e NB |
400 | extent_io_tree_init(fs_info, &cur_trans->pinned_extents, |
401 | IO_TREE_FS_PINNED_EXTENTS, NULL); | |
19ae4e81 JS |
402 | fs_info->generation++; |
403 | cur_trans->transid = fs_info->generation; | |
404 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 405 | cur_trans->aborted = 0; |
19ae4e81 | 406 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 407 | |
79154b1b CM |
408 | return 0; |
409 | } | |
410 | ||
d352ac68 | 411 | /* |
92a7cc42 QW |
412 | * This does all the record keeping required to make sure that a shareable root |
413 | * is properly recorded in a given transaction. This is required to make sure | |
414 | * the old root from before we joined the transaction is deleted when the | |
415 | * transaction commits. | |
d352ac68 | 416 | */ |
7585717f | 417 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
6426c7ad QW |
418 | struct btrfs_root *root, |
419 | int force) | |
6702ed49 | 420 | { |
0b246afa | 421 | struct btrfs_fs_info *fs_info = root->fs_info; |
03a7e111 | 422 | int ret = 0; |
0b246afa | 423 | |
92a7cc42 | 424 | if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
6426c7ad | 425 | root->last_trans < trans->transid) || force) { |
4d31778a | 426 | WARN_ON(!force && root->commit_root != root->node); |
5d4f98a2 | 427 | |
7585717f | 428 | /* |
27cdeb70 | 429 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
430 | * we have the reloc mutex held now, so there |
431 | * is only one writer in this function | |
432 | */ | |
27cdeb70 | 433 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 434 | |
27cdeb70 | 435 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
436 | * they find our root->last_trans update |
437 | */ | |
438 | smp_wmb(); | |
439 | ||
0b246afa | 440 | spin_lock(&fs_info->fs_roots_radix_lock); |
6426c7ad | 441 | if (root->last_trans == trans->transid && !force) { |
0b246afa | 442 | spin_unlock(&fs_info->fs_roots_radix_lock); |
a4abeea4 JB |
443 | return 0; |
444 | } | |
0b246afa JM |
445 | radix_tree_tag_set(&fs_info->fs_roots_radix, |
446 | (unsigned long)root->root_key.objectid, | |
447 | BTRFS_ROOT_TRANS_TAG); | |
448 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
7585717f CM |
449 | root->last_trans = trans->transid; |
450 | ||
451 | /* this is pretty tricky. We don't want to | |
452 | * take the relocation lock in btrfs_record_root_in_trans | |
453 | * unless we're really doing the first setup for this root in | |
454 | * this transaction. | |
455 | * | |
456 | * Normally we'd use root->last_trans as a flag to decide | |
457 | * if we want to take the expensive mutex. | |
458 | * | |
459 | * But, we have to set root->last_trans before we | |
460 | * init the relocation root, otherwise, we trip over warnings | |
461 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 462 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
463 | * fixing up the reloc trees and everyone must wait. |
464 | * | |
465 | * When this is zero, they can trust root->last_trans and fly | |
466 | * through btrfs_record_root_in_trans without having to take the | |
467 | * lock. smp_wmb() makes sure that all the writes above are | |
468 | * done before we pop in the zero below | |
469 | */ | |
03a7e111 | 470 | ret = btrfs_init_reloc_root(trans, root); |
c7548af6 | 471 | smp_mb__before_atomic(); |
27cdeb70 | 472 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 | 473 | } |
03a7e111 | 474 | return ret; |
5d4f98a2 | 475 | } |
bcc63abb | 476 | |
7585717f | 477 | |
2b9dbef2 JB |
478 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
479 | struct btrfs_root *root) | |
480 | { | |
0b246afa | 481 | struct btrfs_fs_info *fs_info = root->fs_info; |
2b9dbef2 JB |
482 | struct btrfs_transaction *cur_trans = trans->transaction; |
483 | ||
484 | /* Add ourselves to the transaction dropped list */ | |
485 | spin_lock(&cur_trans->dropped_roots_lock); | |
486 | list_add_tail(&root->root_list, &cur_trans->dropped_roots); | |
487 | spin_unlock(&cur_trans->dropped_roots_lock); | |
488 | ||
489 | /* Make sure we don't try to update the root at commit time */ | |
0b246afa JM |
490 | spin_lock(&fs_info->fs_roots_radix_lock); |
491 | radix_tree_tag_clear(&fs_info->fs_roots_radix, | |
2b9dbef2 JB |
492 | (unsigned long)root->root_key.objectid, |
493 | BTRFS_ROOT_TRANS_TAG); | |
0b246afa | 494 | spin_unlock(&fs_info->fs_roots_radix_lock); |
2b9dbef2 JB |
495 | } |
496 | ||
7585717f CM |
497 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
498 | struct btrfs_root *root) | |
499 | { | |
0b246afa | 500 | struct btrfs_fs_info *fs_info = root->fs_info; |
1409e6cc | 501 | int ret; |
0b246afa | 502 | |
92a7cc42 | 503 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
7585717f CM |
504 | return 0; |
505 | ||
506 | /* | |
27cdeb70 | 507 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
508 | * and barriers |
509 | */ | |
510 | smp_rmb(); | |
511 | if (root->last_trans == trans->transid && | |
27cdeb70 | 512 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
513 | return 0; |
514 | ||
0b246afa | 515 | mutex_lock(&fs_info->reloc_mutex); |
1409e6cc | 516 | ret = record_root_in_trans(trans, root, 0); |
0b246afa | 517 | mutex_unlock(&fs_info->reloc_mutex); |
7585717f | 518 | |
1409e6cc | 519 | return ret; |
7585717f CM |
520 | } |
521 | ||
4a9d8bde MX |
522 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
523 | { | |
3296bf56 | 524 | return (trans->state >= TRANS_STATE_COMMIT_START && |
501407aa | 525 | trans->state < TRANS_STATE_UNBLOCKED && |
bf31f87f | 526 | !TRANS_ABORTED(trans)); |
4a9d8bde MX |
527 | } |
528 | ||
d352ac68 CM |
529 | /* wait for commit against the current transaction to become unblocked |
530 | * when this is done, it is safe to start a new transaction, but the current | |
531 | * transaction might not be fully on disk. | |
532 | */ | |
2ff7e61e | 533 | static void wait_current_trans(struct btrfs_fs_info *fs_info) |
79154b1b | 534 | { |
f9295749 | 535 | struct btrfs_transaction *cur_trans; |
79154b1b | 536 | |
0b246afa JM |
537 | spin_lock(&fs_info->trans_lock); |
538 | cur_trans = fs_info->running_transaction; | |
4a9d8bde | 539 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
9b64f57d | 540 | refcount_inc(&cur_trans->use_count); |
0b246afa | 541 | spin_unlock(&fs_info->trans_lock); |
72d63ed6 | 542 | |
0b246afa | 543 | wait_event(fs_info->transaction_wait, |
501407aa | 544 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
bf31f87f | 545 | TRANS_ABORTED(cur_trans)); |
724e2315 | 546 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 547 | } else { |
0b246afa | 548 | spin_unlock(&fs_info->trans_lock); |
f9295749 | 549 | } |
37d1aeee CM |
550 | } |
551 | ||
2ff7e61e | 552 | static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) |
a22285a6 | 553 | { |
0b246afa | 554 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
a4abeea4 JB |
555 | return 0; |
556 | ||
92e2f7e3 | 557 | if (type == TRANS_START) |
a22285a6 | 558 | return 1; |
a4abeea4 | 559 | |
a22285a6 YZ |
560 | return 0; |
561 | } | |
562 | ||
20dd2cbf MX |
563 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
564 | { | |
0b246afa JM |
565 | struct btrfs_fs_info *fs_info = root->fs_info; |
566 | ||
567 | if (!fs_info->reloc_ctl || | |
92a7cc42 | 568 | !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || |
20dd2cbf MX |
569 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
570 | root->reloc_root) | |
571 | return false; | |
572 | ||
573 | return true; | |
574 | } | |
575 | ||
08e007d2 | 576 | static struct btrfs_trans_handle * |
5aed1dd8 | 577 | start_transaction(struct btrfs_root *root, unsigned int num_items, |
003d7c59 JM |
578 | unsigned int type, enum btrfs_reserve_flush_enum flush, |
579 | bool enforce_qgroups) | |
37d1aeee | 580 | { |
0b246afa | 581 | struct btrfs_fs_info *fs_info = root->fs_info; |
ba2c4d4e | 582 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
a22285a6 YZ |
583 | struct btrfs_trans_handle *h; |
584 | struct btrfs_transaction *cur_trans; | |
b5009945 | 585 | u64 num_bytes = 0; |
c5567237 | 586 | u64 qgroup_reserved = 0; |
20dd2cbf | 587 | bool reloc_reserved = false; |
9c343784 | 588 | bool do_chunk_alloc = false; |
20dd2cbf | 589 | int ret; |
acce952b | 590 | |
84961539 | 591 | if (BTRFS_FS_ERROR(fs_info)) |
acce952b | 592 | return ERR_PTR(-EROFS); |
2a1eb461 | 593 | |
46c4e71e | 594 | if (current->journal_info) { |
0860adfd | 595 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 | 596 | h = current->journal_info; |
b50fff81 DS |
597 | refcount_inc(&h->use_count); |
598 | WARN_ON(refcount_read(&h->use_count) > 2); | |
2a1eb461 JB |
599 | h->orig_rsv = h->block_rsv; |
600 | h->block_rsv = NULL; | |
601 | goto got_it; | |
602 | } | |
b5009945 JB |
603 | |
604 | /* | |
605 | * Do the reservation before we join the transaction so we can do all | |
606 | * the appropriate flushing if need be. | |
607 | */ | |
003d7c59 | 608 | if (num_items && root != fs_info->chunk_root) { |
ba2c4d4e JB |
609 | struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; |
610 | u64 delayed_refs_bytes = 0; | |
611 | ||
0b246afa | 612 | qgroup_reserved = num_items * fs_info->nodesize; |
733e03a0 QW |
613 | ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, |
614 | enforce_qgroups); | |
7174109c QW |
615 | if (ret) |
616 | return ERR_PTR(ret); | |
c5567237 | 617 | |
ba2c4d4e JB |
618 | /* |
619 | * We want to reserve all the bytes we may need all at once, so | |
620 | * we only do 1 enospc flushing cycle per transaction start. We | |
621 | * accomplish this by simply assuming we'll do 2 x num_items | |
622 | * worth of delayed refs updates in this trans handle, and | |
623 | * refill that amount for whatever is missing in the reserve. | |
624 | */ | |
2bd36e7b | 625 | num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); |
7f9fe614 JB |
626 | if (flush == BTRFS_RESERVE_FLUSH_ALL && |
627 | delayed_refs_rsv->full == 0) { | |
ba2c4d4e JB |
628 | delayed_refs_bytes = num_bytes; |
629 | num_bytes <<= 1; | |
630 | } | |
631 | ||
20dd2cbf MX |
632 | /* |
633 | * Do the reservation for the relocation root creation | |
634 | */ | |
ee39b432 | 635 | if (need_reserve_reloc_root(root)) { |
0b246afa | 636 | num_bytes += fs_info->nodesize; |
20dd2cbf MX |
637 | reloc_reserved = true; |
638 | } | |
639 | ||
9270501c | 640 | ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes, flush); |
ba2c4d4e JB |
641 | if (ret) |
642 | goto reserve_fail; | |
643 | if (delayed_refs_bytes) { | |
644 | btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, | |
645 | delayed_refs_bytes); | |
646 | num_bytes -= delayed_refs_bytes; | |
647 | } | |
9c343784 JB |
648 | |
649 | if (rsv->space_info->force_alloc) | |
650 | do_chunk_alloc = true; | |
ba2c4d4e JB |
651 | } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && |
652 | !delayed_refs_rsv->full) { | |
653 | /* | |
654 | * Some people call with btrfs_start_transaction(root, 0) | |
655 | * because they can be throttled, but have some other mechanism | |
656 | * for reserving space. We still want these guys to refill the | |
657 | * delayed block_rsv so just add 1 items worth of reservation | |
658 | * here. | |
659 | */ | |
660 | ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); | |
b5009945 | 661 | if (ret) |
843fcf35 | 662 | goto reserve_fail; |
b5009945 | 663 | } |
a22285a6 | 664 | again: |
f2f767e7 | 665 | h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); |
843fcf35 MX |
666 | if (!h) { |
667 | ret = -ENOMEM; | |
668 | goto alloc_fail; | |
669 | } | |
37d1aeee | 670 | |
98114659 JB |
671 | /* |
672 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
673 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
674 | * because we're already holding a ref. We need this because we could | |
675 | * have raced in and did an fsync() on a file which can kick a commit | |
676 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
677 | * |
678 | * If we are ATTACH, it means we just want to catch the current | |
679 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 680 | */ |
0860adfd | 681 | if (type & __TRANS_FREEZABLE) |
0b246afa | 682 | sb_start_intwrite(fs_info->sb); |
b2b5ef5c | 683 | |
2ff7e61e JM |
684 | if (may_wait_transaction(fs_info, type)) |
685 | wait_current_trans(fs_info); | |
a22285a6 | 686 | |
a4abeea4 | 687 | do { |
2ff7e61e | 688 | ret = join_transaction(fs_info, type); |
178260b2 | 689 | if (ret == -EBUSY) { |
2ff7e61e | 690 | wait_current_trans(fs_info); |
a6d155d2 FM |
691 | if (unlikely(type == TRANS_ATTACH || |
692 | type == TRANS_JOIN_NOSTART)) | |
178260b2 MX |
693 | ret = -ENOENT; |
694 | } | |
a4abeea4 JB |
695 | } while (ret == -EBUSY); |
696 | ||
a43f7f82 | 697 | if (ret < 0) |
843fcf35 | 698 | goto join_fail; |
0f7d52f4 | 699 | |
0b246afa | 700 | cur_trans = fs_info->running_transaction; |
a22285a6 YZ |
701 | |
702 | h->transid = cur_trans->transid; | |
703 | h->transaction = cur_trans; | |
b50fff81 | 704 | refcount_set(&h->use_count, 1); |
64b63580 | 705 | h->fs_info = root->fs_info; |
7174109c | 706 | |
a698d075 | 707 | h->type = type; |
ea658bad | 708 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 709 | |
a22285a6 | 710 | smp_mb(); |
3296bf56 | 711 | if (cur_trans->state >= TRANS_STATE_COMMIT_START && |
2ff7e61e | 712 | may_wait_transaction(fs_info, type)) { |
abdd2e80 | 713 | current->journal_info = h; |
3a45bb20 | 714 | btrfs_commit_transaction(h); |
a22285a6 YZ |
715 | goto again; |
716 | } | |
717 | ||
b5009945 | 718 | if (num_bytes) { |
0b246afa | 719 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 720 | h->transid, num_bytes, 1); |
0b246afa | 721 | h->block_rsv = &fs_info->trans_block_rsv; |
b5009945 | 722 | h->bytes_reserved = num_bytes; |
20dd2cbf | 723 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 724 | } |
9ed74f2d | 725 | |
2a1eb461 | 726 | got_it: |
bcf3a3e7 | 727 | if (!current->journal_info) |
a22285a6 | 728 | current->journal_info = h; |
fcc99734 | 729 | |
9c343784 JB |
730 | /* |
731 | * If the space_info is marked ALLOC_FORCE then we'll get upgraded to | |
732 | * ALLOC_FORCE the first run through, and then we won't allocate for | |
733 | * anybody else who races in later. We don't care about the return | |
734 | * value here. | |
735 | */ | |
736 | if (do_chunk_alloc && num_bytes) { | |
737 | u64 flags = h->block_rsv->space_info->flags; | |
738 | ||
739 | btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), | |
740 | CHUNK_ALLOC_NO_FORCE); | |
741 | } | |
742 | ||
fcc99734 QW |
743 | /* |
744 | * btrfs_record_root_in_trans() needs to alloc new extents, and may | |
745 | * call btrfs_join_transaction() while we're also starting a | |
746 | * transaction. | |
747 | * | |
748 | * Thus it need to be called after current->journal_info initialized, | |
749 | * or we can deadlock. | |
750 | */ | |
68075ea8 JB |
751 | ret = btrfs_record_root_in_trans(h, root); |
752 | if (ret) { | |
753 | /* | |
754 | * The transaction handle is fully initialized and linked with | |
755 | * other structures so it needs to be ended in case of errors, | |
756 | * not just freed. | |
757 | */ | |
758 | btrfs_end_transaction(h); | |
759 | return ERR_PTR(ret); | |
760 | } | |
fcc99734 | 761 | |
79154b1b | 762 | return h; |
843fcf35 MX |
763 | |
764 | join_fail: | |
0860adfd | 765 | if (type & __TRANS_FREEZABLE) |
0b246afa | 766 | sb_end_intwrite(fs_info->sb); |
843fcf35 MX |
767 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
768 | alloc_fail: | |
769 | if (num_bytes) | |
2ff7e61e | 770 | btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, |
63f018be | 771 | num_bytes, NULL); |
843fcf35 | 772 | reserve_fail: |
733e03a0 | 773 | btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); |
843fcf35 | 774 | return ERR_PTR(ret); |
79154b1b CM |
775 | } |
776 | ||
f9295749 | 777 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
5aed1dd8 | 778 | unsigned int num_items) |
f9295749 | 779 | { |
08e007d2 | 780 | return start_transaction(root, num_items, TRANS_START, |
003d7c59 | 781 | BTRFS_RESERVE_FLUSH_ALL, true); |
f9295749 | 782 | } |
003d7c59 | 783 | |
8eab77ff FM |
784 | struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( |
785 | struct btrfs_root *root, | |
7f9fe614 | 786 | unsigned int num_items) |
8eab77ff | 787 | { |
7f9fe614 JB |
788 | return start_transaction(root, num_items, TRANS_START, |
789 | BTRFS_RESERVE_FLUSH_ALL_STEAL, false); | |
8eab77ff | 790 | } |
8407aa46 | 791 | |
7a7eaa40 | 792 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 793 | { |
003d7c59 JM |
794 | return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, |
795 | true); | |
f9295749 CM |
796 | } |
797 | ||
8d510121 | 798 | struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) |
0af3d00b | 799 | { |
575a75d6 | 800 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, |
003d7c59 | 801 | BTRFS_RESERVE_NO_FLUSH, true); |
0af3d00b JB |
802 | } |
803 | ||
a6d155d2 FM |
804 | /* |
805 | * Similar to regular join but it never starts a transaction when none is | |
806 | * running or after waiting for the current one to finish. | |
807 | */ | |
808 | struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) | |
809 | { | |
810 | return start_transaction(root, 0, TRANS_JOIN_NOSTART, | |
811 | BTRFS_RESERVE_NO_FLUSH, true); | |
812 | } | |
813 | ||
d4edf39b MX |
814 | /* |
815 | * btrfs_attach_transaction() - catch the running transaction | |
816 | * | |
817 | * It is used when we want to commit the current the transaction, but | |
818 | * don't want to start a new one. | |
819 | * | |
820 | * Note: If this function return -ENOENT, it just means there is no | |
821 | * running transaction. But it is possible that the inactive transaction | |
822 | * is still in the memory, not fully on disk. If you hope there is no | |
823 | * inactive transaction in the fs when -ENOENT is returned, you should | |
824 | * invoke | |
825 | * btrfs_attach_transaction_barrier() | |
826 | */ | |
354aa0fb | 827 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 828 | { |
575a75d6 | 829 | return start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 830 | BTRFS_RESERVE_NO_FLUSH, true); |
60376ce4 JB |
831 | } |
832 | ||
d4edf39b | 833 | /* |
90b6d283 | 834 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b | 835 | * |
52042d8e | 836 | * It is similar to the above function, the difference is this one |
d4edf39b MX |
837 | * will wait for all the inactive transactions until they fully |
838 | * complete. | |
839 | */ | |
840 | struct btrfs_trans_handle * | |
841 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
842 | { | |
843 | struct btrfs_trans_handle *trans; | |
844 | ||
575a75d6 | 845 | trans = start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 846 | BTRFS_RESERVE_NO_FLUSH, true); |
8d9e220c | 847 | if (trans == ERR_PTR(-ENOENT)) |
2ff7e61e | 848 | btrfs_wait_for_commit(root->fs_info, 0); |
d4edf39b MX |
849 | |
850 | return trans; | |
851 | } | |
852 | ||
d0c2f4fa FM |
853 | /* Wait for a transaction commit to reach at least the given state. */ |
854 | static noinline void wait_for_commit(struct btrfs_transaction *commit, | |
855 | const enum btrfs_trans_state min_state) | |
89ce8a63 | 856 | { |
5fd76bf3 OS |
857 | struct btrfs_fs_info *fs_info = commit->fs_info; |
858 | u64 transid = commit->transid; | |
859 | bool put = false; | |
860 | ||
861 | while (1) { | |
862 | wait_event(commit->commit_wait, commit->state >= min_state); | |
863 | if (put) | |
864 | btrfs_put_transaction(commit); | |
865 | ||
866 | if (min_state < TRANS_STATE_COMPLETED) | |
867 | break; | |
868 | ||
869 | /* | |
870 | * A transaction isn't really completed until all of the | |
871 | * previous transactions are completed, but with fsync we can | |
872 | * end up with SUPER_COMMITTED transactions before a COMPLETED | |
873 | * transaction. Wait for those. | |
874 | */ | |
875 | ||
876 | spin_lock(&fs_info->trans_lock); | |
877 | commit = list_first_entry_or_null(&fs_info->trans_list, | |
878 | struct btrfs_transaction, | |
879 | list); | |
880 | if (!commit || commit->transid > transid) { | |
881 | spin_unlock(&fs_info->trans_lock); | |
882 | break; | |
883 | } | |
884 | refcount_inc(&commit->use_count); | |
885 | put = true; | |
886 | spin_unlock(&fs_info->trans_lock); | |
887 | } | |
89ce8a63 CM |
888 | } |
889 | ||
2ff7e61e | 890 | int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) |
46204592 SW |
891 | { |
892 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 893 | int ret = 0; |
46204592 | 894 | |
46204592 | 895 | if (transid) { |
0b246afa | 896 | if (transid <= fs_info->last_trans_committed) |
a4abeea4 | 897 | goto out; |
46204592 SW |
898 | |
899 | /* find specified transaction */ | |
0b246afa JM |
900 | spin_lock(&fs_info->trans_lock); |
901 | list_for_each_entry(t, &fs_info->trans_list, list) { | |
46204592 SW |
902 | if (t->transid == transid) { |
903 | cur_trans = t; | |
9b64f57d | 904 | refcount_inc(&cur_trans->use_count); |
8cd2807f | 905 | ret = 0; |
46204592 SW |
906 | break; |
907 | } | |
8cd2807f MX |
908 | if (t->transid > transid) { |
909 | ret = 0; | |
46204592 | 910 | break; |
8cd2807f | 911 | } |
46204592 | 912 | } |
0b246afa | 913 | spin_unlock(&fs_info->trans_lock); |
42383020 SW |
914 | |
915 | /* | |
916 | * The specified transaction doesn't exist, or we | |
917 | * raced with btrfs_commit_transaction | |
918 | */ | |
919 | if (!cur_trans) { | |
0b246afa | 920 | if (transid > fs_info->last_trans_committed) |
42383020 | 921 | ret = -EINVAL; |
8cd2807f | 922 | goto out; |
42383020 | 923 | } |
46204592 SW |
924 | } else { |
925 | /* find newest transaction that is committing | committed */ | |
0b246afa JM |
926 | spin_lock(&fs_info->trans_lock); |
927 | list_for_each_entry_reverse(t, &fs_info->trans_list, | |
46204592 | 928 | list) { |
4a9d8bde MX |
929 | if (t->state >= TRANS_STATE_COMMIT_START) { |
930 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 931 | break; |
46204592 | 932 | cur_trans = t; |
9b64f57d | 933 | refcount_inc(&cur_trans->use_count); |
46204592 SW |
934 | break; |
935 | } | |
936 | } | |
0b246afa | 937 | spin_unlock(&fs_info->trans_lock); |
46204592 | 938 | if (!cur_trans) |
a4abeea4 | 939 | goto out; /* nothing committing|committed */ |
46204592 SW |
940 | } |
941 | ||
d0c2f4fa | 942 | wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); |
724e2315 | 943 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 944 | out: |
46204592 SW |
945 | return ret; |
946 | } | |
947 | ||
2ff7e61e | 948 | void btrfs_throttle(struct btrfs_fs_info *fs_info) |
37d1aeee | 949 | { |
92e2f7e3 | 950 | wait_current_trans(fs_info); |
37d1aeee CM |
951 | } |
952 | ||
8a8f4dea | 953 | static bool should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa | 954 | { |
2ff7e61e | 955 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0b246afa | 956 | |
64403612 | 957 | if (btrfs_check_space_for_delayed_refs(fs_info)) |
8a8f4dea | 958 | return true; |
36ba022a | 959 | |
2ff7e61e | 960 | return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); |
8929ecfa YZ |
961 | } |
962 | ||
a2633b6a | 963 | bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa YZ |
964 | { |
965 | struct btrfs_transaction *cur_trans = trans->transaction; | |
8929ecfa | 966 | |
3296bf56 | 967 | if (cur_trans->state >= TRANS_STATE_COMMIT_START || |
e19eb11f | 968 | test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) |
a2633b6a | 969 | return true; |
8929ecfa | 970 | |
2ff7e61e | 971 | return should_end_transaction(trans); |
8929ecfa YZ |
972 | } |
973 | ||
dc60c525 NB |
974 | static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) |
975 | ||
0e34693f | 976 | { |
dc60c525 NB |
977 | struct btrfs_fs_info *fs_info = trans->fs_info; |
978 | ||
0e34693f NB |
979 | if (!trans->block_rsv) { |
980 | ASSERT(!trans->bytes_reserved); | |
981 | return; | |
982 | } | |
983 | ||
984 | if (!trans->bytes_reserved) | |
985 | return; | |
986 | ||
987 | ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); | |
988 | trace_btrfs_space_reservation(fs_info, "transaction", | |
989 | trans->transid, trans->bytes_reserved, 0); | |
990 | btrfs_block_rsv_release(fs_info, trans->block_rsv, | |
63f018be | 991 | trans->bytes_reserved, NULL); |
0e34693f NB |
992 | trans->bytes_reserved = 0; |
993 | } | |
994 | ||
89ce8a63 | 995 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
3a45bb20 | 996 | int throttle) |
79154b1b | 997 | { |
3a45bb20 | 998 | struct btrfs_fs_info *info = trans->fs_info; |
8929ecfa | 999 | struct btrfs_transaction *cur_trans = trans->transaction; |
4edc2ca3 | 1000 | int err = 0; |
c3e69d58 | 1001 | |
b50fff81 DS |
1002 | if (refcount_read(&trans->use_count) > 1) { |
1003 | refcount_dec(&trans->use_count); | |
2a1eb461 JB |
1004 | trans->block_rsv = trans->orig_rsv; |
1005 | return 0; | |
1006 | } | |
1007 | ||
dc60c525 | 1008 | btrfs_trans_release_metadata(trans); |
4c13d758 | 1009 | trans->block_rsv = NULL; |
c5567237 | 1010 | |
119e80df | 1011 | btrfs_create_pending_block_groups(trans); |
ea658bad | 1012 | |
4fbcdf66 FM |
1013 | btrfs_trans_release_chunk_metadata(trans); |
1014 | ||
0860adfd | 1015 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 1016 | sb_end_intwrite(info->sb); |
6df7881a | 1017 | |
8929ecfa | 1018 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
1019 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
1020 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 1021 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 1022 | |
093258e6 | 1023 | cond_wake_up(&cur_trans->writer_wait); |
724e2315 | 1024 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
1025 | |
1026 | if (current->journal_info == trans) | |
1027 | current->journal_info = NULL; | |
ab78c84d | 1028 | |
24bbcf04 | 1029 | if (throttle) |
2ff7e61e | 1030 | btrfs_run_delayed_iputs(info); |
24bbcf04 | 1031 | |
84961539 | 1032 | if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) { |
4e121c06 | 1033 | wake_up_process(info->transaction_kthread); |
fbabd4a3 JB |
1034 | if (TRANS_ABORTED(trans)) |
1035 | err = trans->aborted; | |
1036 | else | |
1037 | err = -EROFS; | |
4e121c06 | 1038 | } |
49b25e05 | 1039 | |
4edc2ca3 DJ |
1040 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
1041 | return err; | |
79154b1b CM |
1042 | } |
1043 | ||
3a45bb20 | 1044 | int btrfs_end_transaction(struct btrfs_trans_handle *trans) |
89ce8a63 | 1045 | { |
3a45bb20 | 1046 | return __btrfs_end_transaction(trans, 0); |
89ce8a63 CM |
1047 | } |
1048 | ||
3a45bb20 | 1049 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) |
89ce8a63 | 1050 | { |
3a45bb20 | 1051 | return __btrfs_end_transaction(trans, 1); |
16cdcec7 MX |
1052 | } |
1053 | ||
d352ac68 CM |
1054 | /* |
1055 | * when btree blocks are allocated, they have some corresponding bits set for | |
1056 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 1057 | * those extents are sent to disk but does not wait on them |
d352ac68 | 1058 | */ |
2ff7e61e | 1059 | int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, |
8cef4e16 | 1060 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 1061 | { |
777e6bd7 | 1062 | int err = 0; |
7c4452b9 | 1063 | int werr = 0; |
0b246afa | 1064 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 1065 | struct extent_state *cached_state = NULL; |
777e6bd7 | 1066 | u64 start = 0; |
5f39d397 | 1067 | u64 end; |
7c4452b9 | 1068 | |
6300463b | 1069 | atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
1728366e | 1070 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 1071 | mark, &cached_state)) { |
663dfbb0 FM |
1072 | bool wait_writeback = false; |
1073 | ||
1074 | err = convert_extent_bit(dirty_pages, start, end, | |
1075 | EXTENT_NEED_WAIT, | |
210aa277 | 1076 | mark, &cached_state); |
663dfbb0 FM |
1077 | /* |
1078 | * convert_extent_bit can return -ENOMEM, which is most of the | |
1079 | * time a temporary error. So when it happens, ignore the error | |
1080 | * and wait for writeback of this range to finish - because we | |
1081 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
bf89d38f JM |
1082 | * to __btrfs_wait_marked_extents() would not know that |
1083 | * writeback for this range started and therefore wouldn't | |
1084 | * wait for it to finish - we don't want to commit a | |
1085 | * superblock that points to btree nodes/leafs for which | |
1086 | * writeback hasn't finished yet (and without errors). | |
663dfbb0 | 1087 | * We cleanup any entries left in the io tree when committing |
41e7acd3 | 1088 | * the transaction (through extent_io_tree_release()). |
663dfbb0 FM |
1089 | */ |
1090 | if (err == -ENOMEM) { | |
1091 | err = 0; | |
1092 | wait_writeback = true; | |
1093 | } | |
1094 | if (!err) | |
1095 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
1096 | if (err) |
1097 | werr = err; | |
663dfbb0 FM |
1098 | else if (wait_writeback) |
1099 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 1100 | free_extent_state(cached_state); |
663dfbb0 | 1101 | cached_state = NULL; |
1728366e JB |
1102 | cond_resched(); |
1103 | start = end + 1; | |
7c4452b9 | 1104 | } |
6300463b | 1105 | atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
690587d1 CM |
1106 | return werr; |
1107 | } | |
1108 | ||
1109 | /* | |
1110 | * when btree blocks are allocated, they have some corresponding bits set for | |
1111 | * them in one of two extent_io trees. This is used to make sure all of | |
1112 | * those extents are on disk for transaction or log commit. We wait | |
1113 | * on all the pages and clear them from the dirty pages state tree | |
1114 | */ | |
bf89d38f JM |
1115 | static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, |
1116 | struct extent_io_tree *dirty_pages) | |
690587d1 | 1117 | { |
690587d1 CM |
1118 | int err = 0; |
1119 | int werr = 0; | |
0b246afa | 1120 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 1121 | struct extent_state *cached_state = NULL; |
690587d1 CM |
1122 | u64 start = 0; |
1123 | u64 end; | |
777e6bd7 | 1124 | |
1728366e | 1125 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 1126 | EXTENT_NEED_WAIT, &cached_state)) { |
663dfbb0 FM |
1127 | /* |
1128 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
1129 | * When committing the transaction, we'll remove any entries | |
1130 | * left in the io tree. For a log commit, we don't remove them | |
1131 | * after committing the log because the tree can be accessed | |
1132 | * concurrently - we do it only at transaction commit time when | |
41e7acd3 | 1133 | * it's safe to do it (through extent_io_tree_release()). |
663dfbb0 FM |
1134 | */ |
1135 | err = clear_extent_bit(dirty_pages, start, end, | |
ae0f1625 | 1136 | EXTENT_NEED_WAIT, 0, 0, &cached_state); |
663dfbb0 FM |
1137 | if (err == -ENOMEM) |
1138 | err = 0; | |
1139 | if (!err) | |
1140 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
1141 | if (err) |
1142 | werr = err; | |
e38e2ed7 FM |
1143 | free_extent_state(cached_state); |
1144 | cached_state = NULL; | |
1728366e JB |
1145 | cond_resched(); |
1146 | start = end + 1; | |
777e6bd7 | 1147 | } |
7c4452b9 CM |
1148 | if (err) |
1149 | werr = err; | |
bf89d38f JM |
1150 | return werr; |
1151 | } | |
656f30db | 1152 | |
b9fae2eb | 1153 | static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, |
bf89d38f JM |
1154 | struct extent_io_tree *dirty_pages) |
1155 | { | |
1156 | bool errors = false; | |
1157 | int err; | |
656f30db | 1158 | |
bf89d38f JM |
1159 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); |
1160 | if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) | |
1161 | errors = true; | |
1162 | ||
1163 | if (errors && !err) | |
1164 | err = -EIO; | |
1165 | return err; | |
1166 | } | |
656f30db | 1167 | |
bf89d38f JM |
1168 | int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) |
1169 | { | |
1170 | struct btrfs_fs_info *fs_info = log_root->fs_info; | |
1171 | struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; | |
1172 | bool errors = false; | |
1173 | int err; | |
656f30db | 1174 | |
bf89d38f JM |
1175 | ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
1176 | ||
1177 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); | |
1178 | if ((mark & EXTENT_DIRTY) && | |
1179 | test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) | |
1180 | errors = true; | |
1181 | ||
1182 | if ((mark & EXTENT_NEW) && | |
1183 | test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) | |
1184 | errors = true; | |
1185 | ||
1186 | if (errors && !err) | |
1187 | err = -EIO; | |
1188 | return err; | |
79154b1b CM |
1189 | } |
1190 | ||
690587d1 | 1191 | /* |
c9b577c0 NB |
1192 | * When btree blocks are allocated the corresponding extents are marked dirty. |
1193 | * This function ensures such extents are persisted on disk for transaction or | |
1194 | * log commit. | |
1195 | * | |
1196 | * @trans: transaction whose dirty pages we'd like to write | |
690587d1 | 1197 | */ |
70458a58 | 1198 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) |
690587d1 CM |
1199 | { |
1200 | int ret; | |
1201 | int ret2; | |
c9b577c0 | 1202 | struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; |
70458a58 | 1203 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c6adc9cc | 1204 | struct blk_plug plug; |
690587d1 | 1205 | |
c6adc9cc | 1206 | blk_start_plug(&plug); |
c9b577c0 | 1207 | ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); |
c6adc9cc | 1208 | blk_finish_plug(&plug); |
bf89d38f | 1209 | ret2 = btrfs_wait_extents(fs_info, dirty_pages); |
bf0da8c1 | 1210 | |
41e7acd3 | 1211 | extent_io_tree_release(&trans->transaction->dirty_pages); |
c9b577c0 | 1212 | |
bf0da8c1 CM |
1213 | if (ret) |
1214 | return ret; | |
c9b577c0 | 1215 | else if (ret2) |
bf0da8c1 | 1216 | return ret2; |
c9b577c0 NB |
1217 | else |
1218 | return 0; | |
d0c803c4 CM |
1219 | } |
1220 | ||
d352ac68 CM |
1221 | /* |
1222 | * this is used to update the root pointer in the tree of tree roots. | |
1223 | * | |
1224 | * But, in the case of the extent allocation tree, updating the root | |
1225 | * pointer may allocate blocks which may change the root of the extent | |
1226 | * allocation tree. | |
1227 | * | |
1228 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1229 | * change while the root pointer was being updated in the metadata. | |
1230 | */ | |
0b86a832 CM |
1231 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1232 | struct btrfs_root *root) | |
79154b1b CM |
1233 | { |
1234 | int ret; | |
0b86a832 | 1235 | u64 old_root_bytenr; |
86b9f2ec | 1236 | u64 old_root_used; |
0b246afa JM |
1237 | struct btrfs_fs_info *fs_info = root->fs_info; |
1238 | struct btrfs_root *tree_root = fs_info->tree_root; | |
79154b1b | 1239 | |
86b9f2ec | 1240 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1241 | |
d397712b | 1242 | while (1) { |
0b86a832 | 1243 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1244 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1245 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1246 | break; |
87ef2bb4 | 1247 | |
5d4f98a2 | 1248 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1249 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1250 | &root->root_key, |
1251 | &root->root_item); | |
49b25e05 JM |
1252 | if (ret) |
1253 | return ret; | |
56bec294 | 1254 | |
86b9f2ec | 1255 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1256 | } |
276e680d | 1257 | |
0b86a832 CM |
1258 | return 0; |
1259 | } | |
1260 | ||
d352ac68 CM |
1261 | /* |
1262 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1263 | * |
1264 | * The error handling in this function may not be obvious. Any of the | |
1265 | * failures will cause the file system to go offline. We still need | |
1266 | * to clean up the delayed refs. | |
d352ac68 | 1267 | */ |
9386d8bc | 1268 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) |
0b86a832 | 1269 | { |
9386d8bc | 1270 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ea526d18 | 1271 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1272 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1273 | struct list_head *next; |
84234f3a | 1274 | struct extent_buffer *eb; |
56bec294 | 1275 | int ret; |
84234f3a | 1276 | |
dfba78dc FM |
1277 | /* |
1278 | * At this point no one can be using this transaction to modify any tree | |
1279 | * and no one can start another transaction to modify any tree either. | |
1280 | */ | |
1281 | ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); | |
1282 | ||
84234f3a | 1283 | eb = btrfs_lock_root_node(fs_info->tree_root); |
49b25e05 | 1284 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
9631e4cc | 1285 | 0, &eb, BTRFS_NESTING_COW); |
84234f3a YZ |
1286 | btrfs_tree_unlock(eb); |
1287 | free_extent_buffer(eb); | |
0b86a832 | 1288 | |
49b25e05 JM |
1289 | if (ret) |
1290 | return ret; | |
87ef2bb4 | 1291 | |
196c9d8d | 1292 | ret = btrfs_run_dev_stats(trans); |
c16ce190 JB |
1293 | if (ret) |
1294 | return ret; | |
2b584c68 | 1295 | ret = btrfs_run_dev_replace(trans); |
c16ce190 JB |
1296 | if (ret) |
1297 | return ret; | |
280f8bd2 | 1298 | ret = btrfs_run_qgroups(trans); |
c16ce190 JB |
1299 | if (ret) |
1300 | return ret; | |
546adb0d | 1301 | |
bbebb3e0 | 1302 | ret = btrfs_setup_space_cache(trans); |
dcdf7f6d JB |
1303 | if (ret) |
1304 | return ret; | |
1305 | ||
ea526d18 | 1306 | again: |
d397712b | 1307 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
2ff7e61e | 1308 | struct btrfs_root *root; |
0b86a832 CM |
1309 | next = fs_info->dirty_cowonly_roots.next; |
1310 | list_del_init(next); | |
1311 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1312 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1313 | |
826582ca JB |
1314 | list_add_tail(&root->dirty_list, |
1315 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1316 | ret = update_cowonly_root(trans, root); |
1317 | if (ret) | |
1318 | return ret; | |
79154b1b | 1319 | } |
276e680d | 1320 | |
488bc2a2 JB |
1321 | /* Now flush any delayed refs generated by updating all of the roots */ |
1322 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); | |
1323 | if (ret) | |
1324 | return ret; | |
1325 | ||
1bbc621e | 1326 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
5742d15f | 1327 | ret = btrfs_write_dirty_block_groups(trans); |
ea526d18 JB |
1328 | if (ret) |
1329 | return ret; | |
488bc2a2 JB |
1330 | |
1331 | /* | |
1332 | * We're writing the dirty block groups, which could generate | |
1333 | * delayed refs, which could generate more dirty block groups, | |
1334 | * so we want to keep this flushing in this loop to make sure | |
1335 | * everything gets run. | |
1336 | */ | |
c79a70b1 | 1337 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
ea526d18 JB |
1338 | if (ret) |
1339 | return ret; | |
1340 | } | |
1341 | ||
1342 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1343 | goto again; | |
1344 | ||
9f6cbcbb DS |
1345 | /* Update dev-replace pointer once everything is committed */ |
1346 | fs_info->dev_replace.committed_cursor_left = | |
1347 | fs_info->dev_replace.cursor_left_last_write_of_item; | |
8dabb742 | 1348 | |
79154b1b CM |
1349 | return 0; |
1350 | } | |
1351 | ||
b4be6aef JB |
1352 | /* |
1353 | * If we had a pending drop we need to see if there are any others left in our | |
1354 | * dead roots list, and if not clear our bit and wake any waiters. | |
1355 | */ | |
1356 | void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info) | |
1357 | { | |
1358 | /* | |
1359 | * We put the drop in progress roots at the front of the list, so if the | |
1360 | * first entry doesn't have UNFINISHED_DROP set we can wake everybody | |
1361 | * up. | |
1362 | */ | |
1363 | spin_lock(&fs_info->trans_lock); | |
1364 | if (!list_empty(&fs_info->dead_roots)) { | |
1365 | struct btrfs_root *root = list_first_entry(&fs_info->dead_roots, | |
1366 | struct btrfs_root, | |
1367 | root_list); | |
1368 | if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) { | |
1369 | spin_unlock(&fs_info->trans_lock); | |
1370 | return; | |
1371 | } | |
1372 | } | |
1373 | spin_unlock(&fs_info->trans_lock); | |
1374 | ||
1375 | btrfs_wake_unfinished_drop(fs_info); | |
1376 | } | |
1377 | ||
d352ac68 CM |
1378 | /* |
1379 | * dead roots are old snapshots that need to be deleted. This allocates | |
1380 | * a dirty root struct and adds it into the list of dead roots that need to | |
1381 | * be deleted | |
1382 | */ | |
cfad392b | 1383 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1384 | { |
0b246afa JM |
1385 | struct btrfs_fs_info *fs_info = root->fs_info; |
1386 | ||
1387 | spin_lock(&fs_info->trans_lock); | |
dc9492c1 JB |
1388 | if (list_empty(&root->root_list)) { |
1389 | btrfs_grab_root(root); | |
b4be6aef JB |
1390 | |
1391 | /* We want to process the partially complete drops first. */ | |
1392 | if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) | |
1393 | list_add(&root->root_list, &fs_info->dead_roots); | |
1394 | else | |
1395 | list_add_tail(&root->root_list, &fs_info->dead_roots); | |
dc9492c1 | 1396 | } |
0b246afa | 1397 | spin_unlock(&fs_info->trans_lock); |
5eda7b5e CM |
1398 | } |
1399 | ||
d352ac68 | 1400 | /* |
dfba78dc FM |
1401 | * Update each subvolume root and its relocation root, if it exists, in the tree |
1402 | * of tree roots. Also free log roots if they exist. | |
d352ac68 | 1403 | */ |
7e4443d9 | 1404 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) |
0f7d52f4 | 1405 | { |
7e4443d9 | 1406 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0f7d52f4 | 1407 | struct btrfs_root *gang[8]; |
0f7d52f4 CM |
1408 | int i; |
1409 | int ret; | |
54aa1f4d | 1410 | |
dfba78dc FM |
1411 | /* |
1412 | * At this point no one can be using this transaction to modify any tree | |
1413 | * and no one can start another transaction to modify any tree either. | |
1414 | */ | |
1415 | ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); | |
1416 | ||
a4abeea4 | 1417 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1418 | while (1) { |
5d4f98a2 YZ |
1419 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1420 | (void **)gang, 0, | |
0f7d52f4 CM |
1421 | ARRAY_SIZE(gang), |
1422 | BTRFS_ROOT_TRANS_TAG); | |
1423 | if (ret == 0) | |
1424 | break; | |
1425 | for (i = 0; i < ret; i++) { | |
5b4aacef | 1426 | struct btrfs_root *root = gang[i]; |
4f4317c1 JB |
1427 | int ret2; |
1428 | ||
dfba78dc FM |
1429 | /* |
1430 | * At this point we can neither have tasks logging inodes | |
1431 | * from a root nor trying to commit a log tree. | |
1432 | */ | |
1433 | ASSERT(atomic_read(&root->log_writers) == 0); | |
1434 | ASSERT(atomic_read(&root->log_commit[0]) == 0); | |
1435 | ASSERT(atomic_read(&root->log_commit[1]) == 0); | |
1436 | ||
5d4f98a2 YZ |
1437 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1438 | (unsigned long)root->root_key.objectid, | |
1439 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1440 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1441 | |
e02119d5 | 1442 | btrfs_free_log(trans, root); |
2dd8298e JB |
1443 | ret2 = btrfs_update_reloc_root(trans, root); |
1444 | if (ret2) | |
1445 | return ret2; | |
bcc63abb | 1446 | |
f1ebcc74 | 1447 | /* see comments in should_cow_block() */ |
27cdeb70 | 1448 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
c7548af6 | 1449 | smp_mb__after_atomic(); |
f1ebcc74 | 1450 | |
978d910d | 1451 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1452 | list_add_tail(&root->dirty_list, |
1453 | &trans->transaction->switch_commits); | |
978d910d YZ |
1454 | btrfs_set_root_node(&root->root_item, |
1455 | root->node); | |
1456 | } | |
5d4f98a2 | 1457 | |
4f4317c1 | 1458 | ret2 = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1459 | &root->root_key, |
1460 | &root->root_item); | |
4f4317c1 JB |
1461 | if (ret2) |
1462 | return ret2; | |
a4abeea4 | 1463 | spin_lock(&fs_info->fs_roots_radix_lock); |
733e03a0 | 1464 | btrfs_qgroup_free_meta_all_pertrans(root); |
0f7d52f4 CM |
1465 | } |
1466 | } | |
a4abeea4 | 1467 | spin_unlock(&fs_info->fs_roots_radix_lock); |
4f4317c1 | 1468 | return 0; |
0f7d52f4 CM |
1469 | } |
1470 | ||
d352ac68 | 1471 | /* |
de78b51a ES |
1472 | * defrag a given btree. |
1473 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1474 | */ |
de78b51a | 1475 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1476 | { |
1477 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1478 | struct btrfs_trans_handle *trans; |
8929ecfa | 1479 | int ret; |
e9d0b13b | 1480 | |
27cdeb70 | 1481 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1482 | return 0; |
8929ecfa | 1483 | |
6b80053d | 1484 | while (1) { |
8929ecfa | 1485 | trans = btrfs_start_transaction(root, 0); |
6819703f DS |
1486 | if (IS_ERR(trans)) { |
1487 | ret = PTR_ERR(trans); | |
1488 | break; | |
1489 | } | |
8929ecfa | 1490 | |
de78b51a | 1491 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1492 | |
3a45bb20 | 1493 | btrfs_end_transaction(trans); |
2ff7e61e | 1494 | btrfs_btree_balance_dirty(info); |
e9d0b13b CM |
1495 | cond_resched(); |
1496 | ||
ab8d0fc4 | 1497 | if (btrfs_fs_closing(info) || ret != -EAGAIN) |
e9d0b13b | 1498 | break; |
210549eb | 1499 | |
ab8d0fc4 JM |
1500 | if (btrfs_defrag_cancelled(info)) { |
1501 | btrfs_debug(info, "defrag_root cancelled"); | |
210549eb DS |
1502 | ret = -EAGAIN; |
1503 | break; | |
1504 | } | |
e9d0b13b | 1505 | } |
27cdeb70 | 1506 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1507 | return ret; |
e9d0b13b CM |
1508 | } |
1509 | ||
6426c7ad QW |
1510 | /* |
1511 | * Do all special snapshot related qgroup dirty hack. | |
1512 | * | |
1513 | * Will do all needed qgroup inherit and dirty hack like switch commit | |
1514 | * roots inside one transaction and write all btree into disk, to make | |
1515 | * qgroup works. | |
1516 | */ | |
1517 | static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, | |
1518 | struct btrfs_root *src, | |
1519 | struct btrfs_root *parent, | |
1520 | struct btrfs_qgroup_inherit *inherit, | |
1521 | u64 dst_objectid) | |
1522 | { | |
1523 | struct btrfs_fs_info *fs_info = src->fs_info; | |
1524 | int ret; | |
1525 | ||
1526 | /* | |
1527 | * Save some performance in the case that qgroups are not | |
1528 | * enabled. If this check races with the ioctl, rescan will | |
1529 | * kick in anyway. | |
1530 | */ | |
9ea6e2b5 | 1531 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
6426c7ad | 1532 | return 0; |
6426c7ad | 1533 | |
4d31778a | 1534 | /* |
52042d8e | 1535 | * Ensure dirty @src will be committed. Or, after coming |
4d31778a QW |
1536 | * commit_fs_roots() and switch_commit_roots(), any dirty but not |
1537 | * recorded root will never be updated again, causing an outdated root | |
1538 | * item. | |
1539 | */ | |
1c442d22 JB |
1540 | ret = record_root_in_trans(trans, src, 1); |
1541 | if (ret) | |
1542 | return ret; | |
4d31778a | 1543 | |
2a4d84c1 JB |
1544 | /* |
1545 | * btrfs_qgroup_inherit relies on a consistent view of the usage for the | |
1546 | * src root, so we must run the delayed refs here. | |
1547 | * | |
1548 | * However this isn't particularly fool proof, because there's no | |
1549 | * synchronization keeping us from changing the tree after this point | |
1550 | * before we do the qgroup_inherit, or even from making changes while | |
1551 | * we're doing the qgroup_inherit. But that's a problem for the future, | |
1552 | * for now flush the delayed refs to narrow the race window where the | |
1553 | * qgroup counters could end up wrong. | |
1554 | */ | |
1555 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); | |
1556 | if (ret) { | |
1557 | btrfs_abort_transaction(trans, ret); | |
44365827 | 1558 | return ret; |
2a4d84c1 JB |
1559 | } |
1560 | ||
7e4443d9 | 1561 | ret = commit_fs_roots(trans); |
6426c7ad QW |
1562 | if (ret) |
1563 | goto out; | |
460fb20a | 1564 | ret = btrfs_qgroup_account_extents(trans); |
6426c7ad QW |
1565 | if (ret < 0) |
1566 | goto out; | |
1567 | ||
1568 | /* Now qgroup are all updated, we can inherit it to new qgroups */ | |
a9377422 | 1569 | ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, |
6426c7ad QW |
1570 | inherit); |
1571 | if (ret < 0) | |
1572 | goto out; | |
1573 | ||
1574 | /* | |
1575 | * Now we do a simplified commit transaction, which will: | |
1576 | * 1) commit all subvolume and extent tree | |
1577 | * To ensure all subvolume and extent tree have a valid | |
1578 | * commit_root to accounting later insert_dir_item() | |
1579 | * 2) write all btree blocks onto disk | |
1580 | * This is to make sure later btree modification will be cowed | |
1581 | * Or commit_root can be populated and cause wrong qgroup numbers | |
1582 | * In this simplified commit, we don't really care about other trees | |
1583 | * like chunk and root tree, as they won't affect qgroup. | |
1584 | * And we don't write super to avoid half committed status. | |
1585 | */ | |
9386d8bc | 1586 | ret = commit_cowonly_roots(trans); |
6426c7ad QW |
1587 | if (ret) |
1588 | goto out; | |
889bfa39 | 1589 | switch_commit_roots(trans); |
70458a58 | 1590 | ret = btrfs_write_and_wait_transaction(trans); |
6426c7ad | 1591 | if (ret) |
f7af3934 | 1592 | btrfs_handle_fs_error(fs_info, ret, |
6426c7ad QW |
1593 | "Error while writing out transaction for qgroup"); |
1594 | ||
1595 | out: | |
6426c7ad QW |
1596 | /* |
1597 | * Force parent root to be updated, as we recorded it before so its | |
1598 | * last_trans == cur_transid. | |
1599 | * Or it won't be committed again onto disk after later | |
1600 | * insert_dir_item() | |
1601 | */ | |
1602 | if (!ret) | |
1c442d22 | 1603 | ret = record_root_in_trans(trans, parent, 1); |
6426c7ad QW |
1604 | return ret; |
1605 | } | |
1606 | ||
d352ac68 CM |
1607 | /* |
1608 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1609 | * transaction commit. This does the actual creation. |
1610 | * | |
1611 | * Note: | |
1612 | * If the error which may affect the commitment of the current transaction | |
1613 | * happens, we should return the error number. If the error which just affect | |
1614 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1615 | */ |
80b6794d | 1616 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1617 | struct btrfs_pending_snapshot *pending) |
1618 | { | |
08d50ca3 NB |
1619 | |
1620 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3063d29f | 1621 | struct btrfs_key key; |
80b6794d | 1622 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1623 | struct btrfs_root *tree_root = fs_info->tree_root; |
1624 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1625 | struct btrfs_root *parent_root; |
98c9942a | 1626 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1627 | struct inode *parent_inode; |
42874b3d MX |
1628 | struct btrfs_path *path; |
1629 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1630 | struct dentry *dentry; |
3063d29f | 1631 | struct extent_buffer *tmp; |
925baedd | 1632 | struct extent_buffer *old; |
95582b00 | 1633 | struct timespec64 cur_time; |
aec8030a | 1634 | int ret = 0; |
d68fc57b | 1635 | u64 to_reserve = 0; |
6bdb72de | 1636 | u64 index = 0; |
a22285a6 | 1637 | u64 objectid; |
b83cc969 | 1638 | u64 root_flags; |
3063d29f | 1639 | |
8546b570 DS |
1640 | ASSERT(pending->path); |
1641 | path = pending->path; | |
42874b3d | 1642 | |
b0c0ea63 DS |
1643 | ASSERT(pending->root_item); |
1644 | new_root_item = pending->root_item; | |
a22285a6 | 1645 | |
543068a2 | 1646 | pending->error = btrfs_get_free_objectid(tree_root, &objectid); |
aec8030a | 1647 | if (pending->error) |
6fa9700e | 1648 | goto no_free_objectid; |
3063d29f | 1649 | |
d6726335 QW |
1650 | /* |
1651 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1652 | * accounted by later btrfs_qgroup_inherit(). | |
1653 | */ | |
1654 | btrfs_set_skip_qgroup(trans, objectid); | |
1655 | ||
147d256e | 1656 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
d68fc57b YZ |
1657 | |
1658 | if (to_reserve > 0) { | |
9270501c | 1659 | pending->error = btrfs_block_rsv_add(fs_info, |
aec8030a MX |
1660 | &pending->block_rsv, |
1661 | to_reserve, | |
1662 | BTRFS_RESERVE_NO_FLUSH); | |
1663 | if (pending->error) | |
d6726335 | 1664 | goto clear_skip_qgroup; |
d68fc57b YZ |
1665 | } |
1666 | ||
3063d29f | 1667 | key.objectid = objectid; |
a22285a6 YZ |
1668 | key.offset = (u64)-1; |
1669 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1670 | |
6fa9700e | 1671 | rsv = trans->block_rsv; |
a22285a6 | 1672 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1673 | trans->bytes_reserved = trans->block_rsv->reserved; |
0b246afa | 1674 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa JB |
1675 | trans->transid, |
1676 | trans->bytes_reserved, 1); | |
a22285a6 | 1677 | dentry = pending->dentry; |
e9662f70 | 1678 | parent_inode = pending->dir; |
a22285a6 | 1679 | parent_root = BTRFS_I(parent_inode)->root; |
f0118cb6 JB |
1680 | ret = record_root_in_trans(trans, parent_root, 0); |
1681 | if (ret) | |
1682 | goto fail; | |
c2050a45 | 1683 | cur_time = current_time(parent_inode); |
04b285f3 | 1684 | |
3063d29f CM |
1685 | /* |
1686 | * insert the directory item | |
1687 | */ | |
877574e2 | 1688 | ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); |
49b25e05 | 1689 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1690 | |
1691 | /* check if there is a file/dir which has the same name. */ | |
1692 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
4a0cc7ca | 1693 | btrfs_ino(BTRFS_I(parent_inode)), |
42874b3d MX |
1694 | dentry->d_name.name, |
1695 | dentry->d_name.len, 0); | |
1696 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1697 | pending->error = -EEXIST; |
aec8030a | 1698 | goto dir_item_existed; |
42874b3d MX |
1699 | } else if (IS_ERR(dir_item)) { |
1700 | ret = PTR_ERR(dir_item); | |
66642832 | 1701 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1702 | goto fail; |
79787eaa | 1703 | } |
42874b3d | 1704 | btrfs_release_path(path); |
52c26179 | 1705 | |
e999376f CM |
1706 | /* |
1707 | * pull in the delayed directory update | |
1708 | * and the delayed inode item | |
1709 | * otherwise we corrupt the FS during | |
1710 | * snapshot | |
1711 | */ | |
e5c304e6 | 1712 | ret = btrfs_run_delayed_items(trans); |
8732d44f | 1713 | if (ret) { /* Transaction aborted */ |
66642832 | 1714 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1715 | goto fail; |
1716 | } | |
e999376f | 1717 | |
f0118cb6 JB |
1718 | ret = record_root_in_trans(trans, root, 0); |
1719 | if (ret) { | |
1720 | btrfs_abort_transaction(trans, ret); | |
1721 | goto fail; | |
1722 | } | |
6bdb72de SW |
1723 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1724 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1725 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1726 | |
b83cc969 LZ |
1727 | root_flags = btrfs_root_flags(new_root_item); |
1728 | if (pending->readonly) | |
1729 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1730 | else | |
1731 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1732 | btrfs_set_root_flags(new_root_item, root_flags); | |
1733 | ||
8ea05e3a AB |
1734 | btrfs_set_root_generation_v2(new_root_item, |
1735 | trans->transid); | |
807fc790 | 1736 | generate_random_guid(new_root_item->uuid); |
8ea05e3a AB |
1737 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, |
1738 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1739 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1740 | memset(new_root_item->received_uuid, 0, | |
1741 | sizeof(new_root_item->received_uuid)); | |
1742 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1743 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1744 | btrfs_set_root_stransid(new_root_item, 0); | |
1745 | btrfs_set_root_rtransid(new_root_item, 0); | |
1746 | } | |
3cae210f QW |
1747 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1748 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1749 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1750 | |
6bdb72de | 1751 | old = btrfs_lock_root_node(root); |
9631e4cc JB |
1752 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, |
1753 | BTRFS_NESTING_COW); | |
79787eaa JM |
1754 | if (ret) { |
1755 | btrfs_tree_unlock(old); | |
1756 | free_extent_buffer(old); | |
66642832 | 1757 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1758 | goto fail; |
79787eaa | 1759 | } |
49b25e05 | 1760 | |
49b25e05 | 1761 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1762 | /* clean up in any case */ |
6bdb72de SW |
1763 | btrfs_tree_unlock(old); |
1764 | free_extent_buffer(old); | |
8732d44f | 1765 | if (ret) { |
66642832 | 1766 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1767 | goto fail; |
1768 | } | |
f1ebcc74 | 1769 | /* see comments in should_cow_block() */ |
27cdeb70 | 1770 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1771 | smp_wmb(); |
1772 | ||
6bdb72de | 1773 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1774 | /* record when the snapshot was created in key.offset */ |
1775 | key.offset = trans->transid; | |
1776 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1777 | btrfs_tree_unlock(tmp); |
1778 | free_extent_buffer(tmp); | |
8732d44f | 1779 | if (ret) { |
66642832 | 1780 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1781 | goto fail; |
1782 | } | |
6bdb72de | 1783 | |
a22285a6 YZ |
1784 | /* |
1785 | * insert root back/forward references | |
1786 | */ | |
6025c19f | 1787 | ret = btrfs_add_root_ref(trans, objectid, |
0660b5af | 1788 | parent_root->root_key.objectid, |
4a0cc7ca | 1789 | btrfs_ino(BTRFS_I(parent_inode)), index, |
a22285a6 | 1790 | dentry->d_name.name, dentry->d_name.len); |
8732d44f | 1791 | if (ret) { |
66642832 | 1792 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1793 | goto fail; |
1794 | } | |
0660b5af | 1795 | |
a22285a6 | 1796 | key.offset = (u64)-1; |
2dfb1e43 | 1797 | pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); |
79787eaa JM |
1798 | if (IS_ERR(pending->snap)) { |
1799 | ret = PTR_ERR(pending->snap); | |
2d892ccd | 1800 | pending->snap = NULL; |
66642832 | 1801 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1802 | goto fail; |
79787eaa | 1803 | } |
d68fc57b | 1804 | |
49b25e05 | 1805 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f | 1806 | if (ret) { |
66642832 | 1807 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1808 | goto fail; |
1809 | } | |
361048f5 | 1810 | |
6426c7ad QW |
1811 | /* |
1812 | * Do special qgroup accounting for snapshot, as we do some qgroup | |
1813 | * snapshot hack to do fast snapshot. | |
1814 | * To co-operate with that hack, we do hack again. | |
1815 | * Or snapshot will be greatly slowed down by a subtree qgroup rescan | |
1816 | */ | |
1817 | ret = qgroup_account_snapshot(trans, root, parent_root, | |
1818 | pending->inherit, objectid); | |
1819 | if (ret < 0) | |
1820 | goto fail; | |
1821 | ||
684572df LF |
1822 | ret = btrfs_insert_dir_item(trans, dentry->d_name.name, |
1823 | dentry->d_name.len, BTRFS_I(parent_inode), | |
1824 | &key, BTRFS_FT_DIR, index); | |
42874b3d | 1825 | /* We have check then name at the beginning, so it is impossible. */ |
9c52057c | 1826 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f | 1827 | if (ret) { |
66642832 | 1828 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1829 | goto fail; |
1830 | } | |
42874b3d | 1831 | |
6ef06d27 | 1832 | btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + |
42874b3d | 1833 | dentry->d_name.len * 2); |
04b285f3 | 1834 | parent_inode->i_mtime = parent_inode->i_ctime = |
c2050a45 | 1835 | current_time(parent_inode); |
729f7961 | 1836 | ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); |
dd5f9615 | 1837 | if (ret) { |
66642832 | 1838 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1839 | goto fail; |
1840 | } | |
807fc790 AS |
1841 | ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, |
1842 | BTRFS_UUID_KEY_SUBVOL, | |
cdb345a8 | 1843 | objectid); |
dd5f9615 | 1844 | if (ret) { |
66642832 | 1845 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1846 | goto fail; |
1847 | } | |
1848 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
cdb345a8 | 1849 | ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, |
dd5f9615 SB |
1850 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, |
1851 | objectid); | |
1852 | if (ret && ret != -EEXIST) { | |
66642832 | 1853 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1854 | goto fail; |
1855 | } | |
1856 | } | |
d6726335 | 1857 | |
3063d29f | 1858 | fail: |
aec8030a MX |
1859 | pending->error = ret; |
1860 | dir_item_existed: | |
98c9942a | 1861 | trans->block_rsv = rsv; |
2382c5cc | 1862 | trans->bytes_reserved = 0; |
d6726335 QW |
1863 | clear_skip_qgroup: |
1864 | btrfs_clear_skip_qgroup(trans); | |
6fa9700e MX |
1865 | no_free_objectid: |
1866 | kfree(new_root_item); | |
b0c0ea63 | 1867 | pending->root_item = NULL; |
42874b3d | 1868 | btrfs_free_path(path); |
8546b570 DS |
1869 | pending->path = NULL; |
1870 | ||
49b25e05 | 1871 | return ret; |
3063d29f CM |
1872 | } |
1873 | ||
d352ac68 CM |
1874 | /* |
1875 | * create all the snapshots we've scheduled for creation | |
1876 | */ | |
08d50ca3 | 1877 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) |
3de4586c | 1878 | { |
aec8030a | 1879 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1880 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1881 | int ret = 0; |
3de4586c | 1882 | |
aec8030a MX |
1883 | list_for_each_entry_safe(pending, next, head, list) { |
1884 | list_del(&pending->list); | |
08d50ca3 | 1885 | ret = create_pending_snapshot(trans, pending); |
aec8030a MX |
1886 | if (ret) |
1887 | break; | |
1888 | } | |
1889 | return ret; | |
3de4586c CM |
1890 | } |
1891 | ||
2ff7e61e | 1892 | static void update_super_roots(struct btrfs_fs_info *fs_info) |
5d4f98a2 YZ |
1893 | { |
1894 | struct btrfs_root_item *root_item; | |
1895 | struct btrfs_super_block *super; | |
1896 | ||
0b246afa | 1897 | super = fs_info->super_copy; |
5d4f98a2 | 1898 | |
0b246afa | 1899 | root_item = &fs_info->chunk_root->root_item; |
093e037c DS |
1900 | super->chunk_root = root_item->bytenr; |
1901 | super->chunk_root_generation = root_item->generation; | |
1902 | super->chunk_root_level = root_item->level; | |
5d4f98a2 | 1903 | |
0b246afa | 1904 | root_item = &fs_info->tree_root->root_item; |
093e037c DS |
1905 | super->root = root_item->bytenr; |
1906 | super->generation = root_item->generation; | |
1907 | super->root_level = root_item->level; | |
0b246afa | 1908 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) |
093e037c | 1909 | super->cache_generation = root_item->generation; |
94846229 BB |
1910 | else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) |
1911 | super->cache_generation = 0; | |
0b246afa | 1912 | if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) |
093e037c | 1913 | super->uuid_tree_generation = root_item->generation; |
5d4f98a2 YZ |
1914 | } |
1915 | ||
f36f3042 CM |
1916 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1917 | { | |
4a9d8bde | 1918 | struct btrfs_transaction *trans; |
f36f3042 | 1919 | int ret = 0; |
4a9d8bde | 1920 | |
a4abeea4 | 1921 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1922 | trans = info->running_transaction; |
1923 | if (trans) | |
1924 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1925 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1926 | return ret; |
1927 | } | |
1928 | ||
8929ecfa YZ |
1929 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1930 | { | |
4a9d8bde | 1931 | struct btrfs_transaction *trans; |
8929ecfa | 1932 | int ret = 0; |
4a9d8bde | 1933 | |
a4abeea4 | 1934 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1935 | trans = info->running_transaction; |
1936 | if (trans) | |
1937 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1938 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1939 | return ret; |
1940 | } | |
1941 | ||
fdfbf020 | 1942 | void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans) |
bb9c12c9 | 1943 | { |
3a45bb20 | 1944 | struct btrfs_fs_info *fs_info = trans->fs_info; |
bb9c12c9 SW |
1945 | struct btrfs_transaction *cur_trans; |
1946 | ||
fdfbf020 JB |
1947 | /* Kick the transaction kthread. */ |
1948 | set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); | |
1949 | wake_up_process(fs_info->transaction_kthread); | |
bb9c12c9 SW |
1950 | |
1951 | /* take transaction reference */ | |
bb9c12c9 | 1952 | cur_trans = trans->transaction; |
9b64f57d | 1953 | refcount_inc(&cur_trans->use_count); |
bb9c12c9 | 1954 | |
3a45bb20 | 1955 | btrfs_end_transaction(trans); |
6fc4e354 | 1956 | |
ae5d29d4 DS |
1957 | /* |
1958 | * Wait for the current transaction commit to start and block | |
1959 | * subsequent transaction joins | |
1960 | */ | |
1961 | wait_event(fs_info->transaction_blocked_wait, | |
1962 | cur_trans->state >= TRANS_STATE_COMMIT_START || | |
1963 | TRANS_ABORTED(cur_trans)); | |
724e2315 | 1964 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1965 | } |
1966 | ||
97cb39bb | 1967 | static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) |
49b25e05 | 1968 | { |
97cb39bb | 1969 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 JM |
1970 | struct btrfs_transaction *cur_trans = trans->transaction; |
1971 | ||
b50fff81 | 1972 | WARN_ON(refcount_read(&trans->use_count) > 1); |
49b25e05 | 1973 | |
66642832 | 1974 | btrfs_abort_transaction(trans, err); |
7b8b92af | 1975 | |
0b246afa | 1976 | spin_lock(&fs_info->trans_lock); |
66b6135b | 1977 | |
25d8c284 MX |
1978 | /* |
1979 | * If the transaction is removed from the list, it means this | |
1980 | * transaction has been committed successfully, so it is impossible | |
1981 | * to call the cleanup function. | |
1982 | */ | |
1983 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1984 | |
0b246afa | 1985 | if (cur_trans == fs_info->running_transaction) { |
4a9d8bde | 1986 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 1987 | spin_unlock(&fs_info->trans_lock); |
f094ac32 LB |
1988 | wait_event(cur_trans->writer_wait, |
1989 | atomic_read(&cur_trans->num_writers) == 1); | |
1990 | ||
0b246afa | 1991 | spin_lock(&fs_info->trans_lock); |
d7096fc3 | 1992 | } |
061dde82 FM |
1993 | |
1994 | /* | |
1995 | * Now that we know no one else is still using the transaction we can | |
1996 | * remove the transaction from the list of transactions. This avoids | |
1997 | * the transaction kthread from cleaning up the transaction while some | |
1998 | * other task is still using it, which could result in a use-after-free | |
1999 | * on things like log trees, as it forces the transaction kthread to | |
2000 | * wait for this transaction to be cleaned up by us. | |
2001 | */ | |
2002 | list_del_init(&cur_trans->list); | |
2003 | ||
0b246afa | 2004 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 2005 | |
2ff7e61e | 2006 | btrfs_cleanup_one_transaction(trans->transaction, fs_info); |
49b25e05 | 2007 | |
0b246afa JM |
2008 | spin_lock(&fs_info->trans_lock); |
2009 | if (cur_trans == fs_info->running_transaction) | |
2010 | fs_info->running_transaction = NULL; | |
2011 | spin_unlock(&fs_info->trans_lock); | |
4a9d8bde | 2012 | |
e0228285 | 2013 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2014 | sb_end_intwrite(fs_info->sb); |
724e2315 JB |
2015 | btrfs_put_transaction(cur_trans); |
2016 | btrfs_put_transaction(cur_trans); | |
49b25e05 | 2017 | |
2e4e97ab | 2018 | trace_btrfs_transaction_commit(fs_info); |
49b25e05 | 2019 | |
49b25e05 JM |
2020 | if (current->journal_info == trans) |
2021 | current->journal_info = NULL; | |
0b246afa | 2022 | btrfs_scrub_cancel(fs_info); |
49b25e05 JM |
2023 | |
2024 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
2025 | } | |
2026 | ||
c7cc64a9 DS |
2027 | /* |
2028 | * Release reserved delayed ref space of all pending block groups of the | |
2029 | * transaction and remove them from the list | |
2030 | */ | |
2031 | static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) | |
2032 | { | |
2033 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2034 | struct btrfs_block_group *block_group, *tmp; |
c7cc64a9 DS |
2035 | |
2036 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { | |
2037 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
2038 | list_del_init(&block_group->bg_list); | |
2039 | } | |
2040 | } | |
2041 | ||
88090ad3 | 2042 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
82436617 | 2043 | { |
ce8ea7cc | 2044 | /* |
a0f0cf83 | 2045 | * We use try_to_writeback_inodes_sb() here because if we used |
ce8ea7cc JB |
2046 | * btrfs_start_delalloc_roots we would deadlock with fs freeze. |
2047 | * Currently are holding the fs freeze lock, if we do an async flush | |
2048 | * we'll do btrfs_join_transaction() and deadlock because we need to | |
2049 | * wait for the fs freeze lock. Using the direct flushing we benefit | |
2050 | * from already being in a transaction and our join_transaction doesn't | |
2051 | * have to re-take the fs freeze lock. | |
a0f0cf83 FM |
2052 | * |
2053 | * Note that try_to_writeback_inodes_sb() will only trigger writeback | |
2054 | * if it can read lock sb->s_umount. It will always be able to lock it, | |
2055 | * except when the filesystem is being unmounted or being frozen, but in | |
2056 | * those cases sync_filesystem() is called, which results in calling | |
2057 | * writeback_inodes_sb() while holding a write lock on sb->s_umount. | |
2058 | * Note that we don't call writeback_inodes_sb() directly, because it | |
2059 | * will emit a warning if sb->s_umount is not locked. | |
ce8ea7cc | 2060 | */ |
88090ad3 | 2061 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
a0f0cf83 | 2062 | try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); |
82436617 MX |
2063 | return 0; |
2064 | } | |
2065 | ||
88090ad3 | 2066 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) |
82436617 | 2067 | { |
88090ad3 | 2068 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
6374e57a | 2069 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); |
82436617 MX |
2070 | } |
2071 | ||
28b21c55 FM |
2072 | /* |
2073 | * Add a pending snapshot associated with the given transaction handle to the | |
2074 | * respective handle. This must be called after the transaction commit started | |
2075 | * and while holding fs_info->trans_lock. | |
2076 | * This serves to guarantee a caller of btrfs_commit_transaction() that it can | |
2077 | * safely free the pending snapshot pointer in case btrfs_commit_transaction() | |
2078 | * returns an error. | |
2079 | */ | |
2080 | static void add_pending_snapshot(struct btrfs_trans_handle *trans) | |
2081 | { | |
2082 | struct btrfs_transaction *cur_trans = trans->transaction; | |
2083 | ||
2084 | if (!trans->pending_snapshot) | |
2085 | return; | |
2086 | ||
2087 | lockdep_assert_held(&trans->fs_info->trans_lock); | |
2088 | ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_START); | |
2089 | ||
2090 | list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots); | |
2091 | } | |
2092 | ||
3a45bb20 | 2093 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans) |
79154b1b | 2094 | { |
3a45bb20 | 2095 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 | 2096 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 2097 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 2098 | int ret; |
79154b1b | 2099 | |
35b814f3 NB |
2100 | ASSERT(refcount_read(&trans->use_count) == 1); |
2101 | ||
8d25a086 | 2102 | /* Stop the commit early if ->aborted is set */ |
bf31f87f | 2103 | if (TRANS_ABORTED(cur_trans)) { |
25287e0a | 2104 | ret = cur_trans->aborted; |
3a45bb20 | 2105 | btrfs_end_transaction(trans); |
e4a2bcac | 2106 | return ret; |
25287e0a | 2107 | } |
49b25e05 | 2108 | |
f45c752b JB |
2109 | btrfs_trans_release_metadata(trans); |
2110 | trans->block_rsv = NULL; | |
2111 | ||
56bec294 | 2112 | /* |
e19eb11f JB |
2113 | * We only want one transaction commit doing the flushing so we do not |
2114 | * waste a bunch of time on lock contention on the extent root node. | |
56bec294 | 2115 | */ |
e19eb11f JB |
2116 | if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, |
2117 | &cur_trans->delayed_refs.flags)) { | |
2118 | /* | |
2119 | * Make a pass through all the delayed refs we have so far. | |
2120 | * Any running threads may add more while we are here. | |
2121 | */ | |
2122 | ret = btrfs_run_delayed_refs(trans, 0); | |
2123 | if (ret) { | |
2124 | btrfs_end_transaction(trans); | |
2125 | return ret; | |
2126 | } | |
2127 | } | |
56bec294 | 2128 | |
119e80df | 2129 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2130 | |
3204d33c | 2131 | if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { |
1bbc621e CM |
2132 | int run_it = 0; |
2133 | ||
2134 | /* this mutex is also taken before trying to set | |
2135 | * block groups readonly. We need to make sure | |
2136 | * that nobody has set a block group readonly | |
2137 | * after a extents from that block group have been | |
2138 | * allocated for cache files. btrfs_set_block_group_ro | |
2139 | * will wait for the transaction to commit if it | |
3204d33c | 2140 | * finds BTRFS_TRANS_DIRTY_BG_RUN set. |
1bbc621e | 2141 | * |
3204d33c JB |
2142 | * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure |
2143 | * only one process starts all the block group IO. It wouldn't | |
1bbc621e CM |
2144 | * hurt to have more than one go through, but there's no |
2145 | * real advantage to it either. | |
2146 | */ | |
0b246afa | 2147 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c JB |
2148 | if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, |
2149 | &cur_trans->flags)) | |
1bbc621e | 2150 | run_it = 1; |
0b246afa | 2151 | mutex_unlock(&fs_info->ro_block_group_mutex); |
1bbc621e | 2152 | |
f9cacae3 | 2153 | if (run_it) { |
21217054 | 2154 | ret = btrfs_start_dirty_block_groups(trans); |
f9cacae3 NB |
2155 | if (ret) { |
2156 | btrfs_end_transaction(trans); | |
2157 | return ret; | |
2158 | } | |
2159 | } | |
1bbc621e CM |
2160 | } |
2161 | ||
0b246afa | 2162 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2163 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { |
d0c2f4fa FM |
2164 | enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; |
2165 | ||
28b21c55 FM |
2166 | add_pending_snapshot(trans); |
2167 | ||
0b246afa | 2168 | spin_unlock(&fs_info->trans_lock); |
9b64f57d | 2169 | refcount_inc(&cur_trans->use_count); |
ccd467d6 | 2170 | |
d0c2f4fa FM |
2171 | if (trans->in_fsync) |
2172 | want_state = TRANS_STATE_SUPER_COMMITTED; | |
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); |
bb9c12c9 | 2186 | |
0b246afa | 2187 | if (cur_trans->list.prev != &fs_info->trans_list) { |
d0c2f4fa FM |
2188 | enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; |
2189 | ||
2190 | if (trans->in_fsync) | |
2191 | want_state = TRANS_STATE_SUPER_COMMITTED; | |
2192 | ||
ccd467d6 CM |
2193 | prev_trans = list_entry(cur_trans->list.prev, |
2194 | struct btrfs_transaction, list); | |
d0c2f4fa | 2195 | if (prev_trans->state < want_state) { |
9b64f57d | 2196 | refcount_inc(&prev_trans->use_count); |
0b246afa | 2197 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2198 | |
d0c2f4fa FM |
2199 | wait_for_commit(prev_trans, want_state); |
2200 | ||
bf31f87f | 2201 | ret = READ_ONCE(prev_trans->aborted); |
ccd467d6 | 2202 | |
724e2315 | 2203 | btrfs_put_transaction(prev_trans); |
1f9b8c8f FM |
2204 | if (ret) |
2205 | goto cleanup_transaction; | |
a4abeea4 | 2206 | } else { |
0b246afa | 2207 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2208 | } |
a4abeea4 | 2209 | } else { |
0b246afa | 2210 | spin_unlock(&fs_info->trans_lock); |
cb2d3dad FM |
2211 | /* |
2212 | * The previous transaction was aborted and was already removed | |
2213 | * from the list of transactions at fs_info->trans_list. So we | |
2214 | * abort to prevent writing a new superblock that reflects a | |
2215 | * corrupt state (pointing to trees with unwritten nodes/leafs). | |
2216 | */ | |
84961539 | 2217 | if (BTRFS_FS_ERROR(fs_info)) { |
cb2d3dad FM |
2218 | ret = -EROFS; |
2219 | goto cleanup_transaction; | |
2220 | } | |
ccd467d6 | 2221 | } |
15ee9bc7 | 2222 | |
0860adfd MX |
2223 | extwriter_counter_dec(cur_trans, trans->type); |
2224 | ||
88090ad3 | 2225 | ret = btrfs_start_delalloc_flush(fs_info); |
82436617 MX |
2226 | if (ret) |
2227 | goto cleanup_transaction; | |
2228 | ||
e5c304e6 | 2229 | ret = btrfs_run_delayed_items(trans); |
581227d0 MX |
2230 | if (ret) |
2231 | goto cleanup_transaction; | |
15ee9bc7 | 2232 | |
581227d0 MX |
2233 | wait_event(cur_trans->writer_wait, |
2234 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 2235 | |
581227d0 | 2236 | /* some pending stuffs might be added after the previous flush. */ |
e5c304e6 | 2237 | ret = btrfs_run_delayed_items(trans); |
ca469637 MX |
2238 | if (ret) |
2239 | goto cleanup_transaction; | |
2240 | ||
88090ad3 | 2241 | btrfs_wait_delalloc_flush(fs_info); |
cb7ab021 | 2242 | |
48778179 FM |
2243 | /* |
2244 | * Wait for all ordered extents started by a fast fsync that joined this | |
2245 | * transaction. Otherwise if this transaction commits before the ordered | |
2246 | * extents complete we lose logged data after a power failure. | |
2247 | */ | |
2248 | wait_event(cur_trans->pending_wait, | |
2249 | atomic_read(&cur_trans->pending_ordered) == 0); | |
2250 | ||
2ff7e61e | 2251 | btrfs_scrub_pause(fs_info); |
ed0ca140 JB |
2252 | /* |
2253 | * Ok now we need to make sure to block out any other joins while we | |
2254 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 2255 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 | 2256 | */ |
0b246afa | 2257 | spin_lock(&fs_info->trans_lock); |
28b21c55 | 2258 | add_pending_snapshot(trans); |
4a9d8bde | 2259 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 2260 | spin_unlock(&fs_info->trans_lock); |
ed0ca140 JB |
2261 | wait_event(cur_trans->writer_wait, |
2262 | atomic_read(&cur_trans->num_writers) == 1); | |
2263 | ||
fdfbf020 JB |
2264 | /* |
2265 | * We've started the commit, clear the flag in case we were triggered to | |
2266 | * do an async commit but somebody else started before the transaction | |
2267 | * kthread could do the work. | |
2268 | */ | |
2269 | clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); | |
2270 | ||
bf31f87f | 2271 | if (TRANS_ABORTED(cur_trans)) { |
2cba30f1 | 2272 | ret = cur_trans->aborted; |
6cf7f77e | 2273 | goto scrub_continue; |
2cba30f1 | 2274 | } |
7585717f CM |
2275 | /* |
2276 | * the reloc mutex makes sure that we stop | |
2277 | * the balancing code from coming in and moving | |
2278 | * extents around in the middle of the commit | |
2279 | */ | |
0b246afa | 2280 | mutex_lock(&fs_info->reloc_mutex); |
7585717f | 2281 | |
42874b3d MX |
2282 | /* |
2283 | * We needn't worry about the delayed items because we will | |
2284 | * deal with them in create_pending_snapshot(), which is the | |
2285 | * core function of the snapshot creation. | |
2286 | */ | |
08d50ca3 | 2287 | ret = create_pending_snapshots(trans); |
56e9f6ea DS |
2288 | if (ret) |
2289 | goto unlock_reloc; | |
3063d29f | 2290 | |
42874b3d MX |
2291 | /* |
2292 | * We insert the dir indexes of the snapshots and update the inode | |
2293 | * of the snapshots' parents after the snapshot creation, so there | |
2294 | * are some delayed items which are not dealt with. Now deal with | |
2295 | * them. | |
2296 | * | |
2297 | * We needn't worry that this operation will corrupt the snapshots, | |
2298 | * because all the tree which are snapshoted will be forced to COW | |
2299 | * the nodes and leaves. | |
2300 | */ | |
e5c304e6 | 2301 | ret = btrfs_run_delayed_items(trans); |
56e9f6ea DS |
2302 | if (ret) |
2303 | goto unlock_reloc; | |
16cdcec7 | 2304 | |
c79a70b1 | 2305 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
56e9f6ea DS |
2306 | if (ret) |
2307 | goto unlock_reloc; | |
56bec294 | 2308 | |
e999376f CM |
2309 | /* |
2310 | * make sure none of the code above managed to slip in a | |
2311 | * delayed item | |
2312 | */ | |
ccdf9b30 | 2313 | btrfs_assert_delayed_root_empty(fs_info); |
e999376f | 2314 | |
2c90e5d6 | 2315 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 2316 | |
7e4443d9 | 2317 | ret = commit_fs_roots(trans); |
56e9f6ea | 2318 | if (ret) |
dfba78dc | 2319 | goto unlock_reloc; |
54aa1f4d | 2320 | |
3818aea2 | 2321 | /* |
7e1876ac DS |
2322 | * Since the transaction is done, we can apply the pending changes |
2323 | * before the next transaction. | |
3818aea2 | 2324 | */ |
0b246afa | 2325 | btrfs_apply_pending_changes(fs_info); |
3818aea2 | 2326 | |
5d4f98a2 | 2327 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2328 | * safe to free the root of tree log roots |
2329 | */ | |
0b246afa | 2330 | btrfs_free_log_root_tree(trans, fs_info); |
e02119d5 | 2331 | |
0ed4792a QW |
2332 | /* |
2333 | * Since fs roots are all committed, we can get a quite accurate | |
2334 | * new_roots. So let's do quota accounting. | |
2335 | */ | |
460fb20a | 2336 | ret = btrfs_qgroup_account_extents(trans); |
56e9f6ea | 2337 | if (ret < 0) |
dfba78dc | 2338 | goto unlock_reloc; |
0ed4792a | 2339 | |
9386d8bc | 2340 | ret = commit_cowonly_roots(trans); |
56e9f6ea | 2341 | if (ret) |
dfba78dc | 2342 | goto unlock_reloc; |
54aa1f4d | 2343 | |
2cba30f1 MX |
2344 | /* |
2345 | * The tasks which save the space cache and inode cache may also | |
2346 | * update ->aborted, check it. | |
2347 | */ | |
bf31f87f | 2348 | if (TRANS_ABORTED(cur_trans)) { |
2cba30f1 | 2349 | ret = cur_trans->aborted; |
dfba78dc | 2350 | goto unlock_reloc; |
2cba30f1 MX |
2351 | } |
2352 | ||
0b246afa | 2353 | cur_trans = fs_info->running_transaction; |
5d4f98a2 | 2354 | |
0b246afa JM |
2355 | btrfs_set_root_node(&fs_info->tree_root->root_item, |
2356 | fs_info->tree_root->node); | |
2357 | list_add_tail(&fs_info->tree_root->dirty_list, | |
9e351cc8 | 2358 | &cur_trans->switch_commits); |
5d4f98a2 | 2359 | |
0b246afa JM |
2360 | btrfs_set_root_node(&fs_info->chunk_root->root_item, |
2361 | fs_info->chunk_root->node); | |
2362 | list_add_tail(&fs_info->chunk_root->dirty_list, | |
9e351cc8 JB |
2363 | &cur_trans->switch_commits); |
2364 | ||
889bfa39 | 2365 | switch_commit_roots(trans); |
5d4f98a2 | 2366 | |
ce93ec54 | 2367 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2368 | ASSERT(list_empty(&cur_trans->io_bgs)); |
2ff7e61e | 2369 | update_super_roots(fs_info); |
e02119d5 | 2370 | |
0b246afa JM |
2371 | btrfs_set_super_log_root(fs_info->super_copy, 0); |
2372 | btrfs_set_super_log_root_level(fs_info->super_copy, 0); | |
2373 | memcpy(fs_info->super_for_commit, fs_info->super_copy, | |
2374 | sizeof(*fs_info->super_copy)); | |
ccd467d6 | 2375 | |
bbbf7243 | 2376 | btrfs_commit_device_sizes(cur_trans); |
935e5cc9 | 2377 | |
0b246afa JM |
2378 | clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
2379 | clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); | |
656f30db | 2380 | |
4fbcdf66 FM |
2381 | btrfs_trans_release_chunk_metadata(trans); |
2382 | ||
dfba78dc FM |
2383 | /* |
2384 | * Before changing the transaction state to TRANS_STATE_UNBLOCKED and | |
2385 | * setting fs_info->running_transaction to NULL, lock tree_log_mutex to | |
2386 | * make sure that before we commit our superblock, no other task can | |
2387 | * start a new transaction and commit a log tree before we commit our | |
2388 | * superblock. Anyone trying to commit a log tree locks this mutex before | |
2389 | * writing its superblock. | |
2390 | */ | |
2391 | mutex_lock(&fs_info->tree_log_mutex); | |
2392 | ||
0b246afa | 2393 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2394 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
0b246afa JM |
2395 | fs_info->running_transaction = NULL; |
2396 | spin_unlock(&fs_info->trans_lock); | |
2397 | mutex_unlock(&fs_info->reloc_mutex); | |
b7ec40d7 | 2398 | |
0b246afa | 2399 | wake_up(&fs_info->transaction_wait); |
e6dcd2dc | 2400 | |
70458a58 | 2401 | ret = btrfs_write_and_wait_transaction(trans); |
49b25e05 | 2402 | if (ret) { |
0b246afa JM |
2403 | btrfs_handle_fs_error(fs_info, ret, |
2404 | "Error while writing out transaction"); | |
2405 | mutex_unlock(&fs_info->tree_log_mutex); | |
6cf7f77e | 2406 | goto scrub_continue; |
49b25e05 JM |
2407 | } |
2408 | ||
d3575156 NA |
2409 | /* |
2410 | * At this point, we should have written all the tree blocks allocated | |
2411 | * in this transaction. So it's now safe to free the redirtyied extent | |
2412 | * buffers. | |
2413 | */ | |
2414 | btrfs_free_redirty_list(cur_trans); | |
2415 | ||
eece6a9c | 2416 | ret = write_all_supers(fs_info, 0); |
e02119d5 CM |
2417 | /* |
2418 | * the super is written, we can safely allow the tree-loggers | |
2419 | * to go about their business | |
2420 | */ | |
0b246afa | 2421 | mutex_unlock(&fs_info->tree_log_mutex); |
c1f32b7c AJ |
2422 | if (ret) |
2423 | goto scrub_continue; | |
e02119d5 | 2424 | |
d0c2f4fa FM |
2425 | /* |
2426 | * We needn't acquire the lock here because there is no other task | |
2427 | * which can change it. | |
2428 | */ | |
2429 | cur_trans->state = TRANS_STATE_SUPER_COMMITTED; | |
2430 | wake_up(&cur_trans->commit_wait); | |
2431 | ||
5ead2dd0 | 2432 | btrfs_finish_extent_commit(trans); |
4313b399 | 2433 | |
3204d33c | 2434 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) |
0b246afa | 2435 | btrfs_clear_space_info_full(fs_info); |
13212b54 | 2436 | |
0b246afa | 2437 | fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
2438 | /* |
2439 | * We needn't acquire the lock here because there is no other task | |
2440 | * which can change it. | |
2441 | */ | |
2442 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2443 | wake_up(&cur_trans->commit_wait); |
3de4586c | 2444 | |
0b246afa | 2445 | spin_lock(&fs_info->trans_lock); |
13c5a93e | 2446 | list_del_init(&cur_trans->list); |
0b246afa | 2447 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 2448 | |
724e2315 JB |
2449 | btrfs_put_transaction(cur_trans); |
2450 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2451 | |
0860adfd | 2452 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2453 | sb_end_intwrite(fs_info->sb); |
b2b5ef5c | 2454 | |
2e4e97ab | 2455 | trace_btrfs_transaction_commit(fs_info); |
1abe9b8a | 2456 | |
2ff7e61e | 2457 | btrfs_scrub_continue(fs_info); |
a2de733c | 2458 | |
9ed74f2d JB |
2459 | if (current->journal_info == trans) |
2460 | current->journal_info = NULL; | |
2461 | ||
2c90e5d6 | 2462 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2463 | |
79154b1b | 2464 | return ret; |
49b25e05 | 2465 | |
56e9f6ea DS |
2466 | unlock_reloc: |
2467 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2468 | scrub_continue: |
2ff7e61e | 2469 | btrfs_scrub_continue(fs_info); |
49b25e05 | 2470 | cleanup_transaction: |
dc60c525 | 2471 | btrfs_trans_release_metadata(trans); |
c7cc64a9 | 2472 | btrfs_cleanup_pending_block_groups(trans); |
4fbcdf66 | 2473 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2474 | trans->block_rsv = NULL; |
0b246afa | 2475 | btrfs_warn(fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2476 | if (current->journal_info == trans) |
2477 | current->journal_info = NULL; | |
97cb39bb | 2478 | cleanup_transaction(trans, ret); |
49b25e05 JM |
2479 | |
2480 | return ret; | |
79154b1b CM |
2481 | } |
2482 | ||
d352ac68 | 2483 | /* |
9d1a2a3a DS |
2484 | * return < 0 if error |
2485 | * 0 if there are no more dead_roots at the time of call | |
2486 | * 1 there are more to be processed, call me again | |
2487 | * | |
2488 | * The return value indicates there are certainly more snapshots to delete, but | |
2489 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2490 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2491 | * few seconds later. | |
d352ac68 | 2492 | */ |
9d1a2a3a | 2493 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 2494 | { |
9d1a2a3a | 2495 | int ret; |
5d4f98a2 YZ |
2496 | struct btrfs_fs_info *fs_info = root->fs_info; |
2497 | ||
a4abeea4 | 2498 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2499 | if (list_empty(&fs_info->dead_roots)) { |
2500 | spin_unlock(&fs_info->trans_lock); | |
2501 | return 0; | |
2502 | } | |
2503 | root = list_first_entry(&fs_info->dead_roots, | |
2504 | struct btrfs_root, root_list); | |
cfad392b | 2505 | list_del_init(&root->root_list); |
a4abeea4 | 2506 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2507 | |
4fd786e6 | 2508 | btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); |
76dda93c | 2509 | |
9d1a2a3a | 2510 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2511 | |
9d1a2a3a DS |
2512 | if (btrfs_header_backref_rev(root->node) < |
2513 | BTRFS_MIXED_BACKREF_REV) | |
0078a9f9 | 2514 | ret = btrfs_drop_snapshot(root, 0, 0); |
9d1a2a3a | 2515 | else |
0078a9f9 | 2516 | ret = btrfs_drop_snapshot(root, 1, 0); |
32471dc2 | 2517 | |
dc9492c1 | 2518 | btrfs_put_root(root); |
6596a928 | 2519 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2520 | } |
572d9ab7 DS |
2521 | |
2522 | void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) | |
2523 | { | |
2524 | unsigned long prev; | |
2525 | unsigned long bit; | |
2526 | ||
6c9fe14f | 2527 | prev = xchg(&fs_info->pending_changes, 0); |
572d9ab7 DS |
2528 | if (!prev) |
2529 | return; | |
2530 | ||
d51033d0 DS |
2531 | bit = 1 << BTRFS_PENDING_COMMIT; |
2532 | if (prev & bit) | |
2533 | btrfs_debug(fs_info, "pending commit done"); | |
2534 | prev &= ~bit; | |
2535 | ||
572d9ab7 DS |
2536 | if (prev) |
2537 | btrfs_warn(fs_info, | |
2538 | "unknown pending changes left 0x%lx, ignoring", prev); | |
2539 | } |