Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 CM |
4 | */ |
5 | ||
79154b1b | 6 | #include <linux/fs.h> |
5a0e3ad6 | 7 | #include <linux/slab.h> |
34088780 | 8 | #include <linux/sched.h> |
d3c2fdcf | 9 | #include <linux/writeback.h> |
5f39d397 | 10 | #include <linux/pagemap.h> |
5f2cc086 | 11 | #include <linux/blkdev.h> |
8ea05e3a | 12 | #include <linux/uuid.h> |
602cbe91 | 13 | #include "misc.h" |
79154b1b CM |
14 | #include "ctree.h" |
15 | #include "disk-io.h" | |
16 | #include "transaction.h" | |
925baedd | 17 | #include "locking.h" |
e02119d5 | 18 | #include "tree-log.h" |
733f4fbb | 19 | #include "volumes.h" |
8dabb742 | 20 | #include "dev-replace.h" |
fcebe456 | 21 | #include "qgroup.h" |
aac0023c | 22 | #include "block-group.h" |
9c343784 | 23 | #include "space-info.h" |
d3575156 | 24 | #include "zoned.h" |
79154b1b | 25 | |
0f7d52f4 CM |
26 | #define BTRFS_ROOT_TRANS_TAG 0 |
27 | ||
61c047b5 QW |
28 | /* |
29 | * Transaction states and transitions | |
30 | * | |
31 | * No running transaction (fs tree blocks are not modified) | |
32 | * | | |
33 | * | To next stage: | |
34 | * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). | |
35 | * V | |
36 | * Transaction N [[TRANS_STATE_RUNNING]] | |
37 | * | | |
38 | * | New trans handles can be attached to transaction N by calling all | |
39 | * | start_transaction() variants. | |
40 | * | | |
41 | * | To next stage: | |
42 | * | Call btrfs_commit_transaction() on any trans handle attached to | |
43 | * | transaction N | |
44 | * V | |
45 | * Transaction N [[TRANS_STATE_COMMIT_START]] | |
46 | * | | |
47 | * | Will wait for previous running transaction to completely finish if there | |
48 | * | is one | |
49 | * | | |
50 | * | Then one of the following happes: | |
51 | * | - Wait for all other trans handle holders to release. | |
52 | * | The btrfs_commit_transaction() caller will do the commit work. | |
53 | * | - Wait for current transaction to be committed by others. | |
54 | * | Other btrfs_commit_transaction() caller will do the commit work. | |
55 | * | | |
56 | * | At this stage, only btrfs_join_transaction*() variants can attach | |
57 | * | to this running transaction. | |
58 | * | All other variants will wait for current one to finish and attach to | |
59 | * | transaction N+1. | |
60 | * | | |
61 | * | To next stage: | |
62 | * | Caller is chosen to commit transaction N, and all other trans handle | |
63 | * | haven been released. | |
64 | * V | |
65 | * Transaction N [[TRANS_STATE_COMMIT_DOING]] | |
66 | * | | |
67 | * | The heavy lifting transaction work is started. | |
68 | * | From running delayed refs (modifying extent tree) to creating pending | |
69 | * | snapshots, running qgroups. | |
70 | * | In short, modify supporting trees to reflect modifications of subvolume | |
71 | * | trees. | |
72 | * | | |
73 | * | At this stage, all start_transaction() calls will wait for this | |
74 | * | transaction to finish and attach to transaction N+1. | |
75 | * | | |
76 | * | To next stage: | |
77 | * | Until all supporting trees are updated. | |
78 | * V | |
79 | * Transaction N [[TRANS_STATE_UNBLOCKED]] | |
80 | * | Transaction N+1 | |
81 | * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] | |
82 | * | need to write them back to disk and update | | |
83 | * | super blocks. | | |
84 | * | | | |
85 | * | At this stage, new transaction is allowed to | | |
86 | * | start. | | |
87 | * | All new start_transaction() calls will be | | |
88 | * | attached to transid N+1. | | |
89 | * | | | |
90 | * | To next stage: | | |
91 | * | Until all tree blocks are super blocks are | | |
92 | * | written to block devices | | |
93 | * V | | |
94 | * Transaction N [[TRANS_STATE_COMPLETED]] V | |
95 | * All tree blocks and super blocks are written. Transaction N+1 | |
96 | * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] | |
97 | * data structures will be cleaned up. | Life goes on | |
98 | */ | |
e8c9f186 | 99 | static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { |
4a9d8bde | 100 | [TRANS_STATE_RUNNING] = 0U, |
bcf3a3e7 NB |
101 | [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), |
102 | [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | | |
4a9d8bde | 103 | __TRANS_ATTACH | |
a6d155d2 FM |
104 | __TRANS_JOIN | |
105 | __TRANS_JOIN_NOSTART), | |
bcf3a3e7 | 106 | [TRANS_STATE_UNBLOCKED] = (__TRANS_START | |
4a9d8bde MX |
107 | __TRANS_ATTACH | |
108 | __TRANS_JOIN | | |
a6d155d2 FM |
109 | __TRANS_JOIN_NOLOCK | |
110 | __TRANS_JOIN_NOSTART), | |
d0c2f4fa FM |
111 | [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START | |
112 | __TRANS_ATTACH | | |
113 | __TRANS_JOIN | | |
114 | __TRANS_JOIN_NOLOCK | | |
115 | __TRANS_JOIN_NOSTART), | |
bcf3a3e7 | 116 | [TRANS_STATE_COMPLETED] = (__TRANS_START | |
4a9d8bde MX |
117 | __TRANS_ATTACH | |
118 | __TRANS_JOIN | | |
a6d155d2 FM |
119 | __TRANS_JOIN_NOLOCK | |
120 | __TRANS_JOIN_NOSTART), | |
4a9d8bde MX |
121 | }; |
122 | ||
724e2315 | 123 | void btrfs_put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 124 | { |
9b64f57d ER |
125 | WARN_ON(refcount_read(&transaction->use_count) == 0); |
126 | if (refcount_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 127 | BUG_ON(!list_empty(&transaction->list)); |
5c9d028b LB |
128 | WARN_ON(!RB_EMPTY_ROOT( |
129 | &transaction->delayed_refs.href_root.rb_root)); | |
81f7eb00 JM |
130 | WARN_ON(!RB_EMPTY_ROOT( |
131 | &transaction->delayed_refs.dirty_extent_root)); | |
1262133b | 132 | if (transaction->delayed_refs.pending_csums) |
ab8d0fc4 JM |
133 | btrfs_err(transaction->fs_info, |
134 | "pending csums is %llu", | |
135 | transaction->delayed_refs.pending_csums); | |
7785a663 FM |
136 | /* |
137 | * If any block groups are found in ->deleted_bgs then it's | |
138 | * because the transaction was aborted and a commit did not | |
139 | * happen (things failed before writing the new superblock | |
140 | * and calling btrfs_finish_extent_commit()), so we can not | |
141 | * discard the physical locations of the block groups. | |
142 | */ | |
143 | while (!list_empty(&transaction->deleted_bgs)) { | |
32da5386 | 144 | struct btrfs_block_group *cache; |
7785a663 FM |
145 | |
146 | cache = list_first_entry(&transaction->deleted_bgs, | |
32da5386 | 147 | struct btrfs_block_group, |
7785a663 FM |
148 | bg_list); |
149 | list_del_init(&cache->bg_list); | |
6b7304af | 150 | btrfs_unfreeze_block_group(cache); |
7785a663 FM |
151 | btrfs_put_block_group(cache); |
152 | } | |
bbbf7243 | 153 | WARN_ON(!list_empty(&transaction->dev_update_list)); |
4b5faeac | 154 | kfree(transaction); |
78fae27e | 155 | } |
79154b1b CM |
156 | } |
157 | ||
889bfa39 | 158 | static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) |
817d52f8 | 159 | { |
889bfa39 | 160 | struct btrfs_transaction *cur_trans = trans->transaction; |
16916a88 | 161 | struct btrfs_fs_info *fs_info = trans->fs_info; |
9e351cc8 | 162 | struct btrfs_root *root, *tmp; |
27d56e62 | 163 | struct btrfs_caching_control *caching_ctl, *next; |
9e351cc8 JB |
164 | |
165 | down_write(&fs_info->commit_root_sem); | |
889bfa39 | 166 | list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, |
9e351cc8 JB |
167 | dirty_list) { |
168 | list_del_init(&root->dirty_list); | |
169 | free_extent_buffer(root->commit_root); | |
170 | root->commit_root = btrfs_root_node(root); | |
41e7acd3 | 171 | extent_io_tree_release(&root->dirty_log_pages); |
370a11b8 | 172 | btrfs_qgroup_clean_swapped_blocks(root); |
9e351cc8 | 173 | } |
2b9dbef2 JB |
174 | |
175 | /* We can free old roots now. */ | |
889bfa39 JB |
176 | spin_lock(&cur_trans->dropped_roots_lock); |
177 | while (!list_empty(&cur_trans->dropped_roots)) { | |
178 | root = list_first_entry(&cur_trans->dropped_roots, | |
2b9dbef2 JB |
179 | struct btrfs_root, root_list); |
180 | list_del_init(&root->root_list); | |
889bfa39 JB |
181 | spin_unlock(&cur_trans->dropped_roots_lock); |
182 | btrfs_free_log(trans, root); | |
2b9dbef2 | 183 | btrfs_drop_and_free_fs_root(fs_info, root); |
889bfa39 | 184 | spin_lock(&cur_trans->dropped_roots_lock); |
2b9dbef2 | 185 | } |
889bfa39 | 186 | spin_unlock(&cur_trans->dropped_roots_lock); |
27d56e62 JB |
187 | |
188 | /* | |
189 | * We have to update the last_byte_to_unpin under the commit_root_sem, | |
190 | * at the same time we swap out the commit roots. | |
191 | * | |
192 | * This is because we must have a real view of the last spot the caching | |
193 | * kthreads were while caching. Consider the following views of the | |
194 | * extent tree for a block group | |
195 | * | |
196 | * commit root | |
197 | * +----+----+----+----+----+----+----+ | |
198 | * |\\\\| |\\\\|\\\\| |\\\\|\\\\| | |
199 | * +----+----+----+----+----+----+----+ | |
200 | * 0 1 2 3 4 5 6 7 | |
201 | * | |
202 | * new commit root | |
203 | * +----+----+----+----+----+----+----+ | |
204 | * | | | |\\\\| | |\\\\| | |
205 | * +----+----+----+----+----+----+----+ | |
206 | * 0 1 2 3 4 5 6 7 | |
207 | * | |
208 | * If the cache_ctl->progress was at 3, then we are only allowed to | |
209 | * unpin [0,1) and [2,3], because the caching thread has already | |
210 | * processed those extents. We are not allowed to unpin [5,6), because | |
211 | * the caching thread will re-start it's search from 3, and thus find | |
212 | * the hole from [4,6) to add to the free space cache. | |
213 | */ | |
bbb86a37 | 214 | spin_lock(&fs_info->block_group_cache_lock); |
27d56e62 JB |
215 | list_for_each_entry_safe(caching_ctl, next, |
216 | &fs_info->caching_block_groups, list) { | |
217 | struct btrfs_block_group *cache = caching_ctl->block_group; | |
218 | ||
219 | if (btrfs_block_group_done(cache)) { | |
220 | cache->last_byte_to_unpin = (u64)-1; | |
221 | list_del_init(&caching_ctl->list); | |
222 | btrfs_put_caching_control(caching_ctl); | |
223 | } else { | |
224 | cache->last_byte_to_unpin = caching_ctl->progress; | |
225 | } | |
226 | } | |
bbb86a37 | 227 | spin_unlock(&fs_info->block_group_cache_lock); |
9e351cc8 | 228 | up_write(&fs_info->commit_root_sem); |
817d52f8 JB |
229 | } |
230 | ||
0860adfd MX |
231 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
232 | unsigned int type) | |
233 | { | |
234 | if (type & TRANS_EXTWRITERS) | |
235 | atomic_inc(&trans->num_extwriters); | |
236 | } | |
237 | ||
238 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
239 | unsigned int type) | |
240 | { | |
241 | if (type & TRANS_EXTWRITERS) | |
242 | atomic_dec(&trans->num_extwriters); | |
243 | } | |
244 | ||
245 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
246 | unsigned int type) | |
247 | { | |
248 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
249 | } | |
250 | ||
251 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
252 | { | |
253 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
254 | } |
255 | ||
fb6dea26 JB |
256 | /* |
257 | * To be called after all the new block groups attached to the transaction | |
258 | * handle have been created (btrfs_create_pending_block_groups()). | |
259 | */ | |
260 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
261 | { | |
262 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
eafa4fd0 | 263 | struct btrfs_transaction *cur_trans = trans->transaction; |
fb6dea26 JB |
264 | |
265 | if (!trans->chunk_bytes_reserved) | |
266 | return; | |
267 | ||
268 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
269 | ||
270 | btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, | |
63f018be | 271 | trans->chunk_bytes_reserved, NULL); |
eafa4fd0 FM |
272 | atomic64_sub(trans->chunk_bytes_reserved, &cur_trans->chunk_bytes_reserved); |
273 | cond_wake_up(&cur_trans->chunk_reserve_wait); | |
fb6dea26 JB |
274 | trans->chunk_bytes_reserved = 0; |
275 | } | |
276 | ||
d352ac68 CM |
277 | /* |
278 | * either allocate a new transaction or hop into the existing one | |
279 | */ | |
2ff7e61e JM |
280 | static noinline int join_transaction(struct btrfs_fs_info *fs_info, |
281 | unsigned int type) | |
79154b1b CM |
282 | { |
283 | struct btrfs_transaction *cur_trans; | |
a4abeea4 | 284 | |
19ae4e81 | 285 | spin_lock(&fs_info->trans_lock); |
d43317dc | 286 | loop: |
49b25e05 | 287 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 288 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 289 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
290 | return -EROFS; |
291 | } | |
292 | ||
19ae4e81 | 293 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 294 | if (cur_trans) { |
bf31f87f | 295 | if (TRANS_ABORTED(cur_trans)) { |
19ae4e81 | 296 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 297 | return cur_trans->aborted; |
871383be | 298 | } |
4a9d8bde | 299 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
300 | spin_unlock(&fs_info->trans_lock); |
301 | return -EBUSY; | |
302 | } | |
9b64f57d | 303 | refcount_inc(&cur_trans->use_count); |
13c5a93e | 304 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 305 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 306 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 307 | return 0; |
79154b1b | 308 | } |
19ae4e81 | 309 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 310 | |
354aa0fb MX |
311 | /* |
312 | * If we are ATTACH, we just want to catch the current transaction, | |
313 | * and commit it. If there is no transaction, just return ENOENT. | |
314 | */ | |
315 | if (type == TRANS_ATTACH) | |
316 | return -ENOENT; | |
317 | ||
4a9d8bde MX |
318 | /* |
319 | * JOIN_NOLOCK only happens during the transaction commit, so | |
320 | * it is impossible that ->running_transaction is NULL | |
321 | */ | |
322 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
323 | ||
4b5faeac | 324 | cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); |
a4abeea4 JB |
325 | if (!cur_trans) |
326 | return -ENOMEM; | |
d43317dc | 327 | |
19ae4e81 JS |
328 | spin_lock(&fs_info->trans_lock); |
329 | if (fs_info->running_transaction) { | |
d43317dc CM |
330 | /* |
331 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 332 | * to redo the checks above |
d43317dc | 333 | */ |
4b5faeac | 334 | kfree(cur_trans); |
d43317dc | 335 | goto loop; |
87533c47 | 336 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 337 | spin_unlock(&fs_info->trans_lock); |
4b5faeac | 338 | kfree(cur_trans); |
7b8b92af | 339 | return -EROFS; |
79154b1b | 340 | } |
d43317dc | 341 | |
ab8d0fc4 | 342 | cur_trans->fs_info = fs_info; |
48778179 FM |
343 | atomic_set(&cur_trans->pending_ordered, 0); |
344 | init_waitqueue_head(&cur_trans->pending_wait); | |
a4abeea4 | 345 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 346 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
347 | init_waitqueue_head(&cur_trans->writer_wait); |
348 | init_waitqueue_head(&cur_trans->commit_wait); | |
4a9d8bde | 349 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
350 | /* |
351 | * One for this trans handle, one so it will live on until we | |
352 | * commit the transaction. | |
353 | */ | |
9b64f57d | 354 | refcount_set(&cur_trans->use_count, 2); |
3204d33c | 355 | cur_trans->flags = 0; |
afd48513 | 356 | cur_trans->start_time = ktime_get_seconds(); |
a4abeea4 | 357 | |
a099d0fd AM |
358 | memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); |
359 | ||
5c9d028b | 360 | cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; |
3368d001 | 361 | cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; |
d7df2c79 | 362 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
20b297d6 JS |
363 | |
364 | /* | |
365 | * although the tree mod log is per file system and not per transaction, | |
366 | * the log must never go across transaction boundaries. | |
367 | */ | |
368 | smp_mb(); | |
31b1a2bd | 369 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
5d163e0e | 370 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); |
31b1a2bd | 371 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
5d163e0e | 372 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); |
fc36ed7e | 373 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 374 | |
a4abeea4 JB |
375 | spin_lock_init(&cur_trans->delayed_refs.lock); |
376 | ||
377 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
bbbf7243 | 378 | INIT_LIST_HEAD(&cur_trans->dev_update_list); |
9e351cc8 | 379 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
ce93ec54 | 380 | INIT_LIST_HEAD(&cur_trans->dirty_bgs); |
1bbc621e | 381 | INIT_LIST_HEAD(&cur_trans->io_bgs); |
2b9dbef2 | 382 | INIT_LIST_HEAD(&cur_trans->dropped_roots); |
1bbc621e | 383 | mutex_init(&cur_trans->cache_write_mutex); |
ce93ec54 | 384 | spin_lock_init(&cur_trans->dirty_bgs_lock); |
e33e17ee | 385 | INIT_LIST_HEAD(&cur_trans->deleted_bgs); |
2b9dbef2 | 386 | spin_lock_init(&cur_trans->dropped_roots_lock); |
d3575156 NA |
387 | INIT_LIST_HEAD(&cur_trans->releasing_ebs); |
388 | spin_lock_init(&cur_trans->releasing_ebs_lock); | |
eafa4fd0 FM |
389 | atomic64_set(&cur_trans->chunk_bytes_reserved, 0); |
390 | init_waitqueue_head(&cur_trans->chunk_reserve_wait); | |
19ae4e81 | 391 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
c258d6e3 | 392 | extent_io_tree_init(fs_info, &cur_trans->dirty_pages, |
43eb5f29 | 393 | IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode); |
fe119a6e NB |
394 | extent_io_tree_init(fs_info, &cur_trans->pinned_extents, |
395 | IO_TREE_FS_PINNED_EXTENTS, NULL); | |
19ae4e81 JS |
396 | fs_info->generation++; |
397 | cur_trans->transid = fs_info->generation; | |
398 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 399 | cur_trans->aborted = 0; |
19ae4e81 | 400 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 401 | |
79154b1b CM |
402 | return 0; |
403 | } | |
404 | ||
d352ac68 | 405 | /* |
92a7cc42 QW |
406 | * This does all the record keeping required to make sure that a shareable root |
407 | * is properly recorded in a given transaction. This is required to make sure | |
408 | * the old root from before we joined the transaction is deleted when the | |
409 | * transaction commits. | |
d352ac68 | 410 | */ |
7585717f | 411 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
6426c7ad QW |
412 | struct btrfs_root *root, |
413 | int force) | |
6702ed49 | 414 | { |
0b246afa JM |
415 | struct btrfs_fs_info *fs_info = root->fs_info; |
416 | ||
92a7cc42 | 417 | if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
6426c7ad | 418 | root->last_trans < trans->transid) || force) { |
0b246afa | 419 | WARN_ON(root == fs_info->extent_root); |
4d31778a | 420 | WARN_ON(!force && root->commit_root != root->node); |
5d4f98a2 | 421 | |
7585717f | 422 | /* |
27cdeb70 | 423 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
424 | * we have the reloc mutex held now, so there |
425 | * is only one writer in this function | |
426 | */ | |
27cdeb70 | 427 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 428 | |
27cdeb70 | 429 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
430 | * they find our root->last_trans update |
431 | */ | |
432 | smp_wmb(); | |
433 | ||
0b246afa | 434 | spin_lock(&fs_info->fs_roots_radix_lock); |
6426c7ad | 435 | if (root->last_trans == trans->transid && !force) { |
0b246afa | 436 | spin_unlock(&fs_info->fs_roots_radix_lock); |
a4abeea4 JB |
437 | return 0; |
438 | } | |
0b246afa JM |
439 | radix_tree_tag_set(&fs_info->fs_roots_radix, |
440 | (unsigned long)root->root_key.objectid, | |
441 | BTRFS_ROOT_TRANS_TAG); | |
442 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
7585717f CM |
443 | root->last_trans = trans->transid; |
444 | ||
445 | /* this is pretty tricky. We don't want to | |
446 | * take the relocation lock in btrfs_record_root_in_trans | |
447 | * unless we're really doing the first setup for this root in | |
448 | * this transaction. | |
449 | * | |
450 | * Normally we'd use root->last_trans as a flag to decide | |
451 | * if we want to take the expensive mutex. | |
452 | * | |
453 | * But, we have to set root->last_trans before we | |
454 | * init the relocation root, otherwise, we trip over warnings | |
455 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 456 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
457 | * fixing up the reloc trees and everyone must wait. |
458 | * | |
459 | * When this is zero, they can trust root->last_trans and fly | |
460 | * through btrfs_record_root_in_trans without having to take the | |
461 | * lock. smp_wmb() makes sure that all the writes above are | |
462 | * done before we pop in the zero below | |
463 | */ | |
5d4f98a2 | 464 | btrfs_init_reloc_root(trans, root); |
c7548af6 | 465 | smp_mb__before_atomic(); |
27cdeb70 | 466 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 YZ |
467 | } |
468 | return 0; | |
469 | } | |
bcc63abb | 470 | |
7585717f | 471 | |
2b9dbef2 JB |
472 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
473 | struct btrfs_root *root) | |
474 | { | |
0b246afa | 475 | struct btrfs_fs_info *fs_info = root->fs_info; |
2b9dbef2 JB |
476 | struct btrfs_transaction *cur_trans = trans->transaction; |
477 | ||
478 | /* Add ourselves to the transaction dropped list */ | |
479 | spin_lock(&cur_trans->dropped_roots_lock); | |
480 | list_add_tail(&root->root_list, &cur_trans->dropped_roots); | |
481 | spin_unlock(&cur_trans->dropped_roots_lock); | |
482 | ||
483 | /* Make sure we don't try to update the root at commit time */ | |
0b246afa JM |
484 | spin_lock(&fs_info->fs_roots_radix_lock); |
485 | radix_tree_tag_clear(&fs_info->fs_roots_radix, | |
2b9dbef2 JB |
486 | (unsigned long)root->root_key.objectid, |
487 | BTRFS_ROOT_TRANS_TAG); | |
0b246afa | 488 | spin_unlock(&fs_info->fs_roots_radix_lock); |
2b9dbef2 JB |
489 | } |
490 | ||
7585717f CM |
491 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
492 | struct btrfs_root *root) | |
493 | { | |
0b246afa JM |
494 | struct btrfs_fs_info *fs_info = root->fs_info; |
495 | ||
92a7cc42 | 496 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
7585717f CM |
497 | return 0; |
498 | ||
499 | /* | |
27cdeb70 | 500 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
501 | * and barriers |
502 | */ | |
503 | smp_rmb(); | |
504 | if (root->last_trans == trans->transid && | |
27cdeb70 | 505 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
506 | return 0; |
507 | ||
0b246afa | 508 | mutex_lock(&fs_info->reloc_mutex); |
6426c7ad | 509 | record_root_in_trans(trans, root, 0); |
0b246afa | 510 | mutex_unlock(&fs_info->reloc_mutex); |
7585717f CM |
511 | |
512 | return 0; | |
513 | } | |
514 | ||
4a9d8bde MX |
515 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
516 | { | |
3296bf56 | 517 | return (trans->state >= TRANS_STATE_COMMIT_START && |
501407aa | 518 | trans->state < TRANS_STATE_UNBLOCKED && |
bf31f87f | 519 | !TRANS_ABORTED(trans)); |
4a9d8bde MX |
520 | } |
521 | ||
d352ac68 CM |
522 | /* wait for commit against the current transaction to become unblocked |
523 | * when this is done, it is safe to start a new transaction, but the current | |
524 | * transaction might not be fully on disk. | |
525 | */ | |
2ff7e61e | 526 | static void wait_current_trans(struct btrfs_fs_info *fs_info) |
79154b1b | 527 | { |
f9295749 | 528 | struct btrfs_transaction *cur_trans; |
79154b1b | 529 | |
0b246afa JM |
530 | spin_lock(&fs_info->trans_lock); |
531 | cur_trans = fs_info->running_transaction; | |
4a9d8bde | 532 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
9b64f57d | 533 | refcount_inc(&cur_trans->use_count); |
0b246afa | 534 | spin_unlock(&fs_info->trans_lock); |
72d63ed6 | 535 | |
0b246afa | 536 | wait_event(fs_info->transaction_wait, |
501407aa | 537 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
bf31f87f | 538 | TRANS_ABORTED(cur_trans)); |
724e2315 | 539 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 540 | } else { |
0b246afa | 541 | spin_unlock(&fs_info->trans_lock); |
f9295749 | 542 | } |
37d1aeee CM |
543 | } |
544 | ||
2ff7e61e | 545 | static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) |
a22285a6 | 546 | { |
0b246afa | 547 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
a4abeea4 JB |
548 | return 0; |
549 | ||
92e2f7e3 | 550 | if (type == TRANS_START) |
a22285a6 | 551 | return 1; |
a4abeea4 | 552 | |
a22285a6 YZ |
553 | return 0; |
554 | } | |
555 | ||
20dd2cbf MX |
556 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
557 | { | |
0b246afa JM |
558 | struct btrfs_fs_info *fs_info = root->fs_info; |
559 | ||
560 | if (!fs_info->reloc_ctl || | |
92a7cc42 | 561 | !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || |
20dd2cbf MX |
562 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
563 | root->reloc_root) | |
564 | return false; | |
565 | ||
566 | return true; | |
567 | } | |
568 | ||
08e007d2 | 569 | static struct btrfs_trans_handle * |
5aed1dd8 | 570 | start_transaction(struct btrfs_root *root, unsigned int num_items, |
003d7c59 JM |
571 | unsigned int type, enum btrfs_reserve_flush_enum flush, |
572 | bool enforce_qgroups) | |
37d1aeee | 573 | { |
0b246afa | 574 | struct btrfs_fs_info *fs_info = root->fs_info; |
ba2c4d4e | 575 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
a22285a6 YZ |
576 | struct btrfs_trans_handle *h; |
577 | struct btrfs_transaction *cur_trans; | |
b5009945 | 578 | u64 num_bytes = 0; |
c5567237 | 579 | u64 qgroup_reserved = 0; |
20dd2cbf | 580 | bool reloc_reserved = false; |
9c343784 | 581 | bool do_chunk_alloc = false; |
20dd2cbf | 582 | int ret; |
acce952b | 583 | |
46c4e71e | 584 | /* Send isn't supposed to start transactions. */ |
2755a0de | 585 | ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); |
46c4e71e | 586 | |
0b246afa | 587 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
acce952b | 588 | return ERR_PTR(-EROFS); |
2a1eb461 | 589 | |
46c4e71e | 590 | if (current->journal_info) { |
0860adfd | 591 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 | 592 | h = current->journal_info; |
b50fff81 DS |
593 | refcount_inc(&h->use_count); |
594 | WARN_ON(refcount_read(&h->use_count) > 2); | |
2a1eb461 JB |
595 | h->orig_rsv = h->block_rsv; |
596 | h->block_rsv = NULL; | |
597 | goto got_it; | |
598 | } | |
b5009945 JB |
599 | |
600 | /* | |
601 | * Do the reservation before we join the transaction so we can do all | |
602 | * the appropriate flushing if need be. | |
603 | */ | |
003d7c59 | 604 | if (num_items && root != fs_info->chunk_root) { |
ba2c4d4e JB |
605 | struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; |
606 | u64 delayed_refs_bytes = 0; | |
607 | ||
0b246afa | 608 | qgroup_reserved = num_items * fs_info->nodesize; |
733e03a0 QW |
609 | ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, |
610 | enforce_qgroups); | |
7174109c QW |
611 | if (ret) |
612 | return ERR_PTR(ret); | |
c5567237 | 613 | |
ba2c4d4e JB |
614 | /* |
615 | * We want to reserve all the bytes we may need all at once, so | |
616 | * we only do 1 enospc flushing cycle per transaction start. We | |
617 | * accomplish this by simply assuming we'll do 2 x num_items | |
618 | * worth of delayed refs updates in this trans handle, and | |
619 | * refill that amount for whatever is missing in the reserve. | |
620 | */ | |
2bd36e7b | 621 | num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); |
7f9fe614 JB |
622 | if (flush == BTRFS_RESERVE_FLUSH_ALL && |
623 | delayed_refs_rsv->full == 0) { | |
ba2c4d4e JB |
624 | delayed_refs_bytes = num_bytes; |
625 | num_bytes <<= 1; | |
626 | } | |
627 | ||
20dd2cbf MX |
628 | /* |
629 | * Do the reservation for the relocation root creation | |
630 | */ | |
ee39b432 | 631 | if (need_reserve_reloc_root(root)) { |
0b246afa | 632 | num_bytes += fs_info->nodesize; |
20dd2cbf MX |
633 | reloc_reserved = true; |
634 | } | |
635 | ||
ba2c4d4e JB |
636 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush); |
637 | if (ret) | |
638 | goto reserve_fail; | |
639 | if (delayed_refs_bytes) { | |
640 | btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, | |
641 | delayed_refs_bytes); | |
642 | num_bytes -= delayed_refs_bytes; | |
643 | } | |
9c343784 JB |
644 | |
645 | if (rsv->space_info->force_alloc) | |
646 | do_chunk_alloc = true; | |
ba2c4d4e JB |
647 | } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && |
648 | !delayed_refs_rsv->full) { | |
649 | /* | |
650 | * Some people call with btrfs_start_transaction(root, 0) | |
651 | * because they can be throttled, but have some other mechanism | |
652 | * for reserving space. We still want these guys to refill the | |
653 | * delayed block_rsv so just add 1 items worth of reservation | |
654 | * here. | |
655 | */ | |
656 | ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); | |
b5009945 | 657 | if (ret) |
843fcf35 | 658 | goto reserve_fail; |
b5009945 | 659 | } |
a22285a6 | 660 | again: |
f2f767e7 | 661 | h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); |
843fcf35 MX |
662 | if (!h) { |
663 | ret = -ENOMEM; | |
664 | goto alloc_fail; | |
665 | } | |
37d1aeee | 666 | |
98114659 JB |
667 | /* |
668 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
669 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
670 | * because we're already holding a ref. We need this because we could | |
671 | * have raced in and did an fsync() on a file which can kick a commit | |
672 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
673 | * |
674 | * If we are ATTACH, it means we just want to catch the current | |
675 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 676 | */ |
0860adfd | 677 | if (type & __TRANS_FREEZABLE) |
0b246afa | 678 | sb_start_intwrite(fs_info->sb); |
b2b5ef5c | 679 | |
2ff7e61e JM |
680 | if (may_wait_transaction(fs_info, type)) |
681 | wait_current_trans(fs_info); | |
a22285a6 | 682 | |
a4abeea4 | 683 | do { |
2ff7e61e | 684 | ret = join_transaction(fs_info, type); |
178260b2 | 685 | if (ret == -EBUSY) { |
2ff7e61e | 686 | wait_current_trans(fs_info); |
a6d155d2 FM |
687 | if (unlikely(type == TRANS_ATTACH || |
688 | type == TRANS_JOIN_NOSTART)) | |
178260b2 MX |
689 | ret = -ENOENT; |
690 | } | |
a4abeea4 JB |
691 | } while (ret == -EBUSY); |
692 | ||
a43f7f82 | 693 | if (ret < 0) |
843fcf35 | 694 | goto join_fail; |
0f7d52f4 | 695 | |
0b246afa | 696 | cur_trans = fs_info->running_transaction; |
a22285a6 YZ |
697 | |
698 | h->transid = cur_trans->transid; | |
699 | h->transaction = cur_trans; | |
d13603ef | 700 | h->root = root; |
b50fff81 | 701 | refcount_set(&h->use_count, 1); |
64b63580 | 702 | h->fs_info = root->fs_info; |
7174109c | 703 | |
a698d075 | 704 | h->type = type; |
d9a0540a | 705 | h->can_flush_pending_bgs = true; |
ea658bad | 706 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 707 | |
a22285a6 | 708 | smp_mb(); |
3296bf56 | 709 | if (cur_trans->state >= TRANS_STATE_COMMIT_START && |
2ff7e61e | 710 | may_wait_transaction(fs_info, type)) { |
abdd2e80 | 711 | current->journal_info = h; |
3a45bb20 | 712 | btrfs_commit_transaction(h); |
a22285a6 YZ |
713 | goto again; |
714 | } | |
715 | ||
b5009945 | 716 | if (num_bytes) { |
0b246afa | 717 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 718 | h->transid, num_bytes, 1); |
0b246afa | 719 | h->block_rsv = &fs_info->trans_block_rsv; |
b5009945 | 720 | h->bytes_reserved = num_bytes; |
20dd2cbf | 721 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 722 | } |
9ed74f2d | 723 | |
2a1eb461 | 724 | got_it: |
bcf3a3e7 | 725 | if (!current->journal_info) |
a22285a6 | 726 | current->journal_info = h; |
fcc99734 | 727 | |
9c343784 JB |
728 | /* |
729 | * If the space_info is marked ALLOC_FORCE then we'll get upgraded to | |
730 | * ALLOC_FORCE the first run through, and then we won't allocate for | |
731 | * anybody else who races in later. We don't care about the return | |
732 | * value here. | |
733 | */ | |
734 | if (do_chunk_alloc && num_bytes) { | |
735 | u64 flags = h->block_rsv->space_info->flags; | |
736 | ||
737 | btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), | |
738 | CHUNK_ALLOC_NO_FORCE); | |
739 | } | |
740 | ||
fcc99734 QW |
741 | /* |
742 | * btrfs_record_root_in_trans() needs to alloc new extents, and may | |
743 | * call btrfs_join_transaction() while we're also starting a | |
744 | * transaction. | |
745 | * | |
746 | * Thus it need to be called after current->journal_info initialized, | |
747 | * or we can deadlock. | |
748 | */ | |
68075ea8 JB |
749 | ret = btrfs_record_root_in_trans(h, root); |
750 | if (ret) { | |
751 | /* | |
752 | * The transaction handle is fully initialized and linked with | |
753 | * other structures so it needs to be ended in case of errors, | |
754 | * not just freed. | |
755 | */ | |
756 | btrfs_end_transaction(h); | |
757 | return ERR_PTR(ret); | |
758 | } | |
fcc99734 | 759 | |
79154b1b | 760 | return h; |
843fcf35 MX |
761 | |
762 | join_fail: | |
0860adfd | 763 | if (type & __TRANS_FREEZABLE) |
0b246afa | 764 | sb_end_intwrite(fs_info->sb); |
843fcf35 MX |
765 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
766 | alloc_fail: | |
767 | if (num_bytes) | |
2ff7e61e | 768 | btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, |
63f018be | 769 | num_bytes, NULL); |
843fcf35 | 770 | reserve_fail: |
733e03a0 | 771 | btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); |
843fcf35 | 772 | return ERR_PTR(ret); |
79154b1b CM |
773 | } |
774 | ||
f9295749 | 775 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
5aed1dd8 | 776 | unsigned int num_items) |
f9295749 | 777 | { |
08e007d2 | 778 | return start_transaction(root, num_items, TRANS_START, |
003d7c59 | 779 | BTRFS_RESERVE_FLUSH_ALL, true); |
f9295749 | 780 | } |
003d7c59 | 781 | |
8eab77ff FM |
782 | struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( |
783 | struct btrfs_root *root, | |
7f9fe614 | 784 | unsigned int num_items) |
8eab77ff | 785 | { |
7f9fe614 JB |
786 | return start_transaction(root, num_items, TRANS_START, |
787 | BTRFS_RESERVE_FLUSH_ALL_STEAL, false); | |
8eab77ff | 788 | } |
8407aa46 | 789 | |
7a7eaa40 | 790 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 791 | { |
003d7c59 JM |
792 | return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, |
793 | true); | |
f9295749 CM |
794 | } |
795 | ||
8d510121 | 796 | struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) |
0af3d00b | 797 | { |
575a75d6 | 798 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, |
003d7c59 | 799 | BTRFS_RESERVE_NO_FLUSH, true); |
0af3d00b JB |
800 | } |
801 | ||
a6d155d2 FM |
802 | /* |
803 | * Similar to regular join but it never starts a transaction when none is | |
804 | * running or after waiting for the current one to finish. | |
805 | */ | |
806 | struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) | |
807 | { | |
808 | return start_transaction(root, 0, TRANS_JOIN_NOSTART, | |
809 | BTRFS_RESERVE_NO_FLUSH, true); | |
810 | } | |
811 | ||
d4edf39b MX |
812 | /* |
813 | * btrfs_attach_transaction() - catch the running transaction | |
814 | * | |
815 | * It is used when we want to commit the current the transaction, but | |
816 | * don't want to start a new one. | |
817 | * | |
818 | * Note: If this function return -ENOENT, it just means there is no | |
819 | * running transaction. But it is possible that the inactive transaction | |
820 | * is still in the memory, not fully on disk. If you hope there is no | |
821 | * inactive transaction in the fs when -ENOENT is returned, you should | |
822 | * invoke | |
823 | * btrfs_attach_transaction_barrier() | |
824 | */ | |
354aa0fb | 825 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 826 | { |
575a75d6 | 827 | return start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 828 | BTRFS_RESERVE_NO_FLUSH, true); |
60376ce4 JB |
829 | } |
830 | ||
d4edf39b | 831 | /* |
90b6d283 | 832 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b | 833 | * |
52042d8e | 834 | * It is similar to the above function, the difference is this one |
d4edf39b MX |
835 | * will wait for all the inactive transactions until they fully |
836 | * complete. | |
837 | */ | |
838 | struct btrfs_trans_handle * | |
839 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
840 | { | |
841 | struct btrfs_trans_handle *trans; | |
842 | ||
575a75d6 | 843 | trans = start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 844 | BTRFS_RESERVE_NO_FLUSH, true); |
8d9e220c | 845 | if (trans == ERR_PTR(-ENOENT)) |
2ff7e61e | 846 | btrfs_wait_for_commit(root->fs_info, 0); |
d4edf39b MX |
847 | |
848 | return trans; | |
849 | } | |
850 | ||
d0c2f4fa FM |
851 | /* Wait for a transaction commit to reach at least the given state. */ |
852 | static noinline void wait_for_commit(struct btrfs_transaction *commit, | |
853 | const enum btrfs_trans_state min_state) | |
89ce8a63 | 854 | { |
d0c2f4fa | 855 | wait_event(commit->commit_wait, commit->state >= min_state); |
89ce8a63 CM |
856 | } |
857 | ||
2ff7e61e | 858 | int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) |
46204592 SW |
859 | { |
860 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 861 | int ret = 0; |
46204592 | 862 | |
46204592 | 863 | if (transid) { |
0b246afa | 864 | if (transid <= fs_info->last_trans_committed) |
a4abeea4 | 865 | goto out; |
46204592 SW |
866 | |
867 | /* find specified transaction */ | |
0b246afa JM |
868 | spin_lock(&fs_info->trans_lock); |
869 | list_for_each_entry(t, &fs_info->trans_list, list) { | |
46204592 SW |
870 | if (t->transid == transid) { |
871 | cur_trans = t; | |
9b64f57d | 872 | refcount_inc(&cur_trans->use_count); |
8cd2807f | 873 | ret = 0; |
46204592 SW |
874 | break; |
875 | } | |
8cd2807f MX |
876 | if (t->transid > transid) { |
877 | ret = 0; | |
46204592 | 878 | break; |
8cd2807f | 879 | } |
46204592 | 880 | } |
0b246afa | 881 | spin_unlock(&fs_info->trans_lock); |
42383020 SW |
882 | |
883 | /* | |
884 | * The specified transaction doesn't exist, or we | |
885 | * raced with btrfs_commit_transaction | |
886 | */ | |
887 | if (!cur_trans) { | |
0b246afa | 888 | if (transid > fs_info->last_trans_committed) |
42383020 | 889 | ret = -EINVAL; |
8cd2807f | 890 | goto out; |
42383020 | 891 | } |
46204592 SW |
892 | } else { |
893 | /* find newest transaction that is committing | committed */ | |
0b246afa JM |
894 | spin_lock(&fs_info->trans_lock); |
895 | list_for_each_entry_reverse(t, &fs_info->trans_list, | |
46204592 | 896 | list) { |
4a9d8bde MX |
897 | if (t->state >= TRANS_STATE_COMMIT_START) { |
898 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 899 | break; |
46204592 | 900 | cur_trans = t; |
9b64f57d | 901 | refcount_inc(&cur_trans->use_count); |
46204592 SW |
902 | break; |
903 | } | |
904 | } | |
0b246afa | 905 | spin_unlock(&fs_info->trans_lock); |
46204592 | 906 | if (!cur_trans) |
a4abeea4 | 907 | goto out; /* nothing committing|committed */ |
46204592 SW |
908 | } |
909 | ||
d0c2f4fa | 910 | wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); |
724e2315 | 911 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 912 | out: |
46204592 SW |
913 | return ret; |
914 | } | |
915 | ||
2ff7e61e | 916 | void btrfs_throttle(struct btrfs_fs_info *fs_info) |
37d1aeee | 917 | { |
92e2f7e3 | 918 | wait_current_trans(fs_info); |
37d1aeee CM |
919 | } |
920 | ||
8a8f4dea | 921 | static bool should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa | 922 | { |
2ff7e61e | 923 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0b246afa | 924 | |
64403612 | 925 | if (btrfs_check_space_for_delayed_refs(fs_info)) |
8a8f4dea | 926 | return true; |
36ba022a | 927 | |
2ff7e61e | 928 | return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); |
8929ecfa YZ |
929 | } |
930 | ||
a2633b6a | 931 | bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa YZ |
932 | { |
933 | struct btrfs_transaction *cur_trans = trans->transaction; | |
8929ecfa | 934 | |
3296bf56 | 935 | if (cur_trans->state >= TRANS_STATE_COMMIT_START || |
e19eb11f | 936 | test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) |
a2633b6a | 937 | return true; |
8929ecfa | 938 | |
2ff7e61e | 939 | return should_end_transaction(trans); |
8929ecfa YZ |
940 | } |
941 | ||
dc60c525 NB |
942 | static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) |
943 | ||
0e34693f | 944 | { |
dc60c525 NB |
945 | struct btrfs_fs_info *fs_info = trans->fs_info; |
946 | ||
0e34693f NB |
947 | if (!trans->block_rsv) { |
948 | ASSERT(!trans->bytes_reserved); | |
949 | return; | |
950 | } | |
951 | ||
952 | if (!trans->bytes_reserved) | |
953 | return; | |
954 | ||
955 | ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); | |
956 | trace_btrfs_space_reservation(fs_info, "transaction", | |
957 | trans->transid, trans->bytes_reserved, 0); | |
958 | btrfs_block_rsv_release(fs_info, trans->block_rsv, | |
63f018be | 959 | trans->bytes_reserved, NULL); |
0e34693f NB |
960 | trans->bytes_reserved = 0; |
961 | } | |
962 | ||
89ce8a63 | 963 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
3a45bb20 | 964 | int throttle) |
79154b1b | 965 | { |
3a45bb20 | 966 | struct btrfs_fs_info *info = trans->fs_info; |
8929ecfa | 967 | struct btrfs_transaction *cur_trans = trans->transaction; |
4edc2ca3 | 968 | int err = 0; |
c3e69d58 | 969 | |
b50fff81 DS |
970 | if (refcount_read(&trans->use_count) > 1) { |
971 | refcount_dec(&trans->use_count); | |
2a1eb461 JB |
972 | trans->block_rsv = trans->orig_rsv; |
973 | return 0; | |
974 | } | |
975 | ||
dc60c525 | 976 | btrfs_trans_release_metadata(trans); |
4c13d758 | 977 | trans->block_rsv = NULL; |
c5567237 | 978 | |
119e80df | 979 | btrfs_create_pending_block_groups(trans); |
ea658bad | 980 | |
4fbcdf66 FM |
981 | btrfs_trans_release_chunk_metadata(trans); |
982 | ||
0860adfd | 983 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 984 | sb_end_intwrite(info->sb); |
6df7881a | 985 | |
8929ecfa | 986 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
987 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
988 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 989 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 990 | |
093258e6 | 991 | cond_wake_up(&cur_trans->writer_wait); |
724e2315 | 992 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
993 | |
994 | if (current->journal_info == trans) | |
995 | current->journal_info = NULL; | |
ab78c84d | 996 | |
24bbcf04 | 997 | if (throttle) |
2ff7e61e | 998 | btrfs_run_delayed_iputs(info); |
24bbcf04 | 999 | |
bf31f87f | 1000 | if (TRANS_ABORTED(trans) || |
0b246afa | 1001 | test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) { |
4e121c06 | 1002 | wake_up_process(info->transaction_kthread); |
fbabd4a3 JB |
1003 | if (TRANS_ABORTED(trans)) |
1004 | err = trans->aborted; | |
1005 | else | |
1006 | err = -EROFS; | |
4e121c06 | 1007 | } |
49b25e05 | 1008 | |
4edc2ca3 DJ |
1009 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
1010 | return err; | |
79154b1b CM |
1011 | } |
1012 | ||
3a45bb20 | 1013 | int btrfs_end_transaction(struct btrfs_trans_handle *trans) |
89ce8a63 | 1014 | { |
3a45bb20 | 1015 | return __btrfs_end_transaction(trans, 0); |
89ce8a63 CM |
1016 | } |
1017 | ||
3a45bb20 | 1018 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) |
89ce8a63 | 1019 | { |
3a45bb20 | 1020 | return __btrfs_end_transaction(trans, 1); |
16cdcec7 MX |
1021 | } |
1022 | ||
d352ac68 CM |
1023 | /* |
1024 | * when btree blocks are allocated, they have some corresponding bits set for | |
1025 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 1026 | * those extents are sent to disk but does not wait on them |
d352ac68 | 1027 | */ |
2ff7e61e | 1028 | int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, |
8cef4e16 | 1029 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 1030 | { |
777e6bd7 | 1031 | int err = 0; |
7c4452b9 | 1032 | int werr = 0; |
0b246afa | 1033 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 1034 | struct extent_state *cached_state = NULL; |
777e6bd7 | 1035 | u64 start = 0; |
5f39d397 | 1036 | u64 end; |
7c4452b9 | 1037 | |
6300463b | 1038 | atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
1728366e | 1039 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 1040 | mark, &cached_state)) { |
663dfbb0 FM |
1041 | bool wait_writeback = false; |
1042 | ||
1043 | err = convert_extent_bit(dirty_pages, start, end, | |
1044 | EXTENT_NEED_WAIT, | |
210aa277 | 1045 | mark, &cached_state); |
663dfbb0 FM |
1046 | /* |
1047 | * convert_extent_bit can return -ENOMEM, which is most of the | |
1048 | * time a temporary error. So when it happens, ignore the error | |
1049 | * and wait for writeback of this range to finish - because we | |
1050 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
bf89d38f JM |
1051 | * to __btrfs_wait_marked_extents() would not know that |
1052 | * writeback for this range started and therefore wouldn't | |
1053 | * wait for it to finish - we don't want to commit a | |
1054 | * superblock that points to btree nodes/leafs for which | |
1055 | * writeback hasn't finished yet (and without errors). | |
663dfbb0 | 1056 | * We cleanup any entries left in the io tree when committing |
41e7acd3 | 1057 | * the transaction (through extent_io_tree_release()). |
663dfbb0 FM |
1058 | */ |
1059 | if (err == -ENOMEM) { | |
1060 | err = 0; | |
1061 | wait_writeback = true; | |
1062 | } | |
1063 | if (!err) | |
1064 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
1065 | if (err) |
1066 | werr = err; | |
663dfbb0 FM |
1067 | else if (wait_writeback) |
1068 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 1069 | free_extent_state(cached_state); |
663dfbb0 | 1070 | cached_state = NULL; |
1728366e JB |
1071 | cond_resched(); |
1072 | start = end + 1; | |
7c4452b9 | 1073 | } |
6300463b | 1074 | atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
690587d1 CM |
1075 | return werr; |
1076 | } | |
1077 | ||
1078 | /* | |
1079 | * when btree blocks are allocated, they have some corresponding bits set for | |
1080 | * them in one of two extent_io trees. This is used to make sure all of | |
1081 | * those extents are on disk for transaction or log commit. We wait | |
1082 | * on all the pages and clear them from the dirty pages state tree | |
1083 | */ | |
bf89d38f JM |
1084 | static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, |
1085 | struct extent_io_tree *dirty_pages) | |
690587d1 | 1086 | { |
690587d1 CM |
1087 | int err = 0; |
1088 | int werr = 0; | |
0b246afa | 1089 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 1090 | struct extent_state *cached_state = NULL; |
690587d1 CM |
1091 | u64 start = 0; |
1092 | u64 end; | |
777e6bd7 | 1093 | |
1728366e | 1094 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 1095 | EXTENT_NEED_WAIT, &cached_state)) { |
663dfbb0 FM |
1096 | /* |
1097 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
1098 | * When committing the transaction, we'll remove any entries | |
1099 | * left in the io tree. For a log commit, we don't remove them | |
1100 | * after committing the log because the tree can be accessed | |
1101 | * concurrently - we do it only at transaction commit time when | |
41e7acd3 | 1102 | * it's safe to do it (through extent_io_tree_release()). |
663dfbb0 FM |
1103 | */ |
1104 | err = clear_extent_bit(dirty_pages, start, end, | |
ae0f1625 | 1105 | EXTENT_NEED_WAIT, 0, 0, &cached_state); |
663dfbb0 FM |
1106 | if (err == -ENOMEM) |
1107 | err = 0; | |
1108 | if (!err) | |
1109 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
1110 | if (err) |
1111 | werr = err; | |
e38e2ed7 FM |
1112 | free_extent_state(cached_state); |
1113 | cached_state = NULL; | |
1728366e JB |
1114 | cond_resched(); |
1115 | start = end + 1; | |
777e6bd7 | 1116 | } |
7c4452b9 CM |
1117 | if (err) |
1118 | werr = err; | |
bf89d38f JM |
1119 | return werr; |
1120 | } | |
656f30db | 1121 | |
b9fae2eb | 1122 | static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, |
bf89d38f JM |
1123 | struct extent_io_tree *dirty_pages) |
1124 | { | |
1125 | bool errors = false; | |
1126 | int err; | |
656f30db | 1127 | |
bf89d38f JM |
1128 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); |
1129 | if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) | |
1130 | errors = true; | |
1131 | ||
1132 | if (errors && !err) | |
1133 | err = -EIO; | |
1134 | return err; | |
1135 | } | |
656f30db | 1136 | |
bf89d38f JM |
1137 | int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) |
1138 | { | |
1139 | struct btrfs_fs_info *fs_info = log_root->fs_info; | |
1140 | struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; | |
1141 | bool errors = false; | |
1142 | int err; | |
656f30db | 1143 | |
bf89d38f JM |
1144 | ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
1145 | ||
1146 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); | |
1147 | if ((mark & EXTENT_DIRTY) && | |
1148 | test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) | |
1149 | errors = true; | |
1150 | ||
1151 | if ((mark & EXTENT_NEW) && | |
1152 | test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) | |
1153 | errors = true; | |
1154 | ||
1155 | if (errors && !err) | |
1156 | err = -EIO; | |
1157 | return err; | |
79154b1b CM |
1158 | } |
1159 | ||
690587d1 | 1160 | /* |
c9b577c0 NB |
1161 | * When btree blocks are allocated the corresponding extents are marked dirty. |
1162 | * This function ensures such extents are persisted on disk for transaction or | |
1163 | * log commit. | |
1164 | * | |
1165 | * @trans: transaction whose dirty pages we'd like to write | |
690587d1 | 1166 | */ |
70458a58 | 1167 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) |
690587d1 CM |
1168 | { |
1169 | int ret; | |
1170 | int ret2; | |
c9b577c0 | 1171 | struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; |
70458a58 | 1172 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c6adc9cc | 1173 | struct blk_plug plug; |
690587d1 | 1174 | |
c6adc9cc | 1175 | blk_start_plug(&plug); |
c9b577c0 | 1176 | ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); |
c6adc9cc | 1177 | blk_finish_plug(&plug); |
bf89d38f | 1178 | ret2 = btrfs_wait_extents(fs_info, dirty_pages); |
bf0da8c1 | 1179 | |
41e7acd3 | 1180 | extent_io_tree_release(&trans->transaction->dirty_pages); |
c9b577c0 | 1181 | |
bf0da8c1 CM |
1182 | if (ret) |
1183 | return ret; | |
c9b577c0 | 1184 | else if (ret2) |
bf0da8c1 | 1185 | return ret2; |
c9b577c0 NB |
1186 | else |
1187 | return 0; | |
d0c803c4 CM |
1188 | } |
1189 | ||
d352ac68 CM |
1190 | /* |
1191 | * this is used to update the root pointer in the tree of tree roots. | |
1192 | * | |
1193 | * But, in the case of the extent allocation tree, updating the root | |
1194 | * pointer may allocate blocks which may change the root of the extent | |
1195 | * allocation tree. | |
1196 | * | |
1197 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1198 | * change while the root pointer was being updated in the metadata. | |
1199 | */ | |
0b86a832 CM |
1200 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1201 | struct btrfs_root *root) | |
79154b1b CM |
1202 | { |
1203 | int ret; | |
0b86a832 | 1204 | u64 old_root_bytenr; |
86b9f2ec | 1205 | u64 old_root_used; |
0b246afa JM |
1206 | struct btrfs_fs_info *fs_info = root->fs_info; |
1207 | struct btrfs_root *tree_root = fs_info->tree_root; | |
79154b1b | 1208 | |
86b9f2ec | 1209 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1210 | |
d397712b | 1211 | while (1) { |
0b86a832 | 1212 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1213 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1214 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1215 | break; |
87ef2bb4 | 1216 | |
5d4f98a2 | 1217 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1218 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1219 | &root->root_key, |
1220 | &root->root_item); | |
49b25e05 JM |
1221 | if (ret) |
1222 | return ret; | |
56bec294 | 1223 | |
86b9f2ec | 1224 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1225 | } |
276e680d | 1226 | |
0b86a832 CM |
1227 | return 0; |
1228 | } | |
1229 | ||
d352ac68 CM |
1230 | /* |
1231 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1232 | * |
1233 | * The error handling in this function may not be obvious. Any of the | |
1234 | * failures will cause the file system to go offline. We still need | |
1235 | * to clean up the delayed refs. | |
d352ac68 | 1236 | */ |
9386d8bc | 1237 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) |
0b86a832 | 1238 | { |
9386d8bc | 1239 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ea526d18 | 1240 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1241 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1242 | struct list_head *next; |
84234f3a | 1243 | struct extent_buffer *eb; |
56bec294 | 1244 | int ret; |
84234f3a YZ |
1245 | |
1246 | eb = btrfs_lock_root_node(fs_info->tree_root); | |
49b25e05 | 1247 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
9631e4cc | 1248 | 0, &eb, BTRFS_NESTING_COW); |
84234f3a YZ |
1249 | btrfs_tree_unlock(eb); |
1250 | free_extent_buffer(eb); | |
0b86a832 | 1251 | |
49b25e05 JM |
1252 | if (ret) |
1253 | return ret; | |
87ef2bb4 | 1254 | |
196c9d8d | 1255 | ret = btrfs_run_dev_stats(trans); |
c16ce190 JB |
1256 | if (ret) |
1257 | return ret; | |
2b584c68 | 1258 | ret = btrfs_run_dev_replace(trans); |
c16ce190 JB |
1259 | if (ret) |
1260 | return ret; | |
280f8bd2 | 1261 | ret = btrfs_run_qgroups(trans); |
c16ce190 JB |
1262 | if (ret) |
1263 | return ret; | |
546adb0d | 1264 | |
bbebb3e0 | 1265 | ret = btrfs_setup_space_cache(trans); |
dcdf7f6d JB |
1266 | if (ret) |
1267 | return ret; | |
1268 | ||
ea526d18 | 1269 | again: |
d397712b | 1270 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
2ff7e61e | 1271 | struct btrfs_root *root; |
0b86a832 CM |
1272 | next = fs_info->dirty_cowonly_roots.next; |
1273 | list_del_init(next); | |
1274 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1275 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1276 | |
9e351cc8 JB |
1277 | if (root != fs_info->extent_root) |
1278 | list_add_tail(&root->dirty_list, | |
1279 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1280 | ret = update_cowonly_root(trans, root); |
1281 | if (ret) | |
1282 | return ret; | |
79154b1b | 1283 | } |
276e680d | 1284 | |
488bc2a2 JB |
1285 | /* Now flush any delayed refs generated by updating all of the roots */ |
1286 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); | |
1287 | if (ret) | |
1288 | return ret; | |
1289 | ||
1bbc621e | 1290 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
5742d15f | 1291 | ret = btrfs_write_dirty_block_groups(trans); |
ea526d18 JB |
1292 | if (ret) |
1293 | return ret; | |
488bc2a2 JB |
1294 | |
1295 | /* | |
1296 | * We're writing the dirty block groups, which could generate | |
1297 | * delayed refs, which could generate more dirty block groups, | |
1298 | * so we want to keep this flushing in this loop to make sure | |
1299 | * everything gets run. | |
1300 | */ | |
c79a70b1 | 1301 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
ea526d18 JB |
1302 | if (ret) |
1303 | return ret; | |
1304 | } | |
1305 | ||
1306 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1307 | goto again; | |
1308 | ||
9e351cc8 JB |
1309 | list_add_tail(&fs_info->extent_root->dirty_list, |
1310 | &trans->transaction->switch_commits); | |
9f6cbcbb DS |
1311 | |
1312 | /* Update dev-replace pointer once everything is committed */ | |
1313 | fs_info->dev_replace.committed_cursor_left = | |
1314 | fs_info->dev_replace.cursor_left_last_write_of_item; | |
8dabb742 | 1315 | |
79154b1b CM |
1316 | return 0; |
1317 | } | |
1318 | ||
d352ac68 CM |
1319 | /* |
1320 | * dead roots are old snapshots that need to be deleted. This allocates | |
1321 | * a dirty root struct and adds it into the list of dead roots that need to | |
1322 | * be deleted | |
1323 | */ | |
cfad392b | 1324 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1325 | { |
0b246afa JM |
1326 | struct btrfs_fs_info *fs_info = root->fs_info; |
1327 | ||
1328 | spin_lock(&fs_info->trans_lock); | |
dc9492c1 JB |
1329 | if (list_empty(&root->root_list)) { |
1330 | btrfs_grab_root(root); | |
0b246afa | 1331 | list_add_tail(&root->root_list, &fs_info->dead_roots); |
dc9492c1 | 1332 | } |
0b246afa | 1333 | spin_unlock(&fs_info->trans_lock); |
5eda7b5e CM |
1334 | } |
1335 | ||
d352ac68 | 1336 | /* |
5d4f98a2 | 1337 | * update all the cowonly tree roots on disk |
d352ac68 | 1338 | */ |
7e4443d9 | 1339 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) |
0f7d52f4 | 1340 | { |
7e4443d9 | 1341 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0f7d52f4 | 1342 | struct btrfs_root *gang[8]; |
0f7d52f4 CM |
1343 | int i; |
1344 | int ret; | |
54aa1f4d | 1345 | |
a4abeea4 | 1346 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1347 | while (1) { |
5d4f98a2 YZ |
1348 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1349 | (void **)gang, 0, | |
0f7d52f4 CM |
1350 | ARRAY_SIZE(gang), |
1351 | BTRFS_ROOT_TRANS_TAG); | |
1352 | if (ret == 0) | |
1353 | break; | |
1354 | for (i = 0; i < ret; i++) { | |
5b4aacef | 1355 | struct btrfs_root *root = gang[i]; |
4f4317c1 JB |
1356 | int ret2; |
1357 | ||
5d4f98a2 YZ |
1358 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1359 | (unsigned long)root->root_key.objectid, | |
1360 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1361 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1362 | |
e02119d5 | 1363 | btrfs_free_log(trans, root); |
5d4f98a2 | 1364 | btrfs_update_reloc_root(trans, root); |
bcc63abb | 1365 | |
f1ebcc74 | 1366 | /* see comments in should_cow_block() */ |
27cdeb70 | 1367 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
c7548af6 | 1368 | smp_mb__after_atomic(); |
f1ebcc74 | 1369 | |
978d910d | 1370 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1371 | list_add_tail(&root->dirty_list, |
1372 | &trans->transaction->switch_commits); | |
978d910d YZ |
1373 | btrfs_set_root_node(&root->root_item, |
1374 | root->node); | |
1375 | } | |
5d4f98a2 | 1376 | |
4f4317c1 | 1377 | ret2 = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1378 | &root->root_key, |
1379 | &root->root_item); | |
4f4317c1 JB |
1380 | if (ret2) |
1381 | return ret2; | |
a4abeea4 | 1382 | spin_lock(&fs_info->fs_roots_radix_lock); |
733e03a0 | 1383 | btrfs_qgroup_free_meta_all_pertrans(root); |
0f7d52f4 CM |
1384 | } |
1385 | } | |
a4abeea4 | 1386 | spin_unlock(&fs_info->fs_roots_radix_lock); |
4f4317c1 | 1387 | return 0; |
0f7d52f4 CM |
1388 | } |
1389 | ||
d352ac68 | 1390 | /* |
de78b51a ES |
1391 | * defrag a given btree. |
1392 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1393 | */ |
de78b51a | 1394 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1395 | { |
1396 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1397 | struct btrfs_trans_handle *trans; |
8929ecfa | 1398 | int ret; |
e9d0b13b | 1399 | |
27cdeb70 | 1400 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1401 | return 0; |
8929ecfa | 1402 | |
6b80053d | 1403 | while (1) { |
8929ecfa YZ |
1404 | trans = btrfs_start_transaction(root, 0); |
1405 | if (IS_ERR(trans)) | |
1406 | return PTR_ERR(trans); | |
1407 | ||
de78b51a | 1408 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1409 | |
3a45bb20 | 1410 | btrfs_end_transaction(trans); |
2ff7e61e | 1411 | btrfs_btree_balance_dirty(info); |
e9d0b13b CM |
1412 | cond_resched(); |
1413 | ||
ab8d0fc4 | 1414 | if (btrfs_fs_closing(info) || ret != -EAGAIN) |
e9d0b13b | 1415 | break; |
210549eb | 1416 | |
ab8d0fc4 JM |
1417 | if (btrfs_defrag_cancelled(info)) { |
1418 | btrfs_debug(info, "defrag_root cancelled"); | |
210549eb DS |
1419 | ret = -EAGAIN; |
1420 | break; | |
1421 | } | |
e9d0b13b | 1422 | } |
27cdeb70 | 1423 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1424 | return ret; |
e9d0b13b CM |
1425 | } |
1426 | ||
6426c7ad QW |
1427 | /* |
1428 | * Do all special snapshot related qgroup dirty hack. | |
1429 | * | |
1430 | * Will do all needed qgroup inherit and dirty hack like switch commit | |
1431 | * roots inside one transaction and write all btree into disk, to make | |
1432 | * qgroup works. | |
1433 | */ | |
1434 | static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, | |
1435 | struct btrfs_root *src, | |
1436 | struct btrfs_root *parent, | |
1437 | struct btrfs_qgroup_inherit *inherit, | |
1438 | u64 dst_objectid) | |
1439 | { | |
1440 | struct btrfs_fs_info *fs_info = src->fs_info; | |
1441 | int ret; | |
1442 | ||
1443 | /* | |
1444 | * Save some performance in the case that qgroups are not | |
1445 | * enabled. If this check races with the ioctl, rescan will | |
1446 | * kick in anyway. | |
1447 | */ | |
9ea6e2b5 | 1448 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
6426c7ad | 1449 | return 0; |
6426c7ad | 1450 | |
4d31778a | 1451 | /* |
52042d8e | 1452 | * Ensure dirty @src will be committed. Or, after coming |
4d31778a QW |
1453 | * commit_fs_roots() and switch_commit_roots(), any dirty but not |
1454 | * recorded root will never be updated again, causing an outdated root | |
1455 | * item. | |
1456 | */ | |
1c442d22 JB |
1457 | ret = record_root_in_trans(trans, src, 1); |
1458 | if (ret) | |
1459 | return ret; | |
4d31778a | 1460 | |
2a4d84c1 JB |
1461 | /* |
1462 | * btrfs_qgroup_inherit relies on a consistent view of the usage for the | |
1463 | * src root, so we must run the delayed refs here. | |
1464 | * | |
1465 | * However this isn't particularly fool proof, because there's no | |
1466 | * synchronization keeping us from changing the tree after this point | |
1467 | * before we do the qgroup_inherit, or even from making changes while | |
1468 | * we're doing the qgroup_inherit. But that's a problem for the future, | |
1469 | * for now flush the delayed refs to narrow the race window where the | |
1470 | * qgroup counters could end up wrong. | |
1471 | */ | |
1472 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); | |
1473 | if (ret) { | |
1474 | btrfs_abort_transaction(trans, ret); | |
1475 | goto out; | |
1476 | } | |
1477 | ||
6426c7ad QW |
1478 | /* |
1479 | * We are going to commit transaction, see btrfs_commit_transaction() | |
1480 | * comment for reason locking tree_log_mutex | |
1481 | */ | |
1482 | mutex_lock(&fs_info->tree_log_mutex); | |
1483 | ||
7e4443d9 | 1484 | ret = commit_fs_roots(trans); |
6426c7ad QW |
1485 | if (ret) |
1486 | goto out; | |
460fb20a | 1487 | ret = btrfs_qgroup_account_extents(trans); |
6426c7ad QW |
1488 | if (ret < 0) |
1489 | goto out; | |
1490 | ||
1491 | /* Now qgroup are all updated, we can inherit it to new qgroups */ | |
a9377422 | 1492 | ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, |
6426c7ad QW |
1493 | inherit); |
1494 | if (ret < 0) | |
1495 | goto out; | |
1496 | ||
1497 | /* | |
1498 | * Now we do a simplified commit transaction, which will: | |
1499 | * 1) commit all subvolume and extent tree | |
1500 | * To ensure all subvolume and extent tree have a valid | |
1501 | * commit_root to accounting later insert_dir_item() | |
1502 | * 2) write all btree blocks onto disk | |
1503 | * This is to make sure later btree modification will be cowed | |
1504 | * Or commit_root can be populated and cause wrong qgroup numbers | |
1505 | * In this simplified commit, we don't really care about other trees | |
1506 | * like chunk and root tree, as they won't affect qgroup. | |
1507 | * And we don't write super to avoid half committed status. | |
1508 | */ | |
9386d8bc | 1509 | ret = commit_cowonly_roots(trans); |
6426c7ad QW |
1510 | if (ret) |
1511 | goto out; | |
889bfa39 | 1512 | switch_commit_roots(trans); |
70458a58 | 1513 | ret = btrfs_write_and_wait_transaction(trans); |
6426c7ad | 1514 | if (ret) |
f7af3934 | 1515 | btrfs_handle_fs_error(fs_info, ret, |
6426c7ad QW |
1516 | "Error while writing out transaction for qgroup"); |
1517 | ||
1518 | out: | |
1519 | mutex_unlock(&fs_info->tree_log_mutex); | |
1520 | ||
1521 | /* | |
1522 | * Force parent root to be updated, as we recorded it before so its | |
1523 | * last_trans == cur_transid. | |
1524 | * Or it won't be committed again onto disk after later | |
1525 | * insert_dir_item() | |
1526 | */ | |
1527 | if (!ret) | |
1c442d22 | 1528 | ret = record_root_in_trans(trans, parent, 1); |
6426c7ad QW |
1529 | return ret; |
1530 | } | |
1531 | ||
d352ac68 CM |
1532 | /* |
1533 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1534 | * transaction commit. This does the actual creation. |
1535 | * | |
1536 | * Note: | |
1537 | * If the error which may affect the commitment of the current transaction | |
1538 | * happens, we should return the error number. If the error which just affect | |
1539 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1540 | */ |
80b6794d | 1541 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1542 | struct btrfs_pending_snapshot *pending) |
1543 | { | |
08d50ca3 NB |
1544 | |
1545 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3063d29f | 1546 | struct btrfs_key key; |
80b6794d | 1547 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1548 | struct btrfs_root *tree_root = fs_info->tree_root; |
1549 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1550 | struct btrfs_root *parent_root; |
98c9942a | 1551 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1552 | struct inode *parent_inode; |
42874b3d MX |
1553 | struct btrfs_path *path; |
1554 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1555 | struct dentry *dentry; |
3063d29f | 1556 | struct extent_buffer *tmp; |
925baedd | 1557 | struct extent_buffer *old; |
95582b00 | 1558 | struct timespec64 cur_time; |
aec8030a | 1559 | int ret = 0; |
d68fc57b | 1560 | u64 to_reserve = 0; |
6bdb72de | 1561 | u64 index = 0; |
a22285a6 | 1562 | u64 objectid; |
b83cc969 | 1563 | u64 root_flags; |
3063d29f | 1564 | |
8546b570 DS |
1565 | ASSERT(pending->path); |
1566 | path = pending->path; | |
42874b3d | 1567 | |
b0c0ea63 DS |
1568 | ASSERT(pending->root_item); |
1569 | new_root_item = pending->root_item; | |
a22285a6 | 1570 | |
543068a2 | 1571 | pending->error = btrfs_get_free_objectid(tree_root, &objectid); |
aec8030a | 1572 | if (pending->error) |
6fa9700e | 1573 | goto no_free_objectid; |
3063d29f | 1574 | |
d6726335 QW |
1575 | /* |
1576 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1577 | * accounted by later btrfs_qgroup_inherit(). | |
1578 | */ | |
1579 | btrfs_set_skip_qgroup(trans, objectid); | |
1580 | ||
147d256e | 1581 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
d68fc57b YZ |
1582 | |
1583 | if (to_reserve > 0) { | |
aec8030a MX |
1584 | pending->error = btrfs_block_rsv_add(root, |
1585 | &pending->block_rsv, | |
1586 | to_reserve, | |
1587 | BTRFS_RESERVE_NO_FLUSH); | |
1588 | if (pending->error) | |
d6726335 | 1589 | goto clear_skip_qgroup; |
d68fc57b YZ |
1590 | } |
1591 | ||
3063d29f | 1592 | key.objectid = objectid; |
a22285a6 YZ |
1593 | key.offset = (u64)-1; |
1594 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1595 | |
6fa9700e | 1596 | rsv = trans->block_rsv; |
a22285a6 | 1597 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1598 | trans->bytes_reserved = trans->block_rsv->reserved; |
0b246afa | 1599 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa JB |
1600 | trans->transid, |
1601 | trans->bytes_reserved, 1); | |
a22285a6 | 1602 | dentry = pending->dentry; |
e9662f70 | 1603 | parent_inode = pending->dir; |
a22285a6 | 1604 | parent_root = BTRFS_I(parent_inode)->root; |
6426c7ad | 1605 | record_root_in_trans(trans, parent_root, 0); |
a22285a6 | 1606 | |
c2050a45 | 1607 | cur_time = current_time(parent_inode); |
04b285f3 | 1608 | |
3063d29f CM |
1609 | /* |
1610 | * insert the directory item | |
1611 | */ | |
877574e2 | 1612 | ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); |
49b25e05 | 1613 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1614 | |
1615 | /* check if there is a file/dir which has the same name. */ | |
1616 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
4a0cc7ca | 1617 | btrfs_ino(BTRFS_I(parent_inode)), |
42874b3d MX |
1618 | dentry->d_name.name, |
1619 | dentry->d_name.len, 0); | |
1620 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1621 | pending->error = -EEXIST; |
aec8030a | 1622 | goto dir_item_existed; |
42874b3d MX |
1623 | } else if (IS_ERR(dir_item)) { |
1624 | ret = PTR_ERR(dir_item); | |
66642832 | 1625 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1626 | goto fail; |
79787eaa | 1627 | } |
42874b3d | 1628 | btrfs_release_path(path); |
52c26179 | 1629 | |
e999376f CM |
1630 | /* |
1631 | * pull in the delayed directory update | |
1632 | * and the delayed inode item | |
1633 | * otherwise we corrupt the FS during | |
1634 | * snapshot | |
1635 | */ | |
e5c304e6 | 1636 | ret = btrfs_run_delayed_items(trans); |
8732d44f | 1637 | if (ret) { /* Transaction aborted */ |
66642832 | 1638 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1639 | goto fail; |
1640 | } | |
e999376f | 1641 | |
6426c7ad | 1642 | record_root_in_trans(trans, root, 0); |
6bdb72de SW |
1643 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1644 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1645 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1646 | |
b83cc969 LZ |
1647 | root_flags = btrfs_root_flags(new_root_item); |
1648 | if (pending->readonly) | |
1649 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1650 | else | |
1651 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1652 | btrfs_set_root_flags(new_root_item, root_flags); | |
1653 | ||
8ea05e3a AB |
1654 | btrfs_set_root_generation_v2(new_root_item, |
1655 | trans->transid); | |
807fc790 | 1656 | generate_random_guid(new_root_item->uuid); |
8ea05e3a AB |
1657 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, |
1658 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1659 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1660 | memset(new_root_item->received_uuid, 0, | |
1661 | sizeof(new_root_item->received_uuid)); | |
1662 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1663 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1664 | btrfs_set_root_stransid(new_root_item, 0); | |
1665 | btrfs_set_root_rtransid(new_root_item, 0); | |
1666 | } | |
3cae210f QW |
1667 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1668 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1669 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1670 | |
6bdb72de | 1671 | old = btrfs_lock_root_node(root); |
9631e4cc JB |
1672 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, |
1673 | BTRFS_NESTING_COW); | |
79787eaa JM |
1674 | if (ret) { |
1675 | btrfs_tree_unlock(old); | |
1676 | free_extent_buffer(old); | |
66642832 | 1677 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1678 | goto fail; |
79787eaa | 1679 | } |
49b25e05 | 1680 | |
49b25e05 | 1681 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1682 | /* clean up in any case */ |
6bdb72de SW |
1683 | btrfs_tree_unlock(old); |
1684 | free_extent_buffer(old); | |
8732d44f | 1685 | if (ret) { |
66642832 | 1686 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1687 | goto fail; |
1688 | } | |
f1ebcc74 | 1689 | /* see comments in should_cow_block() */ |
27cdeb70 | 1690 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1691 | smp_wmb(); |
1692 | ||
6bdb72de | 1693 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1694 | /* record when the snapshot was created in key.offset */ |
1695 | key.offset = trans->transid; | |
1696 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1697 | btrfs_tree_unlock(tmp); |
1698 | free_extent_buffer(tmp); | |
8732d44f | 1699 | if (ret) { |
66642832 | 1700 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1701 | goto fail; |
1702 | } | |
6bdb72de | 1703 | |
a22285a6 YZ |
1704 | /* |
1705 | * insert root back/forward references | |
1706 | */ | |
6025c19f | 1707 | ret = btrfs_add_root_ref(trans, objectid, |
0660b5af | 1708 | parent_root->root_key.objectid, |
4a0cc7ca | 1709 | btrfs_ino(BTRFS_I(parent_inode)), index, |
a22285a6 | 1710 | dentry->d_name.name, dentry->d_name.len); |
8732d44f | 1711 | if (ret) { |
66642832 | 1712 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1713 | goto fail; |
1714 | } | |
0660b5af | 1715 | |
a22285a6 | 1716 | key.offset = (u64)-1; |
2dfb1e43 | 1717 | pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); |
79787eaa JM |
1718 | if (IS_ERR(pending->snap)) { |
1719 | ret = PTR_ERR(pending->snap); | |
2d892ccd | 1720 | pending->snap = NULL; |
66642832 | 1721 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1722 | goto fail; |
79787eaa | 1723 | } |
d68fc57b | 1724 | |
49b25e05 | 1725 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f | 1726 | if (ret) { |
66642832 | 1727 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1728 | goto fail; |
1729 | } | |
361048f5 | 1730 | |
6426c7ad QW |
1731 | /* |
1732 | * Do special qgroup accounting for snapshot, as we do some qgroup | |
1733 | * snapshot hack to do fast snapshot. | |
1734 | * To co-operate with that hack, we do hack again. | |
1735 | * Or snapshot will be greatly slowed down by a subtree qgroup rescan | |
1736 | */ | |
1737 | ret = qgroup_account_snapshot(trans, root, parent_root, | |
1738 | pending->inherit, objectid); | |
1739 | if (ret < 0) | |
1740 | goto fail; | |
1741 | ||
684572df LF |
1742 | ret = btrfs_insert_dir_item(trans, dentry->d_name.name, |
1743 | dentry->d_name.len, BTRFS_I(parent_inode), | |
1744 | &key, BTRFS_FT_DIR, index); | |
42874b3d | 1745 | /* We have check then name at the beginning, so it is impossible. */ |
9c52057c | 1746 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f | 1747 | if (ret) { |
66642832 | 1748 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1749 | goto fail; |
1750 | } | |
42874b3d | 1751 | |
6ef06d27 | 1752 | btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + |
42874b3d | 1753 | dentry->d_name.len * 2); |
04b285f3 | 1754 | parent_inode->i_mtime = parent_inode->i_ctime = |
c2050a45 | 1755 | current_time(parent_inode); |
729f7961 | 1756 | ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); |
dd5f9615 | 1757 | if (ret) { |
66642832 | 1758 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1759 | goto fail; |
1760 | } | |
807fc790 AS |
1761 | ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, |
1762 | BTRFS_UUID_KEY_SUBVOL, | |
cdb345a8 | 1763 | objectid); |
dd5f9615 | 1764 | if (ret) { |
66642832 | 1765 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1766 | goto fail; |
1767 | } | |
1768 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
cdb345a8 | 1769 | ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, |
dd5f9615 SB |
1770 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, |
1771 | objectid); | |
1772 | if (ret && ret != -EEXIST) { | |
66642832 | 1773 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1774 | goto fail; |
1775 | } | |
1776 | } | |
d6726335 | 1777 | |
3063d29f | 1778 | fail: |
aec8030a MX |
1779 | pending->error = ret; |
1780 | dir_item_existed: | |
98c9942a | 1781 | trans->block_rsv = rsv; |
2382c5cc | 1782 | trans->bytes_reserved = 0; |
d6726335 QW |
1783 | clear_skip_qgroup: |
1784 | btrfs_clear_skip_qgroup(trans); | |
6fa9700e MX |
1785 | no_free_objectid: |
1786 | kfree(new_root_item); | |
b0c0ea63 | 1787 | pending->root_item = NULL; |
42874b3d | 1788 | btrfs_free_path(path); |
8546b570 DS |
1789 | pending->path = NULL; |
1790 | ||
49b25e05 | 1791 | return ret; |
3063d29f CM |
1792 | } |
1793 | ||
d352ac68 CM |
1794 | /* |
1795 | * create all the snapshots we've scheduled for creation | |
1796 | */ | |
08d50ca3 | 1797 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) |
3de4586c | 1798 | { |
aec8030a | 1799 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1800 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1801 | int ret = 0; |
3de4586c | 1802 | |
aec8030a MX |
1803 | list_for_each_entry_safe(pending, next, head, list) { |
1804 | list_del(&pending->list); | |
08d50ca3 | 1805 | ret = create_pending_snapshot(trans, pending); |
aec8030a MX |
1806 | if (ret) |
1807 | break; | |
1808 | } | |
1809 | return ret; | |
3de4586c CM |
1810 | } |
1811 | ||
2ff7e61e | 1812 | static void update_super_roots(struct btrfs_fs_info *fs_info) |
5d4f98a2 YZ |
1813 | { |
1814 | struct btrfs_root_item *root_item; | |
1815 | struct btrfs_super_block *super; | |
1816 | ||
0b246afa | 1817 | super = fs_info->super_copy; |
5d4f98a2 | 1818 | |
0b246afa | 1819 | root_item = &fs_info->chunk_root->root_item; |
093e037c DS |
1820 | super->chunk_root = root_item->bytenr; |
1821 | super->chunk_root_generation = root_item->generation; | |
1822 | super->chunk_root_level = root_item->level; | |
5d4f98a2 | 1823 | |
0b246afa | 1824 | root_item = &fs_info->tree_root->root_item; |
093e037c DS |
1825 | super->root = root_item->bytenr; |
1826 | super->generation = root_item->generation; | |
1827 | super->root_level = root_item->level; | |
0b246afa | 1828 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) |
093e037c | 1829 | super->cache_generation = root_item->generation; |
94846229 BB |
1830 | else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) |
1831 | super->cache_generation = 0; | |
0b246afa | 1832 | if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) |
093e037c | 1833 | super->uuid_tree_generation = root_item->generation; |
5d4f98a2 YZ |
1834 | } |
1835 | ||
f36f3042 CM |
1836 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1837 | { | |
4a9d8bde | 1838 | struct btrfs_transaction *trans; |
f36f3042 | 1839 | int ret = 0; |
4a9d8bde | 1840 | |
a4abeea4 | 1841 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1842 | trans = info->running_transaction; |
1843 | if (trans) | |
1844 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1845 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1846 | return ret; |
1847 | } | |
1848 | ||
8929ecfa YZ |
1849 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1850 | { | |
4a9d8bde | 1851 | struct btrfs_transaction *trans; |
8929ecfa | 1852 | int ret = 0; |
4a9d8bde | 1853 | |
a4abeea4 | 1854 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1855 | trans = info->running_transaction; |
1856 | if (trans) | |
1857 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1858 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1859 | return ret; |
1860 | } | |
1861 | ||
bb9c12c9 SW |
1862 | /* |
1863 | * wait for the current transaction commit to start and block subsequent | |
1864 | * transaction joins | |
1865 | */ | |
2ff7e61e | 1866 | static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info, |
bb9c12c9 SW |
1867 | struct btrfs_transaction *trans) |
1868 | { | |
2ff7e61e | 1869 | wait_event(fs_info->transaction_blocked_wait, |
bf31f87f DS |
1870 | trans->state >= TRANS_STATE_COMMIT_START || |
1871 | TRANS_ABORTED(trans)); | |
bb9c12c9 SW |
1872 | } |
1873 | ||
1874 | /* | |
1875 | * wait for the current transaction to start and then become unblocked. | |
1876 | * caller holds ref. | |
1877 | */ | |
2ff7e61e JM |
1878 | static void wait_current_trans_commit_start_and_unblock( |
1879 | struct btrfs_fs_info *fs_info, | |
1880 | struct btrfs_transaction *trans) | |
bb9c12c9 | 1881 | { |
2ff7e61e | 1882 | wait_event(fs_info->transaction_wait, |
bf31f87f DS |
1883 | trans->state >= TRANS_STATE_UNBLOCKED || |
1884 | TRANS_ABORTED(trans)); | |
bb9c12c9 SW |
1885 | } |
1886 | ||
1887 | /* | |
1888 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1889 | * returns, any subsequent transaction will not be allowed to join. | |
1890 | */ | |
1891 | struct btrfs_async_commit { | |
1892 | struct btrfs_trans_handle *newtrans; | |
7892b5af | 1893 | struct work_struct work; |
bb9c12c9 SW |
1894 | }; |
1895 | ||
1896 | static void do_async_commit(struct work_struct *work) | |
1897 | { | |
1898 | struct btrfs_async_commit *ac = | |
7892b5af | 1899 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1900 | |
6fc4e354 SW |
1901 | /* |
1902 | * We've got freeze protection passed with the transaction. | |
1903 | * Tell lockdep about it. | |
1904 | */ | |
b1a06a4b | 1905 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
3a45bb20 | 1906 | __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1907 | |
e209db7a SW |
1908 | current->journal_info = ac->newtrans; |
1909 | ||
3a45bb20 | 1910 | btrfs_commit_transaction(ac->newtrans); |
bb9c12c9 SW |
1911 | kfree(ac); |
1912 | } | |
1913 | ||
1914 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
bb9c12c9 SW |
1915 | int wait_for_unblock) |
1916 | { | |
3a45bb20 | 1917 | struct btrfs_fs_info *fs_info = trans->fs_info; |
bb9c12c9 SW |
1918 | struct btrfs_async_commit *ac; |
1919 | struct btrfs_transaction *cur_trans; | |
1920 | ||
1921 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1922 | if (!ac) |
1923 | return -ENOMEM; | |
bb9c12c9 | 1924 | |
7892b5af | 1925 | INIT_WORK(&ac->work, do_async_commit); |
3a45bb20 | 1926 | ac->newtrans = btrfs_join_transaction(trans->root); |
3612b495 TI |
1927 | if (IS_ERR(ac->newtrans)) { |
1928 | int err = PTR_ERR(ac->newtrans); | |
1929 | kfree(ac); | |
1930 | return err; | |
1931 | } | |
bb9c12c9 SW |
1932 | |
1933 | /* take transaction reference */ | |
bb9c12c9 | 1934 | cur_trans = trans->transaction; |
9b64f57d | 1935 | refcount_inc(&cur_trans->use_count); |
bb9c12c9 | 1936 | |
3a45bb20 | 1937 | btrfs_end_transaction(trans); |
6fc4e354 SW |
1938 | |
1939 | /* | |
1940 | * Tell lockdep we've released the freeze rwsem, since the | |
1941 | * async commit thread will be the one to unlock it. | |
1942 | */ | |
b1a06a4b | 1943 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
0b246afa | 1944 | __sb_writers_release(fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1945 | |
7892b5af | 1946 | schedule_work(&ac->work); |
bb9c12c9 SW |
1947 | |
1948 | /* wait for transaction to start and unblock */ | |
bb9c12c9 | 1949 | if (wait_for_unblock) |
2ff7e61e | 1950 | wait_current_trans_commit_start_and_unblock(fs_info, cur_trans); |
bb9c12c9 | 1951 | else |
2ff7e61e | 1952 | wait_current_trans_commit_start(fs_info, cur_trans); |
bb9c12c9 | 1953 | |
38e88054 SW |
1954 | if (current->journal_info == trans) |
1955 | current->journal_info = NULL; | |
1956 | ||
724e2315 | 1957 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1958 | return 0; |
1959 | } | |
1960 | ||
49b25e05 | 1961 | |
97cb39bb | 1962 | static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) |
49b25e05 | 1963 | { |
97cb39bb | 1964 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 JM |
1965 | struct btrfs_transaction *cur_trans = trans->transaction; |
1966 | ||
b50fff81 | 1967 | WARN_ON(refcount_read(&trans->use_count) > 1); |
49b25e05 | 1968 | |
66642832 | 1969 | btrfs_abort_transaction(trans, err); |
7b8b92af | 1970 | |
0b246afa | 1971 | spin_lock(&fs_info->trans_lock); |
66b6135b | 1972 | |
25d8c284 MX |
1973 | /* |
1974 | * If the transaction is removed from the list, it means this | |
1975 | * transaction has been committed successfully, so it is impossible | |
1976 | * to call the cleanup function. | |
1977 | */ | |
1978 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1979 | |
49b25e05 | 1980 | list_del_init(&cur_trans->list); |
0b246afa | 1981 | if (cur_trans == fs_info->running_transaction) { |
4a9d8bde | 1982 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 1983 | spin_unlock(&fs_info->trans_lock); |
f094ac32 LB |
1984 | wait_event(cur_trans->writer_wait, |
1985 | atomic_read(&cur_trans->num_writers) == 1); | |
1986 | ||
0b246afa | 1987 | spin_lock(&fs_info->trans_lock); |
d7096fc3 | 1988 | } |
0b246afa | 1989 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 1990 | |
2ff7e61e | 1991 | btrfs_cleanup_one_transaction(trans->transaction, fs_info); |
49b25e05 | 1992 | |
0b246afa JM |
1993 | spin_lock(&fs_info->trans_lock); |
1994 | if (cur_trans == fs_info->running_transaction) | |
1995 | fs_info->running_transaction = NULL; | |
1996 | spin_unlock(&fs_info->trans_lock); | |
4a9d8bde | 1997 | |
e0228285 | 1998 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 1999 | sb_end_intwrite(fs_info->sb); |
724e2315 JB |
2000 | btrfs_put_transaction(cur_trans); |
2001 | btrfs_put_transaction(cur_trans); | |
49b25e05 | 2002 | |
97cb39bb | 2003 | trace_btrfs_transaction_commit(trans->root); |
49b25e05 | 2004 | |
49b25e05 JM |
2005 | if (current->journal_info == trans) |
2006 | current->journal_info = NULL; | |
0b246afa | 2007 | btrfs_scrub_cancel(fs_info); |
49b25e05 JM |
2008 | |
2009 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
2010 | } | |
2011 | ||
c7cc64a9 DS |
2012 | /* |
2013 | * Release reserved delayed ref space of all pending block groups of the | |
2014 | * transaction and remove them from the list | |
2015 | */ | |
2016 | static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) | |
2017 | { | |
2018 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2019 | struct btrfs_block_group *block_group, *tmp; |
c7cc64a9 DS |
2020 | |
2021 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { | |
2022 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
2023 | list_del_init(&block_group->bg_list); | |
2024 | } | |
2025 | } | |
2026 | ||
88090ad3 | 2027 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
82436617 | 2028 | { |
ce8ea7cc JB |
2029 | /* |
2030 | * We use writeback_inodes_sb here because if we used | |
2031 | * btrfs_start_delalloc_roots we would deadlock with fs freeze. | |
2032 | * Currently are holding the fs freeze lock, if we do an async flush | |
2033 | * we'll do btrfs_join_transaction() and deadlock because we need to | |
2034 | * wait for the fs freeze lock. Using the direct flushing we benefit | |
2035 | * from already being in a transaction and our join_transaction doesn't | |
2036 | * have to re-take the fs freeze lock. | |
2037 | */ | |
88090ad3 | 2038 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
ce8ea7cc | 2039 | writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); |
82436617 MX |
2040 | return 0; |
2041 | } | |
2042 | ||
88090ad3 | 2043 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) |
82436617 | 2044 | { |
88090ad3 | 2045 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
6374e57a | 2046 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); |
82436617 MX |
2047 | } |
2048 | ||
3a45bb20 | 2049 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans) |
79154b1b | 2050 | { |
3a45bb20 | 2051 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 | 2052 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 2053 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 2054 | int ret; |
79154b1b | 2055 | |
35b814f3 NB |
2056 | ASSERT(refcount_read(&trans->use_count) == 1); |
2057 | ||
d62b23c9 JB |
2058 | /* |
2059 | * Some places just start a transaction to commit it. We need to make | |
2060 | * sure that if this commit fails that the abort code actually marks the | |
2061 | * transaction as failed, so set trans->dirty to make the abort code do | |
2062 | * the right thing. | |
2063 | */ | |
2064 | trans->dirty = true; | |
2065 | ||
8d25a086 | 2066 | /* Stop the commit early if ->aborted is set */ |
bf31f87f | 2067 | if (TRANS_ABORTED(cur_trans)) { |
25287e0a | 2068 | ret = cur_trans->aborted; |
3a45bb20 | 2069 | btrfs_end_transaction(trans); |
e4a2bcac | 2070 | return ret; |
25287e0a | 2071 | } |
49b25e05 | 2072 | |
f45c752b JB |
2073 | btrfs_trans_release_metadata(trans); |
2074 | trans->block_rsv = NULL; | |
2075 | ||
56bec294 | 2076 | /* |
e19eb11f JB |
2077 | * We only want one transaction commit doing the flushing so we do not |
2078 | * waste a bunch of time on lock contention on the extent root node. | |
56bec294 | 2079 | */ |
e19eb11f JB |
2080 | if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, |
2081 | &cur_trans->delayed_refs.flags)) { | |
2082 | /* | |
2083 | * Make a pass through all the delayed refs we have so far. | |
2084 | * Any running threads may add more while we are here. | |
2085 | */ | |
2086 | ret = btrfs_run_delayed_refs(trans, 0); | |
2087 | if (ret) { | |
2088 | btrfs_end_transaction(trans); | |
2089 | return ret; | |
2090 | } | |
2091 | } | |
56bec294 | 2092 | |
119e80df | 2093 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2094 | |
3204d33c | 2095 | if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { |
1bbc621e CM |
2096 | int run_it = 0; |
2097 | ||
2098 | /* this mutex is also taken before trying to set | |
2099 | * block groups readonly. We need to make sure | |
2100 | * that nobody has set a block group readonly | |
2101 | * after a extents from that block group have been | |
2102 | * allocated for cache files. btrfs_set_block_group_ro | |
2103 | * will wait for the transaction to commit if it | |
3204d33c | 2104 | * finds BTRFS_TRANS_DIRTY_BG_RUN set. |
1bbc621e | 2105 | * |
3204d33c JB |
2106 | * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure |
2107 | * only one process starts all the block group IO. It wouldn't | |
1bbc621e CM |
2108 | * hurt to have more than one go through, but there's no |
2109 | * real advantage to it either. | |
2110 | */ | |
0b246afa | 2111 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c JB |
2112 | if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, |
2113 | &cur_trans->flags)) | |
1bbc621e | 2114 | run_it = 1; |
0b246afa | 2115 | mutex_unlock(&fs_info->ro_block_group_mutex); |
1bbc621e | 2116 | |
f9cacae3 | 2117 | if (run_it) { |
21217054 | 2118 | ret = btrfs_start_dirty_block_groups(trans); |
f9cacae3 NB |
2119 | if (ret) { |
2120 | btrfs_end_transaction(trans); | |
2121 | return ret; | |
2122 | } | |
2123 | } | |
1bbc621e CM |
2124 | } |
2125 | ||
0b246afa | 2126 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2127 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { |
d0c2f4fa FM |
2128 | enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; |
2129 | ||
0b246afa | 2130 | spin_unlock(&fs_info->trans_lock); |
9b64f57d | 2131 | refcount_inc(&cur_trans->use_count); |
ccd467d6 | 2132 | |
d0c2f4fa FM |
2133 | if (trans->in_fsync) |
2134 | want_state = TRANS_STATE_SUPER_COMMITTED; | |
2135 | ret = btrfs_end_transaction(trans); | |
2136 | wait_for_commit(cur_trans, want_state); | |
15ee9bc7 | 2137 | |
bf31f87f | 2138 | if (TRANS_ABORTED(cur_trans)) |
b4924a0f LB |
2139 | ret = cur_trans->aborted; |
2140 | ||
724e2315 | 2141 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 2142 | |
49b25e05 | 2143 | return ret; |
79154b1b | 2144 | } |
4313b399 | 2145 | |
4a9d8bde | 2146 | cur_trans->state = TRANS_STATE_COMMIT_START; |
0b246afa | 2147 | wake_up(&fs_info->transaction_blocked_wait); |
bb9c12c9 | 2148 | |
0b246afa | 2149 | if (cur_trans->list.prev != &fs_info->trans_list) { |
d0c2f4fa FM |
2150 | enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; |
2151 | ||
2152 | if (trans->in_fsync) | |
2153 | want_state = TRANS_STATE_SUPER_COMMITTED; | |
2154 | ||
ccd467d6 CM |
2155 | prev_trans = list_entry(cur_trans->list.prev, |
2156 | struct btrfs_transaction, list); | |
d0c2f4fa | 2157 | if (prev_trans->state < want_state) { |
9b64f57d | 2158 | refcount_inc(&prev_trans->use_count); |
0b246afa | 2159 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2160 | |
d0c2f4fa FM |
2161 | wait_for_commit(prev_trans, want_state); |
2162 | ||
bf31f87f | 2163 | ret = READ_ONCE(prev_trans->aborted); |
ccd467d6 | 2164 | |
724e2315 | 2165 | btrfs_put_transaction(prev_trans); |
1f9b8c8f FM |
2166 | if (ret) |
2167 | goto cleanup_transaction; | |
a4abeea4 | 2168 | } else { |
0b246afa | 2169 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2170 | } |
a4abeea4 | 2171 | } else { |
0b246afa | 2172 | spin_unlock(&fs_info->trans_lock); |
cb2d3dad FM |
2173 | /* |
2174 | * The previous transaction was aborted and was already removed | |
2175 | * from the list of transactions at fs_info->trans_list. So we | |
2176 | * abort to prevent writing a new superblock that reflects a | |
2177 | * corrupt state (pointing to trees with unwritten nodes/leafs). | |
2178 | */ | |
2179 | if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) { | |
2180 | ret = -EROFS; | |
2181 | goto cleanup_transaction; | |
2182 | } | |
ccd467d6 | 2183 | } |
15ee9bc7 | 2184 | |
0860adfd MX |
2185 | extwriter_counter_dec(cur_trans, trans->type); |
2186 | ||
88090ad3 | 2187 | ret = btrfs_start_delalloc_flush(fs_info); |
82436617 MX |
2188 | if (ret) |
2189 | goto cleanup_transaction; | |
2190 | ||
e5c304e6 | 2191 | ret = btrfs_run_delayed_items(trans); |
581227d0 MX |
2192 | if (ret) |
2193 | goto cleanup_transaction; | |
15ee9bc7 | 2194 | |
581227d0 MX |
2195 | wait_event(cur_trans->writer_wait, |
2196 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 2197 | |
581227d0 | 2198 | /* some pending stuffs might be added after the previous flush. */ |
e5c304e6 | 2199 | ret = btrfs_run_delayed_items(trans); |
ca469637 MX |
2200 | if (ret) |
2201 | goto cleanup_transaction; | |
2202 | ||
88090ad3 | 2203 | btrfs_wait_delalloc_flush(fs_info); |
cb7ab021 | 2204 | |
48778179 FM |
2205 | /* |
2206 | * Wait for all ordered extents started by a fast fsync that joined this | |
2207 | * transaction. Otherwise if this transaction commits before the ordered | |
2208 | * extents complete we lose logged data after a power failure. | |
2209 | */ | |
2210 | wait_event(cur_trans->pending_wait, | |
2211 | atomic_read(&cur_trans->pending_ordered) == 0); | |
2212 | ||
2ff7e61e | 2213 | btrfs_scrub_pause(fs_info); |
ed0ca140 JB |
2214 | /* |
2215 | * Ok now we need to make sure to block out any other joins while we | |
2216 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 2217 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 | 2218 | */ |
0b246afa | 2219 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2220 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 2221 | spin_unlock(&fs_info->trans_lock); |
ed0ca140 JB |
2222 | wait_event(cur_trans->writer_wait, |
2223 | atomic_read(&cur_trans->num_writers) == 1); | |
2224 | ||
bf31f87f | 2225 | if (TRANS_ABORTED(cur_trans)) { |
2cba30f1 | 2226 | ret = cur_trans->aborted; |
6cf7f77e | 2227 | goto scrub_continue; |
2cba30f1 | 2228 | } |
7585717f CM |
2229 | /* |
2230 | * the reloc mutex makes sure that we stop | |
2231 | * the balancing code from coming in and moving | |
2232 | * extents around in the middle of the commit | |
2233 | */ | |
0b246afa | 2234 | mutex_lock(&fs_info->reloc_mutex); |
7585717f | 2235 | |
42874b3d MX |
2236 | /* |
2237 | * We needn't worry about the delayed items because we will | |
2238 | * deal with them in create_pending_snapshot(), which is the | |
2239 | * core function of the snapshot creation. | |
2240 | */ | |
08d50ca3 | 2241 | ret = create_pending_snapshots(trans); |
56e9f6ea DS |
2242 | if (ret) |
2243 | goto unlock_reloc; | |
3063d29f | 2244 | |
42874b3d MX |
2245 | /* |
2246 | * We insert the dir indexes of the snapshots and update the inode | |
2247 | * of the snapshots' parents after the snapshot creation, so there | |
2248 | * are some delayed items which are not dealt with. Now deal with | |
2249 | * them. | |
2250 | * | |
2251 | * We needn't worry that this operation will corrupt the snapshots, | |
2252 | * because all the tree which are snapshoted will be forced to COW | |
2253 | * the nodes and leaves. | |
2254 | */ | |
e5c304e6 | 2255 | ret = btrfs_run_delayed_items(trans); |
56e9f6ea DS |
2256 | if (ret) |
2257 | goto unlock_reloc; | |
16cdcec7 | 2258 | |
c79a70b1 | 2259 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
56e9f6ea DS |
2260 | if (ret) |
2261 | goto unlock_reloc; | |
56bec294 | 2262 | |
e999376f CM |
2263 | /* |
2264 | * make sure none of the code above managed to slip in a | |
2265 | * delayed item | |
2266 | */ | |
ccdf9b30 | 2267 | btrfs_assert_delayed_root_empty(fs_info); |
e999376f | 2268 | |
2c90e5d6 | 2269 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 2270 | |
e02119d5 CM |
2271 | /* btrfs_commit_tree_roots is responsible for getting the |
2272 | * various roots consistent with each other. Every pointer | |
2273 | * in the tree of tree roots has to point to the most up to date | |
2274 | * root for every subvolume and other tree. So, we have to keep | |
2275 | * the tree logging code from jumping in and changing any | |
2276 | * of the trees. | |
2277 | * | |
2278 | * At this point in the commit, there can't be any tree-log | |
2279 | * writers, but a little lower down we drop the trans mutex | |
2280 | * and let new people in. By holding the tree_log_mutex | |
2281 | * from now until after the super is written, we avoid races | |
2282 | * with the tree-log code. | |
2283 | */ | |
0b246afa | 2284 | mutex_lock(&fs_info->tree_log_mutex); |
e02119d5 | 2285 | |
7e4443d9 | 2286 | ret = commit_fs_roots(trans); |
56e9f6ea DS |
2287 | if (ret) |
2288 | goto unlock_tree_log; | |
54aa1f4d | 2289 | |
3818aea2 | 2290 | /* |
7e1876ac DS |
2291 | * Since the transaction is done, we can apply the pending changes |
2292 | * before the next transaction. | |
3818aea2 | 2293 | */ |
0b246afa | 2294 | btrfs_apply_pending_changes(fs_info); |
3818aea2 | 2295 | |
5d4f98a2 | 2296 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2297 | * safe to free the root of tree log roots |
2298 | */ | |
0b246afa | 2299 | btrfs_free_log_root_tree(trans, fs_info); |
e02119d5 | 2300 | |
0ed4792a QW |
2301 | /* |
2302 | * Since fs roots are all committed, we can get a quite accurate | |
2303 | * new_roots. So let's do quota accounting. | |
2304 | */ | |
460fb20a | 2305 | ret = btrfs_qgroup_account_extents(trans); |
56e9f6ea DS |
2306 | if (ret < 0) |
2307 | goto unlock_tree_log; | |
0ed4792a | 2308 | |
9386d8bc | 2309 | ret = commit_cowonly_roots(trans); |
56e9f6ea DS |
2310 | if (ret) |
2311 | goto unlock_tree_log; | |
54aa1f4d | 2312 | |
2cba30f1 MX |
2313 | /* |
2314 | * The tasks which save the space cache and inode cache may also | |
2315 | * update ->aborted, check it. | |
2316 | */ | |
bf31f87f | 2317 | if (TRANS_ABORTED(cur_trans)) { |
2cba30f1 | 2318 | ret = cur_trans->aborted; |
56e9f6ea | 2319 | goto unlock_tree_log; |
2cba30f1 MX |
2320 | } |
2321 | ||
0b246afa | 2322 | cur_trans = fs_info->running_transaction; |
5d4f98a2 | 2323 | |
0b246afa JM |
2324 | btrfs_set_root_node(&fs_info->tree_root->root_item, |
2325 | fs_info->tree_root->node); | |
2326 | list_add_tail(&fs_info->tree_root->dirty_list, | |
9e351cc8 | 2327 | &cur_trans->switch_commits); |
5d4f98a2 | 2328 | |
0b246afa JM |
2329 | btrfs_set_root_node(&fs_info->chunk_root->root_item, |
2330 | fs_info->chunk_root->node); | |
2331 | list_add_tail(&fs_info->chunk_root->dirty_list, | |
9e351cc8 JB |
2332 | &cur_trans->switch_commits); |
2333 | ||
889bfa39 | 2334 | switch_commit_roots(trans); |
5d4f98a2 | 2335 | |
ce93ec54 | 2336 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2337 | ASSERT(list_empty(&cur_trans->io_bgs)); |
2ff7e61e | 2338 | update_super_roots(fs_info); |
e02119d5 | 2339 | |
0b246afa JM |
2340 | btrfs_set_super_log_root(fs_info->super_copy, 0); |
2341 | btrfs_set_super_log_root_level(fs_info->super_copy, 0); | |
2342 | memcpy(fs_info->super_for_commit, fs_info->super_copy, | |
2343 | sizeof(*fs_info->super_copy)); | |
ccd467d6 | 2344 | |
bbbf7243 | 2345 | btrfs_commit_device_sizes(cur_trans); |
935e5cc9 | 2346 | |
0b246afa JM |
2347 | clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
2348 | clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); | |
656f30db | 2349 | |
4fbcdf66 FM |
2350 | btrfs_trans_release_chunk_metadata(trans); |
2351 | ||
0b246afa | 2352 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2353 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
0b246afa JM |
2354 | fs_info->running_transaction = NULL; |
2355 | spin_unlock(&fs_info->trans_lock); | |
2356 | mutex_unlock(&fs_info->reloc_mutex); | |
b7ec40d7 | 2357 | |
0b246afa | 2358 | wake_up(&fs_info->transaction_wait); |
e6dcd2dc | 2359 | |
70458a58 | 2360 | ret = btrfs_write_and_wait_transaction(trans); |
49b25e05 | 2361 | if (ret) { |
0b246afa JM |
2362 | btrfs_handle_fs_error(fs_info, ret, |
2363 | "Error while writing out transaction"); | |
56e9f6ea DS |
2364 | /* |
2365 | * reloc_mutex has been unlocked, tree_log_mutex is still held | |
2366 | * but we can't jump to unlock_tree_log causing double unlock | |
2367 | */ | |
0b246afa | 2368 | mutex_unlock(&fs_info->tree_log_mutex); |
6cf7f77e | 2369 | goto scrub_continue; |
49b25e05 JM |
2370 | } |
2371 | ||
d3575156 NA |
2372 | /* |
2373 | * At this point, we should have written all the tree blocks allocated | |
2374 | * in this transaction. So it's now safe to free the redirtyied extent | |
2375 | * buffers. | |
2376 | */ | |
2377 | btrfs_free_redirty_list(cur_trans); | |
2378 | ||
eece6a9c | 2379 | ret = write_all_supers(fs_info, 0); |
e02119d5 CM |
2380 | /* |
2381 | * the super is written, we can safely allow the tree-loggers | |
2382 | * to go about their business | |
2383 | */ | |
0b246afa | 2384 | mutex_unlock(&fs_info->tree_log_mutex); |
c1f32b7c AJ |
2385 | if (ret) |
2386 | goto scrub_continue; | |
e02119d5 | 2387 | |
d0c2f4fa FM |
2388 | /* |
2389 | * We needn't acquire the lock here because there is no other task | |
2390 | * which can change it. | |
2391 | */ | |
2392 | cur_trans->state = TRANS_STATE_SUPER_COMMITTED; | |
2393 | wake_up(&cur_trans->commit_wait); | |
2394 | ||
5ead2dd0 | 2395 | btrfs_finish_extent_commit(trans); |
4313b399 | 2396 | |
3204d33c | 2397 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) |
0b246afa | 2398 | btrfs_clear_space_info_full(fs_info); |
13212b54 | 2399 | |
0b246afa | 2400 | fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
2401 | /* |
2402 | * We needn't acquire the lock here because there is no other task | |
2403 | * which can change it. | |
2404 | */ | |
2405 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2406 | wake_up(&cur_trans->commit_wait); |
3de4586c | 2407 | |
0b246afa | 2408 | spin_lock(&fs_info->trans_lock); |
13c5a93e | 2409 | list_del_init(&cur_trans->list); |
0b246afa | 2410 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 2411 | |
724e2315 JB |
2412 | btrfs_put_transaction(cur_trans); |
2413 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2414 | |
0860adfd | 2415 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2416 | sb_end_intwrite(fs_info->sb); |
b2b5ef5c | 2417 | |
3a45bb20 | 2418 | trace_btrfs_transaction_commit(trans->root); |
1abe9b8a | 2419 | |
2ff7e61e | 2420 | btrfs_scrub_continue(fs_info); |
a2de733c | 2421 | |
9ed74f2d JB |
2422 | if (current->journal_info == trans) |
2423 | current->journal_info = NULL; | |
2424 | ||
2c90e5d6 | 2425 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2426 | |
79154b1b | 2427 | return ret; |
49b25e05 | 2428 | |
56e9f6ea DS |
2429 | unlock_tree_log: |
2430 | mutex_unlock(&fs_info->tree_log_mutex); | |
2431 | unlock_reloc: | |
2432 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2433 | scrub_continue: |
2ff7e61e | 2434 | btrfs_scrub_continue(fs_info); |
49b25e05 | 2435 | cleanup_transaction: |
dc60c525 | 2436 | btrfs_trans_release_metadata(trans); |
c7cc64a9 | 2437 | btrfs_cleanup_pending_block_groups(trans); |
4fbcdf66 | 2438 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2439 | trans->block_rsv = NULL; |
0b246afa | 2440 | btrfs_warn(fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2441 | if (current->journal_info == trans) |
2442 | current->journal_info = NULL; | |
97cb39bb | 2443 | cleanup_transaction(trans, ret); |
49b25e05 JM |
2444 | |
2445 | return ret; | |
79154b1b CM |
2446 | } |
2447 | ||
d352ac68 | 2448 | /* |
9d1a2a3a DS |
2449 | * return < 0 if error |
2450 | * 0 if there are no more dead_roots at the time of call | |
2451 | * 1 there are more to be processed, call me again | |
2452 | * | |
2453 | * The return value indicates there are certainly more snapshots to delete, but | |
2454 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2455 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2456 | * few seconds later. | |
d352ac68 | 2457 | */ |
9d1a2a3a | 2458 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 2459 | { |
9d1a2a3a | 2460 | int ret; |
5d4f98a2 YZ |
2461 | struct btrfs_fs_info *fs_info = root->fs_info; |
2462 | ||
a4abeea4 | 2463 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2464 | if (list_empty(&fs_info->dead_roots)) { |
2465 | spin_unlock(&fs_info->trans_lock); | |
2466 | return 0; | |
2467 | } | |
2468 | root = list_first_entry(&fs_info->dead_roots, | |
2469 | struct btrfs_root, root_list); | |
cfad392b | 2470 | list_del_init(&root->root_list); |
a4abeea4 | 2471 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2472 | |
4fd786e6 | 2473 | btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); |
76dda93c | 2474 | |
9d1a2a3a | 2475 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2476 | |
9d1a2a3a DS |
2477 | if (btrfs_header_backref_rev(root->node) < |
2478 | BTRFS_MIXED_BACKREF_REV) | |
0078a9f9 | 2479 | ret = btrfs_drop_snapshot(root, 0, 0); |
9d1a2a3a | 2480 | else |
0078a9f9 | 2481 | ret = btrfs_drop_snapshot(root, 1, 0); |
32471dc2 | 2482 | |
dc9492c1 | 2483 | btrfs_put_root(root); |
6596a928 | 2484 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2485 | } |
572d9ab7 DS |
2486 | |
2487 | void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) | |
2488 | { | |
2489 | unsigned long prev; | |
2490 | unsigned long bit; | |
2491 | ||
6c9fe14f | 2492 | prev = xchg(&fs_info->pending_changes, 0); |
572d9ab7 DS |
2493 | if (!prev) |
2494 | return; | |
2495 | ||
d51033d0 DS |
2496 | bit = 1 << BTRFS_PENDING_COMMIT; |
2497 | if (prev & bit) | |
2498 | btrfs_debug(fs_info, "pending commit done"); | |
2499 | prev &= ~bit; | |
2500 | ||
572d9ab7 DS |
2501 | if (prev) |
2502 | btrfs_warn(fs_info, | |
2503 | "unknown pending changes left 0x%lx, ignoring", prev); | |
2504 | } |