Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
79154b1b | 19 | #include <linux/fs.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
34088780 | 21 | #include <linux/sched.h> |
d3c2fdcf | 22 | #include <linux/writeback.h> |
5f39d397 | 23 | #include <linux/pagemap.h> |
5f2cc086 | 24 | #include <linux/blkdev.h> |
8ea05e3a | 25 | #include <linux/uuid.h> |
79154b1b CM |
26 | #include "ctree.h" |
27 | #include "disk-io.h" | |
28 | #include "transaction.h" | |
925baedd | 29 | #include "locking.h" |
e02119d5 | 30 | #include "tree-log.h" |
581bb050 | 31 | #include "inode-map.h" |
733f4fbb | 32 | #include "volumes.h" |
8dabb742 | 33 | #include "dev-replace.h" |
fcebe456 | 34 | #include "qgroup.h" |
79154b1b | 35 | |
0f7d52f4 CM |
36 | #define BTRFS_ROOT_TRANS_TAG 0 |
37 | ||
e8c9f186 | 38 | static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { |
4a9d8bde MX |
39 | [TRANS_STATE_RUNNING] = 0U, |
40 | [TRANS_STATE_BLOCKED] = (__TRANS_USERSPACE | | |
41 | __TRANS_START), | |
42 | [TRANS_STATE_COMMIT_START] = (__TRANS_USERSPACE | | |
43 | __TRANS_START | | |
44 | __TRANS_ATTACH), | |
45 | [TRANS_STATE_COMMIT_DOING] = (__TRANS_USERSPACE | | |
46 | __TRANS_START | | |
47 | __TRANS_ATTACH | | |
48 | __TRANS_JOIN), | |
49 | [TRANS_STATE_UNBLOCKED] = (__TRANS_USERSPACE | | |
50 | __TRANS_START | | |
51 | __TRANS_ATTACH | | |
52 | __TRANS_JOIN | | |
53 | __TRANS_JOIN_NOLOCK), | |
54 | [TRANS_STATE_COMPLETED] = (__TRANS_USERSPACE | | |
55 | __TRANS_START | | |
56 | __TRANS_ATTACH | | |
57 | __TRANS_JOIN | | |
58 | __TRANS_JOIN_NOLOCK), | |
59 | }; | |
60 | ||
724e2315 | 61 | void btrfs_put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 62 | { |
13c5a93e JB |
63 | WARN_ON(atomic_read(&transaction->use_count) == 0); |
64 | if (atomic_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 65 | BUG_ON(!list_empty(&transaction->list)); |
c46effa6 | 66 | WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root)); |
1262133b JB |
67 | if (transaction->delayed_refs.pending_csums) |
68 | printk(KERN_ERR "pending csums is %llu\n", | |
69 | transaction->delayed_refs.pending_csums); | |
6df9a95e JB |
70 | while (!list_empty(&transaction->pending_chunks)) { |
71 | struct extent_map *em; | |
72 | ||
73 | em = list_first_entry(&transaction->pending_chunks, | |
74 | struct extent_map, list); | |
75 | list_del_init(&em->list); | |
76 | free_extent_map(em); | |
77 | } | |
2c90e5d6 | 78 | kmem_cache_free(btrfs_transaction_cachep, transaction); |
78fae27e | 79 | } |
79154b1b CM |
80 | } |
81 | ||
663dfbb0 FM |
82 | static void clear_btree_io_tree(struct extent_io_tree *tree) |
83 | { | |
84 | spin_lock(&tree->lock); | |
b666a9cd DS |
85 | /* |
86 | * Do a single barrier for the waitqueue_active check here, the state | |
87 | * of the waitqueue should not change once clear_btree_io_tree is | |
88 | * called. | |
89 | */ | |
90 | smp_mb(); | |
663dfbb0 FM |
91 | while (!RB_EMPTY_ROOT(&tree->state)) { |
92 | struct rb_node *node; | |
93 | struct extent_state *state; | |
94 | ||
95 | node = rb_first(&tree->state); | |
96 | state = rb_entry(node, struct extent_state, rb_node); | |
97 | rb_erase(&state->rb_node, &tree->state); | |
98 | RB_CLEAR_NODE(&state->rb_node); | |
99 | /* | |
100 | * btree io trees aren't supposed to have tasks waiting for | |
101 | * changes in the flags of extent states ever. | |
102 | */ | |
103 | ASSERT(!waitqueue_active(&state->wq)); | |
104 | free_extent_state(state); | |
351810c1 DS |
105 | |
106 | cond_resched_lock(&tree->lock); | |
663dfbb0 FM |
107 | } |
108 | spin_unlock(&tree->lock); | |
109 | } | |
110 | ||
9e351cc8 JB |
111 | static noinline void switch_commit_roots(struct btrfs_transaction *trans, |
112 | struct btrfs_fs_info *fs_info) | |
817d52f8 | 113 | { |
9e351cc8 JB |
114 | struct btrfs_root *root, *tmp; |
115 | ||
116 | down_write(&fs_info->commit_root_sem); | |
117 | list_for_each_entry_safe(root, tmp, &trans->switch_commits, | |
118 | dirty_list) { | |
119 | list_del_init(&root->dirty_list); | |
120 | free_extent_buffer(root->commit_root); | |
121 | root->commit_root = btrfs_root_node(root); | |
122 | if (is_fstree(root->objectid)) | |
123 | btrfs_unpin_free_ino(root); | |
663dfbb0 | 124 | clear_btree_io_tree(&root->dirty_log_pages); |
9e351cc8 | 125 | } |
2b9dbef2 JB |
126 | |
127 | /* We can free old roots now. */ | |
128 | spin_lock(&trans->dropped_roots_lock); | |
129 | while (!list_empty(&trans->dropped_roots)) { | |
130 | root = list_first_entry(&trans->dropped_roots, | |
131 | struct btrfs_root, root_list); | |
132 | list_del_init(&root->root_list); | |
133 | spin_unlock(&trans->dropped_roots_lock); | |
134 | btrfs_drop_and_free_fs_root(fs_info, root); | |
135 | spin_lock(&trans->dropped_roots_lock); | |
136 | } | |
137 | spin_unlock(&trans->dropped_roots_lock); | |
9e351cc8 | 138 | up_write(&fs_info->commit_root_sem); |
817d52f8 JB |
139 | } |
140 | ||
0860adfd MX |
141 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
142 | unsigned int type) | |
143 | { | |
144 | if (type & TRANS_EXTWRITERS) | |
145 | atomic_inc(&trans->num_extwriters); | |
146 | } | |
147 | ||
148 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
149 | unsigned int type) | |
150 | { | |
151 | if (type & TRANS_EXTWRITERS) | |
152 | atomic_dec(&trans->num_extwriters); | |
153 | } | |
154 | ||
155 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
156 | unsigned int type) | |
157 | { | |
158 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
159 | } | |
160 | ||
161 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
162 | { | |
163 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
164 | } |
165 | ||
d352ac68 CM |
166 | /* |
167 | * either allocate a new transaction or hop into the existing one | |
168 | */ | |
0860adfd | 169 | static noinline int join_transaction(struct btrfs_root *root, unsigned int type) |
79154b1b CM |
170 | { |
171 | struct btrfs_transaction *cur_trans; | |
19ae4e81 | 172 | struct btrfs_fs_info *fs_info = root->fs_info; |
a4abeea4 | 173 | |
19ae4e81 | 174 | spin_lock(&fs_info->trans_lock); |
d43317dc | 175 | loop: |
49b25e05 | 176 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 177 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 178 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
179 | return -EROFS; |
180 | } | |
181 | ||
19ae4e81 | 182 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 183 | if (cur_trans) { |
871383be | 184 | if (cur_trans->aborted) { |
19ae4e81 | 185 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 186 | return cur_trans->aborted; |
871383be | 187 | } |
4a9d8bde | 188 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
189 | spin_unlock(&fs_info->trans_lock); |
190 | return -EBUSY; | |
191 | } | |
a4abeea4 | 192 | atomic_inc(&cur_trans->use_count); |
13c5a93e | 193 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 194 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 195 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 196 | return 0; |
79154b1b | 197 | } |
19ae4e81 | 198 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 199 | |
354aa0fb MX |
200 | /* |
201 | * If we are ATTACH, we just want to catch the current transaction, | |
202 | * and commit it. If there is no transaction, just return ENOENT. | |
203 | */ | |
204 | if (type == TRANS_ATTACH) | |
205 | return -ENOENT; | |
206 | ||
4a9d8bde MX |
207 | /* |
208 | * JOIN_NOLOCK only happens during the transaction commit, so | |
209 | * it is impossible that ->running_transaction is NULL | |
210 | */ | |
211 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
212 | ||
a4abeea4 JB |
213 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); |
214 | if (!cur_trans) | |
215 | return -ENOMEM; | |
d43317dc | 216 | |
19ae4e81 JS |
217 | spin_lock(&fs_info->trans_lock); |
218 | if (fs_info->running_transaction) { | |
d43317dc CM |
219 | /* |
220 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 221 | * to redo the checks above |
d43317dc | 222 | */ |
a4abeea4 | 223 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
d43317dc | 224 | goto loop; |
87533c47 | 225 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 226 | spin_unlock(&fs_info->trans_lock); |
7b8b92af JB |
227 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
228 | return -EROFS; | |
79154b1b | 229 | } |
d43317dc | 230 | |
a4abeea4 | 231 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 232 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
233 | init_waitqueue_head(&cur_trans->writer_wait); |
234 | init_waitqueue_head(&cur_trans->commit_wait); | |
161c3549 | 235 | init_waitqueue_head(&cur_trans->pending_wait); |
4a9d8bde | 236 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
237 | /* |
238 | * One for this trans handle, one so it will live on until we | |
239 | * commit the transaction. | |
240 | */ | |
241 | atomic_set(&cur_trans->use_count, 2); | |
161c3549 | 242 | atomic_set(&cur_trans->pending_ordered, 0); |
3204d33c | 243 | cur_trans->flags = 0; |
a4abeea4 JB |
244 | cur_trans->start_time = get_seconds(); |
245 | ||
a099d0fd AM |
246 | memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); |
247 | ||
c46effa6 | 248 | cur_trans->delayed_refs.href_root = RB_ROOT; |
3368d001 | 249 | cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; |
d7df2c79 | 250 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
20b297d6 JS |
251 | |
252 | /* | |
253 | * although the tree mod log is per file system and not per transaction, | |
254 | * the log must never go across transaction boundaries. | |
255 | */ | |
256 | smp_mb(); | |
31b1a2bd | 257 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
efe120a0 | 258 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when " |
20b297d6 | 259 | "creating a fresh transaction\n"); |
31b1a2bd | 260 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
efe120a0 | 261 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when " |
20b297d6 | 262 | "creating a fresh transaction\n"); |
fc36ed7e | 263 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 264 | |
a4abeea4 JB |
265 | spin_lock_init(&cur_trans->delayed_refs.lock); |
266 | ||
267 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
6df9a95e | 268 | INIT_LIST_HEAD(&cur_trans->pending_chunks); |
9e351cc8 | 269 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
ce93ec54 | 270 | INIT_LIST_HEAD(&cur_trans->dirty_bgs); |
1bbc621e | 271 | INIT_LIST_HEAD(&cur_trans->io_bgs); |
2b9dbef2 | 272 | INIT_LIST_HEAD(&cur_trans->dropped_roots); |
1bbc621e | 273 | mutex_init(&cur_trans->cache_write_mutex); |
cb723e49 | 274 | cur_trans->num_dirty_bgs = 0; |
ce93ec54 | 275 | spin_lock_init(&cur_trans->dirty_bgs_lock); |
e33e17ee JM |
276 | INIT_LIST_HEAD(&cur_trans->deleted_bgs); |
277 | spin_lock_init(&cur_trans->deleted_bgs_lock); | |
2b9dbef2 | 278 | spin_lock_init(&cur_trans->dropped_roots_lock); |
19ae4e81 | 279 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 280 | extent_io_tree_init(&cur_trans->dirty_pages, |
19ae4e81 JS |
281 | fs_info->btree_inode->i_mapping); |
282 | fs_info->generation++; | |
283 | cur_trans->transid = fs_info->generation; | |
284 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 285 | cur_trans->aborted = 0; |
19ae4e81 | 286 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 287 | |
79154b1b CM |
288 | return 0; |
289 | } | |
290 | ||
d352ac68 | 291 | /* |
d397712b CM |
292 | * this does all the record keeping required to make sure that a reference |
293 | * counted root is properly recorded in a given transaction. This is required | |
294 | * to make sure the old root from before we joined the transaction is deleted | |
295 | * when the transaction commits | |
d352ac68 | 296 | */ |
7585717f | 297 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 298 | struct btrfs_root *root) |
6702ed49 | 299 | { |
27cdeb70 MX |
300 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
301 | root->last_trans < trans->transid) { | |
6702ed49 | 302 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
303 | WARN_ON(root->commit_root != root->node); |
304 | ||
7585717f | 305 | /* |
27cdeb70 | 306 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
307 | * we have the reloc mutex held now, so there |
308 | * is only one writer in this function | |
309 | */ | |
27cdeb70 | 310 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 311 | |
27cdeb70 | 312 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
313 | * they find our root->last_trans update |
314 | */ | |
315 | smp_wmb(); | |
316 | ||
a4abeea4 JB |
317 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
318 | if (root->last_trans == trans->transid) { | |
319 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
320 | return 0; | |
321 | } | |
5d4f98a2 YZ |
322 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
323 | (unsigned long)root->root_key.objectid, | |
324 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 325 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
326 | root->last_trans = trans->transid; |
327 | ||
328 | /* this is pretty tricky. We don't want to | |
329 | * take the relocation lock in btrfs_record_root_in_trans | |
330 | * unless we're really doing the first setup for this root in | |
331 | * this transaction. | |
332 | * | |
333 | * Normally we'd use root->last_trans as a flag to decide | |
334 | * if we want to take the expensive mutex. | |
335 | * | |
336 | * But, we have to set root->last_trans before we | |
337 | * init the relocation root, otherwise, we trip over warnings | |
338 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 339 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
340 | * fixing up the reloc trees and everyone must wait. |
341 | * | |
342 | * When this is zero, they can trust root->last_trans and fly | |
343 | * through btrfs_record_root_in_trans without having to take the | |
344 | * lock. smp_wmb() makes sure that all the writes above are | |
345 | * done before we pop in the zero below | |
346 | */ | |
5d4f98a2 | 347 | btrfs_init_reloc_root(trans, root); |
c7548af6 | 348 | smp_mb__before_atomic(); |
27cdeb70 | 349 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 YZ |
350 | } |
351 | return 0; | |
352 | } | |
bcc63abb | 353 | |
7585717f | 354 | |
2b9dbef2 JB |
355 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
356 | struct btrfs_root *root) | |
357 | { | |
358 | struct btrfs_transaction *cur_trans = trans->transaction; | |
359 | ||
360 | /* Add ourselves to the transaction dropped list */ | |
361 | spin_lock(&cur_trans->dropped_roots_lock); | |
362 | list_add_tail(&root->root_list, &cur_trans->dropped_roots); | |
363 | spin_unlock(&cur_trans->dropped_roots_lock); | |
364 | ||
365 | /* Make sure we don't try to update the root at commit time */ | |
366 | spin_lock(&root->fs_info->fs_roots_radix_lock); | |
367 | radix_tree_tag_clear(&root->fs_info->fs_roots_radix, | |
368 | (unsigned long)root->root_key.objectid, | |
369 | BTRFS_ROOT_TRANS_TAG); | |
370 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
371 | } | |
372 | ||
7585717f CM |
373 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
374 | struct btrfs_root *root) | |
375 | { | |
27cdeb70 | 376 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
7585717f CM |
377 | return 0; |
378 | ||
379 | /* | |
27cdeb70 | 380 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
381 | * and barriers |
382 | */ | |
383 | smp_rmb(); | |
384 | if (root->last_trans == trans->transid && | |
27cdeb70 | 385 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
386 | return 0; |
387 | ||
388 | mutex_lock(&root->fs_info->reloc_mutex); | |
389 | record_root_in_trans(trans, root); | |
390 | mutex_unlock(&root->fs_info->reloc_mutex); | |
391 | ||
392 | return 0; | |
393 | } | |
394 | ||
4a9d8bde MX |
395 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
396 | { | |
397 | return (trans->state >= TRANS_STATE_BLOCKED && | |
501407aa JB |
398 | trans->state < TRANS_STATE_UNBLOCKED && |
399 | !trans->aborted); | |
4a9d8bde MX |
400 | } |
401 | ||
d352ac68 CM |
402 | /* wait for commit against the current transaction to become unblocked |
403 | * when this is done, it is safe to start a new transaction, but the current | |
404 | * transaction might not be fully on disk. | |
405 | */ | |
37d1aeee | 406 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 407 | { |
f9295749 | 408 | struct btrfs_transaction *cur_trans; |
79154b1b | 409 | |
a4abeea4 | 410 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 411 | cur_trans = root->fs_info->running_transaction; |
4a9d8bde | 412 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
13c5a93e | 413 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 414 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
415 | |
416 | wait_event(root->fs_info->transaction_wait, | |
501407aa JB |
417 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
418 | cur_trans->aborted); | |
724e2315 | 419 | btrfs_put_transaction(cur_trans); |
a4abeea4 JB |
420 | } else { |
421 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 422 | } |
37d1aeee CM |
423 | } |
424 | ||
a22285a6 YZ |
425 | static int may_wait_transaction(struct btrfs_root *root, int type) |
426 | { | |
a4abeea4 JB |
427 | if (root->fs_info->log_root_recovering) |
428 | return 0; | |
429 | ||
430 | if (type == TRANS_USERSPACE) | |
431 | return 1; | |
432 | ||
433 | if (type == TRANS_START && | |
434 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 435 | return 1; |
a4abeea4 | 436 | |
a22285a6 YZ |
437 | return 0; |
438 | } | |
439 | ||
20dd2cbf MX |
440 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
441 | { | |
442 | if (!root->fs_info->reloc_ctl || | |
27cdeb70 | 443 | !test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
20dd2cbf MX |
444 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
445 | root->reloc_root) | |
446 | return false; | |
447 | ||
448 | return true; | |
449 | } | |
450 | ||
08e007d2 | 451 | static struct btrfs_trans_handle * |
5aed1dd8 AM |
452 | start_transaction(struct btrfs_root *root, unsigned int num_items, |
453 | unsigned int type, enum btrfs_reserve_flush_enum flush) | |
37d1aeee | 454 | { |
a22285a6 YZ |
455 | struct btrfs_trans_handle *h; |
456 | struct btrfs_transaction *cur_trans; | |
b5009945 | 457 | u64 num_bytes = 0; |
c5567237 | 458 | u64 qgroup_reserved = 0; |
20dd2cbf MX |
459 | bool reloc_reserved = false; |
460 | int ret; | |
acce952b | 461 | |
46c4e71e | 462 | /* Send isn't supposed to start transactions. */ |
2755a0de | 463 | ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); |
46c4e71e | 464 | |
87533c47 | 465 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
acce952b | 466 | return ERR_PTR(-EROFS); |
2a1eb461 | 467 | |
46c4e71e | 468 | if (current->journal_info) { |
0860adfd | 469 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 JB |
470 | h = current->journal_info; |
471 | h->use_count++; | |
b7d5b0a8 | 472 | WARN_ON(h->use_count > 2); |
2a1eb461 JB |
473 | h->orig_rsv = h->block_rsv; |
474 | h->block_rsv = NULL; | |
475 | goto got_it; | |
476 | } | |
b5009945 JB |
477 | |
478 | /* | |
479 | * Do the reservation before we join the transaction so we can do all | |
480 | * the appropriate flushing if need be. | |
481 | */ | |
482 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
7174109c QW |
483 | qgroup_reserved = num_items * root->nodesize; |
484 | ret = btrfs_qgroup_reserve_meta(root, qgroup_reserved); | |
485 | if (ret) | |
486 | return ERR_PTR(ret); | |
c5567237 | 487 | |
b5009945 | 488 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); |
20dd2cbf MX |
489 | /* |
490 | * Do the reservation for the relocation root creation | |
491 | */ | |
ee39b432 | 492 | if (need_reserve_reloc_root(root)) { |
20dd2cbf MX |
493 | num_bytes += root->nodesize; |
494 | reloc_reserved = true; | |
495 | } | |
496 | ||
08e007d2 MX |
497 | ret = btrfs_block_rsv_add(root, |
498 | &root->fs_info->trans_block_rsv, | |
499 | num_bytes, flush); | |
b5009945 | 500 | if (ret) |
843fcf35 | 501 | goto reserve_fail; |
b5009945 | 502 | } |
a22285a6 | 503 | again: |
f2f767e7 | 504 | h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); |
843fcf35 MX |
505 | if (!h) { |
506 | ret = -ENOMEM; | |
507 | goto alloc_fail; | |
508 | } | |
37d1aeee | 509 | |
98114659 JB |
510 | /* |
511 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
512 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
513 | * because we're already holding a ref. We need this because we could | |
514 | * have raced in and did an fsync() on a file which can kick a commit | |
515 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
516 | * |
517 | * If we are ATTACH, it means we just want to catch the current | |
518 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 519 | */ |
0860adfd | 520 | if (type & __TRANS_FREEZABLE) |
60376ce4 | 521 | sb_start_intwrite(root->fs_info->sb); |
b2b5ef5c | 522 | |
a22285a6 | 523 | if (may_wait_transaction(root, type)) |
37d1aeee | 524 | wait_current_trans(root); |
a22285a6 | 525 | |
a4abeea4 | 526 | do { |
354aa0fb | 527 | ret = join_transaction(root, type); |
178260b2 | 528 | if (ret == -EBUSY) { |
a4abeea4 | 529 | wait_current_trans(root); |
178260b2 MX |
530 | if (unlikely(type == TRANS_ATTACH)) |
531 | ret = -ENOENT; | |
532 | } | |
a4abeea4 JB |
533 | } while (ret == -EBUSY); |
534 | ||
db5b493a | 535 | if (ret < 0) { |
354aa0fb MX |
536 | /* We must get the transaction if we are JOIN_NOLOCK. */ |
537 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
843fcf35 | 538 | goto join_fail; |
db5b493a | 539 | } |
0f7d52f4 | 540 | |
a22285a6 | 541 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
542 | |
543 | h->transid = cur_trans->transid; | |
544 | h->transaction = cur_trans; | |
d13603ef | 545 | h->root = root; |
2a1eb461 | 546 | h->use_count = 1; |
7174109c | 547 | |
a698d075 | 548 | h->type = type; |
d9a0540a | 549 | h->can_flush_pending_bgs = true; |
bed92eae | 550 | INIT_LIST_HEAD(&h->qgroup_ref_list); |
ea658bad | 551 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 552 | |
a22285a6 | 553 | smp_mb(); |
4a9d8bde MX |
554 | if (cur_trans->state >= TRANS_STATE_BLOCKED && |
555 | may_wait_transaction(root, type)) { | |
abdd2e80 | 556 | current->journal_info = h; |
a22285a6 YZ |
557 | btrfs_commit_transaction(h, root); |
558 | goto again; | |
559 | } | |
560 | ||
b5009945 | 561 | if (num_bytes) { |
8c2a3ca2 | 562 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 563 | h->transid, num_bytes, 1); |
b5009945 JB |
564 | h->block_rsv = &root->fs_info->trans_block_rsv; |
565 | h->bytes_reserved = num_bytes; | |
20dd2cbf | 566 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 567 | } |
9ed74f2d | 568 | |
2a1eb461 | 569 | got_it: |
a4abeea4 | 570 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
571 | |
572 | if (!current->journal_info && type != TRANS_USERSPACE) | |
573 | current->journal_info = h; | |
79154b1b | 574 | return h; |
843fcf35 MX |
575 | |
576 | join_fail: | |
0860adfd | 577 | if (type & __TRANS_FREEZABLE) |
843fcf35 MX |
578 | sb_end_intwrite(root->fs_info->sb); |
579 | kmem_cache_free(btrfs_trans_handle_cachep, h); | |
580 | alloc_fail: | |
581 | if (num_bytes) | |
582 | btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv, | |
583 | num_bytes); | |
584 | reserve_fail: | |
7174109c | 585 | btrfs_qgroup_free_meta(root, qgroup_reserved); |
843fcf35 | 586 | return ERR_PTR(ret); |
79154b1b CM |
587 | } |
588 | ||
f9295749 | 589 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
5aed1dd8 | 590 | unsigned int num_items) |
f9295749 | 591 | { |
08e007d2 MX |
592 | return start_transaction(root, num_items, TRANS_START, |
593 | BTRFS_RESERVE_FLUSH_ALL); | |
f9295749 | 594 | } |
8407aa46 | 595 | |
08e007d2 | 596 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
5aed1dd8 AM |
597 | struct btrfs_root *root, |
598 | unsigned int num_items) | |
8407aa46 | 599 | { |
08e007d2 MX |
600 | return start_transaction(root, num_items, TRANS_START, |
601 | BTRFS_RESERVE_FLUSH_LIMIT); | |
8407aa46 MX |
602 | } |
603 | ||
7a7eaa40 | 604 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 605 | { |
8407aa46 | 606 | return start_transaction(root, 0, TRANS_JOIN, 0); |
f9295749 CM |
607 | } |
608 | ||
7a7eaa40 | 609 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 610 | { |
8407aa46 | 611 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0); |
0af3d00b JB |
612 | } |
613 | ||
7a7eaa40 | 614 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 615 | { |
8407aa46 | 616 | return start_transaction(root, 0, TRANS_USERSPACE, 0); |
9ca9ee09 SW |
617 | } |
618 | ||
d4edf39b MX |
619 | /* |
620 | * btrfs_attach_transaction() - catch the running transaction | |
621 | * | |
622 | * It is used when we want to commit the current the transaction, but | |
623 | * don't want to start a new one. | |
624 | * | |
625 | * Note: If this function return -ENOENT, it just means there is no | |
626 | * running transaction. But it is possible that the inactive transaction | |
627 | * is still in the memory, not fully on disk. If you hope there is no | |
628 | * inactive transaction in the fs when -ENOENT is returned, you should | |
629 | * invoke | |
630 | * btrfs_attach_transaction_barrier() | |
631 | */ | |
354aa0fb | 632 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 633 | { |
354aa0fb | 634 | return start_transaction(root, 0, TRANS_ATTACH, 0); |
60376ce4 JB |
635 | } |
636 | ||
d4edf39b | 637 | /* |
90b6d283 | 638 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b MX |
639 | * |
640 | * It is similar to the above function, the differentia is this one | |
641 | * will wait for all the inactive transactions until they fully | |
642 | * complete. | |
643 | */ | |
644 | struct btrfs_trans_handle * | |
645 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
646 | { | |
647 | struct btrfs_trans_handle *trans; | |
648 | ||
649 | trans = start_transaction(root, 0, TRANS_ATTACH, 0); | |
650 | if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT) | |
651 | btrfs_wait_for_commit(root, 0); | |
652 | ||
653 | return trans; | |
654 | } | |
655 | ||
d352ac68 | 656 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 657 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
658 | struct btrfs_transaction *commit) |
659 | { | |
4a9d8bde | 660 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
661 | } |
662 | ||
46204592 SW |
663 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
664 | { | |
665 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 666 | int ret = 0; |
46204592 | 667 | |
46204592 SW |
668 | if (transid) { |
669 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 670 | goto out; |
46204592 SW |
671 | |
672 | /* find specified transaction */ | |
a4abeea4 | 673 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
674 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
675 | if (t->transid == transid) { | |
676 | cur_trans = t; | |
a4abeea4 | 677 | atomic_inc(&cur_trans->use_count); |
8cd2807f | 678 | ret = 0; |
46204592 SW |
679 | break; |
680 | } | |
8cd2807f MX |
681 | if (t->transid > transid) { |
682 | ret = 0; | |
46204592 | 683 | break; |
8cd2807f | 684 | } |
46204592 | 685 | } |
a4abeea4 | 686 | spin_unlock(&root->fs_info->trans_lock); |
42383020 SW |
687 | |
688 | /* | |
689 | * The specified transaction doesn't exist, or we | |
690 | * raced with btrfs_commit_transaction | |
691 | */ | |
692 | if (!cur_trans) { | |
693 | if (transid > root->fs_info->last_trans_committed) | |
694 | ret = -EINVAL; | |
8cd2807f | 695 | goto out; |
42383020 | 696 | } |
46204592 SW |
697 | } else { |
698 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 699 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
700 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
701 | list) { | |
4a9d8bde MX |
702 | if (t->state >= TRANS_STATE_COMMIT_START) { |
703 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 704 | break; |
46204592 | 705 | cur_trans = t; |
a4abeea4 | 706 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
707 | break; |
708 | } | |
709 | } | |
a4abeea4 | 710 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 711 | if (!cur_trans) |
a4abeea4 | 712 | goto out; /* nothing committing|committed */ |
46204592 SW |
713 | } |
714 | ||
46204592 | 715 | wait_for_commit(root, cur_trans); |
724e2315 | 716 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 717 | out: |
46204592 SW |
718 | return ret; |
719 | } | |
720 | ||
37d1aeee CM |
721 | void btrfs_throttle(struct btrfs_root *root) |
722 | { | |
a4abeea4 | 723 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 724 | wait_current_trans(root); |
37d1aeee CM |
725 | } |
726 | ||
8929ecfa YZ |
727 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
728 | struct btrfs_root *root) | |
729 | { | |
1be41b78 | 730 | if (root->fs_info->global_block_rsv.space_info->full && |
0a2b2a84 | 731 | btrfs_check_space_for_delayed_refs(trans, root)) |
1be41b78 | 732 | return 1; |
36ba022a | 733 | |
1be41b78 | 734 | return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); |
8929ecfa YZ |
735 | } |
736 | ||
737 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
738 | struct btrfs_root *root) | |
739 | { | |
740 | struct btrfs_transaction *cur_trans = trans->transaction; | |
741 | int updates; | |
49b25e05 | 742 | int err; |
8929ecfa | 743 | |
a4abeea4 | 744 | smp_mb(); |
4a9d8bde MX |
745 | if (cur_trans->state >= TRANS_STATE_BLOCKED || |
746 | cur_trans->delayed_refs.flushing) | |
8929ecfa YZ |
747 | return 1; |
748 | ||
749 | updates = trans->delayed_ref_updates; | |
750 | trans->delayed_ref_updates = 0; | |
49b25e05 | 751 | if (updates) { |
28ed1345 | 752 | err = btrfs_run_delayed_refs(trans, root, updates * 2); |
49b25e05 JM |
753 | if (err) /* Error code will also eval true */ |
754 | return err; | |
755 | } | |
8929ecfa YZ |
756 | |
757 | return should_end_transaction(trans, root); | |
758 | } | |
759 | ||
89ce8a63 | 760 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
a698d075 | 761 | struct btrfs_root *root, int throttle) |
79154b1b | 762 | { |
8929ecfa | 763 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 764 | struct btrfs_fs_info *info = root->fs_info; |
1be41b78 | 765 | unsigned long cur = trans->delayed_ref_updates; |
a698d075 | 766 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 767 | int err = 0; |
a79b7d4b | 768 | int must_run_delayed_refs = 0; |
c3e69d58 | 769 | |
3bbb24b2 JB |
770 | if (trans->use_count > 1) { |
771 | trans->use_count--; | |
2a1eb461 JB |
772 | trans->block_rsv = trans->orig_rsv; |
773 | return 0; | |
774 | } | |
775 | ||
b24e03db | 776 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 777 | trans->block_rsv = NULL; |
c5567237 | 778 | |
ea658bad JB |
779 | if (!list_empty(&trans->new_bgs)) |
780 | btrfs_create_pending_block_groups(trans, root); | |
781 | ||
1be41b78 | 782 | trans->delayed_ref_updates = 0; |
a79b7d4b CM |
783 | if (!trans->sync) { |
784 | must_run_delayed_refs = | |
785 | btrfs_should_throttle_delayed_refs(trans, root); | |
0a2b2a84 | 786 | cur = max_t(unsigned long, cur, 32); |
a79b7d4b CM |
787 | |
788 | /* | |
789 | * don't make the caller wait if they are from a NOLOCK | |
790 | * or ATTACH transaction, it will deadlock with commit | |
791 | */ | |
792 | if (must_run_delayed_refs == 1 && | |
793 | (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH))) | |
794 | must_run_delayed_refs = 2; | |
56bec294 | 795 | } |
bb721703 | 796 | |
0e721106 JB |
797 | btrfs_trans_release_metadata(trans, root); |
798 | trans->block_rsv = NULL; | |
56bec294 | 799 | |
ea658bad JB |
800 | if (!list_empty(&trans->new_bgs)) |
801 | btrfs_create_pending_block_groups(trans, root); | |
802 | ||
4fbcdf66 FM |
803 | btrfs_trans_release_chunk_metadata(trans); |
804 | ||
a4abeea4 | 805 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
4a9d8bde MX |
806 | should_end_transaction(trans, root) && |
807 | ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) { | |
808 | spin_lock(&info->trans_lock); | |
809 | if (cur_trans->state == TRANS_STATE_RUNNING) | |
810 | cur_trans->state = TRANS_STATE_BLOCKED; | |
811 | spin_unlock(&info->trans_lock); | |
a4abeea4 | 812 | } |
8929ecfa | 813 | |
4a9d8bde | 814 | if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
3bbb24b2 | 815 | if (throttle) |
8929ecfa | 816 | return btrfs_commit_transaction(trans, root); |
3bbb24b2 | 817 | else |
8929ecfa YZ |
818 | wake_up_process(info->transaction_kthread); |
819 | } | |
820 | ||
0860adfd | 821 | if (trans->type & __TRANS_FREEZABLE) |
98114659 | 822 | sb_end_intwrite(root->fs_info->sb); |
6df7881a | 823 | |
8929ecfa | 824 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
825 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
826 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 827 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 828 | |
a83342aa DS |
829 | /* |
830 | * Make sure counter is updated before we wake up waiters. | |
831 | */ | |
99d16cbc | 832 | smp_mb(); |
79154b1b CM |
833 | if (waitqueue_active(&cur_trans->writer_wait)) |
834 | wake_up(&cur_trans->writer_wait); | |
724e2315 | 835 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
836 | |
837 | if (current->journal_info == trans) | |
838 | current->journal_info = NULL; | |
ab78c84d | 839 | |
24bbcf04 YZ |
840 | if (throttle) |
841 | btrfs_run_delayed_iputs(root); | |
842 | ||
49b25e05 | 843 | if (trans->aborted || |
4e121c06 JB |
844 | test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) { |
845 | wake_up_process(info->transaction_kthread); | |
4edc2ca3 | 846 | err = -EIO; |
4e121c06 | 847 | } |
edf39272 | 848 | assert_qgroups_uptodate(trans); |
49b25e05 | 849 | |
4edc2ca3 | 850 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
a79b7d4b CM |
851 | if (must_run_delayed_refs) { |
852 | btrfs_async_run_delayed_refs(root, cur, | |
853 | must_run_delayed_refs == 1); | |
854 | } | |
4edc2ca3 | 855 | return err; |
79154b1b CM |
856 | } |
857 | ||
89ce8a63 CM |
858 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
859 | struct btrfs_root *root) | |
860 | { | |
98ad43be | 861 | return __btrfs_end_transaction(trans, root, 0); |
89ce8a63 CM |
862 | } |
863 | ||
864 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
865 | struct btrfs_root *root) | |
866 | { | |
98ad43be | 867 | return __btrfs_end_transaction(trans, root, 1); |
16cdcec7 MX |
868 | } |
869 | ||
d352ac68 CM |
870 | /* |
871 | * when btree blocks are allocated, they have some corresponding bits set for | |
872 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 873 | * those extents are sent to disk but does not wait on them |
d352ac68 | 874 | */ |
690587d1 | 875 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 876 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 877 | { |
777e6bd7 | 878 | int err = 0; |
7c4452b9 | 879 | int werr = 0; |
1728366e | 880 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 881 | struct extent_state *cached_state = NULL; |
777e6bd7 | 882 | u64 start = 0; |
5f39d397 | 883 | u64 end; |
7c4452b9 | 884 | |
1728366e | 885 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 886 | mark, &cached_state)) { |
663dfbb0 FM |
887 | bool wait_writeback = false; |
888 | ||
889 | err = convert_extent_bit(dirty_pages, start, end, | |
890 | EXTENT_NEED_WAIT, | |
891 | mark, &cached_state, GFP_NOFS); | |
892 | /* | |
893 | * convert_extent_bit can return -ENOMEM, which is most of the | |
894 | * time a temporary error. So when it happens, ignore the error | |
895 | * and wait for writeback of this range to finish - because we | |
896 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
897 | * to btrfs_wait_marked_extents() would not know that writeback | |
898 | * for this range started and therefore wouldn't wait for it to | |
899 | * finish - we don't want to commit a superblock that points to | |
900 | * btree nodes/leafs for which writeback hasn't finished yet | |
901 | * (and without errors). | |
902 | * We cleanup any entries left in the io tree when committing | |
903 | * the transaction (through clear_btree_io_tree()). | |
904 | */ | |
905 | if (err == -ENOMEM) { | |
906 | err = 0; | |
907 | wait_writeback = true; | |
908 | } | |
909 | if (!err) | |
910 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
911 | if (err) |
912 | werr = err; | |
663dfbb0 FM |
913 | else if (wait_writeback) |
914 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 915 | free_extent_state(cached_state); |
663dfbb0 | 916 | cached_state = NULL; |
1728366e JB |
917 | cond_resched(); |
918 | start = end + 1; | |
7c4452b9 | 919 | } |
690587d1 CM |
920 | return werr; |
921 | } | |
922 | ||
923 | /* | |
924 | * when btree blocks are allocated, they have some corresponding bits set for | |
925 | * them in one of two extent_io trees. This is used to make sure all of | |
926 | * those extents are on disk for transaction or log commit. We wait | |
927 | * on all the pages and clear them from the dirty pages state tree | |
928 | */ | |
929 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 930 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 931 | { |
690587d1 CM |
932 | int err = 0; |
933 | int werr = 0; | |
1728366e | 934 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 935 | struct extent_state *cached_state = NULL; |
690587d1 CM |
936 | u64 start = 0; |
937 | u64 end; | |
656f30db FM |
938 | struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode); |
939 | bool errors = false; | |
777e6bd7 | 940 | |
1728366e | 941 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 942 | EXTENT_NEED_WAIT, &cached_state)) { |
663dfbb0 FM |
943 | /* |
944 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
945 | * When committing the transaction, we'll remove any entries | |
946 | * left in the io tree. For a log commit, we don't remove them | |
947 | * after committing the log because the tree can be accessed | |
948 | * concurrently - we do it only at transaction commit time when | |
949 | * it's safe to do it (through clear_btree_io_tree()). | |
950 | */ | |
951 | err = clear_extent_bit(dirty_pages, start, end, | |
952 | EXTENT_NEED_WAIT, | |
953 | 0, 0, &cached_state, GFP_NOFS); | |
954 | if (err == -ENOMEM) | |
955 | err = 0; | |
956 | if (!err) | |
957 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
958 | if (err) |
959 | werr = err; | |
e38e2ed7 FM |
960 | free_extent_state(cached_state); |
961 | cached_state = NULL; | |
1728366e JB |
962 | cond_resched(); |
963 | start = end + 1; | |
777e6bd7 | 964 | } |
7c4452b9 CM |
965 | if (err) |
966 | werr = err; | |
656f30db FM |
967 | |
968 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { | |
969 | if ((mark & EXTENT_DIRTY) && | |
970 | test_and_clear_bit(BTRFS_INODE_BTREE_LOG1_ERR, | |
971 | &btree_ino->runtime_flags)) | |
972 | errors = true; | |
973 | ||
974 | if ((mark & EXTENT_NEW) && | |
975 | test_and_clear_bit(BTRFS_INODE_BTREE_LOG2_ERR, | |
976 | &btree_ino->runtime_flags)) | |
977 | errors = true; | |
978 | } else { | |
979 | if (test_and_clear_bit(BTRFS_INODE_BTREE_ERR, | |
980 | &btree_ino->runtime_flags)) | |
981 | errors = true; | |
982 | } | |
983 | ||
984 | if (errors && !werr) | |
985 | werr = -EIO; | |
986 | ||
7c4452b9 | 987 | return werr; |
79154b1b CM |
988 | } |
989 | ||
690587d1 CM |
990 | /* |
991 | * when btree blocks are allocated, they have some corresponding bits set for | |
992 | * them in one of two extent_io trees. This is used to make sure all of | |
993 | * those extents are on disk for transaction or log commit | |
994 | */ | |
171170c1 | 995 | static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, |
8cef4e16 | 996 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
997 | { |
998 | int ret; | |
999 | int ret2; | |
c6adc9cc | 1000 | struct blk_plug plug; |
690587d1 | 1001 | |
c6adc9cc | 1002 | blk_start_plug(&plug); |
8cef4e16 | 1003 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
c6adc9cc | 1004 | blk_finish_plug(&plug); |
8cef4e16 | 1005 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); |
bf0da8c1 CM |
1006 | |
1007 | if (ret) | |
1008 | return ret; | |
1009 | if (ret2) | |
1010 | return ret2; | |
1011 | return 0; | |
690587d1 CM |
1012 | } |
1013 | ||
663dfbb0 | 1014 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
d0c803c4 CM |
1015 | struct btrfs_root *root) |
1016 | { | |
663dfbb0 FM |
1017 | int ret; |
1018 | ||
1019 | ret = btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
1020 | &trans->transaction->dirty_pages, |
1021 | EXTENT_DIRTY); | |
663dfbb0 FM |
1022 | clear_btree_io_tree(&trans->transaction->dirty_pages); |
1023 | ||
1024 | return ret; | |
d0c803c4 CM |
1025 | } |
1026 | ||
d352ac68 CM |
1027 | /* |
1028 | * this is used to update the root pointer in the tree of tree roots. | |
1029 | * | |
1030 | * But, in the case of the extent allocation tree, updating the root | |
1031 | * pointer may allocate blocks which may change the root of the extent | |
1032 | * allocation tree. | |
1033 | * | |
1034 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1035 | * change while the root pointer was being updated in the metadata. | |
1036 | */ | |
0b86a832 CM |
1037 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1038 | struct btrfs_root *root) | |
79154b1b CM |
1039 | { |
1040 | int ret; | |
0b86a832 | 1041 | u64 old_root_bytenr; |
86b9f2ec | 1042 | u64 old_root_used; |
0b86a832 | 1043 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 1044 | |
86b9f2ec | 1045 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1046 | |
d397712b | 1047 | while (1) { |
0b86a832 | 1048 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1049 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1050 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1051 | break; |
87ef2bb4 | 1052 | |
5d4f98a2 | 1053 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1054 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1055 | &root->root_key, |
1056 | &root->root_item); | |
49b25e05 JM |
1057 | if (ret) |
1058 | return ret; | |
56bec294 | 1059 | |
86b9f2ec | 1060 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1061 | } |
276e680d | 1062 | |
0b86a832 CM |
1063 | return 0; |
1064 | } | |
1065 | ||
d352ac68 CM |
1066 | /* |
1067 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1068 | * |
1069 | * The error handling in this function may not be obvious. Any of the | |
1070 | * failures will cause the file system to go offline. We still need | |
1071 | * to clean up the delayed refs. | |
d352ac68 | 1072 | */ |
5d4f98a2 YZ |
1073 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
1074 | struct btrfs_root *root) | |
0b86a832 CM |
1075 | { |
1076 | struct btrfs_fs_info *fs_info = root->fs_info; | |
ea526d18 | 1077 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1078 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1079 | struct list_head *next; |
84234f3a | 1080 | struct extent_buffer *eb; |
56bec294 | 1081 | int ret; |
84234f3a YZ |
1082 | |
1083 | eb = btrfs_lock_root_node(fs_info->tree_root); | |
49b25e05 JM |
1084 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
1085 | 0, &eb); | |
84234f3a YZ |
1086 | btrfs_tree_unlock(eb); |
1087 | free_extent_buffer(eb); | |
0b86a832 | 1088 | |
49b25e05 JM |
1089 | if (ret) |
1090 | return ret; | |
1091 | ||
56bec294 | 1092 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1093 | if (ret) |
1094 | return ret; | |
87ef2bb4 | 1095 | |
733f4fbb | 1096 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
c16ce190 JB |
1097 | if (ret) |
1098 | return ret; | |
8dabb742 | 1099 | ret = btrfs_run_dev_replace(trans, root->fs_info); |
c16ce190 JB |
1100 | if (ret) |
1101 | return ret; | |
546adb0d | 1102 | ret = btrfs_run_qgroups(trans, root->fs_info); |
c16ce190 JB |
1103 | if (ret) |
1104 | return ret; | |
546adb0d | 1105 | |
dcdf7f6d JB |
1106 | ret = btrfs_setup_space_cache(trans, root); |
1107 | if (ret) | |
1108 | return ret; | |
1109 | ||
546adb0d JS |
1110 | /* run_qgroups might have added some more refs */ |
1111 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
c16ce190 JB |
1112 | if (ret) |
1113 | return ret; | |
ea526d18 | 1114 | again: |
d397712b | 1115 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
1116 | next = fs_info->dirty_cowonly_roots.next; |
1117 | list_del_init(next); | |
1118 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1119 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1120 | |
9e351cc8 JB |
1121 | if (root != fs_info->extent_root) |
1122 | list_add_tail(&root->dirty_list, | |
1123 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1124 | ret = update_cowonly_root(trans, root); |
1125 | if (ret) | |
1126 | return ret; | |
ea526d18 JB |
1127 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1128 | if (ret) | |
1129 | return ret; | |
79154b1b | 1130 | } |
276e680d | 1131 | |
1bbc621e | 1132 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
ea526d18 JB |
1133 | ret = btrfs_write_dirty_block_groups(trans, root); |
1134 | if (ret) | |
1135 | return ret; | |
1136 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
1137 | if (ret) | |
1138 | return ret; | |
1139 | } | |
1140 | ||
1141 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1142 | goto again; | |
1143 | ||
9e351cc8 JB |
1144 | list_add_tail(&fs_info->extent_root->dirty_list, |
1145 | &trans->transaction->switch_commits); | |
8dabb742 SB |
1146 | btrfs_after_dev_replace_commit(fs_info); |
1147 | ||
79154b1b CM |
1148 | return 0; |
1149 | } | |
1150 | ||
d352ac68 CM |
1151 | /* |
1152 | * dead roots are old snapshots that need to be deleted. This allocates | |
1153 | * a dirty root struct and adds it into the list of dead roots that need to | |
1154 | * be deleted | |
1155 | */ | |
cfad392b | 1156 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1157 | { |
a4abeea4 | 1158 | spin_lock(&root->fs_info->trans_lock); |
cfad392b JB |
1159 | if (list_empty(&root->root_list)) |
1160 | list_add_tail(&root->root_list, &root->fs_info->dead_roots); | |
a4abeea4 | 1161 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
1162 | } |
1163 | ||
d352ac68 | 1164 | /* |
5d4f98a2 | 1165 | * update all the cowonly tree roots on disk |
d352ac68 | 1166 | */ |
5d4f98a2 YZ |
1167 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
1168 | struct btrfs_root *root) | |
0f7d52f4 | 1169 | { |
0f7d52f4 | 1170 | struct btrfs_root *gang[8]; |
5d4f98a2 | 1171 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
1172 | int i; |
1173 | int ret; | |
54aa1f4d CM |
1174 | int err = 0; |
1175 | ||
a4abeea4 | 1176 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1177 | while (1) { |
5d4f98a2 YZ |
1178 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1179 | (void **)gang, 0, | |
0f7d52f4 CM |
1180 | ARRAY_SIZE(gang), |
1181 | BTRFS_ROOT_TRANS_TAG); | |
1182 | if (ret == 0) | |
1183 | break; | |
1184 | for (i = 0; i < ret; i++) { | |
1185 | root = gang[i]; | |
5d4f98a2 YZ |
1186 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1187 | (unsigned long)root->root_key.objectid, | |
1188 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1189 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1190 | |
e02119d5 | 1191 | btrfs_free_log(trans, root); |
5d4f98a2 | 1192 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 1193 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 1194 | |
82d5902d LZ |
1195 | btrfs_save_ino_cache(root, trans); |
1196 | ||
f1ebcc74 | 1197 | /* see comments in should_cow_block() */ |
27cdeb70 | 1198 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
c7548af6 | 1199 | smp_mb__after_atomic(); |
f1ebcc74 | 1200 | |
978d910d | 1201 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1202 | list_add_tail(&root->dirty_list, |
1203 | &trans->transaction->switch_commits); | |
978d910d YZ |
1204 | btrfs_set_root_node(&root->root_item, |
1205 | root->node); | |
1206 | } | |
5d4f98a2 | 1207 | |
5d4f98a2 | 1208 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1209 | &root->root_key, |
1210 | &root->root_item); | |
a4abeea4 | 1211 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1212 | if (err) |
1213 | break; | |
7174109c | 1214 | btrfs_qgroup_free_meta_all(root); |
0f7d52f4 CM |
1215 | } |
1216 | } | |
a4abeea4 | 1217 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1218 | return err; |
0f7d52f4 CM |
1219 | } |
1220 | ||
d352ac68 | 1221 | /* |
de78b51a ES |
1222 | * defrag a given btree. |
1223 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1224 | */ |
de78b51a | 1225 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1226 | { |
1227 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1228 | struct btrfs_trans_handle *trans; |
8929ecfa | 1229 | int ret; |
e9d0b13b | 1230 | |
27cdeb70 | 1231 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1232 | return 0; |
8929ecfa | 1233 | |
6b80053d | 1234 | while (1) { |
8929ecfa YZ |
1235 | trans = btrfs_start_transaction(root, 0); |
1236 | if (IS_ERR(trans)) | |
1237 | return PTR_ERR(trans); | |
1238 | ||
de78b51a | 1239 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1240 | |
e9d0b13b | 1241 | btrfs_end_transaction(trans, root); |
b53d3f5d | 1242 | btrfs_btree_balance_dirty(info->tree_root); |
e9d0b13b CM |
1243 | cond_resched(); |
1244 | ||
7841cb28 | 1245 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b | 1246 | break; |
210549eb DS |
1247 | |
1248 | if (btrfs_defrag_cancelled(root->fs_info)) { | |
efe120a0 | 1249 | pr_debug("BTRFS: defrag_root cancelled\n"); |
210549eb DS |
1250 | ret = -EAGAIN; |
1251 | break; | |
1252 | } | |
e9d0b13b | 1253 | } |
27cdeb70 | 1254 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1255 | return ret; |
e9d0b13b CM |
1256 | } |
1257 | ||
d352ac68 CM |
1258 | /* |
1259 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1260 | * transaction commit. This does the actual creation. |
1261 | * | |
1262 | * Note: | |
1263 | * If the error which may affect the commitment of the current transaction | |
1264 | * happens, we should return the error number. If the error which just affect | |
1265 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1266 | */ |
80b6794d | 1267 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1268 | struct btrfs_fs_info *fs_info, |
1269 | struct btrfs_pending_snapshot *pending) | |
1270 | { | |
1271 | struct btrfs_key key; | |
80b6794d | 1272 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1273 | struct btrfs_root *tree_root = fs_info->tree_root; |
1274 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1275 | struct btrfs_root *parent_root; |
98c9942a | 1276 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1277 | struct inode *parent_inode; |
42874b3d MX |
1278 | struct btrfs_path *path; |
1279 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1280 | struct dentry *dentry; |
3063d29f | 1281 | struct extent_buffer *tmp; |
925baedd | 1282 | struct extent_buffer *old; |
8ea05e3a | 1283 | struct timespec cur_time = CURRENT_TIME; |
aec8030a | 1284 | int ret = 0; |
d68fc57b | 1285 | u64 to_reserve = 0; |
6bdb72de | 1286 | u64 index = 0; |
a22285a6 | 1287 | u64 objectid; |
b83cc969 | 1288 | u64 root_flags; |
8ea05e3a | 1289 | uuid_le new_uuid; |
3063d29f | 1290 | |
42874b3d MX |
1291 | path = btrfs_alloc_path(); |
1292 | if (!path) { | |
aec8030a MX |
1293 | pending->error = -ENOMEM; |
1294 | return 0; | |
42874b3d MX |
1295 | } |
1296 | ||
80b6794d CM |
1297 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
1298 | if (!new_root_item) { | |
aec8030a | 1299 | pending->error = -ENOMEM; |
6fa9700e | 1300 | goto root_item_alloc_fail; |
80b6794d | 1301 | } |
a22285a6 | 1302 | |
aec8030a MX |
1303 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1304 | if (pending->error) | |
6fa9700e | 1305 | goto no_free_objectid; |
3063d29f | 1306 | |
d6726335 QW |
1307 | /* |
1308 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1309 | * accounted by later btrfs_qgroup_inherit(). | |
1310 | */ | |
1311 | btrfs_set_skip_qgroup(trans, objectid); | |
1312 | ||
147d256e | 1313 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
d68fc57b YZ |
1314 | |
1315 | if (to_reserve > 0) { | |
aec8030a MX |
1316 | pending->error = btrfs_block_rsv_add(root, |
1317 | &pending->block_rsv, | |
1318 | to_reserve, | |
1319 | BTRFS_RESERVE_NO_FLUSH); | |
1320 | if (pending->error) | |
d6726335 | 1321 | goto clear_skip_qgroup; |
d68fc57b YZ |
1322 | } |
1323 | ||
3063d29f | 1324 | key.objectid = objectid; |
a22285a6 YZ |
1325 | key.offset = (u64)-1; |
1326 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1327 | |
6fa9700e | 1328 | rsv = trans->block_rsv; |
a22285a6 | 1329 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1330 | trans->bytes_reserved = trans->block_rsv->reserved; |
3de4586c | 1331 | |
a22285a6 | 1332 | dentry = pending->dentry; |
e9662f70 | 1333 | parent_inode = pending->dir; |
a22285a6 | 1334 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 1335 | record_root_in_trans(trans, parent_root); |
a22285a6 | 1336 | |
3063d29f CM |
1337 | /* |
1338 | * insert the directory item | |
1339 | */ | |
3de4586c | 1340 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1341 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1342 | |
1343 | /* check if there is a file/dir which has the same name. */ | |
1344 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
1345 | btrfs_ino(parent_inode), | |
1346 | dentry->d_name.name, | |
1347 | dentry->d_name.len, 0); | |
1348 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1349 | pending->error = -EEXIST; |
aec8030a | 1350 | goto dir_item_existed; |
42874b3d MX |
1351 | } else if (IS_ERR(dir_item)) { |
1352 | ret = PTR_ERR(dir_item); | |
8732d44f MX |
1353 | btrfs_abort_transaction(trans, root, ret); |
1354 | goto fail; | |
79787eaa | 1355 | } |
42874b3d | 1356 | btrfs_release_path(path); |
52c26179 | 1357 | |
e999376f CM |
1358 | /* |
1359 | * pull in the delayed directory update | |
1360 | * and the delayed inode item | |
1361 | * otherwise we corrupt the FS during | |
1362 | * snapshot | |
1363 | */ | |
1364 | ret = btrfs_run_delayed_items(trans, root); | |
8732d44f MX |
1365 | if (ret) { /* Transaction aborted */ |
1366 | btrfs_abort_transaction(trans, root, ret); | |
1367 | goto fail; | |
1368 | } | |
e999376f | 1369 | |
7585717f | 1370 | record_root_in_trans(trans, root); |
6bdb72de SW |
1371 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1372 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1373 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1374 | |
b83cc969 LZ |
1375 | root_flags = btrfs_root_flags(new_root_item); |
1376 | if (pending->readonly) | |
1377 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1378 | else | |
1379 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1380 | btrfs_set_root_flags(new_root_item, root_flags); | |
1381 | ||
8ea05e3a AB |
1382 | btrfs_set_root_generation_v2(new_root_item, |
1383 | trans->transid); | |
1384 | uuid_le_gen(&new_uuid); | |
1385 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1386 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1387 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1388 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1389 | memset(new_root_item->received_uuid, 0, | |
1390 | sizeof(new_root_item->received_uuid)); | |
1391 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1392 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1393 | btrfs_set_root_stransid(new_root_item, 0); | |
1394 | btrfs_set_root_rtransid(new_root_item, 0); | |
1395 | } | |
3cae210f QW |
1396 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1397 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1398 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1399 | |
6bdb72de | 1400 | old = btrfs_lock_root_node(root); |
49b25e05 | 1401 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1402 | if (ret) { |
1403 | btrfs_tree_unlock(old); | |
1404 | free_extent_buffer(old); | |
8732d44f MX |
1405 | btrfs_abort_transaction(trans, root, ret); |
1406 | goto fail; | |
79787eaa | 1407 | } |
49b25e05 | 1408 | |
6bdb72de SW |
1409 | btrfs_set_lock_blocking(old); |
1410 | ||
49b25e05 | 1411 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1412 | /* clean up in any case */ |
6bdb72de SW |
1413 | btrfs_tree_unlock(old); |
1414 | free_extent_buffer(old); | |
8732d44f MX |
1415 | if (ret) { |
1416 | btrfs_abort_transaction(trans, root, ret); | |
1417 | goto fail; | |
1418 | } | |
f1ebcc74 | 1419 | /* see comments in should_cow_block() */ |
27cdeb70 | 1420 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1421 | smp_wmb(); |
1422 | ||
6bdb72de | 1423 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1424 | /* record when the snapshot was created in key.offset */ |
1425 | key.offset = trans->transid; | |
1426 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1427 | btrfs_tree_unlock(tmp); |
1428 | free_extent_buffer(tmp); | |
8732d44f MX |
1429 | if (ret) { |
1430 | btrfs_abort_transaction(trans, root, ret); | |
1431 | goto fail; | |
1432 | } | |
6bdb72de | 1433 | |
a22285a6 YZ |
1434 | /* |
1435 | * insert root back/forward references | |
1436 | */ | |
1437 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1438 | parent_root->root_key.objectid, |
33345d01 | 1439 | btrfs_ino(parent_inode), index, |
a22285a6 | 1440 | dentry->d_name.name, dentry->d_name.len); |
8732d44f MX |
1441 | if (ret) { |
1442 | btrfs_abort_transaction(trans, root, ret); | |
1443 | goto fail; | |
1444 | } | |
0660b5af | 1445 | |
a22285a6 YZ |
1446 | key.offset = (u64)-1; |
1447 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1448 | if (IS_ERR(pending->snap)) { |
1449 | ret = PTR_ERR(pending->snap); | |
8732d44f MX |
1450 | btrfs_abort_transaction(trans, root, ret); |
1451 | goto fail; | |
79787eaa | 1452 | } |
d68fc57b | 1453 | |
49b25e05 | 1454 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f MX |
1455 | if (ret) { |
1456 | btrfs_abort_transaction(trans, root, ret); | |
1457 | goto fail; | |
1458 | } | |
361048f5 MX |
1459 | |
1460 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
8732d44f MX |
1461 | if (ret) { |
1462 | btrfs_abort_transaction(trans, root, ret); | |
1463 | goto fail; | |
1464 | } | |
42874b3d MX |
1465 | |
1466 | ret = btrfs_insert_dir_item(trans, parent_root, | |
1467 | dentry->d_name.name, dentry->d_name.len, | |
1468 | parent_inode, &key, | |
1469 | BTRFS_FT_DIR, index); | |
1470 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1471 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f MX |
1472 | if (ret) { |
1473 | btrfs_abort_transaction(trans, root, ret); | |
1474 | goto fail; | |
1475 | } | |
42874b3d MX |
1476 | |
1477 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
1478 | dentry->d_name.len * 2); | |
1479 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; | |
be6aef60 | 1480 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
dd5f9615 SB |
1481 | if (ret) { |
1482 | btrfs_abort_transaction(trans, root, ret); | |
1483 | goto fail; | |
1484 | } | |
1485 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b, | |
1486 | BTRFS_UUID_KEY_SUBVOL, objectid); | |
1487 | if (ret) { | |
8732d44f | 1488 | btrfs_abort_transaction(trans, root, ret); |
dd5f9615 SB |
1489 | goto fail; |
1490 | } | |
1491 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
1492 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, | |
1493 | new_root_item->received_uuid, | |
1494 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
1495 | objectid); | |
1496 | if (ret && ret != -EEXIST) { | |
1497 | btrfs_abort_transaction(trans, root, ret); | |
1498 | goto fail; | |
1499 | } | |
1500 | } | |
d6726335 QW |
1501 | |
1502 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
1503 | if (ret) { | |
1504 | btrfs_abort_transaction(trans, root, ret); | |
1505 | goto fail; | |
1506 | } | |
1507 | ||
1508 | /* | |
1509 | * account qgroup counters before qgroup_inherit() | |
1510 | */ | |
1511 | ret = btrfs_qgroup_prepare_account_extents(trans, fs_info); | |
1512 | if (ret) | |
1513 | goto fail; | |
1514 | ret = btrfs_qgroup_account_extents(trans, fs_info); | |
1515 | if (ret) | |
1516 | goto fail; | |
1517 | ret = btrfs_qgroup_inherit(trans, fs_info, | |
1518 | root->root_key.objectid, | |
1519 | objectid, pending->inherit); | |
1520 | if (ret) { | |
1521 | btrfs_abort_transaction(trans, root, ret); | |
1522 | goto fail; | |
1523 | } | |
1524 | ||
3063d29f | 1525 | fail: |
aec8030a MX |
1526 | pending->error = ret; |
1527 | dir_item_existed: | |
98c9942a | 1528 | trans->block_rsv = rsv; |
2382c5cc | 1529 | trans->bytes_reserved = 0; |
d6726335 QW |
1530 | clear_skip_qgroup: |
1531 | btrfs_clear_skip_qgroup(trans); | |
6fa9700e MX |
1532 | no_free_objectid: |
1533 | kfree(new_root_item); | |
1534 | root_item_alloc_fail: | |
42874b3d | 1535 | btrfs_free_path(path); |
49b25e05 | 1536 | return ret; |
3063d29f CM |
1537 | } |
1538 | ||
d352ac68 CM |
1539 | /* |
1540 | * create all the snapshots we've scheduled for creation | |
1541 | */ | |
80b6794d CM |
1542 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1543 | struct btrfs_fs_info *fs_info) | |
3de4586c | 1544 | { |
aec8030a | 1545 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1546 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1547 | int ret = 0; |
3de4586c | 1548 | |
aec8030a MX |
1549 | list_for_each_entry_safe(pending, next, head, list) { |
1550 | list_del(&pending->list); | |
1551 | ret = create_pending_snapshot(trans, fs_info, pending); | |
1552 | if (ret) | |
1553 | break; | |
1554 | } | |
1555 | return ret; | |
3de4586c CM |
1556 | } |
1557 | ||
5d4f98a2 YZ |
1558 | static void update_super_roots(struct btrfs_root *root) |
1559 | { | |
1560 | struct btrfs_root_item *root_item; | |
1561 | struct btrfs_super_block *super; | |
1562 | ||
6c41761f | 1563 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1564 | |
1565 | root_item = &root->fs_info->chunk_root->root_item; | |
1566 | super->chunk_root = root_item->bytenr; | |
1567 | super->chunk_root_generation = root_item->generation; | |
1568 | super->chunk_root_level = root_item->level; | |
1569 | ||
1570 | root_item = &root->fs_info->tree_root->root_item; | |
1571 | super->root = root_item->bytenr; | |
1572 | super->generation = root_item->generation; | |
1573 | super->root_level = root_item->level; | |
73bc1876 | 1574 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1575 | super->cache_generation = root_item->generation; |
70f80175 SB |
1576 | if (root->fs_info->update_uuid_tree_gen) |
1577 | super->uuid_tree_generation = root_item->generation; | |
5d4f98a2 YZ |
1578 | } |
1579 | ||
f36f3042 CM |
1580 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1581 | { | |
4a9d8bde | 1582 | struct btrfs_transaction *trans; |
f36f3042 | 1583 | int ret = 0; |
4a9d8bde | 1584 | |
a4abeea4 | 1585 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1586 | trans = info->running_transaction; |
1587 | if (trans) | |
1588 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1589 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1590 | return ret; |
1591 | } | |
1592 | ||
8929ecfa YZ |
1593 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1594 | { | |
4a9d8bde | 1595 | struct btrfs_transaction *trans; |
8929ecfa | 1596 | int ret = 0; |
4a9d8bde | 1597 | |
a4abeea4 | 1598 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1599 | trans = info->running_transaction; |
1600 | if (trans) | |
1601 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1602 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1603 | return ret; |
1604 | } | |
1605 | ||
bb9c12c9 SW |
1606 | /* |
1607 | * wait for the current transaction commit to start and block subsequent | |
1608 | * transaction joins | |
1609 | */ | |
1610 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1611 | struct btrfs_transaction *trans) | |
1612 | { | |
4a9d8bde | 1613 | wait_event(root->fs_info->transaction_blocked_wait, |
501407aa JB |
1614 | trans->state >= TRANS_STATE_COMMIT_START || |
1615 | trans->aborted); | |
bb9c12c9 SW |
1616 | } |
1617 | ||
1618 | /* | |
1619 | * wait for the current transaction to start and then become unblocked. | |
1620 | * caller holds ref. | |
1621 | */ | |
1622 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1623 | struct btrfs_transaction *trans) | |
1624 | { | |
72d63ed6 | 1625 | wait_event(root->fs_info->transaction_wait, |
501407aa JB |
1626 | trans->state >= TRANS_STATE_UNBLOCKED || |
1627 | trans->aborted); | |
bb9c12c9 SW |
1628 | } |
1629 | ||
1630 | /* | |
1631 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1632 | * returns, any subsequent transaction will not be allowed to join. | |
1633 | */ | |
1634 | struct btrfs_async_commit { | |
1635 | struct btrfs_trans_handle *newtrans; | |
1636 | struct btrfs_root *root; | |
7892b5af | 1637 | struct work_struct work; |
bb9c12c9 SW |
1638 | }; |
1639 | ||
1640 | static void do_async_commit(struct work_struct *work) | |
1641 | { | |
1642 | struct btrfs_async_commit *ac = | |
7892b5af | 1643 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1644 | |
6fc4e354 SW |
1645 | /* |
1646 | * We've got freeze protection passed with the transaction. | |
1647 | * Tell lockdep about it. | |
1648 | */ | |
b1a06a4b | 1649 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
bee9182d | 1650 | __sb_writers_acquired(ac->root->fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1651 | |
e209db7a SW |
1652 | current->journal_info = ac->newtrans; |
1653 | ||
bb9c12c9 SW |
1654 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1655 | kfree(ac); | |
1656 | } | |
1657 | ||
1658 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1659 | struct btrfs_root *root, | |
1660 | int wait_for_unblock) | |
1661 | { | |
1662 | struct btrfs_async_commit *ac; | |
1663 | struct btrfs_transaction *cur_trans; | |
1664 | ||
1665 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1666 | if (!ac) |
1667 | return -ENOMEM; | |
bb9c12c9 | 1668 | |
7892b5af | 1669 | INIT_WORK(&ac->work, do_async_commit); |
bb9c12c9 | 1670 | ac->root = root; |
7a7eaa40 | 1671 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1672 | if (IS_ERR(ac->newtrans)) { |
1673 | int err = PTR_ERR(ac->newtrans); | |
1674 | kfree(ac); | |
1675 | return err; | |
1676 | } | |
bb9c12c9 SW |
1677 | |
1678 | /* take transaction reference */ | |
bb9c12c9 | 1679 | cur_trans = trans->transaction; |
13c5a93e | 1680 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1681 | |
1682 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1683 | |
1684 | /* | |
1685 | * Tell lockdep we've released the freeze rwsem, since the | |
1686 | * async commit thread will be the one to unlock it. | |
1687 | */ | |
b1a06a4b | 1688 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
bee9182d | 1689 | __sb_writers_release(root->fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1690 | |
7892b5af | 1691 | schedule_work(&ac->work); |
bb9c12c9 SW |
1692 | |
1693 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1694 | if (wait_for_unblock) |
1695 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1696 | else | |
1697 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1698 | |
38e88054 SW |
1699 | if (current->journal_info == trans) |
1700 | current->journal_info = NULL; | |
1701 | ||
724e2315 | 1702 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1703 | return 0; |
1704 | } | |
1705 | ||
49b25e05 JM |
1706 | |
1707 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1708 | struct btrfs_root *root, int err) |
49b25e05 JM |
1709 | { |
1710 | struct btrfs_transaction *cur_trans = trans->transaction; | |
f094ac32 | 1711 | DEFINE_WAIT(wait); |
49b25e05 JM |
1712 | |
1713 | WARN_ON(trans->use_count > 1); | |
1714 | ||
7b8b92af JB |
1715 | btrfs_abort_transaction(trans, root, err); |
1716 | ||
49b25e05 | 1717 | spin_lock(&root->fs_info->trans_lock); |
66b6135b | 1718 | |
25d8c284 MX |
1719 | /* |
1720 | * If the transaction is removed from the list, it means this | |
1721 | * transaction has been committed successfully, so it is impossible | |
1722 | * to call the cleanup function. | |
1723 | */ | |
1724 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1725 | |
49b25e05 | 1726 | list_del_init(&cur_trans->list); |
d7096fc3 | 1727 | if (cur_trans == root->fs_info->running_transaction) { |
4a9d8bde | 1728 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
f094ac32 LB |
1729 | spin_unlock(&root->fs_info->trans_lock); |
1730 | wait_event(cur_trans->writer_wait, | |
1731 | atomic_read(&cur_trans->num_writers) == 1); | |
1732 | ||
1733 | spin_lock(&root->fs_info->trans_lock); | |
d7096fc3 | 1734 | } |
49b25e05 JM |
1735 | spin_unlock(&root->fs_info->trans_lock); |
1736 | ||
1737 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1738 | ||
4a9d8bde MX |
1739 | spin_lock(&root->fs_info->trans_lock); |
1740 | if (cur_trans == root->fs_info->running_transaction) | |
1741 | root->fs_info->running_transaction = NULL; | |
1742 | spin_unlock(&root->fs_info->trans_lock); | |
1743 | ||
e0228285 JB |
1744 | if (trans->type & __TRANS_FREEZABLE) |
1745 | sb_end_intwrite(root->fs_info->sb); | |
724e2315 JB |
1746 | btrfs_put_transaction(cur_trans); |
1747 | btrfs_put_transaction(cur_trans); | |
49b25e05 JM |
1748 | |
1749 | trace_btrfs_transaction_commit(root); | |
1750 | ||
49b25e05 JM |
1751 | if (current->journal_info == trans) |
1752 | current->journal_info = NULL; | |
c0af8f0b | 1753 | btrfs_scrub_cancel(root->fs_info); |
49b25e05 JM |
1754 | |
1755 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1756 | } | |
1757 | ||
82436617 MX |
1758 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
1759 | { | |
1760 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
6c255e67 | 1761 | return btrfs_start_delalloc_roots(fs_info, 1, -1); |
82436617 MX |
1762 | return 0; |
1763 | } | |
1764 | ||
1765 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) | |
1766 | { | |
1767 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
b0244199 | 1768 | btrfs_wait_ordered_roots(fs_info, -1); |
82436617 MX |
1769 | } |
1770 | ||
50d9aa99 | 1771 | static inline void |
161c3549 | 1772 | btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans) |
50d9aa99 | 1773 | { |
161c3549 JB |
1774 | wait_event(cur_trans->pending_wait, |
1775 | atomic_read(&cur_trans->pending_ordered) == 0); | |
50d9aa99 JB |
1776 | } |
1777 | ||
79154b1b CM |
1778 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1779 | struct btrfs_root *root) | |
1780 | { | |
49b25e05 | 1781 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1782 | struct btrfs_transaction *prev_trans = NULL; |
656f30db | 1783 | struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode); |
25287e0a | 1784 | int ret; |
79154b1b | 1785 | |
8d25a086 MX |
1786 | /* Stop the commit early if ->aborted is set */ |
1787 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
25287e0a | 1788 | ret = cur_trans->aborted; |
e4a2bcac JB |
1789 | btrfs_end_transaction(trans, root); |
1790 | return ret; | |
25287e0a | 1791 | } |
49b25e05 | 1792 | |
56bec294 CM |
1793 | /* make a pass through all the delayed refs we have so far |
1794 | * any runnings procs may add more while we are here | |
1795 | */ | |
1796 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
e4a2bcac JB |
1797 | if (ret) { |
1798 | btrfs_end_transaction(trans, root); | |
1799 | return ret; | |
1800 | } | |
56bec294 | 1801 | |
0e721106 JB |
1802 | btrfs_trans_release_metadata(trans, root); |
1803 | trans->block_rsv = NULL; | |
1804 | ||
b7ec40d7 | 1805 | cur_trans = trans->transaction; |
49b25e05 | 1806 | |
56bec294 CM |
1807 | /* |
1808 | * set the flushing flag so procs in this transaction have to | |
1809 | * start sending their work down. | |
1810 | */ | |
b7ec40d7 | 1811 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 1812 | smp_wmb(); |
56bec294 | 1813 | |
ea658bad JB |
1814 | if (!list_empty(&trans->new_bgs)) |
1815 | btrfs_create_pending_block_groups(trans, root); | |
1816 | ||
c3e69d58 | 1817 | ret = btrfs_run_delayed_refs(trans, root, 0); |
e4a2bcac JB |
1818 | if (ret) { |
1819 | btrfs_end_transaction(trans, root); | |
1820 | return ret; | |
1821 | } | |
56bec294 | 1822 | |
3204d33c | 1823 | if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { |
1bbc621e CM |
1824 | int run_it = 0; |
1825 | ||
1826 | /* this mutex is also taken before trying to set | |
1827 | * block groups readonly. We need to make sure | |
1828 | * that nobody has set a block group readonly | |
1829 | * after a extents from that block group have been | |
1830 | * allocated for cache files. btrfs_set_block_group_ro | |
1831 | * will wait for the transaction to commit if it | |
3204d33c | 1832 | * finds BTRFS_TRANS_DIRTY_BG_RUN set. |
1bbc621e | 1833 | * |
3204d33c JB |
1834 | * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure |
1835 | * only one process starts all the block group IO. It wouldn't | |
1bbc621e CM |
1836 | * hurt to have more than one go through, but there's no |
1837 | * real advantage to it either. | |
1838 | */ | |
1839 | mutex_lock(&root->fs_info->ro_block_group_mutex); | |
3204d33c JB |
1840 | if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, |
1841 | &cur_trans->flags)) | |
1bbc621e | 1842 | run_it = 1; |
1bbc621e CM |
1843 | mutex_unlock(&root->fs_info->ro_block_group_mutex); |
1844 | ||
1845 | if (run_it) | |
1846 | ret = btrfs_start_dirty_block_groups(trans, root); | |
1847 | } | |
1848 | if (ret) { | |
1849 | btrfs_end_transaction(trans, root); | |
1850 | return ret; | |
1851 | } | |
1852 | ||
4a9d8bde MX |
1853 | spin_lock(&root->fs_info->trans_lock); |
1854 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { | |
1855 | spin_unlock(&root->fs_info->trans_lock); | |
13c5a93e | 1856 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1857 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1858 | |
b9c8300c | 1859 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1860 | |
b4924a0f LB |
1861 | if (unlikely(cur_trans->aborted)) |
1862 | ret = cur_trans->aborted; | |
1863 | ||
724e2315 | 1864 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 1865 | |
49b25e05 | 1866 | return ret; |
79154b1b | 1867 | } |
4313b399 | 1868 | |
4a9d8bde | 1869 | cur_trans->state = TRANS_STATE_COMMIT_START; |
bb9c12c9 SW |
1870 | wake_up(&root->fs_info->transaction_blocked_wait); |
1871 | ||
ccd467d6 CM |
1872 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1873 | prev_trans = list_entry(cur_trans->list.prev, | |
1874 | struct btrfs_transaction, list); | |
4a9d8bde | 1875 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
13c5a93e | 1876 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1877 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1878 | |
1879 | wait_for_commit(root, prev_trans); | |
1f9b8c8f | 1880 | ret = prev_trans->aborted; |
ccd467d6 | 1881 | |
724e2315 | 1882 | btrfs_put_transaction(prev_trans); |
1f9b8c8f FM |
1883 | if (ret) |
1884 | goto cleanup_transaction; | |
a4abeea4 JB |
1885 | } else { |
1886 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1887 | } |
a4abeea4 JB |
1888 | } else { |
1889 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1890 | } |
15ee9bc7 | 1891 | |
0860adfd MX |
1892 | extwriter_counter_dec(cur_trans, trans->type); |
1893 | ||
82436617 MX |
1894 | ret = btrfs_start_delalloc_flush(root->fs_info); |
1895 | if (ret) | |
1896 | goto cleanup_transaction; | |
1897 | ||
8d875f95 | 1898 | ret = btrfs_run_delayed_items(trans, root); |
581227d0 MX |
1899 | if (ret) |
1900 | goto cleanup_transaction; | |
15ee9bc7 | 1901 | |
581227d0 MX |
1902 | wait_event(cur_trans->writer_wait, |
1903 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 1904 | |
581227d0 | 1905 | /* some pending stuffs might be added after the previous flush. */ |
8d875f95 | 1906 | ret = btrfs_run_delayed_items(trans, root); |
ca469637 MX |
1907 | if (ret) |
1908 | goto cleanup_transaction; | |
1909 | ||
82436617 | 1910 | btrfs_wait_delalloc_flush(root->fs_info); |
cb7ab021 | 1911 | |
161c3549 | 1912 | btrfs_wait_pending_ordered(cur_trans); |
50d9aa99 | 1913 | |
cb7ab021 | 1914 | btrfs_scrub_pause(root); |
ed0ca140 JB |
1915 | /* |
1916 | * Ok now we need to make sure to block out any other joins while we | |
1917 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 1918 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 JB |
1919 | */ |
1920 | spin_lock(&root->fs_info->trans_lock); | |
4a9d8bde | 1921 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
ed0ca140 JB |
1922 | spin_unlock(&root->fs_info->trans_lock); |
1923 | wait_event(cur_trans->writer_wait, | |
1924 | atomic_read(&cur_trans->num_writers) == 1); | |
1925 | ||
2cba30f1 MX |
1926 | /* ->aborted might be set after the previous check, so check it */ |
1927 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1928 | ret = cur_trans->aborted; | |
6cf7f77e | 1929 | goto scrub_continue; |
2cba30f1 | 1930 | } |
7585717f CM |
1931 | /* |
1932 | * the reloc mutex makes sure that we stop | |
1933 | * the balancing code from coming in and moving | |
1934 | * extents around in the middle of the commit | |
1935 | */ | |
1936 | mutex_lock(&root->fs_info->reloc_mutex); | |
1937 | ||
42874b3d MX |
1938 | /* |
1939 | * We needn't worry about the delayed items because we will | |
1940 | * deal with them in create_pending_snapshot(), which is the | |
1941 | * core function of the snapshot creation. | |
1942 | */ | |
1943 | ret = create_pending_snapshots(trans, root->fs_info); | |
49b25e05 JM |
1944 | if (ret) { |
1945 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1946 | goto scrub_continue; |
49b25e05 | 1947 | } |
3063d29f | 1948 | |
42874b3d MX |
1949 | /* |
1950 | * We insert the dir indexes of the snapshots and update the inode | |
1951 | * of the snapshots' parents after the snapshot creation, so there | |
1952 | * are some delayed items which are not dealt with. Now deal with | |
1953 | * them. | |
1954 | * | |
1955 | * We needn't worry that this operation will corrupt the snapshots, | |
1956 | * because all the tree which are snapshoted will be forced to COW | |
1957 | * the nodes and leaves. | |
1958 | */ | |
1959 | ret = btrfs_run_delayed_items(trans, root); | |
49b25e05 JM |
1960 | if (ret) { |
1961 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1962 | goto scrub_continue; |
49b25e05 | 1963 | } |
16cdcec7 | 1964 | |
56bec294 | 1965 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1966 | if (ret) { |
1967 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1968 | goto scrub_continue; |
49b25e05 | 1969 | } |
56bec294 | 1970 | |
0ed4792a QW |
1971 | /* Reocrd old roots for later qgroup accounting */ |
1972 | ret = btrfs_qgroup_prepare_account_extents(trans, root->fs_info); | |
1973 | if (ret) { | |
1974 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1975 | goto scrub_continue; | |
1976 | } | |
1977 | ||
e999376f CM |
1978 | /* |
1979 | * make sure none of the code above managed to slip in a | |
1980 | * delayed item | |
1981 | */ | |
1982 | btrfs_assert_delayed_root_empty(root); | |
1983 | ||
2c90e5d6 | 1984 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1985 | |
e02119d5 CM |
1986 | /* btrfs_commit_tree_roots is responsible for getting the |
1987 | * various roots consistent with each other. Every pointer | |
1988 | * in the tree of tree roots has to point to the most up to date | |
1989 | * root for every subvolume and other tree. So, we have to keep | |
1990 | * the tree logging code from jumping in and changing any | |
1991 | * of the trees. | |
1992 | * | |
1993 | * At this point in the commit, there can't be any tree-log | |
1994 | * writers, but a little lower down we drop the trans mutex | |
1995 | * and let new people in. By holding the tree_log_mutex | |
1996 | * from now until after the super is written, we avoid races | |
1997 | * with the tree-log code. | |
1998 | */ | |
1999 | mutex_lock(&root->fs_info->tree_log_mutex); | |
2000 | ||
5d4f98a2 | 2001 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
2002 | if (ret) { |
2003 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 2004 | mutex_unlock(&root->fs_info->reloc_mutex); |
6cf7f77e | 2005 | goto scrub_continue; |
49b25e05 | 2006 | } |
54aa1f4d | 2007 | |
3818aea2 | 2008 | /* |
7e1876ac DS |
2009 | * Since the transaction is done, we can apply the pending changes |
2010 | * before the next transaction. | |
3818aea2 | 2011 | */ |
572d9ab7 | 2012 | btrfs_apply_pending_changes(root->fs_info); |
3818aea2 | 2013 | |
5d4f98a2 | 2014 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2015 | * safe to free the root of tree log roots |
2016 | */ | |
2017 | btrfs_free_log_root_tree(trans, root->fs_info); | |
2018 | ||
0ed4792a QW |
2019 | /* |
2020 | * Since fs roots are all committed, we can get a quite accurate | |
2021 | * new_roots. So let's do quota accounting. | |
2022 | */ | |
2023 | ret = btrfs_qgroup_account_extents(trans, root->fs_info); | |
2024 | if (ret < 0) { | |
2025 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
2026 | mutex_unlock(&root->fs_info->reloc_mutex); | |
2027 | goto scrub_continue; | |
2028 | } | |
2029 | ||
5d4f98a2 | 2030 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
2031 | if (ret) { |
2032 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 2033 | mutex_unlock(&root->fs_info->reloc_mutex); |
6cf7f77e | 2034 | goto scrub_continue; |
49b25e05 | 2035 | } |
54aa1f4d | 2036 | |
2cba30f1 MX |
2037 | /* |
2038 | * The tasks which save the space cache and inode cache may also | |
2039 | * update ->aborted, check it. | |
2040 | */ | |
2041 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
2042 | ret = cur_trans->aborted; | |
2043 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
2044 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 2045 | goto scrub_continue; |
2cba30f1 MX |
2046 | } |
2047 | ||
11833d66 YZ |
2048 | btrfs_prepare_extent_commit(trans, root); |
2049 | ||
78fae27e | 2050 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
2051 | |
2052 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
2053 | root->fs_info->tree_root->node); | |
9e351cc8 JB |
2054 | list_add_tail(&root->fs_info->tree_root->dirty_list, |
2055 | &cur_trans->switch_commits); | |
5d4f98a2 YZ |
2056 | |
2057 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
2058 | root->fs_info->chunk_root->node); | |
9e351cc8 JB |
2059 | list_add_tail(&root->fs_info->chunk_root->dirty_list, |
2060 | &cur_trans->switch_commits); | |
2061 | ||
2062 | switch_commit_roots(cur_trans, root->fs_info); | |
5d4f98a2 | 2063 | |
edf39272 | 2064 | assert_qgroups_uptodate(trans); |
ce93ec54 | 2065 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2066 | ASSERT(list_empty(&cur_trans->io_bgs)); |
5d4f98a2 | 2067 | update_super_roots(root); |
e02119d5 | 2068 | |
60e7cd3a JB |
2069 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
2070 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
6c41761f DS |
2071 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
2072 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 2073 | |
935e5cc9 | 2074 | btrfs_update_commit_device_size(root->fs_info); |
ce7213c7 | 2075 | btrfs_update_commit_device_bytes_used(root, cur_trans); |
935e5cc9 | 2076 | |
656f30db FM |
2077 | clear_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags); |
2078 | clear_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags); | |
2079 | ||
4fbcdf66 FM |
2080 | btrfs_trans_release_chunk_metadata(trans); |
2081 | ||
a4abeea4 | 2082 | spin_lock(&root->fs_info->trans_lock); |
4a9d8bde | 2083 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
a4abeea4 | 2084 | root->fs_info->running_transaction = NULL; |
a4abeea4 | 2085 | spin_unlock(&root->fs_info->trans_lock); |
7585717f | 2086 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 2087 | |
f9295749 | 2088 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 2089 | |
79154b1b | 2090 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 | 2091 | if (ret) { |
a4553fef | 2092 | btrfs_std_error(root->fs_info, ret, |
08748810 | 2093 | "Error while writing out transaction"); |
49b25e05 | 2094 | mutex_unlock(&root->fs_info->tree_log_mutex); |
6cf7f77e | 2095 | goto scrub_continue; |
49b25e05 JM |
2096 | } |
2097 | ||
2098 | ret = write_ctree_super(trans, root, 0); | |
2099 | if (ret) { | |
2100 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
6cf7f77e | 2101 | goto scrub_continue; |
49b25e05 | 2102 | } |
4313b399 | 2103 | |
e02119d5 CM |
2104 | /* |
2105 | * the super is written, we can safely allow the tree-loggers | |
2106 | * to go about their business | |
2107 | */ | |
2108 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
2109 | ||
11833d66 | 2110 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 2111 | |
3204d33c | 2112 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) |
13212b54 ZL |
2113 | btrfs_clear_space_info_full(root->fs_info); |
2114 | ||
15ee9bc7 | 2115 | root->fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
2116 | /* |
2117 | * We needn't acquire the lock here because there is no other task | |
2118 | * which can change it. | |
2119 | */ | |
2120 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2121 | wake_up(&cur_trans->commit_wait); |
3de4586c | 2122 | |
a4abeea4 | 2123 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 2124 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
2125 | spin_unlock(&root->fs_info->trans_lock); |
2126 | ||
724e2315 JB |
2127 | btrfs_put_transaction(cur_trans); |
2128 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2129 | |
0860adfd | 2130 | if (trans->type & __TRANS_FREEZABLE) |
354aa0fb | 2131 | sb_end_intwrite(root->fs_info->sb); |
b2b5ef5c | 2132 | |
1abe9b8a | 2133 | trace_btrfs_transaction_commit(root); |
2134 | ||
a2de733c AJ |
2135 | btrfs_scrub_continue(root); |
2136 | ||
9ed74f2d JB |
2137 | if (current->journal_info == trans) |
2138 | current->journal_info = NULL; | |
2139 | ||
2c90e5d6 | 2140 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2141 | |
8a733013 ZL |
2142 | if (current != root->fs_info->transaction_kthread && |
2143 | current != root->fs_info->cleaner_kthread) | |
24bbcf04 YZ |
2144 | btrfs_run_delayed_iputs(root); |
2145 | ||
79154b1b | 2146 | return ret; |
49b25e05 | 2147 | |
6cf7f77e WS |
2148 | scrub_continue: |
2149 | btrfs_scrub_continue(root); | |
49b25e05 | 2150 | cleanup_transaction: |
0e721106 | 2151 | btrfs_trans_release_metadata(trans, root); |
4fbcdf66 | 2152 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2153 | trans->block_rsv = NULL; |
c2cf52eb | 2154 | btrfs_warn(root->fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2155 | if (current->journal_info == trans) |
2156 | current->journal_info = NULL; | |
7b8b92af | 2157 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
2158 | |
2159 | return ret; | |
79154b1b CM |
2160 | } |
2161 | ||
d352ac68 | 2162 | /* |
9d1a2a3a DS |
2163 | * return < 0 if error |
2164 | * 0 if there are no more dead_roots at the time of call | |
2165 | * 1 there are more to be processed, call me again | |
2166 | * | |
2167 | * The return value indicates there are certainly more snapshots to delete, but | |
2168 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2169 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2170 | * few seconds later. | |
d352ac68 | 2171 | */ |
9d1a2a3a | 2172 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 2173 | { |
9d1a2a3a | 2174 | int ret; |
5d4f98a2 YZ |
2175 | struct btrfs_fs_info *fs_info = root->fs_info; |
2176 | ||
a4abeea4 | 2177 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2178 | if (list_empty(&fs_info->dead_roots)) { |
2179 | spin_unlock(&fs_info->trans_lock); | |
2180 | return 0; | |
2181 | } | |
2182 | root = list_first_entry(&fs_info->dead_roots, | |
2183 | struct btrfs_root, root_list); | |
cfad392b | 2184 | list_del_init(&root->root_list); |
a4abeea4 | 2185 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2186 | |
efe120a0 | 2187 | pr_debug("BTRFS: cleaner removing %llu\n", root->objectid); |
76dda93c | 2188 | |
9d1a2a3a | 2189 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2190 | |
9d1a2a3a DS |
2191 | if (btrfs_header_backref_rev(root->node) < |
2192 | BTRFS_MIXED_BACKREF_REV) | |
2193 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
2194 | else | |
2195 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
32471dc2 | 2196 | |
6596a928 | 2197 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2198 | } |
572d9ab7 DS |
2199 | |
2200 | void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) | |
2201 | { | |
2202 | unsigned long prev; | |
2203 | unsigned long bit; | |
2204 | ||
6c9fe14f | 2205 | prev = xchg(&fs_info->pending_changes, 0); |
572d9ab7 DS |
2206 | if (!prev) |
2207 | return; | |
2208 | ||
7e1876ac DS |
2209 | bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE; |
2210 | if (prev & bit) | |
2211 | btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2212 | prev &= ~bit; | |
2213 | ||
2214 | bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE; | |
2215 | if (prev & bit) | |
2216 | btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2217 | prev &= ~bit; | |
2218 | ||
d51033d0 DS |
2219 | bit = 1 << BTRFS_PENDING_COMMIT; |
2220 | if (prev & bit) | |
2221 | btrfs_debug(fs_info, "pending commit done"); | |
2222 | prev &= ~bit; | |
2223 | ||
572d9ab7 DS |
2224 | if (prev) |
2225 | btrfs_warn(fs_info, | |
2226 | "unknown pending changes left 0x%lx, ignoring", prev); | |
2227 | } |