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) { |
bf31f87f | 246 | if (TRANS_ABORTED(cur_trans)) { |
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); |
fe119a6e NB |
339 | extent_io_tree_init(fs_info, &cur_trans->pinned_extents, |
340 | IO_TREE_FS_PINNED_EXTENTS, NULL); | |
19ae4e81 JS |
341 | fs_info->generation++; |
342 | cur_trans->transid = fs_info->generation; | |
343 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 344 | cur_trans->aborted = 0; |
19ae4e81 | 345 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 346 | |
79154b1b CM |
347 | return 0; |
348 | } | |
349 | ||
d352ac68 | 350 | /* |
d397712b CM |
351 | * this does all the record keeping required to make sure that a reference |
352 | * counted root is properly recorded in a given transaction. This is required | |
353 | * to make sure the old root from before we joined the transaction is deleted | |
354 | * when the transaction commits | |
d352ac68 | 355 | */ |
7585717f | 356 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
6426c7ad QW |
357 | struct btrfs_root *root, |
358 | int force) | |
6702ed49 | 359 | { |
0b246afa JM |
360 | struct btrfs_fs_info *fs_info = root->fs_info; |
361 | ||
6426c7ad QW |
362 | if ((test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
363 | root->last_trans < trans->transid) || force) { | |
0b246afa | 364 | WARN_ON(root == fs_info->extent_root); |
4d31778a | 365 | WARN_ON(!force && root->commit_root != root->node); |
5d4f98a2 | 366 | |
7585717f | 367 | /* |
27cdeb70 | 368 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
369 | * we have the reloc mutex held now, so there |
370 | * is only one writer in this function | |
371 | */ | |
27cdeb70 | 372 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 373 | |
27cdeb70 | 374 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
375 | * they find our root->last_trans update |
376 | */ | |
377 | smp_wmb(); | |
378 | ||
0b246afa | 379 | spin_lock(&fs_info->fs_roots_radix_lock); |
6426c7ad | 380 | if (root->last_trans == trans->transid && !force) { |
0b246afa | 381 | spin_unlock(&fs_info->fs_roots_radix_lock); |
a4abeea4 JB |
382 | return 0; |
383 | } | |
0b246afa JM |
384 | radix_tree_tag_set(&fs_info->fs_roots_radix, |
385 | (unsigned long)root->root_key.objectid, | |
386 | BTRFS_ROOT_TRANS_TAG); | |
387 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
7585717f CM |
388 | root->last_trans = trans->transid; |
389 | ||
390 | /* this is pretty tricky. We don't want to | |
391 | * take the relocation lock in btrfs_record_root_in_trans | |
392 | * unless we're really doing the first setup for this root in | |
393 | * this transaction. | |
394 | * | |
395 | * Normally we'd use root->last_trans as a flag to decide | |
396 | * if we want to take the expensive mutex. | |
397 | * | |
398 | * But, we have to set root->last_trans before we | |
399 | * init the relocation root, otherwise, we trip over warnings | |
400 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 401 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
402 | * fixing up the reloc trees and everyone must wait. |
403 | * | |
404 | * When this is zero, they can trust root->last_trans and fly | |
405 | * through btrfs_record_root_in_trans without having to take the | |
406 | * lock. smp_wmb() makes sure that all the writes above are | |
407 | * done before we pop in the zero below | |
408 | */ | |
5d4f98a2 | 409 | btrfs_init_reloc_root(trans, root); |
c7548af6 | 410 | smp_mb__before_atomic(); |
27cdeb70 | 411 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 YZ |
412 | } |
413 | return 0; | |
414 | } | |
bcc63abb | 415 | |
7585717f | 416 | |
2b9dbef2 JB |
417 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
418 | struct btrfs_root *root) | |
419 | { | |
0b246afa | 420 | struct btrfs_fs_info *fs_info = root->fs_info; |
2b9dbef2 JB |
421 | struct btrfs_transaction *cur_trans = trans->transaction; |
422 | ||
423 | /* Add ourselves to the transaction dropped list */ | |
424 | spin_lock(&cur_trans->dropped_roots_lock); | |
425 | list_add_tail(&root->root_list, &cur_trans->dropped_roots); | |
426 | spin_unlock(&cur_trans->dropped_roots_lock); | |
427 | ||
428 | /* Make sure we don't try to update the root at commit time */ | |
0b246afa JM |
429 | spin_lock(&fs_info->fs_roots_radix_lock); |
430 | radix_tree_tag_clear(&fs_info->fs_roots_radix, | |
2b9dbef2 JB |
431 | (unsigned long)root->root_key.objectid, |
432 | BTRFS_ROOT_TRANS_TAG); | |
0b246afa | 433 | spin_unlock(&fs_info->fs_roots_radix_lock); |
2b9dbef2 JB |
434 | } |
435 | ||
7585717f CM |
436 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
437 | struct btrfs_root *root) | |
438 | { | |
0b246afa JM |
439 | struct btrfs_fs_info *fs_info = root->fs_info; |
440 | ||
27cdeb70 | 441 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
7585717f CM |
442 | return 0; |
443 | ||
444 | /* | |
27cdeb70 | 445 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
446 | * and barriers |
447 | */ | |
448 | smp_rmb(); | |
449 | if (root->last_trans == trans->transid && | |
27cdeb70 | 450 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
451 | return 0; |
452 | ||
0b246afa | 453 | mutex_lock(&fs_info->reloc_mutex); |
6426c7ad | 454 | record_root_in_trans(trans, root, 0); |
0b246afa | 455 | mutex_unlock(&fs_info->reloc_mutex); |
7585717f CM |
456 | |
457 | return 0; | |
458 | } | |
459 | ||
4a9d8bde MX |
460 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
461 | { | |
3296bf56 | 462 | return (trans->state >= TRANS_STATE_COMMIT_START && |
501407aa | 463 | trans->state < TRANS_STATE_UNBLOCKED && |
bf31f87f | 464 | !TRANS_ABORTED(trans)); |
4a9d8bde MX |
465 | } |
466 | ||
d352ac68 CM |
467 | /* wait for commit against the current transaction to become unblocked |
468 | * when this is done, it is safe to start a new transaction, but the current | |
469 | * transaction might not be fully on disk. | |
470 | */ | |
2ff7e61e | 471 | static void wait_current_trans(struct btrfs_fs_info *fs_info) |
79154b1b | 472 | { |
f9295749 | 473 | struct btrfs_transaction *cur_trans; |
79154b1b | 474 | |
0b246afa JM |
475 | spin_lock(&fs_info->trans_lock); |
476 | cur_trans = fs_info->running_transaction; | |
4a9d8bde | 477 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
9b64f57d | 478 | refcount_inc(&cur_trans->use_count); |
0b246afa | 479 | spin_unlock(&fs_info->trans_lock); |
72d63ed6 | 480 | |
0b246afa | 481 | wait_event(fs_info->transaction_wait, |
501407aa | 482 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
bf31f87f | 483 | TRANS_ABORTED(cur_trans)); |
724e2315 | 484 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 485 | } else { |
0b246afa | 486 | spin_unlock(&fs_info->trans_lock); |
f9295749 | 487 | } |
37d1aeee CM |
488 | } |
489 | ||
2ff7e61e | 490 | static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) |
a22285a6 | 491 | { |
0b246afa | 492 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
a4abeea4 JB |
493 | return 0; |
494 | ||
92e2f7e3 | 495 | if (type == TRANS_START) |
a22285a6 | 496 | return 1; |
a4abeea4 | 497 | |
a22285a6 YZ |
498 | return 0; |
499 | } | |
500 | ||
20dd2cbf MX |
501 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
502 | { | |
0b246afa JM |
503 | struct btrfs_fs_info *fs_info = root->fs_info; |
504 | ||
505 | if (!fs_info->reloc_ctl || | |
27cdeb70 | 506 | !test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
20dd2cbf MX |
507 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
508 | root->reloc_root) | |
509 | return false; | |
510 | ||
511 | return true; | |
512 | } | |
513 | ||
08e007d2 | 514 | static struct btrfs_trans_handle * |
5aed1dd8 | 515 | start_transaction(struct btrfs_root *root, unsigned int num_items, |
003d7c59 JM |
516 | unsigned int type, enum btrfs_reserve_flush_enum flush, |
517 | bool enforce_qgroups) | |
37d1aeee | 518 | { |
0b246afa | 519 | struct btrfs_fs_info *fs_info = root->fs_info; |
ba2c4d4e | 520 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
a22285a6 YZ |
521 | struct btrfs_trans_handle *h; |
522 | struct btrfs_transaction *cur_trans; | |
b5009945 | 523 | u64 num_bytes = 0; |
c5567237 | 524 | u64 qgroup_reserved = 0; |
20dd2cbf MX |
525 | bool reloc_reserved = false; |
526 | int ret; | |
acce952b | 527 | |
46c4e71e | 528 | /* Send isn't supposed to start transactions. */ |
2755a0de | 529 | ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); |
46c4e71e | 530 | |
0b246afa | 531 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
acce952b | 532 | return ERR_PTR(-EROFS); |
2a1eb461 | 533 | |
46c4e71e | 534 | if (current->journal_info) { |
0860adfd | 535 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 | 536 | h = current->journal_info; |
b50fff81 DS |
537 | refcount_inc(&h->use_count); |
538 | WARN_ON(refcount_read(&h->use_count) > 2); | |
2a1eb461 JB |
539 | h->orig_rsv = h->block_rsv; |
540 | h->block_rsv = NULL; | |
541 | goto got_it; | |
542 | } | |
b5009945 JB |
543 | |
544 | /* | |
545 | * Do the reservation before we join the transaction so we can do all | |
546 | * the appropriate flushing if need be. | |
547 | */ | |
003d7c59 | 548 | if (num_items && root != fs_info->chunk_root) { |
ba2c4d4e JB |
549 | struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; |
550 | u64 delayed_refs_bytes = 0; | |
551 | ||
0b246afa | 552 | qgroup_reserved = num_items * fs_info->nodesize; |
733e03a0 QW |
553 | ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, |
554 | enforce_qgroups); | |
7174109c QW |
555 | if (ret) |
556 | return ERR_PTR(ret); | |
c5567237 | 557 | |
ba2c4d4e JB |
558 | /* |
559 | * We want to reserve all the bytes we may need all at once, so | |
560 | * we only do 1 enospc flushing cycle per transaction start. We | |
561 | * accomplish this by simply assuming we'll do 2 x num_items | |
562 | * worth of delayed refs updates in this trans handle, and | |
563 | * refill that amount for whatever is missing in the reserve. | |
564 | */ | |
2bd36e7b | 565 | num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); |
ba2c4d4e JB |
566 | if (delayed_refs_rsv->full == 0) { |
567 | delayed_refs_bytes = num_bytes; | |
568 | num_bytes <<= 1; | |
569 | } | |
570 | ||
20dd2cbf MX |
571 | /* |
572 | * Do the reservation for the relocation root creation | |
573 | */ | |
ee39b432 | 574 | if (need_reserve_reloc_root(root)) { |
0b246afa | 575 | num_bytes += fs_info->nodesize; |
20dd2cbf MX |
576 | reloc_reserved = true; |
577 | } | |
578 | ||
ba2c4d4e JB |
579 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush); |
580 | if (ret) | |
581 | goto reserve_fail; | |
582 | if (delayed_refs_bytes) { | |
583 | btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, | |
584 | delayed_refs_bytes); | |
585 | num_bytes -= delayed_refs_bytes; | |
586 | } | |
587 | } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && | |
588 | !delayed_refs_rsv->full) { | |
589 | /* | |
590 | * Some people call with btrfs_start_transaction(root, 0) | |
591 | * because they can be throttled, but have some other mechanism | |
592 | * for reserving space. We still want these guys to refill the | |
593 | * delayed block_rsv so just add 1 items worth of reservation | |
594 | * here. | |
595 | */ | |
596 | ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); | |
b5009945 | 597 | if (ret) |
843fcf35 | 598 | goto reserve_fail; |
b5009945 | 599 | } |
a22285a6 | 600 | again: |
f2f767e7 | 601 | h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); |
843fcf35 MX |
602 | if (!h) { |
603 | ret = -ENOMEM; | |
604 | goto alloc_fail; | |
605 | } | |
37d1aeee | 606 | |
98114659 JB |
607 | /* |
608 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
609 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
610 | * because we're already holding a ref. We need this because we could | |
611 | * have raced in and did an fsync() on a file which can kick a commit | |
612 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
613 | * |
614 | * If we are ATTACH, it means we just want to catch the current | |
615 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 616 | */ |
0860adfd | 617 | if (type & __TRANS_FREEZABLE) |
0b246afa | 618 | sb_start_intwrite(fs_info->sb); |
b2b5ef5c | 619 | |
2ff7e61e JM |
620 | if (may_wait_transaction(fs_info, type)) |
621 | wait_current_trans(fs_info); | |
a22285a6 | 622 | |
a4abeea4 | 623 | do { |
2ff7e61e | 624 | ret = join_transaction(fs_info, type); |
178260b2 | 625 | if (ret == -EBUSY) { |
2ff7e61e | 626 | wait_current_trans(fs_info); |
a6d155d2 FM |
627 | if (unlikely(type == TRANS_ATTACH || |
628 | type == TRANS_JOIN_NOSTART)) | |
178260b2 MX |
629 | ret = -ENOENT; |
630 | } | |
a4abeea4 JB |
631 | } while (ret == -EBUSY); |
632 | ||
a43f7f82 | 633 | if (ret < 0) |
843fcf35 | 634 | goto join_fail; |
0f7d52f4 | 635 | |
0b246afa | 636 | cur_trans = fs_info->running_transaction; |
a22285a6 YZ |
637 | |
638 | h->transid = cur_trans->transid; | |
639 | h->transaction = cur_trans; | |
d13603ef | 640 | h->root = root; |
b50fff81 | 641 | refcount_set(&h->use_count, 1); |
64b63580 | 642 | h->fs_info = root->fs_info; |
7174109c | 643 | |
a698d075 | 644 | h->type = type; |
d9a0540a | 645 | h->can_flush_pending_bgs = true; |
ea658bad | 646 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 647 | |
a22285a6 | 648 | smp_mb(); |
3296bf56 | 649 | if (cur_trans->state >= TRANS_STATE_COMMIT_START && |
2ff7e61e | 650 | may_wait_transaction(fs_info, type)) { |
abdd2e80 | 651 | current->journal_info = h; |
3a45bb20 | 652 | btrfs_commit_transaction(h); |
a22285a6 YZ |
653 | goto again; |
654 | } | |
655 | ||
b5009945 | 656 | if (num_bytes) { |
0b246afa | 657 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 658 | h->transid, num_bytes, 1); |
0b246afa | 659 | h->block_rsv = &fs_info->trans_block_rsv; |
b5009945 | 660 | h->bytes_reserved = num_bytes; |
20dd2cbf | 661 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 662 | } |
9ed74f2d | 663 | |
2a1eb461 | 664 | got_it: |
a4abeea4 | 665 | btrfs_record_root_in_trans(h, root); |
a22285a6 | 666 | |
bcf3a3e7 | 667 | if (!current->journal_info) |
a22285a6 | 668 | current->journal_info = h; |
79154b1b | 669 | return h; |
843fcf35 MX |
670 | |
671 | join_fail: | |
0860adfd | 672 | if (type & __TRANS_FREEZABLE) |
0b246afa | 673 | sb_end_intwrite(fs_info->sb); |
843fcf35 MX |
674 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
675 | alloc_fail: | |
676 | if (num_bytes) | |
2ff7e61e | 677 | btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, |
843fcf35 MX |
678 | num_bytes); |
679 | reserve_fail: | |
733e03a0 | 680 | btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); |
843fcf35 | 681 | return ERR_PTR(ret); |
79154b1b CM |
682 | } |
683 | ||
f9295749 | 684 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
5aed1dd8 | 685 | unsigned int num_items) |
f9295749 | 686 | { |
08e007d2 | 687 | return start_transaction(root, num_items, TRANS_START, |
003d7c59 | 688 | BTRFS_RESERVE_FLUSH_ALL, true); |
f9295749 | 689 | } |
003d7c59 | 690 | |
8eab77ff FM |
691 | struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( |
692 | struct btrfs_root *root, | |
693 | unsigned int num_items, | |
694 | int min_factor) | |
695 | { | |
0b246afa | 696 | struct btrfs_fs_info *fs_info = root->fs_info; |
8eab77ff FM |
697 | struct btrfs_trans_handle *trans; |
698 | u64 num_bytes; | |
699 | int ret; | |
700 | ||
003d7c59 JM |
701 | /* |
702 | * We have two callers: unlink and block group removal. The | |
703 | * former should succeed even if we will temporarily exceed | |
704 | * quota and the latter operates on the extent root so | |
705 | * qgroup enforcement is ignored anyway. | |
706 | */ | |
707 | trans = start_transaction(root, num_items, TRANS_START, | |
708 | BTRFS_RESERVE_FLUSH_ALL, false); | |
8eab77ff FM |
709 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
710 | return trans; | |
711 | ||
712 | trans = btrfs_start_transaction(root, 0); | |
713 | if (IS_ERR(trans)) | |
714 | return trans; | |
715 | ||
2bd36e7b | 716 | num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); |
0b246afa JM |
717 | ret = btrfs_cond_migrate_bytes(fs_info, &fs_info->trans_block_rsv, |
718 | num_bytes, min_factor); | |
8eab77ff | 719 | if (ret) { |
3a45bb20 | 720 | btrfs_end_transaction(trans); |
8eab77ff FM |
721 | return ERR_PTR(ret); |
722 | } | |
723 | ||
0b246afa | 724 | trans->block_rsv = &fs_info->trans_block_rsv; |
8eab77ff | 725 | trans->bytes_reserved = num_bytes; |
0b246afa | 726 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa | 727 | trans->transid, num_bytes, 1); |
8eab77ff FM |
728 | |
729 | return trans; | |
730 | } | |
8407aa46 | 731 | |
7a7eaa40 | 732 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 733 | { |
003d7c59 JM |
734 | return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, |
735 | true); | |
f9295749 CM |
736 | } |
737 | ||
8d510121 | 738 | struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) |
0af3d00b | 739 | { |
575a75d6 | 740 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, |
003d7c59 | 741 | BTRFS_RESERVE_NO_FLUSH, true); |
0af3d00b JB |
742 | } |
743 | ||
a6d155d2 FM |
744 | /* |
745 | * Similar to regular join but it never starts a transaction when none is | |
746 | * running or after waiting for the current one to finish. | |
747 | */ | |
748 | struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) | |
749 | { | |
750 | return start_transaction(root, 0, TRANS_JOIN_NOSTART, | |
751 | BTRFS_RESERVE_NO_FLUSH, true); | |
752 | } | |
753 | ||
d4edf39b MX |
754 | /* |
755 | * btrfs_attach_transaction() - catch the running transaction | |
756 | * | |
757 | * It is used when we want to commit the current the transaction, but | |
758 | * don't want to start a new one. | |
759 | * | |
760 | * Note: If this function return -ENOENT, it just means there is no | |
761 | * running transaction. But it is possible that the inactive transaction | |
762 | * is still in the memory, not fully on disk. If you hope there is no | |
763 | * inactive transaction in the fs when -ENOENT is returned, you should | |
764 | * invoke | |
765 | * btrfs_attach_transaction_barrier() | |
766 | */ | |
354aa0fb | 767 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 768 | { |
575a75d6 | 769 | return start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 770 | BTRFS_RESERVE_NO_FLUSH, true); |
60376ce4 JB |
771 | } |
772 | ||
d4edf39b | 773 | /* |
90b6d283 | 774 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b | 775 | * |
52042d8e | 776 | * It is similar to the above function, the difference is this one |
d4edf39b MX |
777 | * will wait for all the inactive transactions until they fully |
778 | * complete. | |
779 | */ | |
780 | struct btrfs_trans_handle * | |
781 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
782 | { | |
783 | struct btrfs_trans_handle *trans; | |
784 | ||
575a75d6 | 785 | trans = start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 786 | BTRFS_RESERVE_NO_FLUSH, true); |
8d9e220c | 787 | if (trans == ERR_PTR(-ENOENT)) |
2ff7e61e | 788 | btrfs_wait_for_commit(root->fs_info, 0); |
d4edf39b MX |
789 | |
790 | return trans; | |
791 | } | |
792 | ||
d352ac68 | 793 | /* wait for a transaction commit to be fully complete */ |
2ff7e61e | 794 | static noinline void wait_for_commit(struct btrfs_transaction *commit) |
89ce8a63 | 795 | { |
4a9d8bde | 796 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
797 | } |
798 | ||
2ff7e61e | 799 | int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) |
46204592 SW |
800 | { |
801 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 802 | int ret = 0; |
46204592 | 803 | |
46204592 | 804 | if (transid) { |
0b246afa | 805 | if (transid <= fs_info->last_trans_committed) |
a4abeea4 | 806 | goto out; |
46204592 SW |
807 | |
808 | /* find specified transaction */ | |
0b246afa JM |
809 | spin_lock(&fs_info->trans_lock); |
810 | list_for_each_entry(t, &fs_info->trans_list, list) { | |
46204592 SW |
811 | if (t->transid == transid) { |
812 | cur_trans = t; | |
9b64f57d | 813 | refcount_inc(&cur_trans->use_count); |
8cd2807f | 814 | ret = 0; |
46204592 SW |
815 | break; |
816 | } | |
8cd2807f MX |
817 | if (t->transid > transid) { |
818 | ret = 0; | |
46204592 | 819 | break; |
8cd2807f | 820 | } |
46204592 | 821 | } |
0b246afa | 822 | spin_unlock(&fs_info->trans_lock); |
42383020 SW |
823 | |
824 | /* | |
825 | * The specified transaction doesn't exist, or we | |
826 | * raced with btrfs_commit_transaction | |
827 | */ | |
828 | if (!cur_trans) { | |
0b246afa | 829 | if (transid > fs_info->last_trans_committed) |
42383020 | 830 | ret = -EINVAL; |
8cd2807f | 831 | goto out; |
42383020 | 832 | } |
46204592 SW |
833 | } else { |
834 | /* find newest transaction that is committing | committed */ | |
0b246afa JM |
835 | spin_lock(&fs_info->trans_lock); |
836 | list_for_each_entry_reverse(t, &fs_info->trans_list, | |
46204592 | 837 | list) { |
4a9d8bde MX |
838 | if (t->state >= TRANS_STATE_COMMIT_START) { |
839 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 840 | break; |
46204592 | 841 | cur_trans = t; |
9b64f57d | 842 | refcount_inc(&cur_trans->use_count); |
46204592 SW |
843 | break; |
844 | } | |
845 | } | |
0b246afa | 846 | spin_unlock(&fs_info->trans_lock); |
46204592 | 847 | if (!cur_trans) |
a4abeea4 | 848 | goto out; /* nothing committing|committed */ |
46204592 SW |
849 | } |
850 | ||
2ff7e61e | 851 | wait_for_commit(cur_trans); |
724e2315 | 852 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 853 | out: |
46204592 SW |
854 | return ret; |
855 | } | |
856 | ||
2ff7e61e | 857 | void btrfs_throttle(struct btrfs_fs_info *fs_info) |
37d1aeee | 858 | { |
92e2f7e3 | 859 | wait_current_trans(fs_info); |
37d1aeee CM |
860 | } |
861 | ||
2ff7e61e | 862 | static int should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa | 863 | { |
2ff7e61e | 864 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0b246afa | 865 | |
64403612 | 866 | if (btrfs_check_space_for_delayed_refs(fs_info)) |
1be41b78 | 867 | return 1; |
36ba022a | 868 | |
2ff7e61e | 869 | return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); |
8929ecfa YZ |
870 | } |
871 | ||
3a45bb20 | 872 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa YZ |
873 | { |
874 | struct btrfs_transaction *cur_trans = trans->transaction; | |
8929ecfa | 875 | |
a4abeea4 | 876 | smp_mb(); |
3296bf56 | 877 | if (cur_trans->state >= TRANS_STATE_COMMIT_START || |
4a9d8bde | 878 | cur_trans->delayed_refs.flushing) |
8929ecfa YZ |
879 | return 1; |
880 | ||
2ff7e61e | 881 | return should_end_transaction(trans); |
8929ecfa YZ |
882 | } |
883 | ||
dc60c525 NB |
884 | static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) |
885 | ||
0e34693f | 886 | { |
dc60c525 NB |
887 | struct btrfs_fs_info *fs_info = trans->fs_info; |
888 | ||
0e34693f NB |
889 | if (!trans->block_rsv) { |
890 | ASSERT(!trans->bytes_reserved); | |
891 | return; | |
892 | } | |
893 | ||
894 | if (!trans->bytes_reserved) | |
895 | return; | |
896 | ||
897 | ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); | |
898 | trace_btrfs_space_reservation(fs_info, "transaction", | |
899 | trans->transid, trans->bytes_reserved, 0); | |
900 | btrfs_block_rsv_release(fs_info, trans->block_rsv, | |
901 | trans->bytes_reserved); | |
902 | trans->bytes_reserved = 0; | |
903 | } | |
904 | ||
89ce8a63 | 905 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
3a45bb20 | 906 | int throttle) |
79154b1b | 907 | { |
3a45bb20 | 908 | struct btrfs_fs_info *info = trans->fs_info; |
8929ecfa | 909 | struct btrfs_transaction *cur_trans = trans->transaction; |
4edc2ca3 | 910 | int err = 0; |
c3e69d58 | 911 | |
b50fff81 DS |
912 | if (refcount_read(&trans->use_count) > 1) { |
913 | refcount_dec(&trans->use_count); | |
2a1eb461 JB |
914 | trans->block_rsv = trans->orig_rsv; |
915 | return 0; | |
916 | } | |
917 | ||
dc60c525 | 918 | btrfs_trans_release_metadata(trans); |
4c13d758 | 919 | trans->block_rsv = NULL; |
c5567237 | 920 | |
119e80df | 921 | btrfs_create_pending_block_groups(trans); |
ea658bad | 922 | |
4fbcdf66 FM |
923 | btrfs_trans_release_chunk_metadata(trans); |
924 | ||
0860adfd | 925 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 926 | sb_end_intwrite(info->sb); |
6df7881a | 927 | |
8929ecfa | 928 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
929 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
930 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 931 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 932 | |
093258e6 | 933 | cond_wake_up(&cur_trans->writer_wait); |
724e2315 | 934 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
935 | |
936 | if (current->journal_info == trans) | |
937 | current->journal_info = NULL; | |
ab78c84d | 938 | |
24bbcf04 | 939 | if (throttle) |
2ff7e61e | 940 | btrfs_run_delayed_iputs(info); |
24bbcf04 | 941 | |
bf31f87f | 942 | if (TRANS_ABORTED(trans) || |
0b246afa | 943 | test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) { |
4e121c06 | 944 | wake_up_process(info->transaction_kthread); |
4edc2ca3 | 945 | err = -EIO; |
4e121c06 | 946 | } |
49b25e05 | 947 | |
4edc2ca3 DJ |
948 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
949 | return err; | |
79154b1b CM |
950 | } |
951 | ||
3a45bb20 | 952 | int btrfs_end_transaction(struct btrfs_trans_handle *trans) |
89ce8a63 | 953 | { |
3a45bb20 | 954 | return __btrfs_end_transaction(trans, 0); |
89ce8a63 CM |
955 | } |
956 | ||
3a45bb20 | 957 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) |
89ce8a63 | 958 | { |
3a45bb20 | 959 | return __btrfs_end_transaction(trans, 1); |
16cdcec7 MX |
960 | } |
961 | ||
d352ac68 CM |
962 | /* |
963 | * when btree blocks are allocated, they have some corresponding bits set for | |
964 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 965 | * those extents are sent to disk but does not wait on them |
d352ac68 | 966 | */ |
2ff7e61e | 967 | int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, |
8cef4e16 | 968 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 969 | { |
777e6bd7 | 970 | int err = 0; |
7c4452b9 | 971 | int werr = 0; |
0b246afa | 972 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 973 | struct extent_state *cached_state = NULL; |
777e6bd7 | 974 | u64 start = 0; |
5f39d397 | 975 | u64 end; |
7c4452b9 | 976 | |
6300463b | 977 | atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
1728366e | 978 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 979 | mark, &cached_state)) { |
663dfbb0 FM |
980 | bool wait_writeback = false; |
981 | ||
982 | err = convert_extent_bit(dirty_pages, start, end, | |
983 | EXTENT_NEED_WAIT, | |
210aa277 | 984 | mark, &cached_state); |
663dfbb0 FM |
985 | /* |
986 | * convert_extent_bit can return -ENOMEM, which is most of the | |
987 | * time a temporary error. So when it happens, ignore the error | |
988 | * and wait for writeback of this range to finish - because we | |
989 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
bf89d38f JM |
990 | * to __btrfs_wait_marked_extents() would not know that |
991 | * writeback for this range started and therefore wouldn't | |
992 | * wait for it to finish - we don't want to commit a | |
993 | * superblock that points to btree nodes/leafs for which | |
994 | * writeback hasn't finished yet (and without errors). | |
663dfbb0 | 995 | * We cleanup any entries left in the io tree when committing |
41e7acd3 | 996 | * the transaction (through extent_io_tree_release()). |
663dfbb0 FM |
997 | */ |
998 | if (err == -ENOMEM) { | |
999 | err = 0; | |
1000 | wait_writeback = true; | |
1001 | } | |
1002 | if (!err) | |
1003 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
1004 | if (err) |
1005 | werr = err; | |
663dfbb0 FM |
1006 | else if (wait_writeback) |
1007 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 1008 | free_extent_state(cached_state); |
663dfbb0 | 1009 | cached_state = NULL; |
1728366e JB |
1010 | cond_resched(); |
1011 | start = end + 1; | |
7c4452b9 | 1012 | } |
6300463b | 1013 | atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
690587d1 CM |
1014 | return werr; |
1015 | } | |
1016 | ||
1017 | /* | |
1018 | * when btree blocks are allocated, they have some corresponding bits set for | |
1019 | * them in one of two extent_io trees. This is used to make sure all of | |
1020 | * those extents are on disk for transaction or log commit. We wait | |
1021 | * on all the pages and clear them from the dirty pages state tree | |
1022 | */ | |
bf89d38f JM |
1023 | static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, |
1024 | struct extent_io_tree *dirty_pages) | |
690587d1 | 1025 | { |
690587d1 CM |
1026 | int err = 0; |
1027 | int werr = 0; | |
0b246afa | 1028 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 1029 | struct extent_state *cached_state = NULL; |
690587d1 CM |
1030 | u64 start = 0; |
1031 | u64 end; | |
777e6bd7 | 1032 | |
1728366e | 1033 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 1034 | EXTENT_NEED_WAIT, &cached_state)) { |
663dfbb0 FM |
1035 | /* |
1036 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
1037 | * When committing the transaction, we'll remove any entries | |
1038 | * left in the io tree. For a log commit, we don't remove them | |
1039 | * after committing the log because the tree can be accessed | |
1040 | * concurrently - we do it only at transaction commit time when | |
41e7acd3 | 1041 | * it's safe to do it (through extent_io_tree_release()). |
663dfbb0 FM |
1042 | */ |
1043 | err = clear_extent_bit(dirty_pages, start, end, | |
ae0f1625 | 1044 | EXTENT_NEED_WAIT, 0, 0, &cached_state); |
663dfbb0 FM |
1045 | if (err == -ENOMEM) |
1046 | err = 0; | |
1047 | if (!err) | |
1048 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
1049 | if (err) |
1050 | werr = err; | |
e38e2ed7 FM |
1051 | free_extent_state(cached_state); |
1052 | cached_state = NULL; | |
1728366e JB |
1053 | cond_resched(); |
1054 | start = end + 1; | |
777e6bd7 | 1055 | } |
7c4452b9 CM |
1056 | if (err) |
1057 | werr = err; | |
bf89d38f JM |
1058 | return werr; |
1059 | } | |
656f30db | 1060 | |
b9fae2eb | 1061 | static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, |
bf89d38f JM |
1062 | struct extent_io_tree *dirty_pages) |
1063 | { | |
1064 | bool errors = false; | |
1065 | int err; | |
656f30db | 1066 | |
bf89d38f JM |
1067 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); |
1068 | if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) | |
1069 | errors = true; | |
1070 | ||
1071 | if (errors && !err) | |
1072 | err = -EIO; | |
1073 | return err; | |
1074 | } | |
656f30db | 1075 | |
bf89d38f JM |
1076 | int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) |
1077 | { | |
1078 | struct btrfs_fs_info *fs_info = log_root->fs_info; | |
1079 | struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; | |
1080 | bool errors = false; | |
1081 | int err; | |
656f30db | 1082 | |
bf89d38f JM |
1083 | ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
1084 | ||
1085 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); | |
1086 | if ((mark & EXTENT_DIRTY) && | |
1087 | test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) | |
1088 | errors = true; | |
1089 | ||
1090 | if ((mark & EXTENT_NEW) && | |
1091 | test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) | |
1092 | errors = true; | |
1093 | ||
1094 | if (errors && !err) | |
1095 | err = -EIO; | |
1096 | return err; | |
79154b1b CM |
1097 | } |
1098 | ||
690587d1 | 1099 | /* |
c9b577c0 NB |
1100 | * When btree blocks are allocated the corresponding extents are marked dirty. |
1101 | * This function ensures such extents are persisted on disk for transaction or | |
1102 | * log commit. | |
1103 | * | |
1104 | * @trans: transaction whose dirty pages we'd like to write | |
690587d1 | 1105 | */ |
70458a58 | 1106 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) |
690587d1 CM |
1107 | { |
1108 | int ret; | |
1109 | int ret2; | |
c9b577c0 | 1110 | struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; |
70458a58 | 1111 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c6adc9cc | 1112 | struct blk_plug plug; |
690587d1 | 1113 | |
c6adc9cc | 1114 | blk_start_plug(&plug); |
c9b577c0 | 1115 | ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); |
c6adc9cc | 1116 | blk_finish_plug(&plug); |
bf89d38f | 1117 | ret2 = btrfs_wait_extents(fs_info, dirty_pages); |
bf0da8c1 | 1118 | |
41e7acd3 | 1119 | extent_io_tree_release(&trans->transaction->dirty_pages); |
c9b577c0 | 1120 | |
bf0da8c1 CM |
1121 | if (ret) |
1122 | return ret; | |
c9b577c0 | 1123 | else if (ret2) |
bf0da8c1 | 1124 | return ret2; |
c9b577c0 NB |
1125 | else |
1126 | return 0; | |
d0c803c4 CM |
1127 | } |
1128 | ||
d352ac68 CM |
1129 | /* |
1130 | * this is used to update the root pointer in the tree of tree roots. | |
1131 | * | |
1132 | * But, in the case of the extent allocation tree, updating the root | |
1133 | * pointer may allocate blocks which may change the root of the extent | |
1134 | * allocation tree. | |
1135 | * | |
1136 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1137 | * change while the root pointer was being updated in the metadata. | |
1138 | */ | |
0b86a832 CM |
1139 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1140 | struct btrfs_root *root) | |
79154b1b CM |
1141 | { |
1142 | int ret; | |
0b86a832 | 1143 | u64 old_root_bytenr; |
86b9f2ec | 1144 | u64 old_root_used; |
0b246afa JM |
1145 | struct btrfs_fs_info *fs_info = root->fs_info; |
1146 | struct btrfs_root *tree_root = fs_info->tree_root; | |
79154b1b | 1147 | |
86b9f2ec | 1148 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1149 | |
d397712b | 1150 | while (1) { |
0b86a832 | 1151 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1152 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1153 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1154 | break; |
87ef2bb4 | 1155 | |
5d4f98a2 | 1156 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1157 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1158 | &root->root_key, |
1159 | &root->root_item); | |
49b25e05 JM |
1160 | if (ret) |
1161 | return ret; | |
56bec294 | 1162 | |
86b9f2ec | 1163 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1164 | } |
276e680d | 1165 | |
0b86a832 CM |
1166 | return 0; |
1167 | } | |
1168 | ||
d352ac68 CM |
1169 | /* |
1170 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1171 | * |
1172 | * The error handling in this function may not be obvious. Any of the | |
1173 | * failures will cause the file system to go offline. We still need | |
1174 | * to clean up the delayed refs. | |
d352ac68 | 1175 | */ |
9386d8bc | 1176 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) |
0b86a832 | 1177 | { |
9386d8bc | 1178 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ea526d18 | 1179 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1180 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1181 | struct list_head *next; |
84234f3a | 1182 | struct extent_buffer *eb; |
56bec294 | 1183 | int ret; |
84234f3a YZ |
1184 | |
1185 | eb = btrfs_lock_root_node(fs_info->tree_root); | |
49b25e05 JM |
1186 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
1187 | 0, &eb); | |
84234f3a YZ |
1188 | btrfs_tree_unlock(eb); |
1189 | free_extent_buffer(eb); | |
0b86a832 | 1190 | |
49b25e05 JM |
1191 | if (ret) |
1192 | return ret; | |
1193 | ||
c79a70b1 | 1194 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
49b25e05 JM |
1195 | if (ret) |
1196 | return ret; | |
87ef2bb4 | 1197 | |
196c9d8d | 1198 | ret = btrfs_run_dev_stats(trans); |
c16ce190 JB |
1199 | if (ret) |
1200 | return ret; | |
2b584c68 | 1201 | ret = btrfs_run_dev_replace(trans); |
c16ce190 JB |
1202 | if (ret) |
1203 | return ret; | |
280f8bd2 | 1204 | ret = btrfs_run_qgroups(trans); |
c16ce190 JB |
1205 | if (ret) |
1206 | return ret; | |
546adb0d | 1207 | |
bbebb3e0 | 1208 | ret = btrfs_setup_space_cache(trans); |
dcdf7f6d JB |
1209 | if (ret) |
1210 | return ret; | |
1211 | ||
546adb0d | 1212 | /* run_qgroups might have added some more refs */ |
c79a70b1 | 1213 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
c16ce190 JB |
1214 | if (ret) |
1215 | return ret; | |
ea526d18 | 1216 | again: |
d397712b | 1217 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
2ff7e61e | 1218 | struct btrfs_root *root; |
0b86a832 CM |
1219 | next = fs_info->dirty_cowonly_roots.next; |
1220 | list_del_init(next); | |
1221 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1222 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1223 | |
9e351cc8 JB |
1224 | if (root != fs_info->extent_root) |
1225 | list_add_tail(&root->dirty_list, | |
1226 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1227 | ret = update_cowonly_root(trans, root); |
1228 | if (ret) | |
1229 | return ret; | |
c79a70b1 | 1230 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
ea526d18 JB |
1231 | if (ret) |
1232 | return ret; | |
79154b1b | 1233 | } |
276e680d | 1234 | |
1bbc621e | 1235 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
5742d15f | 1236 | ret = btrfs_write_dirty_block_groups(trans); |
ea526d18 JB |
1237 | if (ret) |
1238 | return ret; | |
c79a70b1 | 1239 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
ea526d18 JB |
1240 | if (ret) |
1241 | return ret; | |
1242 | } | |
1243 | ||
1244 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1245 | goto again; | |
1246 | ||
9e351cc8 JB |
1247 | list_add_tail(&fs_info->extent_root->dirty_list, |
1248 | &trans->transaction->switch_commits); | |
9f6cbcbb DS |
1249 | |
1250 | /* Update dev-replace pointer once everything is committed */ | |
1251 | fs_info->dev_replace.committed_cursor_left = | |
1252 | fs_info->dev_replace.cursor_left_last_write_of_item; | |
8dabb742 | 1253 | |
79154b1b CM |
1254 | return 0; |
1255 | } | |
1256 | ||
d352ac68 CM |
1257 | /* |
1258 | * dead roots are old snapshots that need to be deleted. This allocates | |
1259 | * a dirty root struct and adds it into the list of dead roots that need to | |
1260 | * be deleted | |
1261 | */ | |
cfad392b | 1262 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1263 | { |
0b246afa JM |
1264 | struct btrfs_fs_info *fs_info = root->fs_info; |
1265 | ||
1266 | spin_lock(&fs_info->trans_lock); | |
cfad392b | 1267 | if (list_empty(&root->root_list)) |
0b246afa JM |
1268 | list_add_tail(&root->root_list, &fs_info->dead_roots); |
1269 | spin_unlock(&fs_info->trans_lock); | |
5eda7b5e CM |
1270 | } |
1271 | ||
d352ac68 | 1272 | /* |
5d4f98a2 | 1273 | * update all the cowonly tree roots on disk |
d352ac68 | 1274 | */ |
7e4443d9 | 1275 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) |
0f7d52f4 | 1276 | { |
7e4443d9 | 1277 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0f7d52f4 | 1278 | struct btrfs_root *gang[8]; |
0f7d52f4 CM |
1279 | int i; |
1280 | int ret; | |
54aa1f4d CM |
1281 | int err = 0; |
1282 | ||
a4abeea4 | 1283 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1284 | while (1) { |
5d4f98a2 YZ |
1285 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1286 | (void **)gang, 0, | |
0f7d52f4 CM |
1287 | ARRAY_SIZE(gang), |
1288 | BTRFS_ROOT_TRANS_TAG); | |
1289 | if (ret == 0) | |
1290 | break; | |
1291 | for (i = 0; i < ret; i++) { | |
5b4aacef | 1292 | struct btrfs_root *root = gang[i]; |
5d4f98a2 YZ |
1293 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1294 | (unsigned long)root->root_key.objectid, | |
1295 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1296 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1297 | |
e02119d5 | 1298 | btrfs_free_log(trans, root); |
5d4f98a2 | 1299 | btrfs_update_reloc_root(trans, root); |
bcc63abb | 1300 | |
82d5902d LZ |
1301 | btrfs_save_ino_cache(root, trans); |
1302 | ||
f1ebcc74 | 1303 | /* see comments in should_cow_block() */ |
27cdeb70 | 1304 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
c7548af6 | 1305 | smp_mb__after_atomic(); |
f1ebcc74 | 1306 | |
978d910d | 1307 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1308 | list_add_tail(&root->dirty_list, |
1309 | &trans->transaction->switch_commits); | |
978d910d YZ |
1310 | btrfs_set_root_node(&root->root_item, |
1311 | root->node); | |
1312 | } | |
5d4f98a2 | 1313 | |
5d4f98a2 | 1314 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1315 | &root->root_key, |
1316 | &root->root_item); | |
a4abeea4 | 1317 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1318 | if (err) |
1319 | break; | |
733e03a0 | 1320 | btrfs_qgroup_free_meta_all_pertrans(root); |
0f7d52f4 CM |
1321 | } |
1322 | } | |
a4abeea4 | 1323 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1324 | return err; |
0f7d52f4 CM |
1325 | } |
1326 | ||
d352ac68 | 1327 | /* |
de78b51a ES |
1328 | * defrag a given btree. |
1329 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1330 | */ |
de78b51a | 1331 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1332 | { |
1333 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1334 | struct btrfs_trans_handle *trans; |
8929ecfa | 1335 | int ret; |
e9d0b13b | 1336 | |
27cdeb70 | 1337 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1338 | return 0; |
8929ecfa | 1339 | |
6b80053d | 1340 | while (1) { |
8929ecfa YZ |
1341 | trans = btrfs_start_transaction(root, 0); |
1342 | if (IS_ERR(trans)) | |
1343 | return PTR_ERR(trans); | |
1344 | ||
de78b51a | 1345 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1346 | |
3a45bb20 | 1347 | btrfs_end_transaction(trans); |
2ff7e61e | 1348 | btrfs_btree_balance_dirty(info); |
e9d0b13b CM |
1349 | cond_resched(); |
1350 | ||
ab8d0fc4 | 1351 | if (btrfs_fs_closing(info) || ret != -EAGAIN) |
e9d0b13b | 1352 | break; |
210549eb | 1353 | |
ab8d0fc4 JM |
1354 | if (btrfs_defrag_cancelled(info)) { |
1355 | btrfs_debug(info, "defrag_root cancelled"); | |
210549eb DS |
1356 | ret = -EAGAIN; |
1357 | break; | |
1358 | } | |
e9d0b13b | 1359 | } |
27cdeb70 | 1360 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1361 | return ret; |
e9d0b13b CM |
1362 | } |
1363 | ||
6426c7ad QW |
1364 | /* |
1365 | * Do all special snapshot related qgroup dirty hack. | |
1366 | * | |
1367 | * Will do all needed qgroup inherit and dirty hack like switch commit | |
1368 | * roots inside one transaction and write all btree into disk, to make | |
1369 | * qgroup works. | |
1370 | */ | |
1371 | static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, | |
1372 | struct btrfs_root *src, | |
1373 | struct btrfs_root *parent, | |
1374 | struct btrfs_qgroup_inherit *inherit, | |
1375 | u64 dst_objectid) | |
1376 | { | |
1377 | struct btrfs_fs_info *fs_info = src->fs_info; | |
1378 | int ret; | |
1379 | ||
1380 | /* | |
1381 | * Save some performance in the case that qgroups are not | |
1382 | * enabled. If this check races with the ioctl, rescan will | |
1383 | * kick in anyway. | |
1384 | */ | |
9ea6e2b5 | 1385 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
6426c7ad | 1386 | return 0; |
6426c7ad | 1387 | |
4d31778a | 1388 | /* |
52042d8e | 1389 | * Ensure dirty @src will be committed. Or, after coming |
4d31778a QW |
1390 | * commit_fs_roots() and switch_commit_roots(), any dirty but not |
1391 | * recorded root will never be updated again, causing an outdated root | |
1392 | * item. | |
1393 | */ | |
1394 | record_root_in_trans(trans, src, 1); | |
1395 | ||
6426c7ad QW |
1396 | /* |
1397 | * We are going to commit transaction, see btrfs_commit_transaction() | |
1398 | * comment for reason locking tree_log_mutex | |
1399 | */ | |
1400 | mutex_lock(&fs_info->tree_log_mutex); | |
1401 | ||
7e4443d9 | 1402 | ret = commit_fs_roots(trans); |
6426c7ad QW |
1403 | if (ret) |
1404 | goto out; | |
460fb20a | 1405 | ret = btrfs_qgroup_account_extents(trans); |
6426c7ad QW |
1406 | if (ret < 0) |
1407 | goto out; | |
1408 | ||
1409 | /* Now qgroup are all updated, we can inherit it to new qgroups */ | |
a9377422 | 1410 | ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, |
6426c7ad QW |
1411 | inherit); |
1412 | if (ret < 0) | |
1413 | goto out; | |
1414 | ||
1415 | /* | |
1416 | * Now we do a simplified commit transaction, which will: | |
1417 | * 1) commit all subvolume and extent tree | |
1418 | * To ensure all subvolume and extent tree have a valid | |
1419 | * commit_root to accounting later insert_dir_item() | |
1420 | * 2) write all btree blocks onto disk | |
1421 | * This is to make sure later btree modification will be cowed | |
1422 | * Or commit_root can be populated and cause wrong qgroup numbers | |
1423 | * In this simplified commit, we don't really care about other trees | |
1424 | * like chunk and root tree, as they won't affect qgroup. | |
1425 | * And we don't write super to avoid half committed status. | |
1426 | */ | |
9386d8bc | 1427 | ret = commit_cowonly_roots(trans); |
6426c7ad QW |
1428 | if (ret) |
1429 | goto out; | |
889bfa39 | 1430 | switch_commit_roots(trans); |
70458a58 | 1431 | ret = btrfs_write_and_wait_transaction(trans); |
6426c7ad | 1432 | if (ret) |
f7af3934 | 1433 | btrfs_handle_fs_error(fs_info, ret, |
6426c7ad QW |
1434 | "Error while writing out transaction for qgroup"); |
1435 | ||
1436 | out: | |
1437 | mutex_unlock(&fs_info->tree_log_mutex); | |
1438 | ||
1439 | /* | |
1440 | * Force parent root to be updated, as we recorded it before so its | |
1441 | * last_trans == cur_transid. | |
1442 | * Or it won't be committed again onto disk after later | |
1443 | * insert_dir_item() | |
1444 | */ | |
1445 | if (!ret) | |
1446 | record_root_in_trans(trans, parent, 1); | |
1447 | return ret; | |
1448 | } | |
1449 | ||
d352ac68 CM |
1450 | /* |
1451 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1452 | * transaction commit. This does the actual creation. |
1453 | * | |
1454 | * Note: | |
1455 | * If the error which may affect the commitment of the current transaction | |
1456 | * happens, we should return the error number. If the error which just affect | |
1457 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1458 | */ |
80b6794d | 1459 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1460 | struct btrfs_pending_snapshot *pending) |
1461 | { | |
08d50ca3 NB |
1462 | |
1463 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3063d29f | 1464 | struct btrfs_key key; |
80b6794d | 1465 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1466 | struct btrfs_root *tree_root = fs_info->tree_root; |
1467 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1468 | struct btrfs_root *parent_root; |
98c9942a | 1469 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1470 | struct inode *parent_inode; |
42874b3d MX |
1471 | struct btrfs_path *path; |
1472 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1473 | struct dentry *dentry; |
3063d29f | 1474 | struct extent_buffer *tmp; |
925baedd | 1475 | struct extent_buffer *old; |
95582b00 | 1476 | struct timespec64 cur_time; |
aec8030a | 1477 | int ret = 0; |
d68fc57b | 1478 | u64 to_reserve = 0; |
6bdb72de | 1479 | u64 index = 0; |
a22285a6 | 1480 | u64 objectid; |
b83cc969 | 1481 | u64 root_flags; |
8ea05e3a | 1482 | uuid_le new_uuid; |
3063d29f | 1483 | |
8546b570 DS |
1484 | ASSERT(pending->path); |
1485 | path = pending->path; | |
42874b3d | 1486 | |
b0c0ea63 DS |
1487 | ASSERT(pending->root_item); |
1488 | new_root_item = pending->root_item; | |
a22285a6 | 1489 | |
aec8030a MX |
1490 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1491 | if (pending->error) | |
6fa9700e | 1492 | goto no_free_objectid; |
3063d29f | 1493 | |
d6726335 QW |
1494 | /* |
1495 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1496 | * accounted by later btrfs_qgroup_inherit(). | |
1497 | */ | |
1498 | btrfs_set_skip_qgroup(trans, objectid); | |
1499 | ||
147d256e | 1500 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
d68fc57b YZ |
1501 | |
1502 | if (to_reserve > 0) { | |
aec8030a MX |
1503 | pending->error = btrfs_block_rsv_add(root, |
1504 | &pending->block_rsv, | |
1505 | to_reserve, | |
1506 | BTRFS_RESERVE_NO_FLUSH); | |
1507 | if (pending->error) | |
d6726335 | 1508 | goto clear_skip_qgroup; |
d68fc57b YZ |
1509 | } |
1510 | ||
3063d29f | 1511 | key.objectid = objectid; |
a22285a6 YZ |
1512 | key.offset = (u64)-1; |
1513 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1514 | |
6fa9700e | 1515 | rsv = trans->block_rsv; |
a22285a6 | 1516 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1517 | trans->bytes_reserved = trans->block_rsv->reserved; |
0b246afa | 1518 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa JB |
1519 | trans->transid, |
1520 | trans->bytes_reserved, 1); | |
a22285a6 | 1521 | dentry = pending->dentry; |
e9662f70 | 1522 | parent_inode = pending->dir; |
a22285a6 | 1523 | parent_root = BTRFS_I(parent_inode)->root; |
6426c7ad | 1524 | record_root_in_trans(trans, parent_root, 0); |
a22285a6 | 1525 | |
c2050a45 | 1526 | cur_time = current_time(parent_inode); |
04b285f3 | 1527 | |
3063d29f CM |
1528 | /* |
1529 | * insert the directory item | |
1530 | */ | |
877574e2 | 1531 | ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); |
49b25e05 | 1532 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1533 | |
1534 | /* check if there is a file/dir which has the same name. */ | |
1535 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
4a0cc7ca | 1536 | btrfs_ino(BTRFS_I(parent_inode)), |
42874b3d MX |
1537 | dentry->d_name.name, |
1538 | dentry->d_name.len, 0); | |
1539 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1540 | pending->error = -EEXIST; |
aec8030a | 1541 | goto dir_item_existed; |
42874b3d MX |
1542 | } else if (IS_ERR(dir_item)) { |
1543 | ret = PTR_ERR(dir_item); | |
66642832 | 1544 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1545 | goto fail; |
79787eaa | 1546 | } |
42874b3d | 1547 | btrfs_release_path(path); |
52c26179 | 1548 | |
e999376f CM |
1549 | /* |
1550 | * pull in the delayed directory update | |
1551 | * and the delayed inode item | |
1552 | * otherwise we corrupt the FS during | |
1553 | * snapshot | |
1554 | */ | |
e5c304e6 | 1555 | ret = btrfs_run_delayed_items(trans); |
8732d44f | 1556 | if (ret) { /* Transaction aborted */ |
66642832 | 1557 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1558 | goto fail; |
1559 | } | |
e999376f | 1560 | |
6426c7ad | 1561 | record_root_in_trans(trans, root, 0); |
6bdb72de SW |
1562 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1563 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1564 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1565 | |
b83cc969 LZ |
1566 | root_flags = btrfs_root_flags(new_root_item); |
1567 | if (pending->readonly) | |
1568 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1569 | else | |
1570 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1571 | btrfs_set_root_flags(new_root_item, root_flags); | |
1572 | ||
8ea05e3a AB |
1573 | btrfs_set_root_generation_v2(new_root_item, |
1574 | trans->transid); | |
1575 | uuid_le_gen(&new_uuid); | |
1576 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1577 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1578 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1579 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1580 | memset(new_root_item->received_uuid, 0, | |
1581 | sizeof(new_root_item->received_uuid)); | |
1582 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1583 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1584 | btrfs_set_root_stransid(new_root_item, 0); | |
1585 | btrfs_set_root_rtransid(new_root_item, 0); | |
1586 | } | |
3cae210f QW |
1587 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1588 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1589 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1590 | |
6bdb72de | 1591 | old = btrfs_lock_root_node(root); |
49b25e05 | 1592 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1593 | if (ret) { |
1594 | btrfs_tree_unlock(old); | |
1595 | free_extent_buffer(old); | |
66642832 | 1596 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1597 | goto fail; |
79787eaa | 1598 | } |
49b25e05 | 1599 | |
8bead258 | 1600 | btrfs_set_lock_blocking_write(old); |
6bdb72de | 1601 | |
49b25e05 | 1602 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1603 | /* clean up in any case */ |
6bdb72de SW |
1604 | btrfs_tree_unlock(old); |
1605 | free_extent_buffer(old); | |
8732d44f | 1606 | if (ret) { |
66642832 | 1607 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1608 | goto fail; |
1609 | } | |
f1ebcc74 | 1610 | /* see comments in should_cow_block() */ |
27cdeb70 | 1611 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1612 | smp_wmb(); |
1613 | ||
6bdb72de | 1614 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1615 | /* record when the snapshot was created in key.offset */ |
1616 | key.offset = trans->transid; | |
1617 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1618 | btrfs_tree_unlock(tmp); |
1619 | free_extent_buffer(tmp); | |
8732d44f | 1620 | if (ret) { |
66642832 | 1621 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1622 | goto fail; |
1623 | } | |
6bdb72de | 1624 | |
a22285a6 YZ |
1625 | /* |
1626 | * insert root back/forward references | |
1627 | */ | |
6025c19f | 1628 | ret = btrfs_add_root_ref(trans, objectid, |
0660b5af | 1629 | parent_root->root_key.objectid, |
4a0cc7ca | 1630 | btrfs_ino(BTRFS_I(parent_inode)), index, |
a22285a6 | 1631 | dentry->d_name.name, dentry->d_name.len); |
8732d44f | 1632 | if (ret) { |
66642832 | 1633 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1634 | goto fail; |
1635 | } | |
0660b5af | 1636 | |
a22285a6 | 1637 | key.offset = (u64)-1; |
3619c94f | 1638 | pending->snap = btrfs_get_fs_root(fs_info, &key, true); |
79787eaa JM |
1639 | if (IS_ERR(pending->snap)) { |
1640 | ret = PTR_ERR(pending->snap); | |
66642832 | 1641 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1642 | goto fail; |
79787eaa | 1643 | } |
d68fc57b | 1644 | |
49b25e05 | 1645 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f | 1646 | if (ret) { |
66642832 | 1647 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1648 | goto fail; |
1649 | } | |
361048f5 | 1650 | |
c79a70b1 | 1651 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
8732d44f | 1652 | if (ret) { |
66642832 | 1653 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1654 | goto fail; |
1655 | } | |
42874b3d | 1656 | |
6426c7ad QW |
1657 | /* |
1658 | * Do special qgroup accounting for snapshot, as we do some qgroup | |
1659 | * snapshot hack to do fast snapshot. | |
1660 | * To co-operate with that hack, we do hack again. | |
1661 | * Or snapshot will be greatly slowed down by a subtree qgroup rescan | |
1662 | */ | |
1663 | ret = qgroup_account_snapshot(trans, root, parent_root, | |
1664 | pending->inherit, objectid); | |
1665 | if (ret < 0) | |
1666 | goto fail; | |
1667 | ||
684572df LF |
1668 | ret = btrfs_insert_dir_item(trans, dentry->d_name.name, |
1669 | dentry->d_name.len, BTRFS_I(parent_inode), | |
1670 | &key, BTRFS_FT_DIR, index); | |
42874b3d | 1671 | /* We have check then name at the beginning, so it is impossible. */ |
9c52057c | 1672 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f | 1673 | if (ret) { |
66642832 | 1674 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1675 | goto fail; |
1676 | } | |
42874b3d | 1677 | |
6ef06d27 | 1678 | btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + |
42874b3d | 1679 | dentry->d_name.len * 2); |
04b285f3 | 1680 | parent_inode->i_mtime = parent_inode->i_ctime = |
c2050a45 | 1681 | current_time(parent_inode); |
be6aef60 | 1682 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
dd5f9615 | 1683 | if (ret) { |
66642832 | 1684 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1685 | goto fail; |
1686 | } | |
cdb345a8 LF |
1687 | ret = btrfs_uuid_tree_add(trans, new_uuid.b, BTRFS_UUID_KEY_SUBVOL, |
1688 | objectid); | |
dd5f9615 | 1689 | if (ret) { |
66642832 | 1690 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1691 | goto fail; |
1692 | } | |
1693 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
cdb345a8 | 1694 | ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, |
dd5f9615 SB |
1695 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, |
1696 | objectid); | |
1697 | if (ret && ret != -EEXIST) { | |
66642832 | 1698 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1699 | goto fail; |
1700 | } | |
1701 | } | |
d6726335 | 1702 | |
c79a70b1 | 1703 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
d6726335 | 1704 | if (ret) { |
66642832 | 1705 | btrfs_abort_transaction(trans, ret); |
d6726335 QW |
1706 | goto fail; |
1707 | } | |
1708 | ||
3063d29f | 1709 | fail: |
aec8030a MX |
1710 | pending->error = ret; |
1711 | dir_item_existed: | |
98c9942a | 1712 | trans->block_rsv = rsv; |
2382c5cc | 1713 | trans->bytes_reserved = 0; |
d6726335 QW |
1714 | clear_skip_qgroup: |
1715 | btrfs_clear_skip_qgroup(trans); | |
6fa9700e MX |
1716 | no_free_objectid: |
1717 | kfree(new_root_item); | |
b0c0ea63 | 1718 | pending->root_item = NULL; |
42874b3d | 1719 | btrfs_free_path(path); |
8546b570 DS |
1720 | pending->path = NULL; |
1721 | ||
49b25e05 | 1722 | return ret; |
3063d29f CM |
1723 | } |
1724 | ||
d352ac68 CM |
1725 | /* |
1726 | * create all the snapshots we've scheduled for creation | |
1727 | */ | |
08d50ca3 | 1728 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) |
3de4586c | 1729 | { |
aec8030a | 1730 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1731 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1732 | int ret = 0; |
3de4586c | 1733 | |
aec8030a MX |
1734 | list_for_each_entry_safe(pending, next, head, list) { |
1735 | list_del(&pending->list); | |
08d50ca3 | 1736 | ret = create_pending_snapshot(trans, pending); |
aec8030a MX |
1737 | if (ret) |
1738 | break; | |
1739 | } | |
1740 | return ret; | |
3de4586c CM |
1741 | } |
1742 | ||
2ff7e61e | 1743 | static void update_super_roots(struct btrfs_fs_info *fs_info) |
5d4f98a2 YZ |
1744 | { |
1745 | struct btrfs_root_item *root_item; | |
1746 | struct btrfs_super_block *super; | |
1747 | ||
0b246afa | 1748 | super = fs_info->super_copy; |
5d4f98a2 | 1749 | |
0b246afa | 1750 | root_item = &fs_info->chunk_root->root_item; |
093e037c DS |
1751 | super->chunk_root = root_item->bytenr; |
1752 | super->chunk_root_generation = root_item->generation; | |
1753 | super->chunk_root_level = root_item->level; | |
5d4f98a2 | 1754 | |
0b246afa | 1755 | root_item = &fs_info->tree_root->root_item; |
093e037c DS |
1756 | super->root = root_item->bytenr; |
1757 | super->generation = root_item->generation; | |
1758 | super->root_level = root_item->level; | |
0b246afa | 1759 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) |
093e037c | 1760 | super->cache_generation = root_item->generation; |
0b246afa | 1761 | if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) |
093e037c | 1762 | super->uuid_tree_generation = root_item->generation; |
5d4f98a2 YZ |
1763 | } |
1764 | ||
f36f3042 CM |
1765 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1766 | { | |
4a9d8bde | 1767 | struct btrfs_transaction *trans; |
f36f3042 | 1768 | int ret = 0; |
4a9d8bde | 1769 | |
a4abeea4 | 1770 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1771 | trans = info->running_transaction; |
1772 | if (trans) | |
1773 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1774 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1775 | return ret; |
1776 | } | |
1777 | ||
8929ecfa YZ |
1778 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1779 | { | |
4a9d8bde | 1780 | struct btrfs_transaction *trans; |
8929ecfa | 1781 | int ret = 0; |
4a9d8bde | 1782 | |
a4abeea4 | 1783 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1784 | trans = info->running_transaction; |
1785 | if (trans) | |
1786 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1787 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1788 | return ret; |
1789 | } | |
1790 | ||
bb9c12c9 SW |
1791 | /* |
1792 | * wait for the current transaction commit to start and block subsequent | |
1793 | * transaction joins | |
1794 | */ | |
2ff7e61e | 1795 | static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info, |
bb9c12c9 SW |
1796 | struct btrfs_transaction *trans) |
1797 | { | |
2ff7e61e | 1798 | wait_event(fs_info->transaction_blocked_wait, |
bf31f87f DS |
1799 | trans->state >= TRANS_STATE_COMMIT_START || |
1800 | TRANS_ABORTED(trans)); | |
bb9c12c9 SW |
1801 | } |
1802 | ||
1803 | /* | |
1804 | * wait for the current transaction to start and then become unblocked. | |
1805 | * caller holds ref. | |
1806 | */ | |
2ff7e61e JM |
1807 | static void wait_current_trans_commit_start_and_unblock( |
1808 | struct btrfs_fs_info *fs_info, | |
1809 | struct btrfs_transaction *trans) | |
bb9c12c9 | 1810 | { |
2ff7e61e | 1811 | wait_event(fs_info->transaction_wait, |
bf31f87f DS |
1812 | trans->state >= TRANS_STATE_UNBLOCKED || |
1813 | TRANS_ABORTED(trans)); | |
bb9c12c9 SW |
1814 | } |
1815 | ||
1816 | /* | |
1817 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1818 | * returns, any subsequent transaction will not be allowed to join. | |
1819 | */ | |
1820 | struct btrfs_async_commit { | |
1821 | struct btrfs_trans_handle *newtrans; | |
7892b5af | 1822 | struct work_struct work; |
bb9c12c9 SW |
1823 | }; |
1824 | ||
1825 | static void do_async_commit(struct work_struct *work) | |
1826 | { | |
1827 | struct btrfs_async_commit *ac = | |
7892b5af | 1828 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1829 | |
6fc4e354 SW |
1830 | /* |
1831 | * We've got freeze protection passed with the transaction. | |
1832 | * Tell lockdep about it. | |
1833 | */ | |
b1a06a4b | 1834 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
3a45bb20 | 1835 | __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1836 | |
e209db7a SW |
1837 | current->journal_info = ac->newtrans; |
1838 | ||
3a45bb20 | 1839 | btrfs_commit_transaction(ac->newtrans); |
bb9c12c9 SW |
1840 | kfree(ac); |
1841 | } | |
1842 | ||
1843 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
bb9c12c9 SW |
1844 | int wait_for_unblock) |
1845 | { | |
3a45bb20 | 1846 | struct btrfs_fs_info *fs_info = trans->fs_info; |
bb9c12c9 SW |
1847 | struct btrfs_async_commit *ac; |
1848 | struct btrfs_transaction *cur_trans; | |
1849 | ||
1850 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1851 | if (!ac) |
1852 | return -ENOMEM; | |
bb9c12c9 | 1853 | |
7892b5af | 1854 | INIT_WORK(&ac->work, do_async_commit); |
3a45bb20 | 1855 | ac->newtrans = btrfs_join_transaction(trans->root); |
3612b495 TI |
1856 | if (IS_ERR(ac->newtrans)) { |
1857 | int err = PTR_ERR(ac->newtrans); | |
1858 | kfree(ac); | |
1859 | return err; | |
1860 | } | |
bb9c12c9 SW |
1861 | |
1862 | /* take transaction reference */ | |
bb9c12c9 | 1863 | cur_trans = trans->transaction; |
9b64f57d | 1864 | refcount_inc(&cur_trans->use_count); |
bb9c12c9 | 1865 | |
3a45bb20 | 1866 | btrfs_end_transaction(trans); |
6fc4e354 SW |
1867 | |
1868 | /* | |
1869 | * Tell lockdep we've released the freeze rwsem, since the | |
1870 | * async commit thread will be the one to unlock it. | |
1871 | */ | |
b1a06a4b | 1872 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
0b246afa | 1873 | __sb_writers_release(fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1874 | |
7892b5af | 1875 | schedule_work(&ac->work); |
bb9c12c9 SW |
1876 | |
1877 | /* wait for transaction to start and unblock */ | |
bb9c12c9 | 1878 | if (wait_for_unblock) |
2ff7e61e | 1879 | wait_current_trans_commit_start_and_unblock(fs_info, cur_trans); |
bb9c12c9 | 1880 | else |
2ff7e61e | 1881 | wait_current_trans_commit_start(fs_info, cur_trans); |
bb9c12c9 | 1882 | |
38e88054 SW |
1883 | if (current->journal_info == trans) |
1884 | current->journal_info = NULL; | |
1885 | ||
724e2315 | 1886 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1887 | return 0; |
1888 | } | |
1889 | ||
49b25e05 | 1890 | |
97cb39bb | 1891 | static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) |
49b25e05 | 1892 | { |
97cb39bb | 1893 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 JM |
1894 | struct btrfs_transaction *cur_trans = trans->transaction; |
1895 | ||
b50fff81 | 1896 | WARN_ON(refcount_read(&trans->use_count) > 1); |
49b25e05 | 1897 | |
66642832 | 1898 | btrfs_abort_transaction(trans, err); |
7b8b92af | 1899 | |
0b246afa | 1900 | spin_lock(&fs_info->trans_lock); |
66b6135b | 1901 | |
25d8c284 MX |
1902 | /* |
1903 | * If the transaction is removed from the list, it means this | |
1904 | * transaction has been committed successfully, so it is impossible | |
1905 | * to call the cleanup function. | |
1906 | */ | |
1907 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1908 | |
49b25e05 | 1909 | list_del_init(&cur_trans->list); |
0b246afa | 1910 | if (cur_trans == fs_info->running_transaction) { |
4a9d8bde | 1911 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 1912 | spin_unlock(&fs_info->trans_lock); |
f094ac32 LB |
1913 | wait_event(cur_trans->writer_wait, |
1914 | atomic_read(&cur_trans->num_writers) == 1); | |
1915 | ||
0b246afa | 1916 | spin_lock(&fs_info->trans_lock); |
d7096fc3 | 1917 | } |
0b246afa | 1918 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 1919 | |
2ff7e61e | 1920 | btrfs_cleanup_one_transaction(trans->transaction, fs_info); |
49b25e05 | 1921 | |
0b246afa JM |
1922 | spin_lock(&fs_info->trans_lock); |
1923 | if (cur_trans == fs_info->running_transaction) | |
1924 | fs_info->running_transaction = NULL; | |
1925 | spin_unlock(&fs_info->trans_lock); | |
4a9d8bde | 1926 | |
e0228285 | 1927 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 1928 | sb_end_intwrite(fs_info->sb); |
724e2315 JB |
1929 | btrfs_put_transaction(cur_trans); |
1930 | btrfs_put_transaction(cur_trans); | |
49b25e05 | 1931 | |
97cb39bb | 1932 | trace_btrfs_transaction_commit(trans->root); |
49b25e05 | 1933 | |
49b25e05 JM |
1934 | if (current->journal_info == trans) |
1935 | current->journal_info = NULL; | |
0b246afa | 1936 | btrfs_scrub_cancel(fs_info); |
49b25e05 JM |
1937 | |
1938 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1939 | } | |
1940 | ||
c7cc64a9 DS |
1941 | /* |
1942 | * Release reserved delayed ref space of all pending block groups of the | |
1943 | * transaction and remove them from the list | |
1944 | */ | |
1945 | static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) | |
1946 | { | |
1947 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 1948 | struct btrfs_block_group *block_group, *tmp; |
c7cc64a9 DS |
1949 | |
1950 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { | |
1951 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1952 | list_del_init(&block_group->bg_list); | |
1953 | } | |
1954 | } | |
1955 | ||
609e804d | 1956 | static inline int btrfs_start_delalloc_flush(struct btrfs_trans_handle *trans) |
82436617 | 1957 | { |
609e804d FM |
1958 | struct btrfs_fs_info *fs_info = trans->fs_info; |
1959 | ||
ce8ea7cc JB |
1960 | /* |
1961 | * We use writeback_inodes_sb here because if we used | |
1962 | * btrfs_start_delalloc_roots we would deadlock with fs freeze. | |
1963 | * Currently are holding the fs freeze lock, if we do an async flush | |
1964 | * we'll do btrfs_join_transaction() and deadlock because we need to | |
1965 | * wait for the fs freeze lock. Using the direct flushing we benefit | |
1966 | * from already being in a transaction and our join_transaction doesn't | |
1967 | * have to re-take the fs freeze lock. | |
1968 | */ | |
609e804d | 1969 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) { |
ce8ea7cc | 1970 | writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); |
609e804d FM |
1971 | } else { |
1972 | struct btrfs_pending_snapshot *pending; | |
1973 | struct list_head *head = &trans->transaction->pending_snapshots; | |
1974 | ||
1975 | /* | |
1976 | * Flush dellaloc for any root that is going to be snapshotted. | |
1977 | * This is done to avoid a corrupted version of files, in the | |
1978 | * snapshots, that had both buffered and direct IO writes (even | |
1979 | * if they were done sequentially) due to an unordered update of | |
1980 | * the inode's size on disk. | |
1981 | */ | |
1982 | list_for_each_entry(pending, head, list) { | |
1983 | int ret; | |
1984 | ||
1985 | ret = btrfs_start_delalloc_snapshot(pending->root); | |
1986 | if (ret) | |
1987 | return ret; | |
1988 | } | |
1989 | } | |
82436617 MX |
1990 | return 0; |
1991 | } | |
1992 | ||
609e804d | 1993 | static inline void btrfs_wait_delalloc_flush(struct btrfs_trans_handle *trans) |
82436617 | 1994 | { |
609e804d FM |
1995 | struct btrfs_fs_info *fs_info = trans->fs_info; |
1996 | ||
1997 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) { | |
6374e57a | 1998 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); |
609e804d FM |
1999 | } else { |
2000 | struct btrfs_pending_snapshot *pending; | |
2001 | struct list_head *head = &trans->transaction->pending_snapshots; | |
2002 | ||
2003 | /* | |
2004 | * Wait for any dellaloc that we started previously for the roots | |
2005 | * that are going to be snapshotted. This is to avoid a corrupted | |
2006 | * version of files in the snapshots that had both buffered and | |
2007 | * direct IO writes (even if they were done sequentially). | |
2008 | */ | |
2009 | list_for_each_entry(pending, head, list) | |
2010 | btrfs_wait_ordered_extents(pending->root, | |
2011 | U64_MAX, 0, U64_MAX); | |
2012 | } | |
82436617 MX |
2013 | } |
2014 | ||
3a45bb20 | 2015 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans) |
79154b1b | 2016 | { |
3a45bb20 | 2017 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 | 2018 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 2019 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 2020 | int ret; |
79154b1b | 2021 | |
35b814f3 NB |
2022 | ASSERT(refcount_read(&trans->use_count) == 1); |
2023 | ||
d62b23c9 JB |
2024 | /* |
2025 | * Some places just start a transaction to commit it. We need to make | |
2026 | * sure that if this commit fails that the abort code actually marks the | |
2027 | * transaction as failed, so set trans->dirty to make the abort code do | |
2028 | * the right thing. | |
2029 | */ | |
2030 | trans->dirty = true; | |
2031 | ||
8d25a086 | 2032 | /* Stop the commit early if ->aborted is set */ |
bf31f87f | 2033 | if (TRANS_ABORTED(cur_trans)) { |
25287e0a | 2034 | ret = cur_trans->aborted; |
3a45bb20 | 2035 | btrfs_end_transaction(trans); |
e4a2bcac | 2036 | return ret; |
25287e0a | 2037 | } |
49b25e05 | 2038 | |
f45c752b JB |
2039 | btrfs_trans_release_metadata(trans); |
2040 | trans->block_rsv = NULL; | |
2041 | ||
56bec294 CM |
2042 | /* make a pass through all the delayed refs we have so far |
2043 | * any runnings procs may add more while we are here | |
2044 | */ | |
c79a70b1 | 2045 | ret = btrfs_run_delayed_refs(trans, 0); |
e4a2bcac | 2046 | if (ret) { |
3a45bb20 | 2047 | btrfs_end_transaction(trans); |
e4a2bcac JB |
2048 | return ret; |
2049 | } | |
56bec294 | 2050 | |
b7ec40d7 | 2051 | cur_trans = trans->transaction; |
49b25e05 | 2052 | |
56bec294 CM |
2053 | /* |
2054 | * set the flushing flag so procs in this transaction have to | |
2055 | * start sending their work down. | |
2056 | */ | |
b7ec40d7 | 2057 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 2058 | smp_wmb(); |
56bec294 | 2059 | |
119e80df | 2060 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2061 | |
c79a70b1 | 2062 | ret = btrfs_run_delayed_refs(trans, 0); |
e4a2bcac | 2063 | if (ret) { |
3a45bb20 | 2064 | btrfs_end_transaction(trans); |
e4a2bcac JB |
2065 | return ret; |
2066 | } | |
56bec294 | 2067 | |
3204d33c | 2068 | if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { |
1bbc621e CM |
2069 | int run_it = 0; |
2070 | ||
2071 | /* this mutex is also taken before trying to set | |
2072 | * block groups readonly. We need to make sure | |
2073 | * that nobody has set a block group readonly | |
2074 | * after a extents from that block group have been | |
2075 | * allocated for cache files. btrfs_set_block_group_ro | |
2076 | * will wait for the transaction to commit if it | |
3204d33c | 2077 | * finds BTRFS_TRANS_DIRTY_BG_RUN set. |
1bbc621e | 2078 | * |
3204d33c JB |
2079 | * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure |
2080 | * only one process starts all the block group IO. It wouldn't | |
1bbc621e CM |
2081 | * hurt to have more than one go through, but there's no |
2082 | * real advantage to it either. | |
2083 | */ | |
0b246afa | 2084 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c JB |
2085 | if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, |
2086 | &cur_trans->flags)) | |
1bbc621e | 2087 | run_it = 1; |
0b246afa | 2088 | mutex_unlock(&fs_info->ro_block_group_mutex); |
1bbc621e | 2089 | |
f9cacae3 | 2090 | if (run_it) { |
21217054 | 2091 | ret = btrfs_start_dirty_block_groups(trans); |
f9cacae3 NB |
2092 | if (ret) { |
2093 | btrfs_end_transaction(trans); | |
2094 | return ret; | |
2095 | } | |
2096 | } | |
1bbc621e CM |
2097 | } |
2098 | ||
0b246afa | 2099 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2100 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { |
0b246afa | 2101 | spin_unlock(&fs_info->trans_lock); |
9b64f57d | 2102 | refcount_inc(&cur_trans->use_count); |
3a45bb20 | 2103 | ret = btrfs_end_transaction(trans); |
ccd467d6 | 2104 | |
2ff7e61e | 2105 | wait_for_commit(cur_trans); |
15ee9bc7 | 2106 | |
bf31f87f | 2107 | if (TRANS_ABORTED(cur_trans)) |
b4924a0f LB |
2108 | ret = cur_trans->aborted; |
2109 | ||
724e2315 | 2110 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 2111 | |
49b25e05 | 2112 | return ret; |
79154b1b | 2113 | } |
4313b399 | 2114 | |
4a9d8bde | 2115 | cur_trans->state = TRANS_STATE_COMMIT_START; |
0b246afa | 2116 | wake_up(&fs_info->transaction_blocked_wait); |
bb9c12c9 | 2117 | |
0b246afa | 2118 | if (cur_trans->list.prev != &fs_info->trans_list) { |
ccd467d6 CM |
2119 | prev_trans = list_entry(cur_trans->list.prev, |
2120 | struct btrfs_transaction, list); | |
4a9d8bde | 2121 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
9b64f57d | 2122 | refcount_inc(&prev_trans->use_count); |
0b246afa | 2123 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2124 | |
2ff7e61e | 2125 | wait_for_commit(prev_trans); |
bf31f87f | 2126 | ret = READ_ONCE(prev_trans->aborted); |
ccd467d6 | 2127 | |
724e2315 | 2128 | btrfs_put_transaction(prev_trans); |
1f9b8c8f FM |
2129 | if (ret) |
2130 | goto cleanup_transaction; | |
a4abeea4 | 2131 | } else { |
0b246afa | 2132 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2133 | } |
a4abeea4 | 2134 | } else { |
0b246afa | 2135 | spin_unlock(&fs_info->trans_lock); |
cb2d3dad FM |
2136 | /* |
2137 | * The previous transaction was aborted and was already removed | |
2138 | * from the list of transactions at fs_info->trans_list. So we | |
2139 | * abort to prevent writing a new superblock that reflects a | |
2140 | * corrupt state (pointing to trees with unwritten nodes/leafs). | |
2141 | */ | |
2142 | if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) { | |
2143 | ret = -EROFS; | |
2144 | goto cleanup_transaction; | |
2145 | } | |
ccd467d6 | 2146 | } |
15ee9bc7 | 2147 | |
0860adfd MX |
2148 | extwriter_counter_dec(cur_trans, trans->type); |
2149 | ||
609e804d | 2150 | ret = btrfs_start_delalloc_flush(trans); |
82436617 MX |
2151 | if (ret) |
2152 | goto cleanup_transaction; | |
2153 | ||
e5c304e6 | 2154 | ret = btrfs_run_delayed_items(trans); |
581227d0 MX |
2155 | if (ret) |
2156 | goto cleanup_transaction; | |
15ee9bc7 | 2157 | |
581227d0 MX |
2158 | wait_event(cur_trans->writer_wait, |
2159 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 2160 | |
581227d0 | 2161 | /* some pending stuffs might be added after the previous flush. */ |
e5c304e6 | 2162 | ret = btrfs_run_delayed_items(trans); |
ca469637 MX |
2163 | if (ret) |
2164 | goto cleanup_transaction; | |
2165 | ||
609e804d | 2166 | btrfs_wait_delalloc_flush(trans); |
cb7ab021 | 2167 | |
2ff7e61e | 2168 | btrfs_scrub_pause(fs_info); |
ed0ca140 JB |
2169 | /* |
2170 | * Ok now we need to make sure to block out any other joins while we | |
2171 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 2172 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 | 2173 | */ |
0b246afa | 2174 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2175 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 2176 | spin_unlock(&fs_info->trans_lock); |
ed0ca140 JB |
2177 | wait_event(cur_trans->writer_wait, |
2178 | atomic_read(&cur_trans->num_writers) == 1); | |
2179 | ||
bf31f87f | 2180 | if (TRANS_ABORTED(cur_trans)) { |
2cba30f1 | 2181 | ret = cur_trans->aborted; |
6cf7f77e | 2182 | goto scrub_continue; |
2cba30f1 | 2183 | } |
7585717f CM |
2184 | /* |
2185 | * the reloc mutex makes sure that we stop | |
2186 | * the balancing code from coming in and moving | |
2187 | * extents around in the middle of the commit | |
2188 | */ | |
0b246afa | 2189 | mutex_lock(&fs_info->reloc_mutex); |
7585717f | 2190 | |
42874b3d MX |
2191 | /* |
2192 | * We needn't worry about the delayed items because we will | |
2193 | * deal with them in create_pending_snapshot(), which is the | |
2194 | * core function of the snapshot creation. | |
2195 | */ | |
08d50ca3 | 2196 | ret = create_pending_snapshots(trans); |
56e9f6ea DS |
2197 | if (ret) |
2198 | goto unlock_reloc; | |
3063d29f | 2199 | |
42874b3d MX |
2200 | /* |
2201 | * We insert the dir indexes of the snapshots and update the inode | |
2202 | * of the snapshots' parents after the snapshot creation, so there | |
2203 | * are some delayed items which are not dealt with. Now deal with | |
2204 | * them. | |
2205 | * | |
2206 | * We needn't worry that this operation will corrupt the snapshots, | |
2207 | * because all the tree which are snapshoted will be forced to COW | |
2208 | * the nodes and leaves. | |
2209 | */ | |
e5c304e6 | 2210 | ret = btrfs_run_delayed_items(trans); |
56e9f6ea DS |
2211 | if (ret) |
2212 | goto unlock_reloc; | |
16cdcec7 | 2213 | |
c79a70b1 | 2214 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
56e9f6ea DS |
2215 | if (ret) |
2216 | goto unlock_reloc; | |
56bec294 | 2217 | |
e999376f CM |
2218 | /* |
2219 | * make sure none of the code above managed to slip in a | |
2220 | * delayed item | |
2221 | */ | |
ccdf9b30 | 2222 | btrfs_assert_delayed_root_empty(fs_info); |
e999376f | 2223 | |
2c90e5d6 | 2224 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 2225 | |
e02119d5 CM |
2226 | /* btrfs_commit_tree_roots is responsible for getting the |
2227 | * various roots consistent with each other. Every pointer | |
2228 | * in the tree of tree roots has to point to the most up to date | |
2229 | * root for every subvolume and other tree. So, we have to keep | |
2230 | * the tree logging code from jumping in and changing any | |
2231 | * of the trees. | |
2232 | * | |
2233 | * At this point in the commit, there can't be any tree-log | |
2234 | * writers, but a little lower down we drop the trans mutex | |
2235 | * and let new people in. By holding the tree_log_mutex | |
2236 | * from now until after the super is written, we avoid races | |
2237 | * with the tree-log code. | |
2238 | */ | |
0b246afa | 2239 | mutex_lock(&fs_info->tree_log_mutex); |
e02119d5 | 2240 | |
7e4443d9 | 2241 | ret = commit_fs_roots(trans); |
56e9f6ea DS |
2242 | if (ret) |
2243 | goto unlock_tree_log; | |
54aa1f4d | 2244 | |
3818aea2 | 2245 | /* |
7e1876ac DS |
2246 | * Since the transaction is done, we can apply the pending changes |
2247 | * before the next transaction. | |
3818aea2 | 2248 | */ |
0b246afa | 2249 | btrfs_apply_pending_changes(fs_info); |
3818aea2 | 2250 | |
5d4f98a2 | 2251 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2252 | * safe to free the root of tree log roots |
2253 | */ | |
0b246afa | 2254 | btrfs_free_log_root_tree(trans, fs_info); |
e02119d5 | 2255 | |
82bafb38 QW |
2256 | /* |
2257 | * commit_fs_roots() can call btrfs_save_ino_cache(), which generates | |
2258 | * new delayed refs. Must handle them or qgroup can be wrong. | |
2259 | */ | |
c79a70b1 | 2260 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
56e9f6ea DS |
2261 | if (ret) |
2262 | goto unlock_tree_log; | |
82bafb38 | 2263 | |
0ed4792a QW |
2264 | /* |
2265 | * Since fs roots are all committed, we can get a quite accurate | |
2266 | * new_roots. So let's do quota accounting. | |
2267 | */ | |
460fb20a | 2268 | ret = btrfs_qgroup_account_extents(trans); |
56e9f6ea DS |
2269 | if (ret < 0) |
2270 | goto unlock_tree_log; | |
0ed4792a | 2271 | |
9386d8bc | 2272 | ret = commit_cowonly_roots(trans); |
56e9f6ea DS |
2273 | if (ret) |
2274 | goto unlock_tree_log; | |
54aa1f4d | 2275 | |
2cba30f1 MX |
2276 | /* |
2277 | * The tasks which save the space cache and inode cache may also | |
2278 | * update ->aborted, check it. | |
2279 | */ | |
bf31f87f | 2280 | if (TRANS_ABORTED(cur_trans)) { |
2cba30f1 | 2281 | ret = cur_trans->aborted; |
56e9f6ea | 2282 | goto unlock_tree_log; |
2cba30f1 MX |
2283 | } |
2284 | ||
8b74c03e | 2285 | btrfs_prepare_extent_commit(fs_info); |
11833d66 | 2286 | |
0b246afa | 2287 | cur_trans = fs_info->running_transaction; |
5d4f98a2 | 2288 | |
0b246afa JM |
2289 | btrfs_set_root_node(&fs_info->tree_root->root_item, |
2290 | fs_info->tree_root->node); | |
2291 | list_add_tail(&fs_info->tree_root->dirty_list, | |
9e351cc8 | 2292 | &cur_trans->switch_commits); |
5d4f98a2 | 2293 | |
0b246afa JM |
2294 | btrfs_set_root_node(&fs_info->chunk_root->root_item, |
2295 | fs_info->chunk_root->node); | |
2296 | list_add_tail(&fs_info->chunk_root->dirty_list, | |
9e351cc8 JB |
2297 | &cur_trans->switch_commits); |
2298 | ||
889bfa39 | 2299 | switch_commit_roots(trans); |
5d4f98a2 | 2300 | |
ce93ec54 | 2301 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2302 | ASSERT(list_empty(&cur_trans->io_bgs)); |
2ff7e61e | 2303 | update_super_roots(fs_info); |
e02119d5 | 2304 | |
0b246afa JM |
2305 | btrfs_set_super_log_root(fs_info->super_copy, 0); |
2306 | btrfs_set_super_log_root_level(fs_info->super_copy, 0); | |
2307 | memcpy(fs_info->super_for_commit, fs_info->super_copy, | |
2308 | sizeof(*fs_info->super_copy)); | |
ccd467d6 | 2309 | |
bbbf7243 | 2310 | btrfs_commit_device_sizes(cur_trans); |
935e5cc9 | 2311 | |
0b246afa JM |
2312 | clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
2313 | clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); | |
656f30db | 2314 | |
4fbcdf66 FM |
2315 | btrfs_trans_release_chunk_metadata(trans); |
2316 | ||
0b246afa | 2317 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2318 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
0b246afa JM |
2319 | fs_info->running_transaction = NULL; |
2320 | spin_unlock(&fs_info->trans_lock); | |
2321 | mutex_unlock(&fs_info->reloc_mutex); | |
b7ec40d7 | 2322 | |
0b246afa | 2323 | wake_up(&fs_info->transaction_wait); |
e6dcd2dc | 2324 | |
70458a58 | 2325 | ret = btrfs_write_and_wait_transaction(trans); |
49b25e05 | 2326 | if (ret) { |
0b246afa JM |
2327 | btrfs_handle_fs_error(fs_info, ret, |
2328 | "Error while writing out transaction"); | |
56e9f6ea DS |
2329 | /* |
2330 | * reloc_mutex has been unlocked, tree_log_mutex is still held | |
2331 | * but we can't jump to unlock_tree_log causing double unlock | |
2332 | */ | |
0b246afa | 2333 | mutex_unlock(&fs_info->tree_log_mutex); |
6cf7f77e | 2334 | goto scrub_continue; |
49b25e05 JM |
2335 | } |
2336 | ||
eece6a9c | 2337 | ret = write_all_supers(fs_info, 0); |
e02119d5 CM |
2338 | /* |
2339 | * the super is written, we can safely allow the tree-loggers | |
2340 | * to go about their business | |
2341 | */ | |
0b246afa | 2342 | mutex_unlock(&fs_info->tree_log_mutex); |
c1f32b7c AJ |
2343 | if (ret) |
2344 | goto scrub_continue; | |
e02119d5 | 2345 | |
5ead2dd0 | 2346 | btrfs_finish_extent_commit(trans); |
4313b399 | 2347 | |
3204d33c | 2348 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) |
0b246afa | 2349 | btrfs_clear_space_info_full(fs_info); |
13212b54 | 2350 | |
0b246afa | 2351 | fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
2352 | /* |
2353 | * We needn't acquire the lock here because there is no other task | |
2354 | * which can change it. | |
2355 | */ | |
2356 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2357 | wake_up(&cur_trans->commit_wait); |
a514d638 | 2358 | clear_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags); |
3de4586c | 2359 | |
0b246afa | 2360 | spin_lock(&fs_info->trans_lock); |
13c5a93e | 2361 | list_del_init(&cur_trans->list); |
0b246afa | 2362 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 2363 | |
724e2315 JB |
2364 | btrfs_put_transaction(cur_trans); |
2365 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2366 | |
0860adfd | 2367 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2368 | sb_end_intwrite(fs_info->sb); |
b2b5ef5c | 2369 | |
3a45bb20 | 2370 | trace_btrfs_transaction_commit(trans->root); |
1abe9b8a | 2371 | |
2ff7e61e | 2372 | btrfs_scrub_continue(fs_info); |
a2de733c | 2373 | |
9ed74f2d JB |
2374 | if (current->journal_info == trans) |
2375 | current->journal_info = NULL; | |
2376 | ||
2c90e5d6 | 2377 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2378 | |
79154b1b | 2379 | return ret; |
49b25e05 | 2380 | |
56e9f6ea DS |
2381 | unlock_tree_log: |
2382 | mutex_unlock(&fs_info->tree_log_mutex); | |
2383 | unlock_reloc: | |
2384 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2385 | scrub_continue: |
2ff7e61e | 2386 | btrfs_scrub_continue(fs_info); |
49b25e05 | 2387 | cleanup_transaction: |
dc60c525 | 2388 | btrfs_trans_release_metadata(trans); |
c7cc64a9 | 2389 | btrfs_cleanup_pending_block_groups(trans); |
4fbcdf66 | 2390 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2391 | trans->block_rsv = NULL; |
0b246afa | 2392 | btrfs_warn(fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2393 | if (current->journal_info == trans) |
2394 | current->journal_info = NULL; | |
97cb39bb | 2395 | cleanup_transaction(trans, ret); |
49b25e05 JM |
2396 | |
2397 | return ret; | |
79154b1b CM |
2398 | } |
2399 | ||
d352ac68 | 2400 | /* |
9d1a2a3a DS |
2401 | * return < 0 if error |
2402 | * 0 if there are no more dead_roots at the time of call | |
2403 | * 1 there are more to be processed, call me again | |
2404 | * | |
2405 | * The return value indicates there are certainly more snapshots to delete, but | |
2406 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2407 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2408 | * few seconds later. | |
d352ac68 | 2409 | */ |
9d1a2a3a | 2410 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 2411 | { |
9d1a2a3a | 2412 | int ret; |
5d4f98a2 YZ |
2413 | struct btrfs_fs_info *fs_info = root->fs_info; |
2414 | ||
a4abeea4 | 2415 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2416 | if (list_empty(&fs_info->dead_roots)) { |
2417 | spin_unlock(&fs_info->trans_lock); | |
2418 | return 0; | |
2419 | } | |
2420 | root = list_first_entry(&fs_info->dead_roots, | |
2421 | struct btrfs_root, root_list); | |
cfad392b | 2422 | list_del_init(&root->root_list); |
a4abeea4 | 2423 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2424 | |
4fd786e6 | 2425 | btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); |
76dda93c | 2426 | |
9d1a2a3a | 2427 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2428 | |
9d1a2a3a DS |
2429 | if (btrfs_header_backref_rev(root->node) < |
2430 | BTRFS_MIXED_BACKREF_REV) | |
2431 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
2432 | else | |
2433 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
32471dc2 | 2434 | |
6596a928 | 2435 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2436 | } |
572d9ab7 DS |
2437 | |
2438 | void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) | |
2439 | { | |
2440 | unsigned long prev; | |
2441 | unsigned long bit; | |
2442 | ||
6c9fe14f | 2443 | prev = xchg(&fs_info->pending_changes, 0); |
572d9ab7 DS |
2444 | if (!prev) |
2445 | return; | |
2446 | ||
7e1876ac DS |
2447 | bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE; |
2448 | if (prev & bit) | |
2449 | btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2450 | prev &= ~bit; | |
2451 | ||
2452 | bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE; | |
2453 | if (prev & bit) | |
2454 | btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2455 | prev &= ~bit; | |
2456 | ||
d51033d0 DS |
2457 | bit = 1 << BTRFS_PENDING_COMMIT; |
2458 | if (prev & bit) | |
2459 | btrfs_debug(fs_info, "pending commit done"); | |
2460 | prev &= ~bit; | |
2461 | ||
572d9ab7 DS |
2462 | if (prev) |
2463 | btrfs_warn(fs_info, | |
2464 | "unknown pending changes left 0x%lx, ignoring", prev); | |
2465 | } |