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 | 67 | if (transaction->delayed_refs.pending_csums) |
62e85577 | 68 | pr_err("pending csums is %llu\n", |
1262133b | 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 | } | |
7785a663 FM |
78 | /* |
79 | * If any block groups are found in ->deleted_bgs then it's | |
80 | * because the transaction was aborted and a commit did not | |
81 | * happen (things failed before writing the new superblock | |
82 | * and calling btrfs_finish_extent_commit()), so we can not | |
83 | * discard the physical locations of the block groups. | |
84 | */ | |
85 | while (!list_empty(&transaction->deleted_bgs)) { | |
86 | struct btrfs_block_group_cache *cache; | |
87 | ||
88 | cache = list_first_entry(&transaction->deleted_bgs, | |
89 | struct btrfs_block_group_cache, | |
90 | bg_list); | |
91 | list_del_init(&cache->bg_list); | |
92 | btrfs_put_block_group_trimming(cache); | |
93 | btrfs_put_block_group(cache); | |
94 | } | |
2c90e5d6 | 95 | kmem_cache_free(btrfs_transaction_cachep, transaction); |
78fae27e | 96 | } |
79154b1b CM |
97 | } |
98 | ||
663dfbb0 FM |
99 | static void clear_btree_io_tree(struct extent_io_tree *tree) |
100 | { | |
101 | spin_lock(&tree->lock); | |
b666a9cd DS |
102 | /* |
103 | * Do a single barrier for the waitqueue_active check here, the state | |
104 | * of the waitqueue should not change once clear_btree_io_tree is | |
105 | * called. | |
106 | */ | |
107 | smp_mb(); | |
663dfbb0 FM |
108 | while (!RB_EMPTY_ROOT(&tree->state)) { |
109 | struct rb_node *node; | |
110 | struct extent_state *state; | |
111 | ||
112 | node = rb_first(&tree->state); | |
113 | state = rb_entry(node, struct extent_state, rb_node); | |
114 | rb_erase(&state->rb_node, &tree->state); | |
115 | RB_CLEAR_NODE(&state->rb_node); | |
116 | /* | |
117 | * btree io trees aren't supposed to have tasks waiting for | |
118 | * changes in the flags of extent states ever. | |
119 | */ | |
120 | ASSERT(!waitqueue_active(&state->wq)); | |
121 | free_extent_state(state); | |
351810c1 DS |
122 | |
123 | cond_resched_lock(&tree->lock); | |
663dfbb0 FM |
124 | } |
125 | spin_unlock(&tree->lock); | |
126 | } | |
127 | ||
9e351cc8 JB |
128 | static noinline void switch_commit_roots(struct btrfs_transaction *trans, |
129 | struct btrfs_fs_info *fs_info) | |
817d52f8 | 130 | { |
9e351cc8 JB |
131 | struct btrfs_root *root, *tmp; |
132 | ||
133 | down_write(&fs_info->commit_root_sem); | |
134 | list_for_each_entry_safe(root, tmp, &trans->switch_commits, | |
135 | dirty_list) { | |
136 | list_del_init(&root->dirty_list); | |
137 | free_extent_buffer(root->commit_root); | |
138 | root->commit_root = btrfs_root_node(root); | |
139 | if (is_fstree(root->objectid)) | |
140 | btrfs_unpin_free_ino(root); | |
663dfbb0 | 141 | clear_btree_io_tree(&root->dirty_log_pages); |
9e351cc8 | 142 | } |
2b9dbef2 JB |
143 | |
144 | /* We can free old roots now. */ | |
145 | spin_lock(&trans->dropped_roots_lock); | |
146 | while (!list_empty(&trans->dropped_roots)) { | |
147 | root = list_first_entry(&trans->dropped_roots, | |
148 | struct btrfs_root, root_list); | |
149 | list_del_init(&root->root_list); | |
150 | spin_unlock(&trans->dropped_roots_lock); | |
151 | btrfs_drop_and_free_fs_root(fs_info, root); | |
152 | spin_lock(&trans->dropped_roots_lock); | |
153 | } | |
154 | spin_unlock(&trans->dropped_roots_lock); | |
9e351cc8 | 155 | up_write(&fs_info->commit_root_sem); |
817d52f8 JB |
156 | } |
157 | ||
0860adfd MX |
158 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
159 | unsigned int type) | |
160 | { | |
161 | if (type & TRANS_EXTWRITERS) | |
162 | atomic_inc(&trans->num_extwriters); | |
163 | } | |
164 | ||
165 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
166 | unsigned int type) | |
167 | { | |
168 | if (type & TRANS_EXTWRITERS) | |
169 | atomic_dec(&trans->num_extwriters); | |
170 | } | |
171 | ||
172 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
173 | unsigned int type) | |
174 | { | |
175 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
176 | } | |
177 | ||
178 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
179 | { | |
180 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
181 | } |
182 | ||
d352ac68 CM |
183 | /* |
184 | * either allocate a new transaction or hop into the existing one | |
185 | */ | |
0860adfd | 186 | static noinline int join_transaction(struct btrfs_root *root, unsigned int type) |
79154b1b CM |
187 | { |
188 | struct btrfs_transaction *cur_trans; | |
19ae4e81 | 189 | struct btrfs_fs_info *fs_info = root->fs_info; |
a4abeea4 | 190 | |
19ae4e81 | 191 | spin_lock(&fs_info->trans_lock); |
d43317dc | 192 | loop: |
49b25e05 | 193 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 194 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 195 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
196 | return -EROFS; |
197 | } | |
198 | ||
19ae4e81 | 199 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 200 | if (cur_trans) { |
871383be | 201 | if (cur_trans->aborted) { |
19ae4e81 | 202 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 203 | return cur_trans->aborted; |
871383be | 204 | } |
4a9d8bde | 205 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
206 | spin_unlock(&fs_info->trans_lock); |
207 | return -EBUSY; | |
208 | } | |
a4abeea4 | 209 | atomic_inc(&cur_trans->use_count); |
13c5a93e | 210 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 211 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 212 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 213 | return 0; |
79154b1b | 214 | } |
19ae4e81 | 215 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 216 | |
354aa0fb MX |
217 | /* |
218 | * If we are ATTACH, we just want to catch the current transaction, | |
219 | * and commit it. If there is no transaction, just return ENOENT. | |
220 | */ | |
221 | if (type == TRANS_ATTACH) | |
222 | return -ENOENT; | |
223 | ||
4a9d8bde MX |
224 | /* |
225 | * JOIN_NOLOCK only happens during the transaction commit, so | |
226 | * it is impossible that ->running_transaction is NULL | |
227 | */ | |
228 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
229 | ||
a4abeea4 JB |
230 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); |
231 | if (!cur_trans) | |
232 | return -ENOMEM; | |
d43317dc | 233 | |
19ae4e81 JS |
234 | spin_lock(&fs_info->trans_lock); |
235 | if (fs_info->running_transaction) { | |
d43317dc CM |
236 | /* |
237 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 238 | * to redo the checks above |
d43317dc | 239 | */ |
a4abeea4 | 240 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
d43317dc | 241 | goto loop; |
87533c47 | 242 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 243 | spin_unlock(&fs_info->trans_lock); |
7b8b92af JB |
244 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
245 | return -EROFS; | |
79154b1b | 246 | } |
d43317dc | 247 | |
a4abeea4 | 248 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 249 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
250 | init_waitqueue_head(&cur_trans->writer_wait); |
251 | init_waitqueue_head(&cur_trans->commit_wait); | |
161c3549 | 252 | init_waitqueue_head(&cur_trans->pending_wait); |
4a9d8bde | 253 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
254 | /* |
255 | * One for this trans handle, one so it will live on until we | |
256 | * commit the transaction. | |
257 | */ | |
258 | atomic_set(&cur_trans->use_count, 2); | |
161c3549 | 259 | atomic_set(&cur_trans->pending_ordered, 0); |
3204d33c | 260 | cur_trans->flags = 0; |
a4abeea4 JB |
261 | cur_trans->start_time = get_seconds(); |
262 | ||
a099d0fd AM |
263 | memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); |
264 | ||
c46effa6 | 265 | cur_trans->delayed_refs.href_root = RB_ROOT; |
3368d001 | 266 | cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; |
d7df2c79 | 267 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
20b297d6 JS |
268 | |
269 | /* | |
270 | * although the tree mod log is per file system and not per transaction, | |
271 | * the log must never go across transaction boundaries. | |
272 | */ | |
273 | smp_mb(); | |
31b1a2bd | 274 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
5d163e0e | 275 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); |
31b1a2bd | 276 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
5d163e0e | 277 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); |
fc36ed7e | 278 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 279 | |
a4abeea4 JB |
280 | spin_lock_init(&cur_trans->delayed_refs.lock); |
281 | ||
282 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
6df9a95e | 283 | INIT_LIST_HEAD(&cur_trans->pending_chunks); |
9e351cc8 | 284 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
ce93ec54 | 285 | INIT_LIST_HEAD(&cur_trans->dirty_bgs); |
1bbc621e | 286 | INIT_LIST_HEAD(&cur_trans->io_bgs); |
2b9dbef2 | 287 | INIT_LIST_HEAD(&cur_trans->dropped_roots); |
1bbc621e | 288 | mutex_init(&cur_trans->cache_write_mutex); |
cb723e49 | 289 | cur_trans->num_dirty_bgs = 0; |
ce93ec54 | 290 | spin_lock_init(&cur_trans->dirty_bgs_lock); |
e33e17ee | 291 | INIT_LIST_HEAD(&cur_trans->deleted_bgs); |
2b9dbef2 | 292 | spin_lock_init(&cur_trans->dropped_roots_lock); |
19ae4e81 | 293 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 294 | extent_io_tree_init(&cur_trans->dirty_pages, |
19ae4e81 JS |
295 | fs_info->btree_inode->i_mapping); |
296 | fs_info->generation++; | |
297 | cur_trans->transid = fs_info->generation; | |
298 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 299 | cur_trans->aborted = 0; |
19ae4e81 | 300 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 301 | |
79154b1b CM |
302 | return 0; |
303 | } | |
304 | ||
d352ac68 | 305 | /* |
d397712b CM |
306 | * this does all the record keeping required to make sure that a reference |
307 | * counted root is properly recorded in a given transaction. This is required | |
308 | * to make sure the old root from before we joined the transaction is deleted | |
309 | * when the transaction commits | |
d352ac68 | 310 | */ |
7585717f | 311 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
6426c7ad QW |
312 | struct btrfs_root *root, |
313 | int force) | |
6702ed49 | 314 | { |
6426c7ad QW |
315 | if ((test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
316 | root->last_trans < trans->transid) || force) { | |
6702ed49 | 317 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
318 | WARN_ON(root->commit_root != root->node); |
319 | ||
7585717f | 320 | /* |
27cdeb70 | 321 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
322 | * we have the reloc mutex held now, so there |
323 | * is only one writer in this function | |
324 | */ | |
27cdeb70 | 325 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 326 | |
27cdeb70 | 327 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
328 | * they find our root->last_trans update |
329 | */ | |
330 | smp_wmb(); | |
331 | ||
a4abeea4 | 332 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
6426c7ad | 333 | if (root->last_trans == trans->transid && !force) { |
a4abeea4 JB |
334 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
335 | return 0; | |
336 | } | |
5d4f98a2 YZ |
337 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
338 | (unsigned long)root->root_key.objectid, | |
339 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 340 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
341 | root->last_trans = trans->transid; |
342 | ||
343 | /* this is pretty tricky. We don't want to | |
344 | * take the relocation lock in btrfs_record_root_in_trans | |
345 | * unless we're really doing the first setup for this root in | |
346 | * this transaction. | |
347 | * | |
348 | * Normally we'd use root->last_trans as a flag to decide | |
349 | * if we want to take the expensive mutex. | |
350 | * | |
351 | * But, we have to set root->last_trans before we | |
352 | * init the relocation root, otherwise, we trip over warnings | |
353 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 354 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
355 | * fixing up the reloc trees and everyone must wait. |
356 | * | |
357 | * When this is zero, they can trust root->last_trans and fly | |
358 | * through btrfs_record_root_in_trans without having to take the | |
359 | * lock. smp_wmb() makes sure that all the writes above are | |
360 | * done before we pop in the zero below | |
361 | */ | |
5d4f98a2 | 362 | btrfs_init_reloc_root(trans, root); |
c7548af6 | 363 | smp_mb__before_atomic(); |
27cdeb70 | 364 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 YZ |
365 | } |
366 | return 0; | |
367 | } | |
bcc63abb | 368 | |
7585717f | 369 | |
2b9dbef2 JB |
370 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
371 | struct btrfs_root *root) | |
372 | { | |
373 | struct btrfs_transaction *cur_trans = trans->transaction; | |
374 | ||
375 | /* Add ourselves to the transaction dropped list */ | |
376 | spin_lock(&cur_trans->dropped_roots_lock); | |
377 | list_add_tail(&root->root_list, &cur_trans->dropped_roots); | |
378 | spin_unlock(&cur_trans->dropped_roots_lock); | |
379 | ||
380 | /* Make sure we don't try to update the root at commit time */ | |
381 | spin_lock(&root->fs_info->fs_roots_radix_lock); | |
382 | radix_tree_tag_clear(&root->fs_info->fs_roots_radix, | |
383 | (unsigned long)root->root_key.objectid, | |
384 | BTRFS_ROOT_TRANS_TAG); | |
385 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
386 | } | |
387 | ||
7585717f CM |
388 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
389 | struct btrfs_root *root) | |
390 | { | |
27cdeb70 | 391 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
7585717f CM |
392 | return 0; |
393 | ||
394 | /* | |
27cdeb70 | 395 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
396 | * and barriers |
397 | */ | |
398 | smp_rmb(); | |
399 | if (root->last_trans == trans->transid && | |
27cdeb70 | 400 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
401 | return 0; |
402 | ||
403 | mutex_lock(&root->fs_info->reloc_mutex); | |
6426c7ad | 404 | record_root_in_trans(trans, root, 0); |
7585717f CM |
405 | mutex_unlock(&root->fs_info->reloc_mutex); |
406 | ||
407 | return 0; | |
408 | } | |
409 | ||
4a9d8bde MX |
410 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
411 | { | |
412 | return (trans->state >= TRANS_STATE_BLOCKED && | |
501407aa JB |
413 | trans->state < TRANS_STATE_UNBLOCKED && |
414 | !trans->aborted); | |
4a9d8bde MX |
415 | } |
416 | ||
d352ac68 CM |
417 | /* wait for commit against the current transaction to become unblocked |
418 | * when this is done, it is safe to start a new transaction, but the current | |
419 | * transaction might not be fully on disk. | |
420 | */ | |
37d1aeee | 421 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 422 | { |
f9295749 | 423 | struct btrfs_transaction *cur_trans; |
79154b1b | 424 | |
a4abeea4 | 425 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 426 | cur_trans = root->fs_info->running_transaction; |
4a9d8bde | 427 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
13c5a93e | 428 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 429 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
430 | |
431 | wait_event(root->fs_info->transaction_wait, | |
501407aa JB |
432 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
433 | cur_trans->aborted); | |
724e2315 | 434 | btrfs_put_transaction(cur_trans); |
a4abeea4 JB |
435 | } else { |
436 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 437 | } |
37d1aeee CM |
438 | } |
439 | ||
a22285a6 YZ |
440 | static int may_wait_transaction(struct btrfs_root *root, int type) |
441 | { | |
afcdd129 | 442 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) |
a4abeea4 JB |
443 | return 0; |
444 | ||
445 | if (type == TRANS_USERSPACE) | |
446 | return 1; | |
447 | ||
448 | if (type == TRANS_START && | |
449 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 450 | return 1; |
a4abeea4 | 451 | |
a22285a6 YZ |
452 | return 0; |
453 | } | |
454 | ||
20dd2cbf MX |
455 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
456 | { | |
457 | if (!root->fs_info->reloc_ctl || | |
27cdeb70 | 458 | !test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
20dd2cbf MX |
459 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
460 | root->reloc_root) | |
461 | return false; | |
462 | ||
463 | return true; | |
464 | } | |
465 | ||
08e007d2 | 466 | static struct btrfs_trans_handle * |
5aed1dd8 AM |
467 | start_transaction(struct btrfs_root *root, unsigned int num_items, |
468 | unsigned int type, enum btrfs_reserve_flush_enum flush) | |
37d1aeee | 469 | { |
a22285a6 YZ |
470 | struct btrfs_trans_handle *h; |
471 | struct btrfs_transaction *cur_trans; | |
b5009945 | 472 | u64 num_bytes = 0; |
c5567237 | 473 | u64 qgroup_reserved = 0; |
20dd2cbf MX |
474 | bool reloc_reserved = false; |
475 | int ret; | |
acce952b | 476 | |
46c4e71e | 477 | /* Send isn't supposed to start transactions. */ |
2755a0de | 478 | ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); |
46c4e71e | 479 | |
87533c47 | 480 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
acce952b | 481 | return ERR_PTR(-EROFS); |
2a1eb461 | 482 | |
46c4e71e | 483 | if (current->journal_info) { |
0860adfd | 484 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 JB |
485 | h = current->journal_info; |
486 | h->use_count++; | |
b7d5b0a8 | 487 | WARN_ON(h->use_count > 2); |
2a1eb461 JB |
488 | h->orig_rsv = h->block_rsv; |
489 | h->block_rsv = NULL; | |
490 | goto got_it; | |
491 | } | |
b5009945 JB |
492 | |
493 | /* | |
494 | * Do the reservation before we join the transaction so we can do all | |
495 | * the appropriate flushing if need be. | |
496 | */ | |
497 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
7174109c QW |
498 | qgroup_reserved = num_items * root->nodesize; |
499 | ret = btrfs_qgroup_reserve_meta(root, qgroup_reserved); | |
500 | if (ret) | |
501 | return ERR_PTR(ret); | |
c5567237 | 502 | |
b5009945 | 503 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); |
20dd2cbf MX |
504 | /* |
505 | * Do the reservation for the relocation root creation | |
506 | */ | |
ee39b432 | 507 | if (need_reserve_reloc_root(root)) { |
20dd2cbf MX |
508 | num_bytes += root->nodesize; |
509 | reloc_reserved = true; | |
510 | } | |
511 | ||
08e007d2 MX |
512 | ret = btrfs_block_rsv_add(root, |
513 | &root->fs_info->trans_block_rsv, | |
514 | num_bytes, flush); | |
b5009945 | 515 | if (ret) |
843fcf35 | 516 | goto reserve_fail; |
b5009945 | 517 | } |
a22285a6 | 518 | again: |
f2f767e7 | 519 | h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); |
843fcf35 MX |
520 | if (!h) { |
521 | ret = -ENOMEM; | |
522 | goto alloc_fail; | |
523 | } | |
37d1aeee | 524 | |
98114659 JB |
525 | /* |
526 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
527 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
528 | * because we're already holding a ref. We need this because we could | |
529 | * have raced in and did an fsync() on a file which can kick a commit | |
530 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
531 | * |
532 | * If we are ATTACH, it means we just want to catch the current | |
533 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 534 | */ |
0860adfd | 535 | if (type & __TRANS_FREEZABLE) |
60376ce4 | 536 | sb_start_intwrite(root->fs_info->sb); |
b2b5ef5c | 537 | |
a22285a6 | 538 | if (may_wait_transaction(root, type)) |
37d1aeee | 539 | wait_current_trans(root); |
a22285a6 | 540 | |
a4abeea4 | 541 | do { |
354aa0fb | 542 | ret = join_transaction(root, type); |
178260b2 | 543 | if (ret == -EBUSY) { |
a4abeea4 | 544 | wait_current_trans(root); |
178260b2 MX |
545 | if (unlikely(type == TRANS_ATTACH)) |
546 | ret = -ENOENT; | |
547 | } | |
a4abeea4 JB |
548 | } while (ret == -EBUSY); |
549 | ||
a43f7f82 | 550 | if (ret < 0) |
843fcf35 | 551 | goto join_fail; |
0f7d52f4 | 552 | |
a22285a6 | 553 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
554 | |
555 | h->transid = cur_trans->transid; | |
556 | h->transaction = cur_trans; | |
d13603ef | 557 | h->root = root; |
2a1eb461 | 558 | h->use_count = 1; |
64b63580 | 559 | h->fs_info = root->fs_info; |
7174109c | 560 | |
a698d075 | 561 | h->type = type; |
d9a0540a | 562 | h->can_flush_pending_bgs = true; |
bed92eae | 563 | INIT_LIST_HEAD(&h->qgroup_ref_list); |
ea658bad | 564 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 565 | |
a22285a6 | 566 | smp_mb(); |
4a9d8bde MX |
567 | if (cur_trans->state >= TRANS_STATE_BLOCKED && |
568 | may_wait_transaction(root, type)) { | |
abdd2e80 | 569 | current->journal_info = h; |
a22285a6 YZ |
570 | btrfs_commit_transaction(h, root); |
571 | goto again; | |
572 | } | |
573 | ||
b5009945 | 574 | if (num_bytes) { |
8c2a3ca2 | 575 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 576 | h->transid, num_bytes, 1); |
b5009945 JB |
577 | h->block_rsv = &root->fs_info->trans_block_rsv; |
578 | h->bytes_reserved = num_bytes; | |
20dd2cbf | 579 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 580 | } |
9ed74f2d | 581 | |
2a1eb461 | 582 | got_it: |
a4abeea4 | 583 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
584 | |
585 | if (!current->journal_info && type != TRANS_USERSPACE) | |
586 | current->journal_info = h; | |
79154b1b | 587 | return h; |
843fcf35 MX |
588 | |
589 | join_fail: | |
0860adfd | 590 | if (type & __TRANS_FREEZABLE) |
843fcf35 MX |
591 | sb_end_intwrite(root->fs_info->sb); |
592 | kmem_cache_free(btrfs_trans_handle_cachep, h); | |
593 | alloc_fail: | |
594 | if (num_bytes) | |
595 | btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv, | |
596 | num_bytes); | |
597 | reserve_fail: | |
7174109c | 598 | btrfs_qgroup_free_meta(root, qgroup_reserved); |
843fcf35 | 599 | return ERR_PTR(ret); |
79154b1b CM |
600 | } |
601 | ||
f9295749 | 602 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
5aed1dd8 | 603 | unsigned int num_items) |
f9295749 | 604 | { |
08e007d2 MX |
605 | return start_transaction(root, num_items, TRANS_START, |
606 | BTRFS_RESERVE_FLUSH_ALL); | |
f9295749 | 607 | } |
8eab77ff FM |
608 | struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( |
609 | struct btrfs_root *root, | |
610 | unsigned int num_items, | |
611 | int min_factor) | |
612 | { | |
613 | struct btrfs_trans_handle *trans; | |
614 | u64 num_bytes; | |
615 | int ret; | |
616 | ||
617 | trans = btrfs_start_transaction(root, num_items); | |
618 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) | |
619 | return trans; | |
620 | ||
621 | trans = btrfs_start_transaction(root, 0); | |
622 | if (IS_ERR(trans)) | |
623 | return trans; | |
624 | ||
625 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); | |
626 | ret = btrfs_cond_migrate_bytes(root->fs_info, | |
627 | &root->fs_info->trans_block_rsv, | |
628 | num_bytes, | |
629 | min_factor); | |
630 | if (ret) { | |
631 | btrfs_end_transaction(trans, root); | |
632 | return ERR_PTR(ret); | |
633 | } | |
634 | ||
635 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
636 | trans->bytes_reserved = num_bytes; | |
88d3a5aa JB |
637 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
638 | trans->transid, num_bytes, 1); | |
8eab77ff FM |
639 | |
640 | return trans; | |
641 | } | |
8407aa46 | 642 | |
08e007d2 | 643 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
5aed1dd8 AM |
644 | struct btrfs_root *root, |
645 | unsigned int num_items) | |
8407aa46 | 646 | { |
08e007d2 MX |
647 | return start_transaction(root, num_items, TRANS_START, |
648 | BTRFS_RESERVE_FLUSH_LIMIT); | |
8407aa46 MX |
649 | } |
650 | ||
7a7eaa40 | 651 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 652 | { |
575a75d6 AM |
653 | return start_transaction(root, 0, TRANS_JOIN, |
654 | BTRFS_RESERVE_NO_FLUSH); | |
f9295749 CM |
655 | } |
656 | ||
7a7eaa40 | 657 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 658 | { |
575a75d6 AM |
659 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, |
660 | BTRFS_RESERVE_NO_FLUSH); | |
0af3d00b JB |
661 | } |
662 | ||
7a7eaa40 | 663 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 664 | { |
575a75d6 AM |
665 | return start_transaction(root, 0, TRANS_USERSPACE, |
666 | BTRFS_RESERVE_NO_FLUSH); | |
9ca9ee09 SW |
667 | } |
668 | ||
d4edf39b MX |
669 | /* |
670 | * btrfs_attach_transaction() - catch the running transaction | |
671 | * | |
672 | * It is used when we want to commit the current the transaction, but | |
673 | * don't want to start a new one. | |
674 | * | |
675 | * Note: If this function return -ENOENT, it just means there is no | |
676 | * running transaction. But it is possible that the inactive transaction | |
677 | * is still in the memory, not fully on disk. If you hope there is no | |
678 | * inactive transaction in the fs when -ENOENT is returned, you should | |
679 | * invoke | |
680 | * btrfs_attach_transaction_barrier() | |
681 | */ | |
354aa0fb | 682 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 683 | { |
575a75d6 AM |
684 | return start_transaction(root, 0, TRANS_ATTACH, |
685 | BTRFS_RESERVE_NO_FLUSH); | |
60376ce4 JB |
686 | } |
687 | ||
d4edf39b | 688 | /* |
90b6d283 | 689 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b MX |
690 | * |
691 | * It is similar to the above function, the differentia is this one | |
692 | * will wait for all the inactive transactions until they fully | |
693 | * complete. | |
694 | */ | |
695 | struct btrfs_trans_handle * | |
696 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
697 | { | |
698 | struct btrfs_trans_handle *trans; | |
699 | ||
575a75d6 AM |
700 | trans = start_transaction(root, 0, TRANS_ATTACH, |
701 | BTRFS_RESERVE_NO_FLUSH); | |
d4edf39b MX |
702 | if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT) |
703 | btrfs_wait_for_commit(root, 0); | |
704 | ||
705 | return trans; | |
706 | } | |
707 | ||
d352ac68 | 708 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 709 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
710 | struct btrfs_transaction *commit) |
711 | { | |
4a9d8bde | 712 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
713 | } |
714 | ||
46204592 SW |
715 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
716 | { | |
717 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 718 | int ret = 0; |
46204592 | 719 | |
46204592 SW |
720 | if (transid) { |
721 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 722 | goto out; |
46204592 SW |
723 | |
724 | /* find specified transaction */ | |
a4abeea4 | 725 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
726 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
727 | if (t->transid == transid) { | |
728 | cur_trans = t; | |
a4abeea4 | 729 | atomic_inc(&cur_trans->use_count); |
8cd2807f | 730 | ret = 0; |
46204592 SW |
731 | break; |
732 | } | |
8cd2807f MX |
733 | if (t->transid > transid) { |
734 | ret = 0; | |
46204592 | 735 | break; |
8cd2807f | 736 | } |
46204592 | 737 | } |
a4abeea4 | 738 | spin_unlock(&root->fs_info->trans_lock); |
42383020 SW |
739 | |
740 | /* | |
741 | * The specified transaction doesn't exist, or we | |
742 | * raced with btrfs_commit_transaction | |
743 | */ | |
744 | if (!cur_trans) { | |
745 | if (transid > root->fs_info->last_trans_committed) | |
746 | ret = -EINVAL; | |
8cd2807f | 747 | goto out; |
42383020 | 748 | } |
46204592 SW |
749 | } else { |
750 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 751 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
752 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
753 | list) { | |
4a9d8bde MX |
754 | if (t->state >= TRANS_STATE_COMMIT_START) { |
755 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 756 | break; |
46204592 | 757 | cur_trans = t; |
a4abeea4 | 758 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
759 | break; |
760 | } | |
761 | } | |
a4abeea4 | 762 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 763 | if (!cur_trans) |
a4abeea4 | 764 | goto out; /* nothing committing|committed */ |
46204592 SW |
765 | } |
766 | ||
46204592 | 767 | wait_for_commit(root, cur_trans); |
724e2315 | 768 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 769 | out: |
46204592 SW |
770 | return ret; |
771 | } | |
772 | ||
37d1aeee CM |
773 | void btrfs_throttle(struct btrfs_root *root) |
774 | { | |
a4abeea4 | 775 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 776 | wait_current_trans(root); |
37d1aeee CM |
777 | } |
778 | ||
8929ecfa YZ |
779 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
780 | struct btrfs_root *root) | |
781 | { | |
1be41b78 | 782 | if (root->fs_info->global_block_rsv.space_info->full && |
0a2b2a84 | 783 | btrfs_check_space_for_delayed_refs(trans, root)) |
1be41b78 | 784 | return 1; |
36ba022a | 785 | |
1be41b78 | 786 | return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); |
8929ecfa YZ |
787 | } |
788 | ||
789 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
790 | struct btrfs_root *root) | |
791 | { | |
792 | struct btrfs_transaction *cur_trans = trans->transaction; | |
793 | int updates; | |
49b25e05 | 794 | int err; |
8929ecfa | 795 | |
a4abeea4 | 796 | smp_mb(); |
4a9d8bde MX |
797 | if (cur_trans->state >= TRANS_STATE_BLOCKED || |
798 | cur_trans->delayed_refs.flushing) | |
8929ecfa YZ |
799 | return 1; |
800 | ||
801 | updates = trans->delayed_ref_updates; | |
802 | trans->delayed_ref_updates = 0; | |
49b25e05 | 803 | if (updates) { |
28ed1345 | 804 | err = btrfs_run_delayed_refs(trans, root, updates * 2); |
49b25e05 JM |
805 | if (err) /* Error code will also eval true */ |
806 | return err; | |
807 | } | |
8929ecfa YZ |
808 | |
809 | return should_end_transaction(trans, root); | |
810 | } | |
811 | ||
89ce8a63 | 812 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
a698d075 | 813 | struct btrfs_root *root, int throttle) |
79154b1b | 814 | { |
8929ecfa | 815 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 816 | struct btrfs_fs_info *info = root->fs_info; |
31b9655f | 817 | u64 transid = trans->transid; |
1be41b78 | 818 | unsigned long cur = trans->delayed_ref_updates; |
a698d075 | 819 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 820 | int err = 0; |
a79b7d4b | 821 | int must_run_delayed_refs = 0; |
c3e69d58 | 822 | |
3bbb24b2 JB |
823 | if (trans->use_count > 1) { |
824 | trans->use_count--; | |
2a1eb461 JB |
825 | trans->block_rsv = trans->orig_rsv; |
826 | return 0; | |
827 | } | |
828 | ||
b24e03db | 829 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 830 | trans->block_rsv = NULL; |
c5567237 | 831 | |
ea658bad JB |
832 | if (!list_empty(&trans->new_bgs)) |
833 | btrfs_create_pending_block_groups(trans, root); | |
834 | ||
1be41b78 | 835 | trans->delayed_ref_updates = 0; |
a79b7d4b CM |
836 | if (!trans->sync) { |
837 | must_run_delayed_refs = | |
838 | btrfs_should_throttle_delayed_refs(trans, root); | |
0a2b2a84 | 839 | cur = max_t(unsigned long, cur, 32); |
a79b7d4b CM |
840 | |
841 | /* | |
842 | * don't make the caller wait if they are from a NOLOCK | |
843 | * or ATTACH transaction, it will deadlock with commit | |
844 | */ | |
845 | if (must_run_delayed_refs == 1 && | |
846 | (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH))) | |
847 | must_run_delayed_refs = 2; | |
56bec294 | 848 | } |
bb721703 | 849 | |
0e721106 JB |
850 | btrfs_trans_release_metadata(trans, root); |
851 | trans->block_rsv = NULL; | |
56bec294 | 852 | |
ea658bad JB |
853 | if (!list_empty(&trans->new_bgs)) |
854 | btrfs_create_pending_block_groups(trans, root); | |
855 | ||
4fbcdf66 FM |
856 | btrfs_trans_release_chunk_metadata(trans); |
857 | ||
a4abeea4 | 858 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
4a9d8bde MX |
859 | should_end_transaction(trans, root) && |
860 | ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) { | |
861 | spin_lock(&info->trans_lock); | |
862 | if (cur_trans->state == TRANS_STATE_RUNNING) | |
863 | cur_trans->state = TRANS_STATE_BLOCKED; | |
864 | spin_unlock(&info->trans_lock); | |
a4abeea4 | 865 | } |
8929ecfa | 866 | |
4a9d8bde | 867 | if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
3bbb24b2 | 868 | if (throttle) |
8929ecfa | 869 | return btrfs_commit_transaction(trans, root); |
3bbb24b2 | 870 | else |
8929ecfa YZ |
871 | wake_up_process(info->transaction_kthread); |
872 | } | |
873 | ||
0860adfd | 874 | if (trans->type & __TRANS_FREEZABLE) |
98114659 | 875 | sb_end_intwrite(root->fs_info->sb); |
6df7881a | 876 | |
8929ecfa | 877 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
878 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
879 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 880 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 881 | |
a83342aa DS |
882 | /* |
883 | * Make sure counter is updated before we wake up waiters. | |
884 | */ | |
99d16cbc | 885 | smp_mb(); |
79154b1b CM |
886 | if (waitqueue_active(&cur_trans->writer_wait)) |
887 | wake_up(&cur_trans->writer_wait); | |
724e2315 | 888 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
889 | |
890 | if (current->journal_info == trans) | |
891 | current->journal_info = NULL; | |
ab78c84d | 892 | |
24bbcf04 YZ |
893 | if (throttle) |
894 | btrfs_run_delayed_iputs(root); | |
895 | ||
49b25e05 | 896 | if (trans->aborted || |
4e121c06 JB |
897 | test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) { |
898 | wake_up_process(info->transaction_kthread); | |
4edc2ca3 | 899 | err = -EIO; |
4e121c06 | 900 | } |
edf39272 | 901 | assert_qgroups_uptodate(trans); |
49b25e05 | 902 | |
4edc2ca3 | 903 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
a79b7d4b | 904 | if (must_run_delayed_refs) { |
31b9655f | 905 | btrfs_async_run_delayed_refs(root, cur, transid, |
a79b7d4b CM |
906 | must_run_delayed_refs == 1); |
907 | } | |
4edc2ca3 | 908 | return err; |
79154b1b CM |
909 | } |
910 | ||
89ce8a63 CM |
911 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
912 | struct btrfs_root *root) | |
913 | { | |
98ad43be | 914 | return __btrfs_end_transaction(trans, root, 0); |
89ce8a63 CM |
915 | } |
916 | ||
917 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
918 | struct btrfs_root *root) | |
919 | { | |
98ad43be | 920 | return __btrfs_end_transaction(trans, root, 1); |
16cdcec7 MX |
921 | } |
922 | ||
d352ac68 CM |
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 | |
690587d1 | 926 | * those extents are sent to disk but does not wait on them |
d352ac68 | 927 | */ |
690587d1 | 928 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 929 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 930 | { |
777e6bd7 | 931 | int err = 0; |
7c4452b9 | 932 | int werr = 0; |
1728366e | 933 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 934 | struct extent_state *cached_state = NULL; |
777e6bd7 | 935 | u64 start = 0; |
5f39d397 | 936 | u64 end; |
7c4452b9 | 937 | |
1728366e | 938 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 939 | mark, &cached_state)) { |
663dfbb0 FM |
940 | bool wait_writeback = false; |
941 | ||
942 | err = convert_extent_bit(dirty_pages, start, end, | |
943 | EXTENT_NEED_WAIT, | |
210aa277 | 944 | mark, &cached_state); |
663dfbb0 FM |
945 | /* |
946 | * convert_extent_bit can return -ENOMEM, which is most of the | |
947 | * time a temporary error. So when it happens, ignore the error | |
948 | * and wait for writeback of this range to finish - because we | |
949 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
950 | * to btrfs_wait_marked_extents() would not know that writeback | |
951 | * for this range started and therefore wouldn't wait for it to | |
952 | * finish - we don't want to commit a superblock that points to | |
953 | * btree nodes/leafs for which writeback hasn't finished yet | |
954 | * (and without errors). | |
955 | * We cleanup any entries left in the io tree when committing | |
956 | * the transaction (through clear_btree_io_tree()). | |
957 | */ | |
958 | if (err == -ENOMEM) { | |
959 | err = 0; | |
960 | wait_writeback = true; | |
961 | } | |
962 | if (!err) | |
963 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
964 | if (err) |
965 | werr = err; | |
663dfbb0 FM |
966 | else if (wait_writeback) |
967 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 968 | free_extent_state(cached_state); |
663dfbb0 | 969 | cached_state = NULL; |
1728366e JB |
970 | cond_resched(); |
971 | start = end + 1; | |
7c4452b9 | 972 | } |
690587d1 CM |
973 | return werr; |
974 | } | |
975 | ||
976 | /* | |
977 | * when btree blocks are allocated, they have some corresponding bits set for | |
978 | * them in one of two extent_io trees. This is used to make sure all of | |
979 | * those extents are on disk for transaction or log commit. We wait | |
980 | * on all the pages and clear them from the dirty pages state tree | |
981 | */ | |
982 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 983 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 984 | { |
690587d1 CM |
985 | int err = 0; |
986 | int werr = 0; | |
1728366e | 987 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 988 | struct extent_state *cached_state = NULL; |
690587d1 CM |
989 | u64 start = 0; |
990 | u64 end; | |
656f30db | 991 | bool errors = false; |
777e6bd7 | 992 | |
1728366e | 993 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 994 | EXTENT_NEED_WAIT, &cached_state)) { |
663dfbb0 FM |
995 | /* |
996 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
997 | * When committing the transaction, we'll remove any entries | |
998 | * left in the io tree. For a log commit, we don't remove them | |
999 | * after committing the log because the tree can be accessed | |
1000 | * concurrently - we do it only at transaction commit time when | |
1001 | * it's safe to do it (through clear_btree_io_tree()). | |
1002 | */ | |
1003 | err = clear_extent_bit(dirty_pages, start, end, | |
1004 | EXTENT_NEED_WAIT, | |
1005 | 0, 0, &cached_state, GFP_NOFS); | |
1006 | if (err == -ENOMEM) | |
1007 | err = 0; | |
1008 | if (!err) | |
1009 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
1010 | if (err) |
1011 | werr = err; | |
e38e2ed7 FM |
1012 | free_extent_state(cached_state); |
1013 | cached_state = NULL; | |
1728366e JB |
1014 | cond_resched(); |
1015 | start = end + 1; | |
777e6bd7 | 1016 | } |
7c4452b9 CM |
1017 | if (err) |
1018 | werr = err; | |
656f30db FM |
1019 | |
1020 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { | |
1021 | if ((mark & EXTENT_DIRTY) && | |
afcdd129 JB |
1022 | test_and_clear_bit(BTRFS_FS_LOG1_ERR, |
1023 | &root->fs_info->flags)) | |
656f30db FM |
1024 | errors = true; |
1025 | ||
1026 | if ((mark & EXTENT_NEW) && | |
afcdd129 JB |
1027 | test_and_clear_bit(BTRFS_FS_LOG2_ERR, |
1028 | &root->fs_info->flags)) | |
656f30db FM |
1029 | errors = true; |
1030 | } else { | |
afcdd129 JB |
1031 | if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, |
1032 | &root->fs_info->flags)) | |
656f30db FM |
1033 | errors = true; |
1034 | } | |
1035 | ||
1036 | if (errors && !werr) | |
1037 | werr = -EIO; | |
1038 | ||
7c4452b9 | 1039 | return werr; |
79154b1b CM |
1040 | } |
1041 | ||
690587d1 CM |
1042 | /* |
1043 | * when btree blocks are allocated, they have some corresponding bits set for | |
1044 | * them in one of two extent_io trees. This is used to make sure all of | |
1045 | * those extents are on disk for transaction or log commit | |
1046 | */ | |
171170c1 | 1047 | static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, |
8cef4e16 | 1048 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
1049 | { |
1050 | int ret; | |
1051 | int ret2; | |
c6adc9cc | 1052 | struct blk_plug plug; |
690587d1 | 1053 | |
c6adc9cc | 1054 | blk_start_plug(&plug); |
8cef4e16 | 1055 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
c6adc9cc | 1056 | blk_finish_plug(&plug); |
8cef4e16 | 1057 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); |
bf0da8c1 CM |
1058 | |
1059 | if (ret) | |
1060 | return ret; | |
1061 | if (ret2) | |
1062 | return ret2; | |
1063 | return 0; | |
690587d1 CM |
1064 | } |
1065 | ||
663dfbb0 | 1066 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
d0c803c4 CM |
1067 | struct btrfs_root *root) |
1068 | { | |
663dfbb0 FM |
1069 | int ret; |
1070 | ||
1071 | ret = btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
1072 | &trans->transaction->dirty_pages, |
1073 | EXTENT_DIRTY); | |
663dfbb0 FM |
1074 | clear_btree_io_tree(&trans->transaction->dirty_pages); |
1075 | ||
1076 | return ret; | |
d0c803c4 CM |
1077 | } |
1078 | ||
d352ac68 CM |
1079 | /* |
1080 | * this is used to update the root pointer in the tree of tree roots. | |
1081 | * | |
1082 | * But, in the case of the extent allocation tree, updating the root | |
1083 | * pointer may allocate blocks which may change the root of the extent | |
1084 | * allocation tree. | |
1085 | * | |
1086 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1087 | * change while the root pointer was being updated in the metadata. | |
1088 | */ | |
0b86a832 CM |
1089 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1090 | struct btrfs_root *root) | |
79154b1b CM |
1091 | { |
1092 | int ret; | |
0b86a832 | 1093 | u64 old_root_bytenr; |
86b9f2ec | 1094 | u64 old_root_used; |
0b86a832 | 1095 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 1096 | |
86b9f2ec | 1097 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1098 | |
d397712b | 1099 | while (1) { |
0b86a832 | 1100 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1101 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1102 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1103 | break; |
87ef2bb4 | 1104 | |
5d4f98a2 | 1105 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1106 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1107 | &root->root_key, |
1108 | &root->root_item); | |
49b25e05 JM |
1109 | if (ret) |
1110 | return ret; | |
56bec294 | 1111 | |
86b9f2ec | 1112 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1113 | } |
276e680d | 1114 | |
0b86a832 CM |
1115 | return 0; |
1116 | } | |
1117 | ||
d352ac68 CM |
1118 | /* |
1119 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1120 | * |
1121 | * The error handling in this function may not be obvious. Any of the | |
1122 | * failures will cause the file system to go offline. We still need | |
1123 | * to clean up the delayed refs. | |
d352ac68 | 1124 | */ |
5d4f98a2 YZ |
1125 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
1126 | struct btrfs_root *root) | |
0b86a832 CM |
1127 | { |
1128 | struct btrfs_fs_info *fs_info = root->fs_info; | |
ea526d18 | 1129 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1130 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1131 | struct list_head *next; |
84234f3a | 1132 | struct extent_buffer *eb; |
56bec294 | 1133 | int ret; |
84234f3a YZ |
1134 | |
1135 | eb = btrfs_lock_root_node(fs_info->tree_root); | |
49b25e05 JM |
1136 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
1137 | 0, &eb); | |
84234f3a YZ |
1138 | btrfs_tree_unlock(eb); |
1139 | free_extent_buffer(eb); | |
0b86a832 | 1140 | |
49b25e05 JM |
1141 | if (ret) |
1142 | return ret; | |
1143 | ||
56bec294 | 1144 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1145 | if (ret) |
1146 | return ret; | |
87ef2bb4 | 1147 | |
733f4fbb | 1148 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
c16ce190 JB |
1149 | if (ret) |
1150 | return ret; | |
8dabb742 | 1151 | ret = btrfs_run_dev_replace(trans, root->fs_info); |
c16ce190 JB |
1152 | if (ret) |
1153 | return ret; | |
546adb0d | 1154 | ret = btrfs_run_qgroups(trans, root->fs_info); |
c16ce190 JB |
1155 | if (ret) |
1156 | return ret; | |
546adb0d | 1157 | |
dcdf7f6d JB |
1158 | ret = btrfs_setup_space_cache(trans, root); |
1159 | if (ret) | |
1160 | return ret; | |
1161 | ||
546adb0d JS |
1162 | /* run_qgroups might have added some more refs */ |
1163 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
c16ce190 JB |
1164 | if (ret) |
1165 | return ret; | |
ea526d18 | 1166 | again: |
d397712b | 1167 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
1168 | next = fs_info->dirty_cowonly_roots.next; |
1169 | list_del_init(next); | |
1170 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1171 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1172 | |
9e351cc8 JB |
1173 | if (root != fs_info->extent_root) |
1174 | list_add_tail(&root->dirty_list, | |
1175 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1176 | ret = update_cowonly_root(trans, root); |
1177 | if (ret) | |
1178 | return ret; | |
ea526d18 JB |
1179 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1180 | if (ret) | |
1181 | return ret; | |
79154b1b | 1182 | } |
276e680d | 1183 | |
1bbc621e | 1184 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
ea526d18 JB |
1185 | ret = btrfs_write_dirty_block_groups(trans, root); |
1186 | if (ret) | |
1187 | return ret; | |
1188 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
1189 | if (ret) | |
1190 | return ret; | |
1191 | } | |
1192 | ||
1193 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1194 | goto again; | |
1195 | ||
9e351cc8 JB |
1196 | list_add_tail(&fs_info->extent_root->dirty_list, |
1197 | &trans->transaction->switch_commits); | |
8dabb742 SB |
1198 | btrfs_after_dev_replace_commit(fs_info); |
1199 | ||
79154b1b CM |
1200 | return 0; |
1201 | } | |
1202 | ||
d352ac68 CM |
1203 | /* |
1204 | * dead roots are old snapshots that need to be deleted. This allocates | |
1205 | * a dirty root struct and adds it into the list of dead roots that need to | |
1206 | * be deleted | |
1207 | */ | |
cfad392b | 1208 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1209 | { |
a4abeea4 | 1210 | spin_lock(&root->fs_info->trans_lock); |
cfad392b JB |
1211 | if (list_empty(&root->root_list)) |
1212 | list_add_tail(&root->root_list, &root->fs_info->dead_roots); | |
a4abeea4 | 1213 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
1214 | } |
1215 | ||
d352ac68 | 1216 | /* |
5d4f98a2 | 1217 | * update all the cowonly tree roots on disk |
d352ac68 | 1218 | */ |
5d4f98a2 YZ |
1219 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
1220 | struct btrfs_root *root) | |
0f7d52f4 | 1221 | { |
0f7d52f4 | 1222 | struct btrfs_root *gang[8]; |
5d4f98a2 | 1223 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
1224 | int i; |
1225 | int ret; | |
54aa1f4d CM |
1226 | int err = 0; |
1227 | ||
a4abeea4 | 1228 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1229 | while (1) { |
5d4f98a2 YZ |
1230 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1231 | (void **)gang, 0, | |
0f7d52f4 CM |
1232 | ARRAY_SIZE(gang), |
1233 | BTRFS_ROOT_TRANS_TAG); | |
1234 | if (ret == 0) | |
1235 | break; | |
1236 | for (i = 0; i < ret; i++) { | |
1237 | root = gang[i]; | |
5d4f98a2 YZ |
1238 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1239 | (unsigned long)root->root_key.objectid, | |
1240 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1241 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1242 | |
e02119d5 | 1243 | btrfs_free_log(trans, root); |
5d4f98a2 | 1244 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 1245 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 1246 | |
82d5902d LZ |
1247 | btrfs_save_ino_cache(root, trans); |
1248 | ||
f1ebcc74 | 1249 | /* see comments in should_cow_block() */ |
27cdeb70 | 1250 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
c7548af6 | 1251 | smp_mb__after_atomic(); |
f1ebcc74 | 1252 | |
978d910d | 1253 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1254 | list_add_tail(&root->dirty_list, |
1255 | &trans->transaction->switch_commits); | |
978d910d YZ |
1256 | btrfs_set_root_node(&root->root_item, |
1257 | root->node); | |
1258 | } | |
5d4f98a2 | 1259 | |
5d4f98a2 | 1260 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1261 | &root->root_key, |
1262 | &root->root_item); | |
a4abeea4 | 1263 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1264 | if (err) |
1265 | break; | |
7174109c | 1266 | btrfs_qgroup_free_meta_all(root); |
0f7d52f4 CM |
1267 | } |
1268 | } | |
a4abeea4 | 1269 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1270 | return err; |
0f7d52f4 CM |
1271 | } |
1272 | ||
d352ac68 | 1273 | /* |
de78b51a ES |
1274 | * defrag a given btree. |
1275 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1276 | */ |
de78b51a | 1277 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1278 | { |
1279 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1280 | struct btrfs_trans_handle *trans; |
8929ecfa | 1281 | int ret; |
e9d0b13b | 1282 | |
27cdeb70 | 1283 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1284 | return 0; |
8929ecfa | 1285 | |
6b80053d | 1286 | while (1) { |
8929ecfa YZ |
1287 | trans = btrfs_start_transaction(root, 0); |
1288 | if (IS_ERR(trans)) | |
1289 | return PTR_ERR(trans); | |
1290 | ||
de78b51a | 1291 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1292 | |
e9d0b13b | 1293 | btrfs_end_transaction(trans, root); |
b53d3f5d | 1294 | btrfs_btree_balance_dirty(info->tree_root); |
e9d0b13b CM |
1295 | cond_resched(); |
1296 | ||
7841cb28 | 1297 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b | 1298 | break; |
210549eb DS |
1299 | |
1300 | if (btrfs_defrag_cancelled(root->fs_info)) { | |
efe120a0 | 1301 | pr_debug("BTRFS: defrag_root cancelled\n"); |
210549eb DS |
1302 | ret = -EAGAIN; |
1303 | break; | |
1304 | } | |
e9d0b13b | 1305 | } |
27cdeb70 | 1306 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1307 | return ret; |
e9d0b13b CM |
1308 | } |
1309 | ||
6426c7ad QW |
1310 | /* |
1311 | * Do all special snapshot related qgroup dirty hack. | |
1312 | * | |
1313 | * Will do all needed qgroup inherit and dirty hack like switch commit | |
1314 | * roots inside one transaction and write all btree into disk, to make | |
1315 | * qgroup works. | |
1316 | */ | |
1317 | static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, | |
1318 | struct btrfs_root *src, | |
1319 | struct btrfs_root *parent, | |
1320 | struct btrfs_qgroup_inherit *inherit, | |
1321 | u64 dst_objectid) | |
1322 | { | |
1323 | struct btrfs_fs_info *fs_info = src->fs_info; | |
1324 | int ret; | |
1325 | ||
1326 | /* | |
1327 | * Save some performance in the case that qgroups are not | |
1328 | * enabled. If this check races with the ioctl, rescan will | |
1329 | * kick in anyway. | |
1330 | */ | |
1331 | mutex_lock(&fs_info->qgroup_ioctl_lock); | |
afcdd129 | 1332 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
6426c7ad QW |
1333 | mutex_unlock(&fs_info->qgroup_ioctl_lock); |
1334 | return 0; | |
1335 | } | |
1336 | mutex_unlock(&fs_info->qgroup_ioctl_lock); | |
1337 | ||
1338 | /* | |
1339 | * We are going to commit transaction, see btrfs_commit_transaction() | |
1340 | * comment for reason locking tree_log_mutex | |
1341 | */ | |
1342 | mutex_lock(&fs_info->tree_log_mutex); | |
1343 | ||
1344 | ret = commit_fs_roots(trans, src); | |
1345 | if (ret) | |
1346 | goto out; | |
1347 | ret = btrfs_qgroup_prepare_account_extents(trans, fs_info); | |
1348 | if (ret < 0) | |
1349 | goto out; | |
1350 | ret = btrfs_qgroup_account_extents(trans, fs_info); | |
1351 | if (ret < 0) | |
1352 | goto out; | |
1353 | ||
1354 | /* Now qgroup are all updated, we can inherit it to new qgroups */ | |
1355 | ret = btrfs_qgroup_inherit(trans, fs_info, | |
1356 | src->root_key.objectid, dst_objectid, | |
1357 | inherit); | |
1358 | if (ret < 0) | |
1359 | goto out; | |
1360 | ||
1361 | /* | |
1362 | * Now we do a simplified commit transaction, which will: | |
1363 | * 1) commit all subvolume and extent tree | |
1364 | * To ensure all subvolume and extent tree have a valid | |
1365 | * commit_root to accounting later insert_dir_item() | |
1366 | * 2) write all btree blocks onto disk | |
1367 | * This is to make sure later btree modification will be cowed | |
1368 | * Or commit_root can be populated and cause wrong qgroup numbers | |
1369 | * In this simplified commit, we don't really care about other trees | |
1370 | * like chunk and root tree, as they won't affect qgroup. | |
1371 | * And we don't write super to avoid half committed status. | |
1372 | */ | |
1373 | ret = commit_cowonly_roots(trans, src); | |
1374 | if (ret) | |
1375 | goto out; | |
1376 | switch_commit_roots(trans->transaction, fs_info); | |
1377 | ret = btrfs_write_and_wait_transaction(trans, src); | |
1378 | if (ret) | |
f7af3934 | 1379 | btrfs_handle_fs_error(fs_info, ret, |
6426c7ad QW |
1380 | "Error while writing out transaction for qgroup"); |
1381 | ||
1382 | out: | |
1383 | mutex_unlock(&fs_info->tree_log_mutex); | |
1384 | ||
1385 | /* | |
1386 | * Force parent root to be updated, as we recorded it before so its | |
1387 | * last_trans == cur_transid. | |
1388 | * Or it won't be committed again onto disk after later | |
1389 | * insert_dir_item() | |
1390 | */ | |
1391 | if (!ret) | |
1392 | record_root_in_trans(trans, parent, 1); | |
1393 | return ret; | |
1394 | } | |
1395 | ||
d352ac68 CM |
1396 | /* |
1397 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1398 | * transaction commit. This does the actual creation. |
1399 | * | |
1400 | * Note: | |
1401 | * If the error which may affect the commitment of the current transaction | |
1402 | * happens, we should return the error number. If the error which just affect | |
1403 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1404 | */ |
80b6794d | 1405 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1406 | struct btrfs_fs_info *fs_info, |
1407 | struct btrfs_pending_snapshot *pending) | |
1408 | { | |
1409 | struct btrfs_key key; | |
80b6794d | 1410 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1411 | struct btrfs_root *tree_root = fs_info->tree_root; |
1412 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1413 | struct btrfs_root *parent_root; |
98c9942a | 1414 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1415 | struct inode *parent_inode; |
42874b3d MX |
1416 | struct btrfs_path *path; |
1417 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1418 | struct dentry *dentry; |
3063d29f | 1419 | struct extent_buffer *tmp; |
925baedd | 1420 | struct extent_buffer *old; |
04b285f3 | 1421 | struct timespec cur_time; |
aec8030a | 1422 | int ret = 0; |
d68fc57b | 1423 | u64 to_reserve = 0; |
6bdb72de | 1424 | u64 index = 0; |
a22285a6 | 1425 | u64 objectid; |
b83cc969 | 1426 | u64 root_flags; |
8ea05e3a | 1427 | uuid_le new_uuid; |
3063d29f | 1428 | |
8546b570 DS |
1429 | ASSERT(pending->path); |
1430 | path = pending->path; | |
42874b3d | 1431 | |
b0c0ea63 DS |
1432 | ASSERT(pending->root_item); |
1433 | new_root_item = pending->root_item; | |
a22285a6 | 1434 | |
aec8030a MX |
1435 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1436 | if (pending->error) | |
6fa9700e | 1437 | goto no_free_objectid; |
3063d29f | 1438 | |
d6726335 QW |
1439 | /* |
1440 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1441 | * accounted by later btrfs_qgroup_inherit(). | |
1442 | */ | |
1443 | btrfs_set_skip_qgroup(trans, objectid); | |
1444 | ||
147d256e | 1445 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
d68fc57b YZ |
1446 | |
1447 | if (to_reserve > 0) { | |
aec8030a MX |
1448 | pending->error = btrfs_block_rsv_add(root, |
1449 | &pending->block_rsv, | |
1450 | to_reserve, | |
1451 | BTRFS_RESERVE_NO_FLUSH); | |
1452 | if (pending->error) | |
d6726335 | 1453 | goto clear_skip_qgroup; |
d68fc57b YZ |
1454 | } |
1455 | ||
3063d29f | 1456 | key.objectid = objectid; |
a22285a6 YZ |
1457 | key.offset = (u64)-1; |
1458 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1459 | |
6fa9700e | 1460 | rsv = trans->block_rsv; |
a22285a6 | 1461 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1462 | trans->bytes_reserved = trans->block_rsv->reserved; |
88d3a5aa JB |
1463 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
1464 | trans->transid, | |
1465 | trans->bytes_reserved, 1); | |
a22285a6 | 1466 | dentry = pending->dentry; |
e9662f70 | 1467 | parent_inode = pending->dir; |
a22285a6 | 1468 | parent_root = BTRFS_I(parent_inode)->root; |
6426c7ad | 1469 | record_root_in_trans(trans, parent_root, 0); |
a22285a6 | 1470 | |
04b285f3 DD |
1471 | cur_time = current_fs_time(parent_inode->i_sb); |
1472 | ||
3063d29f CM |
1473 | /* |
1474 | * insert the directory item | |
1475 | */ | |
3de4586c | 1476 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1477 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1478 | |
1479 | /* check if there is a file/dir which has the same name. */ | |
1480 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
1481 | btrfs_ino(parent_inode), | |
1482 | dentry->d_name.name, | |
1483 | dentry->d_name.len, 0); | |
1484 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1485 | pending->error = -EEXIST; |
aec8030a | 1486 | goto dir_item_existed; |
42874b3d MX |
1487 | } else if (IS_ERR(dir_item)) { |
1488 | ret = PTR_ERR(dir_item); | |
66642832 | 1489 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1490 | goto fail; |
79787eaa | 1491 | } |
42874b3d | 1492 | btrfs_release_path(path); |
52c26179 | 1493 | |
e999376f CM |
1494 | /* |
1495 | * pull in the delayed directory update | |
1496 | * and the delayed inode item | |
1497 | * otherwise we corrupt the FS during | |
1498 | * snapshot | |
1499 | */ | |
1500 | ret = btrfs_run_delayed_items(trans, root); | |
8732d44f | 1501 | if (ret) { /* Transaction aborted */ |
66642832 | 1502 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1503 | goto fail; |
1504 | } | |
e999376f | 1505 | |
6426c7ad | 1506 | record_root_in_trans(trans, root, 0); |
6bdb72de SW |
1507 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1508 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1509 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1510 | |
b83cc969 LZ |
1511 | root_flags = btrfs_root_flags(new_root_item); |
1512 | if (pending->readonly) | |
1513 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1514 | else | |
1515 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1516 | btrfs_set_root_flags(new_root_item, root_flags); | |
1517 | ||
8ea05e3a AB |
1518 | btrfs_set_root_generation_v2(new_root_item, |
1519 | trans->transid); | |
1520 | uuid_le_gen(&new_uuid); | |
1521 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1522 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1523 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1524 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1525 | memset(new_root_item->received_uuid, 0, | |
1526 | sizeof(new_root_item->received_uuid)); | |
1527 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1528 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1529 | btrfs_set_root_stransid(new_root_item, 0); | |
1530 | btrfs_set_root_rtransid(new_root_item, 0); | |
1531 | } | |
3cae210f QW |
1532 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1533 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1534 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1535 | |
6bdb72de | 1536 | old = btrfs_lock_root_node(root); |
49b25e05 | 1537 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1538 | if (ret) { |
1539 | btrfs_tree_unlock(old); | |
1540 | free_extent_buffer(old); | |
66642832 | 1541 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1542 | goto fail; |
79787eaa | 1543 | } |
49b25e05 | 1544 | |
6bdb72de SW |
1545 | btrfs_set_lock_blocking(old); |
1546 | ||
49b25e05 | 1547 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1548 | /* clean up in any case */ |
6bdb72de SW |
1549 | btrfs_tree_unlock(old); |
1550 | free_extent_buffer(old); | |
8732d44f | 1551 | if (ret) { |
66642832 | 1552 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1553 | goto fail; |
1554 | } | |
f1ebcc74 | 1555 | /* see comments in should_cow_block() */ |
27cdeb70 | 1556 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1557 | smp_wmb(); |
1558 | ||
6bdb72de | 1559 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1560 | /* record when the snapshot was created in key.offset */ |
1561 | key.offset = trans->transid; | |
1562 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1563 | btrfs_tree_unlock(tmp); |
1564 | free_extent_buffer(tmp); | |
8732d44f | 1565 | if (ret) { |
66642832 | 1566 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1567 | goto fail; |
1568 | } | |
6bdb72de | 1569 | |
a22285a6 YZ |
1570 | /* |
1571 | * insert root back/forward references | |
1572 | */ | |
1573 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1574 | parent_root->root_key.objectid, |
33345d01 | 1575 | btrfs_ino(parent_inode), index, |
a22285a6 | 1576 | dentry->d_name.name, dentry->d_name.len); |
8732d44f | 1577 | if (ret) { |
66642832 | 1578 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1579 | goto fail; |
1580 | } | |
0660b5af | 1581 | |
a22285a6 YZ |
1582 | key.offset = (u64)-1; |
1583 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1584 | if (IS_ERR(pending->snap)) { |
1585 | ret = PTR_ERR(pending->snap); | |
66642832 | 1586 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1587 | goto fail; |
79787eaa | 1588 | } |
d68fc57b | 1589 | |
49b25e05 | 1590 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f | 1591 | if (ret) { |
66642832 | 1592 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1593 | goto fail; |
1594 | } | |
361048f5 MX |
1595 | |
1596 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
8732d44f | 1597 | if (ret) { |
66642832 | 1598 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1599 | goto fail; |
1600 | } | |
42874b3d | 1601 | |
6426c7ad QW |
1602 | /* |
1603 | * Do special qgroup accounting for snapshot, as we do some qgroup | |
1604 | * snapshot hack to do fast snapshot. | |
1605 | * To co-operate with that hack, we do hack again. | |
1606 | * Or snapshot will be greatly slowed down by a subtree qgroup rescan | |
1607 | */ | |
1608 | ret = qgroup_account_snapshot(trans, root, parent_root, | |
1609 | pending->inherit, objectid); | |
1610 | if (ret < 0) | |
1611 | goto fail; | |
1612 | ||
42874b3d MX |
1613 | ret = btrfs_insert_dir_item(trans, parent_root, |
1614 | dentry->d_name.name, dentry->d_name.len, | |
1615 | parent_inode, &key, | |
1616 | BTRFS_FT_DIR, index); | |
1617 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1618 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f | 1619 | if (ret) { |
66642832 | 1620 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1621 | goto fail; |
1622 | } | |
42874b3d MX |
1623 | |
1624 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
1625 | dentry->d_name.len * 2); | |
04b285f3 DD |
1626 | parent_inode->i_mtime = parent_inode->i_ctime = |
1627 | current_fs_time(parent_inode->i_sb); | |
be6aef60 | 1628 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
dd5f9615 | 1629 | if (ret) { |
66642832 | 1630 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1631 | goto fail; |
1632 | } | |
1633 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b, | |
1634 | BTRFS_UUID_KEY_SUBVOL, objectid); | |
1635 | if (ret) { | |
66642832 | 1636 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1637 | goto fail; |
1638 | } | |
1639 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
1640 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, | |
1641 | new_root_item->received_uuid, | |
1642 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
1643 | objectid); | |
1644 | if (ret && ret != -EEXIST) { | |
66642832 | 1645 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1646 | goto fail; |
1647 | } | |
1648 | } | |
d6726335 QW |
1649 | |
1650 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
1651 | if (ret) { | |
66642832 | 1652 | btrfs_abort_transaction(trans, ret); |
d6726335 QW |
1653 | goto fail; |
1654 | } | |
1655 | ||
3063d29f | 1656 | fail: |
aec8030a MX |
1657 | pending->error = ret; |
1658 | dir_item_existed: | |
98c9942a | 1659 | trans->block_rsv = rsv; |
2382c5cc | 1660 | trans->bytes_reserved = 0; |
d6726335 QW |
1661 | clear_skip_qgroup: |
1662 | btrfs_clear_skip_qgroup(trans); | |
6fa9700e MX |
1663 | no_free_objectid: |
1664 | kfree(new_root_item); | |
b0c0ea63 | 1665 | pending->root_item = NULL; |
42874b3d | 1666 | btrfs_free_path(path); |
8546b570 DS |
1667 | pending->path = NULL; |
1668 | ||
49b25e05 | 1669 | return ret; |
3063d29f CM |
1670 | } |
1671 | ||
d352ac68 CM |
1672 | /* |
1673 | * create all the snapshots we've scheduled for creation | |
1674 | */ | |
80b6794d CM |
1675 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1676 | struct btrfs_fs_info *fs_info) | |
3de4586c | 1677 | { |
aec8030a | 1678 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1679 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1680 | int ret = 0; |
3de4586c | 1681 | |
aec8030a MX |
1682 | list_for_each_entry_safe(pending, next, head, list) { |
1683 | list_del(&pending->list); | |
1684 | ret = create_pending_snapshot(trans, fs_info, pending); | |
1685 | if (ret) | |
1686 | break; | |
1687 | } | |
1688 | return ret; | |
3de4586c CM |
1689 | } |
1690 | ||
5d4f98a2 YZ |
1691 | static void update_super_roots(struct btrfs_root *root) |
1692 | { | |
1693 | struct btrfs_root_item *root_item; | |
1694 | struct btrfs_super_block *super; | |
1695 | ||
6c41761f | 1696 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1697 | |
1698 | root_item = &root->fs_info->chunk_root->root_item; | |
1699 | super->chunk_root = root_item->bytenr; | |
1700 | super->chunk_root_generation = root_item->generation; | |
1701 | super->chunk_root_level = root_item->level; | |
1702 | ||
1703 | root_item = &root->fs_info->tree_root->root_item; | |
1704 | super->root = root_item->bytenr; | |
1705 | super->generation = root_item->generation; | |
1706 | super->root_level = root_item->level; | |
3cdde224 | 1707 | if (btrfs_test_opt(root->fs_info, SPACE_CACHE)) |
0af3d00b | 1708 | super->cache_generation = root_item->generation; |
afcdd129 | 1709 | if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &root->fs_info->flags)) |
70f80175 | 1710 | super->uuid_tree_generation = root_item->generation; |
5d4f98a2 YZ |
1711 | } |
1712 | ||
f36f3042 CM |
1713 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1714 | { | |
4a9d8bde | 1715 | struct btrfs_transaction *trans; |
f36f3042 | 1716 | int ret = 0; |
4a9d8bde | 1717 | |
a4abeea4 | 1718 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1719 | trans = info->running_transaction; |
1720 | if (trans) | |
1721 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1722 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1723 | return ret; |
1724 | } | |
1725 | ||
8929ecfa YZ |
1726 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1727 | { | |
4a9d8bde | 1728 | struct btrfs_transaction *trans; |
8929ecfa | 1729 | int ret = 0; |
4a9d8bde | 1730 | |
a4abeea4 | 1731 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1732 | trans = info->running_transaction; |
1733 | if (trans) | |
1734 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1735 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1736 | return ret; |
1737 | } | |
1738 | ||
bb9c12c9 SW |
1739 | /* |
1740 | * wait for the current transaction commit to start and block subsequent | |
1741 | * transaction joins | |
1742 | */ | |
1743 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1744 | struct btrfs_transaction *trans) | |
1745 | { | |
4a9d8bde | 1746 | wait_event(root->fs_info->transaction_blocked_wait, |
501407aa JB |
1747 | trans->state >= TRANS_STATE_COMMIT_START || |
1748 | trans->aborted); | |
bb9c12c9 SW |
1749 | } |
1750 | ||
1751 | /* | |
1752 | * wait for the current transaction to start and then become unblocked. | |
1753 | * caller holds ref. | |
1754 | */ | |
1755 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1756 | struct btrfs_transaction *trans) | |
1757 | { | |
72d63ed6 | 1758 | wait_event(root->fs_info->transaction_wait, |
501407aa JB |
1759 | trans->state >= TRANS_STATE_UNBLOCKED || |
1760 | trans->aborted); | |
bb9c12c9 SW |
1761 | } |
1762 | ||
1763 | /* | |
1764 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1765 | * returns, any subsequent transaction will not be allowed to join. | |
1766 | */ | |
1767 | struct btrfs_async_commit { | |
1768 | struct btrfs_trans_handle *newtrans; | |
1769 | struct btrfs_root *root; | |
7892b5af | 1770 | struct work_struct work; |
bb9c12c9 SW |
1771 | }; |
1772 | ||
1773 | static void do_async_commit(struct work_struct *work) | |
1774 | { | |
1775 | struct btrfs_async_commit *ac = | |
7892b5af | 1776 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1777 | |
6fc4e354 SW |
1778 | /* |
1779 | * We've got freeze protection passed with the transaction. | |
1780 | * Tell lockdep about it. | |
1781 | */ | |
b1a06a4b | 1782 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
bee9182d | 1783 | __sb_writers_acquired(ac->root->fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1784 | |
e209db7a SW |
1785 | current->journal_info = ac->newtrans; |
1786 | ||
bb9c12c9 SW |
1787 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1788 | kfree(ac); | |
1789 | } | |
1790 | ||
1791 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1792 | struct btrfs_root *root, | |
1793 | int wait_for_unblock) | |
1794 | { | |
1795 | struct btrfs_async_commit *ac; | |
1796 | struct btrfs_transaction *cur_trans; | |
1797 | ||
1798 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1799 | if (!ac) |
1800 | return -ENOMEM; | |
bb9c12c9 | 1801 | |
7892b5af | 1802 | INIT_WORK(&ac->work, do_async_commit); |
bb9c12c9 | 1803 | ac->root = root; |
7a7eaa40 | 1804 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1805 | if (IS_ERR(ac->newtrans)) { |
1806 | int err = PTR_ERR(ac->newtrans); | |
1807 | kfree(ac); | |
1808 | return err; | |
1809 | } | |
bb9c12c9 SW |
1810 | |
1811 | /* take transaction reference */ | |
bb9c12c9 | 1812 | cur_trans = trans->transaction; |
13c5a93e | 1813 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1814 | |
1815 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1816 | |
1817 | /* | |
1818 | * Tell lockdep we've released the freeze rwsem, since the | |
1819 | * async commit thread will be the one to unlock it. | |
1820 | */ | |
b1a06a4b | 1821 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
bee9182d | 1822 | __sb_writers_release(root->fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1823 | |
7892b5af | 1824 | schedule_work(&ac->work); |
bb9c12c9 SW |
1825 | |
1826 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1827 | if (wait_for_unblock) |
1828 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1829 | else | |
1830 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1831 | |
38e88054 SW |
1832 | if (current->journal_info == trans) |
1833 | current->journal_info = NULL; | |
1834 | ||
724e2315 | 1835 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1836 | return 0; |
1837 | } | |
1838 | ||
49b25e05 JM |
1839 | |
1840 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1841 | struct btrfs_root *root, int err) |
49b25e05 JM |
1842 | { |
1843 | struct btrfs_transaction *cur_trans = trans->transaction; | |
f094ac32 | 1844 | DEFINE_WAIT(wait); |
49b25e05 JM |
1845 | |
1846 | WARN_ON(trans->use_count > 1); | |
1847 | ||
66642832 | 1848 | btrfs_abort_transaction(trans, err); |
7b8b92af | 1849 | |
49b25e05 | 1850 | spin_lock(&root->fs_info->trans_lock); |
66b6135b | 1851 | |
25d8c284 MX |
1852 | /* |
1853 | * If the transaction is removed from the list, it means this | |
1854 | * transaction has been committed successfully, so it is impossible | |
1855 | * to call the cleanup function. | |
1856 | */ | |
1857 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1858 | |
49b25e05 | 1859 | list_del_init(&cur_trans->list); |
d7096fc3 | 1860 | if (cur_trans == root->fs_info->running_transaction) { |
4a9d8bde | 1861 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
f094ac32 LB |
1862 | spin_unlock(&root->fs_info->trans_lock); |
1863 | wait_event(cur_trans->writer_wait, | |
1864 | atomic_read(&cur_trans->num_writers) == 1); | |
1865 | ||
1866 | spin_lock(&root->fs_info->trans_lock); | |
d7096fc3 | 1867 | } |
49b25e05 JM |
1868 | spin_unlock(&root->fs_info->trans_lock); |
1869 | ||
1870 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1871 | ||
4a9d8bde MX |
1872 | spin_lock(&root->fs_info->trans_lock); |
1873 | if (cur_trans == root->fs_info->running_transaction) | |
1874 | root->fs_info->running_transaction = NULL; | |
1875 | spin_unlock(&root->fs_info->trans_lock); | |
1876 | ||
e0228285 JB |
1877 | if (trans->type & __TRANS_FREEZABLE) |
1878 | sb_end_intwrite(root->fs_info->sb); | |
724e2315 JB |
1879 | btrfs_put_transaction(cur_trans); |
1880 | btrfs_put_transaction(cur_trans); | |
49b25e05 JM |
1881 | |
1882 | trace_btrfs_transaction_commit(root); | |
1883 | ||
49b25e05 JM |
1884 | if (current->journal_info == trans) |
1885 | current->journal_info = NULL; | |
c0af8f0b | 1886 | btrfs_scrub_cancel(root->fs_info); |
49b25e05 JM |
1887 | |
1888 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1889 | } | |
1890 | ||
82436617 MX |
1891 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
1892 | { | |
3cdde224 | 1893 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
6c255e67 | 1894 | return btrfs_start_delalloc_roots(fs_info, 1, -1); |
82436617 MX |
1895 | return 0; |
1896 | } | |
1897 | ||
1898 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) | |
1899 | { | |
3cdde224 | 1900 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
578def7c | 1901 | btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1); |
82436617 MX |
1902 | } |
1903 | ||
50d9aa99 | 1904 | static inline void |
161c3549 | 1905 | btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans) |
50d9aa99 | 1906 | { |
161c3549 JB |
1907 | wait_event(cur_trans->pending_wait, |
1908 | atomic_read(&cur_trans->pending_ordered) == 0); | |
50d9aa99 JB |
1909 | } |
1910 | ||
79154b1b CM |
1911 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1912 | struct btrfs_root *root) | |
1913 | { | |
49b25e05 | 1914 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1915 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 1916 | int ret; |
79154b1b | 1917 | |
8d25a086 MX |
1918 | /* Stop the commit early if ->aborted is set */ |
1919 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
25287e0a | 1920 | ret = cur_trans->aborted; |
e4a2bcac JB |
1921 | btrfs_end_transaction(trans, root); |
1922 | return ret; | |
25287e0a | 1923 | } |
49b25e05 | 1924 | |
56bec294 CM |
1925 | /* make a pass through all the delayed refs we have so far |
1926 | * any runnings procs may add more while we are here | |
1927 | */ | |
1928 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
e4a2bcac JB |
1929 | if (ret) { |
1930 | btrfs_end_transaction(trans, root); | |
1931 | return ret; | |
1932 | } | |
56bec294 | 1933 | |
0e721106 JB |
1934 | btrfs_trans_release_metadata(trans, root); |
1935 | trans->block_rsv = NULL; | |
1936 | ||
b7ec40d7 | 1937 | cur_trans = trans->transaction; |
49b25e05 | 1938 | |
56bec294 CM |
1939 | /* |
1940 | * set the flushing flag so procs in this transaction have to | |
1941 | * start sending their work down. | |
1942 | */ | |
b7ec40d7 | 1943 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 1944 | smp_wmb(); |
56bec294 | 1945 | |
ea658bad JB |
1946 | if (!list_empty(&trans->new_bgs)) |
1947 | btrfs_create_pending_block_groups(trans, root); | |
1948 | ||
c3e69d58 | 1949 | ret = btrfs_run_delayed_refs(trans, root, 0); |
e4a2bcac JB |
1950 | if (ret) { |
1951 | btrfs_end_transaction(trans, root); | |
1952 | return ret; | |
1953 | } | |
56bec294 | 1954 | |
3204d33c | 1955 | if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { |
1bbc621e CM |
1956 | int run_it = 0; |
1957 | ||
1958 | /* this mutex is also taken before trying to set | |
1959 | * block groups readonly. We need to make sure | |
1960 | * that nobody has set a block group readonly | |
1961 | * after a extents from that block group have been | |
1962 | * allocated for cache files. btrfs_set_block_group_ro | |
1963 | * will wait for the transaction to commit if it | |
3204d33c | 1964 | * finds BTRFS_TRANS_DIRTY_BG_RUN set. |
1bbc621e | 1965 | * |
3204d33c JB |
1966 | * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure |
1967 | * only one process starts all the block group IO. It wouldn't | |
1bbc621e CM |
1968 | * hurt to have more than one go through, but there's no |
1969 | * real advantage to it either. | |
1970 | */ | |
1971 | mutex_lock(&root->fs_info->ro_block_group_mutex); | |
3204d33c JB |
1972 | if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, |
1973 | &cur_trans->flags)) | |
1bbc621e | 1974 | run_it = 1; |
1bbc621e CM |
1975 | mutex_unlock(&root->fs_info->ro_block_group_mutex); |
1976 | ||
1977 | if (run_it) | |
1978 | ret = btrfs_start_dirty_block_groups(trans, root); | |
1979 | } | |
1980 | if (ret) { | |
1981 | btrfs_end_transaction(trans, root); | |
1982 | return ret; | |
1983 | } | |
1984 | ||
4a9d8bde MX |
1985 | spin_lock(&root->fs_info->trans_lock); |
1986 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { | |
1987 | spin_unlock(&root->fs_info->trans_lock); | |
13c5a93e | 1988 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1989 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1990 | |
b9c8300c | 1991 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1992 | |
b4924a0f LB |
1993 | if (unlikely(cur_trans->aborted)) |
1994 | ret = cur_trans->aborted; | |
1995 | ||
724e2315 | 1996 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 1997 | |
49b25e05 | 1998 | return ret; |
79154b1b | 1999 | } |
4313b399 | 2000 | |
4a9d8bde | 2001 | cur_trans->state = TRANS_STATE_COMMIT_START; |
bb9c12c9 SW |
2002 | wake_up(&root->fs_info->transaction_blocked_wait); |
2003 | ||
ccd467d6 CM |
2004 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
2005 | prev_trans = list_entry(cur_trans->list.prev, | |
2006 | struct btrfs_transaction, list); | |
4a9d8bde | 2007 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
13c5a93e | 2008 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 2009 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
2010 | |
2011 | wait_for_commit(root, prev_trans); | |
1f9b8c8f | 2012 | ret = prev_trans->aborted; |
ccd467d6 | 2013 | |
724e2315 | 2014 | btrfs_put_transaction(prev_trans); |
1f9b8c8f FM |
2015 | if (ret) |
2016 | goto cleanup_transaction; | |
a4abeea4 JB |
2017 | } else { |
2018 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 2019 | } |
a4abeea4 JB |
2020 | } else { |
2021 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 2022 | } |
15ee9bc7 | 2023 | |
0860adfd MX |
2024 | extwriter_counter_dec(cur_trans, trans->type); |
2025 | ||
82436617 MX |
2026 | ret = btrfs_start_delalloc_flush(root->fs_info); |
2027 | if (ret) | |
2028 | goto cleanup_transaction; | |
2029 | ||
8d875f95 | 2030 | ret = btrfs_run_delayed_items(trans, root); |
581227d0 MX |
2031 | if (ret) |
2032 | goto cleanup_transaction; | |
15ee9bc7 | 2033 | |
581227d0 MX |
2034 | wait_event(cur_trans->writer_wait, |
2035 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 2036 | |
581227d0 | 2037 | /* some pending stuffs might be added after the previous flush. */ |
8d875f95 | 2038 | ret = btrfs_run_delayed_items(trans, root); |
ca469637 MX |
2039 | if (ret) |
2040 | goto cleanup_transaction; | |
2041 | ||
82436617 | 2042 | btrfs_wait_delalloc_flush(root->fs_info); |
cb7ab021 | 2043 | |
161c3549 | 2044 | btrfs_wait_pending_ordered(cur_trans); |
50d9aa99 | 2045 | |
cb7ab021 | 2046 | btrfs_scrub_pause(root); |
ed0ca140 JB |
2047 | /* |
2048 | * Ok now we need to make sure to block out any other joins while we | |
2049 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 2050 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 JB |
2051 | */ |
2052 | spin_lock(&root->fs_info->trans_lock); | |
4a9d8bde | 2053 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
ed0ca140 JB |
2054 | spin_unlock(&root->fs_info->trans_lock); |
2055 | wait_event(cur_trans->writer_wait, | |
2056 | atomic_read(&cur_trans->num_writers) == 1); | |
2057 | ||
2cba30f1 MX |
2058 | /* ->aborted might be set after the previous check, so check it */ |
2059 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
2060 | ret = cur_trans->aborted; | |
6cf7f77e | 2061 | goto scrub_continue; |
2cba30f1 | 2062 | } |
7585717f CM |
2063 | /* |
2064 | * the reloc mutex makes sure that we stop | |
2065 | * the balancing code from coming in and moving | |
2066 | * extents around in the middle of the commit | |
2067 | */ | |
2068 | mutex_lock(&root->fs_info->reloc_mutex); | |
2069 | ||
42874b3d MX |
2070 | /* |
2071 | * We needn't worry about the delayed items because we will | |
2072 | * deal with them in create_pending_snapshot(), which is the | |
2073 | * core function of the snapshot creation. | |
2074 | */ | |
2075 | ret = create_pending_snapshots(trans, root->fs_info); | |
49b25e05 JM |
2076 | if (ret) { |
2077 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 2078 | goto scrub_continue; |
49b25e05 | 2079 | } |
3063d29f | 2080 | |
42874b3d MX |
2081 | /* |
2082 | * We insert the dir indexes of the snapshots and update the inode | |
2083 | * of the snapshots' parents after the snapshot creation, so there | |
2084 | * are some delayed items which are not dealt with. Now deal with | |
2085 | * them. | |
2086 | * | |
2087 | * We needn't worry that this operation will corrupt the snapshots, | |
2088 | * because all the tree which are snapshoted will be forced to COW | |
2089 | * the nodes and leaves. | |
2090 | */ | |
2091 | ret = btrfs_run_delayed_items(trans, root); | |
49b25e05 JM |
2092 | if (ret) { |
2093 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 2094 | goto scrub_continue; |
49b25e05 | 2095 | } |
16cdcec7 | 2096 | |
56bec294 | 2097 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
2098 | if (ret) { |
2099 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 2100 | goto scrub_continue; |
49b25e05 | 2101 | } |
56bec294 | 2102 | |
0ed4792a QW |
2103 | /* Reocrd old roots for later qgroup accounting */ |
2104 | ret = btrfs_qgroup_prepare_account_extents(trans, root->fs_info); | |
2105 | if (ret) { | |
2106 | mutex_unlock(&root->fs_info->reloc_mutex); | |
2107 | goto scrub_continue; | |
2108 | } | |
2109 | ||
e999376f CM |
2110 | /* |
2111 | * make sure none of the code above managed to slip in a | |
2112 | * delayed item | |
2113 | */ | |
2114 | btrfs_assert_delayed_root_empty(root); | |
2115 | ||
2c90e5d6 | 2116 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 2117 | |
e02119d5 CM |
2118 | /* btrfs_commit_tree_roots is responsible for getting the |
2119 | * various roots consistent with each other. Every pointer | |
2120 | * in the tree of tree roots has to point to the most up to date | |
2121 | * root for every subvolume and other tree. So, we have to keep | |
2122 | * the tree logging code from jumping in and changing any | |
2123 | * of the trees. | |
2124 | * | |
2125 | * At this point in the commit, there can't be any tree-log | |
2126 | * writers, but a little lower down we drop the trans mutex | |
2127 | * and let new people in. By holding the tree_log_mutex | |
2128 | * from now until after the super is written, we avoid races | |
2129 | * with the tree-log code. | |
2130 | */ | |
2131 | mutex_lock(&root->fs_info->tree_log_mutex); | |
2132 | ||
5d4f98a2 | 2133 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
2134 | if (ret) { |
2135 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 2136 | mutex_unlock(&root->fs_info->reloc_mutex); |
6cf7f77e | 2137 | goto scrub_continue; |
49b25e05 | 2138 | } |
54aa1f4d | 2139 | |
3818aea2 | 2140 | /* |
7e1876ac DS |
2141 | * Since the transaction is done, we can apply the pending changes |
2142 | * before the next transaction. | |
3818aea2 | 2143 | */ |
572d9ab7 | 2144 | btrfs_apply_pending_changes(root->fs_info); |
3818aea2 | 2145 | |
5d4f98a2 | 2146 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2147 | * safe to free the root of tree log roots |
2148 | */ | |
2149 | btrfs_free_log_root_tree(trans, root->fs_info); | |
2150 | ||
0ed4792a QW |
2151 | /* |
2152 | * Since fs roots are all committed, we can get a quite accurate | |
2153 | * new_roots. So let's do quota accounting. | |
2154 | */ | |
2155 | ret = btrfs_qgroup_account_extents(trans, root->fs_info); | |
2156 | if (ret < 0) { | |
2157 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
2158 | mutex_unlock(&root->fs_info->reloc_mutex); | |
2159 | goto scrub_continue; | |
2160 | } | |
2161 | ||
5d4f98a2 | 2162 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
2163 | if (ret) { |
2164 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 2165 | mutex_unlock(&root->fs_info->reloc_mutex); |
6cf7f77e | 2166 | goto scrub_continue; |
49b25e05 | 2167 | } |
54aa1f4d | 2168 | |
2cba30f1 MX |
2169 | /* |
2170 | * The tasks which save the space cache and inode cache may also | |
2171 | * update ->aborted, check it. | |
2172 | */ | |
2173 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
2174 | ret = cur_trans->aborted; | |
2175 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
2176 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 2177 | goto scrub_continue; |
2cba30f1 MX |
2178 | } |
2179 | ||
11833d66 YZ |
2180 | btrfs_prepare_extent_commit(trans, root); |
2181 | ||
78fae27e | 2182 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
2183 | |
2184 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
2185 | root->fs_info->tree_root->node); | |
9e351cc8 JB |
2186 | list_add_tail(&root->fs_info->tree_root->dirty_list, |
2187 | &cur_trans->switch_commits); | |
5d4f98a2 YZ |
2188 | |
2189 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
2190 | root->fs_info->chunk_root->node); | |
9e351cc8 JB |
2191 | list_add_tail(&root->fs_info->chunk_root->dirty_list, |
2192 | &cur_trans->switch_commits); | |
2193 | ||
2194 | switch_commit_roots(cur_trans, root->fs_info); | |
5d4f98a2 | 2195 | |
edf39272 | 2196 | assert_qgroups_uptodate(trans); |
ce93ec54 | 2197 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2198 | ASSERT(list_empty(&cur_trans->io_bgs)); |
5d4f98a2 | 2199 | update_super_roots(root); |
e02119d5 | 2200 | |
60e7cd3a JB |
2201 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
2202 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
6c41761f DS |
2203 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
2204 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 2205 | |
935e5cc9 | 2206 | btrfs_update_commit_device_size(root->fs_info); |
ce7213c7 | 2207 | btrfs_update_commit_device_bytes_used(root, cur_trans); |
935e5cc9 | 2208 | |
afcdd129 JB |
2209 | clear_bit(BTRFS_FS_LOG1_ERR, &root->fs_info->flags); |
2210 | clear_bit(BTRFS_FS_LOG2_ERR, &root->fs_info->flags); | |
656f30db | 2211 | |
4fbcdf66 FM |
2212 | btrfs_trans_release_chunk_metadata(trans); |
2213 | ||
a4abeea4 | 2214 | spin_lock(&root->fs_info->trans_lock); |
4a9d8bde | 2215 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
a4abeea4 | 2216 | root->fs_info->running_transaction = NULL; |
a4abeea4 | 2217 | spin_unlock(&root->fs_info->trans_lock); |
7585717f | 2218 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 2219 | |
f9295749 | 2220 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 2221 | |
79154b1b | 2222 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 | 2223 | if (ret) { |
34d97007 | 2224 | btrfs_handle_fs_error(root->fs_info, ret, |
08748810 | 2225 | "Error while writing out transaction"); |
49b25e05 | 2226 | mutex_unlock(&root->fs_info->tree_log_mutex); |
6cf7f77e | 2227 | goto scrub_continue; |
49b25e05 JM |
2228 | } |
2229 | ||
2230 | ret = write_ctree_super(trans, root, 0); | |
2231 | if (ret) { | |
2232 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
6cf7f77e | 2233 | goto scrub_continue; |
49b25e05 | 2234 | } |
4313b399 | 2235 | |
e02119d5 CM |
2236 | /* |
2237 | * the super is written, we can safely allow the tree-loggers | |
2238 | * to go about their business | |
2239 | */ | |
2240 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
2241 | ||
11833d66 | 2242 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 2243 | |
3204d33c | 2244 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) |
13212b54 ZL |
2245 | btrfs_clear_space_info_full(root->fs_info); |
2246 | ||
15ee9bc7 | 2247 | root->fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
2248 | /* |
2249 | * We needn't acquire the lock here because there is no other task | |
2250 | * which can change it. | |
2251 | */ | |
2252 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2253 | wake_up(&cur_trans->commit_wait); |
3de4586c | 2254 | |
a4abeea4 | 2255 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 2256 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
2257 | spin_unlock(&root->fs_info->trans_lock); |
2258 | ||
724e2315 JB |
2259 | btrfs_put_transaction(cur_trans); |
2260 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2261 | |
0860adfd | 2262 | if (trans->type & __TRANS_FREEZABLE) |
354aa0fb | 2263 | sb_end_intwrite(root->fs_info->sb); |
b2b5ef5c | 2264 | |
1abe9b8a | 2265 | trace_btrfs_transaction_commit(root); |
2266 | ||
a2de733c AJ |
2267 | btrfs_scrub_continue(root); |
2268 | ||
9ed74f2d JB |
2269 | if (current->journal_info == trans) |
2270 | current->journal_info = NULL; | |
2271 | ||
2c90e5d6 | 2272 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2273 | |
9e7cc91a WX |
2274 | /* |
2275 | * If fs has been frozen, we can not handle delayed iputs, otherwise | |
2276 | * it'll result in deadlock about SB_FREEZE_FS. | |
2277 | */ | |
8a733013 | 2278 | if (current != root->fs_info->transaction_kthread && |
9e7cc91a WX |
2279 | current != root->fs_info->cleaner_kthread && |
2280 | !root->fs_info->fs_frozen) | |
24bbcf04 YZ |
2281 | btrfs_run_delayed_iputs(root); |
2282 | ||
79154b1b | 2283 | return ret; |
49b25e05 | 2284 | |
6cf7f77e WS |
2285 | scrub_continue: |
2286 | btrfs_scrub_continue(root); | |
49b25e05 | 2287 | cleanup_transaction: |
0e721106 | 2288 | btrfs_trans_release_metadata(trans, root); |
4fbcdf66 | 2289 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2290 | trans->block_rsv = NULL; |
c2cf52eb | 2291 | btrfs_warn(root->fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2292 | if (current->journal_info == trans) |
2293 | current->journal_info = NULL; | |
7b8b92af | 2294 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
2295 | |
2296 | return ret; | |
79154b1b CM |
2297 | } |
2298 | ||
d352ac68 | 2299 | /* |
9d1a2a3a DS |
2300 | * return < 0 if error |
2301 | * 0 if there are no more dead_roots at the time of call | |
2302 | * 1 there are more to be processed, call me again | |
2303 | * | |
2304 | * The return value indicates there are certainly more snapshots to delete, but | |
2305 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2306 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2307 | * few seconds later. | |
d352ac68 | 2308 | */ |
9d1a2a3a | 2309 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 2310 | { |
9d1a2a3a | 2311 | int ret; |
5d4f98a2 YZ |
2312 | struct btrfs_fs_info *fs_info = root->fs_info; |
2313 | ||
a4abeea4 | 2314 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2315 | if (list_empty(&fs_info->dead_roots)) { |
2316 | spin_unlock(&fs_info->trans_lock); | |
2317 | return 0; | |
2318 | } | |
2319 | root = list_first_entry(&fs_info->dead_roots, | |
2320 | struct btrfs_root, root_list); | |
cfad392b | 2321 | list_del_init(&root->root_list); |
a4abeea4 | 2322 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2323 | |
efe120a0 | 2324 | pr_debug("BTRFS: cleaner removing %llu\n", root->objectid); |
76dda93c | 2325 | |
9d1a2a3a | 2326 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2327 | |
9d1a2a3a DS |
2328 | if (btrfs_header_backref_rev(root->node) < |
2329 | BTRFS_MIXED_BACKREF_REV) | |
2330 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
2331 | else | |
2332 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
32471dc2 | 2333 | |
6596a928 | 2334 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2335 | } |
572d9ab7 DS |
2336 | |
2337 | void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) | |
2338 | { | |
2339 | unsigned long prev; | |
2340 | unsigned long bit; | |
2341 | ||
6c9fe14f | 2342 | prev = xchg(&fs_info->pending_changes, 0); |
572d9ab7 DS |
2343 | if (!prev) |
2344 | return; | |
2345 | ||
7e1876ac DS |
2346 | bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE; |
2347 | if (prev & bit) | |
2348 | btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2349 | prev &= ~bit; | |
2350 | ||
2351 | bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE; | |
2352 | if (prev & bit) | |
2353 | btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2354 | prev &= ~bit; | |
2355 | ||
d51033d0 DS |
2356 | bit = 1 << BTRFS_PENDING_COMMIT; |
2357 | if (prev & bit) | |
2358 | btrfs_debug(fs_info, "pending commit done"); | |
2359 | prev &= ~bit; | |
2360 | ||
572d9ab7 DS |
2361 | if (prev) |
2362 | btrfs_warn(fs_info, | |
2363 | "unknown pending changes left 0x%lx, ignoring", prev); | |
2364 | } |