Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 CM |
4 | */ |
5 | ||
79154b1b | 6 | #include <linux/fs.h> |
5a0e3ad6 | 7 | #include <linux/slab.h> |
34088780 | 8 | #include <linux/sched.h> |
d3c2fdcf | 9 | #include <linux/writeback.h> |
5f39d397 | 10 | #include <linux/pagemap.h> |
5f2cc086 | 11 | #include <linux/blkdev.h> |
8ea05e3a | 12 | #include <linux/uuid.h> |
602cbe91 | 13 | #include "misc.h" |
79154b1b CM |
14 | #include "ctree.h" |
15 | #include "disk-io.h" | |
16 | #include "transaction.h" | |
925baedd | 17 | #include "locking.h" |
e02119d5 | 18 | #include "tree-log.h" |
733f4fbb | 19 | #include "volumes.h" |
8dabb742 | 20 | #include "dev-replace.h" |
fcebe456 | 21 | #include "qgroup.h" |
aac0023c | 22 | #include "block-group.h" |
9c343784 | 23 | #include "space-info.h" |
79154b1b | 24 | |
0f7d52f4 CM |
25 | #define BTRFS_ROOT_TRANS_TAG 0 |
26 | ||
61c047b5 QW |
27 | /* |
28 | * Transaction states and transitions | |
29 | * | |
30 | * No running transaction (fs tree blocks are not modified) | |
31 | * | | |
32 | * | To next stage: | |
33 | * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). | |
34 | * V | |
35 | * Transaction N [[TRANS_STATE_RUNNING]] | |
36 | * | | |
37 | * | New trans handles can be attached to transaction N by calling all | |
38 | * | start_transaction() variants. | |
39 | * | | |
40 | * | To next stage: | |
41 | * | Call btrfs_commit_transaction() on any trans handle attached to | |
42 | * | transaction N | |
43 | * V | |
44 | * Transaction N [[TRANS_STATE_COMMIT_START]] | |
45 | * | | |
46 | * | Will wait for previous running transaction to completely finish if there | |
47 | * | is one | |
48 | * | | |
49 | * | Then one of the following happes: | |
50 | * | - Wait for all other trans handle holders to release. | |
51 | * | The btrfs_commit_transaction() caller will do the commit work. | |
52 | * | - Wait for current transaction to be committed by others. | |
53 | * | Other btrfs_commit_transaction() caller will do the commit work. | |
54 | * | | |
55 | * | At this stage, only btrfs_join_transaction*() variants can attach | |
56 | * | to this running transaction. | |
57 | * | All other variants will wait for current one to finish and attach to | |
58 | * | transaction N+1. | |
59 | * | | |
60 | * | To next stage: | |
61 | * | Caller is chosen to commit transaction N, and all other trans handle | |
62 | * | haven been released. | |
63 | * V | |
64 | * Transaction N [[TRANS_STATE_COMMIT_DOING]] | |
65 | * | | |
66 | * | The heavy lifting transaction work is started. | |
67 | * | From running delayed refs (modifying extent tree) to creating pending | |
68 | * | snapshots, running qgroups. | |
69 | * | In short, modify supporting trees to reflect modifications of subvolume | |
70 | * | trees. | |
71 | * | | |
72 | * | At this stage, all start_transaction() calls will wait for this | |
73 | * | transaction to finish and attach to transaction N+1. | |
74 | * | | |
75 | * | To next stage: | |
76 | * | Until all supporting trees are updated. | |
77 | * V | |
78 | * Transaction N [[TRANS_STATE_UNBLOCKED]] | |
79 | * | Transaction N+1 | |
80 | * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] | |
81 | * | need to write them back to disk and update | | |
82 | * | super blocks. | | |
83 | * | | | |
84 | * | At this stage, new transaction is allowed to | | |
85 | * | start. | | |
86 | * | All new start_transaction() calls will be | | |
87 | * | attached to transid N+1. | | |
88 | * | | | |
89 | * | To next stage: | | |
90 | * | Until all tree blocks are super blocks are | | |
91 | * | written to block devices | | |
92 | * V | | |
93 | * Transaction N [[TRANS_STATE_COMPLETED]] V | |
94 | * All tree blocks and super blocks are written. Transaction N+1 | |
95 | * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] | |
96 | * data structures will be cleaned up. | Life goes on | |
97 | */ | |
e8c9f186 | 98 | static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { |
4a9d8bde | 99 | [TRANS_STATE_RUNNING] = 0U, |
bcf3a3e7 NB |
100 | [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), |
101 | [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | | |
4a9d8bde | 102 | __TRANS_ATTACH | |
a6d155d2 FM |
103 | __TRANS_JOIN | |
104 | __TRANS_JOIN_NOSTART), | |
bcf3a3e7 | 105 | [TRANS_STATE_UNBLOCKED] = (__TRANS_START | |
4a9d8bde MX |
106 | __TRANS_ATTACH | |
107 | __TRANS_JOIN | | |
a6d155d2 FM |
108 | __TRANS_JOIN_NOLOCK | |
109 | __TRANS_JOIN_NOSTART), | |
d0c2f4fa FM |
110 | [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START | |
111 | __TRANS_ATTACH | | |
112 | __TRANS_JOIN | | |
113 | __TRANS_JOIN_NOLOCK | | |
114 | __TRANS_JOIN_NOSTART), | |
bcf3a3e7 | 115 | [TRANS_STATE_COMPLETED] = (__TRANS_START | |
4a9d8bde MX |
116 | __TRANS_ATTACH | |
117 | __TRANS_JOIN | | |
a6d155d2 FM |
118 | __TRANS_JOIN_NOLOCK | |
119 | __TRANS_JOIN_NOSTART), | |
4a9d8bde MX |
120 | }; |
121 | ||
724e2315 | 122 | void btrfs_put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 123 | { |
9b64f57d ER |
124 | WARN_ON(refcount_read(&transaction->use_count) == 0); |
125 | if (refcount_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 126 | BUG_ON(!list_empty(&transaction->list)); |
5c9d028b LB |
127 | WARN_ON(!RB_EMPTY_ROOT( |
128 | &transaction->delayed_refs.href_root.rb_root)); | |
81f7eb00 JM |
129 | WARN_ON(!RB_EMPTY_ROOT( |
130 | &transaction->delayed_refs.dirty_extent_root)); | |
1262133b | 131 | if (transaction->delayed_refs.pending_csums) |
ab8d0fc4 JM |
132 | btrfs_err(transaction->fs_info, |
133 | "pending csums is %llu", | |
134 | transaction->delayed_refs.pending_csums); | |
7785a663 FM |
135 | /* |
136 | * If any block groups are found in ->deleted_bgs then it's | |
137 | * because the transaction was aborted and a commit did not | |
138 | * happen (things failed before writing the new superblock | |
139 | * and calling btrfs_finish_extent_commit()), so we can not | |
140 | * discard the physical locations of the block groups. | |
141 | */ | |
142 | while (!list_empty(&transaction->deleted_bgs)) { | |
32da5386 | 143 | struct btrfs_block_group *cache; |
7785a663 FM |
144 | |
145 | cache = list_first_entry(&transaction->deleted_bgs, | |
32da5386 | 146 | struct btrfs_block_group, |
7785a663 FM |
147 | bg_list); |
148 | list_del_init(&cache->bg_list); | |
6b7304af | 149 | btrfs_unfreeze_block_group(cache); |
7785a663 FM |
150 | btrfs_put_block_group(cache); |
151 | } | |
bbbf7243 | 152 | WARN_ON(!list_empty(&transaction->dev_update_list)); |
4b5faeac | 153 | kfree(transaction); |
78fae27e | 154 | } |
79154b1b CM |
155 | } |
156 | ||
889bfa39 | 157 | static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) |
817d52f8 | 158 | { |
889bfa39 | 159 | struct btrfs_transaction *cur_trans = trans->transaction; |
16916a88 | 160 | struct btrfs_fs_info *fs_info = trans->fs_info; |
9e351cc8 | 161 | struct btrfs_root *root, *tmp; |
27d56e62 | 162 | struct btrfs_caching_control *caching_ctl, *next; |
9e351cc8 JB |
163 | |
164 | down_write(&fs_info->commit_root_sem); | |
889bfa39 | 165 | list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, |
9e351cc8 JB |
166 | dirty_list) { |
167 | list_del_init(&root->dirty_list); | |
168 | free_extent_buffer(root->commit_root); | |
169 | root->commit_root = btrfs_root_node(root); | |
41e7acd3 | 170 | extent_io_tree_release(&root->dirty_log_pages); |
370a11b8 | 171 | btrfs_qgroup_clean_swapped_blocks(root); |
9e351cc8 | 172 | } |
2b9dbef2 JB |
173 | |
174 | /* We can free old roots now. */ | |
889bfa39 JB |
175 | spin_lock(&cur_trans->dropped_roots_lock); |
176 | while (!list_empty(&cur_trans->dropped_roots)) { | |
177 | root = list_first_entry(&cur_trans->dropped_roots, | |
2b9dbef2 JB |
178 | struct btrfs_root, root_list); |
179 | list_del_init(&root->root_list); | |
889bfa39 JB |
180 | spin_unlock(&cur_trans->dropped_roots_lock); |
181 | btrfs_free_log(trans, root); | |
2b9dbef2 | 182 | btrfs_drop_and_free_fs_root(fs_info, root); |
889bfa39 | 183 | spin_lock(&cur_trans->dropped_roots_lock); |
2b9dbef2 | 184 | } |
889bfa39 | 185 | spin_unlock(&cur_trans->dropped_roots_lock); |
27d56e62 JB |
186 | |
187 | /* | |
188 | * We have to update the last_byte_to_unpin under the commit_root_sem, | |
189 | * at the same time we swap out the commit roots. | |
190 | * | |
191 | * This is because we must have a real view of the last spot the caching | |
192 | * kthreads were while caching. Consider the following views of the | |
193 | * extent tree for a block group | |
194 | * | |
195 | * commit root | |
196 | * +----+----+----+----+----+----+----+ | |
197 | * |\\\\| |\\\\|\\\\| |\\\\|\\\\| | |
198 | * +----+----+----+----+----+----+----+ | |
199 | * 0 1 2 3 4 5 6 7 | |
200 | * | |
201 | * new commit root | |
202 | * +----+----+----+----+----+----+----+ | |
203 | * | | | |\\\\| | |\\\\| | |
204 | * +----+----+----+----+----+----+----+ | |
205 | * 0 1 2 3 4 5 6 7 | |
206 | * | |
207 | * If the cache_ctl->progress was at 3, then we are only allowed to | |
208 | * unpin [0,1) and [2,3], because the caching thread has already | |
209 | * processed those extents. We are not allowed to unpin [5,6), because | |
210 | * the caching thread will re-start it's search from 3, and thus find | |
211 | * the hole from [4,6) to add to the free space cache. | |
212 | */ | |
bbb86a37 | 213 | spin_lock(&fs_info->block_group_cache_lock); |
27d56e62 JB |
214 | list_for_each_entry_safe(caching_ctl, next, |
215 | &fs_info->caching_block_groups, list) { | |
216 | struct btrfs_block_group *cache = caching_ctl->block_group; | |
217 | ||
218 | if (btrfs_block_group_done(cache)) { | |
219 | cache->last_byte_to_unpin = (u64)-1; | |
220 | list_del_init(&caching_ctl->list); | |
221 | btrfs_put_caching_control(caching_ctl); | |
222 | } else { | |
223 | cache->last_byte_to_unpin = caching_ctl->progress; | |
224 | } | |
225 | } | |
bbb86a37 | 226 | spin_unlock(&fs_info->block_group_cache_lock); |
9e351cc8 | 227 | up_write(&fs_info->commit_root_sem); |
817d52f8 JB |
228 | } |
229 | ||
0860adfd MX |
230 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
231 | unsigned int type) | |
232 | { | |
233 | if (type & TRANS_EXTWRITERS) | |
234 | atomic_inc(&trans->num_extwriters); | |
235 | } | |
236 | ||
237 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
238 | unsigned int type) | |
239 | { | |
240 | if (type & TRANS_EXTWRITERS) | |
241 | atomic_dec(&trans->num_extwriters); | |
242 | } | |
243 | ||
244 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
245 | unsigned int type) | |
246 | { | |
247 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
248 | } | |
249 | ||
250 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
251 | { | |
252 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
253 | } |
254 | ||
fb6dea26 JB |
255 | /* |
256 | * To be called after all the new block groups attached to the transaction | |
257 | * handle have been created (btrfs_create_pending_block_groups()). | |
258 | */ | |
259 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
260 | { | |
261 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
262 | ||
263 | if (!trans->chunk_bytes_reserved) | |
264 | return; | |
265 | ||
266 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
267 | ||
268 | btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, | |
63f018be | 269 | trans->chunk_bytes_reserved, NULL); |
fb6dea26 JB |
270 | trans->chunk_bytes_reserved = 0; |
271 | } | |
272 | ||
d352ac68 CM |
273 | /* |
274 | * either allocate a new transaction or hop into the existing one | |
275 | */ | |
2ff7e61e JM |
276 | static noinline int join_transaction(struct btrfs_fs_info *fs_info, |
277 | unsigned int type) | |
79154b1b CM |
278 | { |
279 | struct btrfs_transaction *cur_trans; | |
a4abeea4 | 280 | |
19ae4e81 | 281 | spin_lock(&fs_info->trans_lock); |
d43317dc | 282 | loop: |
49b25e05 | 283 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 284 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 285 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
286 | return -EROFS; |
287 | } | |
288 | ||
19ae4e81 | 289 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 290 | if (cur_trans) { |
bf31f87f | 291 | if (TRANS_ABORTED(cur_trans)) { |
19ae4e81 | 292 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 293 | return cur_trans->aborted; |
871383be | 294 | } |
4a9d8bde | 295 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
296 | spin_unlock(&fs_info->trans_lock); |
297 | return -EBUSY; | |
298 | } | |
9b64f57d | 299 | refcount_inc(&cur_trans->use_count); |
13c5a93e | 300 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 301 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 302 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 303 | return 0; |
79154b1b | 304 | } |
19ae4e81 | 305 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 306 | |
354aa0fb MX |
307 | /* |
308 | * If we are ATTACH, we just want to catch the current transaction, | |
309 | * and commit it. If there is no transaction, just return ENOENT. | |
310 | */ | |
311 | if (type == TRANS_ATTACH) | |
312 | return -ENOENT; | |
313 | ||
4a9d8bde MX |
314 | /* |
315 | * JOIN_NOLOCK only happens during the transaction commit, so | |
316 | * it is impossible that ->running_transaction is NULL | |
317 | */ | |
318 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
319 | ||
4b5faeac | 320 | cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); |
a4abeea4 JB |
321 | if (!cur_trans) |
322 | return -ENOMEM; | |
d43317dc | 323 | |
19ae4e81 JS |
324 | spin_lock(&fs_info->trans_lock); |
325 | if (fs_info->running_transaction) { | |
d43317dc CM |
326 | /* |
327 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 328 | * to redo the checks above |
d43317dc | 329 | */ |
4b5faeac | 330 | kfree(cur_trans); |
d43317dc | 331 | goto loop; |
87533c47 | 332 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 333 | spin_unlock(&fs_info->trans_lock); |
4b5faeac | 334 | kfree(cur_trans); |
7b8b92af | 335 | return -EROFS; |
79154b1b | 336 | } |
d43317dc | 337 | |
ab8d0fc4 | 338 | cur_trans->fs_info = fs_info; |
48778179 FM |
339 | atomic_set(&cur_trans->pending_ordered, 0); |
340 | init_waitqueue_head(&cur_trans->pending_wait); | |
a4abeea4 | 341 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 342 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
343 | init_waitqueue_head(&cur_trans->writer_wait); |
344 | init_waitqueue_head(&cur_trans->commit_wait); | |
4a9d8bde | 345 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
346 | /* |
347 | * One for this trans handle, one so it will live on until we | |
348 | * commit the transaction. | |
349 | */ | |
9b64f57d | 350 | refcount_set(&cur_trans->use_count, 2); |
3204d33c | 351 | cur_trans->flags = 0; |
afd48513 | 352 | cur_trans->start_time = ktime_get_seconds(); |
a4abeea4 | 353 | |
a099d0fd AM |
354 | memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); |
355 | ||
5c9d028b | 356 | cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; |
3368d001 | 357 | cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; |
d7df2c79 | 358 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
20b297d6 JS |
359 | |
360 | /* | |
361 | * although the tree mod log is per file system and not per transaction, | |
362 | * the log must never go across transaction boundaries. | |
363 | */ | |
364 | smp_mb(); | |
31b1a2bd | 365 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
5d163e0e | 366 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); |
31b1a2bd | 367 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
5d163e0e | 368 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); |
fc36ed7e | 369 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 370 | |
a4abeea4 JB |
371 | spin_lock_init(&cur_trans->delayed_refs.lock); |
372 | ||
373 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
bbbf7243 | 374 | INIT_LIST_HEAD(&cur_trans->dev_update_list); |
9e351cc8 | 375 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
ce93ec54 | 376 | INIT_LIST_HEAD(&cur_trans->dirty_bgs); |
1bbc621e | 377 | INIT_LIST_HEAD(&cur_trans->io_bgs); |
2b9dbef2 | 378 | INIT_LIST_HEAD(&cur_trans->dropped_roots); |
1bbc621e | 379 | mutex_init(&cur_trans->cache_write_mutex); |
ce93ec54 | 380 | spin_lock_init(&cur_trans->dirty_bgs_lock); |
e33e17ee | 381 | INIT_LIST_HEAD(&cur_trans->deleted_bgs); |
2b9dbef2 | 382 | spin_lock_init(&cur_trans->dropped_roots_lock); |
19ae4e81 | 383 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
c258d6e3 | 384 | extent_io_tree_init(fs_info, &cur_trans->dirty_pages, |
43eb5f29 | 385 | IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode); |
fe119a6e NB |
386 | extent_io_tree_init(fs_info, &cur_trans->pinned_extents, |
387 | IO_TREE_FS_PINNED_EXTENTS, NULL); | |
19ae4e81 JS |
388 | fs_info->generation++; |
389 | cur_trans->transid = fs_info->generation; | |
390 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 391 | cur_trans->aborted = 0; |
19ae4e81 | 392 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 393 | |
79154b1b CM |
394 | return 0; |
395 | } | |
396 | ||
d352ac68 | 397 | /* |
92a7cc42 QW |
398 | * This does all the record keeping required to make sure that a shareable root |
399 | * is properly recorded in a given transaction. This is required to make sure | |
400 | * the old root from before we joined the transaction is deleted when the | |
401 | * transaction commits. | |
d352ac68 | 402 | */ |
7585717f | 403 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
6426c7ad QW |
404 | struct btrfs_root *root, |
405 | int force) | |
6702ed49 | 406 | { |
0b246afa JM |
407 | struct btrfs_fs_info *fs_info = root->fs_info; |
408 | ||
92a7cc42 | 409 | if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
6426c7ad | 410 | root->last_trans < trans->transid) || force) { |
0b246afa | 411 | WARN_ON(root == fs_info->extent_root); |
4d31778a | 412 | WARN_ON(!force && root->commit_root != root->node); |
5d4f98a2 | 413 | |
7585717f | 414 | /* |
27cdeb70 | 415 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
416 | * we have the reloc mutex held now, so there |
417 | * is only one writer in this function | |
418 | */ | |
27cdeb70 | 419 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 420 | |
27cdeb70 | 421 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
422 | * they find our root->last_trans update |
423 | */ | |
424 | smp_wmb(); | |
425 | ||
0b246afa | 426 | spin_lock(&fs_info->fs_roots_radix_lock); |
6426c7ad | 427 | if (root->last_trans == trans->transid && !force) { |
0b246afa | 428 | spin_unlock(&fs_info->fs_roots_radix_lock); |
a4abeea4 JB |
429 | return 0; |
430 | } | |
0b246afa JM |
431 | radix_tree_tag_set(&fs_info->fs_roots_radix, |
432 | (unsigned long)root->root_key.objectid, | |
433 | BTRFS_ROOT_TRANS_TAG); | |
434 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
7585717f CM |
435 | root->last_trans = trans->transid; |
436 | ||
437 | /* this is pretty tricky. We don't want to | |
438 | * take the relocation lock in btrfs_record_root_in_trans | |
439 | * unless we're really doing the first setup for this root in | |
440 | * this transaction. | |
441 | * | |
442 | * Normally we'd use root->last_trans as a flag to decide | |
443 | * if we want to take the expensive mutex. | |
444 | * | |
445 | * But, we have to set root->last_trans before we | |
446 | * init the relocation root, otherwise, we trip over warnings | |
447 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 448 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
449 | * fixing up the reloc trees and everyone must wait. |
450 | * | |
451 | * When this is zero, they can trust root->last_trans and fly | |
452 | * through btrfs_record_root_in_trans without having to take the | |
453 | * lock. smp_wmb() makes sure that all the writes above are | |
454 | * done before we pop in the zero below | |
455 | */ | |
5d4f98a2 | 456 | btrfs_init_reloc_root(trans, root); |
c7548af6 | 457 | smp_mb__before_atomic(); |
27cdeb70 | 458 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 YZ |
459 | } |
460 | return 0; | |
461 | } | |
bcc63abb | 462 | |
7585717f | 463 | |
2b9dbef2 JB |
464 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
465 | struct btrfs_root *root) | |
466 | { | |
0b246afa | 467 | struct btrfs_fs_info *fs_info = root->fs_info; |
2b9dbef2 JB |
468 | struct btrfs_transaction *cur_trans = trans->transaction; |
469 | ||
470 | /* Add ourselves to the transaction dropped list */ | |
471 | spin_lock(&cur_trans->dropped_roots_lock); | |
472 | list_add_tail(&root->root_list, &cur_trans->dropped_roots); | |
473 | spin_unlock(&cur_trans->dropped_roots_lock); | |
474 | ||
475 | /* Make sure we don't try to update the root at commit time */ | |
0b246afa JM |
476 | spin_lock(&fs_info->fs_roots_radix_lock); |
477 | radix_tree_tag_clear(&fs_info->fs_roots_radix, | |
2b9dbef2 JB |
478 | (unsigned long)root->root_key.objectid, |
479 | BTRFS_ROOT_TRANS_TAG); | |
0b246afa | 480 | spin_unlock(&fs_info->fs_roots_radix_lock); |
2b9dbef2 JB |
481 | } |
482 | ||
7585717f CM |
483 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
484 | struct btrfs_root *root) | |
485 | { | |
0b246afa JM |
486 | struct btrfs_fs_info *fs_info = root->fs_info; |
487 | ||
92a7cc42 | 488 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
7585717f CM |
489 | return 0; |
490 | ||
491 | /* | |
27cdeb70 | 492 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
493 | * and barriers |
494 | */ | |
495 | smp_rmb(); | |
496 | if (root->last_trans == trans->transid && | |
27cdeb70 | 497 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
498 | return 0; |
499 | ||
0b246afa | 500 | mutex_lock(&fs_info->reloc_mutex); |
6426c7ad | 501 | record_root_in_trans(trans, root, 0); |
0b246afa | 502 | mutex_unlock(&fs_info->reloc_mutex); |
7585717f CM |
503 | |
504 | return 0; | |
505 | } | |
506 | ||
4a9d8bde MX |
507 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
508 | { | |
3296bf56 | 509 | return (trans->state >= TRANS_STATE_COMMIT_START && |
501407aa | 510 | trans->state < TRANS_STATE_UNBLOCKED && |
bf31f87f | 511 | !TRANS_ABORTED(trans)); |
4a9d8bde MX |
512 | } |
513 | ||
d352ac68 CM |
514 | /* wait for commit against the current transaction to become unblocked |
515 | * when this is done, it is safe to start a new transaction, but the current | |
516 | * transaction might not be fully on disk. | |
517 | */ | |
2ff7e61e | 518 | static void wait_current_trans(struct btrfs_fs_info *fs_info) |
79154b1b | 519 | { |
f9295749 | 520 | struct btrfs_transaction *cur_trans; |
79154b1b | 521 | |
0b246afa JM |
522 | spin_lock(&fs_info->trans_lock); |
523 | cur_trans = fs_info->running_transaction; | |
4a9d8bde | 524 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
9b64f57d | 525 | refcount_inc(&cur_trans->use_count); |
0b246afa | 526 | spin_unlock(&fs_info->trans_lock); |
72d63ed6 | 527 | |
0b246afa | 528 | wait_event(fs_info->transaction_wait, |
501407aa | 529 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
bf31f87f | 530 | TRANS_ABORTED(cur_trans)); |
724e2315 | 531 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 532 | } else { |
0b246afa | 533 | spin_unlock(&fs_info->trans_lock); |
f9295749 | 534 | } |
37d1aeee CM |
535 | } |
536 | ||
2ff7e61e | 537 | static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) |
a22285a6 | 538 | { |
0b246afa | 539 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
a4abeea4 JB |
540 | return 0; |
541 | ||
92e2f7e3 | 542 | if (type == TRANS_START) |
a22285a6 | 543 | return 1; |
a4abeea4 | 544 | |
a22285a6 YZ |
545 | return 0; |
546 | } | |
547 | ||
20dd2cbf MX |
548 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
549 | { | |
0b246afa JM |
550 | struct btrfs_fs_info *fs_info = root->fs_info; |
551 | ||
552 | if (!fs_info->reloc_ctl || | |
92a7cc42 | 553 | !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || |
20dd2cbf MX |
554 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
555 | root->reloc_root) | |
556 | return false; | |
557 | ||
558 | return true; | |
559 | } | |
560 | ||
08e007d2 | 561 | static struct btrfs_trans_handle * |
5aed1dd8 | 562 | start_transaction(struct btrfs_root *root, unsigned int num_items, |
003d7c59 JM |
563 | unsigned int type, enum btrfs_reserve_flush_enum flush, |
564 | bool enforce_qgroups) | |
37d1aeee | 565 | { |
0b246afa | 566 | struct btrfs_fs_info *fs_info = root->fs_info; |
ba2c4d4e | 567 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
a22285a6 YZ |
568 | struct btrfs_trans_handle *h; |
569 | struct btrfs_transaction *cur_trans; | |
b5009945 | 570 | u64 num_bytes = 0; |
c5567237 | 571 | u64 qgroup_reserved = 0; |
20dd2cbf | 572 | bool reloc_reserved = false; |
9c343784 | 573 | bool do_chunk_alloc = false; |
20dd2cbf | 574 | int ret; |
acce952b | 575 | |
46c4e71e | 576 | /* Send isn't supposed to start transactions. */ |
2755a0de | 577 | ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); |
46c4e71e | 578 | |
0b246afa | 579 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
acce952b | 580 | return ERR_PTR(-EROFS); |
2a1eb461 | 581 | |
46c4e71e | 582 | if (current->journal_info) { |
0860adfd | 583 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 | 584 | h = current->journal_info; |
b50fff81 DS |
585 | refcount_inc(&h->use_count); |
586 | WARN_ON(refcount_read(&h->use_count) > 2); | |
2a1eb461 JB |
587 | h->orig_rsv = h->block_rsv; |
588 | h->block_rsv = NULL; | |
589 | goto got_it; | |
590 | } | |
b5009945 JB |
591 | |
592 | /* | |
593 | * Do the reservation before we join the transaction so we can do all | |
594 | * the appropriate flushing if need be. | |
595 | */ | |
003d7c59 | 596 | if (num_items && root != fs_info->chunk_root) { |
ba2c4d4e JB |
597 | struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; |
598 | u64 delayed_refs_bytes = 0; | |
599 | ||
0b246afa | 600 | qgroup_reserved = num_items * fs_info->nodesize; |
733e03a0 QW |
601 | ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, |
602 | enforce_qgroups); | |
7174109c QW |
603 | if (ret) |
604 | return ERR_PTR(ret); | |
c5567237 | 605 | |
ba2c4d4e JB |
606 | /* |
607 | * We want to reserve all the bytes we may need all at once, so | |
608 | * we only do 1 enospc flushing cycle per transaction start. We | |
609 | * accomplish this by simply assuming we'll do 2 x num_items | |
610 | * worth of delayed refs updates in this trans handle, and | |
611 | * refill that amount for whatever is missing in the reserve. | |
612 | */ | |
2bd36e7b | 613 | num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); |
7f9fe614 JB |
614 | if (flush == BTRFS_RESERVE_FLUSH_ALL && |
615 | delayed_refs_rsv->full == 0) { | |
ba2c4d4e JB |
616 | delayed_refs_bytes = num_bytes; |
617 | num_bytes <<= 1; | |
618 | } | |
619 | ||
20dd2cbf MX |
620 | /* |
621 | * Do the reservation for the relocation root creation | |
622 | */ | |
ee39b432 | 623 | if (need_reserve_reloc_root(root)) { |
0b246afa | 624 | num_bytes += fs_info->nodesize; |
20dd2cbf MX |
625 | reloc_reserved = true; |
626 | } | |
627 | ||
ba2c4d4e JB |
628 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush); |
629 | if (ret) | |
630 | goto reserve_fail; | |
631 | if (delayed_refs_bytes) { | |
632 | btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, | |
633 | delayed_refs_bytes); | |
634 | num_bytes -= delayed_refs_bytes; | |
635 | } | |
9c343784 JB |
636 | |
637 | if (rsv->space_info->force_alloc) | |
638 | do_chunk_alloc = true; | |
ba2c4d4e JB |
639 | } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && |
640 | !delayed_refs_rsv->full) { | |
641 | /* | |
642 | * Some people call with btrfs_start_transaction(root, 0) | |
643 | * because they can be throttled, but have some other mechanism | |
644 | * for reserving space. We still want these guys to refill the | |
645 | * delayed block_rsv so just add 1 items worth of reservation | |
646 | * here. | |
647 | */ | |
648 | ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); | |
b5009945 | 649 | if (ret) |
843fcf35 | 650 | goto reserve_fail; |
b5009945 | 651 | } |
a22285a6 | 652 | again: |
f2f767e7 | 653 | h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); |
843fcf35 MX |
654 | if (!h) { |
655 | ret = -ENOMEM; | |
656 | goto alloc_fail; | |
657 | } | |
37d1aeee | 658 | |
98114659 JB |
659 | /* |
660 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
661 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
662 | * because we're already holding a ref. We need this because we could | |
663 | * have raced in and did an fsync() on a file which can kick a commit | |
664 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
665 | * |
666 | * If we are ATTACH, it means we just want to catch the current | |
667 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 668 | */ |
0860adfd | 669 | if (type & __TRANS_FREEZABLE) |
0b246afa | 670 | sb_start_intwrite(fs_info->sb); |
b2b5ef5c | 671 | |
2ff7e61e JM |
672 | if (may_wait_transaction(fs_info, type)) |
673 | wait_current_trans(fs_info); | |
a22285a6 | 674 | |
a4abeea4 | 675 | do { |
2ff7e61e | 676 | ret = join_transaction(fs_info, type); |
178260b2 | 677 | if (ret == -EBUSY) { |
2ff7e61e | 678 | wait_current_trans(fs_info); |
a6d155d2 FM |
679 | if (unlikely(type == TRANS_ATTACH || |
680 | type == TRANS_JOIN_NOSTART)) | |
178260b2 MX |
681 | ret = -ENOENT; |
682 | } | |
a4abeea4 JB |
683 | } while (ret == -EBUSY); |
684 | ||
a43f7f82 | 685 | if (ret < 0) |
843fcf35 | 686 | goto join_fail; |
0f7d52f4 | 687 | |
0b246afa | 688 | cur_trans = fs_info->running_transaction; |
a22285a6 YZ |
689 | |
690 | h->transid = cur_trans->transid; | |
691 | h->transaction = cur_trans; | |
d13603ef | 692 | h->root = root; |
b50fff81 | 693 | refcount_set(&h->use_count, 1); |
64b63580 | 694 | h->fs_info = root->fs_info; |
7174109c | 695 | |
a698d075 | 696 | h->type = type; |
d9a0540a | 697 | h->can_flush_pending_bgs = true; |
ea658bad | 698 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 699 | |
a22285a6 | 700 | smp_mb(); |
3296bf56 | 701 | if (cur_trans->state >= TRANS_STATE_COMMIT_START && |
2ff7e61e | 702 | may_wait_transaction(fs_info, type)) { |
abdd2e80 | 703 | current->journal_info = h; |
3a45bb20 | 704 | btrfs_commit_transaction(h); |
a22285a6 YZ |
705 | goto again; |
706 | } | |
707 | ||
b5009945 | 708 | if (num_bytes) { |
0b246afa | 709 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 710 | h->transid, num_bytes, 1); |
0b246afa | 711 | h->block_rsv = &fs_info->trans_block_rsv; |
b5009945 | 712 | h->bytes_reserved = num_bytes; |
20dd2cbf | 713 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 714 | } |
9ed74f2d | 715 | |
2a1eb461 | 716 | got_it: |
bcf3a3e7 | 717 | if (!current->journal_info) |
a22285a6 | 718 | current->journal_info = h; |
fcc99734 | 719 | |
9c343784 JB |
720 | /* |
721 | * If the space_info is marked ALLOC_FORCE then we'll get upgraded to | |
722 | * ALLOC_FORCE the first run through, and then we won't allocate for | |
723 | * anybody else who races in later. We don't care about the return | |
724 | * value here. | |
725 | */ | |
726 | if (do_chunk_alloc && num_bytes) { | |
727 | u64 flags = h->block_rsv->space_info->flags; | |
728 | ||
729 | btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), | |
730 | CHUNK_ALLOC_NO_FORCE); | |
731 | } | |
732 | ||
fcc99734 QW |
733 | /* |
734 | * btrfs_record_root_in_trans() needs to alloc new extents, and may | |
735 | * call btrfs_join_transaction() while we're also starting a | |
736 | * transaction. | |
737 | * | |
738 | * Thus it need to be called after current->journal_info initialized, | |
739 | * or we can deadlock. | |
740 | */ | |
741 | btrfs_record_root_in_trans(h, root); | |
742 | ||
79154b1b | 743 | return h; |
843fcf35 MX |
744 | |
745 | join_fail: | |
0860adfd | 746 | if (type & __TRANS_FREEZABLE) |
0b246afa | 747 | sb_end_intwrite(fs_info->sb); |
843fcf35 MX |
748 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
749 | alloc_fail: | |
750 | if (num_bytes) | |
2ff7e61e | 751 | btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, |
63f018be | 752 | num_bytes, NULL); |
843fcf35 | 753 | reserve_fail: |
733e03a0 | 754 | btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); |
843fcf35 | 755 | return ERR_PTR(ret); |
79154b1b CM |
756 | } |
757 | ||
f9295749 | 758 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
5aed1dd8 | 759 | unsigned int num_items) |
f9295749 | 760 | { |
08e007d2 | 761 | return start_transaction(root, num_items, TRANS_START, |
003d7c59 | 762 | BTRFS_RESERVE_FLUSH_ALL, true); |
f9295749 | 763 | } |
003d7c59 | 764 | |
8eab77ff FM |
765 | struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( |
766 | struct btrfs_root *root, | |
7f9fe614 | 767 | unsigned int num_items) |
8eab77ff | 768 | { |
7f9fe614 JB |
769 | return start_transaction(root, num_items, TRANS_START, |
770 | BTRFS_RESERVE_FLUSH_ALL_STEAL, false); | |
8eab77ff | 771 | } |
8407aa46 | 772 | |
7a7eaa40 | 773 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 774 | { |
003d7c59 JM |
775 | return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, |
776 | true); | |
f9295749 CM |
777 | } |
778 | ||
8d510121 | 779 | struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) |
0af3d00b | 780 | { |
575a75d6 | 781 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, |
003d7c59 | 782 | BTRFS_RESERVE_NO_FLUSH, true); |
0af3d00b JB |
783 | } |
784 | ||
a6d155d2 FM |
785 | /* |
786 | * Similar to regular join but it never starts a transaction when none is | |
787 | * running or after waiting for the current one to finish. | |
788 | */ | |
789 | struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) | |
790 | { | |
791 | return start_transaction(root, 0, TRANS_JOIN_NOSTART, | |
792 | BTRFS_RESERVE_NO_FLUSH, true); | |
793 | } | |
794 | ||
d4edf39b MX |
795 | /* |
796 | * btrfs_attach_transaction() - catch the running transaction | |
797 | * | |
798 | * It is used when we want to commit the current the transaction, but | |
799 | * don't want to start a new one. | |
800 | * | |
801 | * Note: If this function return -ENOENT, it just means there is no | |
802 | * running transaction. But it is possible that the inactive transaction | |
803 | * is still in the memory, not fully on disk. If you hope there is no | |
804 | * inactive transaction in the fs when -ENOENT is returned, you should | |
805 | * invoke | |
806 | * btrfs_attach_transaction_barrier() | |
807 | */ | |
354aa0fb | 808 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 809 | { |
575a75d6 | 810 | return start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 811 | BTRFS_RESERVE_NO_FLUSH, true); |
60376ce4 JB |
812 | } |
813 | ||
d4edf39b | 814 | /* |
90b6d283 | 815 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b | 816 | * |
52042d8e | 817 | * It is similar to the above function, the difference is this one |
d4edf39b MX |
818 | * will wait for all the inactive transactions until they fully |
819 | * complete. | |
820 | */ | |
821 | struct btrfs_trans_handle * | |
822 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
823 | { | |
824 | struct btrfs_trans_handle *trans; | |
825 | ||
575a75d6 | 826 | trans = start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 827 | BTRFS_RESERVE_NO_FLUSH, true); |
8d9e220c | 828 | if (trans == ERR_PTR(-ENOENT)) |
2ff7e61e | 829 | btrfs_wait_for_commit(root->fs_info, 0); |
d4edf39b MX |
830 | |
831 | return trans; | |
832 | } | |
833 | ||
d0c2f4fa FM |
834 | /* Wait for a transaction commit to reach at least the given state. */ |
835 | static noinline void wait_for_commit(struct btrfs_transaction *commit, | |
836 | const enum btrfs_trans_state min_state) | |
89ce8a63 | 837 | { |
d0c2f4fa | 838 | wait_event(commit->commit_wait, commit->state >= min_state); |
89ce8a63 CM |
839 | } |
840 | ||
2ff7e61e | 841 | int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) |
46204592 SW |
842 | { |
843 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 844 | int ret = 0; |
46204592 | 845 | |
46204592 | 846 | if (transid) { |
0b246afa | 847 | if (transid <= fs_info->last_trans_committed) |
a4abeea4 | 848 | goto out; |
46204592 SW |
849 | |
850 | /* find specified transaction */ | |
0b246afa JM |
851 | spin_lock(&fs_info->trans_lock); |
852 | list_for_each_entry(t, &fs_info->trans_list, list) { | |
46204592 SW |
853 | if (t->transid == transid) { |
854 | cur_trans = t; | |
9b64f57d | 855 | refcount_inc(&cur_trans->use_count); |
8cd2807f | 856 | ret = 0; |
46204592 SW |
857 | break; |
858 | } | |
8cd2807f MX |
859 | if (t->transid > transid) { |
860 | ret = 0; | |
46204592 | 861 | break; |
8cd2807f | 862 | } |
46204592 | 863 | } |
0b246afa | 864 | spin_unlock(&fs_info->trans_lock); |
42383020 SW |
865 | |
866 | /* | |
867 | * The specified transaction doesn't exist, or we | |
868 | * raced with btrfs_commit_transaction | |
869 | */ | |
870 | if (!cur_trans) { | |
0b246afa | 871 | if (transid > fs_info->last_trans_committed) |
42383020 | 872 | ret = -EINVAL; |
8cd2807f | 873 | goto out; |
42383020 | 874 | } |
46204592 SW |
875 | } else { |
876 | /* find newest transaction that is committing | committed */ | |
0b246afa JM |
877 | spin_lock(&fs_info->trans_lock); |
878 | list_for_each_entry_reverse(t, &fs_info->trans_list, | |
46204592 | 879 | list) { |
4a9d8bde MX |
880 | if (t->state >= TRANS_STATE_COMMIT_START) { |
881 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 882 | break; |
46204592 | 883 | cur_trans = t; |
9b64f57d | 884 | refcount_inc(&cur_trans->use_count); |
46204592 SW |
885 | break; |
886 | } | |
887 | } | |
0b246afa | 888 | spin_unlock(&fs_info->trans_lock); |
46204592 | 889 | if (!cur_trans) |
a4abeea4 | 890 | goto out; /* nothing committing|committed */ |
46204592 SW |
891 | } |
892 | ||
d0c2f4fa | 893 | wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); |
724e2315 | 894 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 895 | out: |
46204592 SW |
896 | return ret; |
897 | } | |
898 | ||
2ff7e61e | 899 | void btrfs_throttle(struct btrfs_fs_info *fs_info) |
37d1aeee | 900 | { |
92e2f7e3 | 901 | wait_current_trans(fs_info); |
37d1aeee CM |
902 | } |
903 | ||
8a8f4dea | 904 | static bool should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa | 905 | { |
2ff7e61e | 906 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0b246afa | 907 | |
64403612 | 908 | if (btrfs_check_space_for_delayed_refs(fs_info)) |
8a8f4dea | 909 | return true; |
36ba022a | 910 | |
2ff7e61e | 911 | return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); |
8929ecfa YZ |
912 | } |
913 | ||
a2633b6a | 914 | bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa YZ |
915 | { |
916 | struct btrfs_transaction *cur_trans = trans->transaction; | |
8929ecfa | 917 | |
3296bf56 | 918 | if (cur_trans->state >= TRANS_STATE_COMMIT_START || |
e19eb11f | 919 | test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) |
a2633b6a | 920 | return true; |
8929ecfa | 921 | |
2ff7e61e | 922 | return should_end_transaction(trans); |
8929ecfa YZ |
923 | } |
924 | ||
dc60c525 NB |
925 | static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) |
926 | ||
0e34693f | 927 | { |
dc60c525 NB |
928 | struct btrfs_fs_info *fs_info = trans->fs_info; |
929 | ||
0e34693f NB |
930 | if (!trans->block_rsv) { |
931 | ASSERT(!trans->bytes_reserved); | |
932 | return; | |
933 | } | |
934 | ||
935 | if (!trans->bytes_reserved) | |
936 | return; | |
937 | ||
938 | ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); | |
939 | trace_btrfs_space_reservation(fs_info, "transaction", | |
940 | trans->transid, trans->bytes_reserved, 0); | |
941 | btrfs_block_rsv_release(fs_info, trans->block_rsv, | |
63f018be | 942 | trans->bytes_reserved, NULL); |
0e34693f NB |
943 | trans->bytes_reserved = 0; |
944 | } | |
945 | ||
89ce8a63 | 946 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
3a45bb20 | 947 | int throttle) |
79154b1b | 948 | { |
3a45bb20 | 949 | struct btrfs_fs_info *info = trans->fs_info; |
8929ecfa | 950 | struct btrfs_transaction *cur_trans = trans->transaction; |
4edc2ca3 | 951 | int err = 0; |
c3e69d58 | 952 | |
b50fff81 DS |
953 | if (refcount_read(&trans->use_count) > 1) { |
954 | refcount_dec(&trans->use_count); | |
2a1eb461 JB |
955 | trans->block_rsv = trans->orig_rsv; |
956 | return 0; | |
957 | } | |
958 | ||
dc60c525 | 959 | btrfs_trans_release_metadata(trans); |
4c13d758 | 960 | trans->block_rsv = NULL; |
c5567237 | 961 | |
119e80df | 962 | btrfs_create_pending_block_groups(trans); |
ea658bad | 963 | |
4fbcdf66 FM |
964 | btrfs_trans_release_chunk_metadata(trans); |
965 | ||
0860adfd | 966 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 967 | sb_end_intwrite(info->sb); |
6df7881a | 968 | |
8929ecfa | 969 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
970 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
971 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 972 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 973 | |
093258e6 | 974 | cond_wake_up(&cur_trans->writer_wait); |
724e2315 | 975 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
976 | |
977 | if (current->journal_info == trans) | |
978 | current->journal_info = NULL; | |
ab78c84d | 979 | |
24bbcf04 | 980 | if (throttle) |
2ff7e61e | 981 | btrfs_run_delayed_iputs(info); |
24bbcf04 | 982 | |
bf31f87f | 983 | if (TRANS_ABORTED(trans) || |
0b246afa | 984 | test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) { |
4e121c06 | 985 | wake_up_process(info->transaction_kthread); |
fbabd4a3 JB |
986 | if (TRANS_ABORTED(trans)) |
987 | err = trans->aborted; | |
988 | else | |
989 | err = -EROFS; | |
4e121c06 | 990 | } |
49b25e05 | 991 | |
4edc2ca3 DJ |
992 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
993 | return err; | |
79154b1b CM |
994 | } |
995 | ||
3a45bb20 | 996 | int btrfs_end_transaction(struct btrfs_trans_handle *trans) |
89ce8a63 | 997 | { |
3a45bb20 | 998 | return __btrfs_end_transaction(trans, 0); |
89ce8a63 CM |
999 | } |
1000 | ||
3a45bb20 | 1001 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) |
89ce8a63 | 1002 | { |
3a45bb20 | 1003 | return __btrfs_end_transaction(trans, 1); |
16cdcec7 MX |
1004 | } |
1005 | ||
d352ac68 CM |
1006 | /* |
1007 | * when btree blocks are allocated, they have some corresponding bits set for | |
1008 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 1009 | * those extents are sent to disk but does not wait on them |
d352ac68 | 1010 | */ |
2ff7e61e | 1011 | int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, |
8cef4e16 | 1012 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 1013 | { |
777e6bd7 | 1014 | int err = 0; |
7c4452b9 | 1015 | int werr = 0; |
0b246afa | 1016 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 1017 | struct extent_state *cached_state = NULL; |
777e6bd7 | 1018 | u64 start = 0; |
5f39d397 | 1019 | u64 end; |
7c4452b9 | 1020 | |
6300463b | 1021 | atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
1728366e | 1022 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 1023 | mark, &cached_state)) { |
663dfbb0 FM |
1024 | bool wait_writeback = false; |
1025 | ||
1026 | err = convert_extent_bit(dirty_pages, start, end, | |
1027 | EXTENT_NEED_WAIT, | |
210aa277 | 1028 | mark, &cached_state); |
663dfbb0 FM |
1029 | /* |
1030 | * convert_extent_bit can return -ENOMEM, which is most of the | |
1031 | * time a temporary error. So when it happens, ignore the error | |
1032 | * and wait for writeback of this range to finish - because we | |
1033 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
bf89d38f JM |
1034 | * to __btrfs_wait_marked_extents() would not know that |
1035 | * writeback for this range started and therefore wouldn't | |
1036 | * wait for it to finish - we don't want to commit a | |
1037 | * superblock that points to btree nodes/leafs for which | |
1038 | * writeback hasn't finished yet (and without errors). | |
663dfbb0 | 1039 | * We cleanup any entries left in the io tree when committing |
41e7acd3 | 1040 | * the transaction (through extent_io_tree_release()). |
663dfbb0 FM |
1041 | */ |
1042 | if (err == -ENOMEM) { | |
1043 | err = 0; | |
1044 | wait_writeback = true; | |
1045 | } | |
1046 | if (!err) | |
1047 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
1048 | if (err) |
1049 | werr = err; | |
663dfbb0 FM |
1050 | else if (wait_writeback) |
1051 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 1052 | free_extent_state(cached_state); |
663dfbb0 | 1053 | cached_state = NULL; |
1728366e JB |
1054 | cond_resched(); |
1055 | start = end + 1; | |
7c4452b9 | 1056 | } |
6300463b | 1057 | atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
690587d1 CM |
1058 | return werr; |
1059 | } | |
1060 | ||
1061 | /* | |
1062 | * when btree blocks are allocated, they have some corresponding bits set for | |
1063 | * them in one of two extent_io trees. This is used to make sure all of | |
1064 | * those extents are on disk for transaction or log commit. We wait | |
1065 | * on all the pages and clear them from the dirty pages state tree | |
1066 | */ | |
bf89d38f JM |
1067 | static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, |
1068 | struct extent_io_tree *dirty_pages) | |
690587d1 | 1069 | { |
690587d1 CM |
1070 | int err = 0; |
1071 | int werr = 0; | |
0b246afa | 1072 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 1073 | struct extent_state *cached_state = NULL; |
690587d1 CM |
1074 | u64 start = 0; |
1075 | u64 end; | |
777e6bd7 | 1076 | |
1728366e | 1077 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 1078 | EXTENT_NEED_WAIT, &cached_state)) { |
663dfbb0 FM |
1079 | /* |
1080 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
1081 | * When committing the transaction, we'll remove any entries | |
1082 | * left in the io tree. For a log commit, we don't remove them | |
1083 | * after committing the log because the tree can be accessed | |
1084 | * concurrently - we do it only at transaction commit time when | |
41e7acd3 | 1085 | * it's safe to do it (through extent_io_tree_release()). |
663dfbb0 FM |
1086 | */ |
1087 | err = clear_extent_bit(dirty_pages, start, end, | |
ae0f1625 | 1088 | EXTENT_NEED_WAIT, 0, 0, &cached_state); |
663dfbb0 FM |
1089 | if (err == -ENOMEM) |
1090 | err = 0; | |
1091 | if (!err) | |
1092 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
1093 | if (err) |
1094 | werr = err; | |
e38e2ed7 FM |
1095 | free_extent_state(cached_state); |
1096 | cached_state = NULL; | |
1728366e JB |
1097 | cond_resched(); |
1098 | start = end + 1; | |
777e6bd7 | 1099 | } |
7c4452b9 CM |
1100 | if (err) |
1101 | werr = err; | |
bf89d38f JM |
1102 | return werr; |
1103 | } | |
656f30db | 1104 | |
b9fae2eb | 1105 | static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, |
bf89d38f JM |
1106 | struct extent_io_tree *dirty_pages) |
1107 | { | |
1108 | bool errors = false; | |
1109 | int err; | |
656f30db | 1110 | |
bf89d38f JM |
1111 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); |
1112 | if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) | |
1113 | errors = true; | |
1114 | ||
1115 | if (errors && !err) | |
1116 | err = -EIO; | |
1117 | return err; | |
1118 | } | |
656f30db | 1119 | |
bf89d38f JM |
1120 | int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) |
1121 | { | |
1122 | struct btrfs_fs_info *fs_info = log_root->fs_info; | |
1123 | struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; | |
1124 | bool errors = false; | |
1125 | int err; | |
656f30db | 1126 | |
bf89d38f JM |
1127 | ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
1128 | ||
1129 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); | |
1130 | if ((mark & EXTENT_DIRTY) && | |
1131 | test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) | |
1132 | errors = true; | |
1133 | ||
1134 | if ((mark & EXTENT_NEW) && | |
1135 | test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) | |
1136 | errors = true; | |
1137 | ||
1138 | if (errors && !err) | |
1139 | err = -EIO; | |
1140 | return err; | |
79154b1b CM |
1141 | } |
1142 | ||
690587d1 | 1143 | /* |
c9b577c0 NB |
1144 | * When btree blocks are allocated the corresponding extents are marked dirty. |
1145 | * This function ensures such extents are persisted on disk for transaction or | |
1146 | * log commit. | |
1147 | * | |
1148 | * @trans: transaction whose dirty pages we'd like to write | |
690587d1 | 1149 | */ |
70458a58 | 1150 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) |
690587d1 CM |
1151 | { |
1152 | int ret; | |
1153 | int ret2; | |
c9b577c0 | 1154 | struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; |
70458a58 | 1155 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c6adc9cc | 1156 | struct blk_plug plug; |
690587d1 | 1157 | |
c6adc9cc | 1158 | blk_start_plug(&plug); |
c9b577c0 | 1159 | ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); |
c6adc9cc | 1160 | blk_finish_plug(&plug); |
bf89d38f | 1161 | ret2 = btrfs_wait_extents(fs_info, dirty_pages); |
bf0da8c1 | 1162 | |
41e7acd3 | 1163 | extent_io_tree_release(&trans->transaction->dirty_pages); |
c9b577c0 | 1164 | |
bf0da8c1 CM |
1165 | if (ret) |
1166 | return ret; | |
c9b577c0 | 1167 | else if (ret2) |
bf0da8c1 | 1168 | return ret2; |
c9b577c0 NB |
1169 | else |
1170 | return 0; | |
d0c803c4 CM |
1171 | } |
1172 | ||
d352ac68 CM |
1173 | /* |
1174 | * this is used to update the root pointer in the tree of tree roots. | |
1175 | * | |
1176 | * But, in the case of the extent allocation tree, updating the root | |
1177 | * pointer may allocate blocks which may change the root of the extent | |
1178 | * allocation tree. | |
1179 | * | |
1180 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1181 | * change while the root pointer was being updated in the metadata. | |
1182 | */ | |
0b86a832 CM |
1183 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1184 | struct btrfs_root *root) | |
79154b1b CM |
1185 | { |
1186 | int ret; | |
0b86a832 | 1187 | u64 old_root_bytenr; |
86b9f2ec | 1188 | u64 old_root_used; |
0b246afa JM |
1189 | struct btrfs_fs_info *fs_info = root->fs_info; |
1190 | struct btrfs_root *tree_root = fs_info->tree_root; | |
79154b1b | 1191 | |
86b9f2ec | 1192 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1193 | |
d397712b | 1194 | while (1) { |
0b86a832 | 1195 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1196 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1197 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1198 | break; |
87ef2bb4 | 1199 | |
5d4f98a2 | 1200 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1201 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1202 | &root->root_key, |
1203 | &root->root_item); | |
49b25e05 JM |
1204 | if (ret) |
1205 | return ret; | |
56bec294 | 1206 | |
86b9f2ec | 1207 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1208 | } |
276e680d | 1209 | |
0b86a832 CM |
1210 | return 0; |
1211 | } | |
1212 | ||
d352ac68 CM |
1213 | /* |
1214 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1215 | * |
1216 | * The error handling in this function may not be obvious. Any of the | |
1217 | * failures will cause the file system to go offline. We still need | |
1218 | * to clean up the delayed refs. | |
d352ac68 | 1219 | */ |
9386d8bc | 1220 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) |
0b86a832 | 1221 | { |
9386d8bc | 1222 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ea526d18 | 1223 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1224 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1225 | struct list_head *next; |
84234f3a | 1226 | struct extent_buffer *eb; |
56bec294 | 1227 | int ret; |
84234f3a YZ |
1228 | |
1229 | eb = btrfs_lock_root_node(fs_info->tree_root); | |
49b25e05 | 1230 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
9631e4cc | 1231 | 0, &eb, BTRFS_NESTING_COW); |
84234f3a YZ |
1232 | btrfs_tree_unlock(eb); |
1233 | free_extent_buffer(eb); | |
0b86a832 | 1234 | |
49b25e05 JM |
1235 | if (ret) |
1236 | return ret; | |
87ef2bb4 | 1237 | |
196c9d8d | 1238 | ret = btrfs_run_dev_stats(trans); |
c16ce190 JB |
1239 | if (ret) |
1240 | return ret; | |
2b584c68 | 1241 | ret = btrfs_run_dev_replace(trans); |
c16ce190 JB |
1242 | if (ret) |
1243 | return ret; | |
280f8bd2 | 1244 | ret = btrfs_run_qgroups(trans); |
c16ce190 JB |
1245 | if (ret) |
1246 | return ret; | |
546adb0d | 1247 | |
bbebb3e0 | 1248 | ret = btrfs_setup_space_cache(trans); |
dcdf7f6d JB |
1249 | if (ret) |
1250 | return ret; | |
1251 | ||
ea526d18 | 1252 | again: |
d397712b | 1253 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
2ff7e61e | 1254 | struct btrfs_root *root; |
0b86a832 CM |
1255 | next = fs_info->dirty_cowonly_roots.next; |
1256 | list_del_init(next); | |
1257 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1258 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1259 | |
9e351cc8 JB |
1260 | if (root != fs_info->extent_root) |
1261 | list_add_tail(&root->dirty_list, | |
1262 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1263 | ret = update_cowonly_root(trans, root); |
1264 | if (ret) | |
1265 | return ret; | |
79154b1b | 1266 | } |
276e680d | 1267 | |
488bc2a2 JB |
1268 | /* Now flush any delayed refs generated by updating all of the roots */ |
1269 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); | |
1270 | if (ret) | |
1271 | return ret; | |
1272 | ||
1bbc621e | 1273 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
5742d15f | 1274 | ret = btrfs_write_dirty_block_groups(trans); |
ea526d18 JB |
1275 | if (ret) |
1276 | return ret; | |
488bc2a2 JB |
1277 | |
1278 | /* | |
1279 | * We're writing the dirty block groups, which could generate | |
1280 | * delayed refs, which could generate more dirty block groups, | |
1281 | * so we want to keep this flushing in this loop to make sure | |
1282 | * everything gets run. | |
1283 | */ | |
c79a70b1 | 1284 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
ea526d18 JB |
1285 | if (ret) |
1286 | return ret; | |
1287 | } | |
1288 | ||
1289 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1290 | goto again; | |
1291 | ||
9e351cc8 JB |
1292 | list_add_tail(&fs_info->extent_root->dirty_list, |
1293 | &trans->transaction->switch_commits); | |
9f6cbcbb DS |
1294 | |
1295 | /* Update dev-replace pointer once everything is committed */ | |
1296 | fs_info->dev_replace.committed_cursor_left = | |
1297 | fs_info->dev_replace.cursor_left_last_write_of_item; | |
8dabb742 | 1298 | |
79154b1b CM |
1299 | return 0; |
1300 | } | |
1301 | ||
d352ac68 CM |
1302 | /* |
1303 | * dead roots are old snapshots that need to be deleted. This allocates | |
1304 | * a dirty root struct and adds it into the list of dead roots that need to | |
1305 | * be deleted | |
1306 | */ | |
cfad392b | 1307 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1308 | { |
0b246afa JM |
1309 | struct btrfs_fs_info *fs_info = root->fs_info; |
1310 | ||
1311 | spin_lock(&fs_info->trans_lock); | |
dc9492c1 JB |
1312 | if (list_empty(&root->root_list)) { |
1313 | btrfs_grab_root(root); | |
0b246afa | 1314 | list_add_tail(&root->root_list, &fs_info->dead_roots); |
dc9492c1 | 1315 | } |
0b246afa | 1316 | spin_unlock(&fs_info->trans_lock); |
5eda7b5e CM |
1317 | } |
1318 | ||
d352ac68 | 1319 | /* |
5d4f98a2 | 1320 | * update all the cowonly tree roots on disk |
d352ac68 | 1321 | */ |
7e4443d9 | 1322 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) |
0f7d52f4 | 1323 | { |
7e4443d9 | 1324 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0f7d52f4 | 1325 | struct btrfs_root *gang[8]; |
0f7d52f4 CM |
1326 | int i; |
1327 | int ret; | |
54aa1f4d | 1328 | |
a4abeea4 | 1329 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1330 | while (1) { |
5d4f98a2 YZ |
1331 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1332 | (void **)gang, 0, | |
0f7d52f4 CM |
1333 | ARRAY_SIZE(gang), |
1334 | BTRFS_ROOT_TRANS_TAG); | |
1335 | if (ret == 0) | |
1336 | break; | |
1337 | for (i = 0; i < ret; i++) { | |
5b4aacef | 1338 | struct btrfs_root *root = gang[i]; |
4f4317c1 JB |
1339 | int ret2; |
1340 | ||
5d4f98a2 YZ |
1341 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1342 | (unsigned long)root->root_key.objectid, | |
1343 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1344 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1345 | |
e02119d5 | 1346 | btrfs_free_log(trans, root); |
5d4f98a2 | 1347 | btrfs_update_reloc_root(trans, root); |
bcc63abb | 1348 | |
f1ebcc74 | 1349 | /* see comments in should_cow_block() */ |
27cdeb70 | 1350 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
c7548af6 | 1351 | smp_mb__after_atomic(); |
f1ebcc74 | 1352 | |
978d910d | 1353 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1354 | list_add_tail(&root->dirty_list, |
1355 | &trans->transaction->switch_commits); | |
978d910d YZ |
1356 | btrfs_set_root_node(&root->root_item, |
1357 | root->node); | |
1358 | } | |
5d4f98a2 | 1359 | |
4f4317c1 | 1360 | ret2 = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1361 | &root->root_key, |
1362 | &root->root_item); | |
4f4317c1 JB |
1363 | if (ret2) |
1364 | return ret2; | |
a4abeea4 | 1365 | spin_lock(&fs_info->fs_roots_radix_lock); |
733e03a0 | 1366 | btrfs_qgroup_free_meta_all_pertrans(root); |
0f7d52f4 CM |
1367 | } |
1368 | } | |
a4abeea4 | 1369 | spin_unlock(&fs_info->fs_roots_radix_lock); |
4f4317c1 | 1370 | return 0; |
0f7d52f4 CM |
1371 | } |
1372 | ||
d352ac68 | 1373 | /* |
de78b51a ES |
1374 | * defrag a given btree. |
1375 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1376 | */ |
de78b51a | 1377 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1378 | { |
1379 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1380 | struct btrfs_trans_handle *trans; |
8929ecfa | 1381 | int ret; |
e9d0b13b | 1382 | |
27cdeb70 | 1383 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1384 | return 0; |
8929ecfa | 1385 | |
6b80053d | 1386 | while (1) { |
8929ecfa YZ |
1387 | trans = btrfs_start_transaction(root, 0); |
1388 | if (IS_ERR(trans)) | |
1389 | return PTR_ERR(trans); | |
1390 | ||
de78b51a | 1391 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1392 | |
3a45bb20 | 1393 | btrfs_end_transaction(trans); |
2ff7e61e | 1394 | btrfs_btree_balance_dirty(info); |
e9d0b13b CM |
1395 | cond_resched(); |
1396 | ||
ab8d0fc4 | 1397 | if (btrfs_fs_closing(info) || ret != -EAGAIN) |
e9d0b13b | 1398 | break; |
210549eb | 1399 | |
ab8d0fc4 JM |
1400 | if (btrfs_defrag_cancelled(info)) { |
1401 | btrfs_debug(info, "defrag_root cancelled"); | |
210549eb DS |
1402 | ret = -EAGAIN; |
1403 | break; | |
1404 | } | |
e9d0b13b | 1405 | } |
27cdeb70 | 1406 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1407 | return ret; |
e9d0b13b CM |
1408 | } |
1409 | ||
6426c7ad QW |
1410 | /* |
1411 | * Do all special snapshot related qgroup dirty hack. | |
1412 | * | |
1413 | * Will do all needed qgroup inherit and dirty hack like switch commit | |
1414 | * roots inside one transaction and write all btree into disk, to make | |
1415 | * qgroup works. | |
1416 | */ | |
1417 | static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, | |
1418 | struct btrfs_root *src, | |
1419 | struct btrfs_root *parent, | |
1420 | struct btrfs_qgroup_inherit *inherit, | |
1421 | u64 dst_objectid) | |
1422 | { | |
1423 | struct btrfs_fs_info *fs_info = src->fs_info; | |
1424 | int ret; | |
1425 | ||
1426 | /* | |
1427 | * Save some performance in the case that qgroups are not | |
1428 | * enabled. If this check races with the ioctl, rescan will | |
1429 | * kick in anyway. | |
1430 | */ | |
9ea6e2b5 | 1431 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
6426c7ad | 1432 | return 0; |
6426c7ad | 1433 | |
4d31778a | 1434 | /* |
52042d8e | 1435 | * Ensure dirty @src will be committed. Or, after coming |
4d31778a QW |
1436 | * commit_fs_roots() and switch_commit_roots(), any dirty but not |
1437 | * recorded root will never be updated again, causing an outdated root | |
1438 | * item. | |
1439 | */ | |
1440 | record_root_in_trans(trans, src, 1); | |
1441 | ||
2a4d84c1 JB |
1442 | /* |
1443 | * btrfs_qgroup_inherit relies on a consistent view of the usage for the | |
1444 | * src root, so we must run the delayed refs here. | |
1445 | * | |
1446 | * However this isn't particularly fool proof, because there's no | |
1447 | * synchronization keeping us from changing the tree after this point | |
1448 | * before we do the qgroup_inherit, or even from making changes while | |
1449 | * we're doing the qgroup_inherit. But that's a problem for the future, | |
1450 | * for now flush the delayed refs to narrow the race window where the | |
1451 | * qgroup counters could end up wrong. | |
1452 | */ | |
1453 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); | |
1454 | if (ret) { | |
1455 | btrfs_abort_transaction(trans, ret); | |
1456 | goto out; | |
1457 | } | |
1458 | ||
6426c7ad QW |
1459 | /* |
1460 | * We are going to commit transaction, see btrfs_commit_transaction() | |
1461 | * comment for reason locking tree_log_mutex | |
1462 | */ | |
1463 | mutex_lock(&fs_info->tree_log_mutex); | |
1464 | ||
7e4443d9 | 1465 | ret = commit_fs_roots(trans); |
6426c7ad QW |
1466 | if (ret) |
1467 | goto out; | |
460fb20a | 1468 | ret = btrfs_qgroup_account_extents(trans); |
6426c7ad QW |
1469 | if (ret < 0) |
1470 | goto out; | |
1471 | ||
1472 | /* Now qgroup are all updated, we can inherit it to new qgroups */ | |
a9377422 | 1473 | ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, |
6426c7ad QW |
1474 | inherit); |
1475 | if (ret < 0) | |
1476 | goto out; | |
1477 | ||
1478 | /* | |
1479 | * Now we do a simplified commit transaction, which will: | |
1480 | * 1) commit all subvolume and extent tree | |
1481 | * To ensure all subvolume and extent tree have a valid | |
1482 | * commit_root to accounting later insert_dir_item() | |
1483 | * 2) write all btree blocks onto disk | |
1484 | * This is to make sure later btree modification will be cowed | |
1485 | * Or commit_root can be populated and cause wrong qgroup numbers | |
1486 | * In this simplified commit, we don't really care about other trees | |
1487 | * like chunk and root tree, as they won't affect qgroup. | |
1488 | * And we don't write super to avoid half committed status. | |
1489 | */ | |
9386d8bc | 1490 | ret = commit_cowonly_roots(trans); |
6426c7ad QW |
1491 | if (ret) |
1492 | goto out; | |
889bfa39 | 1493 | switch_commit_roots(trans); |
70458a58 | 1494 | ret = btrfs_write_and_wait_transaction(trans); |
6426c7ad | 1495 | if (ret) |
f7af3934 | 1496 | btrfs_handle_fs_error(fs_info, ret, |
6426c7ad QW |
1497 | "Error while writing out transaction for qgroup"); |
1498 | ||
1499 | out: | |
1500 | mutex_unlock(&fs_info->tree_log_mutex); | |
1501 | ||
1502 | /* | |
1503 | * Force parent root to be updated, as we recorded it before so its | |
1504 | * last_trans == cur_transid. | |
1505 | * Or it won't be committed again onto disk after later | |
1506 | * insert_dir_item() | |
1507 | */ | |
1508 | if (!ret) | |
1509 | record_root_in_trans(trans, parent, 1); | |
1510 | return ret; | |
1511 | } | |
1512 | ||
d352ac68 CM |
1513 | /* |
1514 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1515 | * transaction commit. This does the actual creation. |
1516 | * | |
1517 | * Note: | |
1518 | * If the error which may affect the commitment of the current transaction | |
1519 | * happens, we should return the error number. If the error which just affect | |
1520 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1521 | */ |
80b6794d | 1522 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1523 | struct btrfs_pending_snapshot *pending) |
1524 | { | |
08d50ca3 NB |
1525 | |
1526 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3063d29f | 1527 | struct btrfs_key key; |
80b6794d | 1528 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1529 | struct btrfs_root *tree_root = fs_info->tree_root; |
1530 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1531 | struct btrfs_root *parent_root; |
98c9942a | 1532 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1533 | struct inode *parent_inode; |
42874b3d MX |
1534 | struct btrfs_path *path; |
1535 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1536 | struct dentry *dentry; |
3063d29f | 1537 | struct extent_buffer *tmp; |
925baedd | 1538 | struct extent_buffer *old; |
95582b00 | 1539 | struct timespec64 cur_time; |
aec8030a | 1540 | int ret = 0; |
d68fc57b | 1541 | u64 to_reserve = 0; |
6bdb72de | 1542 | u64 index = 0; |
a22285a6 | 1543 | u64 objectid; |
b83cc969 | 1544 | u64 root_flags; |
3063d29f | 1545 | |
8546b570 DS |
1546 | ASSERT(pending->path); |
1547 | path = pending->path; | |
42874b3d | 1548 | |
b0c0ea63 DS |
1549 | ASSERT(pending->root_item); |
1550 | new_root_item = pending->root_item; | |
a22285a6 | 1551 | |
543068a2 | 1552 | pending->error = btrfs_get_free_objectid(tree_root, &objectid); |
aec8030a | 1553 | if (pending->error) |
6fa9700e | 1554 | goto no_free_objectid; |
3063d29f | 1555 | |
d6726335 QW |
1556 | /* |
1557 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1558 | * accounted by later btrfs_qgroup_inherit(). | |
1559 | */ | |
1560 | btrfs_set_skip_qgroup(trans, objectid); | |
1561 | ||
147d256e | 1562 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
d68fc57b YZ |
1563 | |
1564 | if (to_reserve > 0) { | |
aec8030a MX |
1565 | pending->error = btrfs_block_rsv_add(root, |
1566 | &pending->block_rsv, | |
1567 | to_reserve, | |
1568 | BTRFS_RESERVE_NO_FLUSH); | |
1569 | if (pending->error) | |
d6726335 | 1570 | goto clear_skip_qgroup; |
d68fc57b YZ |
1571 | } |
1572 | ||
3063d29f | 1573 | key.objectid = objectid; |
a22285a6 YZ |
1574 | key.offset = (u64)-1; |
1575 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1576 | |
6fa9700e | 1577 | rsv = trans->block_rsv; |
a22285a6 | 1578 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1579 | trans->bytes_reserved = trans->block_rsv->reserved; |
0b246afa | 1580 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa JB |
1581 | trans->transid, |
1582 | trans->bytes_reserved, 1); | |
a22285a6 | 1583 | dentry = pending->dentry; |
e9662f70 | 1584 | parent_inode = pending->dir; |
a22285a6 | 1585 | parent_root = BTRFS_I(parent_inode)->root; |
6426c7ad | 1586 | record_root_in_trans(trans, parent_root, 0); |
a22285a6 | 1587 | |
c2050a45 | 1588 | cur_time = current_time(parent_inode); |
04b285f3 | 1589 | |
3063d29f CM |
1590 | /* |
1591 | * insert the directory item | |
1592 | */ | |
877574e2 | 1593 | ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); |
49b25e05 | 1594 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1595 | |
1596 | /* check if there is a file/dir which has the same name. */ | |
1597 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
4a0cc7ca | 1598 | btrfs_ino(BTRFS_I(parent_inode)), |
42874b3d MX |
1599 | dentry->d_name.name, |
1600 | dentry->d_name.len, 0); | |
1601 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1602 | pending->error = -EEXIST; |
aec8030a | 1603 | goto dir_item_existed; |
42874b3d MX |
1604 | } else if (IS_ERR(dir_item)) { |
1605 | ret = PTR_ERR(dir_item); | |
66642832 | 1606 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1607 | goto fail; |
79787eaa | 1608 | } |
42874b3d | 1609 | btrfs_release_path(path); |
52c26179 | 1610 | |
e999376f CM |
1611 | /* |
1612 | * pull in the delayed directory update | |
1613 | * and the delayed inode item | |
1614 | * otherwise we corrupt the FS during | |
1615 | * snapshot | |
1616 | */ | |
e5c304e6 | 1617 | ret = btrfs_run_delayed_items(trans); |
8732d44f | 1618 | if (ret) { /* Transaction aborted */ |
66642832 | 1619 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1620 | goto fail; |
1621 | } | |
e999376f | 1622 | |
6426c7ad | 1623 | record_root_in_trans(trans, root, 0); |
6bdb72de SW |
1624 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1625 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1626 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1627 | |
b83cc969 LZ |
1628 | root_flags = btrfs_root_flags(new_root_item); |
1629 | if (pending->readonly) | |
1630 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1631 | else | |
1632 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1633 | btrfs_set_root_flags(new_root_item, root_flags); | |
1634 | ||
8ea05e3a AB |
1635 | btrfs_set_root_generation_v2(new_root_item, |
1636 | trans->transid); | |
807fc790 | 1637 | generate_random_guid(new_root_item->uuid); |
8ea05e3a AB |
1638 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, |
1639 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1640 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1641 | memset(new_root_item->received_uuid, 0, | |
1642 | sizeof(new_root_item->received_uuid)); | |
1643 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1644 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1645 | btrfs_set_root_stransid(new_root_item, 0); | |
1646 | btrfs_set_root_rtransid(new_root_item, 0); | |
1647 | } | |
3cae210f QW |
1648 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1649 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1650 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1651 | |
6bdb72de | 1652 | old = btrfs_lock_root_node(root); |
9631e4cc JB |
1653 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, |
1654 | BTRFS_NESTING_COW); | |
79787eaa JM |
1655 | if (ret) { |
1656 | btrfs_tree_unlock(old); | |
1657 | free_extent_buffer(old); | |
66642832 | 1658 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1659 | goto fail; |
79787eaa | 1660 | } |
49b25e05 | 1661 | |
49b25e05 | 1662 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1663 | /* clean up in any case */ |
6bdb72de SW |
1664 | btrfs_tree_unlock(old); |
1665 | free_extent_buffer(old); | |
8732d44f | 1666 | if (ret) { |
66642832 | 1667 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1668 | goto fail; |
1669 | } | |
f1ebcc74 | 1670 | /* see comments in should_cow_block() */ |
27cdeb70 | 1671 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1672 | smp_wmb(); |
1673 | ||
6bdb72de | 1674 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1675 | /* record when the snapshot was created in key.offset */ |
1676 | key.offset = trans->transid; | |
1677 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1678 | btrfs_tree_unlock(tmp); |
1679 | free_extent_buffer(tmp); | |
8732d44f | 1680 | if (ret) { |
66642832 | 1681 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1682 | goto fail; |
1683 | } | |
6bdb72de | 1684 | |
a22285a6 YZ |
1685 | /* |
1686 | * insert root back/forward references | |
1687 | */ | |
6025c19f | 1688 | ret = btrfs_add_root_ref(trans, objectid, |
0660b5af | 1689 | parent_root->root_key.objectid, |
4a0cc7ca | 1690 | btrfs_ino(BTRFS_I(parent_inode)), index, |
a22285a6 | 1691 | dentry->d_name.name, dentry->d_name.len); |
8732d44f | 1692 | if (ret) { |
66642832 | 1693 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1694 | goto fail; |
1695 | } | |
0660b5af | 1696 | |
a22285a6 | 1697 | key.offset = (u64)-1; |
2dfb1e43 | 1698 | pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); |
79787eaa JM |
1699 | if (IS_ERR(pending->snap)) { |
1700 | ret = PTR_ERR(pending->snap); | |
2d892ccd | 1701 | pending->snap = NULL; |
66642832 | 1702 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1703 | goto fail; |
79787eaa | 1704 | } |
d68fc57b | 1705 | |
49b25e05 | 1706 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f | 1707 | if (ret) { |
66642832 | 1708 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1709 | goto fail; |
1710 | } | |
361048f5 | 1711 | |
6426c7ad QW |
1712 | /* |
1713 | * Do special qgroup accounting for snapshot, as we do some qgroup | |
1714 | * snapshot hack to do fast snapshot. | |
1715 | * To co-operate with that hack, we do hack again. | |
1716 | * Or snapshot will be greatly slowed down by a subtree qgroup rescan | |
1717 | */ | |
1718 | ret = qgroup_account_snapshot(trans, root, parent_root, | |
1719 | pending->inherit, objectid); | |
1720 | if (ret < 0) | |
1721 | goto fail; | |
1722 | ||
684572df LF |
1723 | ret = btrfs_insert_dir_item(trans, dentry->d_name.name, |
1724 | dentry->d_name.len, BTRFS_I(parent_inode), | |
1725 | &key, BTRFS_FT_DIR, index); | |
42874b3d | 1726 | /* We have check then name at the beginning, so it is impossible. */ |
9c52057c | 1727 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f | 1728 | if (ret) { |
66642832 | 1729 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1730 | goto fail; |
1731 | } | |
42874b3d | 1732 | |
6ef06d27 | 1733 | btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + |
42874b3d | 1734 | dentry->d_name.len * 2); |
04b285f3 | 1735 | parent_inode->i_mtime = parent_inode->i_ctime = |
c2050a45 | 1736 | current_time(parent_inode); |
729f7961 | 1737 | ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); |
dd5f9615 | 1738 | if (ret) { |
66642832 | 1739 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1740 | goto fail; |
1741 | } | |
807fc790 AS |
1742 | ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, |
1743 | BTRFS_UUID_KEY_SUBVOL, | |
cdb345a8 | 1744 | objectid); |
dd5f9615 | 1745 | if (ret) { |
66642832 | 1746 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1747 | goto fail; |
1748 | } | |
1749 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
cdb345a8 | 1750 | ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, |
dd5f9615 SB |
1751 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, |
1752 | objectid); | |
1753 | if (ret && ret != -EEXIST) { | |
66642832 | 1754 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1755 | goto fail; |
1756 | } | |
1757 | } | |
d6726335 | 1758 | |
3063d29f | 1759 | fail: |
aec8030a MX |
1760 | pending->error = ret; |
1761 | dir_item_existed: | |
98c9942a | 1762 | trans->block_rsv = rsv; |
2382c5cc | 1763 | trans->bytes_reserved = 0; |
d6726335 QW |
1764 | clear_skip_qgroup: |
1765 | btrfs_clear_skip_qgroup(trans); | |
6fa9700e MX |
1766 | no_free_objectid: |
1767 | kfree(new_root_item); | |
b0c0ea63 | 1768 | pending->root_item = NULL; |
42874b3d | 1769 | btrfs_free_path(path); |
8546b570 DS |
1770 | pending->path = NULL; |
1771 | ||
49b25e05 | 1772 | return ret; |
3063d29f CM |
1773 | } |
1774 | ||
d352ac68 CM |
1775 | /* |
1776 | * create all the snapshots we've scheduled for creation | |
1777 | */ | |
08d50ca3 | 1778 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) |
3de4586c | 1779 | { |
aec8030a | 1780 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1781 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1782 | int ret = 0; |
3de4586c | 1783 | |
aec8030a MX |
1784 | list_for_each_entry_safe(pending, next, head, list) { |
1785 | list_del(&pending->list); | |
08d50ca3 | 1786 | ret = create_pending_snapshot(trans, pending); |
aec8030a MX |
1787 | if (ret) |
1788 | break; | |
1789 | } | |
1790 | return ret; | |
3de4586c CM |
1791 | } |
1792 | ||
2ff7e61e | 1793 | static void update_super_roots(struct btrfs_fs_info *fs_info) |
5d4f98a2 YZ |
1794 | { |
1795 | struct btrfs_root_item *root_item; | |
1796 | struct btrfs_super_block *super; | |
1797 | ||
0b246afa | 1798 | super = fs_info->super_copy; |
5d4f98a2 | 1799 | |
0b246afa | 1800 | root_item = &fs_info->chunk_root->root_item; |
093e037c DS |
1801 | super->chunk_root = root_item->bytenr; |
1802 | super->chunk_root_generation = root_item->generation; | |
1803 | super->chunk_root_level = root_item->level; | |
5d4f98a2 | 1804 | |
0b246afa | 1805 | root_item = &fs_info->tree_root->root_item; |
093e037c DS |
1806 | super->root = root_item->bytenr; |
1807 | super->generation = root_item->generation; | |
1808 | super->root_level = root_item->level; | |
0b246afa | 1809 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) |
093e037c | 1810 | super->cache_generation = root_item->generation; |
94846229 BB |
1811 | else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) |
1812 | super->cache_generation = 0; | |
0b246afa | 1813 | if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) |
093e037c | 1814 | super->uuid_tree_generation = root_item->generation; |
5d4f98a2 YZ |
1815 | } |
1816 | ||
f36f3042 CM |
1817 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1818 | { | |
4a9d8bde | 1819 | struct btrfs_transaction *trans; |
f36f3042 | 1820 | int ret = 0; |
4a9d8bde | 1821 | |
a4abeea4 | 1822 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1823 | trans = info->running_transaction; |
1824 | if (trans) | |
1825 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1826 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1827 | return ret; |
1828 | } | |
1829 | ||
8929ecfa YZ |
1830 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1831 | { | |
4a9d8bde | 1832 | struct btrfs_transaction *trans; |
8929ecfa | 1833 | int ret = 0; |
4a9d8bde | 1834 | |
a4abeea4 | 1835 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1836 | trans = info->running_transaction; |
1837 | if (trans) | |
1838 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1839 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1840 | return ret; |
1841 | } | |
1842 | ||
bb9c12c9 SW |
1843 | /* |
1844 | * wait for the current transaction commit to start and block subsequent | |
1845 | * transaction joins | |
1846 | */ | |
2ff7e61e | 1847 | static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info, |
bb9c12c9 SW |
1848 | struct btrfs_transaction *trans) |
1849 | { | |
2ff7e61e | 1850 | wait_event(fs_info->transaction_blocked_wait, |
bf31f87f DS |
1851 | trans->state >= TRANS_STATE_COMMIT_START || |
1852 | TRANS_ABORTED(trans)); | |
bb9c12c9 SW |
1853 | } |
1854 | ||
1855 | /* | |
1856 | * wait for the current transaction to start and then become unblocked. | |
1857 | * caller holds ref. | |
1858 | */ | |
2ff7e61e JM |
1859 | static void wait_current_trans_commit_start_and_unblock( |
1860 | struct btrfs_fs_info *fs_info, | |
1861 | struct btrfs_transaction *trans) | |
bb9c12c9 | 1862 | { |
2ff7e61e | 1863 | wait_event(fs_info->transaction_wait, |
bf31f87f DS |
1864 | trans->state >= TRANS_STATE_UNBLOCKED || |
1865 | TRANS_ABORTED(trans)); | |
bb9c12c9 SW |
1866 | } |
1867 | ||
1868 | /* | |
1869 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1870 | * returns, any subsequent transaction will not be allowed to join. | |
1871 | */ | |
1872 | struct btrfs_async_commit { | |
1873 | struct btrfs_trans_handle *newtrans; | |
7892b5af | 1874 | struct work_struct work; |
bb9c12c9 SW |
1875 | }; |
1876 | ||
1877 | static void do_async_commit(struct work_struct *work) | |
1878 | { | |
1879 | struct btrfs_async_commit *ac = | |
7892b5af | 1880 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1881 | |
6fc4e354 SW |
1882 | /* |
1883 | * We've got freeze protection passed with the transaction. | |
1884 | * Tell lockdep about it. | |
1885 | */ | |
b1a06a4b | 1886 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
3a45bb20 | 1887 | __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1888 | |
e209db7a SW |
1889 | current->journal_info = ac->newtrans; |
1890 | ||
3a45bb20 | 1891 | btrfs_commit_transaction(ac->newtrans); |
bb9c12c9 SW |
1892 | kfree(ac); |
1893 | } | |
1894 | ||
1895 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
bb9c12c9 SW |
1896 | int wait_for_unblock) |
1897 | { | |
3a45bb20 | 1898 | struct btrfs_fs_info *fs_info = trans->fs_info; |
bb9c12c9 SW |
1899 | struct btrfs_async_commit *ac; |
1900 | struct btrfs_transaction *cur_trans; | |
1901 | ||
1902 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1903 | if (!ac) |
1904 | return -ENOMEM; | |
bb9c12c9 | 1905 | |
7892b5af | 1906 | INIT_WORK(&ac->work, do_async_commit); |
3a45bb20 | 1907 | ac->newtrans = btrfs_join_transaction(trans->root); |
3612b495 TI |
1908 | if (IS_ERR(ac->newtrans)) { |
1909 | int err = PTR_ERR(ac->newtrans); | |
1910 | kfree(ac); | |
1911 | return err; | |
1912 | } | |
bb9c12c9 SW |
1913 | |
1914 | /* take transaction reference */ | |
bb9c12c9 | 1915 | cur_trans = trans->transaction; |
9b64f57d | 1916 | refcount_inc(&cur_trans->use_count); |
bb9c12c9 | 1917 | |
3a45bb20 | 1918 | btrfs_end_transaction(trans); |
6fc4e354 SW |
1919 | |
1920 | /* | |
1921 | * Tell lockdep we've released the freeze rwsem, since the | |
1922 | * async commit thread will be the one to unlock it. | |
1923 | */ | |
b1a06a4b | 1924 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
0b246afa | 1925 | __sb_writers_release(fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1926 | |
7892b5af | 1927 | schedule_work(&ac->work); |
bb9c12c9 SW |
1928 | |
1929 | /* wait for transaction to start and unblock */ | |
bb9c12c9 | 1930 | if (wait_for_unblock) |
2ff7e61e | 1931 | wait_current_trans_commit_start_and_unblock(fs_info, cur_trans); |
bb9c12c9 | 1932 | else |
2ff7e61e | 1933 | wait_current_trans_commit_start(fs_info, cur_trans); |
bb9c12c9 | 1934 | |
38e88054 SW |
1935 | if (current->journal_info == trans) |
1936 | current->journal_info = NULL; | |
1937 | ||
724e2315 | 1938 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1939 | return 0; |
1940 | } | |
1941 | ||
49b25e05 | 1942 | |
97cb39bb | 1943 | static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) |
49b25e05 | 1944 | { |
97cb39bb | 1945 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 JM |
1946 | struct btrfs_transaction *cur_trans = trans->transaction; |
1947 | ||
b50fff81 | 1948 | WARN_ON(refcount_read(&trans->use_count) > 1); |
49b25e05 | 1949 | |
66642832 | 1950 | btrfs_abort_transaction(trans, err); |
7b8b92af | 1951 | |
0b246afa | 1952 | spin_lock(&fs_info->trans_lock); |
66b6135b | 1953 | |
25d8c284 MX |
1954 | /* |
1955 | * If the transaction is removed from the list, it means this | |
1956 | * transaction has been committed successfully, so it is impossible | |
1957 | * to call the cleanup function. | |
1958 | */ | |
1959 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1960 | |
49b25e05 | 1961 | list_del_init(&cur_trans->list); |
0b246afa | 1962 | if (cur_trans == fs_info->running_transaction) { |
4a9d8bde | 1963 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 1964 | spin_unlock(&fs_info->trans_lock); |
f094ac32 LB |
1965 | wait_event(cur_trans->writer_wait, |
1966 | atomic_read(&cur_trans->num_writers) == 1); | |
1967 | ||
0b246afa | 1968 | spin_lock(&fs_info->trans_lock); |
d7096fc3 | 1969 | } |
0b246afa | 1970 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 1971 | |
2ff7e61e | 1972 | btrfs_cleanup_one_transaction(trans->transaction, fs_info); |
49b25e05 | 1973 | |
0b246afa JM |
1974 | spin_lock(&fs_info->trans_lock); |
1975 | if (cur_trans == fs_info->running_transaction) | |
1976 | fs_info->running_transaction = NULL; | |
1977 | spin_unlock(&fs_info->trans_lock); | |
4a9d8bde | 1978 | |
e0228285 | 1979 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 1980 | sb_end_intwrite(fs_info->sb); |
724e2315 JB |
1981 | btrfs_put_transaction(cur_trans); |
1982 | btrfs_put_transaction(cur_trans); | |
49b25e05 | 1983 | |
97cb39bb | 1984 | trace_btrfs_transaction_commit(trans->root); |
49b25e05 | 1985 | |
49b25e05 JM |
1986 | if (current->journal_info == trans) |
1987 | current->journal_info = NULL; | |
0b246afa | 1988 | btrfs_scrub_cancel(fs_info); |
49b25e05 JM |
1989 | |
1990 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1991 | } | |
1992 | ||
c7cc64a9 DS |
1993 | /* |
1994 | * Release reserved delayed ref space of all pending block groups of the | |
1995 | * transaction and remove them from the list | |
1996 | */ | |
1997 | static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) | |
1998 | { | |
1999 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2000 | struct btrfs_block_group *block_group, *tmp; |
c7cc64a9 DS |
2001 | |
2002 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { | |
2003 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
2004 | list_del_init(&block_group->bg_list); | |
2005 | } | |
2006 | } | |
2007 | ||
88090ad3 | 2008 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
82436617 | 2009 | { |
ce8ea7cc JB |
2010 | /* |
2011 | * We use writeback_inodes_sb here because if we used | |
2012 | * btrfs_start_delalloc_roots we would deadlock with fs freeze. | |
2013 | * Currently are holding the fs freeze lock, if we do an async flush | |
2014 | * we'll do btrfs_join_transaction() and deadlock because we need to | |
2015 | * wait for the fs freeze lock. Using the direct flushing we benefit | |
2016 | * from already being in a transaction and our join_transaction doesn't | |
2017 | * have to re-take the fs freeze lock. | |
2018 | */ | |
88090ad3 | 2019 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
ce8ea7cc | 2020 | writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); |
82436617 MX |
2021 | return 0; |
2022 | } | |
2023 | ||
88090ad3 | 2024 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) |
82436617 | 2025 | { |
88090ad3 | 2026 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
6374e57a | 2027 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); |
82436617 MX |
2028 | } |
2029 | ||
3a45bb20 | 2030 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans) |
79154b1b | 2031 | { |
3a45bb20 | 2032 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 | 2033 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 2034 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 2035 | int ret; |
79154b1b | 2036 | |
35b814f3 NB |
2037 | ASSERT(refcount_read(&trans->use_count) == 1); |
2038 | ||
d62b23c9 JB |
2039 | /* |
2040 | * Some places just start a transaction to commit it. We need to make | |
2041 | * sure that if this commit fails that the abort code actually marks the | |
2042 | * transaction as failed, so set trans->dirty to make the abort code do | |
2043 | * the right thing. | |
2044 | */ | |
2045 | trans->dirty = true; | |
2046 | ||
8d25a086 | 2047 | /* Stop the commit early if ->aborted is set */ |
bf31f87f | 2048 | if (TRANS_ABORTED(cur_trans)) { |
25287e0a | 2049 | ret = cur_trans->aborted; |
3a45bb20 | 2050 | btrfs_end_transaction(trans); |
e4a2bcac | 2051 | return ret; |
25287e0a | 2052 | } |
49b25e05 | 2053 | |
f45c752b JB |
2054 | btrfs_trans_release_metadata(trans); |
2055 | trans->block_rsv = NULL; | |
2056 | ||
56bec294 | 2057 | /* |
e19eb11f JB |
2058 | * We only want one transaction commit doing the flushing so we do not |
2059 | * waste a bunch of time on lock contention on the extent root node. | |
56bec294 | 2060 | */ |
e19eb11f JB |
2061 | if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, |
2062 | &cur_trans->delayed_refs.flags)) { | |
2063 | /* | |
2064 | * Make a pass through all the delayed refs we have so far. | |
2065 | * Any running threads may add more while we are here. | |
2066 | */ | |
2067 | ret = btrfs_run_delayed_refs(trans, 0); | |
2068 | if (ret) { | |
2069 | btrfs_end_transaction(trans); | |
2070 | return ret; | |
2071 | } | |
2072 | } | |
56bec294 | 2073 | |
119e80df | 2074 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2075 | |
3204d33c | 2076 | if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { |
1bbc621e CM |
2077 | int run_it = 0; |
2078 | ||
2079 | /* this mutex is also taken before trying to set | |
2080 | * block groups readonly. We need to make sure | |
2081 | * that nobody has set a block group readonly | |
2082 | * after a extents from that block group have been | |
2083 | * allocated for cache files. btrfs_set_block_group_ro | |
2084 | * will wait for the transaction to commit if it | |
3204d33c | 2085 | * finds BTRFS_TRANS_DIRTY_BG_RUN set. |
1bbc621e | 2086 | * |
3204d33c JB |
2087 | * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure |
2088 | * only one process starts all the block group IO. It wouldn't | |
1bbc621e CM |
2089 | * hurt to have more than one go through, but there's no |
2090 | * real advantage to it either. | |
2091 | */ | |
0b246afa | 2092 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c JB |
2093 | if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, |
2094 | &cur_trans->flags)) | |
1bbc621e | 2095 | run_it = 1; |
0b246afa | 2096 | mutex_unlock(&fs_info->ro_block_group_mutex); |
1bbc621e | 2097 | |
f9cacae3 | 2098 | if (run_it) { |
21217054 | 2099 | ret = btrfs_start_dirty_block_groups(trans); |
f9cacae3 NB |
2100 | if (ret) { |
2101 | btrfs_end_transaction(trans); | |
2102 | return ret; | |
2103 | } | |
2104 | } | |
1bbc621e CM |
2105 | } |
2106 | ||
0b246afa | 2107 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2108 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { |
d0c2f4fa FM |
2109 | enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; |
2110 | ||
0b246afa | 2111 | spin_unlock(&fs_info->trans_lock); |
9b64f57d | 2112 | refcount_inc(&cur_trans->use_count); |
ccd467d6 | 2113 | |
d0c2f4fa FM |
2114 | if (trans->in_fsync) |
2115 | want_state = TRANS_STATE_SUPER_COMMITTED; | |
2116 | ret = btrfs_end_transaction(trans); | |
2117 | wait_for_commit(cur_trans, want_state); | |
15ee9bc7 | 2118 | |
bf31f87f | 2119 | if (TRANS_ABORTED(cur_trans)) |
b4924a0f LB |
2120 | ret = cur_trans->aborted; |
2121 | ||
724e2315 | 2122 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 2123 | |
49b25e05 | 2124 | return ret; |
79154b1b | 2125 | } |
4313b399 | 2126 | |
4a9d8bde | 2127 | cur_trans->state = TRANS_STATE_COMMIT_START; |
0b246afa | 2128 | wake_up(&fs_info->transaction_blocked_wait); |
bb9c12c9 | 2129 | |
0b246afa | 2130 | if (cur_trans->list.prev != &fs_info->trans_list) { |
d0c2f4fa FM |
2131 | enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; |
2132 | ||
2133 | if (trans->in_fsync) | |
2134 | want_state = TRANS_STATE_SUPER_COMMITTED; | |
2135 | ||
ccd467d6 CM |
2136 | prev_trans = list_entry(cur_trans->list.prev, |
2137 | struct btrfs_transaction, list); | |
d0c2f4fa | 2138 | if (prev_trans->state < want_state) { |
9b64f57d | 2139 | refcount_inc(&prev_trans->use_count); |
0b246afa | 2140 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2141 | |
d0c2f4fa FM |
2142 | wait_for_commit(prev_trans, want_state); |
2143 | ||
bf31f87f | 2144 | ret = READ_ONCE(prev_trans->aborted); |
ccd467d6 | 2145 | |
724e2315 | 2146 | btrfs_put_transaction(prev_trans); |
1f9b8c8f FM |
2147 | if (ret) |
2148 | goto cleanup_transaction; | |
a4abeea4 | 2149 | } else { |
0b246afa | 2150 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2151 | } |
a4abeea4 | 2152 | } else { |
0b246afa | 2153 | spin_unlock(&fs_info->trans_lock); |
cb2d3dad FM |
2154 | /* |
2155 | * The previous transaction was aborted and was already removed | |
2156 | * from the list of transactions at fs_info->trans_list. So we | |
2157 | * abort to prevent writing a new superblock that reflects a | |
2158 | * corrupt state (pointing to trees with unwritten nodes/leafs). | |
2159 | */ | |
2160 | if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) { | |
2161 | ret = -EROFS; | |
2162 | goto cleanup_transaction; | |
2163 | } | |
ccd467d6 | 2164 | } |
15ee9bc7 | 2165 | |
0860adfd MX |
2166 | extwriter_counter_dec(cur_trans, trans->type); |
2167 | ||
88090ad3 | 2168 | ret = btrfs_start_delalloc_flush(fs_info); |
82436617 MX |
2169 | if (ret) |
2170 | goto cleanup_transaction; | |
2171 | ||
e5c304e6 | 2172 | ret = btrfs_run_delayed_items(trans); |
581227d0 MX |
2173 | if (ret) |
2174 | goto cleanup_transaction; | |
15ee9bc7 | 2175 | |
581227d0 MX |
2176 | wait_event(cur_trans->writer_wait, |
2177 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 2178 | |
581227d0 | 2179 | /* some pending stuffs might be added after the previous flush. */ |
e5c304e6 | 2180 | ret = btrfs_run_delayed_items(trans); |
ca469637 MX |
2181 | if (ret) |
2182 | goto cleanup_transaction; | |
2183 | ||
88090ad3 | 2184 | btrfs_wait_delalloc_flush(fs_info); |
cb7ab021 | 2185 | |
48778179 FM |
2186 | /* |
2187 | * Wait for all ordered extents started by a fast fsync that joined this | |
2188 | * transaction. Otherwise if this transaction commits before the ordered | |
2189 | * extents complete we lose logged data after a power failure. | |
2190 | */ | |
2191 | wait_event(cur_trans->pending_wait, | |
2192 | atomic_read(&cur_trans->pending_ordered) == 0); | |
2193 | ||
2ff7e61e | 2194 | btrfs_scrub_pause(fs_info); |
ed0ca140 JB |
2195 | /* |
2196 | * Ok now we need to make sure to block out any other joins while we | |
2197 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 2198 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 | 2199 | */ |
0b246afa | 2200 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2201 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 2202 | spin_unlock(&fs_info->trans_lock); |
ed0ca140 JB |
2203 | wait_event(cur_trans->writer_wait, |
2204 | atomic_read(&cur_trans->num_writers) == 1); | |
2205 | ||
bf31f87f | 2206 | if (TRANS_ABORTED(cur_trans)) { |
2cba30f1 | 2207 | ret = cur_trans->aborted; |
6cf7f77e | 2208 | goto scrub_continue; |
2cba30f1 | 2209 | } |
7585717f CM |
2210 | /* |
2211 | * the reloc mutex makes sure that we stop | |
2212 | * the balancing code from coming in and moving | |
2213 | * extents around in the middle of the commit | |
2214 | */ | |
0b246afa | 2215 | mutex_lock(&fs_info->reloc_mutex); |
7585717f | 2216 | |
42874b3d MX |
2217 | /* |
2218 | * We needn't worry about the delayed items because we will | |
2219 | * deal with them in create_pending_snapshot(), which is the | |
2220 | * core function of the snapshot creation. | |
2221 | */ | |
08d50ca3 | 2222 | ret = create_pending_snapshots(trans); |
56e9f6ea DS |
2223 | if (ret) |
2224 | goto unlock_reloc; | |
3063d29f | 2225 | |
42874b3d MX |
2226 | /* |
2227 | * We insert the dir indexes of the snapshots and update the inode | |
2228 | * of the snapshots' parents after the snapshot creation, so there | |
2229 | * are some delayed items which are not dealt with. Now deal with | |
2230 | * them. | |
2231 | * | |
2232 | * We needn't worry that this operation will corrupt the snapshots, | |
2233 | * because all the tree which are snapshoted will be forced to COW | |
2234 | * the nodes and leaves. | |
2235 | */ | |
e5c304e6 | 2236 | ret = btrfs_run_delayed_items(trans); |
56e9f6ea DS |
2237 | if (ret) |
2238 | goto unlock_reloc; | |
16cdcec7 | 2239 | |
c79a70b1 | 2240 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
56e9f6ea DS |
2241 | if (ret) |
2242 | goto unlock_reloc; | |
56bec294 | 2243 | |
e999376f CM |
2244 | /* |
2245 | * make sure none of the code above managed to slip in a | |
2246 | * delayed item | |
2247 | */ | |
ccdf9b30 | 2248 | btrfs_assert_delayed_root_empty(fs_info); |
e999376f | 2249 | |
2c90e5d6 | 2250 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 2251 | |
e02119d5 CM |
2252 | /* btrfs_commit_tree_roots is responsible for getting the |
2253 | * various roots consistent with each other. Every pointer | |
2254 | * in the tree of tree roots has to point to the most up to date | |
2255 | * root for every subvolume and other tree. So, we have to keep | |
2256 | * the tree logging code from jumping in and changing any | |
2257 | * of the trees. | |
2258 | * | |
2259 | * At this point in the commit, there can't be any tree-log | |
2260 | * writers, but a little lower down we drop the trans mutex | |
2261 | * and let new people in. By holding the tree_log_mutex | |
2262 | * from now until after the super is written, we avoid races | |
2263 | * with the tree-log code. | |
2264 | */ | |
0b246afa | 2265 | mutex_lock(&fs_info->tree_log_mutex); |
e02119d5 | 2266 | |
7e4443d9 | 2267 | ret = commit_fs_roots(trans); |
56e9f6ea DS |
2268 | if (ret) |
2269 | goto unlock_tree_log; | |
54aa1f4d | 2270 | |
3818aea2 | 2271 | /* |
7e1876ac DS |
2272 | * Since the transaction is done, we can apply the pending changes |
2273 | * before the next transaction. | |
3818aea2 | 2274 | */ |
0b246afa | 2275 | btrfs_apply_pending_changes(fs_info); |
3818aea2 | 2276 | |
5d4f98a2 | 2277 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2278 | * safe to free the root of tree log roots |
2279 | */ | |
0b246afa | 2280 | btrfs_free_log_root_tree(trans, fs_info); |
e02119d5 | 2281 | |
0ed4792a QW |
2282 | /* |
2283 | * Since fs roots are all committed, we can get a quite accurate | |
2284 | * new_roots. So let's do quota accounting. | |
2285 | */ | |
460fb20a | 2286 | ret = btrfs_qgroup_account_extents(trans); |
56e9f6ea DS |
2287 | if (ret < 0) |
2288 | goto unlock_tree_log; | |
0ed4792a | 2289 | |
9386d8bc | 2290 | ret = commit_cowonly_roots(trans); |
56e9f6ea DS |
2291 | if (ret) |
2292 | goto unlock_tree_log; | |
54aa1f4d | 2293 | |
2cba30f1 MX |
2294 | /* |
2295 | * The tasks which save the space cache and inode cache may also | |
2296 | * update ->aborted, check it. | |
2297 | */ | |
bf31f87f | 2298 | if (TRANS_ABORTED(cur_trans)) { |
2cba30f1 | 2299 | ret = cur_trans->aborted; |
56e9f6ea | 2300 | goto unlock_tree_log; |
2cba30f1 MX |
2301 | } |
2302 | ||
0b246afa | 2303 | cur_trans = fs_info->running_transaction; |
5d4f98a2 | 2304 | |
0b246afa JM |
2305 | btrfs_set_root_node(&fs_info->tree_root->root_item, |
2306 | fs_info->tree_root->node); | |
2307 | list_add_tail(&fs_info->tree_root->dirty_list, | |
9e351cc8 | 2308 | &cur_trans->switch_commits); |
5d4f98a2 | 2309 | |
0b246afa JM |
2310 | btrfs_set_root_node(&fs_info->chunk_root->root_item, |
2311 | fs_info->chunk_root->node); | |
2312 | list_add_tail(&fs_info->chunk_root->dirty_list, | |
9e351cc8 JB |
2313 | &cur_trans->switch_commits); |
2314 | ||
889bfa39 | 2315 | switch_commit_roots(trans); |
5d4f98a2 | 2316 | |
ce93ec54 | 2317 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2318 | ASSERT(list_empty(&cur_trans->io_bgs)); |
2ff7e61e | 2319 | update_super_roots(fs_info); |
e02119d5 | 2320 | |
0b246afa JM |
2321 | btrfs_set_super_log_root(fs_info->super_copy, 0); |
2322 | btrfs_set_super_log_root_level(fs_info->super_copy, 0); | |
2323 | memcpy(fs_info->super_for_commit, fs_info->super_copy, | |
2324 | sizeof(*fs_info->super_copy)); | |
ccd467d6 | 2325 | |
bbbf7243 | 2326 | btrfs_commit_device_sizes(cur_trans); |
935e5cc9 | 2327 | |
0b246afa JM |
2328 | clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
2329 | clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); | |
656f30db | 2330 | |
4fbcdf66 FM |
2331 | btrfs_trans_release_chunk_metadata(trans); |
2332 | ||
0b246afa | 2333 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2334 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
0b246afa JM |
2335 | fs_info->running_transaction = NULL; |
2336 | spin_unlock(&fs_info->trans_lock); | |
2337 | mutex_unlock(&fs_info->reloc_mutex); | |
b7ec40d7 | 2338 | |
0b246afa | 2339 | wake_up(&fs_info->transaction_wait); |
e6dcd2dc | 2340 | |
70458a58 | 2341 | ret = btrfs_write_and_wait_transaction(trans); |
49b25e05 | 2342 | if (ret) { |
0b246afa JM |
2343 | btrfs_handle_fs_error(fs_info, ret, |
2344 | "Error while writing out transaction"); | |
56e9f6ea DS |
2345 | /* |
2346 | * reloc_mutex has been unlocked, tree_log_mutex is still held | |
2347 | * but we can't jump to unlock_tree_log causing double unlock | |
2348 | */ | |
0b246afa | 2349 | mutex_unlock(&fs_info->tree_log_mutex); |
6cf7f77e | 2350 | goto scrub_continue; |
49b25e05 JM |
2351 | } |
2352 | ||
eece6a9c | 2353 | ret = write_all_supers(fs_info, 0); |
e02119d5 CM |
2354 | /* |
2355 | * the super is written, we can safely allow the tree-loggers | |
2356 | * to go about their business | |
2357 | */ | |
0b246afa | 2358 | mutex_unlock(&fs_info->tree_log_mutex); |
c1f32b7c AJ |
2359 | if (ret) |
2360 | goto scrub_continue; | |
e02119d5 | 2361 | |
d0c2f4fa FM |
2362 | /* |
2363 | * We needn't acquire the lock here because there is no other task | |
2364 | * which can change it. | |
2365 | */ | |
2366 | cur_trans->state = TRANS_STATE_SUPER_COMMITTED; | |
2367 | wake_up(&cur_trans->commit_wait); | |
2368 | ||
5ead2dd0 | 2369 | btrfs_finish_extent_commit(trans); |
4313b399 | 2370 | |
3204d33c | 2371 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) |
0b246afa | 2372 | btrfs_clear_space_info_full(fs_info); |
13212b54 | 2373 | |
0b246afa | 2374 | fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
2375 | /* |
2376 | * We needn't acquire the lock here because there is no other task | |
2377 | * which can change it. | |
2378 | */ | |
2379 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2380 | wake_up(&cur_trans->commit_wait); |
3de4586c | 2381 | |
0b246afa | 2382 | spin_lock(&fs_info->trans_lock); |
13c5a93e | 2383 | list_del_init(&cur_trans->list); |
0b246afa | 2384 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 2385 | |
724e2315 JB |
2386 | btrfs_put_transaction(cur_trans); |
2387 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2388 | |
0860adfd | 2389 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2390 | sb_end_intwrite(fs_info->sb); |
b2b5ef5c | 2391 | |
3a45bb20 | 2392 | trace_btrfs_transaction_commit(trans->root); |
1abe9b8a | 2393 | |
2ff7e61e | 2394 | btrfs_scrub_continue(fs_info); |
a2de733c | 2395 | |
9ed74f2d JB |
2396 | if (current->journal_info == trans) |
2397 | current->journal_info = NULL; | |
2398 | ||
2c90e5d6 | 2399 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2400 | |
79154b1b | 2401 | return ret; |
49b25e05 | 2402 | |
56e9f6ea DS |
2403 | unlock_tree_log: |
2404 | mutex_unlock(&fs_info->tree_log_mutex); | |
2405 | unlock_reloc: | |
2406 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2407 | scrub_continue: |
2ff7e61e | 2408 | btrfs_scrub_continue(fs_info); |
49b25e05 | 2409 | cleanup_transaction: |
dc60c525 | 2410 | btrfs_trans_release_metadata(trans); |
c7cc64a9 | 2411 | btrfs_cleanup_pending_block_groups(trans); |
4fbcdf66 | 2412 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2413 | trans->block_rsv = NULL; |
0b246afa | 2414 | btrfs_warn(fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2415 | if (current->journal_info == trans) |
2416 | current->journal_info = NULL; | |
97cb39bb | 2417 | cleanup_transaction(trans, ret); |
49b25e05 JM |
2418 | |
2419 | return ret; | |
79154b1b CM |
2420 | } |
2421 | ||
d352ac68 | 2422 | /* |
9d1a2a3a DS |
2423 | * return < 0 if error |
2424 | * 0 if there are no more dead_roots at the time of call | |
2425 | * 1 there are more to be processed, call me again | |
2426 | * | |
2427 | * The return value indicates there are certainly more snapshots to delete, but | |
2428 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2429 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2430 | * few seconds later. | |
d352ac68 | 2431 | */ |
9d1a2a3a | 2432 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 2433 | { |
9d1a2a3a | 2434 | int ret; |
5d4f98a2 YZ |
2435 | struct btrfs_fs_info *fs_info = root->fs_info; |
2436 | ||
a4abeea4 | 2437 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2438 | if (list_empty(&fs_info->dead_roots)) { |
2439 | spin_unlock(&fs_info->trans_lock); | |
2440 | return 0; | |
2441 | } | |
2442 | root = list_first_entry(&fs_info->dead_roots, | |
2443 | struct btrfs_root, root_list); | |
cfad392b | 2444 | list_del_init(&root->root_list); |
a4abeea4 | 2445 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2446 | |
4fd786e6 | 2447 | btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); |
76dda93c | 2448 | |
9d1a2a3a | 2449 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2450 | |
9d1a2a3a DS |
2451 | if (btrfs_header_backref_rev(root->node) < |
2452 | BTRFS_MIXED_BACKREF_REV) | |
0078a9f9 | 2453 | ret = btrfs_drop_snapshot(root, 0, 0); |
9d1a2a3a | 2454 | else |
0078a9f9 | 2455 | ret = btrfs_drop_snapshot(root, 1, 0); |
32471dc2 | 2456 | |
dc9492c1 | 2457 | btrfs_put_root(root); |
6596a928 | 2458 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2459 | } |
572d9ab7 DS |
2460 | |
2461 | void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) | |
2462 | { | |
2463 | unsigned long prev; | |
2464 | unsigned long bit; | |
2465 | ||
6c9fe14f | 2466 | prev = xchg(&fs_info->pending_changes, 0); |
572d9ab7 DS |
2467 | if (!prev) |
2468 | return; | |
2469 | ||
d51033d0 DS |
2470 | bit = 1 << BTRFS_PENDING_COMMIT; |
2471 | if (prev & bit) | |
2472 | btrfs_debug(fs_info, "pending commit done"); | |
2473 | prev &= ~bit; | |
2474 | ||
572d9ab7 DS |
2475 | if (prev) |
2476 | btrfs_warn(fs_info, | |
2477 | "unknown pending changes left 0x%lx, ignoring", prev); | |
2478 | } |