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" |
581bb050 | 19 | #include "inode-map.h" |
733f4fbb | 20 | #include "volumes.h" |
8dabb742 | 21 | #include "dev-replace.h" |
fcebe456 | 22 | #include "qgroup.h" |
aac0023c | 23 | #include "block-group.h" |
79154b1b | 24 | |
0f7d52f4 CM |
25 | #define BTRFS_ROOT_TRANS_TAG 0 |
26 | ||
61c047b5 QW |
27 | /* |
28 | * Transaction states and transitions | |
29 | * | |
30 | * No running transaction (fs tree blocks are not modified) | |
31 | * | | |
32 | * | To next stage: | |
33 | * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). | |
34 | * V | |
35 | * Transaction N [[TRANS_STATE_RUNNING]] | |
36 | * | | |
37 | * | New trans handles can be attached to transaction N by calling all | |
38 | * | start_transaction() variants. | |
39 | * | | |
40 | * | To next stage: | |
41 | * | Call btrfs_commit_transaction() on any trans handle attached to | |
42 | * | transaction N | |
43 | * V | |
44 | * Transaction N [[TRANS_STATE_COMMIT_START]] | |
45 | * | | |
46 | * | Will wait for previous running transaction to completely finish if there | |
47 | * | is one | |
48 | * | | |
49 | * | Then one of the following happes: | |
50 | * | - Wait for all other trans handle holders to release. | |
51 | * | The btrfs_commit_transaction() caller will do the commit work. | |
52 | * | - Wait for current transaction to be committed by others. | |
53 | * | Other btrfs_commit_transaction() caller will do the commit work. | |
54 | * | | |
55 | * | At this stage, only btrfs_join_transaction*() variants can attach | |
56 | * | to this running transaction. | |
57 | * | All other variants will wait for current one to finish and attach to | |
58 | * | transaction N+1. | |
59 | * | | |
60 | * | To next stage: | |
61 | * | Caller is chosen to commit transaction N, and all other trans handle | |
62 | * | haven been released. | |
63 | * V | |
64 | * Transaction N [[TRANS_STATE_COMMIT_DOING]] | |
65 | * | | |
66 | * | The heavy lifting transaction work is started. | |
67 | * | From running delayed refs (modifying extent tree) to creating pending | |
68 | * | snapshots, running qgroups. | |
69 | * | In short, modify supporting trees to reflect modifications of subvolume | |
70 | * | trees. | |
71 | * | | |
72 | * | At this stage, all start_transaction() calls will wait for this | |
73 | * | transaction to finish and attach to transaction N+1. | |
74 | * | | |
75 | * | To next stage: | |
76 | * | Until all supporting trees are updated. | |
77 | * V | |
78 | * Transaction N [[TRANS_STATE_UNBLOCKED]] | |
79 | * | Transaction N+1 | |
80 | * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] | |
81 | * | need to write them back to disk and update | | |
82 | * | super blocks. | | |
83 | * | | | |
84 | * | At this stage, new transaction is allowed to | | |
85 | * | start. | | |
86 | * | All new start_transaction() calls will be | | |
87 | * | attached to transid N+1. | | |
88 | * | | | |
89 | * | To next stage: | | |
90 | * | Until all tree blocks are super blocks are | | |
91 | * | written to block devices | | |
92 | * V | | |
93 | * Transaction N [[TRANS_STATE_COMPLETED]] V | |
94 | * All tree blocks and super blocks are written. Transaction N+1 | |
95 | * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] | |
96 | * data structures will be cleaned up. | Life goes on | |
97 | */ | |
e8c9f186 | 98 | static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { |
4a9d8bde | 99 | [TRANS_STATE_RUNNING] = 0U, |
bcf3a3e7 NB |
100 | [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), |
101 | [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | | |
4a9d8bde | 102 | __TRANS_ATTACH | |
a6d155d2 FM |
103 | __TRANS_JOIN | |
104 | __TRANS_JOIN_NOSTART), | |
bcf3a3e7 | 105 | [TRANS_STATE_UNBLOCKED] = (__TRANS_START | |
4a9d8bde MX |
106 | __TRANS_ATTACH | |
107 | __TRANS_JOIN | | |
a6d155d2 FM |
108 | __TRANS_JOIN_NOLOCK | |
109 | __TRANS_JOIN_NOSTART), | |
bcf3a3e7 | 110 | [TRANS_STATE_COMPLETED] = (__TRANS_START | |
4a9d8bde MX |
111 | __TRANS_ATTACH | |
112 | __TRANS_JOIN | | |
a6d155d2 FM |
113 | __TRANS_JOIN_NOLOCK | |
114 | __TRANS_JOIN_NOSTART), | |
4a9d8bde MX |
115 | }; |
116 | ||
724e2315 | 117 | void btrfs_put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 118 | { |
9b64f57d ER |
119 | WARN_ON(refcount_read(&transaction->use_count) == 0); |
120 | if (refcount_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 121 | BUG_ON(!list_empty(&transaction->list)); |
5c9d028b LB |
122 | WARN_ON(!RB_EMPTY_ROOT( |
123 | &transaction->delayed_refs.href_root.rb_root)); | |
81f7eb00 JM |
124 | WARN_ON(!RB_EMPTY_ROOT( |
125 | &transaction->delayed_refs.dirty_extent_root)); | |
1262133b | 126 | if (transaction->delayed_refs.pending_csums) |
ab8d0fc4 JM |
127 | btrfs_err(transaction->fs_info, |
128 | "pending csums is %llu", | |
129 | transaction->delayed_refs.pending_csums); | |
7785a663 FM |
130 | /* |
131 | * If any block groups are found in ->deleted_bgs then it's | |
132 | * because the transaction was aborted and a commit did not | |
133 | * happen (things failed before writing the new superblock | |
134 | * and calling btrfs_finish_extent_commit()), so we can not | |
135 | * discard the physical locations of the block groups. | |
136 | */ | |
137 | while (!list_empty(&transaction->deleted_bgs)) { | |
32da5386 | 138 | struct btrfs_block_group *cache; |
7785a663 FM |
139 | |
140 | cache = list_first_entry(&transaction->deleted_bgs, | |
32da5386 | 141 | struct btrfs_block_group, |
7785a663 FM |
142 | bg_list); |
143 | list_del_init(&cache->bg_list); | |
144 | btrfs_put_block_group_trimming(cache); | |
145 | btrfs_put_block_group(cache); | |
146 | } | |
bbbf7243 | 147 | WARN_ON(!list_empty(&transaction->dev_update_list)); |
4b5faeac | 148 | kfree(transaction); |
78fae27e | 149 | } |
79154b1b CM |
150 | } |
151 | ||
889bfa39 | 152 | static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) |
817d52f8 | 153 | { |
889bfa39 | 154 | struct btrfs_transaction *cur_trans = trans->transaction; |
16916a88 | 155 | struct btrfs_fs_info *fs_info = trans->fs_info; |
9e351cc8 JB |
156 | struct btrfs_root *root, *tmp; |
157 | ||
158 | down_write(&fs_info->commit_root_sem); | |
889bfa39 | 159 | list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, |
9e351cc8 JB |
160 | dirty_list) { |
161 | list_del_init(&root->dirty_list); | |
162 | free_extent_buffer(root->commit_root); | |
163 | root->commit_root = btrfs_root_node(root); | |
4fd786e6 | 164 | if (is_fstree(root->root_key.objectid)) |
9e351cc8 | 165 | btrfs_unpin_free_ino(root); |
41e7acd3 | 166 | extent_io_tree_release(&root->dirty_log_pages); |
370a11b8 | 167 | btrfs_qgroup_clean_swapped_blocks(root); |
9e351cc8 | 168 | } |
2b9dbef2 JB |
169 | |
170 | /* We can free old roots now. */ | |
889bfa39 JB |
171 | spin_lock(&cur_trans->dropped_roots_lock); |
172 | while (!list_empty(&cur_trans->dropped_roots)) { | |
173 | root = list_first_entry(&cur_trans->dropped_roots, | |
2b9dbef2 JB |
174 | struct btrfs_root, root_list); |
175 | list_del_init(&root->root_list); | |
889bfa39 JB |
176 | spin_unlock(&cur_trans->dropped_roots_lock); |
177 | btrfs_free_log(trans, root); | |
2b9dbef2 | 178 | btrfs_drop_and_free_fs_root(fs_info, root); |
889bfa39 | 179 | spin_lock(&cur_trans->dropped_roots_lock); |
2b9dbef2 | 180 | } |
889bfa39 | 181 | spin_unlock(&cur_trans->dropped_roots_lock); |
9e351cc8 | 182 | up_write(&fs_info->commit_root_sem); |
817d52f8 JB |
183 | } |
184 | ||
0860adfd MX |
185 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
186 | unsigned int type) | |
187 | { | |
188 | if (type & TRANS_EXTWRITERS) | |
189 | atomic_inc(&trans->num_extwriters); | |
190 | } | |
191 | ||
192 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
193 | unsigned int type) | |
194 | { | |
195 | if (type & TRANS_EXTWRITERS) | |
196 | atomic_dec(&trans->num_extwriters); | |
197 | } | |
198 | ||
199 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
200 | unsigned int type) | |
201 | { | |
202 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
203 | } | |
204 | ||
205 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
206 | { | |
207 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
208 | } |
209 | ||
fb6dea26 JB |
210 | /* |
211 | * To be called after all the new block groups attached to the transaction | |
212 | * handle have been created (btrfs_create_pending_block_groups()). | |
213 | */ | |
214 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
215 | { | |
216 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
217 | ||
218 | if (!trans->chunk_bytes_reserved) | |
219 | return; | |
220 | ||
221 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
222 | ||
223 | btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, | |
224 | trans->chunk_bytes_reserved); | |
225 | trans->chunk_bytes_reserved = 0; | |
226 | } | |
227 | ||
d352ac68 CM |
228 | /* |
229 | * either allocate a new transaction or hop into the existing one | |
230 | */ | |
2ff7e61e JM |
231 | static noinline int join_transaction(struct btrfs_fs_info *fs_info, |
232 | unsigned int type) | |
79154b1b CM |
233 | { |
234 | struct btrfs_transaction *cur_trans; | |
a4abeea4 | 235 | |
19ae4e81 | 236 | spin_lock(&fs_info->trans_lock); |
d43317dc | 237 | loop: |
49b25e05 | 238 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 239 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 240 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
241 | return -EROFS; |
242 | } | |
243 | ||
19ae4e81 | 244 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 245 | if (cur_trans) { |
871383be | 246 | if (cur_trans->aborted) { |
19ae4e81 | 247 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 248 | return cur_trans->aborted; |
871383be | 249 | } |
4a9d8bde | 250 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
251 | spin_unlock(&fs_info->trans_lock); |
252 | return -EBUSY; | |
253 | } | |
9b64f57d | 254 | refcount_inc(&cur_trans->use_count); |
13c5a93e | 255 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 256 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 257 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 258 | return 0; |
79154b1b | 259 | } |
19ae4e81 | 260 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 261 | |
354aa0fb MX |
262 | /* |
263 | * If we are ATTACH, we just want to catch the current transaction, | |
264 | * and commit it. If there is no transaction, just return ENOENT. | |
265 | */ | |
266 | if (type == TRANS_ATTACH) | |
267 | return -ENOENT; | |
268 | ||
4a9d8bde MX |
269 | /* |
270 | * JOIN_NOLOCK only happens during the transaction commit, so | |
271 | * it is impossible that ->running_transaction is NULL | |
272 | */ | |
273 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
274 | ||
4b5faeac | 275 | cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); |
a4abeea4 JB |
276 | if (!cur_trans) |
277 | return -ENOMEM; | |
d43317dc | 278 | |
19ae4e81 JS |
279 | spin_lock(&fs_info->trans_lock); |
280 | if (fs_info->running_transaction) { | |
d43317dc CM |
281 | /* |
282 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 283 | * to redo the checks above |
d43317dc | 284 | */ |
4b5faeac | 285 | kfree(cur_trans); |
d43317dc | 286 | goto loop; |
87533c47 | 287 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 288 | spin_unlock(&fs_info->trans_lock); |
4b5faeac | 289 | kfree(cur_trans); |
7b8b92af | 290 | return -EROFS; |
79154b1b | 291 | } |
d43317dc | 292 | |
ab8d0fc4 | 293 | cur_trans->fs_info = fs_info; |
a4abeea4 | 294 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 295 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
296 | init_waitqueue_head(&cur_trans->writer_wait); |
297 | init_waitqueue_head(&cur_trans->commit_wait); | |
4a9d8bde | 298 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
299 | /* |
300 | * One for this trans handle, one so it will live on until we | |
301 | * commit the transaction. | |
302 | */ | |
9b64f57d | 303 | refcount_set(&cur_trans->use_count, 2); |
3204d33c | 304 | cur_trans->flags = 0; |
afd48513 | 305 | cur_trans->start_time = ktime_get_seconds(); |
a4abeea4 | 306 | |
a099d0fd AM |
307 | memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); |
308 | ||
5c9d028b | 309 | cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; |
3368d001 | 310 | cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; |
d7df2c79 | 311 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
20b297d6 JS |
312 | |
313 | /* | |
314 | * although the tree mod log is per file system and not per transaction, | |
315 | * the log must never go across transaction boundaries. | |
316 | */ | |
317 | smp_mb(); | |
31b1a2bd | 318 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
5d163e0e | 319 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); |
31b1a2bd | 320 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
5d163e0e | 321 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); |
fc36ed7e | 322 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 323 | |
a4abeea4 JB |
324 | spin_lock_init(&cur_trans->delayed_refs.lock); |
325 | ||
326 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
bbbf7243 | 327 | INIT_LIST_HEAD(&cur_trans->dev_update_list); |
9e351cc8 | 328 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
ce93ec54 | 329 | INIT_LIST_HEAD(&cur_trans->dirty_bgs); |
1bbc621e | 330 | INIT_LIST_HEAD(&cur_trans->io_bgs); |
2b9dbef2 | 331 | INIT_LIST_HEAD(&cur_trans->dropped_roots); |
1bbc621e | 332 | mutex_init(&cur_trans->cache_write_mutex); |
ce93ec54 | 333 | spin_lock_init(&cur_trans->dirty_bgs_lock); |
e33e17ee | 334 | INIT_LIST_HEAD(&cur_trans->deleted_bgs); |
2b9dbef2 | 335 | spin_lock_init(&cur_trans->dropped_roots_lock); |
19ae4e81 | 336 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
c258d6e3 | 337 | extent_io_tree_init(fs_info, &cur_trans->dirty_pages, |
43eb5f29 | 338 | IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode); |
19ae4e81 JS |
339 | fs_info->generation++; |
340 | cur_trans->transid = fs_info->generation; | |
341 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 342 | cur_trans->aborted = 0; |
19ae4e81 | 343 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 344 | |
79154b1b CM |
345 | return 0; |
346 | } | |
347 | ||
d352ac68 | 348 | /* |
d397712b CM |
349 | * this does all the record keeping required to make sure that a reference |
350 | * counted root is properly recorded in a given transaction. This is required | |
351 | * to make sure the old root from before we joined the transaction is deleted | |
352 | * when the transaction commits | |
d352ac68 | 353 | */ |
7585717f | 354 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
6426c7ad QW |
355 | struct btrfs_root *root, |
356 | int force) | |
6702ed49 | 357 | { |
0b246afa JM |
358 | struct btrfs_fs_info *fs_info = root->fs_info; |
359 | ||
6426c7ad QW |
360 | if ((test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
361 | root->last_trans < trans->transid) || force) { | |
0b246afa | 362 | WARN_ON(root == fs_info->extent_root); |
4d31778a | 363 | WARN_ON(!force && root->commit_root != root->node); |
5d4f98a2 | 364 | |
7585717f | 365 | /* |
27cdeb70 | 366 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
367 | * we have the reloc mutex held now, so there |
368 | * is only one writer in this function | |
369 | */ | |
27cdeb70 | 370 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 371 | |
27cdeb70 | 372 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
373 | * they find our root->last_trans update |
374 | */ | |
375 | smp_wmb(); | |
376 | ||
0b246afa | 377 | spin_lock(&fs_info->fs_roots_radix_lock); |
6426c7ad | 378 | if (root->last_trans == trans->transid && !force) { |
0b246afa | 379 | spin_unlock(&fs_info->fs_roots_radix_lock); |
a4abeea4 JB |
380 | return 0; |
381 | } | |
0b246afa JM |
382 | radix_tree_tag_set(&fs_info->fs_roots_radix, |
383 | (unsigned long)root->root_key.objectid, | |
384 | BTRFS_ROOT_TRANS_TAG); | |
385 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
7585717f CM |
386 | root->last_trans = trans->transid; |
387 | ||
388 | /* this is pretty tricky. We don't want to | |
389 | * take the relocation lock in btrfs_record_root_in_trans | |
390 | * unless we're really doing the first setup for this root in | |
391 | * this transaction. | |
392 | * | |
393 | * Normally we'd use root->last_trans as a flag to decide | |
394 | * if we want to take the expensive mutex. | |
395 | * | |
396 | * But, we have to set root->last_trans before we | |
397 | * init the relocation root, otherwise, we trip over warnings | |
398 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 399 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
400 | * fixing up the reloc trees and everyone must wait. |
401 | * | |
402 | * When this is zero, they can trust root->last_trans and fly | |
403 | * through btrfs_record_root_in_trans without having to take the | |
404 | * lock. smp_wmb() makes sure that all the writes above are | |
405 | * done before we pop in the zero below | |
406 | */ | |
5d4f98a2 | 407 | btrfs_init_reloc_root(trans, root); |
c7548af6 | 408 | smp_mb__before_atomic(); |
27cdeb70 | 409 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 YZ |
410 | } |
411 | return 0; | |
412 | } | |
bcc63abb | 413 | |
7585717f | 414 | |
2b9dbef2 JB |
415 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
416 | struct btrfs_root *root) | |
417 | { | |
0b246afa | 418 | struct btrfs_fs_info *fs_info = root->fs_info; |
2b9dbef2 JB |
419 | struct btrfs_transaction *cur_trans = trans->transaction; |
420 | ||
421 | /* Add ourselves to the transaction dropped list */ | |
422 | spin_lock(&cur_trans->dropped_roots_lock); | |
423 | list_add_tail(&root->root_list, &cur_trans->dropped_roots); | |
424 | spin_unlock(&cur_trans->dropped_roots_lock); | |
425 | ||
426 | /* Make sure we don't try to update the root at commit time */ | |
0b246afa JM |
427 | spin_lock(&fs_info->fs_roots_radix_lock); |
428 | radix_tree_tag_clear(&fs_info->fs_roots_radix, | |
2b9dbef2 JB |
429 | (unsigned long)root->root_key.objectid, |
430 | BTRFS_ROOT_TRANS_TAG); | |
0b246afa | 431 | spin_unlock(&fs_info->fs_roots_radix_lock); |
2b9dbef2 JB |
432 | } |
433 | ||
7585717f CM |
434 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
435 | struct btrfs_root *root) | |
436 | { | |
0b246afa JM |
437 | struct btrfs_fs_info *fs_info = root->fs_info; |
438 | ||
27cdeb70 | 439 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
7585717f CM |
440 | return 0; |
441 | ||
442 | /* | |
27cdeb70 | 443 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
444 | * and barriers |
445 | */ | |
446 | smp_rmb(); | |
447 | if (root->last_trans == trans->transid && | |
27cdeb70 | 448 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
449 | return 0; |
450 | ||
0b246afa | 451 | mutex_lock(&fs_info->reloc_mutex); |
6426c7ad | 452 | record_root_in_trans(trans, root, 0); |
0b246afa | 453 | mutex_unlock(&fs_info->reloc_mutex); |
7585717f CM |
454 | |
455 | return 0; | |
456 | } | |
457 | ||
4a9d8bde MX |
458 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
459 | { | |
3296bf56 | 460 | return (trans->state >= TRANS_STATE_COMMIT_START && |
501407aa JB |
461 | trans->state < TRANS_STATE_UNBLOCKED && |
462 | !trans->aborted); | |
4a9d8bde MX |
463 | } |
464 | ||
d352ac68 CM |
465 | /* wait for commit against the current transaction to become unblocked |
466 | * when this is done, it is safe to start a new transaction, but the current | |
467 | * transaction might not be fully on disk. | |
468 | */ | |
2ff7e61e | 469 | static void wait_current_trans(struct btrfs_fs_info *fs_info) |
79154b1b | 470 | { |
f9295749 | 471 | struct btrfs_transaction *cur_trans; |
79154b1b | 472 | |
0b246afa JM |
473 | spin_lock(&fs_info->trans_lock); |
474 | cur_trans = fs_info->running_transaction; | |
4a9d8bde | 475 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
9b64f57d | 476 | refcount_inc(&cur_trans->use_count); |
0b246afa | 477 | spin_unlock(&fs_info->trans_lock); |
72d63ed6 | 478 | |
0b246afa | 479 | wait_event(fs_info->transaction_wait, |
501407aa JB |
480 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
481 | cur_trans->aborted); | |
724e2315 | 482 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 483 | } else { |
0b246afa | 484 | spin_unlock(&fs_info->trans_lock); |
f9295749 | 485 | } |
37d1aeee CM |
486 | } |
487 | ||
2ff7e61e | 488 | static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) |
a22285a6 | 489 | { |
0b246afa | 490 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
a4abeea4 JB |
491 | return 0; |
492 | ||
92e2f7e3 | 493 | if (type == TRANS_START) |
a22285a6 | 494 | return 1; |
a4abeea4 | 495 | |
a22285a6 YZ |
496 | return 0; |
497 | } | |
498 | ||
20dd2cbf MX |
499 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
500 | { | |
0b246afa JM |
501 | struct btrfs_fs_info *fs_info = root->fs_info; |
502 | ||
503 | if (!fs_info->reloc_ctl || | |
27cdeb70 | 504 | !test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
20dd2cbf MX |
505 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
506 | root->reloc_root) | |
507 | return false; | |
508 | ||
509 | return true; | |
510 | } | |
511 | ||
08e007d2 | 512 | static struct btrfs_trans_handle * |
5aed1dd8 | 513 | start_transaction(struct btrfs_root *root, unsigned int num_items, |
003d7c59 JM |
514 | unsigned int type, enum btrfs_reserve_flush_enum flush, |
515 | bool enforce_qgroups) | |
37d1aeee | 516 | { |
0b246afa | 517 | struct btrfs_fs_info *fs_info = root->fs_info; |
ba2c4d4e | 518 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
a22285a6 YZ |
519 | struct btrfs_trans_handle *h; |
520 | struct btrfs_transaction *cur_trans; | |
b5009945 | 521 | u64 num_bytes = 0; |
c5567237 | 522 | u64 qgroup_reserved = 0; |
20dd2cbf MX |
523 | bool reloc_reserved = false; |
524 | int ret; | |
acce952b | 525 | |
46c4e71e | 526 | /* Send isn't supposed to start transactions. */ |
2755a0de | 527 | ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); |
46c4e71e | 528 | |
0b246afa | 529 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
acce952b | 530 | return ERR_PTR(-EROFS); |
2a1eb461 | 531 | |
46c4e71e | 532 | if (current->journal_info) { |
0860adfd | 533 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 | 534 | h = current->journal_info; |
b50fff81 DS |
535 | refcount_inc(&h->use_count); |
536 | WARN_ON(refcount_read(&h->use_count) > 2); | |
2a1eb461 JB |
537 | h->orig_rsv = h->block_rsv; |
538 | h->block_rsv = NULL; | |
539 | goto got_it; | |
540 | } | |
b5009945 JB |
541 | |
542 | /* | |
543 | * Do the reservation before we join the transaction so we can do all | |
544 | * the appropriate flushing if need be. | |
545 | */ | |
003d7c59 | 546 | if (num_items && root != fs_info->chunk_root) { |
ba2c4d4e JB |
547 | struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; |
548 | u64 delayed_refs_bytes = 0; | |
549 | ||
0b246afa | 550 | qgroup_reserved = num_items * fs_info->nodesize; |
733e03a0 QW |
551 | ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, |
552 | enforce_qgroups); | |
7174109c QW |
553 | if (ret) |
554 | return ERR_PTR(ret); | |
c5567237 | 555 | |
ba2c4d4e JB |
556 | /* |
557 | * We want to reserve all the bytes we may need all at once, so | |
558 | * we only do 1 enospc flushing cycle per transaction start. We | |
559 | * accomplish this by simply assuming we'll do 2 x num_items | |
560 | * worth of delayed refs updates in this trans handle, and | |
561 | * refill that amount for whatever is missing in the reserve. | |
562 | */ | |
2bd36e7b | 563 | num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); |
ba2c4d4e JB |
564 | if (delayed_refs_rsv->full == 0) { |
565 | delayed_refs_bytes = num_bytes; | |
566 | num_bytes <<= 1; | |
567 | } | |
568 | ||
20dd2cbf MX |
569 | /* |
570 | * Do the reservation for the relocation root creation | |
571 | */ | |
ee39b432 | 572 | if (need_reserve_reloc_root(root)) { |
0b246afa | 573 | num_bytes += fs_info->nodesize; |
20dd2cbf MX |
574 | reloc_reserved = true; |
575 | } | |
576 | ||
ba2c4d4e JB |
577 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush); |
578 | if (ret) | |
579 | goto reserve_fail; | |
580 | if (delayed_refs_bytes) { | |
581 | btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, | |
582 | delayed_refs_bytes); | |
583 | num_bytes -= delayed_refs_bytes; | |
584 | } | |
585 | } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && | |
586 | !delayed_refs_rsv->full) { | |
587 | /* | |
588 | * Some people call with btrfs_start_transaction(root, 0) | |
589 | * because they can be throttled, but have some other mechanism | |
590 | * for reserving space. We still want these guys to refill the | |
591 | * delayed block_rsv so just add 1 items worth of reservation | |
592 | * here. | |
593 | */ | |
594 | ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); | |
b5009945 | 595 | if (ret) |
843fcf35 | 596 | goto reserve_fail; |
b5009945 | 597 | } |
a22285a6 | 598 | again: |
f2f767e7 | 599 | h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); |
843fcf35 MX |
600 | if (!h) { |
601 | ret = -ENOMEM; | |
602 | goto alloc_fail; | |
603 | } | |
37d1aeee | 604 | |
98114659 JB |
605 | /* |
606 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
607 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
608 | * because we're already holding a ref. We need this because we could | |
609 | * have raced in and did an fsync() on a file which can kick a commit | |
610 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
611 | * |
612 | * If we are ATTACH, it means we just want to catch the current | |
613 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 614 | */ |
0860adfd | 615 | if (type & __TRANS_FREEZABLE) |
0b246afa | 616 | sb_start_intwrite(fs_info->sb); |
b2b5ef5c | 617 | |
2ff7e61e JM |
618 | if (may_wait_transaction(fs_info, type)) |
619 | wait_current_trans(fs_info); | |
a22285a6 | 620 | |
a4abeea4 | 621 | do { |
2ff7e61e | 622 | ret = join_transaction(fs_info, type); |
178260b2 | 623 | if (ret == -EBUSY) { |
2ff7e61e | 624 | wait_current_trans(fs_info); |
a6d155d2 FM |
625 | if (unlikely(type == TRANS_ATTACH || |
626 | type == TRANS_JOIN_NOSTART)) | |
178260b2 MX |
627 | ret = -ENOENT; |
628 | } | |
a4abeea4 JB |
629 | } while (ret == -EBUSY); |
630 | ||
a43f7f82 | 631 | if (ret < 0) |
843fcf35 | 632 | goto join_fail; |
0f7d52f4 | 633 | |
0b246afa | 634 | cur_trans = fs_info->running_transaction; |
a22285a6 YZ |
635 | |
636 | h->transid = cur_trans->transid; | |
637 | h->transaction = cur_trans; | |
d13603ef | 638 | h->root = root; |
b50fff81 | 639 | refcount_set(&h->use_count, 1); |
64b63580 | 640 | h->fs_info = root->fs_info; |
7174109c | 641 | |
a698d075 | 642 | h->type = type; |
d9a0540a | 643 | h->can_flush_pending_bgs = true; |
ea658bad | 644 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 645 | |
a22285a6 | 646 | smp_mb(); |
3296bf56 | 647 | if (cur_trans->state >= TRANS_STATE_COMMIT_START && |
2ff7e61e | 648 | may_wait_transaction(fs_info, type)) { |
abdd2e80 | 649 | current->journal_info = h; |
3a45bb20 | 650 | btrfs_commit_transaction(h); |
a22285a6 YZ |
651 | goto again; |
652 | } | |
653 | ||
b5009945 | 654 | if (num_bytes) { |
0b246afa | 655 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 656 | h->transid, num_bytes, 1); |
0b246afa | 657 | h->block_rsv = &fs_info->trans_block_rsv; |
b5009945 | 658 | h->bytes_reserved = num_bytes; |
20dd2cbf | 659 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 660 | } |
9ed74f2d | 661 | |
2a1eb461 | 662 | got_it: |
a4abeea4 | 663 | btrfs_record_root_in_trans(h, root); |
a22285a6 | 664 | |
bcf3a3e7 | 665 | if (!current->journal_info) |
a22285a6 | 666 | current->journal_info = h; |
79154b1b | 667 | return h; |
843fcf35 MX |
668 | |
669 | join_fail: | |
0860adfd | 670 | if (type & __TRANS_FREEZABLE) |
0b246afa | 671 | sb_end_intwrite(fs_info->sb); |
843fcf35 MX |
672 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
673 | alloc_fail: | |
674 | if (num_bytes) | |
2ff7e61e | 675 | btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, |
843fcf35 MX |
676 | num_bytes); |
677 | reserve_fail: | |
733e03a0 | 678 | btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); |
843fcf35 | 679 | return ERR_PTR(ret); |
79154b1b CM |
680 | } |
681 | ||
f9295749 | 682 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
5aed1dd8 | 683 | unsigned int num_items) |
f9295749 | 684 | { |
08e007d2 | 685 | return start_transaction(root, num_items, TRANS_START, |
003d7c59 | 686 | BTRFS_RESERVE_FLUSH_ALL, true); |
f9295749 | 687 | } |
003d7c59 | 688 | |
8eab77ff FM |
689 | struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( |
690 | struct btrfs_root *root, | |
691 | unsigned int num_items, | |
692 | int min_factor) | |
693 | { | |
0b246afa | 694 | struct btrfs_fs_info *fs_info = root->fs_info; |
8eab77ff FM |
695 | struct btrfs_trans_handle *trans; |
696 | u64 num_bytes; | |
697 | int ret; | |
698 | ||
003d7c59 JM |
699 | /* |
700 | * We have two callers: unlink and block group removal. The | |
701 | * former should succeed even if we will temporarily exceed | |
702 | * quota and the latter operates on the extent root so | |
703 | * qgroup enforcement is ignored anyway. | |
704 | */ | |
705 | trans = start_transaction(root, num_items, TRANS_START, | |
706 | BTRFS_RESERVE_FLUSH_ALL, false); | |
8eab77ff FM |
707 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
708 | return trans; | |
709 | ||
710 | trans = btrfs_start_transaction(root, 0); | |
711 | if (IS_ERR(trans)) | |
712 | return trans; | |
713 | ||
2bd36e7b | 714 | num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); |
0b246afa JM |
715 | ret = btrfs_cond_migrate_bytes(fs_info, &fs_info->trans_block_rsv, |
716 | num_bytes, min_factor); | |
8eab77ff | 717 | if (ret) { |
3a45bb20 | 718 | btrfs_end_transaction(trans); |
8eab77ff FM |
719 | return ERR_PTR(ret); |
720 | } | |
721 | ||
0b246afa | 722 | trans->block_rsv = &fs_info->trans_block_rsv; |
8eab77ff | 723 | trans->bytes_reserved = num_bytes; |
0b246afa | 724 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa | 725 | trans->transid, num_bytes, 1); |
8eab77ff FM |
726 | |
727 | return trans; | |
728 | } | |
8407aa46 | 729 | |
7a7eaa40 | 730 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 731 | { |
003d7c59 JM |
732 | return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, |
733 | true); | |
f9295749 CM |
734 | } |
735 | ||
8d510121 | 736 | struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) |
0af3d00b | 737 | { |
575a75d6 | 738 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, |
003d7c59 | 739 | BTRFS_RESERVE_NO_FLUSH, true); |
0af3d00b JB |
740 | } |
741 | ||
a6d155d2 FM |
742 | /* |
743 | * Similar to regular join but it never starts a transaction when none is | |
744 | * running or after waiting for the current one to finish. | |
745 | */ | |
746 | struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) | |
747 | { | |
748 | return start_transaction(root, 0, TRANS_JOIN_NOSTART, | |
749 | BTRFS_RESERVE_NO_FLUSH, true); | |
750 | } | |
751 | ||
d4edf39b MX |
752 | /* |
753 | * btrfs_attach_transaction() - catch the running transaction | |
754 | * | |
755 | * It is used when we want to commit the current the transaction, but | |
756 | * don't want to start a new one. | |
757 | * | |
758 | * Note: If this function return -ENOENT, it just means there is no | |
759 | * running transaction. But it is possible that the inactive transaction | |
760 | * is still in the memory, not fully on disk. If you hope there is no | |
761 | * inactive transaction in the fs when -ENOENT is returned, you should | |
762 | * invoke | |
763 | * btrfs_attach_transaction_barrier() | |
764 | */ | |
354aa0fb | 765 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 766 | { |
575a75d6 | 767 | return start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 768 | BTRFS_RESERVE_NO_FLUSH, true); |
60376ce4 JB |
769 | } |
770 | ||
d4edf39b | 771 | /* |
90b6d283 | 772 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b | 773 | * |
52042d8e | 774 | * It is similar to the above function, the difference is this one |
d4edf39b MX |
775 | * will wait for all the inactive transactions until they fully |
776 | * complete. | |
777 | */ | |
778 | struct btrfs_trans_handle * | |
779 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
780 | { | |
781 | struct btrfs_trans_handle *trans; | |
782 | ||
575a75d6 | 783 | trans = start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 784 | BTRFS_RESERVE_NO_FLUSH, true); |
8d9e220c | 785 | if (trans == ERR_PTR(-ENOENT)) |
2ff7e61e | 786 | btrfs_wait_for_commit(root->fs_info, 0); |
d4edf39b MX |
787 | |
788 | return trans; | |
789 | } | |
790 | ||
d352ac68 | 791 | /* wait for a transaction commit to be fully complete */ |
2ff7e61e | 792 | static noinline void wait_for_commit(struct btrfs_transaction *commit) |
89ce8a63 | 793 | { |
4a9d8bde | 794 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
795 | } |
796 | ||
2ff7e61e | 797 | int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) |
46204592 SW |
798 | { |
799 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 800 | int ret = 0; |
46204592 | 801 | |
46204592 | 802 | if (transid) { |
0b246afa | 803 | if (transid <= fs_info->last_trans_committed) |
a4abeea4 | 804 | goto out; |
46204592 SW |
805 | |
806 | /* find specified transaction */ | |
0b246afa JM |
807 | spin_lock(&fs_info->trans_lock); |
808 | list_for_each_entry(t, &fs_info->trans_list, list) { | |
46204592 SW |
809 | if (t->transid == transid) { |
810 | cur_trans = t; | |
9b64f57d | 811 | refcount_inc(&cur_trans->use_count); |
8cd2807f | 812 | ret = 0; |
46204592 SW |
813 | break; |
814 | } | |
8cd2807f MX |
815 | if (t->transid > transid) { |
816 | ret = 0; | |
46204592 | 817 | break; |
8cd2807f | 818 | } |
46204592 | 819 | } |
0b246afa | 820 | spin_unlock(&fs_info->trans_lock); |
42383020 SW |
821 | |
822 | /* | |
823 | * The specified transaction doesn't exist, or we | |
824 | * raced with btrfs_commit_transaction | |
825 | */ | |
826 | if (!cur_trans) { | |
0b246afa | 827 | if (transid > fs_info->last_trans_committed) |
42383020 | 828 | ret = -EINVAL; |
8cd2807f | 829 | goto out; |
42383020 | 830 | } |
46204592 SW |
831 | } else { |
832 | /* find newest transaction that is committing | committed */ | |
0b246afa JM |
833 | spin_lock(&fs_info->trans_lock); |
834 | list_for_each_entry_reverse(t, &fs_info->trans_list, | |
46204592 | 835 | list) { |
4a9d8bde MX |
836 | if (t->state >= TRANS_STATE_COMMIT_START) { |
837 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 838 | break; |
46204592 | 839 | cur_trans = t; |
9b64f57d | 840 | refcount_inc(&cur_trans->use_count); |
46204592 SW |
841 | break; |
842 | } | |
843 | } | |
0b246afa | 844 | spin_unlock(&fs_info->trans_lock); |
46204592 | 845 | if (!cur_trans) |
a4abeea4 | 846 | goto out; /* nothing committing|committed */ |
46204592 SW |
847 | } |
848 | ||
2ff7e61e | 849 | wait_for_commit(cur_trans); |
724e2315 | 850 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 851 | out: |
46204592 SW |
852 | return ret; |
853 | } | |
854 | ||
2ff7e61e | 855 | void btrfs_throttle(struct btrfs_fs_info *fs_info) |
37d1aeee | 856 | { |
92e2f7e3 | 857 | wait_current_trans(fs_info); |
37d1aeee CM |
858 | } |
859 | ||
2ff7e61e | 860 | static int should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa | 861 | { |
2ff7e61e | 862 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0b246afa | 863 | |
64403612 | 864 | if (btrfs_check_space_for_delayed_refs(fs_info)) |
1be41b78 | 865 | return 1; |
36ba022a | 866 | |
2ff7e61e | 867 | return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); |
8929ecfa YZ |
868 | } |
869 | ||
3a45bb20 | 870 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa YZ |
871 | { |
872 | struct btrfs_transaction *cur_trans = trans->transaction; | |
8929ecfa | 873 | |
a4abeea4 | 874 | smp_mb(); |
3296bf56 | 875 | if (cur_trans->state >= TRANS_STATE_COMMIT_START || |
4a9d8bde | 876 | cur_trans->delayed_refs.flushing) |
8929ecfa YZ |
877 | return 1; |
878 | ||
2ff7e61e | 879 | return should_end_transaction(trans); |
8929ecfa YZ |
880 | } |
881 | ||
dc60c525 NB |
882 | static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) |
883 | ||
0e34693f | 884 | { |
dc60c525 NB |
885 | struct btrfs_fs_info *fs_info = trans->fs_info; |
886 | ||
0e34693f NB |
887 | if (!trans->block_rsv) { |
888 | ASSERT(!trans->bytes_reserved); | |
889 | return; | |
890 | } | |
891 | ||
892 | if (!trans->bytes_reserved) | |
893 | return; | |
894 | ||
895 | ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); | |
896 | trace_btrfs_space_reservation(fs_info, "transaction", | |
897 | trans->transid, trans->bytes_reserved, 0); | |
898 | btrfs_block_rsv_release(fs_info, trans->block_rsv, | |
899 | trans->bytes_reserved); | |
900 | trans->bytes_reserved = 0; | |
901 | } | |
902 | ||
89ce8a63 | 903 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
3a45bb20 | 904 | int throttle) |
79154b1b | 905 | { |
3a45bb20 | 906 | struct btrfs_fs_info *info = trans->fs_info; |
8929ecfa | 907 | struct btrfs_transaction *cur_trans = trans->transaction; |
4edc2ca3 | 908 | int err = 0; |
c3e69d58 | 909 | |
b50fff81 DS |
910 | if (refcount_read(&trans->use_count) > 1) { |
911 | refcount_dec(&trans->use_count); | |
2a1eb461 JB |
912 | trans->block_rsv = trans->orig_rsv; |
913 | return 0; | |
914 | } | |
915 | ||
dc60c525 | 916 | btrfs_trans_release_metadata(trans); |
4c13d758 | 917 | trans->block_rsv = NULL; |
c5567237 | 918 | |
119e80df | 919 | btrfs_create_pending_block_groups(trans); |
ea658bad | 920 | |
4fbcdf66 FM |
921 | btrfs_trans_release_chunk_metadata(trans); |
922 | ||
0860adfd | 923 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 924 | sb_end_intwrite(info->sb); |
6df7881a | 925 | |
8929ecfa | 926 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
927 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
928 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 929 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 930 | |
093258e6 | 931 | cond_wake_up(&cur_trans->writer_wait); |
724e2315 | 932 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
933 | |
934 | if (current->journal_info == trans) | |
935 | current->journal_info = NULL; | |
ab78c84d | 936 | |
24bbcf04 | 937 | if (throttle) |
2ff7e61e | 938 | btrfs_run_delayed_iputs(info); |
24bbcf04 | 939 | |
49b25e05 | 940 | if (trans->aborted || |
0b246afa | 941 | test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) { |
4e121c06 | 942 | wake_up_process(info->transaction_kthread); |
4edc2ca3 | 943 | err = -EIO; |
4e121c06 | 944 | } |
49b25e05 | 945 | |
4edc2ca3 DJ |
946 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
947 | return err; | |
79154b1b CM |
948 | } |
949 | ||
3a45bb20 | 950 | int btrfs_end_transaction(struct btrfs_trans_handle *trans) |
89ce8a63 | 951 | { |
3a45bb20 | 952 | return __btrfs_end_transaction(trans, 0); |
89ce8a63 CM |
953 | } |
954 | ||
3a45bb20 | 955 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) |
89ce8a63 | 956 | { |
3a45bb20 | 957 | return __btrfs_end_transaction(trans, 1); |
16cdcec7 MX |
958 | } |
959 | ||
d352ac68 CM |
960 | /* |
961 | * when btree blocks are allocated, they have some corresponding bits set for | |
962 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 963 | * those extents are sent to disk but does not wait on them |
d352ac68 | 964 | */ |
2ff7e61e | 965 | int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, |
8cef4e16 | 966 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 967 | { |
777e6bd7 | 968 | int err = 0; |
7c4452b9 | 969 | int werr = 0; |
0b246afa | 970 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 971 | struct extent_state *cached_state = NULL; |
777e6bd7 | 972 | u64 start = 0; |
5f39d397 | 973 | u64 end; |
7c4452b9 | 974 | |
6300463b | 975 | atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
1728366e | 976 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 977 | mark, &cached_state)) { |
663dfbb0 FM |
978 | bool wait_writeback = false; |
979 | ||
980 | err = convert_extent_bit(dirty_pages, start, end, | |
981 | EXTENT_NEED_WAIT, | |
210aa277 | 982 | mark, &cached_state); |
663dfbb0 FM |
983 | /* |
984 | * convert_extent_bit can return -ENOMEM, which is most of the | |
985 | * time a temporary error. So when it happens, ignore the error | |
986 | * and wait for writeback of this range to finish - because we | |
987 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
bf89d38f JM |
988 | * to __btrfs_wait_marked_extents() would not know that |
989 | * writeback for this range started and therefore wouldn't | |
990 | * wait for it to finish - we don't want to commit a | |
991 | * superblock that points to btree nodes/leafs for which | |
992 | * writeback hasn't finished yet (and without errors). | |
663dfbb0 | 993 | * We cleanup any entries left in the io tree when committing |
41e7acd3 | 994 | * the transaction (through extent_io_tree_release()). |
663dfbb0 FM |
995 | */ |
996 | if (err == -ENOMEM) { | |
997 | err = 0; | |
998 | wait_writeback = true; | |
999 | } | |
1000 | if (!err) | |
1001 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
1002 | if (err) |
1003 | werr = err; | |
663dfbb0 FM |
1004 | else if (wait_writeback) |
1005 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 1006 | free_extent_state(cached_state); |
663dfbb0 | 1007 | cached_state = NULL; |
1728366e JB |
1008 | cond_resched(); |
1009 | start = end + 1; | |
7c4452b9 | 1010 | } |
6300463b | 1011 | atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
690587d1 CM |
1012 | return werr; |
1013 | } | |
1014 | ||
1015 | /* | |
1016 | * when btree blocks are allocated, they have some corresponding bits set for | |
1017 | * them in one of two extent_io trees. This is used to make sure all of | |
1018 | * those extents are on disk for transaction or log commit. We wait | |
1019 | * on all the pages and clear them from the dirty pages state tree | |
1020 | */ | |
bf89d38f JM |
1021 | static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, |
1022 | struct extent_io_tree *dirty_pages) | |
690587d1 | 1023 | { |
690587d1 CM |
1024 | int err = 0; |
1025 | int werr = 0; | |
0b246afa | 1026 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 1027 | struct extent_state *cached_state = NULL; |
690587d1 CM |
1028 | u64 start = 0; |
1029 | u64 end; | |
777e6bd7 | 1030 | |
1728366e | 1031 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 1032 | EXTENT_NEED_WAIT, &cached_state)) { |
663dfbb0 FM |
1033 | /* |
1034 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
1035 | * When committing the transaction, we'll remove any entries | |
1036 | * left in the io tree. For a log commit, we don't remove them | |
1037 | * after committing the log because the tree can be accessed | |
1038 | * concurrently - we do it only at transaction commit time when | |
41e7acd3 | 1039 | * it's safe to do it (through extent_io_tree_release()). |
663dfbb0 FM |
1040 | */ |
1041 | err = clear_extent_bit(dirty_pages, start, end, | |
ae0f1625 | 1042 | EXTENT_NEED_WAIT, 0, 0, &cached_state); |
663dfbb0 FM |
1043 | if (err == -ENOMEM) |
1044 | err = 0; | |
1045 | if (!err) | |
1046 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
1047 | if (err) |
1048 | werr = err; | |
e38e2ed7 FM |
1049 | free_extent_state(cached_state); |
1050 | cached_state = NULL; | |
1728366e JB |
1051 | cond_resched(); |
1052 | start = end + 1; | |
777e6bd7 | 1053 | } |
7c4452b9 CM |
1054 | if (err) |
1055 | werr = err; | |
bf89d38f JM |
1056 | return werr; |
1057 | } | |
656f30db | 1058 | |
b9fae2eb | 1059 | static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, |
bf89d38f JM |
1060 | struct extent_io_tree *dirty_pages) |
1061 | { | |
1062 | bool errors = false; | |
1063 | int err; | |
656f30db | 1064 | |
bf89d38f JM |
1065 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); |
1066 | if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) | |
1067 | errors = true; | |
1068 | ||
1069 | if (errors && !err) | |
1070 | err = -EIO; | |
1071 | return err; | |
1072 | } | |
656f30db | 1073 | |
bf89d38f JM |
1074 | int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) |
1075 | { | |
1076 | struct btrfs_fs_info *fs_info = log_root->fs_info; | |
1077 | struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; | |
1078 | bool errors = false; | |
1079 | int err; | |
656f30db | 1080 | |
bf89d38f JM |
1081 | ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
1082 | ||
1083 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); | |
1084 | if ((mark & EXTENT_DIRTY) && | |
1085 | test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) | |
1086 | errors = true; | |
1087 | ||
1088 | if ((mark & EXTENT_NEW) && | |
1089 | test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) | |
1090 | errors = true; | |
1091 | ||
1092 | if (errors && !err) | |
1093 | err = -EIO; | |
1094 | return err; | |
79154b1b CM |
1095 | } |
1096 | ||
690587d1 | 1097 | /* |
c9b577c0 NB |
1098 | * When btree blocks are allocated the corresponding extents are marked dirty. |
1099 | * This function ensures such extents are persisted on disk for transaction or | |
1100 | * log commit. | |
1101 | * | |
1102 | * @trans: transaction whose dirty pages we'd like to write | |
690587d1 | 1103 | */ |
70458a58 | 1104 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) |
690587d1 CM |
1105 | { |
1106 | int ret; | |
1107 | int ret2; | |
c9b577c0 | 1108 | struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; |
70458a58 | 1109 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c6adc9cc | 1110 | struct blk_plug plug; |
690587d1 | 1111 | |
c6adc9cc | 1112 | blk_start_plug(&plug); |
c9b577c0 | 1113 | ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); |
c6adc9cc | 1114 | blk_finish_plug(&plug); |
bf89d38f | 1115 | ret2 = btrfs_wait_extents(fs_info, dirty_pages); |
bf0da8c1 | 1116 | |
41e7acd3 | 1117 | extent_io_tree_release(&trans->transaction->dirty_pages); |
c9b577c0 | 1118 | |
bf0da8c1 CM |
1119 | if (ret) |
1120 | return ret; | |
c9b577c0 | 1121 | else if (ret2) |
bf0da8c1 | 1122 | return ret2; |
c9b577c0 NB |
1123 | else |
1124 | return 0; | |
d0c803c4 CM |
1125 | } |
1126 | ||
d352ac68 CM |
1127 | /* |
1128 | * this is used to update the root pointer in the tree of tree roots. | |
1129 | * | |
1130 | * But, in the case of the extent allocation tree, updating the root | |
1131 | * pointer may allocate blocks which may change the root of the extent | |
1132 | * allocation tree. | |
1133 | * | |
1134 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1135 | * change while the root pointer was being updated in the metadata. | |
1136 | */ | |
0b86a832 CM |
1137 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1138 | struct btrfs_root *root) | |
79154b1b CM |
1139 | { |
1140 | int ret; | |
0b86a832 | 1141 | u64 old_root_bytenr; |
86b9f2ec | 1142 | u64 old_root_used; |
0b246afa JM |
1143 | struct btrfs_fs_info *fs_info = root->fs_info; |
1144 | struct btrfs_root *tree_root = fs_info->tree_root; | |
79154b1b | 1145 | |
86b9f2ec | 1146 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1147 | |
d397712b | 1148 | while (1) { |
0b86a832 | 1149 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1150 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1151 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1152 | break; |
87ef2bb4 | 1153 | |
5d4f98a2 | 1154 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1155 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1156 | &root->root_key, |
1157 | &root->root_item); | |
49b25e05 JM |
1158 | if (ret) |
1159 | return ret; | |
56bec294 | 1160 | |
86b9f2ec | 1161 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1162 | } |
276e680d | 1163 | |
0b86a832 CM |
1164 | return 0; |
1165 | } | |
1166 | ||
d352ac68 CM |
1167 | /* |
1168 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1169 | * |
1170 | * The error handling in this function may not be obvious. Any of the | |
1171 | * failures will cause the file system to go offline. We still need | |
1172 | * to clean up the delayed refs. | |
d352ac68 | 1173 | */ |
9386d8bc | 1174 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) |
0b86a832 | 1175 | { |
9386d8bc | 1176 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ea526d18 | 1177 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1178 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1179 | struct list_head *next; |
84234f3a | 1180 | struct extent_buffer *eb; |
56bec294 | 1181 | int ret; |
84234f3a YZ |
1182 | |
1183 | eb = btrfs_lock_root_node(fs_info->tree_root); | |
49b25e05 JM |
1184 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
1185 | 0, &eb); | |
84234f3a YZ |
1186 | btrfs_tree_unlock(eb); |
1187 | free_extent_buffer(eb); | |
0b86a832 | 1188 | |
49b25e05 JM |
1189 | if (ret) |
1190 | return ret; | |
1191 | ||
c79a70b1 | 1192 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
49b25e05 JM |
1193 | if (ret) |
1194 | return ret; | |
87ef2bb4 | 1195 | |
196c9d8d | 1196 | ret = btrfs_run_dev_stats(trans); |
c16ce190 JB |
1197 | if (ret) |
1198 | return ret; | |
2b584c68 | 1199 | ret = btrfs_run_dev_replace(trans); |
c16ce190 JB |
1200 | if (ret) |
1201 | return ret; | |
280f8bd2 | 1202 | ret = btrfs_run_qgroups(trans); |
c16ce190 JB |
1203 | if (ret) |
1204 | return ret; | |
546adb0d | 1205 | |
bbebb3e0 | 1206 | ret = btrfs_setup_space_cache(trans); |
dcdf7f6d JB |
1207 | if (ret) |
1208 | return ret; | |
1209 | ||
546adb0d | 1210 | /* run_qgroups might have added some more refs */ |
c79a70b1 | 1211 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
c16ce190 JB |
1212 | if (ret) |
1213 | return ret; | |
ea526d18 | 1214 | again: |
d397712b | 1215 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
2ff7e61e | 1216 | struct btrfs_root *root; |
0b86a832 CM |
1217 | next = fs_info->dirty_cowonly_roots.next; |
1218 | list_del_init(next); | |
1219 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1220 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1221 | |
9e351cc8 JB |
1222 | if (root != fs_info->extent_root) |
1223 | list_add_tail(&root->dirty_list, | |
1224 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1225 | ret = update_cowonly_root(trans, root); |
1226 | if (ret) | |
1227 | return ret; | |
c79a70b1 | 1228 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
ea526d18 JB |
1229 | if (ret) |
1230 | return ret; | |
79154b1b | 1231 | } |
276e680d | 1232 | |
1bbc621e | 1233 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
5742d15f | 1234 | ret = btrfs_write_dirty_block_groups(trans); |
ea526d18 JB |
1235 | if (ret) |
1236 | return ret; | |
c79a70b1 | 1237 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
ea526d18 JB |
1238 | if (ret) |
1239 | return ret; | |
1240 | } | |
1241 | ||
1242 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1243 | goto again; | |
1244 | ||
9e351cc8 JB |
1245 | list_add_tail(&fs_info->extent_root->dirty_list, |
1246 | &trans->transaction->switch_commits); | |
9f6cbcbb DS |
1247 | |
1248 | /* Update dev-replace pointer once everything is committed */ | |
1249 | fs_info->dev_replace.committed_cursor_left = | |
1250 | fs_info->dev_replace.cursor_left_last_write_of_item; | |
8dabb742 | 1251 | |
79154b1b CM |
1252 | return 0; |
1253 | } | |
1254 | ||
d352ac68 CM |
1255 | /* |
1256 | * dead roots are old snapshots that need to be deleted. This allocates | |
1257 | * a dirty root struct and adds it into the list of dead roots that need to | |
1258 | * be deleted | |
1259 | */ | |
cfad392b | 1260 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1261 | { |
0b246afa JM |
1262 | struct btrfs_fs_info *fs_info = root->fs_info; |
1263 | ||
1264 | spin_lock(&fs_info->trans_lock); | |
cfad392b | 1265 | if (list_empty(&root->root_list)) |
0b246afa JM |
1266 | list_add_tail(&root->root_list, &fs_info->dead_roots); |
1267 | spin_unlock(&fs_info->trans_lock); | |
5eda7b5e CM |
1268 | } |
1269 | ||
d352ac68 | 1270 | /* |
5d4f98a2 | 1271 | * update all the cowonly tree roots on disk |
d352ac68 | 1272 | */ |
7e4443d9 | 1273 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) |
0f7d52f4 | 1274 | { |
7e4443d9 | 1275 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0f7d52f4 | 1276 | struct btrfs_root *gang[8]; |
0f7d52f4 CM |
1277 | int i; |
1278 | int ret; | |
54aa1f4d CM |
1279 | int err = 0; |
1280 | ||
a4abeea4 | 1281 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1282 | while (1) { |
5d4f98a2 YZ |
1283 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1284 | (void **)gang, 0, | |
0f7d52f4 CM |
1285 | ARRAY_SIZE(gang), |
1286 | BTRFS_ROOT_TRANS_TAG); | |
1287 | if (ret == 0) | |
1288 | break; | |
1289 | for (i = 0; i < ret; i++) { | |
5b4aacef | 1290 | struct btrfs_root *root = gang[i]; |
5d4f98a2 YZ |
1291 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1292 | (unsigned long)root->root_key.objectid, | |
1293 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1294 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1295 | |
e02119d5 | 1296 | btrfs_free_log(trans, root); |
5d4f98a2 | 1297 | btrfs_update_reloc_root(trans, root); |
bcc63abb | 1298 | |
82d5902d LZ |
1299 | btrfs_save_ino_cache(root, trans); |
1300 | ||
f1ebcc74 | 1301 | /* see comments in should_cow_block() */ |
27cdeb70 | 1302 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
c7548af6 | 1303 | smp_mb__after_atomic(); |
f1ebcc74 | 1304 | |
978d910d | 1305 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1306 | list_add_tail(&root->dirty_list, |
1307 | &trans->transaction->switch_commits); | |
978d910d YZ |
1308 | btrfs_set_root_node(&root->root_item, |
1309 | root->node); | |
1310 | } | |
5d4f98a2 | 1311 | |
5d4f98a2 | 1312 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1313 | &root->root_key, |
1314 | &root->root_item); | |
a4abeea4 | 1315 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1316 | if (err) |
1317 | break; | |
733e03a0 | 1318 | btrfs_qgroup_free_meta_all_pertrans(root); |
0f7d52f4 CM |
1319 | } |
1320 | } | |
a4abeea4 | 1321 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1322 | return err; |
0f7d52f4 CM |
1323 | } |
1324 | ||
d352ac68 | 1325 | /* |
de78b51a ES |
1326 | * defrag a given btree. |
1327 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1328 | */ |
de78b51a | 1329 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1330 | { |
1331 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1332 | struct btrfs_trans_handle *trans; |
8929ecfa | 1333 | int ret; |
e9d0b13b | 1334 | |
27cdeb70 | 1335 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1336 | return 0; |
8929ecfa | 1337 | |
6b80053d | 1338 | while (1) { |
8929ecfa YZ |
1339 | trans = btrfs_start_transaction(root, 0); |
1340 | if (IS_ERR(trans)) | |
1341 | return PTR_ERR(trans); | |
1342 | ||
de78b51a | 1343 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1344 | |
3a45bb20 | 1345 | btrfs_end_transaction(trans); |
2ff7e61e | 1346 | btrfs_btree_balance_dirty(info); |
e9d0b13b CM |
1347 | cond_resched(); |
1348 | ||
ab8d0fc4 | 1349 | if (btrfs_fs_closing(info) || ret != -EAGAIN) |
e9d0b13b | 1350 | break; |
210549eb | 1351 | |
ab8d0fc4 JM |
1352 | if (btrfs_defrag_cancelled(info)) { |
1353 | btrfs_debug(info, "defrag_root cancelled"); | |
210549eb DS |
1354 | ret = -EAGAIN; |
1355 | break; | |
1356 | } | |
e9d0b13b | 1357 | } |
27cdeb70 | 1358 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1359 | return ret; |
e9d0b13b CM |
1360 | } |
1361 | ||
6426c7ad QW |
1362 | /* |
1363 | * Do all special snapshot related qgroup dirty hack. | |
1364 | * | |
1365 | * Will do all needed qgroup inherit and dirty hack like switch commit | |
1366 | * roots inside one transaction and write all btree into disk, to make | |
1367 | * qgroup works. | |
1368 | */ | |
1369 | static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, | |
1370 | struct btrfs_root *src, | |
1371 | struct btrfs_root *parent, | |
1372 | struct btrfs_qgroup_inherit *inherit, | |
1373 | u64 dst_objectid) | |
1374 | { | |
1375 | struct btrfs_fs_info *fs_info = src->fs_info; | |
1376 | int ret; | |
1377 | ||
1378 | /* | |
1379 | * Save some performance in the case that qgroups are not | |
1380 | * enabled. If this check races with the ioctl, rescan will | |
1381 | * kick in anyway. | |
1382 | */ | |
9ea6e2b5 | 1383 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
6426c7ad | 1384 | return 0; |
6426c7ad | 1385 | |
4d31778a | 1386 | /* |
52042d8e | 1387 | * Ensure dirty @src will be committed. Or, after coming |
4d31778a QW |
1388 | * commit_fs_roots() and switch_commit_roots(), any dirty but not |
1389 | * recorded root will never be updated again, causing an outdated root | |
1390 | * item. | |
1391 | */ | |
1392 | record_root_in_trans(trans, src, 1); | |
1393 | ||
6426c7ad QW |
1394 | /* |
1395 | * We are going to commit transaction, see btrfs_commit_transaction() | |
1396 | * comment for reason locking tree_log_mutex | |
1397 | */ | |
1398 | mutex_lock(&fs_info->tree_log_mutex); | |
1399 | ||
7e4443d9 | 1400 | ret = commit_fs_roots(trans); |
6426c7ad QW |
1401 | if (ret) |
1402 | goto out; | |
460fb20a | 1403 | ret = btrfs_qgroup_account_extents(trans); |
6426c7ad QW |
1404 | if (ret < 0) |
1405 | goto out; | |
1406 | ||
1407 | /* Now qgroup are all updated, we can inherit it to new qgroups */ | |
a9377422 | 1408 | ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, |
6426c7ad QW |
1409 | inherit); |
1410 | if (ret < 0) | |
1411 | goto out; | |
1412 | ||
1413 | /* | |
1414 | * Now we do a simplified commit transaction, which will: | |
1415 | * 1) commit all subvolume and extent tree | |
1416 | * To ensure all subvolume and extent tree have a valid | |
1417 | * commit_root to accounting later insert_dir_item() | |
1418 | * 2) write all btree blocks onto disk | |
1419 | * This is to make sure later btree modification will be cowed | |
1420 | * Or commit_root can be populated and cause wrong qgroup numbers | |
1421 | * In this simplified commit, we don't really care about other trees | |
1422 | * like chunk and root tree, as they won't affect qgroup. | |
1423 | * And we don't write super to avoid half committed status. | |
1424 | */ | |
9386d8bc | 1425 | ret = commit_cowonly_roots(trans); |
6426c7ad QW |
1426 | if (ret) |
1427 | goto out; | |
889bfa39 | 1428 | switch_commit_roots(trans); |
70458a58 | 1429 | ret = btrfs_write_and_wait_transaction(trans); |
6426c7ad | 1430 | if (ret) |
f7af3934 | 1431 | btrfs_handle_fs_error(fs_info, ret, |
6426c7ad QW |
1432 | "Error while writing out transaction for qgroup"); |
1433 | ||
1434 | out: | |
1435 | mutex_unlock(&fs_info->tree_log_mutex); | |
1436 | ||
1437 | /* | |
1438 | * Force parent root to be updated, as we recorded it before so its | |
1439 | * last_trans == cur_transid. | |
1440 | * Or it won't be committed again onto disk after later | |
1441 | * insert_dir_item() | |
1442 | */ | |
1443 | if (!ret) | |
1444 | record_root_in_trans(trans, parent, 1); | |
1445 | return ret; | |
1446 | } | |
1447 | ||
d352ac68 CM |
1448 | /* |
1449 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1450 | * transaction commit. This does the actual creation. |
1451 | * | |
1452 | * Note: | |
1453 | * If the error which may affect the commitment of the current transaction | |
1454 | * happens, we should return the error number. If the error which just affect | |
1455 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1456 | */ |
80b6794d | 1457 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1458 | struct btrfs_pending_snapshot *pending) |
1459 | { | |
08d50ca3 NB |
1460 | |
1461 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3063d29f | 1462 | struct btrfs_key key; |
80b6794d | 1463 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1464 | struct btrfs_root *tree_root = fs_info->tree_root; |
1465 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1466 | struct btrfs_root *parent_root; |
98c9942a | 1467 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1468 | struct inode *parent_inode; |
42874b3d MX |
1469 | struct btrfs_path *path; |
1470 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1471 | struct dentry *dentry; |
3063d29f | 1472 | struct extent_buffer *tmp; |
925baedd | 1473 | struct extent_buffer *old; |
95582b00 | 1474 | struct timespec64 cur_time; |
aec8030a | 1475 | int ret = 0; |
d68fc57b | 1476 | u64 to_reserve = 0; |
6bdb72de | 1477 | u64 index = 0; |
a22285a6 | 1478 | u64 objectid; |
b83cc969 | 1479 | u64 root_flags; |
8ea05e3a | 1480 | uuid_le new_uuid; |
3063d29f | 1481 | |
8546b570 DS |
1482 | ASSERT(pending->path); |
1483 | path = pending->path; | |
42874b3d | 1484 | |
b0c0ea63 DS |
1485 | ASSERT(pending->root_item); |
1486 | new_root_item = pending->root_item; | |
a22285a6 | 1487 | |
aec8030a MX |
1488 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1489 | if (pending->error) | |
6fa9700e | 1490 | goto no_free_objectid; |
3063d29f | 1491 | |
d6726335 QW |
1492 | /* |
1493 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1494 | * accounted by later btrfs_qgroup_inherit(). | |
1495 | */ | |
1496 | btrfs_set_skip_qgroup(trans, objectid); | |
1497 | ||
147d256e | 1498 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
d68fc57b YZ |
1499 | |
1500 | if (to_reserve > 0) { | |
aec8030a MX |
1501 | pending->error = btrfs_block_rsv_add(root, |
1502 | &pending->block_rsv, | |
1503 | to_reserve, | |
1504 | BTRFS_RESERVE_NO_FLUSH); | |
1505 | if (pending->error) | |
d6726335 | 1506 | goto clear_skip_qgroup; |
d68fc57b YZ |
1507 | } |
1508 | ||
3063d29f | 1509 | key.objectid = objectid; |
a22285a6 YZ |
1510 | key.offset = (u64)-1; |
1511 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1512 | |
6fa9700e | 1513 | rsv = trans->block_rsv; |
a22285a6 | 1514 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1515 | trans->bytes_reserved = trans->block_rsv->reserved; |
0b246afa | 1516 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa JB |
1517 | trans->transid, |
1518 | trans->bytes_reserved, 1); | |
a22285a6 | 1519 | dentry = pending->dentry; |
e9662f70 | 1520 | parent_inode = pending->dir; |
a22285a6 | 1521 | parent_root = BTRFS_I(parent_inode)->root; |
6426c7ad | 1522 | record_root_in_trans(trans, parent_root, 0); |
a22285a6 | 1523 | |
c2050a45 | 1524 | cur_time = current_time(parent_inode); |
04b285f3 | 1525 | |
3063d29f CM |
1526 | /* |
1527 | * insert the directory item | |
1528 | */ | |
877574e2 | 1529 | ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); |
49b25e05 | 1530 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1531 | |
1532 | /* check if there is a file/dir which has the same name. */ | |
1533 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
4a0cc7ca | 1534 | btrfs_ino(BTRFS_I(parent_inode)), |
42874b3d MX |
1535 | dentry->d_name.name, |
1536 | dentry->d_name.len, 0); | |
1537 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1538 | pending->error = -EEXIST; |
aec8030a | 1539 | goto dir_item_existed; |
42874b3d MX |
1540 | } else if (IS_ERR(dir_item)) { |
1541 | ret = PTR_ERR(dir_item); | |
66642832 | 1542 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1543 | goto fail; |
79787eaa | 1544 | } |
42874b3d | 1545 | btrfs_release_path(path); |
52c26179 | 1546 | |
e999376f CM |
1547 | /* |
1548 | * pull in the delayed directory update | |
1549 | * and the delayed inode item | |
1550 | * otherwise we corrupt the FS during | |
1551 | * snapshot | |
1552 | */ | |
e5c304e6 | 1553 | ret = btrfs_run_delayed_items(trans); |
8732d44f | 1554 | if (ret) { /* Transaction aborted */ |
66642832 | 1555 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1556 | goto fail; |
1557 | } | |
e999376f | 1558 | |
6426c7ad | 1559 | record_root_in_trans(trans, root, 0); |
6bdb72de SW |
1560 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1561 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1562 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1563 | |
b83cc969 LZ |
1564 | root_flags = btrfs_root_flags(new_root_item); |
1565 | if (pending->readonly) | |
1566 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1567 | else | |
1568 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1569 | btrfs_set_root_flags(new_root_item, root_flags); | |
1570 | ||
8ea05e3a AB |
1571 | btrfs_set_root_generation_v2(new_root_item, |
1572 | trans->transid); | |
1573 | uuid_le_gen(&new_uuid); | |
1574 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1575 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1576 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1577 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1578 | memset(new_root_item->received_uuid, 0, | |
1579 | sizeof(new_root_item->received_uuid)); | |
1580 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1581 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1582 | btrfs_set_root_stransid(new_root_item, 0); | |
1583 | btrfs_set_root_rtransid(new_root_item, 0); | |
1584 | } | |
3cae210f QW |
1585 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1586 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1587 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1588 | |
6bdb72de | 1589 | old = btrfs_lock_root_node(root); |
49b25e05 | 1590 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1591 | if (ret) { |
1592 | btrfs_tree_unlock(old); | |
1593 | free_extent_buffer(old); | |
66642832 | 1594 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1595 | goto fail; |
79787eaa | 1596 | } |
49b25e05 | 1597 | |
8bead258 | 1598 | btrfs_set_lock_blocking_write(old); |
6bdb72de | 1599 | |
49b25e05 | 1600 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1601 | /* clean up in any case */ |
6bdb72de SW |
1602 | btrfs_tree_unlock(old); |
1603 | free_extent_buffer(old); | |
8732d44f | 1604 | if (ret) { |
66642832 | 1605 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1606 | goto fail; |
1607 | } | |
f1ebcc74 | 1608 | /* see comments in should_cow_block() */ |
27cdeb70 | 1609 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1610 | smp_wmb(); |
1611 | ||
6bdb72de | 1612 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1613 | /* record when the snapshot was created in key.offset */ |
1614 | key.offset = trans->transid; | |
1615 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1616 | btrfs_tree_unlock(tmp); |
1617 | free_extent_buffer(tmp); | |
8732d44f | 1618 | if (ret) { |
66642832 | 1619 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1620 | goto fail; |
1621 | } | |
6bdb72de | 1622 | |
a22285a6 YZ |
1623 | /* |
1624 | * insert root back/forward references | |
1625 | */ | |
6025c19f | 1626 | ret = btrfs_add_root_ref(trans, objectid, |
0660b5af | 1627 | parent_root->root_key.objectid, |
4a0cc7ca | 1628 | btrfs_ino(BTRFS_I(parent_inode)), index, |
a22285a6 | 1629 | dentry->d_name.name, dentry->d_name.len); |
8732d44f | 1630 | if (ret) { |
66642832 | 1631 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1632 | goto fail; |
1633 | } | |
0660b5af | 1634 | |
a22285a6 | 1635 | key.offset = (u64)-1; |
0b246afa | 1636 | pending->snap = btrfs_read_fs_root_no_name(fs_info, &key); |
79787eaa JM |
1637 | if (IS_ERR(pending->snap)) { |
1638 | ret = PTR_ERR(pending->snap); | |
66642832 | 1639 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1640 | goto fail; |
79787eaa | 1641 | } |
d68fc57b | 1642 | |
49b25e05 | 1643 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f | 1644 | if (ret) { |
66642832 | 1645 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1646 | goto fail; |
1647 | } | |
361048f5 | 1648 | |
c79a70b1 | 1649 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
8732d44f | 1650 | if (ret) { |
66642832 | 1651 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1652 | goto fail; |
1653 | } | |
42874b3d | 1654 | |
6426c7ad QW |
1655 | /* |
1656 | * Do special qgroup accounting for snapshot, as we do some qgroup | |
1657 | * snapshot hack to do fast snapshot. | |
1658 | * To co-operate with that hack, we do hack again. | |
1659 | * Or snapshot will be greatly slowed down by a subtree qgroup rescan | |
1660 | */ | |
1661 | ret = qgroup_account_snapshot(trans, root, parent_root, | |
1662 | pending->inherit, objectid); | |
1663 | if (ret < 0) | |
1664 | goto fail; | |
1665 | ||
684572df LF |
1666 | ret = btrfs_insert_dir_item(trans, dentry->d_name.name, |
1667 | dentry->d_name.len, BTRFS_I(parent_inode), | |
1668 | &key, BTRFS_FT_DIR, index); | |
42874b3d | 1669 | /* We have check then name at the beginning, so it is impossible. */ |
9c52057c | 1670 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f | 1671 | if (ret) { |
66642832 | 1672 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1673 | goto fail; |
1674 | } | |
42874b3d | 1675 | |
6ef06d27 | 1676 | btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + |
42874b3d | 1677 | dentry->d_name.len * 2); |
04b285f3 | 1678 | parent_inode->i_mtime = parent_inode->i_ctime = |
c2050a45 | 1679 | current_time(parent_inode); |
be6aef60 | 1680 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
dd5f9615 | 1681 | if (ret) { |
66642832 | 1682 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1683 | goto fail; |
1684 | } | |
cdb345a8 LF |
1685 | ret = btrfs_uuid_tree_add(trans, new_uuid.b, BTRFS_UUID_KEY_SUBVOL, |
1686 | objectid); | |
dd5f9615 | 1687 | if (ret) { |
66642832 | 1688 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1689 | goto fail; |
1690 | } | |
1691 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
cdb345a8 | 1692 | ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, |
dd5f9615 SB |
1693 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, |
1694 | objectid); | |
1695 | if (ret && ret != -EEXIST) { | |
66642832 | 1696 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1697 | goto fail; |
1698 | } | |
1699 | } | |
d6726335 | 1700 | |
c79a70b1 | 1701 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
d6726335 | 1702 | if (ret) { |
66642832 | 1703 | btrfs_abort_transaction(trans, ret); |
d6726335 QW |
1704 | goto fail; |
1705 | } | |
1706 | ||
3063d29f | 1707 | fail: |
aec8030a MX |
1708 | pending->error = ret; |
1709 | dir_item_existed: | |
98c9942a | 1710 | trans->block_rsv = rsv; |
2382c5cc | 1711 | trans->bytes_reserved = 0; |
d6726335 QW |
1712 | clear_skip_qgroup: |
1713 | btrfs_clear_skip_qgroup(trans); | |
6fa9700e MX |
1714 | no_free_objectid: |
1715 | kfree(new_root_item); | |
b0c0ea63 | 1716 | pending->root_item = NULL; |
42874b3d | 1717 | btrfs_free_path(path); |
8546b570 DS |
1718 | pending->path = NULL; |
1719 | ||
49b25e05 | 1720 | return ret; |
3063d29f CM |
1721 | } |
1722 | ||
d352ac68 CM |
1723 | /* |
1724 | * create all the snapshots we've scheduled for creation | |
1725 | */ | |
08d50ca3 | 1726 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) |
3de4586c | 1727 | { |
aec8030a | 1728 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1729 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1730 | int ret = 0; |
3de4586c | 1731 | |
aec8030a MX |
1732 | list_for_each_entry_safe(pending, next, head, list) { |
1733 | list_del(&pending->list); | |
08d50ca3 | 1734 | ret = create_pending_snapshot(trans, pending); |
aec8030a MX |
1735 | if (ret) |
1736 | break; | |
1737 | } | |
1738 | return ret; | |
3de4586c CM |
1739 | } |
1740 | ||
2ff7e61e | 1741 | static void update_super_roots(struct btrfs_fs_info *fs_info) |
5d4f98a2 YZ |
1742 | { |
1743 | struct btrfs_root_item *root_item; | |
1744 | struct btrfs_super_block *super; | |
1745 | ||
0b246afa | 1746 | super = fs_info->super_copy; |
5d4f98a2 | 1747 | |
0b246afa | 1748 | root_item = &fs_info->chunk_root->root_item; |
093e037c DS |
1749 | super->chunk_root = root_item->bytenr; |
1750 | super->chunk_root_generation = root_item->generation; | |
1751 | super->chunk_root_level = root_item->level; | |
5d4f98a2 | 1752 | |
0b246afa | 1753 | root_item = &fs_info->tree_root->root_item; |
093e037c DS |
1754 | super->root = root_item->bytenr; |
1755 | super->generation = root_item->generation; | |
1756 | super->root_level = root_item->level; | |
0b246afa | 1757 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) |
093e037c | 1758 | super->cache_generation = root_item->generation; |
0b246afa | 1759 | if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) |
093e037c | 1760 | super->uuid_tree_generation = root_item->generation; |
5d4f98a2 YZ |
1761 | } |
1762 | ||
f36f3042 CM |
1763 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1764 | { | |
4a9d8bde | 1765 | struct btrfs_transaction *trans; |
f36f3042 | 1766 | int ret = 0; |
4a9d8bde | 1767 | |
a4abeea4 | 1768 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1769 | trans = info->running_transaction; |
1770 | if (trans) | |
1771 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1772 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1773 | return ret; |
1774 | } | |
1775 | ||
8929ecfa YZ |
1776 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1777 | { | |
4a9d8bde | 1778 | struct btrfs_transaction *trans; |
8929ecfa | 1779 | int ret = 0; |
4a9d8bde | 1780 | |
a4abeea4 | 1781 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1782 | trans = info->running_transaction; |
1783 | if (trans) | |
1784 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1785 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1786 | return ret; |
1787 | } | |
1788 | ||
bb9c12c9 SW |
1789 | /* |
1790 | * wait for the current transaction commit to start and block subsequent | |
1791 | * transaction joins | |
1792 | */ | |
2ff7e61e | 1793 | static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info, |
bb9c12c9 SW |
1794 | struct btrfs_transaction *trans) |
1795 | { | |
2ff7e61e JM |
1796 | wait_event(fs_info->transaction_blocked_wait, |
1797 | trans->state >= TRANS_STATE_COMMIT_START || trans->aborted); | |
bb9c12c9 SW |
1798 | } |
1799 | ||
1800 | /* | |
1801 | * wait for the current transaction to start and then become unblocked. | |
1802 | * caller holds ref. | |
1803 | */ | |
2ff7e61e JM |
1804 | static void wait_current_trans_commit_start_and_unblock( |
1805 | struct btrfs_fs_info *fs_info, | |
1806 | struct btrfs_transaction *trans) | |
bb9c12c9 | 1807 | { |
2ff7e61e JM |
1808 | wait_event(fs_info->transaction_wait, |
1809 | trans->state >= TRANS_STATE_UNBLOCKED || trans->aborted); | |
bb9c12c9 SW |
1810 | } |
1811 | ||
1812 | /* | |
1813 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1814 | * returns, any subsequent transaction will not be allowed to join. | |
1815 | */ | |
1816 | struct btrfs_async_commit { | |
1817 | struct btrfs_trans_handle *newtrans; | |
7892b5af | 1818 | struct work_struct work; |
bb9c12c9 SW |
1819 | }; |
1820 | ||
1821 | static void do_async_commit(struct work_struct *work) | |
1822 | { | |
1823 | struct btrfs_async_commit *ac = | |
7892b5af | 1824 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1825 | |
6fc4e354 SW |
1826 | /* |
1827 | * We've got freeze protection passed with the transaction. | |
1828 | * Tell lockdep about it. | |
1829 | */ | |
b1a06a4b | 1830 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
3a45bb20 | 1831 | __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1832 | |
e209db7a SW |
1833 | current->journal_info = ac->newtrans; |
1834 | ||
3a45bb20 | 1835 | btrfs_commit_transaction(ac->newtrans); |
bb9c12c9 SW |
1836 | kfree(ac); |
1837 | } | |
1838 | ||
1839 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
bb9c12c9 SW |
1840 | int wait_for_unblock) |
1841 | { | |
3a45bb20 | 1842 | struct btrfs_fs_info *fs_info = trans->fs_info; |
bb9c12c9 SW |
1843 | struct btrfs_async_commit *ac; |
1844 | struct btrfs_transaction *cur_trans; | |
1845 | ||
1846 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1847 | if (!ac) |
1848 | return -ENOMEM; | |
bb9c12c9 | 1849 | |
7892b5af | 1850 | INIT_WORK(&ac->work, do_async_commit); |
3a45bb20 | 1851 | ac->newtrans = btrfs_join_transaction(trans->root); |
3612b495 TI |
1852 | if (IS_ERR(ac->newtrans)) { |
1853 | int err = PTR_ERR(ac->newtrans); | |
1854 | kfree(ac); | |
1855 | return err; | |
1856 | } | |
bb9c12c9 SW |
1857 | |
1858 | /* take transaction reference */ | |
bb9c12c9 | 1859 | cur_trans = trans->transaction; |
9b64f57d | 1860 | refcount_inc(&cur_trans->use_count); |
bb9c12c9 | 1861 | |
3a45bb20 | 1862 | btrfs_end_transaction(trans); |
6fc4e354 SW |
1863 | |
1864 | /* | |
1865 | * Tell lockdep we've released the freeze rwsem, since the | |
1866 | * async commit thread will be the one to unlock it. | |
1867 | */ | |
b1a06a4b | 1868 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
0b246afa | 1869 | __sb_writers_release(fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1870 | |
7892b5af | 1871 | schedule_work(&ac->work); |
bb9c12c9 SW |
1872 | |
1873 | /* wait for transaction to start and unblock */ | |
bb9c12c9 | 1874 | if (wait_for_unblock) |
2ff7e61e | 1875 | wait_current_trans_commit_start_and_unblock(fs_info, cur_trans); |
bb9c12c9 | 1876 | else |
2ff7e61e | 1877 | wait_current_trans_commit_start(fs_info, cur_trans); |
bb9c12c9 | 1878 | |
38e88054 SW |
1879 | if (current->journal_info == trans) |
1880 | current->journal_info = NULL; | |
1881 | ||
724e2315 | 1882 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1883 | return 0; |
1884 | } | |
1885 | ||
49b25e05 | 1886 | |
97cb39bb | 1887 | static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) |
49b25e05 | 1888 | { |
97cb39bb | 1889 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 JM |
1890 | struct btrfs_transaction *cur_trans = trans->transaction; |
1891 | ||
b50fff81 | 1892 | WARN_ON(refcount_read(&trans->use_count) > 1); |
49b25e05 | 1893 | |
66642832 | 1894 | btrfs_abort_transaction(trans, err); |
7b8b92af | 1895 | |
0b246afa | 1896 | spin_lock(&fs_info->trans_lock); |
66b6135b | 1897 | |
25d8c284 MX |
1898 | /* |
1899 | * If the transaction is removed from the list, it means this | |
1900 | * transaction has been committed successfully, so it is impossible | |
1901 | * to call the cleanup function. | |
1902 | */ | |
1903 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1904 | |
49b25e05 | 1905 | list_del_init(&cur_trans->list); |
0b246afa | 1906 | if (cur_trans == fs_info->running_transaction) { |
4a9d8bde | 1907 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 1908 | spin_unlock(&fs_info->trans_lock); |
f094ac32 LB |
1909 | wait_event(cur_trans->writer_wait, |
1910 | atomic_read(&cur_trans->num_writers) == 1); | |
1911 | ||
0b246afa | 1912 | spin_lock(&fs_info->trans_lock); |
d7096fc3 | 1913 | } |
0b246afa | 1914 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 1915 | |
2ff7e61e | 1916 | btrfs_cleanup_one_transaction(trans->transaction, fs_info); |
49b25e05 | 1917 | |
0b246afa JM |
1918 | spin_lock(&fs_info->trans_lock); |
1919 | if (cur_trans == fs_info->running_transaction) | |
1920 | fs_info->running_transaction = NULL; | |
1921 | spin_unlock(&fs_info->trans_lock); | |
4a9d8bde | 1922 | |
e0228285 | 1923 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 1924 | sb_end_intwrite(fs_info->sb); |
724e2315 JB |
1925 | btrfs_put_transaction(cur_trans); |
1926 | btrfs_put_transaction(cur_trans); | |
49b25e05 | 1927 | |
97cb39bb | 1928 | trace_btrfs_transaction_commit(trans->root); |
49b25e05 | 1929 | |
49b25e05 JM |
1930 | if (current->journal_info == trans) |
1931 | current->journal_info = NULL; | |
0b246afa | 1932 | btrfs_scrub_cancel(fs_info); |
49b25e05 JM |
1933 | |
1934 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1935 | } | |
1936 | ||
c7cc64a9 DS |
1937 | /* |
1938 | * Release reserved delayed ref space of all pending block groups of the | |
1939 | * transaction and remove them from the list | |
1940 | */ | |
1941 | static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) | |
1942 | { | |
1943 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 1944 | struct btrfs_block_group *block_group, *tmp; |
c7cc64a9 DS |
1945 | |
1946 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { | |
1947 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1948 | list_del_init(&block_group->bg_list); | |
1949 | } | |
1950 | } | |
1951 | ||
609e804d | 1952 | static inline int btrfs_start_delalloc_flush(struct btrfs_trans_handle *trans) |
82436617 | 1953 | { |
609e804d FM |
1954 | struct btrfs_fs_info *fs_info = trans->fs_info; |
1955 | ||
ce8ea7cc JB |
1956 | /* |
1957 | * We use writeback_inodes_sb here because if we used | |
1958 | * btrfs_start_delalloc_roots we would deadlock with fs freeze. | |
1959 | * Currently are holding the fs freeze lock, if we do an async flush | |
1960 | * we'll do btrfs_join_transaction() and deadlock because we need to | |
1961 | * wait for the fs freeze lock. Using the direct flushing we benefit | |
1962 | * from already being in a transaction and our join_transaction doesn't | |
1963 | * have to re-take the fs freeze lock. | |
1964 | */ | |
609e804d | 1965 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) { |
ce8ea7cc | 1966 | writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); |
609e804d FM |
1967 | } else { |
1968 | struct btrfs_pending_snapshot *pending; | |
1969 | struct list_head *head = &trans->transaction->pending_snapshots; | |
1970 | ||
1971 | /* | |
1972 | * Flush dellaloc for any root that is going to be snapshotted. | |
1973 | * This is done to avoid a corrupted version of files, in the | |
1974 | * snapshots, that had both buffered and direct IO writes (even | |
1975 | * if they were done sequentially) due to an unordered update of | |
1976 | * the inode's size on disk. | |
1977 | */ | |
1978 | list_for_each_entry(pending, head, list) { | |
1979 | int ret; | |
1980 | ||
1981 | ret = btrfs_start_delalloc_snapshot(pending->root); | |
1982 | if (ret) | |
1983 | return ret; | |
1984 | } | |
1985 | } | |
82436617 MX |
1986 | return 0; |
1987 | } | |
1988 | ||
609e804d | 1989 | static inline void btrfs_wait_delalloc_flush(struct btrfs_trans_handle *trans) |
82436617 | 1990 | { |
609e804d FM |
1991 | struct btrfs_fs_info *fs_info = trans->fs_info; |
1992 | ||
1993 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) { | |
6374e57a | 1994 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); |
609e804d FM |
1995 | } else { |
1996 | struct btrfs_pending_snapshot *pending; | |
1997 | struct list_head *head = &trans->transaction->pending_snapshots; | |
1998 | ||
1999 | /* | |
2000 | * Wait for any dellaloc that we started previously for the roots | |
2001 | * that are going to be snapshotted. This is to avoid a corrupted | |
2002 | * version of files in the snapshots that had both buffered and | |
2003 | * direct IO writes (even if they were done sequentially). | |
2004 | */ | |
2005 | list_for_each_entry(pending, head, list) | |
2006 | btrfs_wait_ordered_extents(pending->root, | |
2007 | U64_MAX, 0, U64_MAX); | |
2008 | } | |
82436617 MX |
2009 | } |
2010 | ||
3a45bb20 | 2011 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans) |
79154b1b | 2012 | { |
3a45bb20 | 2013 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 | 2014 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 2015 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 2016 | int ret; |
79154b1b | 2017 | |
35b814f3 NB |
2018 | ASSERT(refcount_read(&trans->use_count) == 1); |
2019 | ||
d62b23c9 JB |
2020 | /* |
2021 | * Some places just start a transaction to commit it. We need to make | |
2022 | * sure that if this commit fails that the abort code actually marks the | |
2023 | * transaction as failed, so set trans->dirty to make the abort code do | |
2024 | * the right thing. | |
2025 | */ | |
2026 | trans->dirty = true; | |
2027 | ||
8d25a086 | 2028 | /* Stop the commit early if ->aborted is set */ |
20c7bcec | 2029 | if (unlikely(READ_ONCE(cur_trans->aborted))) { |
25287e0a | 2030 | ret = cur_trans->aborted; |
3a45bb20 | 2031 | btrfs_end_transaction(trans); |
e4a2bcac | 2032 | return ret; |
25287e0a | 2033 | } |
49b25e05 | 2034 | |
f45c752b JB |
2035 | btrfs_trans_release_metadata(trans); |
2036 | trans->block_rsv = NULL; | |
2037 | ||
56bec294 CM |
2038 | /* make a pass through all the delayed refs we have so far |
2039 | * any runnings procs may add more while we are here | |
2040 | */ | |
c79a70b1 | 2041 | ret = btrfs_run_delayed_refs(trans, 0); |
e4a2bcac | 2042 | if (ret) { |
3a45bb20 | 2043 | btrfs_end_transaction(trans); |
e4a2bcac JB |
2044 | return ret; |
2045 | } | |
56bec294 | 2046 | |
b7ec40d7 | 2047 | cur_trans = trans->transaction; |
49b25e05 | 2048 | |
56bec294 CM |
2049 | /* |
2050 | * set the flushing flag so procs in this transaction have to | |
2051 | * start sending their work down. | |
2052 | */ | |
b7ec40d7 | 2053 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 2054 | smp_wmb(); |
56bec294 | 2055 | |
119e80df | 2056 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2057 | |
c79a70b1 | 2058 | ret = btrfs_run_delayed_refs(trans, 0); |
e4a2bcac | 2059 | if (ret) { |
3a45bb20 | 2060 | btrfs_end_transaction(trans); |
e4a2bcac JB |
2061 | return ret; |
2062 | } | |
56bec294 | 2063 | |
3204d33c | 2064 | if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { |
1bbc621e CM |
2065 | int run_it = 0; |
2066 | ||
2067 | /* this mutex is also taken before trying to set | |
2068 | * block groups readonly. We need to make sure | |
2069 | * that nobody has set a block group readonly | |
2070 | * after a extents from that block group have been | |
2071 | * allocated for cache files. btrfs_set_block_group_ro | |
2072 | * will wait for the transaction to commit if it | |
3204d33c | 2073 | * finds BTRFS_TRANS_DIRTY_BG_RUN set. |
1bbc621e | 2074 | * |
3204d33c JB |
2075 | * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure |
2076 | * only one process starts all the block group IO. It wouldn't | |
1bbc621e CM |
2077 | * hurt to have more than one go through, but there's no |
2078 | * real advantage to it either. | |
2079 | */ | |
0b246afa | 2080 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c JB |
2081 | if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, |
2082 | &cur_trans->flags)) | |
1bbc621e | 2083 | run_it = 1; |
0b246afa | 2084 | mutex_unlock(&fs_info->ro_block_group_mutex); |
1bbc621e | 2085 | |
f9cacae3 | 2086 | if (run_it) { |
21217054 | 2087 | ret = btrfs_start_dirty_block_groups(trans); |
f9cacae3 NB |
2088 | if (ret) { |
2089 | btrfs_end_transaction(trans); | |
2090 | return ret; | |
2091 | } | |
2092 | } | |
1bbc621e CM |
2093 | } |
2094 | ||
0b246afa | 2095 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2096 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { |
0b246afa | 2097 | spin_unlock(&fs_info->trans_lock); |
9b64f57d | 2098 | refcount_inc(&cur_trans->use_count); |
3a45bb20 | 2099 | ret = btrfs_end_transaction(trans); |
ccd467d6 | 2100 | |
2ff7e61e | 2101 | wait_for_commit(cur_trans); |
15ee9bc7 | 2102 | |
b4924a0f LB |
2103 | if (unlikely(cur_trans->aborted)) |
2104 | ret = cur_trans->aborted; | |
2105 | ||
724e2315 | 2106 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 2107 | |
49b25e05 | 2108 | return ret; |
79154b1b | 2109 | } |
4313b399 | 2110 | |
4a9d8bde | 2111 | cur_trans->state = TRANS_STATE_COMMIT_START; |
0b246afa | 2112 | wake_up(&fs_info->transaction_blocked_wait); |
bb9c12c9 | 2113 | |
0b246afa | 2114 | if (cur_trans->list.prev != &fs_info->trans_list) { |
ccd467d6 CM |
2115 | prev_trans = list_entry(cur_trans->list.prev, |
2116 | struct btrfs_transaction, list); | |
4a9d8bde | 2117 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
9b64f57d | 2118 | refcount_inc(&prev_trans->use_count); |
0b246afa | 2119 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2120 | |
2ff7e61e | 2121 | wait_for_commit(prev_trans); |
1f9b8c8f | 2122 | ret = prev_trans->aborted; |
ccd467d6 | 2123 | |
724e2315 | 2124 | btrfs_put_transaction(prev_trans); |
1f9b8c8f FM |
2125 | if (ret) |
2126 | goto cleanup_transaction; | |
a4abeea4 | 2127 | } else { |
0b246afa | 2128 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2129 | } |
a4abeea4 | 2130 | } else { |
0b246afa | 2131 | spin_unlock(&fs_info->trans_lock); |
cb2d3dad FM |
2132 | /* |
2133 | * The previous transaction was aborted and was already removed | |
2134 | * from the list of transactions at fs_info->trans_list. So we | |
2135 | * abort to prevent writing a new superblock that reflects a | |
2136 | * corrupt state (pointing to trees with unwritten nodes/leafs). | |
2137 | */ | |
2138 | if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) { | |
2139 | ret = -EROFS; | |
2140 | goto cleanup_transaction; | |
2141 | } | |
ccd467d6 | 2142 | } |
15ee9bc7 | 2143 | |
0860adfd MX |
2144 | extwriter_counter_dec(cur_trans, trans->type); |
2145 | ||
609e804d | 2146 | ret = btrfs_start_delalloc_flush(trans); |
82436617 MX |
2147 | if (ret) |
2148 | goto cleanup_transaction; | |
2149 | ||
e5c304e6 | 2150 | ret = btrfs_run_delayed_items(trans); |
581227d0 MX |
2151 | if (ret) |
2152 | goto cleanup_transaction; | |
15ee9bc7 | 2153 | |
581227d0 MX |
2154 | wait_event(cur_trans->writer_wait, |
2155 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 2156 | |
581227d0 | 2157 | /* some pending stuffs might be added after the previous flush. */ |
e5c304e6 | 2158 | ret = btrfs_run_delayed_items(trans); |
ca469637 MX |
2159 | if (ret) |
2160 | goto cleanup_transaction; | |
2161 | ||
609e804d | 2162 | btrfs_wait_delalloc_flush(trans); |
cb7ab021 | 2163 | |
2ff7e61e | 2164 | btrfs_scrub_pause(fs_info); |
ed0ca140 JB |
2165 | /* |
2166 | * Ok now we need to make sure to block out any other joins while we | |
2167 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 2168 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 | 2169 | */ |
0b246afa | 2170 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2171 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 2172 | spin_unlock(&fs_info->trans_lock); |
ed0ca140 JB |
2173 | wait_event(cur_trans->writer_wait, |
2174 | atomic_read(&cur_trans->num_writers) == 1); | |
2175 | ||
2cba30f1 | 2176 | /* ->aborted might be set after the previous check, so check it */ |
20c7bcec | 2177 | if (unlikely(READ_ONCE(cur_trans->aborted))) { |
2cba30f1 | 2178 | ret = cur_trans->aborted; |
6cf7f77e | 2179 | goto scrub_continue; |
2cba30f1 | 2180 | } |
7585717f CM |
2181 | /* |
2182 | * the reloc mutex makes sure that we stop | |
2183 | * the balancing code from coming in and moving | |
2184 | * extents around in the middle of the commit | |
2185 | */ | |
0b246afa | 2186 | mutex_lock(&fs_info->reloc_mutex); |
7585717f | 2187 | |
42874b3d MX |
2188 | /* |
2189 | * We needn't worry about the delayed items because we will | |
2190 | * deal with them in create_pending_snapshot(), which is the | |
2191 | * core function of the snapshot creation. | |
2192 | */ | |
08d50ca3 | 2193 | ret = create_pending_snapshots(trans); |
49b25e05 | 2194 | if (ret) { |
0b246afa | 2195 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2196 | goto scrub_continue; |
49b25e05 | 2197 | } |
3063d29f | 2198 | |
42874b3d MX |
2199 | /* |
2200 | * We insert the dir indexes of the snapshots and update the inode | |
2201 | * of the snapshots' parents after the snapshot creation, so there | |
2202 | * are some delayed items which are not dealt with. Now deal with | |
2203 | * them. | |
2204 | * | |
2205 | * We needn't worry that this operation will corrupt the snapshots, | |
2206 | * because all the tree which are snapshoted will be forced to COW | |
2207 | * the nodes and leaves. | |
2208 | */ | |
e5c304e6 | 2209 | ret = btrfs_run_delayed_items(trans); |
49b25e05 | 2210 | if (ret) { |
0b246afa | 2211 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2212 | goto scrub_continue; |
49b25e05 | 2213 | } |
16cdcec7 | 2214 | |
c79a70b1 | 2215 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
49b25e05 | 2216 | if (ret) { |
0b246afa | 2217 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2218 | goto scrub_continue; |
49b25e05 | 2219 | } |
56bec294 | 2220 | |
e999376f CM |
2221 | /* |
2222 | * make sure none of the code above managed to slip in a | |
2223 | * delayed item | |
2224 | */ | |
ccdf9b30 | 2225 | btrfs_assert_delayed_root_empty(fs_info); |
e999376f | 2226 | |
2c90e5d6 | 2227 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 2228 | |
e02119d5 CM |
2229 | /* btrfs_commit_tree_roots is responsible for getting the |
2230 | * various roots consistent with each other. Every pointer | |
2231 | * in the tree of tree roots has to point to the most up to date | |
2232 | * root for every subvolume and other tree. So, we have to keep | |
2233 | * the tree logging code from jumping in and changing any | |
2234 | * of the trees. | |
2235 | * | |
2236 | * At this point in the commit, there can't be any tree-log | |
2237 | * writers, but a little lower down we drop the trans mutex | |
2238 | * and let new people in. By holding the tree_log_mutex | |
2239 | * from now until after the super is written, we avoid races | |
2240 | * with the tree-log code. | |
2241 | */ | |
0b246afa | 2242 | mutex_lock(&fs_info->tree_log_mutex); |
e02119d5 | 2243 | |
7e4443d9 | 2244 | ret = commit_fs_roots(trans); |
49b25e05 | 2245 | if (ret) { |
0b246afa JM |
2246 | mutex_unlock(&fs_info->tree_log_mutex); |
2247 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2248 | goto scrub_continue; |
49b25e05 | 2249 | } |
54aa1f4d | 2250 | |
3818aea2 | 2251 | /* |
7e1876ac DS |
2252 | * Since the transaction is done, we can apply the pending changes |
2253 | * before the next transaction. | |
3818aea2 | 2254 | */ |
0b246afa | 2255 | btrfs_apply_pending_changes(fs_info); |
3818aea2 | 2256 | |
5d4f98a2 | 2257 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2258 | * safe to free the root of tree log roots |
2259 | */ | |
0b246afa | 2260 | btrfs_free_log_root_tree(trans, fs_info); |
e02119d5 | 2261 | |
82bafb38 QW |
2262 | /* |
2263 | * commit_fs_roots() can call btrfs_save_ino_cache(), which generates | |
2264 | * new delayed refs. Must handle them or qgroup can be wrong. | |
2265 | */ | |
c79a70b1 | 2266 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
82bafb38 QW |
2267 | if (ret) { |
2268 | mutex_unlock(&fs_info->tree_log_mutex); | |
2269 | mutex_unlock(&fs_info->reloc_mutex); | |
2270 | goto scrub_continue; | |
2271 | } | |
2272 | ||
0ed4792a QW |
2273 | /* |
2274 | * Since fs roots are all committed, we can get a quite accurate | |
2275 | * new_roots. So let's do quota accounting. | |
2276 | */ | |
460fb20a | 2277 | ret = btrfs_qgroup_account_extents(trans); |
0ed4792a | 2278 | if (ret < 0) { |
0b246afa JM |
2279 | mutex_unlock(&fs_info->tree_log_mutex); |
2280 | mutex_unlock(&fs_info->reloc_mutex); | |
0ed4792a QW |
2281 | goto scrub_continue; |
2282 | } | |
2283 | ||
9386d8bc | 2284 | ret = commit_cowonly_roots(trans); |
49b25e05 | 2285 | if (ret) { |
0b246afa JM |
2286 | mutex_unlock(&fs_info->tree_log_mutex); |
2287 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2288 | goto scrub_continue; |
49b25e05 | 2289 | } |
54aa1f4d | 2290 | |
2cba30f1 MX |
2291 | /* |
2292 | * The tasks which save the space cache and inode cache may also | |
2293 | * update ->aborted, check it. | |
2294 | */ | |
20c7bcec | 2295 | if (unlikely(READ_ONCE(cur_trans->aborted))) { |
2cba30f1 | 2296 | ret = cur_trans->aborted; |
0b246afa JM |
2297 | mutex_unlock(&fs_info->tree_log_mutex); |
2298 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2299 | goto scrub_continue; |
2cba30f1 MX |
2300 | } |
2301 | ||
8b74c03e | 2302 | btrfs_prepare_extent_commit(fs_info); |
11833d66 | 2303 | |
0b246afa | 2304 | cur_trans = fs_info->running_transaction; |
5d4f98a2 | 2305 | |
0b246afa JM |
2306 | btrfs_set_root_node(&fs_info->tree_root->root_item, |
2307 | fs_info->tree_root->node); | |
2308 | list_add_tail(&fs_info->tree_root->dirty_list, | |
9e351cc8 | 2309 | &cur_trans->switch_commits); |
5d4f98a2 | 2310 | |
0b246afa JM |
2311 | btrfs_set_root_node(&fs_info->chunk_root->root_item, |
2312 | fs_info->chunk_root->node); | |
2313 | list_add_tail(&fs_info->chunk_root->dirty_list, | |
9e351cc8 JB |
2314 | &cur_trans->switch_commits); |
2315 | ||
889bfa39 | 2316 | switch_commit_roots(trans); |
5d4f98a2 | 2317 | |
ce93ec54 | 2318 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2319 | ASSERT(list_empty(&cur_trans->io_bgs)); |
2ff7e61e | 2320 | update_super_roots(fs_info); |
e02119d5 | 2321 | |
0b246afa JM |
2322 | btrfs_set_super_log_root(fs_info->super_copy, 0); |
2323 | btrfs_set_super_log_root_level(fs_info->super_copy, 0); | |
2324 | memcpy(fs_info->super_for_commit, fs_info->super_copy, | |
2325 | sizeof(*fs_info->super_copy)); | |
ccd467d6 | 2326 | |
bbbf7243 | 2327 | btrfs_commit_device_sizes(cur_trans); |
935e5cc9 | 2328 | |
0b246afa JM |
2329 | clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
2330 | clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); | |
656f30db | 2331 | |
4fbcdf66 FM |
2332 | btrfs_trans_release_chunk_metadata(trans); |
2333 | ||
0b246afa | 2334 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2335 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
0b246afa JM |
2336 | fs_info->running_transaction = NULL; |
2337 | spin_unlock(&fs_info->trans_lock); | |
2338 | mutex_unlock(&fs_info->reloc_mutex); | |
b7ec40d7 | 2339 | |
0b246afa | 2340 | wake_up(&fs_info->transaction_wait); |
e6dcd2dc | 2341 | |
70458a58 | 2342 | ret = btrfs_write_and_wait_transaction(trans); |
49b25e05 | 2343 | if (ret) { |
0b246afa JM |
2344 | btrfs_handle_fs_error(fs_info, ret, |
2345 | "Error while writing out transaction"); | |
2346 | mutex_unlock(&fs_info->tree_log_mutex); | |
6cf7f77e | 2347 | goto scrub_continue; |
49b25e05 JM |
2348 | } |
2349 | ||
eece6a9c | 2350 | ret = write_all_supers(fs_info, 0); |
e02119d5 CM |
2351 | /* |
2352 | * the super is written, we can safely allow the tree-loggers | |
2353 | * to go about their business | |
2354 | */ | |
0b246afa | 2355 | mutex_unlock(&fs_info->tree_log_mutex); |
c1f32b7c AJ |
2356 | if (ret) |
2357 | goto scrub_continue; | |
e02119d5 | 2358 | |
5ead2dd0 | 2359 | btrfs_finish_extent_commit(trans); |
4313b399 | 2360 | |
3204d33c | 2361 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) |
0b246afa | 2362 | btrfs_clear_space_info_full(fs_info); |
13212b54 | 2363 | |
0b246afa | 2364 | fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
2365 | /* |
2366 | * We needn't acquire the lock here because there is no other task | |
2367 | * which can change it. | |
2368 | */ | |
2369 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2370 | wake_up(&cur_trans->commit_wait); |
a514d638 | 2371 | clear_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags); |
3de4586c | 2372 | |
0b246afa | 2373 | spin_lock(&fs_info->trans_lock); |
13c5a93e | 2374 | list_del_init(&cur_trans->list); |
0b246afa | 2375 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 2376 | |
724e2315 JB |
2377 | btrfs_put_transaction(cur_trans); |
2378 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2379 | |
0860adfd | 2380 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2381 | sb_end_intwrite(fs_info->sb); |
b2b5ef5c | 2382 | |
3a45bb20 | 2383 | trace_btrfs_transaction_commit(trans->root); |
1abe9b8a | 2384 | |
2ff7e61e | 2385 | btrfs_scrub_continue(fs_info); |
a2de733c | 2386 | |
9ed74f2d JB |
2387 | if (current->journal_info == trans) |
2388 | current->journal_info = NULL; | |
2389 | ||
2c90e5d6 | 2390 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2391 | |
79154b1b | 2392 | return ret; |
49b25e05 | 2393 | |
6cf7f77e | 2394 | scrub_continue: |
2ff7e61e | 2395 | btrfs_scrub_continue(fs_info); |
49b25e05 | 2396 | cleanup_transaction: |
dc60c525 | 2397 | btrfs_trans_release_metadata(trans); |
c7cc64a9 | 2398 | btrfs_cleanup_pending_block_groups(trans); |
4fbcdf66 | 2399 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2400 | trans->block_rsv = NULL; |
0b246afa | 2401 | btrfs_warn(fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2402 | if (current->journal_info == trans) |
2403 | current->journal_info = NULL; | |
97cb39bb | 2404 | cleanup_transaction(trans, ret); |
49b25e05 JM |
2405 | |
2406 | return ret; | |
79154b1b CM |
2407 | } |
2408 | ||
d352ac68 | 2409 | /* |
9d1a2a3a DS |
2410 | * return < 0 if error |
2411 | * 0 if there are no more dead_roots at the time of call | |
2412 | * 1 there are more to be processed, call me again | |
2413 | * | |
2414 | * The return value indicates there are certainly more snapshots to delete, but | |
2415 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2416 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2417 | * few seconds later. | |
d352ac68 | 2418 | */ |
9d1a2a3a | 2419 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 2420 | { |
9d1a2a3a | 2421 | int ret; |
5d4f98a2 YZ |
2422 | struct btrfs_fs_info *fs_info = root->fs_info; |
2423 | ||
a4abeea4 | 2424 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2425 | if (list_empty(&fs_info->dead_roots)) { |
2426 | spin_unlock(&fs_info->trans_lock); | |
2427 | return 0; | |
2428 | } | |
2429 | root = list_first_entry(&fs_info->dead_roots, | |
2430 | struct btrfs_root, root_list); | |
cfad392b | 2431 | list_del_init(&root->root_list); |
a4abeea4 | 2432 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2433 | |
4fd786e6 | 2434 | btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); |
76dda93c | 2435 | |
9d1a2a3a | 2436 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2437 | |
9d1a2a3a DS |
2438 | if (btrfs_header_backref_rev(root->node) < |
2439 | BTRFS_MIXED_BACKREF_REV) | |
2440 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
2441 | else | |
2442 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
32471dc2 | 2443 | |
6596a928 | 2444 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2445 | } |
572d9ab7 DS |
2446 | |
2447 | void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) | |
2448 | { | |
2449 | unsigned long prev; | |
2450 | unsigned long bit; | |
2451 | ||
6c9fe14f | 2452 | prev = xchg(&fs_info->pending_changes, 0); |
572d9ab7 DS |
2453 | if (!prev) |
2454 | return; | |
2455 | ||
7e1876ac DS |
2456 | bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE; |
2457 | if (prev & bit) | |
2458 | btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2459 | prev &= ~bit; | |
2460 | ||
2461 | bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE; | |
2462 | if (prev & bit) | |
2463 | btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2464 | prev &= ~bit; | |
2465 | ||
d51033d0 DS |
2466 | bit = 1 << BTRFS_PENDING_COMMIT; |
2467 | if (prev & bit) | |
2468 | btrfs_debug(fs_info, "pending commit done"); | |
2469 | prev &= ~bit; | |
2470 | ||
572d9ab7 DS |
2471 | if (prev) |
2472 | btrfs_warn(fs_info, | |
2473 | "unknown pending changes left 0x%lx, ignoring", prev); | |
2474 | } |